1d61ea1cbSPeter Xu 2457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 31da177e4SLinus Torvalds /* 41da177e4SLinus Torvalds * linux/mm/memory.c 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds /* 101da177e4SLinus Torvalds * demand-loading started 01.12.91 - seems it is high on the list of 111da177e4SLinus Torvalds * things wanted, and it should be easy to implement. - Linus 121da177e4SLinus Torvalds */ 131da177e4SLinus Torvalds 141da177e4SLinus Torvalds /* 151da177e4SLinus Torvalds * Ok, demand-loading was easy, shared pages a little bit tricker. Shared 161da177e4SLinus Torvalds * pages started 02.12.91, seems to work. - Linus. 171da177e4SLinus Torvalds * 181da177e4SLinus Torvalds * Tested sharing by executing about 30 /bin/sh: under the old kernel it 191da177e4SLinus Torvalds * would have taken more than the 6M I have free, but it worked well as 201da177e4SLinus Torvalds * far as I could see. 211da177e4SLinus Torvalds * 221da177e4SLinus Torvalds * Also corrected some "invalidate()"s - I wasn't doing enough of them. 231da177e4SLinus Torvalds */ 241da177e4SLinus Torvalds 251da177e4SLinus Torvalds /* 261da177e4SLinus Torvalds * Real VM (paging to/from disk) started 18.12.91. Much more work and 271da177e4SLinus Torvalds * thought has to go into this. Oh, well.. 281da177e4SLinus Torvalds * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. 291da177e4SLinus Torvalds * Found it. Everything seems to work now. 301da177e4SLinus Torvalds * 20.12.91 - Ok, making the swap-device changeable like the root. 311da177e4SLinus Torvalds */ 321da177e4SLinus Torvalds 331da177e4SLinus Torvalds /* 341da177e4SLinus Torvalds * 05.04.94 - Multi-page memory management added for v1.1. 351da177e4SLinus Torvalds * Idea by Alex Bligh (alex@cconcepts.co.uk) 361da177e4SLinus Torvalds * 371da177e4SLinus Torvalds * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG 381da177e4SLinus Torvalds * (Gerhard.Wichert@pdb.siemens.de) 391da177e4SLinus Torvalds * 401da177e4SLinus Torvalds * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) 411da177e4SLinus Torvalds */ 421da177e4SLinus Torvalds 431da177e4SLinus Torvalds #include <linux/kernel_stat.h> 441da177e4SLinus Torvalds #include <linux/mm.h> 4536090defSArnd Bergmann #include <linux/mm_inline.h> 466e84f315SIngo Molnar #include <linux/sched/mm.h> 47f7ccbae4SIngo Molnar #include <linux/sched/coredump.h> 486a3827d7SIngo Molnar #include <linux/sched/numa_balancing.h> 4929930025SIngo Molnar #include <linux/sched/task.h> 501da177e4SLinus Torvalds #include <linux/hugetlb.h> 511da177e4SLinus Torvalds #include <linux/mman.h> 521da177e4SLinus Torvalds #include <linux/swap.h> 531da177e4SLinus Torvalds #include <linux/highmem.h> 541da177e4SLinus Torvalds #include <linux/pagemap.h> 555042db43SJérôme Glisse #include <linux/memremap.h> 56b073d7f8SAlexander Potapenko #include <linux/kmsan.h> 579a840895SHugh Dickins #include <linux/ksm.h> 581da177e4SLinus Torvalds #include <linux/rmap.h> 59b95f1b31SPaul Gortmaker #include <linux/export.h> 600ff92245SShailabh Nagar #include <linux/delayacct.h> 611da177e4SLinus Torvalds #include <linux/init.h> 6201c8f1c4SDan Williams #include <linux/pfn_t.h> 63edc79b2aSPeter Zijlstra #include <linux/writeback.h> 648a9f3ccdSBalbir Singh #include <linux/memcontrol.h> 65cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 663dc14741SHugh Dickins #include <linux/swapops.h> 673dc14741SHugh Dickins #include <linux/elf.h> 685a0e3ad6STejun Heo #include <linux/gfp.h> 694daae3b4SMel Gorman #include <linux/migrate.h> 702fbc57c5SAndy Shevchenko #include <linux/string.h> 71467b171aSAneesh Kumar K.V #include <linux/memory-tiers.h> 721592eef0SKirill A. Shutemov #include <linux/debugfs.h> 736b251fc9SAndrea Arcangeli #include <linux/userfaultfd_k.h> 74bc2466e4SJan Kara #include <linux/dax.h> 756b31d595SMichal Hocko #include <linux/oom.h> 7698fa15f3SAnshuman Khandual #include <linux/numa.h> 77bce617edSPeter Xu #include <linux/perf_event.h> 78bce617edSPeter Xu #include <linux/ptrace.h> 79e80d3909SJoerg Roedel #include <linux/vmalloc.h> 8033024536SHuang Ying #include <linux/sched/sysctl.h> 811da177e4SLinus Torvalds 82b3d1411bSJoel Fernandes (Google) #include <trace/events/kmem.h> 83b3d1411bSJoel Fernandes (Google) 846952b61dSAlexey Dobriyan #include <asm/io.h> 8533a709b2SDave Hansen #include <asm/mmu_context.h> 861da177e4SLinus Torvalds #include <asm/pgalloc.h> 877c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 881da177e4SLinus Torvalds #include <asm/tlb.h> 891da177e4SLinus Torvalds #include <asm/tlbflush.h> 901da177e4SLinus Torvalds 91e80d3909SJoerg Roedel #include "pgalloc-track.h" 9242b77728SJan Beulich #include "internal.h" 93014bb1deSNeilBrown #include "swap.h" 9442b77728SJan Beulich 95af27d940SArnd Bergmann #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST) 9690572890SPeter Zijlstra #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. 9775980e97SPeter Zijlstra #endif 9875980e97SPeter Zijlstra 99a9ee6cf5SMike Rapoport #ifndef CONFIG_NUMA 1001da177e4SLinus Torvalds unsigned long max_mapnr; 1011da177e4SLinus Torvalds EXPORT_SYMBOL(max_mapnr); 102166f61b9STobin C Harding 103166f61b9STobin C Harding struct page *mem_map; 1041da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 1051da177e4SLinus Torvalds #endif 1061da177e4SLinus Torvalds 1075c041f5dSPeter Xu static vm_fault_t do_fault(struct vm_fault *vmf); 1082bad466cSPeter Xu static vm_fault_t do_anonymous_page(struct vm_fault *vmf); 1092bad466cSPeter Xu static bool vmf_pte_changed(struct vm_fault *vmf); 1102bad466cSPeter Xu 1112bad466cSPeter Xu /* 1122bad466cSPeter Xu * Return true if the original pte was a uffd-wp pte marker (so the pte was 1132bad466cSPeter Xu * wr-protected). 1142bad466cSPeter Xu */ 1152bad466cSPeter Xu static bool vmf_orig_pte_uffd_wp(struct vm_fault *vmf) 1162bad466cSPeter Xu { 1172bad466cSPeter Xu if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID)) 1182bad466cSPeter Xu return false; 1192bad466cSPeter Xu 1202bad466cSPeter Xu return pte_marker_uffd_wp(vmf->orig_pte); 1212bad466cSPeter Xu } 1225c041f5dSPeter Xu 1231da177e4SLinus Torvalds /* 1241da177e4SLinus Torvalds * A number of key systems in x86 including ioremap() rely on the assumption 1251da177e4SLinus Torvalds * that high_memory defines the upper bound on direct map memory, then end 126e99fb98dSKefeng Wang * of ZONE_NORMAL. 1271da177e4SLinus Torvalds */ 1281da177e4SLinus Torvalds void *high_memory; 1291da177e4SLinus Torvalds EXPORT_SYMBOL(high_memory); 1301da177e4SLinus Torvalds 13132a93233SIngo Molnar /* 13232a93233SIngo Molnar * Randomize the address space (stacks, mmaps, brk, etc.). 13332a93233SIngo Molnar * 13432a93233SIngo Molnar * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, 13532a93233SIngo Molnar * as ancient (libc5 based) binaries can segfault. ) 13632a93233SIngo Molnar */ 13732a93233SIngo Molnar int randomize_va_space __read_mostly = 13832a93233SIngo Molnar #ifdef CONFIG_COMPAT_BRK 13932a93233SIngo Molnar 1; 14032a93233SIngo Molnar #else 14132a93233SIngo Molnar 2; 14232a93233SIngo Molnar #endif 143a62eaf15SAndi Kleen 14446bdb427SWill Deacon #ifndef arch_wants_old_prefaulted_pte 14546bdb427SWill Deacon static inline bool arch_wants_old_prefaulted_pte(void) 14646bdb427SWill Deacon { 14746bdb427SWill Deacon /* 14846bdb427SWill Deacon * Transitioning a PTE from 'old' to 'young' can be expensive on 14946bdb427SWill Deacon * some architectures, even if it's performed in hardware. By 15046bdb427SWill Deacon * default, "false" means prefaulted entries will be 'young'. 15146bdb427SWill Deacon */ 15246bdb427SWill Deacon return false; 15346bdb427SWill Deacon } 15446bdb427SWill Deacon #endif 15546bdb427SWill Deacon 156a62eaf15SAndi Kleen static int __init disable_randmaps(char *s) 157a62eaf15SAndi Kleen { 158a62eaf15SAndi Kleen randomize_va_space = 0; 1599b41046cSOGAWA Hirofumi return 1; 160a62eaf15SAndi Kleen } 161a62eaf15SAndi Kleen __setup("norandmaps", disable_randmaps); 162a62eaf15SAndi Kleen 16362eede62SHugh Dickins unsigned long zero_pfn __read_mostly; 1640b70068eSArd Biesheuvel EXPORT_SYMBOL(zero_pfn); 1650b70068eSArd Biesheuvel 166166f61b9STobin C Harding unsigned long highest_memmap_pfn __read_mostly; 167166f61b9STobin C Harding 168a13ea5b7SHugh Dickins /* 169a13ea5b7SHugh Dickins * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() 170a13ea5b7SHugh Dickins */ 171a13ea5b7SHugh Dickins static int __init init_zero_pfn(void) 172a13ea5b7SHugh Dickins { 173a13ea5b7SHugh Dickins zero_pfn = page_to_pfn(ZERO_PAGE(0)); 174a13ea5b7SHugh Dickins return 0; 175a13ea5b7SHugh Dickins } 176e720e7d0SIlya Lipnitskiy early_initcall(init_zero_pfn); 177a62eaf15SAndi Kleen 178f1a79412SShakeel Butt void mm_trace_rss_stat(struct mm_struct *mm, int member) 179b3d1411bSJoel Fernandes (Google) { 180f1a79412SShakeel Butt trace_rss_stat(mm, member); 181b3d1411bSJoel Fernandes (Google) } 182d559db08SKAMEZAWA Hiroyuki 1831da177e4SLinus Torvalds /* 1841da177e4SLinus Torvalds * Note: this doesn't free the actual pages themselves. That 1851da177e4SLinus Torvalds * has been handled earlier when unmapping all the memory regions. 1861da177e4SLinus Torvalds */ 1879e1b32caSBenjamin Herrenschmidt static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, 1889e1b32caSBenjamin Herrenschmidt unsigned long addr) 1891da177e4SLinus Torvalds { 1902f569afdSMartin Schwidefsky pgtable_t token = pmd_pgtable(*pmd); 1911da177e4SLinus Torvalds pmd_clear(pmd); 1929e1b32caSBenjamin Herrenschmidt pte_free_tlb(tlb, token, addr); 193c4812909SKirill A. Shutemov mm_dec_nr_ptes(tlb->mm); 1941da177e4SLinus Torvalds } 1951da177e4SLinus Torvalds 196e0da382cSHugh Dickins static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, 197e0da382cSHugh Dickins unsigned long addr, unsigned long end, 198e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 1991da177e4SLinus Torvalds { 2001da177e4SLinus Torvalds pmd_t *pmd; 2011da177e4SLinus Torvalds unsigned long next; 202e0da382cSHugh Dickins unsigned long start; 2031da177e4SLinus Torvalds 204e0da382cSHugh Dickins start = addr; 2051da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 2061da177e4SLinus Torvalds do { 2071da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 2081da177e4SLinus Torvalds if (pmd_none_or_clear_bad(pmd)) 2091da177e4SLinus Torvalds continue; 2109e1b32caSBenjamin Herrenschmidt free_pte_range(tlb, pmd, addr); 2111da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 2121da177e4SLinus Torvalds 213e0da382cSHugh Dickins start &= PUD_MASK; 214e0da382cSHugh Dickins if (start < floor) 215e0da382cSHugh Dickins return; 216e0da382cSHugh Dickins if (ceiling) { 217e0da382cSHugh Dickins ceiling &= PUD_MASK; 218e0da382cSHugh Dickins if (!ceiling) 219e0da382cSHugh Dickins return; 2201da177e4SLinus Torvalds } 221e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 222e0da382cSHugh Dickins return; 223e0da382cSHugh Dickins 224e0da382cSHugh Dickins pmd = pmd_offset(pud, start); 225e0da382cSHugh Dickins pud_clear(pud); 2269e1b32caSBenjamin Herrenschmidt pmd_free_tlb(tlb, pmd, start); 227dc6c9a35SKirill A. Shutemov mm_dec_nr_pmds(tlb->mm); 2281da177e4SLinus Torvalds } 2291da177e4SLinus Torvalds 230c2febafcSKirill A. Shutemov static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, 231e0da382cSHugh Dickins unsigned long addr, unsigned long end, 232e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 2331da177e4SLinus Torvalds { 2341da177e4SLinus Torvalds pud_t *pud; 2351da177e4SLinus Torvalds unsigned long next; 236e0da382cSHugh Dickins unsigned long start; 2371da177e4SLinus Torvalds 238e0da382cSHugh Dickins start = addr; 239c2febafcSKirill A. Shutemov pud = pud_offset(p4d, addr); 2401da177e4SLinus Torvalds do { 2411da177e4SLinus Torvalds next = pud_addr_end(addr, end); 2421da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 2431da177e4SLinus Torvalds continue; 244e0da382cSHugh Dickins free_pmd_range(tlb, pud, addr, next, floor, ceiling); 2451da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 2461da177e4SLinus Torvalds 247c2febafcSKirill A. Shutemov start &= P4D_MASK; 248c2febafcSKirill A. Shutemov if (start < floor) 249c2febafcSKirill A. Shutemov return; 250c2febafcSKirill A. Shutemov if (ceiling) { 251c2febafcSKirill A. Shutemov ceiling &= P4D_MASK; 252c2febafcSKirill A. Shutemov if (!ceiling) 253c2febafcSKirill A. Shutemov return; 254c2febafcSKirill A. Shutemov } 255c2febafcSKirill A. Shutemov if (end - 1 > ceiling - 1) 256c2febafcSKirill A. Shutemov return; 257c2febafcSKirill A. Shutemov 258c2febafcSKirill A. Shutemov pud = pud_offset(p4d, start); 259c2febafcSKirill A. Shutemov p4d_clear(p4d); 260c2febafcSKirill A. Shutemov pud_free_tlb(tlb, pud, start); 261b4e98d9aSKirill A. Shutemov mm_dec_nr_puds(tlb->mm); 262c2febafcSKirill A. Shutemov } 263c2febafcSKirill A. Shutemov 264c2febafcSKirill A. Shutemov static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, 265c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 266c2febafcSKirill A. Shutemov unsigned long floor, unsigned long ceiling) 267c2febafcSKirill A. Shutemov { 268c2febafcSKirill A. Shutemov p4d_t *p4d; 269c2febafcSKirill A. Shutemov unsigned long next; 270c2febafcSKirill A. Shutemov unsigned long start; 271c2febafcSKirill A. Shutemov 272c2febafcSKirill A. Shutemov start = addr; 273c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, addr); 274c2febafcSKirill A. Shutemov do { 275c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 276c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(p4d)) 277c2febafcSKirill A. Shutemov continue; 278c2febafcSKirill A. Shutemov free_pud_range(tlb, p4d, addr, next, floor, ceiling); 279c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 280c2febafcSKirill A. Shutemov 281e0da382cSHugh Dickins start &= PGDIR_MASK; 282e0da382cSHugh Dickins if (start < floor) 283e0da382cSHugh Dickins return; 284e0da382cSHugh Dickins if (ceiling) { 285e0da382cSHugh Dickins ceiling &= PGDIR_MASK; 286e0da382cSHugh Dickins if (!ceiling) 287e0da382cSHugh Dickins return; 2881da177e4SLinus Torvalds } 289e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 290e0da382cSHugh Dickins return; 291e0da382cSHugh Dickins 292c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, start); 293e0da382cSHugh Dickins pgd_clear(pgd); 294c2febafcSKirill A. Shutemov p4d_free_tlb(tlb, p4d, start); 2951da177e4SLinus Torvalds } 2961da177e4SLinus Torvalds 2971da177e4SLinus Torvalds /* 298e0da382cSHugh Dickins * This function frees user-level page tables of a process. 2991da177e4SLinus Torvalds */ 30042b77728SJan Beulich void free_pgd_range(struct mmu_gather *tlb, 301e0da382cSHugh Dickins unsigned long addr, unsigned long end, 302e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 3031da177e4SLinus Torvalds { 3041da177e4SLinus Torvalds pgd_t *pgd; 3051da177e4SLinus Torvalds unsigned long next; 3061da177e4SLinus Torvalds 307e0da382cSHugh Dickins /* 308e0da382cSHugh Dickins * The next few lines have given us lots of grief... 309e0da382cSHugh Dickins * 310e0da382cSHugh Dickins * Why are we testing PMD* at this top level? Because often 311e0da382cSHugh Dickins * there will be no work to do at all, and we'd prefer not to 312e0da382cSHugh Dickins * go all the way down to the bottom just to discover that. 313e0da382cSHugh Dickins * 314e0da382cSHugh Dickins * Why all these "- 1"s? Because 0 represents both the bottom 315e0da382cSHugh Dickins * of the address space and the top of it (using -1 for the 316e0da382cSHugh Dickins * top wouldn't help much: the masks would do the wrong thing). 317e0da382cSHugh Dickins * The rule is that addr 0 and floor 0 refer to the bottom of 318e0da382cSHugh Dickins * the address space, but end 0 and ceiling 0 refer to the top 319e0da382cSHugh Dickins * Comparisons need to use "end - 1" and "ceiling - 1" (though 320e0da382cSHugh Dickins * that end 0 case should be mythical). 321e0da382cSHugh Dickins * 322e0da382cSHugh Dickins * Wherever addr is brought up or ceiling brought down, we must 323e0da382cSHugh Dickins * be careful to reject "the opposite 0" before it confuses the 324e0da382cSHugh Dickins * subsequent tests. But what about where end is brought down 325e0da382cSHugh Dickins * by PMD_SIZE below? no, end can't go down to 0 there. 326e0da382cSHugh Dickins * 327e0da382cSHugh Dickins * Whereas we round start (addr) and ceiling down, by different 328e0da382cSHugh Dickins * masks at different levels, in order to test whether a table 329e0da382cSHugh Dickins * now has no other vmas using it, so can be freed, we don't 330e0da382cSHugh Dickins * bother to round floor or end up - the tests don't need that. 331e0da382cSHugh Dickins */ 332e0da382cSHugh Dickins 333e0da382cSHugh Dickins addr &= PMD_MASK; 334e0da382cSHugh Dickins if (addr < floor) { 335e0da382cSHugh Dickins addr += PMD_SIZE; 336e0da382cSHugh Dickins if (!addr) 337e0da382cSHugh Dickins return; 338e0da382cSHugh Dickins } 339e0da382cSHugh Dickins if (ceiling) { 340e0da382cSHugh Dickins ceiling &= PMD_MASK; 341e0da382cSHugh Dickins if (!ceiling) 342e0da382cSHugh Dickins return; 343e0da382cSHugh Dickins } 344e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 345e0da382cSHugh Dickins end -= PMD_SIZE; 346e0da382cSHugh Dickins if (addr > end - 1) 347e0da382cSHugh Dickins return; 34807e32661SAneesh Kumar K.V /* 34907e32661SAneesh Kumar K.V * We add page table cache pages with PAGE_SIZE, 35007e32661SAneesh Kumar K.V * (see pte_free_tlb()), flush the tlb if we need 35107e32661SAneesh Kumar K.V */ 352ed6a7935SPeter Zijlstra tlb_change_page_size(tlb, PAGE_SIZE); 35342b77728SJan Beulich pgd = pgd_offset(tlb->mm, addr); 3541da177e4SLinus Torvalds do { 3551da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 3561da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 3571da177e4SLinus Torvalds continue; 358c2febafcSKirill A. Shutemov free_p4d_range(tlb, pgd, addr, next, floor, ceiling); 3591da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 360e0da382cSHugh Dickins } 361e0da382cSHugh Dickins 362fd892593SLiam R. Howlett void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas, 363763ecb03SLiam R. Howlett struct vm_area_struct *vma, unsigned long floor, 36498e51a22SSuren Baghdasaryan unsigned long ceiling, bool mm_wr_locked) 365e0da382cSHugh Dickins { 366763ecb03SLiam R. Howlett do { 367e0da382cSHugh Dickins unsigned long addr = vma->vm_start; 368763ecb03SLiam R. Howlett struct vm_area_struct *next; 369763ecb03SLiam R. Howlett 370763ecb03SLiam R. Howlett /* 371763ecb03SLiam R. Howlett * Note: USER_PGTABLES_CEILING may be passed as ceiling and may 372763ecb03SLiam R. Howlett * be 0. This will underflow and is okay. 373763ecb03SLiam R. Howlett */ 374fd892593SLiam R. Howlett next = mas_find(mas, ceiling - 1); 375d2406291SPeng Zhang if (unlikely(xa_is_zero(next))) 376d2406291SPeng Zhang next = NULL; 377e0da382cSHugh Dickins 3788f4f8c16SHugh Dickins /* 37925d9e2d1Snpiggin@suse.de * Hide vma from rmap and truncate_pagecache before freeing 38025d9e2d1Snpiggin@suse.de * pgtables 3818f4f8c16SHugh Dickins */ 38298e51a22SSuren Baghdasaryan if (mm_wr_locked) 38398e51a22SSuren Baghdasaryan vma_start_write(vma); 3845beb4930SRik van Riel unlink_anon_vmas(vma); 3858f4f8c16SHugh Dickins unlink_file_vma(vma); 3868f4f8c16SHugh Dickins 3879da61aefSDavid Gibson if (is_vm_hugetlb_page(vma)) { 3883bf5ee95SHugh Dickins hugetlb_free_pgd_range(tlb, addr, vma->vm_end, 3893bf5ee95SHugh Dickins floor, next ? next->vm_start : ceiling); 3903bf5ee95SHugh Dickins } else { 3913bf5ee95SHugh Dickins /* 3923bf5ee95SHugh Dickins * Optimization: gather nearby vmas into one call down 3933bf5ee95SHugh Dickins */ 3943bf5ee95SHugh Dickins while (next && next->vm_start <= vma->vm_end + PMD_SIZE 3954866920bSDavid Gibson && !is_vm_hugetlb_page(next)) { 396e0da382cSHugh Dickins vma = next; 397fd892593SLiam R. Howlett next = mas_find(mas, ceiling - 1); 398d2406291SPeng Zhang if (unlikely(xa_is_zero(next))) 399d2406291SPeng Zhang next = NULL; 40098e51a22SSuren Baghdasaryan if (mm_wr_locked) 40198e51a22SSuren Baghdasaryan vma_start_write(vma); 4025beb4930SRik van Riel unlink_anon_vmas(vma); 4038f4f8c16SHugh Dickins unlink_file_vma(vma); 404e0da382cSHugh Dickins } 4053bf5ee95SHugh Dickins free_pgd_range(tlb, addr, vma->vm_end, 406e0da382cSHugh Dickins floor, next ? next->vm_start : ceiling); 4073bf5ee95SHugh Dickins } 408e0da382cSHugh Dickins vma = next; 409763ecb03SLiam R. Howlett } while (vma); 4101da177e4SLinus Torvalds } 4111da177e4SLinus Torvalds 41203c4f204SQi Zheng void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte) 4131da177e4SLinus Torvalds { 41403c4f204SQi Zheng spinlock_t *ptl = pmd_lock(mm, pmd); 4151bb3630eSHugh Dickins 41603c4f204SQi Zheng if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 41703c4f204SQi Zheng mm_inc_nr_ptes(mm); 418362a61adSNick Piggin /* 419362a61adSNick Piggin * Ensure all pte setup (eg. pte page lock and page clearing) are 420362a61adSNick Piggin * visible before the pte is made visible to other CPUs by being 421362a61adSNick Piggin * put into page tables. 422362a61adSNick Piggin * 423362a61adSNick Piggin * The other side of the story is the pointer chasing in the page 424362a61adSNick Piggin * table walking code (when walking the page table without locking; 425362a61adSNick Piggin * ie. most of the time). Fortunately, these data accesses consist 426362a61adSNick Piggin * of a chain of data-dependent loads, meaning most CPUs (alpha 427362a61adSNick Piggin * being the notable exception) will already guarantee loads are 428362a61adSNick Piggin * seen in-order. See the alpha page table accessors for the 429bb7cdd38SWill Deacon * smp_rmb() barriers in page table walking code. 430362a61adSNick Piggin */ 431362a61adSNick Piggin smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ 432ed33b5a6SQi Zheng pmd_populate(mm, pmd, *pte); 433ed33b5a6SQi Zheng *pte = NULL; 4344b471e88SKirill A. Shutemov } 435c4088ebdSKirill A. Shutemov spin_unlock(ptl); 436ed33b5a6SQi Zheng } 437ed33b5a6SQi Zheng 438ed33b5a6SQi Zheng int __pte_alloc(struct mm_struct *mm, pmd_t *pmd) 439ed33b5a6SQi Zheng { 440ed33b5a6SQi Zheng pgtable_t new = pte_alloc_one(mm); 441ed33b5a6SQi Zheng if (!new) 442ed33b5a6SQi Zheng return -ENOMEM; 4438ac1f832SAndrea Arcangeli 44403c4f204SQi Zheng pmd_install(mm, pmd, &new); 4452f569afdSMartin Schwidefsky if (new) 4462f569afdSMartin Schwidefsky pte_free(mm, new); 4471bb3630eSHugh Dickins return 0; 4481da177e4SLinus Torvalds } 4491da177e4SLinus Torvalds 4504cf58924SJoel Fernandes (Google) int __pte_alloc_kernel(pmd_t *pmd) 4511da177e4SLinus Torvalds { 4524cf58924SJoel Fernandes (Google) pte_t *new = pte_alloc_one_kernel(&init_mm); 4531da177e4SLinus Torvalds if (!new) 4541bb3630eSHugh Dickins return -ENOMEM; 4551da177e4SLinus Torvalds 456872fec16SHugh Dickins spin_lock(&init_mm.page_table_lock); 4578ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 458ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 459872fec16SHugh Dickins pmd_populate_kernel(&init_mm, pmd, new); 4602f569afdSMartin Schwidefsky new = NULL; 4614b471e88SKirill A. Shutemov } 462872fec16SHugh Dickins spin_unlock(&init_mm.page_table_lock); 4632f569afdSMartin Schwidefsky if (new) 4642f569afdSMartin Schwidefsky pte_free_kernel(&init_mm, new); 4651bb3630eSHugh Dickins return 0; 4661da177e4SLinus Torvalds } 4671da177e4SLinus Torvalds 468d559db08SKAMEZAWA Hiroyuki static inline void init_rss_vec(int *rss) 469ae859762SHugh Dickins { 470d559db08SKAMEZAWA Hiroyuki memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); 471d559db08SKAMEZAWA Hiroyuki } 472d559db08SKAMEZAWA Hiroyuki 473d559db08SKAMEZAWA Hiroyuki static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) 474d559db08SKAMEZAWA Hiroyuki { 475d559db08SKAMEZAWA Hiroyuki int i; 476d559db08SKAMEZAWA Hiroyuki 477d559db08SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) 478d559db08SKAMEZAWA Hiroyuki if (rss[i]) 479d559db08SKAMEZAWA Hiroyuki add_mm_counter(mm, i, rss[i]); 480ae859762SHugh Dickins } 481ae859762SHugh Dickins 4821da177e4SLinus Torvalds /* 4836aab341eSLinus Torvalds * This function is called to print an error when a bad pte 4846aab341eSLinus Torvalds * is found. For example, we might have a PFN-mapped pte in 4856aab341eSLinus Torvalds * a region that doesn't allow it. 486b5810039SNick Piggin * 487b5810039SNick Piggin * The calling function must still handle the error. 488b5810039SNick Piggin */ 4893dc14741SHugh Dickins static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, 4903dc14741SHugh Dickins pte_t pte, struct page *page) 491b5810039SNick Piggin { 4923dc14741SHugh Dickins pgd_t *pgd = pgd_offset(vma->vm_mm, addr); 493c2febafcSKirill A. Shutemov p4d_t *p4d = p4d_offset(pgd, addr); 494c2febafcSKirill A. Shutemov pud_t *pud = pud_offset(p4d, addr); 4953dc14741SHugh Dickins pmd_t *pmd = pmd_offset(pud, addr); 4963dc14741SHugh Dickins struct address_space *mapping; 4973dc14741SHugh Dickins pgoff_t index; 498d936cf9bSHugh Dickins static unsigned long resume; 499d936cf9bSHugh Dickins static unsigned long nr_shown; 500d936cf9bSHugh Dickins static unsigned long nr_unshown; 501d936cf9bSHugh Dickins 502d936cf9bSHugh Dickins /* 503d936cf9bSHugh Dickins * Allow a burst of 60 reports, then keep quiet for that minute; 504d936cf9bSHugh Dickins * or allow a steady drip of one report per second. 505d936cf9bSHugh Dickins */ 506d936cf9bSHugh Dickins if (nr_shown == 60) { 507d936cf9bSHugh Dickins if (time_before(jiffies, resume)) { 508d936cf9bSHugh Dickins nr_unshown++; 509d936cf9bSHugh Dickins return; 510d936cf9bSHugh Dickins } 511d936cf9bSHugh Dickins if (nr_unshown) { 5121170532bSJoe Perches pr_alert("BUG: Bad page map: %lu messages suppressed\n", 513d936cf9bSHugh Dickins nr_unshown); 514d936cf9bSHugh Dickins nr_unshown = 0; 515d936cf9bSHugh Dickins } 516d936cf9bSHugh Dickins nr_shown = 0; 517d936cf9bSHugh Dickins } 518d936cf9bSHugh Dickins if (nr_shown++ == 0) 519d936cf9bSHugh Dickins resume = jiffies + 60 * HZ; 5203dc14741SHugh Dickins 5213dc14741SHugh Dickins mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; 5223dc14741SHugh Dickins index = linear_page_index(vma, addr); 5233dc14741SHugh Dickins 5241170532bSJoe Perches pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", 5253dc14741SHugh Dickins current->comm, 5263dc14741SHugh Dickins (long long)pte_val(pte), (long long)pmd_val(*pmd)); 527718a3821SWu Fengguang if (page) 528f0b791a3SDave Hansen dump_page(page, "bad pte"); 5296aa9b8b2SKefeng Wang pr_alert("addr:%px vm_flags:%08lx anon_vma:%px mapping:%px index:%lx\n", 5303dc14741SHugh Dickins (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); 5317e0a1265SMatthew Wilcox (Oracle) pr_alert("file:%pD fault:%ps mmap:%ps read_folio:%ps\n", 5322682582aSKonstantin Khlebnikov vma->vm_file, 5332682582aSKonstantin Khlebnikov vma->vm_ops ? vma->vm_ops->fault : NULL, 5342682582aSKonstantin Khlebnikov vma->vm_file ? vma->vm_file->f_op->mmap : NULL, 5357e0a1265SMatthew Wilcox (Oracle) mapping ? mapping->a_ops->read_folio : NULL); 536b5810039SNick Piggin dump_stack(); 537373d4d09SRusty Russell add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); 538b5810039SNick Piggin } 539b5810039SNick Piggin 540b5810039SNick Piggin /* 5417e675137SNick Piggin * vm_normal_page -- This function gets the "struct page" associated with a pte. 5426aab341eSLinus Torvalds * 5437e675137SNick Piggin * "Special" mappings do not wish to be associated with a "struct page" (either 5447e675137SNick Piggin * it doesn't exist, or it exists but they don't want to touch it). In this 5457e675137SNick Piggin * case, NULL is returned here. "Normal" mappings do have a struct page. 546b379d790SJared Hulbert * 5477e675137SNick Piggin * There are 2 broad cases. Firstly, an architecture may define a pte_special() 5487e675137SNick Piggin * pte bit, in which case this function is trivial. Secondly, an architecture 5497e675137SNick Piggin * may not have a spare pte bit, which requires a more complicated scheme, 5507e675137SNick Piggin * described below. 5517e675137SNick Piggin * 5527e675137SNick Piggin * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a 5537e675137SNick Piggin * special mapping (even if there are underlying and valid "struct pages"). 5547e675137SNick Piggin * COWed pages of a VM_PFNMAP are always normal. 5556aab341eSLinus Torvalds * 556b379d790SJared Hulbert * The way we recognize COWed pages within VM_PFNMAP mappings is through the 557b379d790SJared Hulbert * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit 5587e675137SNick Piggin * set, and the vm_pgoff will point to the first PFN mapped: thus every special 5597e675137SNick Piggin * mapping will always honor the rule 5606aab341eSLinus Torvalds * 5616aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 5626aab341eSLinus Torvalds * 5637e675137SNick Piggin * And for normal mappings this is false. 564b379d790SJared Hulbert * 5657e675137SNick Piggin * This restricts such mappings to be a linear translation from virtual address 5667e675137SNick Piggin * to pfn. To get around this restriction, we allow arbitrary mappings so long 5677e675137SNick Piggin * as the vma is not a COW mapping; in that case, we know that all ptes are 5687e675137SNick Piggin * special (because none can have been COWed). 569b379d790SJared Hulbert * 570b379d790SJared Hulbert * 5717e675137SNick Piggin * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. 5727e675137SNick Piggin * 573b379d790SJared Hulbert * VM_MIXEDMAP mappings can likewise contain memory with or without "struct 574b379d790SJared Hulbert * page" backing, however the difference is that _all_ pages with a struct 575b379d790SJared Hulbert * page (that is, those where pfn_valid is true) are refcounted and considered 576b379d790SJared Hulbert * normal pages by the VM. The disadvantage is that pages are refcounted 577b379d790SJared Hulbert * (which can be slower and simply not an option for some PFNMAP users). The 578b379d790SJared Hulbert * advantage is that we don't have to follow the strict linearity rule of 579b379d790SJared Hulbert * PFNMAP mappings in order to support COWable mappings. 580b379d790SJared Hulbert * 581ee498ed7SHugh Dickins */ 58225b2995aSChristoph Hellwig struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 58325b2995aSChristoph Hellwig pte_t pte) 584ee498ed7SHugh Dickins { 58522b31eecSHugh Dickins unsigned long pfn = pte_pfn(pte); 5867e675137SNick Piggin 58700b3a331SLaurent Dufour if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) { 588b38af472SHugh Dickins if (likely(!pte_special(pte))) 58922b31eecSHugh Dickins goto check_pfn; 590667a0a06SDavid Vrabel if (vma->vm_ops && vma->vm_ops->find_special_page) 591667a0a06SDavid Vrabel return vma->vm_ops->find_special_page(vma, addr); 592a13ea5b7SHugh Dickins if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 593a13ea5b7SHugh Dickins return NULL; 594df6ad698SJérôme Glisse if (is_zero_pfn(pfn)) 595df6ad698SJérôme Glisse return NULL; 596e1fb4a08SDave Jiang if (pte_devmap(pte)) 5973218f871SAlex Sierra /* 5983218f871SAlex Sierra * NOTE: New users of ZONE_DEVICE will not set pte_devmap() 5993218f871SAlex Sierra * and will have refcounts incremented on their struct pages 6003218f871SAlex Sierra * when they are inserted into PTEs, thus they are safe to 6013218f871SAlex Sierra * return here. Legacy ZONE_DEVICE pages that set pte_devmap() 6023218f871SAlex Sierra * do not have refcounts. Example of legacy ZONE_DEVICE is 6033218f871SAlex Sierra * MEMORY_DEVICE_FS_DAX type in pmem or virtio_fs drivers. 6043218f871SAlex Sierra */ 605e1fb4a08SDave Jiang return NULL; 606e1fb4a08SDave Jiang 60722b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 6087e675137SNick Piggin return NULL; 6097e675137SNick Piggin } 6107e675137SNick Piggin 61100b3a331SLaurent Dufour /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */ 6127e675137SNick Piggin 613b379d790SJared Hulbert if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 614b379d790SJared Hulbert if (vma->vm_flags & VM_MIXEDMAP) { 615b379d790SJared Hulbert if (!pfn_valid(pfn)) 616b379d790SJared Hulbert return NULL; 617b379d790SJared Hulbert goto out; 618b379d790SJared Hulbert } else { 6197e675137SNick Piggin unsigned long off; 6207e675137SNick Piggin off = (addr - vma->vm_start) >> PAGE_SHIFT; 6216aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 6226aab341eSLinus Torvalds return NULL; 62367121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 624fb155c16SLinus Torvalds return NULL; 6256aab341eSLinus Torvalds } 626b379d790SJared Hulbert } 6276aab341eSLinus Torvalds 628b38af472SHugh Dickins if (is_zero_pfn(pfn)) 629b38af472SHugh Dickins return NULL; 63000b3a331SLaurent Dufour 63122b31eecSHugh Dickins check_pfn: 63222b31eecSHugh Dickins if (unlikely(pfn > highest_memmap_pfn)) { 63322b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 63422b31eecSHugh Dickins return NULL; 63522b31eecSHugh Dickins } 6366aab341eSLinus Torvalds 6376aab341eSLinus Torvalds /* 6387e675137SNick Piggin * NOTE! We still have PageReserved() pages in the page tables. 6397e675137SNick Piggin * eg. VDSO mappings can cause them to exist. 6406aab341eSLinus Torvalds */ 641b379d790SJared Hulbert out: 6426aab341eSLinus Torvalds return pfn_to_page(pfn); 643ee498ed7SHugh Dickins } 644ee498ed7SHugh Dickins 645318e9342SVishal Moola (Oracle) struct folio *vm_normal_folio(struct vm_area_struct *vma, unsigned long addr, 646318e9342SVishal Moola (Oracle) pte_t pte) 647318e9342SVishal Moola (Oracle) { 648318e9342SVishal Moola (Oracle) struct page *page = vm_normal_page(vma, addr, pte); 649318e9342SVishal Moola (Oracle) 650318e9342SVishal Moola (Oracle) if (page) 651318e9342SVishal Moola (Oracle) return page_folio(page); 652318e9342SVishal Moola (Oracle) return NULL; 653318e9342SVishal Moola (Oracle) } 654318e9342SVishal Moola (Oracle) 65528093f9fSGerald Schaefer #ifdef CONFIG_TRANSPARENT_HUGEPAGE 65628093f9fSGerald Schaefer struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, 65728093f9fSGerald Schaefer pmd_t pmd) 65828093f9fSGerald Schaefer { 65928093f9fSGerald Schaefer unsigned long pfn = pmd_pfn(pmd); 66028093f9fSGerald Schaefer 66128093f9fSGerald Schaefer /* 66228093f9fSGerald Schaefer * There is no pmd_special() but there may be special pmds, e.g. 66328093f9fSGerald Schaefer * in a direct-access (dax) mapping, so let's just replicate the 66400b3a331SLaurent Dufour * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here. 66528093f9fSGerald Schaefer */ 66628093f9fSGerald Schaefer if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 66728093f9fSGerald Schaefer if (vma->vm_flags & VM_MIXEDMAP) { 66828093f9fSGerald Schaefer if (!pfn_valid(pfn)) 66928093f9fSGerald Schaefer return NULL; 67028093f9fSGerald Schaefer goto out; 67128093f9fSGerald Schaefer } else { 67228093f9fSGerald Schaefer unsigned long off; 67328093f9fSGerald Schaefer off = (addr - vma->vm_start) >> PAGE_SHIFT; 67428093f9fSGerald Schaefer if (pfn == vma->vm_pgoff + off) 67528093f9fSGerald Schaefer return NULL; 67628093f9fSGerald Schaefer if (!is_cow_mapping(vma->vm_flags)) 67728093f9fSGerald Schaefer return NULL; 67828093f9fSGerald Schaefer } 67928093f9fSGerald Schaefer } 68028093f9fSGerald Schaefer 681e1fb4a08SDave Jiang if (pmd_devmap(pmd)) 682e1fb4a08SDave Jiang return NULL; 6833cde287bSYu Zhao if (is_huge_zero_pmd(pmd)) 68428093f9fSGerald Schaefer return NULL; 68528093f9fSGerald Schaefer if (unlikely(pfn > highest_memmap_pfn)) 68628093f9fSGerald Schaefer return NULL; 68728093f9fSGerald Schaefer 68828093f9fSGerald Schaefer /* 68928093f9fSGerald Schaefer * NOTE! We still have PageReserved() pages in the page tables. 69028093f9fSGerald Schaefer * eg. VDSO mappings can cause them to exist. 69128093f9fSGerald Schaefer */ 69228093f9fSGerald Schaefer out: 69328093f9fSGerald Schaefer return pfn_to_page(pfn); 69428093f9fSGerald Schaefer } 69565610453SKefeng Wang 69665610453SKefeng Wang struct folio *vm_normal_folio_pmd(struct vm_area_struct *vma, 69765610453SKefeng Wang unsigned long addr, pmd_t pmd) 69865610453SKefeng Wang { 69965610453SKefeng Wang struct page *page = vm_normal_page_pmd(vma, addr, pmd); 70065610453SKefeng Wang 70165610453SKefeng Wang if (page) 70265610453SKefeng Wang return page_folio(page); 70365610453SKefeng Wang return NULL; 70465610453SKefeng Wang } 70528093f9fSGerald Schaefer #endif 70628093f9fSGerald Schaefer 707b756a3b5SAlistair Popple static void restore_exclusive_pte(struct vm_area_struct *vma, 708b756a3b5SAlistair Popple struct page *page, unsigned long address, 709b756a3b5SAlistair Popple pte_t *ptep) 710b756a3b5SAlistair Popple { 711b832a354SDavid Hildenbrand struct folio *folio = page_folio(page); 712c33c7948SRyan Roberts pte_t orig_pte; 713b756a3b5SAlistair Popple pte_t pte; 714b756a3b5SAlistair Popple swp_entry_t entry; 715b756a3b5SAlistair Popple 716c33c7948SRyan Roberts orig_pte = ptep_get(ptep); 717b756a3b5SAlistair Popple pte = pte_mkold(mk_pte(page, READ_ONCE(vma->vm_page_prot))); 718c33c7948SRyan Roberts if (pte_swp_soft_dirty(orig_pte)) 719b756a3b5SAlistair Popple pte = pte_mksoft_dirty(pte); 720b756a3b5SAlistair Popple 721c33c7948SRyan Roberts entry = pte_to_swp_entry(orig_pte); 722c33c7948SRyan Roberts if (pte_swp_uffd_wp(orig_pte)) 723b756a3b5SAlistair Popple pte = pte_mkuffd_wp(pte); 724b756a3b5SAlistair Popple else if (is_writable_device_exclusive_entry(entry)) 725b756a3b5SAlistair Popple pte = maybe_mkwrite(pte_mkdirty(pte), vma); 726b756a3b5SAlistair Popple 727b832a354SDavid Hildenbrand VM_BUG_ON_FOLIO(pte_write(pte) && (!folio_test_anon(folio) && 728b832a354SDavid Hildenbrand PageAnonExclusive(page)), folio); 7296c287605SDavid Hildenbrand 730b756a3b5SAlistair Popple /* 731b756a3b5SAlistair Popple * No need to take a page reference as one was already 732b756a3b5SAlistair Popple * created when the swap entry was made. 733b756a3b5SAlistair Popple */ 734b832a354SDavid Hildenbrand if (folio_test_anon(folio)) 735b832a354SDavid Hildenbrand folio_add_anon_rmap_pte(folio, page, vma, address, RMAP_NONE); 736b756a3b5SAlistair Popple else 737b756a3b5SAlistair Popple /* 738b756a3b5SAlistair Popple * Currently device exclusive access only supports anonymous 739b756a3b5SAlistair Popple * memory so the entry shouldn't point to a filebacked page. 740b756a3b5SAlistair Popple */ 7414d8ff640SMiaohe Lin WARN_ON_ONCE(1); 742b756a3b5SAlistair Popple 7431eba86c0SPasha Tatashin set_pte_at(vma->vm_mm, address, ptep, pte); 7441eba86c0SPasha Tatashin 745b756a3b5SAlistair Popple /* 746b756a3b5SAlistair Popple * No need to invalidate - it was non-present before. However 747b756a3b5SAlistair Popple * secondary CPUs may have mappings that need invalidating. 748b756a3b5SAlistair Popple */ 749b756a3b5SAlistair Popple update_mmu_cache(vma, address, ptep); 750b756a3b5SAlistair Popple } 751b756a3b5SAlistair Popple 752b756a3b5SAlistair Popple /* 753b756a3b5SAlistair Popple * Tries to restore an exclusive pte if the page lock can be acquired without 754b756a3b5SAlistair Popple * sleeping. 755b756a3b5SAlistair Popple */ 756b756a3b5SAlistair Popple static int 757b756a3b5SAlistair Popple try_restore_exclusive_pte(pte_t *src_pte, struct vm_area_struct *vma, 758b756a3b5SAlistair Popple unsigned long addr) 759b756a3b5SAlistair Popple { 760c33c7948SRyan Roberts swp_entry_t entry = pte_to_swp_entry(ptep_get(src_pte)); 761b756a3b5SAlistair Popple struct page *page = pfn_swap_entry_to_page(entry); 762b756a3b5SAlistair Popple 763b756a3b5SAlistair Popple if (trylock_page(page)) { 764b756a3b5SAlistair Popple restore_exclusive_pte(vma, page, addr, src_pte); 765b756a3b5SAlistair Popple unlock_page(page); 766b756a3b5SAlistair Popple return 0; 767b756a3b5SAlistair Popple } 768b756a3b5SAlistair Popple 769b756a3b5SAlistair Popple return -EBUSY; 770b756a3b5SAlistair Popple } 771b756a3b5SAlistair Popple 772ee498ed7SHugh Dickins /* 7731da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 7741da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 7751da177e4SLinus Torvalds * covered by this vma. 7761da177e4SLinus Torvalds */ 7771da177e4SLinus Torvalds 778df3a57d1SLinus Torvalds static unsigned long 779df3a57d1SLinus Torvalds copy_nonpresent_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 7808f34f1eaSPeter Xu pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *dst_vma, 7818f34f1eaSPeter Xu struct vm_area_struct *src_vma, unsigned long addr, int *rss) 7821da177e4SLinus Torvalds { 7838f34f1eaSPeter Xu unsigned long vm_flags = dst_vma->vm_flags; 784c33c7948SRyan Roberts pte_t orig_pte = ptep_get(src_pte); 785c33c7948SRyan Roberts pte_t pte = orig_pte; 78608e7795eSDavid Hildenbrand struct folio *folio; 7871da177e4SLinus Torvalds struct page *page; 788c33c7948SRyan Roberts swp_entry_t entry = pte_to_swp_entry(orig_pte); 7890697212aSChristoph Lameter 7902022b4d1SHugh Dickins if (likely(!non_swap_entry(entry))) { 791570a335bSHugh Dickins if (swap_duplicate(entry) < 0) 7929a5cc85cSAlistair Popple return -EIO; 793570a335bSHugh Dickins 7941da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 7951da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 7961da177e4SLinus Torvalds spin_lock(&mmlist_lock); 797f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 798f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 799f412ac08SHugh Dickins &src_mm->mmlist); 8001da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 8011da177e4SLinus Torvalds } 8021493a191SDavid Hildenbrand /* Mark the swap entry as shared. */ 803c33c7948SRyan Roberts if (pte_swp_exclusive(orig_pte)) { 804c33c7948SRyan Roberts pte = pte_swp_clear_exclusive(orig_pte); 8051493a191SDavid Hildenbrand set_pte_at(src_mm, addr, src_pte, pte); 8061493a191SDavid Hildenbrand } 807b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]++; 8082022b4d1SHugh Dickins } else if (is_migration_entry(entry)) { 809530c2a0dSKefeng Wang folio = pfn_swap_entry_folio(entry); 8109f9f1acdSKonstantin Khlebnikov 811a23f517bSKefeng Wang rss[mm_counter(folio)]++; 8129f9f1acdSKonstantin Khlebnikov 8136c287605SDavid Hildenbrand if (!is_readable_migration_entry(entry) && 8140697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 8150697212aSChristoph Lameter /* 8166c287605SDavid Hildenbrand * COW mappings require pages in both parent and child 8176c287605SDavid Hildenbrand * to be set to read. A previously exclusive entry is 8186c287605SDavid Hildenbrand * now shared. 8190697212aSChristoph Lameter */ 8204dd845b5SAlistair Popple entry = make_readable_migration_entry( 8214dd845b5SAlistair Popple swp_offset(entry)); 8220697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 823c33c7948SRyan Roberts if (pte_swp_soft_dirty(orig_pte)) 824c3d16e16SCyrill Gorcunov pte = pte_swp_mksoft_dirty(pte); 825c33c7948SRyan Roberts if (pte_swp_uffd_wp(orig_pte)) 826f45ec5ffSPeter Xu pte = pte_swp_mkuffd_wp(pte); 8270697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 8280697212aSChristoph Lameter } 8295042db43SJérôme Glisse } else if (is_device_private_entry(entry)) { 830af5cdaf8SAlistair Popple page = pfn_swap_entry_to_page(entry); 83108e7795eSDavid Hildenbrand folio = page_folio(page); 8325042db43SJérôme Glisse 8335042db43SJérôme Glisse /* 8345042db43SJérôme Glisse * Update rss count even for unaddressable pages, as 8355042db43SJérôme Glisse * they should treated just like normal pages in this 8365042db43SJérôme Glisse * respect. 8375042db43SJérôme Glisse * 8385042db43SJérôme Glisse * We will likely want to have some new rss counters 8395042db43SJérôme Glisse * for unaddressable pages, at some point. But for now 8405042db43SJérôme Glisse * keep things as they are. 8415042db43SJérôme Glisse */ 84208e7795eSDavid Hildenbrand folio_get(folio); 843a23f517bSKefeng Wang rss[mm_counter(folio)]++; 844fb3d824dSDavid Hildenbrand /* Cannot fail as these pages cannot get pinned. */ 84508e7795eSDavid Hildenbrand folio_try_dup_anon_rmap_pte(folio, page, src_vma); 8465042db43SJérôme Glisse 8475042db43SJérôme Glisse /* 8485042db43SJérôme Glisse * We do not preserve soft-dirty information, because so 8495042db43SJérôme Glisse * far, checkpoint/restore is the only feature that 8505042db43SJérôme Glisse * requires that. And checkpoint/restore does not work 8515042db43SJérôme Glisse * when a device driver is involved (you cannot easily 8525042db43SJérôme Glisse * save and restore device driver state). 8535042db43SJérôme Glisse */ 8544dd845b5SAlistair Popple if (is_writable_device_private_entry(entry) && 8555042db43SJérôme Glisse is_cow_mapping(vm_flags)) { 8564dd845b5SAlistair Popple entry = make_readable_device_private_entry( 8574dd845b5SAlistair Popple swp_offset(entry)); 8585042db43SJérôme Glisse pte = swp_entry_to_pte(entry); 859c33c7948SRyan Roberts if (pte_swp_uffd_wp(orig_pte)) 860f45ec5ffSPeter Xu pte = pte_swp_mkuffd_wp(pte); 8615042db43SJérôme Glisse set_pte_at(src_mm, addr, src_pte, pte); 8625042db43SJérôme Glisse } 863b756a3b5SAlistair Popple } else if (is_device_exclusive_entry(entry)) { 864b756a3b5SAlistair Popple /* 865b756a3b5SAlistair Popple * Make device exclusive entries present by restoring the 866b756a3b5SAlistair Popple * original entry then copying as for a present pte. Device 867b756a3b5SAlistair Popple * exclusive entries currently only support private writable 868b756a3b5SAlistair Popple * (ie. COW) mappings. 869b756a3b5SAlistair Popple */ 870b756a3b5SAlistair Popple VM_BUG_ON(!is_cow_mapping(src_vma->vm_flags)); 871b756a3b5SAlistair Popple if (try_restore_exclusive_pte(src_pte, src_vma, addr)) 872b756a3b5SAlistair Popple return -EBUSY; 873b756a3b5SAlistair Popple return -ENOENT; 874c56d1b62SPeter Xu } else if (is_pte_marker_entry(entry)) { 875af19487fSAxel Rasmussen pte_marker marker = copy_pte_marker(entry, dst_vma); 876af19487fSAxel Rasmussen 877af19487fSAxel Rasmussen if (marker) 878af19487fSAxel Rasmussen set_pte_at(dst_mm, addr, dst_pte, 879af19487fSAxel Rasmussen make_pte_marker(marker)); 880c56d1b62SPeter Xu return 0; 8811da177e4SLinus Torvalds } 8828f34f1eaSPeter Xu if (!userfaultfd_wp(dst_vma)) 8838f34f1eaSPeter Xu pte = pte_swp_clear_uffd_wp(pte); 884df3a57d1SLinus Torvalds set_pte_at(dst_mm, addr, dst_pte, pte); 885df3a57d1SLinus Torvalds return 0; 8861da177e4SLinus Torvalds } 8871da177e4SLinus Torvalds 88870e806e4SPeter Xu /* 889b51ad4f8SDavid Hildenbrand * Copy a present and normal page. 89070e806e4SPeter Xu * 891b51ad4f8SDavid Hildenbrand * NOTE! The usual case is that this isn't required; 892b51ad4f8SDavid Hildenbrand * instead, the caller can just increase the page refcount 893b51ad4f8SDavid Hildenbrand * and re-use the pte the traditional way. 89470e806e4SPeter Xu * 89570e806e4SPeter Xu * And if we need a pre-allocated page but don't yet have 89670e806e4SPeter Xu * one, return a negative error to let the preallocation 89770e806e4SPeter Xu * code know so that it can do so outside the page table 89870e806e4SPeter Xu * lock. 89970e806e4SPeter Xu */ 90070e806e4SPeter Xu static inline int 901c78f4636SPeter Xu copy_present_page(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 902c78f4636SPeter Xu pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, 903edf50470SMatthew Wilcox (Oracle) struct folio **prealloc, struct page *page) 90470e806e4SPeter Xu { 905edf50470SMatthew Wilcox (Oracle) struct folio *new_folio; 906b51ad4f8SDavid Hildenbrand pte_t pte; 90770e806e4SPeter Xu 908edf50470SMatthew Wilcox (Oracle) new_folio = *prealloc; 909edf50470SMatthew Wilcox (Oracle) if (!new_folio) 91070e806e4SPeter Xu return -EAGAIN; 91170e806e4SPeter Xu 91270e806e4SPeter Xu /* 91370e806e4SPeter Xu * We have a prealloc page, all good! Take it 91470e806e4SPeter Xu * over and copy the page & arm it. 91570e806e4SPeter Xu */ 91670e806e4SPeter Xu *prealloc = NULL; 917edf50470SMatthew Wilcox (Oracle) copy_user_highpage(&new_folio->page, page, addr, src_vma); 918edf50470SMatthew Wilcox (Oracle) __folio_mark_uptodate(new_folio); 919edf50470SMatthew Wilcox (Oracle) folio_add_new_anon_rmap(new_folio, dst_vma, addr); 920edf50470SMatthew Wilcox (Oracle) folio_add_lru_vma(new_folio, dst_vma); 921edf50470SMatthew Wilcox (Oracle) rss[MM_ANONPAGES]++; 92270e806e4SPeter Xu 92370e806e4SPeter Xu /* All done, just insert the new page copy in the child */ 924edf50470SMatthew Wilcox (Oracle) pte = mk_pte(&new_folio->page, dst_vma->vm_page_prot); 925c78f4636SPeter Xu pte = maybe_mkwrite(pte_mkdirty(pte), dst_vma); 926c33c7948SRyan Roberts if (userfaultfd_pte_wp(dst_vma, ptep_get(src_pte))) 9278f34f1eaSPeter Xu /* Uffd-wp needs to be delivered to dest pte as well */ 928f1eb1bacSPeter Xu pte = pte_mkuffd_wp(pte); 929c78f4636SPeter Xu set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); 93070e806e4SPeter Xu return 0; 93170e806e4SPeter Xu } 93270e806e4SPeter Xu 933f8d93776SDavid Hildenbrand static __always_inline void __copy_present_ptes(struct vm_area_struct *dst_vma, 93423ed1908SDavid Hildenbrand struct vm_area_struct *src_vma, pte_t *dst_pte, pte_t *src_pte, 935f8d93776SDavid Hildenbrand pte_t pte, unsigned long addr, int nr) 93623ed1908SDavid Hildenbrand { 93723ed1908SDavid Hildenbrand struct mm_struct *src_mm = src_vma->vm_mm; 93823ed1908SDavid Hildenbrand 93923ed1908SDavid Hildenbrand /* If it's a COW mapping, write protect it both processes. */ 94023ed1908SDavid Hildenbrand if (is_cow_mapping(src_vma->vm_flags) && pte_write(pte)) { 941f8d93776SDavid Hildenbrand wrprotect_ptes(src_mm, addr, src_pte, nr); 94223ed1908SDavid Hildenbrand pte = pte_wrprotect(pte); 94323ed1908SDavid Hildenbrand } 94423ed1908SDavid Hildenbrand 94523ed1908SDavid Hildenbrand /* If it's a shared mapping, mark it clean in the child. */ 94623ed1908SDavid Hildenbrand if (src_vma->vm_flags & VM_SHARED) 94723ed1908SDavid Hildenbrand pte = pte_mkclean(pte); 94823ed1908SDavid Hildenbrand pte = pte_mkold(pte); 94923ed1908SDavid Hildenbrand 95023ed1908SDavid Hildenbrand if (!userfaultfd_wp(dst_vma)) 95123ed1908SDavid Hildenbrand pte = pte_clear_uffd_wp(pte); 95223ed1908SDavid Hildenbrand 953f8d93776SDavid Hildenbrand set_ptes(dst_vma->vm_mm, addr, dst_pte, pte, nr); 95423ed1908SDavid Hildenbrand } 95523ed1908SDavid Hildenbrand 956f8d93776SDavid Hildenbrand /* 957f8d93776SDavid Hildenbrand * Copy one present PTE, trying to batch-process subsequent PTEs that map 958f8d93776SDavid Hildenbrand * consecutive pages of the same folio by copying them as well. 959f8d93776SDavid Hildenbrand * 960f8d93776SDavid Hildenbrand * Returns -EAGAIN if one preallocated page is required to copy the next PTE. 961f8d93776SDavid Hildenbrand * Otherwise, returns the number of copied PTEs (at least 1). 96270e806e4SPeter Xu */ 96370e806e4SPeter Xu static inline int 964f8d93776SDavid Hildenbrand copy_present_ptes(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 96553723298SDavid Hildenbrand pte_t *dst_pte, pte_t *src_pte, pte_t pte, unsigned long addr, 966f8d93776SDavid Hildenbrand int max_nr, int *rss, struct folio **prealloc) 967df3a57d1SLinus Torvalds { 968df3a57d1SLinus Torvalds struct page *page; 96914ddee41SMatthew Wilcox (Oracle) struct folio *folio; 970d7c0e5f7SDavid Hildenbrand bool any_writable; 97125365e10SDavid Hildenbrand fpb_t flags = 0; 972f8d93776SDavid Hildenbrand int err, nr; 973df3a57d1SLinus Torvalds 974c78f4636SPeter Xu page = vm_normal_page(src_vma, addr, pte); 97523ed1908SDavid Hildenbrand if (unlikely(!page)) 97623ed1908SDavid Hildenbrand goto copy_pte; 97723ed1908SDavid Hildenbrand 97814ddee41SMatthew Wilcox (Oracle) folio = page_folio(page); 979f8d93776SDavid Hildenbrand 980f8d93776SDavid Hildenbrand /* 981f8d93776SDavid Hildenbrand * If we likely have to copy, just don't bother with batching. Make 982f8d93776SDavid Hildenbrand * sure that the common "small folio" case is as fast as possible 983f8d93776SDavid Hildenbrand * by keeping the batching logic separate. 984f8d93776SDavid Hildenbrand */ 985f8d93776SDavid Hildenbrand if (unlikely(!*prealloc && folio_test_large(folio) && max_nr != 1)) { 98625365e10SDavid Hildenbrand if (src_vma->vm_flags & VM_SHARED) 98725365e10SDavid Hildenbrand flags |= FPB_IGNORE_DIRTY; 98825365e10SDavid Hildenbrand if (!vma_soft_dirty_enabled(src_vma)) 98925365e10SDavid Hildenbrand flags |= FPB_IGNORE_SOFT_DIRTY; 99025365e10SDavid Hildenbrand 991d7c0e5f7SDavid Hildenbrand nr = folio_pte_batch(folio, addr, src_pte, pte, max_nr, flags, 992d7c0e5f7SDavid Hildenbrand &any_writable); 993f8d93776SDavid Hildenbrand folio_ref_add(folio, nr); 994f8d93776SDavid Hildenbrand if (folio_test_anon(folio)) { 995f8d93776SDavid Hildenbrand if (unlikely(folio_try_dup_anon_rmap_ptes(folio, page, 996f8d93776SDavid Hildenbrand nr, src_vma))) { 997f8d93776SDavid Hildenbrand folio_ref_sub(folio, nr); 998f8d93776SDavid Hildenbrand return -EAGAIN; 999f8d93776SDavid Hildenbrand } 1000f8d93776SDavid Hildenbrand rss[MM_ANONPAGES] += nr; 1001f8d93776SDavid Hildenbrand VM_WARN_ON_FOLIO(PageAnonExclusive(page), folio); 1002f8d93776SDavid Hildenbrand } else { 1003f8d93776SDavid Hildenbrand folio_dup_file_rmap_ptes(folio, page, nr); 1004f8d93776SDavid Hildenbrand rss[mm_counter_file(folio)] += nr; 1005f8d93776SDavid Hildenbrand } 1006d7c0e5f7SDavid Hildenbrand if (any_writable) 1007d7c0e5f7SDavid Hildenbrand pte = pte_mkwrite(pte, src_vma); 1008f8d93776SDavid Hildenbrand __copy_present_ptes(dst_vma, src_vma, dst_pte, src_pte, pte, 1009f8d93776SDavid Hildenbrand addr, nr); 1010f8d93776SDavid Hildenbrand return nr; 1011f8d93776SDavid Hildenbrand } 1012f8d93776SDavid Hildenbrand 101323ed1908SDavid Hildenbrand folio_get(folio); 101423ed1908SDavid Hildenbrand if (folio_test_anon(folio)) { 1015b51ad4f8SDavid Hildenbrand /* 1016b51ad4f8SDavid Hildenbrand * If this page may have been pinned by the parent process, 1017b51ad4f8SDavid Hildenbrand * copy the page immediately for the child so that we'll always 1018b51ad4f8SDavid Hildenbrand * guarantee the pinned page won't be randomly replaced in the 1019b51ad4f8SDavid Hildenbrand * future. 1020b51ad4f8SDavid Hildenbrand */ 102108e7795eSDavid Hildenbrand if (unlikely(folio_try_dup_anon_rmap_pte(folio, page, src_vma))) { 1022fb3d824dSDavid Hildenbrand /* Page may be pinned, we have to copy. */ 102314ddee41SMatthew Wilcox (Oracle) folio_put(folio); 1024f8d93776SDavid Hildenbrand err = copy_present_page(dst_vma, src_vma, dst_pte, src_pte, 1025b51ad4f8SDavid Hildenbrand addr, rss, prealloc, page); 1026f8d93776SDavid Hildenbrand return err ? err : 1; 1027fb3d824dSDavid Hildenbrand } 1028edf50470SMatthew Wilcox (Oracle) rss[MM_ANONPAGES]++; 102923ed1908SDavid Hildenbrand VM_WARN_ON_FOLIO(PageAnonExclusive(page), folio); 103023ed1908SDavid Hildenbrand } else { 1031d8ef5e31SDavid Hildenbrand folio_dup_file_rmap_pte(folio, page); 10326b27cc6cSKefeng Wang rss[mm_counter_file(folio)]++; 103370e806e4SPeter Xu } 103470e806e4SPeter Xu 103523ed1908SDavid Hildenbrand copy_pte: 1036f8d93776SDavid Hildenbrand __copy_present_ptes(dst_vma, src_vma, dst_pte, src_pte, pte, addr, 1); 1037f8d93776SDavid Hildenbrand return 1; 10386aab341eSLinus Torvalds } 1039ae859762SHugh Dickins 1040294de6d8SKefeng Wang static inline struct folio *folio_prealloc(struct mm_struct *src_mm, 1041294de6d8SKefeng Wang struct vm_area_struct *vma, unsigned long addr, bool need_zero) 104270e806e4SPeter Xu { 1043edf50470SMatthew Wilcox (Oracle) struct folio *new_folio; 104470e806e4SPeter Xu 1045294de6d8SKefeng Wang if (need_zero) 1046294de6d8SKefeng Wang new_folio = vma_alloc_zeroed_movable_folio(vma, addr); 1047294de6d8SKefeng Wang else 1048294de6d8SKefeng Wang new_folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma, 1049294de6d8SKefeng Wang addr, false); 1050294de6d8SKefeng Wang 1051edf50470SMatthew Wilcox (Oracle) if (!new_folio) 105270e806e4SPeter Xu return NULL; 105370e806e4SPeter Xu 1054edf50470SMatthew Wilcox (Oracle) if (mem_cgroup_charge(new_folio, src_mm, GFP_KERNEL)) { 1055edf50470SMatthew Wilcox (Oracle) folio_put(new_folio); 105670e806e4SPeter Xu return NULL; 105770e806e4SPeter Xu } 1058e601ded4SKefeng Wang folio_throttle_swaprate(new_folio, GFP_KERNEL); 105970e806e4SPeter Xu 1060edf50470SMatthew Wilcox (Oracle) return new_folio; 10611da177e4SLinus Torvalds } 10621da177e4SLinus Torvalds 1063c78f4636SPeter Xu static int 1064c78f4636SPeter Xu copy_pte_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1065c78f4636SPeter Xu pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr, 1066c78f4636SPeter Xu unsigned long end) 10671da177e4SLinus Torvalds { 1068c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1069c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 1070c36987e2SDaisuke Nishimura pte_t *orig_src_pte, *orig_dst_pte; 10711da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 1072c33c7948SRyan Roberts pte_t ptent; 1073c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 1074f8d93776SDavid Hildenbrand int progress, max_nr, ret = 0; 1075d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 1076570a335bSHugh Dickins swp_entry_t entry = (swp_entry_t){0}; 1077edf50470SMatthew Wilcox (Oracle) struct folio *prealloc = NULL; 1078f8d93776SDavid Hildenbrand int nr; 10791da177e4SLinus Torvalds 10801da177e4SLinus Torvalds again: 108170e806e4SPeter Xu progress = 0; 1082d559db08SKAMEZAWA Hiroyuki init_rss_vec(rss); 1083d559db08SKAMEZAWA Hiroyuki 10843db82b93SHugh Dickins /* 10853db82b93SHugh Dickins * copy_pmd_range()'s prior pmd_none_or_clear_bad(src_pmd), and the 10863db82b93SHugh Dickins * error handling here, assume that exclusive mmap_lock on dst and src 10873db82b93SHugh Dickins * protects anon from unexpected THP transitions; with shmem and file 10883db82b93SHugh Dickins * protected by mmap_lock-less collapse skipping areas with anon_vma 10893db82b93SHugh Dickins * (whereas vma_needs_copy() skips areas without anon_vma). A rework 10903db82b93SHugh Dickins * can remove such assumptions later, but this is good enough for now. 10913db82b93SHugh Dickins */ 1092c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 109370e806e4SPeter Xu if (!dst_pte) { 109470e806e4SPeter Xu ret = -ENOMEM; 109570e806e4SPeter Xu goto out; 109670e806e4SPeter Xu } 10973db82b93SHugh Dickins src_pte = pte_offset_map_nolock(src_mm, src_pmd, addr, &src_ptl); 10983db82b93SHugh Dickins if (!src_pte) { 10993db82b93SHugh Dickins pte_unmap_unlock(dst_pte, dst_ptl); 11003db82b93SHugh Dickins /* ret == 0 */ 11013db82b93SHugh Dickins goto out; 11023db82b93SHugh Dickins } 1103f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 1104c36987e2SDaisuke Nishimura orig_src_pte = src_pte; 1105c36987e2SDaisuke Nishimura orig_dst_pte = dst_pte; 11066606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 11071da177e4SLinus Torvalds 11081da177e4SLinus Torvalds do { 1109f8d93776SDavid Hildenbrand nr = 1; 1110f8d93776SDavid Hildenbrand 11111da177e4SLinus Torvalds /* 11121da177e4SLinus Torvalds * We are holding two locks at this point - either of them 11131da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 11141da177e4SLinus Torvalds */ 1115e040f218SHugh Dickins if (progress >= 32) { 1116e040f218SHugh Dickins progress = 0; 1117e040f218SHugh Dickins if (need_resched() || 111895c354feSNick Piggin spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) 11191da177e4SLinus Torvalds break; 1120e040f218SHugh Dickins } 1121c33c7948SRyan Roberts ptent = ptep_get(src_pte); 1122c33c7948SRyan Roberts if (pte_none(ptent)) { 11231da177e4SLinus Torvalds progress++; 11241da177e4SLinus Torvalds continue; 11251da177e4SLinus Torvalds } 1126c33c7948SRyan Roberts if (unlikely(!pte_present(ptent))) { 11279a5cc85cSAlistair Popple ret = copy_nonpresent_pte(dst_mm, src_mm, 112879a1971cSLinus Torvalds dst_pte, src_pte, 11298f34f1eaSPeter Xu dst_vma, src_vma, 11308f34f1eaSPeter Xu addr, rss); 11319a5cc85cSAlistair Popple if (ret == -EIO) { 1132c33c7948SRyan Roberts entry = pte_to_swp_entry(ptep_get(src_pte)); 1133570a335bSHugh Dickins break; 1134b756a3b5SAlistair Popple } else if (ret == -EBUSY) { 1135b756a3b5SAlistair Popple break; 1136b756a3b5SAlistair Popple } else if (!ret) { 11371da177e4SLinus Torvalds progress += 8; 113879a1971cSLinus Torvalds continue; 113979a1971cSLinus Torvalds } 114053723298SDavid Hildenbrand ptent = ptep_get(src_pte); 114153723298SDavid Hildenbrand VM_WARN_ON_ONCE(!pte_present(ptent)); 1142b756a3b5SAlistair Popple 1143b756a3b5SAlistair Popple /* 1144b756a3b5SAlistair Popple * Device exclusive entry restored, continue by copying 1145b756a3b5SAlistair Popple * the now present pte. 1146b756a3b5SAlistair Popple */ 1147b756a3b5SAlistair Popple WARN_ON_ONCE(ret != -ENOENT); 1148b756a3b5SAlistair Popple } 1149f8d93776SDavid Hildenbrand /* copy_present_ptes() will clear `*prealloc' if consumed */ 1150f8d93776SDavid Hildenbrand max_nr = (end - addr) / PAGE_SIZE; 1151f8d93776SDavid Hildenbrand ret = copy_present_ptes(dst_vma, src_vma, dst_pte, src_pte, 1152f8d93776SDavid Hildenbrand ptent, addr, max_nr, rss, &prealloc); 115370e806e4SPeter Xu /* 115470e806e4SPeter Xu * If we need a pre-allocated page for this pte, drop the 115570e806e4SPeter Xu * locks, allocate, and try again. 115670e806e4SPeter Xu */ 115770e806e4SPeter Xu if (unlikely(ret == -EAGAIN)) 115870e806e4SPeter Xu break; 115970e806e4SPeter Xu if (unlikely(prealloc)) { 116070e806e4SPeter Xu /* 116170e806e4SPeter Xu * pre-alloc page cannot be reused by next time so as 116270e806e4SPeter Xu * to strictly follow mempolicy (e.g., alloc_page_vma() 116370e806e4SPeter Xu * will allocate page according to address). This 116470e806e4SPeter Xu * could only happen if one pinned pte changed. 116570e806e4SPeter Xu */ 1166edf50470SMatthew Wilcox (Oracle) folio_put(prealloc); 116770e806e4SPeter Xu prealloc = NULL; 116870e806e4SPeter Xu } 1169f8d93776SDavid Hildenbrand nr = ret; 1170f8d93776SDavid Hildenbrand progress += 8 * nr; 1171f8d93776SDavid Hildenbrand } while (dst_pte += nr, src_pte += nr, addr += PAGE_SIZE * nr, 1172f8d93776SDavid Hildenbrand addr != end); 11731da177e4SLinus Torvalds 11746606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 11753db82b93SHugh Dickins pte_unmap_unlock(orig_src_pte, src_ptl); 1176d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(dst_mm, rss); 1177c36987e2SDaisuke Nishimura pte_unmap_unlock(orig_dst_pte, dst_ptl); 1178c74df32cSHugh Dickins cond_resched(); 1179570a335bSHugh Dickins 11809a5cc85cSAlistair Popple if (ret == -EIO) { 11819a5cc85cSAlistair Popple VM_WARN_ON_ONCE(!entry.val); 118270e806e4SPeter Xu if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) { 118370e806e4SPeter Xu ret = -ENOMEM; 118470e806e4SPeter Xu goto out; 118570e806e4SPeter Xu } 118670e806e4SPeter Xu entry.val = 0; 1187b756a3b5SAlistair Popple } else if (ret == -EBUSY) { 1188b756a3b5SAlistair Popple goto out; 11899a5cc85cSAlistair Popple } else if (ret == -EAGAIN) { 1190294de6d8SKefeng Wang prealloc = folio_prealloc(src_mm, src_vma, addr, false); 119170e806e4SPeter Xu if (!prealloc) 1192570a335bSHugh Dickins return -ENOMEM; 1193f8d93776SDavid Hildenbrand } else if (ret < 0) { 11949a5cc85cSAlistair Popple VM_WARN_ON_ONCE(1); 11959a5cc85cSAlistair Popple } 11969a5cc85cSAlistair Popple 119770e806e4SPeter Xu /* We've captured and resolved the error. Reset, try again. */ 119870e806e4SPeter Xu ret = 0; 11999a5cc85cSAlistair Popple 12001da177e4SLinus Torvalds if (addr != end) 12011da177e4SLinus Torvalds goto again; 120270e806e4SPeter Xu out: 120370e806e4SPeter Xu if (unlikely(prealloc)) 1204edf50470SMatthew Wilcox (Oracle) folio_put(prealloc); 120570e806e4SPeter Xu return ret; 12061da177e4SLinus Torvalds } 12071da177e4SLinus Torvalds 1208c78f4636SPeter Xu static inline int 1209c78f4636SPeter Xu copy_pmd_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1210c78f4636SPeter Xu pud_t *dst_pud, pud_t *src_pud, unsigned long addr, 1211c78f4636SPeter Xu unsigned long end) 12121da177e4SLinus Torvalds { 1213c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1214c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 12151da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 12161da177e4SLinus Torvalds unsigned long next; 12171da177e4SLinus Torvalds 12181da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 12191da177e4SLinus Torvalds if (!dst_pmd) 12201da177e4SLinus Torvalds return -ENOMEM; 12211da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 12221da177e4SLinus Torvalds do { 12231da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 122484c3fc4eSZi Yan if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) 122584c3fc4eSZi Yan || pmd_devmap(*src_pmd)) { 122671e3aac0SAndrea Arcangeli int err; 1227c78f4636SPeter Xu VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, src_vma); 12288f34f1eaSPeter Xu err = copy_huge_pmd(dst_mm, src_mm, dst_pmd, src_pmd, 12298f34f1eaSPeter Xu addr, dst_vma, src_vma); 123071e3aac0SAndrea Arcangeli if (err == -ENOMEM) 123171e3aac0SAndrea Arcangeli return -ENOMEM; 123271e3aac0SAndrea Arcangeli if (!err) 123371e3aac0SAndrea Arcangeli continue; 123471e3aac0SAndrea Arcangeli /* fall through */ 123571e3aac0SAndrea Arcangeli } 12361da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 12371da177e4SLinus Torvalds continue; 1238c78f4636SPeter Xu if (copy_pte_range(dst_vma, src_vma, dst_pmd, src_pmd, 1239c78f4636SPeter Xu addr, next)) 12401da177e4SLinus Torvalds return -ENOMEM; 12411da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 12421da177e4SLinus Torvalds return 0; 12431da177e4SLinus Torvalds } 12441da177e4SLinus Torvalds 1245c78f4636SPeter Xu static inline int 1246c78f4636SPeter Xu copy_pud_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1247c78f4636SPeter Xu p4d_t *dst_p4d, p4d_t *src_p4d, unsigned long addr, 1248c78f4636SPeter Xu unsigned long end) 12491da177e4SLinus Torvalds { 1250c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1251c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 12521da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 12531da177e4SLinus Torvalds unsigned long next; 12541da177e4SLinus Torvalds 1255c2febafcSKirill A. Shutemov dst_pud = pud_alloc(dst_mm, dst_p4d, addr); 12561da177e4SLinus Torvalds if (!dst_pud) 12571da177e4SLinus Torvalds return -ENOMEM; 1258c2febafcSKirill A. Shutemov src_pud = pud_offset(src_p4d, addr); 12591da177e4SLinus Torvalds do { 12601da177e4SLinus Torvalds next = pud_addr_end(addr, end); 1261a00cc7d9SMatthew Wilcox if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { 1262a00cc7d9SMatthew Wilcox int err; 1263a00cc7d9SMatthew Wilcox 1264c78f4636SPeter Xu VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, src_vma); 1265a00cc7d9SMatthew Wilcox err = copy_huge_pud(dst_mm, src_mm, 1266c78f4636SPeter Xu dst_pud, src_pud, addr, src_vma); 1267a00cc7d9SMatthew Wilcox if (err == -ENOMEM) 1268a00cc7d9SMatthew Wilcox return -ENOMEM; 1269a00cc7d9SMatthew Wilcox if (!err) 1270a00cc7d9SMatthew Wilcox continue; 1271a00cc7d9SMatthew Wilcox /* fall through */ 1272a00cc7d9SMatthew Wilcox } 12731da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 12741da177e4SLinus Torvalds continue; 1275c78f4636SPeter Xu if (copy_pmd_range(dst_vma, src_vma, dst_pud, src_pud, 1276c78f4636SPeter Xu addr, next)) 12771da177e4SLinus Torvalds return -ENOMEM; 12781da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 12791da177e4SLinus Torvalds return 0; 12801da177e4SLinus Torvalds } 12811da177e4SLinus Torvalds 1282c78f4636SPeter Xu static inline int 1283c78f4636SPeter Xu copy_p4d_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1284c78f4636SPeter Xu pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long addr, 1285c78f4636SPeter Xu unsigned long end) 1286c2febafcSKirill A. Shutemov { 1287c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1288c2febafcSKirill A. Shutemov p4d_t *src_p4d, *dst_p4d; 1289c2febafcSKirill A. Shutemov unsigned long next; 1290c2febafcSKirill A. Shutemov 1291c2febafcSKirill A. Shutemov dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); 1292c2febafcSKirill A. Shutemov if (!dst_p4d) 1293c2febafcSKirill A. Shutemov return -ENOMEM; 1294c2febafcSKirill A. Shutemov src_p4d = p4d_offset(src_pgd, addr); 1295c2febafcSKirill A. Shutemov do { 1296c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 1297c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(src_p4d)) 1298c2febafcSKirill A. Shutemov continue; 1299c78f4636SPeter Xu if (copy_pud_range(dst_vma, src_vma, dst_p4d, src_p4d, 1300c78f4636SPeter Xu addr, next)) 1301c2febafcSKirill A. Shutemov return -ENOMEM; 1302c2febafcSKirill A. Shutemov } while (dst_p4d++, src_p4d++, addr = next, addr != end); 1303c2febafcSKirill A. Shutemov return 0; 1304c2febafcSKirill A. Shutemov } 1305c2febafcSKirill A. Shutemov 1306c56d1b62SPeter Xu /* 1307c56d1b62SPeter Xu * Return true if the vma needs to copy the pgtable during this fork(). Return 1308c56d1b62SPeter Xu * false when we can speed up fork() by allowing lazy page faults later until 1309c56d1b62SPeter Xu * when the child accesses the memory range. 1310c56d1b62SPeter Xu */ 1311bc70fbf2SPeter Xu static bool 1312c56d1b62SPeter Xu vma_needs_copy(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) 1313c56d1b62SPeter Xu { 1314c56d1b62SPeter Xu /* 1315c56d1b62SPeter Xu * Always copy pgtables when dst_vma has uffd-wp enabled even if it's 1316c56d1b62SPeter Xu * file-backed (e.g. shmem). Because when uffd-wp is enabled, pgtable 1317c56d1b62SPeter Xu * contains uffd-wp protection information, that's something we can't 1318c56d1b62SPeter Xu * retrieve from page cache, and skip copying will lose those info. 1319c56d1b62SPeter Xu */ 1320c56d1b62SPeter Xu if (userfaultfd_wp(dst_vma)) 1321c56d1b62SPeter Xu return true; 1322c56d1b62SPeter Xu 1323bcd51a3cSMike Kravetz if (src_vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 1324c56d1b62SPeter Xu return true; 1325c56d1b62SPeter Xu 1326c56d1b62SPeter Xu if (src_vma->anon_vma) 1327c56d1b62SPeter Xu return true; 1328c56d1b62SPeter Xu 1329c56d1b62SPeter Xu /* 1330c56d1b62SPeter Xu * Don't copy ptes where a page fault will fill them correctly. Fork 1331c56d1b62SPeter Xu * becomes much lighter when there are big shared or private readonly 1332c56d1b62SPeter Xu * mappings. The tradeoff is that copy_page_range is more efficient 1333c56d1b62SPeter Xu * than faulting. 1334c56d1b62SPeter Xu */ 1335c56d1b62SPeter Xu return false; 1336c56d1b62SPeter Xu } 1337c56d1b62SPeter Xu 1338c78f4636SPeter Xu int 1339c78f4636SPeter Xu copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) 13401da177e4SLinus Torvalds { 13411da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 13421da177e4SLinus Torvalds unsigned long next; 1343c78f4636SPeter Xu unsigned long addr = src_vma->vm_start; 1344c78f4636SPeter Xu unsigned long end = src_vma->vm_end; 1345c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1346c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 1347ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 13482ec74c3eSSagi Grimberg bool is_cow; 1349cddb8a5cSAndrea Arcangeli int ret; 13501da177e4SLinus Torvalds 1351c56d1b62SPeter Xu if (!vma_needs_copy(dst_vma, src_vma)) 1352d992895bSNick Piggin return 0; 1353d992895bSNick Piggin 1354c78f4636SPeter Xu if (is_vm_hugetlb_page(src_vma)) 1355bc70fbf2SPeter Xu return copy_hugetlb_page_range(dst_mm, src_mm, dst_vma, src_vma); 13561da177e4SLinus Torvalds 1357c78f4636SPeter Xu if (unlikely(src_vma->vm_flags & VM_PFNMAP)) { 13582ab64037Svenkatesh.pallipadi@intel.com /* 13592ab64037Svenkatesh.pallipadi@intel.com * We do not free on error cases below as remove_vma 13602ab64037Svenkatesh.pallipadi@intel.com * gets called on error from higher level routine 13612ab64037Svenkatesh.pallipadi@intel.com */ 1362c78f4636SPeter Xu ret = track_pfn_copy(src_vma); 13632ab64037Svenkatesh.pallipadi@intel.com if (ret) 13642ab64037Svenkatesh.pallipadi@intel.com return ret; 13652ab64037Svenkatesh.pallipadi@intel.com } 13662ab64037Svenkatesh.pallipadi@intel.com 1367cddb8a5cSAndrea Arcangeli /* 1368cddb8a5cSAndrea Arcangeli * We need to invalidate the secondary MMU mappings only when 1369cddb8a5cSAndrea Arcangeli * there could be a permission downgrade on the ptes of the 1370cddb8a5cSAndrea Arcangeli * parent mm. And a permission downgrade will only happen if 1371cddb8a5cSAndrea Arcangeli * is_cow_mapping() returns true. 1372cddb8a5cSAndrea Arcangeli */ 1373c78f4636SPeter Xu is_cow = is_cow_mapping(src_vma->vm_flags); 1374ac46d4f3SJérôme Glisse 1375ac46d4f3SJérôme Glisse if (is_cow) { 13767269f999SJérôme Glisse mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_PAGE, 13777d4a8be0SAlistair Popple 0, src_mm, addr, end); 1378ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 137957efa1feSJason Gunthorpe /* 138057efa1feSJason Gunthorpe * Disabling preemption is not needed for the write side, as 138157efa1feSJason Gunthorpe * the read side doesn't spin, but goes to the mmap_lock. 138257efa1feSJason Gunthorpe * 138357efa1feSJason Gunthorpe * Use the raw variant of the seqcount_t write API to avoid 138457efa1feSJason Gunthorpe * lockdep complaining about preemptibility. 138557efa1feSJason Gunthorpe */ 1386e727bfd5SSuren Baghdasaryan vma_assert_write_locked(src_vma); 138757efa1feSJason Gunthorpe raw_write_seqcount_begin(&src_mm->write_protect_seq); 1388ac46d4f3SJérôme Glisse } 1389cddb8a5cSAndrea Arcangeli 1390cddb8a5cSAndrea Arcangeli ret = 0; 13911da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 13921da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 13931da177e4SLinus Torvalds do { 13941da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 13951da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 13961da177e4SLinus Torvalds continue; 1397c78f4636SPeter Xu if (unlikely(copy_p4d_range(dst_vma, src_vma, dst_pgd, src_pgd, 1398c78f4636SPeter Xu addr, next))) { 1399d155df53SMa Wupeng untrack_pfn_clear(dst_vma); 1400cddb8a5cSAndrea Arcangeli ret = -ENOMEM; 1401cddb8a5cSAndrea Arcangeli break; 1402cddb8a5cSAndrea Arcangeli } 14031da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 1404cddb8a5cSAndrea Arcangeli 140557efa1feSJason Gunthorpe if (is_cow) { 140657efa1feSJason Gunthorpe raw_write_seqcount_end(&src_mm->write_protect_seq); 1407ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 140857efa1feSJason Gunthorpe } 1409cddb8a5cSAndrea Arcangeli return ret; 14101da177e4SLinus Torvalds } 14111da177e4SLinus Torvalds 14125abfd71dSPeter Xu /* Whether we should zap all COWed (private) pages too */ 14135abfd71dSPeter Xu static inline bool should_zap_cows(struct zap_details *details) 14145abfd71dSPeter Xu { 14155abfd71dSPeter Xu /* By default, zap all pages */ 14165abfd71dSPeter Xu if (!details) 14175abfd71dSPeter Xu return true; 14185abfd71dSPeter Xu 14195abfd71dSPeter Xu /* Or, we zap COWed pages only if the caller wants to */ 14202e148f1eSPeter Xu return details->even_cows; 14215abfd71dSPeter Xu } 14225abfd71dSPeter Xu 1423eabafaaaSKefeng Wang /* Decides whether we should zap this folio with the folio pointer specified */ 1424eabafaaaSKefeng Wang static inline bool should_zap_folio(struct zap_details *details, 1425eabafaaaSKefeng Wang struct folio *folio) 14263506659eSMatthew Wilcox (Oracle) { 1427eabafaaaSKefeng Wang /* If we can make a decision without *folio.. */ 14285abfd71dSPeter Xu if (should_zap_cows(details)) 1429254ab940SPeter Xu return true; 14303506659eSMatthew Wilcox (Oracle) 1431eabafaaaSKefeng Wang /* Otherwise we should only zap non-anon folios */ 1432eabafaaaSKefeng Wang return !folio_test_anon(folio); 14333506659eSMatthew Wilcox (Oracle) } 14343506659eSMatthew Wilcox (Oracle) 1435999dad82SPeter Xu static inline bool zap_drop_file_uffd_wp(struct zap_details *details) 1436999dad82SPeter Xu { 1437999dad82SPeter Xu if (!details) 1438999dad82SPeter Xu return false; 1439999dad82SPeter Xu 1440999dad82SPeter Xu return details->zap_flags & ZAP_FLAG_DROP_MARKER; 1441999dad82SPeter Xu } 1442999dad82SPeter Xu 1443999dad82SPeter Xu /* 1444999dad82SPeter Xu * This function makes sure that we'll replace the none pte with an uffd-wp 1445999dad82SPeter Xu * swap special pte marker when necessary. Must be with the pgtable lock held. 1446999dad82SPeter Xu */ 1447999dad82SPeter Xu static inline void 1448999dad82SPeter Xu zap_install_uffd_wp_if_needed(struct vm_area_struct *vma, 144910ebac4fSDavid Hildenbrand unsigned long addr, pte_t *pte, int nr, 1450999dad82SPeter Xu struct zap_details *details, pte_t pteval) 1451999dad82SPeter Xu { 14522bad466cSPeter Xu /* Zap on anonymous always means dropping everything */ 14532bad466cSPeter Xu if (vma_is_anonymous(vma)) 14542bad466cSPeter Xu return; 14552bad466cSPeter Xu 1456999dad82SPeter Xu if (zap_drop_file_uffd_wp(details)) 1457999dad82SPeter Xu return; 1458999dad82SPeter Xu 145910ebac4fSDavid Hildenbrand for (;;) { 146010ebac4fSDavid Hildenbrand /* the PFN in the PTE is irrelevant. */ 1461999dad82SPeter Xu pte_install_uffd_wp_if_needed(vma, addr, pte, pteval); 146210ebac4fSDavid Hildenbrand if (--nr == 0) 146310ebac4fSDavid Hildenbrand break; 146410ebac4fSDavid Hildenbrand pte++; 146510ebac4fSDavid Hildenbrand addr += PAGE_SIZE; 146610ebac4fSDavid Hildenbrand } 1467999dad82SPeter Xu } 1468999dad82SPeter Xu 146910ebac4fSDavid Hildenbrand static __always_inline void zap_present_folio_ptes(struct mmu_gather *tlb, 14702b42a7e5SDavid Hildenbrand struct vm_area_struct *vma, struct folio *folio, 147110ebac4fSDavid Hildenbrand struct page *page, pte_t *pte, pte_t ptent, unsigned int nr, 147210ebac4fSDavid Hildenbrand unsigned long addr, struct zap_details *details, int *rss, 147310ebac4fSDavid Hildenbrand bool *force_flush, bool *force_break) 1474789753e1SDavid Hildenbrand { 1475789753e1SDavid Hildenbrand struct mm_struct *mm = tlb->mm; 1476789753e1SDavid Hildenbrand bool delay_rmap = false; 1477789753e1SDavid Hildenbrand 1478789753e1SDavid Hildenbrand if (!folio_test_anon(folio)) { 147910ebac4fSDavid Hildenbrand ptent = get_and_clear_full_ptes(mm, addr, pte, nr, tlb->fullmm); 1480789753e1SDavid Hildenbrand if (pte_dirty(ptent)) { 1481789753e1SDavid Hildenbrand folio_mark_dirty(folio); 1482789753e1SDavid Hildenbrand if (tlb_delay_rmap(tlb)) { 1483789753e1SDavid Hildenbrand delay_rmap = true; 1484789753e1SDavid Hildenbrand *force_flush = true; 1485789753e1SDavid Hildenbrand } 1486789753e1SDavid Hildenbrand } 1487789753e1SDavid Hildenbrand if (pte_young(ptent) && likely(vma_has_recency(vma))) 1488789753e1SDavid Hildenbrand folio_mark_accessed(folio); 148910ebac4fSDavid Hildenbrand rss[mm_counter(folio)] -= nr; 1490d11838edSDavid Hildenbrand } else { 1491d11838edSDavid Hildenbrand /* We don't need up-to-date accessed/dirty bits. */ 149210ebac4fSDavid Hildenbrand clear_full_ptes(mm, addr, pte, nr, tlb->fullmm); 149310ebac4fSDavid Hildenbrand rss[MM_ANONPAGES] -= nr; 1494d11838edSDavid Hildenbrand } 149510ebac4fSDavid Hildenbrand /* Checking a single PTE in a batch is sufficient. */ 1496d11838edSDavid Hildenbrand arch_check_zapped_pte(vma, ptent); 149710ebac4fSDavid Hildenbrand tlb_remove_tlb_entries(tlb, pte, nr, addr); 1498d11838edSDavid Hildenbrand if (unlikely(userfaultfd_pte_wp(vma, ptent))) 149910ebac4fSDavid Hildenbrand zap_install_uffd_wp_if_needed(vma, addr, pte, nr, details, 150010ebac4fSDavid Hildenbrand ptent); 1501d11838edSDavid Hildenbrand 1502789753e1SDavid Hildenbrand if (!delay_rmap) { 150310ebac4fSDavid Hildenbrand folio_remove_rmap_ptes(folio, page, nr, vma); 150410ebac4fSDavid Hildenbrand 150510ebac4fSDavid Hildenbrand /* Only sanity-check the first page in a batch. */ 1506789753e1SDavid Hildenbrand if (unlikely(page_mapcount(page) < 0)) 1507789753e1SDavid Hildenbrand print_bad_pte(vma, addr, ptent, page); 1508789753e1SDavid Hildenbrand } 150910ebac4fSDavid Hildenbrand if (unlikely(__tlb_remove_folio_pages(tlb, page, nr, delay_rmap))) { 1510789753e1SDavid Hildenbrand *force_flush = true; 1511789753e1SDavid Hildenbrand *force_break = true; 1512789753e1SDavid Hildenbrand } 1513789753e1SDavid Hildenbrand } 1514789753e1SDavid Hildenbrand 151510ebac4fSDavid Hildenbrand /* 151610ebac4fSDavid Hildenbrand * Zap or skip at least one present PTE, trying to batch-process subsequent 151710ebac4fSDavid Hildenbrand * PTEs that map consecutive pages of the same folio. 151810ebac4fSDavid Hildenbrand * 151910ebac4fSDavid Hildenbrand * Returns the number of processed (skipped or zapped) PTEs (at least 1). 152010ebac4fSDavid Hildenbrand */ 152110ebac4fSDavid Hildenbrand static inline int zap_present_ptes(struct mmu_gather *tlb, 15222b42a7e5SDavid Hildenbrand struct vm_area_struct *vma, pte_t *pte, pte_t ptent, 152310ebac4fSDavid Hildenbrand unsigned int max_nr, unsigned long addr, 152410ebac4fSDavid Hildenbrand struct zap_details *details, int *rss, bool *force_flush, 152510ebac4fSDavid Hildenbrand bool *force_break) 15262b42a7e5SDavid Hildenbrand { 152710ebac4fSDavid Hildenbrand const fpb_t fpb_flags = FPB_IGNORE_DIRTY | FPB_IGNORE_SOFT_DIRTY; 15282b42a7e5SDavid Hildenbrand struct mm_struct *mm = tlb->mm; 15292b42a7e5SDavid Hildenbrand struct folio *folio; 15302b42a7e5SDavid Hildenbrand struct page *page; 153110ebac4fSDavid Hildenbrand int nr; 15322b42a7e5SDavid Hildenbrand 15332b42a7e5SDavid Hildenbrand page = vm_normal_page(vma, addr, ptent); 15342b42a7e5SDavid Hildenbrand if (!page) { 15352b42a7e5SDavid Hildenbrand /* We don't need up-to-date accessed/dirty bits. */ 15362b42a7e5SDavid Hildenbrand ptep_get_and_clear_full(mm, addr, pte, tlb->fullmm); 15372b42a7e5SDavid Hildenbrand arch_check_zapped_pte(vma, ptent); 15382b42a7e5SDavid Hildenbrand tlb_remove_tlb_entry(tlb, pte, addr); 1539f8572367SPeter Xu if (userfaultfd_pte_wp(vma, ptent)) 1540f8572367SPeter Xu zap_install_uffd_wp_if_needed(vma, addr, pte, 1, 1541f8572367SPeter Xu details, ptent); 15422b42a7e5SDavid Hildenbrand ksm_might_unmap_zero_page(mm, ptent); 154310ebac4fSDavid Hildenbrand return 1; 15442b42a7e5SDavid Hildenbrand } 15452b42a7e5SDavid Hildenbrand 15462b42a7e5SDavid Hildenbrand folio = page_folio(page); 15472b42a7e5SDavid Hildenbrand if (unlikely(!should_zap_folio(details, folio))) 154810ebac4fSDavid Hildenbrand return 1; 154910ebac4fSDavid Hildenbrand 155010ebac4fSDavid Hildenbrand /* 155110ebac4fSDavid Hildenbrand * Make sure that the common "small folio" case is as fast as possible 155210ebac4fSDavid Hildenbrand * by keeping the batching logic separate. 155310ebac4fSDavid Hildenbrand */ 155410ebac4fSDavid Hildenbrand if (unlikely(folio_test_large(folio) && max_nr != 1)) { 155510ebac4fSDavid Hildenbrand nr = folio_pte_batch(folio, addr, pte, ptent, max_nr, fpb_flags, 155610ebac4fSDavid Hildenbrand NULL); 155710ebac4fSDavid Hildenbrand 155810ebac4fSDavid Hildenbrand zap_present_folio_ptes(tlb, vma, folio, page, pte, ptent, nr, 155910ebac4fSDavid Hildenbrand addr, details, rss, force_flush, 156010ebac4fSDavid Hildenbrand force_break); 156110ebac4fSDavid Hildenbrand return nr; 156210ebac4fSDavid Hildenbrand } 156310ebac4fSDavid Hildenbrand zap_present_folio_ptes(tlb, vma, folio, page, pte, ptent, 1, addr, 156410ebac4fSDavid Hildenbrand details, rss, force_flush, force_break); 156510ebac4fSDavid Hildenbrand return 1; 15662b42a7e5SDavid Hildenbrand } 15672b42a7e5SDavid Hildenbrand 156851c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 1569b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 15701da177e4SLinus Torvalds unsigned long addr, unsigned long end, 157197a89413SPeter Zijlstra struct zap_details *details) 15721da177e4SLinus Torvalds { 1573789753e1SDavid Hildenbrand bool force_flush = false, force_break = false; 1574b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 1575d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 157697a89413SPeter Zijlstra spinlock_t *ptl; 15775f1a1907SSteven Rostedt pte_t *start_pte; 157897a89413SPeter Zijlstra pte_t *pte; 15798a5f14a2SKirill A. Shutemov swp_entry_t entry; 158010ebac4fSDavid Hildenbrand int nr; 1581d559db08SKAMEZAWA Hiroyuki 1582ed6a7935SPeter Zijlstra tlb_change_page_size(tlb, PAGE_SIZE); 1583e303297eSPeter Zijlstra init_rss_vec(rss); 15843db82b93SHugh Dickins start_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 15853db82b93SHugh Dickins if (!pte) 15863db82b93SHugh Dickins return addr; 15873db82b93SHugh Dickins 15883ea27719SMel Gorman flush_tlb_batched_pending(mm); 15896606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 15901da177e4SLinus Torvalds do { 1591c33c7948SRyan Roberts pte_t ptent = ptep_get(pte); 1592789753e1SDavid Hildenbrand struct folio *folio; 15938018db85SPeter Xu struct page *page; 159410ebac4fSDavid Hildenbrand int max_nr; 15958018db85SPeter Xu 159610ebac4fSDavid Hildenbrand nr = 1; 1597166f61b9STobin C Harding if (pte_none(ptent)) 15981da177e4SLinus Torvalds continue; 159951c6f666SRobin Holt 16007b167b68SMinchan Kim if (need_resched()) 16017b167b68SMinchan Kim break; 16027b167b68SMinchan Kim 16036f5e6b9eSHugh Dickins if (pte_present(ptent)) { 160410ebac4fSDavid Hildenbrand max_nr = (end - addr) / PAGE_SIZE; 160510ebac4fSDavid Hildenbrand nr = zap_present_ptes(tlb, vma, pte, ptent, max_nr, 160610ebac4fSDavid Hildenbrand addr, details, rss, &force_flush, 160710ebac4fSDavid Hildenbrand &force_break); 1608789753e1SDavid Hildenbrand if (unlikely(force_break)) { 160910ebac4fSDavid Hildenbrand addr += nr * PAGE_SIZE; 1610d16dfc55SPeter Zijlstra break; 16111cf35d47SLinus Torvalds } 16121da177e4SLinus Torvalds continue; 16131da177e4SLinus Torvalds } 16145042db43SJérôme Glisse 16155042db43SJérôme Glisse entry = pte_to_swp_entry(ptent); 1616b756a3b5SAlistair Popple if (is_device_private_entry(entry) || 1617b756a3b5SAlistair Popple is_device_exclusive_entry(entry)) { 16188018db85SPeter Xu page = pfn_swap_entry_to_page(entry); 1619c4626503SDavid Hildenbrand folio = page_folio(page); 1620eabafaaaSKefeng Wang if (unlikely(!should_zap_folio(details, folio))) 16215042db43SJérôme Glisse continue; 1622999dad82SPeter Xu /* 1623999dad82SPeter Xu * Both device private/exclusive mappings should only 1624999dad82SPeter Xu * work with anonymous page so far, so we don't need to 1625999dad82SPeter Xu * consider uffd-wp bit when zap. For more information, 1626999dad82SPeter Xu * see zap_install_uffd_wp_if_needed(). 1627999dad82SPeter Xu */ 1628999dad82SPeter Xu WARN_ON_ONCE(!vma_is_anonymous(vma)); 1629a23f517bSKefeng Wang rss[mm_counter(folio)]--; 1630b756a3b5SAlistair Popple if (is_device_private_entry(entry)) 1631c4626503SDavid Hildenbrand folio_remove_rmap_pte(folio, page, vma); 1632c4626503SDavid Hildenbrand folio_put(folio); 16338018db85SPeter Xu } else if (!non_swap_entry(entry)) { 16345abfd71dSPeter Xu /* Genuine swap entry, hence a private anon page */ 16355abfd71dSPeter Xu if (!should_zap_cows(details)) 16361da177e4SLinus Torvalds continue; 1637b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 16388018db85SPeter Xu if (unlikely(!free_swap_and_cache(entry))) 16398018db85SPeter Xu print_bad_pte(vma, addr, ptent, NULL); 16405abfd71dSPeter Xu } else if (is_migration_entry(entry)) { 1641eabafaaaSKefeng Wang folio = pfn_swap_entry_folio(entry); 1642eabafaaaSKefeng Wang if (!should_zap_folio(details, folio)) 16435abfd71dSPeter Xu continue; 1644a23f517bSKefeng Wang rss[mm_counter(folio)]--; 1645999dad82SPeter Xu } else if (pte_marker_entry_uffd_wp(entry)) { 16462bad466cSPeter Xu /* 16472bad466cSPeter Xu * For anon: always drop the marker; for file: only 16482bad466cSPeter Xu * drop the marker if explicitly requested. 16492bad466cSPeter Xu */ 16502bad466cSPeter Xu if (!vma_is_anonymous(vma) && 16512bad466cSPeter Xu !zap_drop_file_uffd_wp(details)) 1652999dad82SPeter Xu continue; 16539f186f9eSMiaohe Lin } else if (is_hwpoison_entry(entry) || 1654af19487fSAxel Rasmussen is_poisoned_swp_entry(entry)) { 16555abfd71dSPeter Xu if (!should_zap_cows(details)) 16565abfd71dSPeter Xu continue; 16575abfd71dSPeter Xu } else { 16585abfd71dSPeter Xu /* We should have covered all the swap entry types */ 1659727d16f1SAndrew Morton pr_alert("unrecognized swap entry 0x%lx\n", entry.val); 16605abfd71dSPeter Xu WARN_ON_ONCE(1); 16619f9f1acdSKonstantin Khlebnikov } 16629888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 166310ebac4fSDavid Hildenbrand zap_install_uffd_wp_if_needed(vma, addr, pte, 1, details, ptent); 166410ebac4fSDavid Hildenbrand } while (pte += nr, addr += PAGE_SIZE * nr, addr != end); 1665ae859762SHugh Dickins 1666d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 16676606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 166851c6f666SRobin Holt 16691cf35d47SLinus Torvalds /* Do the actual TLB flush before dropping ptl */ 16705df397deSLinus Torvalds if (force_flush) { 16711cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 16725df397deSLinus Torvalds tlb_flush_rmaps(tlb, vma); 16735df397deSLinus Torvalds } 16741cf35d47SLinus Torvalds pte_unmap_unlock(start_pte, ptl); 16751cf35d47SLinus Torvalds 16761cf35d47SLinus Torvalds /* 16771cf35d47SLinus Torvalds * If we forced a TLB flush (either due to running out of 16781cf35d47SLinus Torvalds * batch buffers or because we needed to flush dirty TLB 16791cf35d47SLinus Torvalds * entries before releasing the ptl), free the batched 16803db82b93SHugh Dickins * memory too. Come back again if we didn't do everything. 16811cf35d47SLinus Torvalds */ 16823db82b93SHugh Dickins if (force_flush) 1683fa0aafb8SPeter Zijlstra tlb_flush_mmu(tlb); 1684d16dfc55SPeter Zijlstra 168551c6f666SRobin Holt return addr; 16861da177e4SLinus Torvalds } 16871da177e4SLinus Torvalds 168851c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1689b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 16901da177e4SLinus Torvalds unsigned long addr, unsigned long end, 169197a89413SPeter Zijlstra struct zap_details *details) 16921da177e4SLinus Torvalds { 16931da177e4SLinus Torvalds pmd_t *pmd; 16941da177e4SLinus Torvalds unsigned long next; 16951da177e4SLinus Torvalds 16961da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 16971da177e4SLinus Torvalds do { 16981da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 169984c3fc4eSZi Yan if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { 170053406ed1SHugh Dickins if (next - addr != HPAGE_PMD_SIZE) 1701fd60775aSDavid Rientjes __split_huge_pmd(vma, pmd, addr, false, NULL); 17023db82b93SHugh Dickins else if (zap_huge_pmd(tlb, vma, pmd, addr)) { 17033db82b93SHugh Dickins addr = next; 17043db82b93SHugh Dickins continue; 17053db82b93SHugh Dickins } 170671e3aac0SAndrea Arcangeli /* fall through */ 17073506659eSMatthew Wilcox (Oracle) } else if (details && details->single_folio && 17083506659eSMatthew Wilcox (Oracle) folio_test_pmd_mappable(details->single_folio) && 170922061a1fSHugh Dickins next - addr == HPAGE_PMD_SIZE && pmd_none(*pmd)) { 171022061a1fSHugh Dickins spinlock_t *ptl = pmd_lock(tlb->mm, pmd); 171122061a1fSHugh Dickins /* 171222061a1fSHugh Dickins * Take and drop THP pmd lock so that we cannot return 171322061a1fSHugh Dickins * prematurely, while zap_huge_pmd() has cleared *pmd, 171422061a1fSHugh Dickins * but not yet decremented compound_mapcount(). 171522061a1fSHugh Dickins */ 171622061a1fSHugh Dickins spin_unlock(ptl); 171771e3aac0SAndrea Arcangeli } 17183db82b93SHugh Dickins if (pmd_none(*pmd)) { 17193db82b93SHugh Dickins addr = next; 17203db82b93SHugh Dickins continue; 17213db82b93SHugh Dickins } 17223db82b93SHugh Dickins addr = zap_pte_range(tlb, vma, pmd, addr, next, details); 17233db82b93SHugh Dickins if (addr != next) 17243db82b93SHugh Dickins pmd--; 17253db82b93SHugh Dickins } while (pmd++, cond_resched(), addr != end); 172651c6f666SRobin Holt 172751c6f666SRobin Holt return addr; 17281da177e4SLinus Torvalds } 17291da177e4SLinus Torvalds 173051c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1731c2febafcSKirill A. Shutemov struct vm_area_struct *vma, p4d_t *p4d, 17321da177e4SLinus Torvalds unsigned long addr, unsigned long end, 173397a89413SPeter Zijlstra struct zap_details *details) 17341da177e4SLinus Torvalds { 17351da177e4SLinus Torvalds pud_t *pud; 17361da177e4SLinus Torvalds unsigned long next; 17371da177e4SLinus Torvalds 1738c2febafcSKirill A. Shutemov pud = pud_offset(p4d, addr); 17391da177e4SLinus Torvalds do { 17401da177e4SLinus Torvalds next = pud_addr_end(addr, end); 1741a00cc7d9SMatthew Wilcox if (pud_trans_huge(*pud) || pud_devmap(*pud)) { 1742a00cc7d9SMatthew Wilcox if (next - addr != HPAGE_PUD_SIZE) { 174342fc5414SMichel Lespinasse mmap_assert_locked(tlb->mm); 1744a00cc7d9SMatthew Wilcox split_huge_pud(vma, pud, addr); 1745a00cc7d9SMatthew Wilcox } else if (zap_huge_pud(tlb, vma, pud, addr)) 1746a00cc7d9SMatthew Wilcox goto next; 1747a00cc7d9SMatthew Wilcox /* fall through */ 1748a00cc7d9SMatthew Wilcox } 174997a89413SPeter Zijlstra if (pud_none_or_clear_bad(pud)) 17501da177e4SLinus Torvalds continue; 175197a89413SPeter Zijlstra next = zap_pmd_range(tlb, vma, pud, addr, next, details); 1752a00cc7d9SMatthew Wilcox next: 1753a00cc7d9SMatthew Wilcox cond_resched(); 175497a89413SPeter Zijlstra } while (pud++, addr = next, addr != end); 175551c6f666SRobin Holt 175651c6f666SRobin Holt return addr; 17571da177e4SLinus Torvalds } 17581da177e4SLinus Torvalds 1759c2febafcSKirill A. Shutemov static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, 1760c2febafcSKirill A. Shutemov struct vm_area_struct *vma, pgd_t *pgd, 1761c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 1762c2febafcSKirill A. Shutemov struct zap_details *details) 1763c2febafcSKirill A. Shutemov { 1764c2febafcSKirill A. Shutemov p4d_t *p4d; 1765c2febafcSKirill A. Shutemov unsigned long next; 1766c2febafcSKirill A. Shutemov 1767c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, addr); 1768c2febafcSKirill A. Shutemov do { 1769c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 1770c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(p4d)) 1771c2febafcSKirill A. Shutemov continue; 1772c2febafcSKirill A. Shutemov next = zap_pud_range(tlb, vma, p4d, addr, next, details); 1773c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 1774c2febafcSKirill A. Shutemov 1775c2febafcSKirill A. Shutemov return addr; 1776c2febafcSKirill A. Shutemov } 1777c2febafcSKirill A. Shutemov 1778aac45363SMichal Hocko void unmap_page_range(struct mmu_gather *tlb, 177951c6f666SRobin Holt struct vm_area_struct *vma, 17801da177e4SLinus Torvalds unsigned long addr, unsigned long end, 178197a89413SPeter Zijlstra struct zap_details *details) 17821da177e4SLinus Torvalds { 17831da177e4SLinus Torvalds pgd_t *pgd; 17841da177e4SLinus Torvalds unsigned long next; 17851da177e4SLinus Torvalds 17861da177e4SLinus Torvalds BUG_ON(addr >= end); 17871da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 17881da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 17891da177e4SLinus Torvalds do { 17901da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 179197a89413SPeter Zijlstra if (pgd_none_or_clear_bad(pgd)) 17921da177e4SLinus Torvalds continue; 1793c2febafcSKirill A. Shutemov next = zap_p4d_range(tlb, vma, pgd, addr, next, details); 179497a89413SPeter Zijlstra } while (pgd++, addr = next, addr != end); 17951da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 17961da177e4SLinus Torvalds } 17971da177e4SLinus Torvalds 1798f5cc4eefSAl Viro 1799f5cc4eefSAl Viro static void unmap_single_vma(struct mmu_gather *tlb, 18001da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 18014f74d2c8SLinus Torvalds unsigned long end_addr, 180268f48381SSuren Baghdasaryan struct zap_details *details, bool mm_wr_locked) 18031da177e4SLinus Torvalds { 1804f5cc4eefSAl Viro unsigned long start = max(vma->vm_start, start_addr); 18051da177e4SLinus Torvalds unsigned long end; 18061da177e4SLinus Torvalds 18071da177e4SLinus Torvalds if (start >= vma->vm_end) 1808f5cc4eefSAl Viro return; 18091da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 18101da177e4SLinus Torvalds if (end <= vma->vm_start) 1811f5cc4eefSAl Viro return; 18121da177e4SLinus Torvalds 1813cbc91f71SSrikar Dronamraju if (vma->vm_file) 1814cbc91f71SSrikar Dronamraju uprobe_munmap(vma, start, end); 1815cbc91f71SSrikar Dronamraju 1816b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) 181768f48381SSuren Baghdasaryan untrack_pfn(vma, 0, 0, mm_wr_locked); 18182ab64037Svenkatesh.pallipadi@intel.com 18198b2a1238SAl Viro if (start != end) { 182051c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1821a137e1ccSAndi Kleen /* 1822a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1823a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1824a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 18257aa6b4adSDavidlohr Bueso * cleanup path of mmap_region. When 1826a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 18277aa6b4adSDavidlohr Bueso * mmap_region() nullifies vma->vm_file 1828a137e1ccSAndi Kleen * before calling this function to clean up. 1829a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1830a137e1ccSAndi Kleen * safe to do nothing in this case. 1831a137e1ccSAndi Kleen */ 183224669e58SAneesh Kumar K.V if (vma->vm_file) { 183305e90bd0SPeter Xu zap_flags_t zap_flags = details ? 183405e90bd0SPeter Xu details->zap_flags : 0; 18352820b0f0SRik van Riel __unmap_hugepage_range(tlb, vma, start, end, 183605e90bd0SPeter Xu NULL, zap_flags); 183724669e58SAneesh Kumar K.V } 183851c6f666SRobin Holt } else 1839038c7aa1SAl Viro unmap_page_range(tlb, vma, start, end, details); 184097a89413SPeter Zijlstra } 184151c6f666SRobin Holt } 18421da177e4SLinus Torvalds 1843f5cc4eefSAl Viro /** 1844f5cc4eefSAl Viro * unmap_vmas - unmap a range of memory covered by a list of vma's 1845f5cc4eefSAl Viro * @tlb: address of the caller's struct mmu_gather 18466e412203SYang Li * @mas: the maple state 1847f5cc4eefSAl Viro * @vma: the starting vma 1848f5cc4eefSAl Viro * @start_addr: virtual address at which to start unmapping 1849f5cc4eefSAl Viro * @end_addr: virtual address at which to end unmapping 18506e412203SYang Li * @tree_end: The maximum index to check 1851809ef83cSYang Li * @mm_wr_locked: lock flag 1852f5cc4eefSAl Viro * 1853f5cc4eefSAl Viro * Unmap all pages in the vma list. 1854f5cc4eefSAl Viro * 1855f5cc4eefSAl Viro * Only addresses between `start' and `end' will be unmapped. 1856f5cc4eefSAl Viro * 1857f5cc4eefSAl Viro * The VMA list must be sorted in ascending virtual address order. 1858f5cc4eefSAl Viro * 1859f5cc4eefSAl Viro * unmap_vmas() assumes that the caller will flush the whole unmapped address 1860f5cc4eefSAl Viro * range after unmap_vmas() returns. So the only responsibility here is to 1861f5cc4eefSAl Viro * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 1862f5cc4eefSAl Viro * drops the lock and schedules. 1863f5cc4eefSAl Viro */ 1864fd892593SLiam R. Howlett void unmap_vmas(struct mmu_gather *tlb, struct ma_state *mas, 1865f5cc4eefSAl Viro struct vm_area_struct *vma, unsigned long start_addr, 1866fd892593SLiam R. Howlett unsigned long end_addr, unsigned long tree_end, 1867fd892593SLiam R. Howlett bool mm_wr_locked) 1868f5cc4eefSAl Viro { 1869ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1870999dad82SPeter Xu struct zap_details details = { 187104ada095SMike Kravetz .zap_flags = ZAP_FLAG_DROP_MARKER | ZAP_FLAG_UNMAP, 1872999dad82SPeter Xu /* Careful - we need to zap private pages too! */ 1873999dad82SPeter Xu .even_cows = true, 1874999dad82SPeter Xu }; 1875f5cc4eefSAl Viro 18767d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma->vm_mm, 18776f4f13e8SJérôme Glisse start_addr, end_addr); 1878ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 1879763ecb03SLiam R. Howlett do { 18802820b0f0SRik van Riel unsigned long start = start_addr; 18812820b0f0SRik van Riel unsigned long end = end_addr; 18822820b0f0SRik van Riel hugetlb_zap_begin(vma, &start, &end); 18832820b0f0SRik van Riel unmap_single_vma(tlb, vma, start, end, &details, 188468f48381SSuren Baghdasaryan mm_wr_locked); 18852820b0f0SRik van Riel hugetlb_zap_end(vma, &details); 1886d2406291SPeng Zhang vma = mas_find(mas, tree_end - 1); 1887d2406291SPeng Zhang } while (vma && likely(!xa_is_zero(vma))); 1888ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 18891da177e4SLinus Torvalds } 18901da177e4SLinus Torvalds 18911da177e4SLinus Torvalds /** 1892f5cc4eefSAl Viro * zap_page_range_single - remove user pages in a given range 1893f5cc4eefSAl Viro * @vma: vm_area_struct holding the applicable pages 1894f5cc4eefSAl Viro * @address: starting address of pages to zap 1895f5cc4eefSAl Viro * @size: number of bytes to zap 18968a5f14a2SKirill A. Shutemov * @details: details of shared cache invalidation 1897f5cc4eefSAl Viro * 1898f5cc4eefSAl Viro * The range must fit into one VMA. 1899f5cc4eefSAl Viro */ 190021b85b09SMike Kravetz void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 1901f5cc4eefSAl Viro unsigned long size, struct zap_details *details) 1902f5cc4eefSAl Viro { 190321b85b09SMike Kravetz const unsigned long end = address + size; 1904ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1905f5cc4eefSAl Viro struct mmu_gather tlb; 1906f5cc4eefSAl Viro 1907f5cc4eefSAl Viro lru_add_drain(); 19087d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma->vm_mm, 190921b85b09SMike Kravetz address, end); 19102820b0f0SRik van Riel hugetlb_zap_begin(vma, &range.start, &range.end); 1911a72afd87SWill Deacon tlb_gather_mmu(&tlb, vma->vm_mm); 1912ac46d4f3SJérôme Glisse update_hiwater_rss(vma->vm_mm); 1913ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 191421b85b09SMike Kravetz /* 191521b85b09SMike Kravetz * unmap 'address-end' not 'range.start-range.end' as range 191621b85b09SMike Kravetz * could have been expanded for hugetlb pmd sharing. 191721b85b09SMike Kravetz */ 191868f48381SSuren Baghdasaryan unmap_single_vma(&tlb, vma, address, end, details, false); 1919ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 1920ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 19212820b0f0SRik van Riel hugetlb_zap_end(vma, details); 19221da177e4SLinus Torvalds } 19231da177e4SLinus Torvalds 1924c627f9ccSJack Steiner /** 1925c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1926c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1927c627f9ccSJack Steiner * @address: starting address of pages to zap 1928c627f9ccSJack Steiner * @size: number of bytes to zap 1929c627f9ccSJack Steiner * 1930c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1931c627f9ccSJack Steiner * 1932c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1933c627f9ccSJack Steiner * 1934c627f9ccSJack Steiner */ 193527d036e3SLeon Romanovsky void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1936c627f9ccSJack Steiner unsigned long size) 1937c627f9ccSJack Steiner { 193888a35912SMiaohe Lin if (!range_in_vma(vma, address, address + size) || 1939c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 194027d036e3SLeon Romanovsky return; 194127d036e3SLeon Romanovsky 1942f5cc4eefSAl Viro zap_page_range_single(vma, address, size, NULL); 1943c627f9ccSJack Steiner } 1944c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1945c627f9ccSJack Steiner 19468cd3984dSArjun Roy static pmd_t *walk_to_pmd(struct mm_struct *mm, unsigned long addr) 1947c9cfcddfSLinus Torvalds { 1948c2febafcSKirill A. Shutemov pgd_t *pgd; 1949c2febafcSKirill A. Shutemov p4d_t *p4d; 1950c2febafcSKirill A. Shutemov pud_t *pud; 1951c2febafcSKirill A. Shutemov pmd_t *pmd; 1952c2febafcSKirill A. Shutemov 1953c2febafcSKirill A. Shutemov pgd = pgd_offset(mm, addr); 1954c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 1955c2febafcSKirill A. Shutemov if (!p4d) 1956c2febafcSKirill A. Shutemov return NULL; 1957c2febafcSKirill A. Shutemov pud = pud_alloc(mm, p4d, addr); 1958c2febafcSKirill A. Shutemov if (!pud) 1959c2febafcSKirill A. Shutemov return NULL; 1960c2febafcSKirill A. Shutemov pmd = pmd_alloc(mm, pud, addr); 1961c2febafcSKirill A. Shutemov if (!pmd) 1962c2febafcSKirill A. Shutemov return NULL; 1963c2febafcSKirill A. Shutemov 1964f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 19658cd3984dSArjun Roy return pmd; 19668cd3984dSArjun Roy } 19678cd3984dSArjun Roy 19688cd3984dSArjun Roy pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 19698cd3984dSArjun Roy spinlock_t **ptl) 19708cd3984dSArjun Roy { 19718cd3984dSArjun Roy pmd_t *pmd = walk_to_pmd(mm, addr); 19728cd3984dSArjun Roy 19738cd3984dSArjun Roy if (!pmd) 19748cd3984dSArjun Roy return NULL; 1975c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 1976c9cfcddfSLinus Torvalds } 1977c9cfcddfSLinus Torvalds 19788efd6f5bSArjun Roy static int validate_page_before_insert(struct page *page) 19798efd6f5bSArjun Roy { 1980f8b6187dSKefeng Wang struct folio *folio = page_folio(page); 1981f8b6187dSKefeng Wang 1982f8b6187dSKefeng Wang if (folio_test_anon(folio) || folio_test_slab(folio) || 1983f8b6187dSKefeng Wang page_has_type(page)) 19848efd6f5bSArjun Roy return -EINVAL; 1985f8b6187dSKefeng Wang flush_dcache_folio(folio); 19868efd6f5bSArjun Roy return 0; 19878efd6f5bSArjun Roy } 19888efd6f5bSArjun Roy 1989cea86fe2SHugh Dickins static int insert_page_into_pte_locked(struct vm_area_struct *vma, pte_t *pte, 19908efd6f5bSArjun Roy unsigned long addr, struct page *page, pgprot_t prot) 19918efd6f5bSArjun Roy { 1992ef37b2eaSDavid Hildenbrand struct folio *folio = page_folio(page); 1993ef37b2eaSDavid Hildenbrand 1994c33c7948SRyan Roberts if (!pte_none(ptep_get(pte))) 19958efd6f5bSArjun Roy return -EBUSY; 19968efd6f5bSArjun Roy /* Ok, finally just insert the thing.. */ 1997ef37b2eaSDavid Hildenbrand folio_get(folio); 19986b27cc6cSKefeng Wang inc_mm_counter(vma->vm_mm, mm_counter_file(folio)); 1999ef37b2eaSDavid Hildenbrand folio_add_file_rmap_pte(folio, page, vma); 2000cea86fe2SHugh Dickins set_pte_at(vma->vm_mm, addr, pte, mk_pte(page, prot)); 20018efd6f5bSArjun Roy return 0; 20028efd6f5bSArjun Roy } 20038efd6f5bSArjun Roy 20041da177e4SLinus Torvalds /* 2005238f58d8SLinus Torvalds * This is the old fallback for page remapping. 2006238f58d8SLinus Torvalds * 2007238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 2008238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 2009238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 2010238f58d8SLinus Torvalds */ 2011423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 2012423bad60SNick Piggin struct page *page, pgprot_t prot) 2013238f58d8SLinus Torvalds { 2014238f58d8SLinus Torvalds int retval; 2015238f58d8SLinus Torvalds pte_t *pte; 2016238f58d8SLinus Torvalds spinlock_t *ptl; 2017238f58d8SLinus Torvalds 20188efd6f5bSArjun Roy retval = validate_page_before_insert(page); 20198efd6f5bSArjun Roy if (retval) 20205b4e655eSKAMEZAWA Hiroyuki goto out; 2021238f58d8SLinus Torvalds retval = -ENOMEM; 2022cea86fe2SHugh Dickins pte = get_locked_pte(vma->vm_mm, addr, &ptl); 2023238f58d8SLinus Torvalds if (!pte) 20245b4e655eSKAMEZAWA Hiroyuki goto out; 2025cea86fe2SHugh Dickins retval = insert_page_into_pte_locked(vma, pte, addr, page, prot); 2026238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 2027238f58d8SLinus Torvalds out: 2028238f58d8SLinus Torvalds return retval; 2029238f58d8SLinus Torvalds } 2030238f58d8SLinus Torvalds 2031cea86fe2SHugh Dickins static int insert_page_in_batch_locked(struct vm_area_struct *vma, pte_t *pte, 20328cd3984dSArjun Roy unsigned long addr, struct page *page, pgprot_t prot) 20338cd3984dSArjun Roy { 20348cd3984dSArjun Roy int err; 20358cd3984dSArjun Roy 20368cd3984dSArjun Roy if (!page_count(page)) 20378cd3984dSArjun Roy return -EINVAL; 20388cd3984dSArjun Roy err = validate_page_before_insert(page); 20397f70c2a6SArjun Roy if (err) 20407f70c2a6SArjun Roy return err; 2041cea86fe2SHugh Dickins return insert_page_into_pte_locked(vma, pte, addr, page, prot); 20428cd3984dSArjun Roy } 20438cd3984dSArjun Roy 20448cd3984dSArjun Roy /* insert_pages() amortizes the cost of spinlock operations 2045bb7dbaafSMatthew Wilcox (Oracle) * when inserting pages in a loop. 20468cd3984dSArjun Roy */ 20478cd3984dSArjun Roy static int insert_pages(struct vm_area_struct *vma, unsigned long addr, 20488cd3984dSArjun Roy struct page **pages, unsigned long *num, pgprot_t prot) 20498cd3984dSArjun Roy { 20508cd3984dSArjun Roy pmd_t *pmd = NULL; 20517f70c2a6SArjun Roy pte_t *start_pte, *pte; 20527f70c2a6SArjun Roy spinlock_t *pte_lock; 20538cd3984dSArjun Roy struct mm_struct *const mm = vma->vm_mm; 20548cd3984dSArjun Roy unsigned long curr_page_idx = 0; 20558cd3984dSArjun Roy unsigned long remaining_pages_total = *num; 20568cd3984dSArjun Roy unsigned long pages_to_write_in_pmd; 20578cd3984dSArjun Roy int ret; 20588cd3984dSArjun Roy more: 20598cd3984dSArjun Roy ret = -EFAULT; 20608cd3984dSArjun Roy pmd = walk_to_pmd(mm, addr); 20618cd3984dSArjun Roy if (!pmd) 20628cd3984dSArjun Roy goto out; 20638cd3984dSArjun Roy 20648cd3984dSArjun Roy pages_to_write_in_pmd = min_t(unsigned long, 20658cd3984dSArjun Roy remaining_pages_total, PTRS_PER_PTE - pte_index(addr)); 20668cd3984dSArjun Roy 20678cd3984dSArjun Roy /* Allocate the PTE if necessary; takes PMD lock once only. */ 20688cd3984dSArjun Roy ret = -ENOMEM; 20698cd3984dSArjun Roy if (pte_alloc(mm, pmd)) 20708cd3984dSArjun Roy goto out; 20718cd3984dSArjun Roy 20728cd3984dSArjun Roy while (pages_to_write_in_pmd) { 20738cd3984dSArjun Roy int pte_idx = 0; 20748cd3984dSArjun Roy const int batch_size = min_t(int, pages_to_write_in_pmd, 8); 20758cd3984dSArjun Roy 20767f70c2a6SArjun Roy start_pte = pte_offset_map_lock(mm, pmd, addr, &pte_lock); 20773db82b93SHugh Dickins if (!start_pte) { 20783db82b93SHugh Dickins ret = -EFAULT; 20793db82b93SHugh Dickins goto out; 20803db82b93SHugh Dickins } 20817f70c2a6SArjun Roy for (pte = start_pte; pte_idx < batch_size; ++pte, ++pte_idx) { 2082cea86fe2SHugh Dickins int err = insert_page_in_batch_locked(vma, pte, 20838cd3984dSArjun Roy addr, pages[curr_page_idx], prot); 20848cd3984dSArjun Roy if (unlikely(err)) { 20857f70c2a6SArjun Roy pte_unmap_unlock(start_pte, pte_lock); 20868cd3984dSArjun Roy ret = err; 20878cd3984dSArjun Roy remaining_pages_total -= pte_idx; 20888cd3984dSArjun Roy goto out; 20898cd3984dSArjun Roy } 20908cd3984dSArjun Roy addr += PAGE_SIZE; 20918cd3984dSArjun Roy ++curr_page_idx; 20928cd3984dSArjun Roy } 20937f70c2a6SArjun Roy pte_unmap_unlock(start_pte, pte_lock); 20948cd3984dSArjun Roy pages_to_write_in_pmd -= batch_size; 20958cd3984dSArjun Roy remaining_pages_total -= batch_size; 20968cd3984dSArjun Roy } 20978cd3984dSArjun Roy if (remaining_pages_total) 20988cd3984dSArjun Roy goto more; 20998cd3984dSArjun Roy ret = 0; 21008cd3984dSArjun Roy out: 21018cd3984dSArjun Roy *num = remaining_pages_total; 21028cd3984dSArjun Roy return ret; 21038cd3984dSArjun Roy } 21048cd3984dSArjun Roy 21058cd3984dSArjun Roy /** 21068cd3984dSArjun Roy * vm_insert_pages - insert multiple pages into user vma, batching the pmd lock. 21078cd3984dSArjun Roy * @vma: user vma to map to 21088cd3984dSArjun Roy * @addr: target start user address of these pages 21098cd3984dSArjun Roy * @pages: source kernel pages 21108cd3984dSArjun Roy * @num: in: number of pages to map. out: number of pages that were *not* 21118cd3984dSArjun Roy * mapped. (0 means all pages were successfully mapped). 21128cd3984dSArjun Roy * 21138cd3984dSArjun Roy * Preferred over vm_insert_page() when inserting multiple pages. 21148cd3984dSArjun Roy * 21158cd3984dSArjun Roy * In case of error, we may have mapped a subset of the provided 21168cd3984dSArjun Roy * pages. It is the caller's responsibility to account for this case. 21178cd3984dSArjun Roy * 21188cd3984dSArjun Roy * The same restrictions apply as in vm_insert_page(). 21198cd3984dSArjun Roy */ 21208cd3984dSArjun Roy int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr, 21218cd3984dSArjun Roy struct page **pages, unsigned long *num) 21228cd3984dSArjun Roy { 21238cd3984dSArjun Roy const unsigned long end_addr = addr + (*num * PAGE_SIZE) - 1; 21248cd3984dSArjun Roy 21258cd3984dSArjun Roy if (addr < vma->vm_start || end_addr >= vma->vm_end) 21268cd3984dSArjun Roy return -EFAULT; 21278cd3984dSArjun Roy if (!(vma->vm_flags & VM_MIXEDMAP)) { 2128d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(vma->vm_mm)); 21298cd3984dSArjun Roy BUG_ON(vma->vm_flags & VM_PFNMAP); 21301c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_MIXEDMAP); 21318cd3984dSArjun Roy } 21328cd3984dSArjun Roy /* Defer page refcount checking till we're about to map that page. */ 21338cd3984dSArjun Roy return insert_pages(vma, addr, pages, num, vma->vm_page_prot); 21348cd3984dSArjun Roy } 21358cd3984dSArjun Roy EXPORT_SYMBOL(vm_insert_pages); 21368cd3984dSArjun Roy 2137bfa5bf6dSRolf Eike Beer /** 2138bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 2139bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 2140bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 2141bfa5bf6dSRolf Eike Beer * @page: source kernel page 2142bfa5bf6dSRolf Eike Beer * 2143a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 2144a145dd41SLinus Torvalds * into a user vma. 2145a145dd41SLinus Torvalds * 2146a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 2147a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 2148a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 21498dfcc9baSNick Piggin * (see split_page()). 2150a145dd41SLinus Torvalds * 2151a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 2152a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 2153a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 2154a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 2155a145dd41SLinus Torvalds * ask for a shared writable mapping! 2156a145dd41SLinus Torvalds * 2157a145dd41SLinus Torvalds * The page does not need to be reserved. 21584b6e1e37SKonstantin Khlebnikov * 21594b6e1e37SKonstantin Khlebnikov * Usually this function is called from f_op->mmap() handler 2160c1e8d7c6SMichel Lespinasse * under mm->mmap_lock write-lock, so it can change vma->vm_flags. 21614b6e1e37SKonstantin Khlebnikov * Caller must set VM_MIXEDMAP on vma if it wants to call this 21624b6e1e37SKonstantin Khlebnikov * function from other places, for example from page-fault handler. 2163a862f68aSMike Rapoport * 2164a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 2165a145dd41SLinus Torvalds */ 2166423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 2167423bad60SNick Piggin struct page *page) 2168a145dd41SLinus Torvalds { 2169a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 2170a145dd41SLinus Torvalds return -EFAULT; 2171a145dd41SLinus Torvalds if (!page_count(page)) 2172a145dd41SLinus Torvalds return -EINVAL; 21734b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & VM_MIXEDMAP)) { 2174d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(vma->vm_mm)); 21754b6e1e37SKonstantin Khlebnikov BUG_ON(vma->vm_flags & VM_PFNMAP); 21761c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_MIXEDMAP); 21774b6e1e37SKonstantin Khlebnikov } 2178423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 2179a145dd41SLinus Torvalds } 2180e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 2181a145dd41SLinus Torvalds 2182a667d745SSouptick Joarder /* 2183a667d745SSouptick Joarder * __vm_map_pages - maps range of kernel pages into user vma 2184a667d745SSouptick Joarder * @vma: user vma to map to 2185a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2186a667d745SSouptick Joarder * @num: number of pages in page array 2187a667d745SSouptick Joarder * @offset: user's requested vm_pgoff 2188a667d745SSouptick Joarder * 2189a667d745SSouptick Joarder * This allows drivers to map range of kernel pages into a user vma. 2190a667d745SSouptick Joarder * 2191a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2192a667d745SSouptick Joarder */ 2193a667d745SSouptick Joarder static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages, 2194a667d745SSouptick Joarder unsigned long num, unsigned long offset) 2195a667d745SSouptick Joarder { 2196a667d745SSouptick Joarder unsigned long count = vma_pages(vma); 2197a667d745SSouptick Joarder unsigned long uaddr = vma->vm_start; 2198a667d745SSouptick Joarder int ret, i; 2199a667d745SSouptick Joarder 2200a667d745SSouptick Joarder /* Fail if the user requested offset is beyond the end of the object */ 220196756fcbSMiguel Ojeda if (offset >= num) 2202a667d745SSouptick Joarder return -ENXIO; 2203a667d745SSouptick Joarder 2204a667d745SSouptick Joarder /* Fail if the user requested size exceeds available object size */ 2205a667d745SSouptick Joarder if (count > num - offset) 2206a667d745SSouptick Joarder return -ENXIO; 2207a667d745SSouptick Joarder 2208a667d745SSouptick Joarder for (i = 0; i < count; i++) { 2209a667d745SSouptick Joarder ret = vm_insert_page(vma, uaddr, pages[offset + i]); 2210a667d745SSouptick Joarder if (ret < 0) 2211a667d745SSouptick Joarder return ret; 2212a667d745SSouptick Joarder uaddr += PAGE_SIZE; 2213a667d745SSouptick Joarder } 2214a667d745SSouptick Joarder 2215a667d745SSouptick Joarder return 0; 2216a667d745SSouptick Joarder } 2217a667d745SSouptick Joarder 2218a667d745SSouptick Joarder /** 2219a667d745SSouptick Joarder * vm_map_pages - maps range of kernel pages starts with non zero offset 2220a667d745SSouptick Joarder * @vma: user vma to map to 2221a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2222a667d745SSouptick Joarder * @num: number of pages in page array 2223a667d745SSouptick Joarder * 2224a667d745SSouptick Joarder * Maps an object consisting of @num pages, catering for the user's 2225a667d745SSouptick Joarder * requested vm_pgoff 2226a667d745SSouptick Joarder * 2227a667d745SSouptick Joarder * If we fail to insert any page into the vma, the function will return 2228a667d745SSouptick Joarder * immediately leaving any previously inserted pages present. Callers 2229a667d745SSouptick Joarder * from the mmap handler may immediately return the error as their caller 2230a667d745SSouptick Joarder * will destroy the vma, removing any successfully inserted pages. Other 2231a667d745SSouptick Joarder * callers should make their own arrangements for calling unmap_region(). 2232a667d745SSouptick Joarder * 2233a667d745SSouptick Joarder * Context: Process context. Called by mmap handlers. 2234a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2235a667d745SSouptick Joarder */ 2236a667d745SSouptick Joarder int vm_map_pages(struct vm_area_struct *vma, struct page **pages, 2237a667d745SSouptick Joarder unsigned long num) 2238a667d745SSouptick Joarder { 2239a667d745SSouptick Joarder return __vm_map_pages(vma, pages, num, vma->vm_pgoff); 2240a667d745SSouptick Joarder } 2241a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages); 2242a667d745SSouptick Joarder 2243a667d745SSouptick Joarder /** 2244a667d745SSouptick Joarder * vm_map_pages_zero - map range of kernel pages starts with zero offset 2245a667d745SSouptick Joarder * @vma: user vma to map to 2246a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2247a667d745SSouptick Joarder * @num: number of pages in page array 2248a667d745SSouptick Joarder * 2249a667d745SSouptick Joarder * Similar to vm_map_pages(), except that it explicitly sets the offset 2250a667d745SSouptick Joarder * to 0. This function is intended for the drivers that did not consider 2251a667d745SSouptick Joarder * vm_pgoff. 2252a667d745SSouptick Joarder * 2253a667d745SSouptick Joarder * Context: Process context. Called by mmap handlers. 2254a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2255a667d745SSouptick Joarder */ 2256a667d745SSouptick Joarder int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, 2257a667d745SSouptick Joarder unsigned long num) 2258a667d745SSouptick Joarder { 2259a667d745SSouptick Joarder return __vm_map_pages(vma, pages, num, 0); 2260a667d745SSouptick Joarder } 2261a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages_zero); 2262a667d745SSouptick Joarder 22639b5a8e00SMatthew Wilcox static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2264b2770da6SRoss Zwisler pfn_t pfn, pgprot_t prot, bool mkwrite) 2265423bad60SNick Piggin { 2266423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 2267423bad60SNick Piggin pte_t *pte, entry; 2268423bad60SNick Piggin spinlock_t *ptl; 2269423bad60SNick Piggin 2270423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 2271423bad60SNick Piggin if (!pte) 22729b5a8e00SMatthew Wilcox return VM_FAULT_OOM; 2273c33c7948SRyan Roberts entry = ptep_get(pte); 2274c33c7948SRyan Roberts if (!pte_none(entry)) { 2275b2770da6SRoss Zwisler if (mkwrite) { 2276b2770da6SRoss Zwisler /* 2277b2770da6SRoss Zwisler * For read faults on private mappings the PFN passed 2278b2770da6SRoss Zwisler * in may not match the PFN we have mapped if the 2279b2770da6SRoss Zwisler * mapped PFN is a writeable COW page. In the mkwrite 2280b2770da6SRoss Zwisler * case we are creating a writable PTE for a shared 2281f2c57d91SJan Kara * mapping and we expect the PFNs to match. If they 2282f2c57d91SJan Kara * don't match, we are likely racing with block 2283f2c57d91SJan Kara * allocation and mapping invalidation so just skip the 2284f2c57d91SJan Kara * update. 2285b2770da6SRoss Zwisler */ 2286c33c7948SRyan Roberts if (pte_pfn(entry) != pfn_t_to_pfn(pfn)) { 2287c33c7948SRyan Roberts WARN_ON_ONCE(!is_zero_pfn(pte_pfn(entry))); 2288423bad60SNick Piggin goto out_unlock; 2289f2c57d91SJan Kara } 2290c33c7948SRyan Roberts entry = pte_mkyoung(entry); 2291cae85cb8SJan Kara entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2292cae85cb8SJan Kara if (ptep_set_access_flags(vma, addr, pte, entry, 1)) 2293cae85cb8SJan Kara update_mmu_cache(vma, addr, pte); 2294cae85cb8SJan Kara } 2295b2770da6SRoss Zwisler goto out_unlock; 2296b2770da6SRoss Zwisler } 2297423bad60SNick Piggin 2298423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 229901c8f1c4SDan Williams if (pfn_t_devmap(pfn)) 230001c8f1c4SDan Williams entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); 230101c8f1c4SDan Williams else 230201c8f1c4SDan Williams entry = pte_mkspecial(pfn_t_pte(pfn, prot)); 2303b2770da6SRoss Zwisler 2304b2770da6SRoss Zwisler if (mkwrite) { 2305b2770da6SRoss Zwisler entry = pte_mkyoung(entry); 2306b2770da6SRoss Zwisler entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2307b2770da6SRoss Zwisler } 2308b2770da6SRoss Zwisler 2309423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 23104b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 2311423bad60SNick Piggin 2312423bad60SNick Piggin out_unlock: 2313423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 23149b5a8e00SMatthew Wilcox return VM_FAULT_NOPAGE; 2315423bad60SNick Piggin } 2316423bad60SNick Piggin 2317f5e6d1d5SMatthew Wilcox /** 2318f5e6d1d5SMatthew Wilcox * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot 2319f5e6d1d5SMatthew Wilcox * @vma: user vma to map to 2320f5e6d1d5SMatthew Wilcox * @addr: target user address of this page 2321f5e6d1d5SMatthew Wilcox * @pfn: source kernel pfn 2322f5e6d1d5SMatthew Wilcox * @pgprot: pgprot flags for the inserted page 2323f5e6d1d5SMatthew Wilcox * 2324a1a0aea5SRandy Dunlap * This is exactly like vmf_insert_pfn(), except that it allows drivers 2325f5e6d1d5SMatthew Wilcox * to override pgprot on a per-page basis. 2326f5e6d1d5SMatthew Wilcox * 2327f5e6d1d5SMatthew Wilcox * This only makes sense for IO mappings, and it makes no sense for 2328f5e6d1d5SMatthew Wilcox * COW mappings. In general, using multiple vmas is preferable; 2329ae2b01f3SMatthew Wilcox * vmf_insert_pfn_prot should only be used if using multiple VMAs is 2330f5e6d1d5SMatthew Wilcox * impractical. 2331f5e6d1d5SMatthew Wilcox * 233228d8b812SLorenzo Stoakes * pgprot typically only differs from @vma->vm_page_prot when drivers set 233328d8b812SLorenzo Stoakes * caching- and encryption bits different than those of @vma->vm_page_prot, 233428d8b812SLorenzo Stoakes * because the caching- or encryption mode may not be known at mmap() time. 233528d8b812SLorenzo Stoakes * 233628d8b812SLorenzo Stoakes * This is ok as long as @vma->vm_page_prot is not used by the core vm 233728d8b812SLorenzo Stoakes * to set caching and encryption bits for those vmas (except for COW pages). 233828d8b812SLorenzo Stoakes * This is ensured by core vm only modifying these page table entries using 233928d8b812SLorenzo Stoakes * functions that don't touch caching- or encryption bits, using pte_modify() 234028d8b812SLorenzo Stoakes * if needed. (See for example mprotect()). 234128d8b812SLorenzo Stoakes * 234228d8b812SLorenzo Stoakes * Also when new page-table entries are created, this is only done using the 234328d8b812SLorenzo Stoakes * fault() callback, and never using the value of vma->vm_page_prot, 234428d8b812SLorenzo Stoakes * except for page-table entries that point to anonymous pages as the result 234528d8b812SLorenzo Stoakes * of COW. 2346574c5b3dSThomas Hellstrom * 2347ae2b01f3SMatthew Wilcox * Context: Process context. May allocate using %GFP_KERNEL. 2348f5e6d1d5SMatthew Wilcox * Return: vm_fault_t value. 2349f5e6d1d5SMatthew Wilcox */ 2350f5e6d1d5SMatthew Wilcox vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, 2351f5e6d1d5SMatthew Wilcox unsigned long pfn, pgprot_t pgprot) 2352f5e6d1d5SMatthew Wilcox { 23536d958546SMatthew Wilcox /* 23546d958546SMatthew Wilcox * Technically, architectures with pte_special can avoid all these 23556d958546SMatthew Wilcox * restrictions (same for remap_pfn_range). However we would like 23566d958546SMatthew Wilcox * consistency in testing and feature parity among all, so we should 23576d958546SMatthew Wilcox * try to keep these invariants in place for everybody. 23586d958546SMatthew Wilcox */ 23596d958546SMatthew Wilcox BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 23606d958546SMatthew Wilcox BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 23616d958546SMatthew Wilcox (VM_PFNMAP|VM_MIXEDMAP)); 23626d958546SMatthew Wilcox BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 23636d958546SMatthew Wilcox BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 23646d958546SMatthew Wilcox 23656d958546SMatthew Wilcox if (addr < vma->vm_start || addr >= vma->vm_end) 23666d958546SMatthew Wilcox return VM_FAULT_SIGBUS; 23676d958546SMatthew Wilcox 23686d958546SMatthew Wilcox if (!pfn_modify_allowed(pfn, pgprot)) 23696d958546SMatthew Wilcox return VM_FAULT_SIGBUS; 23706d958546SMatthew Wilcox 23716d958546SMatthew Wilcox track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); 23726d958546SMatthew Wilcox 23739b5a8e00SMatthew Wilcox return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, 23746d958546SMatthew Wilcox false); 2375f5e6d1d5SMatthew Wilcox } 2376f5e6d1d5SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_pfn_prot); 2377e0dc0d8fSNick Piggin 2378ae2b01f3SMatthew Wilcox /** 2379ae2b01f3SMatthew Wilcox * vmf_insert_pfn - insert single pfn into user vma 2380ae2b01f3SMatthew Wilcox * @vma: user vma to map to 2381ae2b01f3SMatthew Wilcox * @addr: target user address of this page 2382ae2b01f3SMatthew Wilcox * @pfn: source kernel pfn 2383ae2b01f3SMatthew Wilcox * 2384ae2b01f3SMatthew Wilcox * Similar to vm_insert_page, this allows drivers to insert individual pages 2385ae2b01f3SMatthew Wilcox * they've allocated into a user vma. Same comments apply. 2386ae2b01f3SMatthew Wilcox * 2387ae2b01f3SMatthew Wilcox * This function should only be called from a vm_ops->fault handler, and 2388ae2b01f3SMatthew Wilcox * in that case the handler should return the result of this function. 2389ae2b01f3SMatthew Wilcox * 2390ae2b01f3SMatthew Wilcox * vma cannot be a COW mapping. 2391ae2b01f3SMatthew Wilcox * 2392ae2b01f3SMatthew Wilcox * As this is called only for pages that do not currently exist, we 2393ae2b01f3SMatthew Wilcox * do not need to flush old virtual caches or the TLB. 2394ae2b01f3SMatthew Wilcox * 2395ae2b01f3SMatthew Wilcox * Context: Process context. May allocate using %GFP_KERNEL. 2396ae2b01f3SMatthew Wilcox * Return: vm_fault_t value. 2397ae2b01f3SMatthew Wilcox */ 2398ae2b01f3SMatthew Wilcox vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2399ae2b01f3SMatthew Wilcox unsigned long pfn) 2400ae2b01f3SMatthew Wilcox { 2401ae2b01f3SMatthew Wilcox return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); 2402ae2b01f3SMatthew Wilcox } 2403ae2b01f3SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_pfn); 2404ae2b01f3SMatthew Wilcox 2405785a3fabSDan Williams static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) 2406785a3fabSDan Williams { 2407785a3fabSDan Williams /* these checks mirror the abort conditions in vm_normal_page */ 2408785a3fabSDan Williams if (vma->vm_flags & VM_MIXEDMAP) 2409785a3fabSDan Williams return true; 2410785a3fabSDan Williams if (pfn_t_devmap(pfn)) 2411785a3fabSDan Williams return true; 2412785a3fabSDan Williams if (pfn_t_special(pfn)) 2413785a3fabSDan Williams return true; 2414785a3fabSDan Williams if (is_zero_pfn(pfn_t_to_pfn(pfn))) 2415785a3fabSDan Williams return true; 2416785a3fabSDan Williams return false; 2417785a3fabSDan Williams } 2418785a3fabSDan Williams 241979f3aa5bSMatthew Wilcox static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma, 242028d8b812SLorenzo Stoakes unsigned long addr, pfn_t pfn, bool mkwrite) 2421423bad60SNick Piggin { 242228d8b812SLorenzo Stoakes pgprot_t pgprot = vma->vm_page_prot; 242379f3aa5bSMatthew Wilcox int err; 242487744ab3SDan Williams 2425785a3fabSDan Williams BUG_ON(!vm_mixed_ok(vma, pfn)); 2426423bad60SNick Piggin 2427423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 242879f3aa5bSMatthew Wilcox return VM_FAULT_SIGBUS; 2429308a047cSBorislav Petkov 2430308a047cSBorislav Petkov track_pfn_insert(vma, &pgprot, pfn); 2431423bad60SNick Piggin 243242e4089cSAndi Kleen if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) 243379f3aa5bSMatthew Wilcox return VM_FAULT_SIGBUS; 243442e4089cSAndi Kleen 2435423bad60SNick Piggin /* 2436423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 2437423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 2438423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 243962eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 244062eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 2441423bad60SNick Piggin */ 244200b3a331SLaurent Dufour if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && 244300b3a331SLaurent Dufour !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { 2444423bad60SNick Piggin struct page *page; 2445423bad60SNick Piggin 244603fc2da6SDan Williams /* 244703fc2da6SDan Williams * At this point we are committed to insert_page() 244803fc2da6SDan Williams * regardless of whether the caller specified flags that 244903fc2da6SDan Williams * result in pfn_t_has_page() == false. 245003fc2da6SDan Williams */ 245103fc2da6SDan Williams page = pfn_to_page(pfn_t_to_pfn(pfn)); 245279f3aa5bSMatthew Wilcox err = insert_page(vma, addr, page, pgprot); 245379f3aa5bSMatthew Wilcox } else { 24549b5a8e00SMatthew Wilcox return insert_pfn(vma, addr, pfn, pgprot, mkwrite); 2455423bad60SNick Piggin } 2456b2770da6SRoss Zwisler 24575d747637SMatthew Wilcox if (err == -ENOMEM) 24585d747637SMatthew Wilcox return VM_FAULT_OOM; 24595d747637SMatthew Wilcox if (err < 0 && err != -EBUSY) 24605d747637SMatthew Wilcox return VM_FAULT_SIGBUS; 24615d747637SMatthew Wilcox 24625d747637SMatthew Wilcox return VM_FAULT_NOPAGE; 2463423bad60SNick Piggin } 246479f3aa5bSMatthew Wilcox 246579f3aa5bSMatthew Wilcox vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 246679f3aa5bSMatthew Wilcox pfn_t pfn) 246779f3aa5bSMatthew Wilcox { 246828d8b812SLorenzo Stoakes return __vm_insert_mixed(vma, addr, pfn, false); 246979f3aa5bSMatthew Wilcox } 24705d747637SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_mixed); 2471423bad60SNick Piggin 2472ab77dab4SSouptick Joarder /* 2473ab77dab4SSouptick Joarder * If the insertion of PTE failed because someone else already added a 2474ab77dab4SSouptick Joarder * different entry in the mean time, we treat that as success as we assume 2475ab77dab4SSouptick Joarder * the same entry was actually inserted. 2476ab77dab4SSouptick Joarder */ 2477ab77dab4SSouptick Joarder vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, 2478ab77dab4SSouptick Joarder unsigned long addr, pfn_t pfn) 2479b2770da6SRoss Zwisler { 248028d8b812SLorenzo Stoakes return __vm_insert_mixed(vma, addr, pfn, true); 2481b2770da6SRoss Zwisler } 2482ab77dab4SSouptick Joarder EXPORT_SYMBOL(vmf_insert_mixed_mkwrite); 2483b2770da6SRoss Zwisler 2484a145dd41SLinus Torvalds /* 24851da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 24861da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 24871da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 24881da177e4SLinus Torvalds */ 24891da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 24901da177e4SLinus Torvalds unsigned long addr, unsigned long end, 24911da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 24921da177e4SLinus Torvalds { 249390a3e375SMiaohe Lin pte_t *pte, *mapped_pte; 2494c74df32cSHugh Dickins spinlock_t *ptl; 249542e4089cSAndi Kleen int err = 0; 24961da177e4SLinus Torvalds 249790a3e375SMiaohe Lin mapped_pte = pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 24981da177e4SLinus Torvalds if (!pte) 24991da177e4SLinus Torvalds return -ENOMEM; 25006606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 25011da177e4SLinus Torvalds do { 2502c33c7948SRyan Roberts BUG_ON(!pte_none(ptep_get(pte))); 250342e4089cSAndi Kleen if (!pfn_modify_allowed(pfn, prot)) { 250442e4089cSAndi Kleen err = -EACCES; 250542e4089cSAndi Kleen break; 250642e4089cSAndi Kleen } 25077e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 25081da177e4SLinus Torvalds pfn++; 25091da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 25106606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 251190a3e375SMiaohe Lin pte_unmap_unlock(mapped_pte, ptl); 251242e4089cSAndi Kleen return err; 25131da177e4SLinus Torvalds } 25141da177e4SLinus Torvalds 25151da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 25161da177e4SLinus Torvalds unsigned long addr, unsigned long end, 25171da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 25181da177e4SLinus Torvalds { 25191da177e4SLinus Torvalds pmd_t *pmd; 25201da177e4SLinus Torvalds unsigned long next; 252142e4089cSAndi Kleen int err; 25221da177e4SLinus Torvalds 25231da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 25241da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 25251da177e4SLinus Torvalds if (!pmd) 25261da177e4SLinus Torvalds return -ENOMEM; 2527f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 25281da177e4SLinus Torvalds do { 25291da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 253042e4089cSAndi Kleen err = remap_pte_range(mm, pmd, addr, next, 253142e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 253242e4089cSAndi Kleen if (err) 253342e4089cSAndi Kleen return err; 25341da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 25351da177e4SLinus Torvalds return 0; 25361da177e4SLinus Torvalds } 25371da177e4SLinus Torvalds 2538c2febafcSKirill A. Shutemov static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, 25391da177e4SLinus Torvalds unsigned long addr, unsigned long end, 25401da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 25411da177e4SLinus Torvalds { 25421da177e4SLinus Torvalds pud_t *pud; 25431da177e4SLinus Torvalds unsigned long next; 254442e4089cSAndi Kleen int err; 25451da177e4SLinus Torvalds 25461da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 2547c2febafcSKirill A. Shutemov pud = pud_alloc(mm, p4d, addr); 25481da177e4SLinus Torvalds if (!pud) 25491da177e4SLinus Torvalds return -ENOMEM; 25501da177e4SLinus Torvalds do { 25511da177e4SLinus Torvalds next = pud_addr_end(addr, end); 255242e4089cSAndi Kleen err = remap_pmd_range(mm, pud, addr, next, 255342e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 255442e4089cSAndi Kleen if (err) 255542e4089cSAndi Kleen return err; 25561da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 25571da177e4SLinus Torvalds return 0; 25581da177e4SLinus Torvalds } 25591da177e4SLinus Torvalds 2560c2febafcSKirill A. Shutemov static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, 2561c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 2562c2febafcSKirill A. Shutemov unsigned long pfn, pgprot_t prot) 2563c2febafcSKirill A. Shutemov { 2564c2febafcSKirill A. Shutemov p4d_t *p4d; 2565c2febafcSKirill A. Shutemov unsigned long next; 256642e4089cSAndi Kleen int err; 2567c2febafcSKirill A. Shutemov 2568c2febafcSKirill A. Shutemov pfn -= addr >> PAGE_SHIFT; 2569c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 2570c2febafcSKirill A. Shutemov if (!p4d) 2571c2febafcSKirill A. Shutemov return -ENOMEM; 2572c2febafcSKirill A. Shutemov do { 2573c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 257442e4089cSAndi Kleen err = remap_pud_range(mm, p4d, addr, next, 257542e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 257642e4089cSAndi Kleen if (err) 257742e4089cSAndi Kleen return err; 2578c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 2579c2febafcSKirill A. Shutemov return 0; 2580c2febafcSKirill A. Shutemov } 2581c2febafcSKirill A. Shutemov 258274ffa5a3SChristoph Hellwig /* 258374ffa5a3SChristoph Hellwig * Variant of remap_pfn_range that does not call track_pfn_remap. The caller 258474ffa5a3SChristoph Hellwig * must have pre-validated the caching bits of the pgprot_t. 2585bfa5bf6dSRolf Eike Beer */ 258674ffa5a3SChristoph Hellwig int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr, 25871da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 25881da177e4SLinus Torvalds { 25891da177e4SLinus Torvalds pgd_t *pgd; 25901da177e4SLinus Torvalds unsigned long next; 25912d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 25921da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 25931da177e4SLinus Torvalds int err; 25941da177e4SLinus Torvalds 25950c4123e3SAlex Zhang if (WARN_ON_ONCE(!PAGE_ALIGNED(addr))) 25960c4123e3SAlex Zhang return -EINVAL; 25970c4123e3SAlex Zhang 25981da177e4SLinus Torvalds /* 25991da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 26001da177e4SLinus Torvalds * rest of the world about it: 26011da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 26021da177e4SLinus Torvalds * (accesses can have side effects). 26036aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 26046aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 26056aab341eSLinus Torvalds * with them. 2606314e51b9SKonstantin Khlebnikov * VM_DONTEXPAND 2607314e51b9SKonstantin Khlebnikov * Disable vma merging and expanding with mremap(). 2608314e51b9SKonstantin Khlebnikov * VM_DONTDUMP 2609314e51b9SKonstantin Khlebnikov * Omit vma from core dump, even when VM_IO turned off. 2610fb155c16SLinus Torvalds * 2611fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 2612fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 2613fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 2614b3b9c293SKonstantin Khlebnikov * See vm_normal_page() for details. 26151da177e4SLinus Torvalds */ 2616b3b9c293SKonstantin Khlebnikov if (is_cow_mapping(vma->vm_flags)) { 2617b3b9c293SKonstantin Khlebnikov if (addr != vma->vm_start || end != vma->vm_end) 2618b3b9c293SKonstantin Khlebnikov return -EINVAL; 26196aab341eSLinus Torvalds vma->vm_pgoff = pfn; 2620b3b9c293SKonstantin Khlebnikov } 2621b3b9c293SKonstantin Khlebnikov 26221c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP); 26231da177e4SLinus Torvalds 26241da177e4SLinus Torvalds BUG_ON(addr >= end); 26251da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 26261da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 26271da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 26281da177e4SLinus Torvalds do { 26291da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 2630c2febafcSKirill A. Shutemov err = remap_p4d_range(mm, pgd, addr, next, 26311da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 26321da177e4SLinus Torvalds if (err) 263374ffa5a3SChristoph Hellwig return err; 26341da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 26352ab64037Svenkatesh.pallipadi@intel.com 263674ffa5a3SChristoph Hellwig return 0; 263774ffa5a3SChristoph Hellwig } 26382ab64037Svenkatesh.pallipadi@intel.com 263974ffa5a3SChristoph Hellwig /** 264074ffa5a3SChristoph Hellwig * remap_pfn_range - remap kernel memory to userspace 264174ffa5a3SChristoph Hellwig * @vma: user vma to map to 264274ffa5a3SChristoph Hellwig * @addr: target page aligned user address to start at 264374ffa5a3SChristoph Hellwig * @pfn: page frame number of kernel physical memory address 264474ffa5a3SChristoph Hellwig * @size: size of mapping area 264574ffa5a3SChristoph Hellwig * @prot: page protection flags for this mapping 264674ffa5a3SChristoph Hellwig * 264774ffa5a3SChristoph Hellwig * Note: this is only safe if the mm semaphore is held when called. 264874ffa5a3SChristoph Hellwig * 264974ffa5a3SChristoph Hellwig * Return: %0 on success, negative error code otherwise. 265074ffa5a3SChristoph Hellwig */ 265174ffa5a3SChristoph Hellwig int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 265274ffa5a3SChristoph Hellwig unsigned long pfn, unsigned long size, pgprot_t prot) 265374ffa5a3SChristoph Hellwig { 265474ffa5a3SChristoph Hellwig int err; 265574ffa5a3SChristoph Hellwig 265674ffa5a3SChristoph Hellwig err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); 265774ffa5a3SChristoph Hellwig if (err) 265874ffa5a3SChristoph Hellwig return -EINVAL; 265974ffa5a3SChristoph Hellwig 266074ffa5a3SChristoph Hellwig err = remap_pfn_range_notrack(vma, addr, pfn, size, prot); 266174ffa5a3SChristoph Hellwig if (err) 266268f48381SSuren Baghdasaryan untrack_pfn(vma, pfn, PAGE_ALIGN(size), true); 26631da177e4SLinus Torvalds return err; 26641da177e4SLinus Torvalds } 26651da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 26661da177e4SLinus Torvalds 2667b4cbb197SLinus Torvalds /** 2668b4cbb197SLinus Torvalds * vm_iomap_memory - remap memory to userspace 2669b4cbb197SLinus Torvalds * @vma: user vma to map to 2670abd69b9eSWang Wenhu * @start: start of the physical memory to be mapped 2671b4cbb197SLinus Torvalds * @len: size of area 2672b4cbb197SLinus Torvalds * 2673b4cbb197SLinus Torvalds * This is a simplified io_remap_pfn_range() for common driver use. The 2674b4cbb197SLinus Torvalds * driver just needs to give us the physical memory range to be mapped, 2675b4cbb197SLinus Torvalds * we'll figure out the rest from the vma information. 2676b4cbb197SLinus Torvalds * 2677b4cbb197SLinus Torvalds * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get 2678b4cbb197SLinus Torvalds * whatever write-combining details or similar. 2679a862f68aSMike Rapoport * 2680a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 2681b4cbb197SLinus Torvalds */ 2682b4cbb197SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 2683b4cbb197SLinus Torvalds { 2684b4cbb197SLinus Torvalds unsigned long vm_len, pfn, pages; 2685b4cbb197SLinus Torvalds 2686b4cbb197SLinus Torvalds /* Check that the physical memory area passed in looks valid */ 2687b4cbb197SLinus Torvalds if (start + len < start) 2688b4cbb197SLinus Torvalds return -EINVAL; 2689b4cbb197SLinus Torvalds /* 2690b4cbb197SLinus Torvalds * You *really* shouldn't map things that aren't page-aligned, 2691b4cbb197SLinus Torvalds * but we've historically allowed it because IO memory might 2692b4cbb197SLinus Torvalds * just have smaller alignment. 2693b4cbb197SLinus Torvalds */ 2694b4cbb197SLinus Torvalds len += start & ~PAGE_MASK; 2695b4cbb197SLinus Torvalds pfn = start >> PAGE_SHIFT; 2696b4cbb197SLinus Torvalds pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; 2697b4cbb197SLinus Torvalds if (pfn + pages < pfn) 2698b4cbb197SLinus Torvalds return -EINVAL; 2699b4cbb197SLinus Torvalds 2700b4cbb197SLinus Torvalds /* We start the mapping 'vm_pgoff' pages into the area */ 2701b4cbb197SLinus Torvalds if (vma->vm_pgoff > pages) 2702b4cbb197SLinus Torvalds return -EINVAL; 2703b4cbb197SLinus Torvalds pfn += vma->vm_pgoff; 2704b4cbb197SLinus Torvalds pages -= vma->vm_pgoff; 2705b4cbb197SLinus Torvalds 2706b4cbb197SLinus Torvalds /* Can we fit all of the mapping? */ 2707b4cbb197SLinus Torvalds vm_len = vma->vm_end - vma->vm_start; 2708b4cbb197SLinus Torvalds if (vm_len >> PAGE_SHIFT > pages) 2709b4cbb197SLinus Torvalds return -EINVAL; 2710b4cbb197SLinus Torvalds 2711b4cbb197SLinus Torvalds /* Ok, let it rip */ 2712b4cbb197SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 2713b4cbb197SLinus Torvalds } 2714b4cbb197SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 2715b4cbb197SLinus Torvalds 2716aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 2717aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2718e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2719e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2720aee16b3cSJeremy Fitzhardinge { 27218abb50c7SMiaohe Lin pte_t *pte, *mapped_pte; 2722be1db475SDaniel Axtens int err = 0; 27233f649ab7SKees Cook spinlock_t *ptl; 2724aee16b3cSJeremy Fitzhardinge 2725be1db475SDaniel Axtens if (create) { 27268abb50c7SMiaohe Lin mapped_pte = pte = (mm == &init_mm) ? 2727e80d3909SJoerg Roedel pte_alloc_kernel_track(pmd, addr, mask) : 2728aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 2729aee16b3cSJeremy Fitzhardinge if (!pte) 2730aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2731be1db475SDaniel Axtens } else { 27328abb50c7SMiaohe Lin mapped_pte = pte = (mm == &init_mm) ? 2733be1db475SDaniel Axtens pte_offset_kernel(pmd, addr) : 2734be1db475SDaniel Axtens pte_offset_map_lock(mm, pmd, addr, &ptl); 27353db82b93SHugh Dickins if (!pte) 27363db82b93SHugh Dickins return -EINVAL; 2737be1db475SDaniel Axtens } 2738aee16b3cSJeremy Fitzhardinge 273938e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 274038e0edb1SJeremy Fitzhardinge 2741eeb4a05fSChristoph Hellwig if (fn) { 2742aee16b3cSJeremy Fitzhardinge do { 2743c33c7948SRyan Roberts if (create || !pte_none(ptep_get(pte))) { 27448b1e0f81SAnshuman Khandual err = fn(pte++, addr, data); 2745aee16b3cSJeremy Fitzhardinge if (err) 2746aee16b3cSJeremy Fitzhardinge break; 2747be1db475SDaniel Axtens } 2748c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 2749eeb4a05fSChristoph Hellwig } 2750e80d3909SJoerg Roedel *mask |= PGTBL_PTE_MODIFIED; 2751aee16b3cSJeremy Fitzhardinge 275238e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 275338e0edb1SJeremy Fitzhardinge 2754aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 27558abb50c7SMiaohe Lin pte_unmap_unlock(mapped_pte, ptl); 2756aee16b3cSJeremy Fitzhardinge return err; 2757aee16b3cSJeremy Fitzhardinge } 2758aee16b3cSJeremy Fitzhardinge 2759aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 2760aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2761e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2762e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2763aee16b3cSJeremy Fitzhardinge { 2764aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 2765aee16b3cSJeremy Fitzhardinge unsigned long next; 2766be1db475SDaniel Axtens int err = 0; 2767aee16b3cSJeremy Fitzhardinge 2768ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 2769ceb86879SAndi Kleen 2770be1db475SDaniel Axtens if (create) { 2771e80d3909SJoerg Roedel pmd = pmd_alloc_track(mm, pud, addr, mask); 2772aee16b3cSJeremy Fitzhardinge if (!pmd) 2773aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2774be1db475SDaniel Axtens } else { 2775be1db475SDaniel Axtens pmd = pmd_offset(pud, addr); 2776be1db475SDaniel Axtens } 2777aee16b3cSJeremy Fitzhardinge do { 2778aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 27790c95cba4SNicholas Piggin if (pmd_none(*pmd) && !create) 27800c95cba4SNicholas Piggin continue; 27810c95cba4SNicholas Piggin if (WARN_ON_ONCE(pmd_leaf(*pmd))) 27820c95cba4SNicholas Piggin return -EINVAL; 27830c95cba4SNicholas Piggin if (!pmd_none(*pmd) && WARN_ON_ONCE(pmd_bad(*pmd))) { 27840c95cba4SNicholas Piggin if (!create) 27850c95cba4SNicholas Piggin continue; 27860c95cba4SNicholas Piggin pmd_clear_bad(pmd); 27870c95cba4SNicholas Piggin } 27880c95cba4SNicholas Piggin err = apply_to_pte_range(mm, pmd, addr, next, 27890c95cba4SNicholas Piggin fn, data, create, mask); 2790aee16b3cSJeremy Fitzhardinge if (err) 2791aee16b3cSJeremy Fitzhardinge break; 2792aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 27930c95cba4SNicholas Piggin 2794aee16b3cSJeremy Fitzhardinge return err; 2795aee16b3cSJeremy Fitzhardinge } 2796aee16b3cSJeremy Fitzhardinge 2797c2febafcSKirill A. Shutemov static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, 2798aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2799e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2800e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2801aee16b3cSJeremy Fitzhardinge { 2802aee16b3cSJeremy Fitzhardinge pud_t *pud; 2803aee16b3cSJeremy Fitzhardinge unsigned long next; 2804be1db475SDaniel Axtens int err = 0; 2805aee16b3cSJeremy Fitzhardinge 2806be1db475SDaniel Axtens if (create) { 2807e80d3909SJoerg Roedel pud = pud_alloc_track(mm, p4d, addr, mask); 2808aee16b3cSJeremy Fitzhardinge if (!pud) 2809aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2810be1db475SDaniel Axtens } else { 2811be1db475SDaniel Axtens pud = pud_offset(p4d, addr); 2812be1db475SDaniel Axtens } 2813aee16b3cSJeremy Fitzhardinge do { 2814aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 28150c95cba4SNicholas Piggin if (pud_none(*pud) && !create) 28160c95cba4SNicholas Piggin continue; 28170c95cba4SNicholas Piggin if (WARN_ON_ONCE(pud_leaf(*pud))) 28180c95cba4SNicholas Piggin return -EINVAL; 28190c95cba4SNicholas Piggin if (!pud_none(*pud) && WARN_ON_ONCE(pud_bad(*pud))) { 28200c95cba4SNicholas Piggin if (!create) 28210c95cba4SNicholas Piggin continue; 28220c95cba4SNicholas Piggin pud_clear_bad(pud); 28230c95cba4SNicholas Piggin } 28240c95cba4SNicholas Piggin err = apply_to_pmd_range(mm, pud, addr, next, 28250c95cba4SNicholas Piggin fn, data, create, mask); 2826aee16b3cSJeremy Fitzhardinge if (err) 2827aee16b3cSJeremy Fitzhardinge break; 2828aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 28290c95cba4SNicholas Piggin 2830aee16b3cSJeremy Fitzhardinge return err; 2831aee16b3cSJeremy Fitzhardinge } 2832aee16b3cSJeremy Fitzhardinge 2833c2febafcSKirill A. Shutemov static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, 2834c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 2835e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2836e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2837c2febafcSKirill A. Shutemov { 2838c2febafcSKirill A. Shutemov p4d_t *p4d; 2839c2febafcSKirill A. Shutemov unsigned long next; 2840be1db475SDaniel Axtens int err = 0; 2841c2febafcSKirill A. Shutemov 2842be1db475SDaniel Axtens if (create) { 2843e80d3909SJoerg Roedel p4d = p4d_alloc_track(mm, pgd, addr, mask); 2844c2febafcSKirill A. Shutemov if (!p4d) 2845c2febafcSKirill A. Shutemov return -ENOMEM; 2846be1db475SDaniel Axtens } else { 2847be1db475SDaniel Axtens p4d = p4d_offset(pgd, addr); 2848be1db475SDaniel Axtens } 2849c2febafcSKirill A. Shutemov do { 2850c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 28510c95cba4SNicholas Piggin if (p4d_none(*p4d) && !create) 28520c95cba4SNicholas Piggin continue; 28530c95cba4SNicholas Piggin if (WARN_ON_ONCE(p4d_leaf(*p4d))) 28540c95cba4SNicholas Piggin return -EINVAL; 28550c95cba4SNicholas Piggin if (!p4d_none(*p4d) && WARN_ON_ONCE(p4d_bad(*p4d))) { 28560c95cba4SNicholas Piggin if (!create) 28570c95cba4SNicholas Piggin continue; 28580c95cba4SNicholas Piggin p4d_clear_bad(p4d); 28590c95cba4SNicholas Piggin } 28600c95cba4SNicholas Piggin err = apply_to_pud_range(mm, p4d, addr, next, 28610c95cba4SNicholas Piggin fn, data, create, mask); 2862c2febafcSKirill A. Shutemov if (err) 2863c2febafcSKirill A. Shutemov break; 2864c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 28650c95cba4SNicholas Piggin 2866c2febafcSKirill A. Shutemov return err; 2867c2febafcSKirill A. Shutemov } 2868c2febafcSKirill A. Shutemov 2869be1db475SDaniel Axtens static int __apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2870be1db475SDaniel Axtens unsigned long size, pte_fn_t fn, 2871be1db475SDaniel Axtens void *data, bool create) 2872be1db475SDaniel Axtens { 2873be1db475SDaniel Axtens pgd_t *pgd; 2874e80d3909SJoerg Roedel unsigned long start = addr, next; 2875be1db475SDaniel Axtens unsigned long end = addr + size; 2876e80d3909SJoerg Roedel pgtbl_mod_mask mask = 0; 2877be1db475SDaniel Axtens int err = 0; 2878be1db475SDaniel Axtens 2879be1db475SDaniel Axtens if (WARN_ON(addr >= end)) 2880be1db475SDaniel Axtens return -EINVAL; 2881be1db475SDaniel Axtens 2882be1db475SDaniel Axtens pgd = pgd_offset(mm, addr); 2883be1db475SDaniel Axtens do { 2884be1db475SDaniel Axtens next = pgd_addr_end(addr, end); 28850c95cba4SNicholas Piggin if (pgd_none(*pgd) && !create) 2886be1db475SDaniel Axtens continue; 28870c95cba4SNicholas Piggin if (WARN_ON_ONCE(pgd_leaf(*pgd))) 28880c95cba4SNicholas Piggin return -EINVAL; 28890c95cba4SNicholas Piggin if (!pgd_none(*pgd) && WARN_ON_ONCE(pgd_bad(*pgd))) { 28900c95cba4SNicholas Piggin if (!create) 28910c95cba4SNicholas Piggin continue; 28920c95cba4SNicholas Piggin pgd_clear_bad(pgd); 28930c95cba4SNicholas Piggin } 28940c95cba4SNicholas Piggin err = apply_to_p4d_range(mm, pgd, addr, next, 28950c95cba4SNicholas Piggin fn, data, create, &mask); 2896be1db475SDaniel Axtens if (err) 2897be1db475SDaniel Axtens break; 2898be1db475SDaniel Axtens } while (pgd++, addr = next, addr != end); 2899be1db475SDaniel Axtens 2900e80d3909SJoerg Roedel if (mask & ARCH_PAGE_TABLE_SYNC_MASK) 2901e80d3909SJoerg Roedel arch_sync_kernel_mappings(start, start + size); 2902e80d3909SJoerg Roedel 2903be1db475SDaniel Axtens return err; 2904be1db475SDaniel Axtens } 2905be1db475SDaniel Axtens 2906aee16b3cSJeremy Fitzhardinge /* 2907aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 2908aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 2909aee16b3cSJeremy Fitzhardinge */ 2910aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2911aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 2912aee16b3cSJeremy Fitzhardinge { 2913be1db475SDaniel Axtens return __apply_to_page_range(mm, addr, size, fn, data, true); 2914aee16b3cSJeremy Fitzhardinge } 2915aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 2916aee16b3cSJeremy Fitzhardinge 29171da177e4SLinus Torvalds /* 2918be1db475SDaniel Axtens * Scan a region of virtual memory, calling a provided function on 2919be1db475SDaniel Axtens * each leaf page table where it exists. 2920be1db475SDaniel Axtens * 2921be1db475SDaniel Axtens * Unlike apply_to_page_range, this does _not_ fill in page tables 2922be1db475SDaniel Axtens * where they are absent. 2923be1db475SDaniel Axtens */ 2924be1db475SDaniel Axtens int apply_to_existing_page_range(struct mm_struct *mm, unsigned long addr, 2925be1db475SDaniel Axtens unsigned long size, pte_fn_t fn, void *data) 2926be1db475SDaniel Axtens { 2927be1db475SDaniel Axtens return __apply_to_page_range(mm, addr, size, fn, data, false); 2928be1db475SDaniel Axtens } 2929be1db475SDaniel Axtens EXPORT_SYMBOL_GPL(apply_to_existing_page_range); 2930be1db475SDaniel Axtens 2931be1db475SDaniel Axtens /* 29329b4bdd2fSKirill A. Shutemov * handle_pte_fault chooses page fault handler according to an entry which was 29339b4bdd2fSKirill A. Shutemov * read non-atomically. Before making any commitment, on those architectures 29349b4bdd2fSKirill A. Shutemov * or configurations (e.g. i386 with PAE) which might give a mix of unmatched 29359b4bdd2fSKirill A. Shutemov * parts, do_swap_page must check under lock before unmapping the pte and 29369b4bdd2fSKirill A. Shutemov * proceeding (but do_wp_page is only called after already making such a check; 2937a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 29388f4e2101SHugh Dickins */ 29392ca99358SPeter Xu static inline int pte_unmap_same(struct vm_fault *vmf) 29408f4e2101SHugh Dickins { 29418f4e2101SHugh Dickins int same = 1; 2942923717cbSThomas Gleixner #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPTION) 29438f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 2944c7ad0880SHugh Dickins spin_lock(vmf->ptl); 2945c33c7948SRyan Roberts same = pte_same(ptep_get(vmf->pte), vmf->orig_pte); 2946c7ad0880SHugh Dickins spin_unlock(vmf->ptl); 29478f4e2101SHugh Dickins } 29488f4e2101SHugh Dickins #endif 29492ca99358SPeter Xu pte_unmap(vmf->pte); 29502ca99358SPeter Xu vmf->pte = NULL; 29518f4e2101SHugh Dickins return same; 29528f4e2101SHugh Dickins } 29538f4e2101SHugh Dickins 2954a873dfe1STony Luck /* 2955a873dfe1STony Luck * Return: 2956a873dfe1STony Luck * 0: copied succeeded 2957a873dfe1STony Luck * -EHWPOISON: copy failed due to hwpoison in source page 2958a873dfe1STony Luck * -EAGAIN: copied failed (some other reason) 2959a873dfe1STony Luck */ 2960a873dfe1STony Luck static inline int __wp_page_copy_user(struct page *dst, struct page *src, 296183d116c5SJia He struct vm_fault *vmf) 29626aab341eSLinus Torvalds { 2963a873dfe1STony Luck int ret; 296483d116c5SJia He void *kaddr; 296583d116c5SJia He void __user *uaddr; 296683d116c5SJia He struct vm_area_struct *vma = vmf->vma; 296783d116c5SJia He struct mm_struct *mm = vma->vm_mm; 296883d116c5SJia He unsigned long addr = vmf->address; 296983d116c5SJia He 297083d116c5SJia He if (likely(src)) { 2971d302c239STony Luck if (copy_mc_user_highpage(dst, src, addr, vma)) { 2972d302c239STony Luck memory_failure_queue(page_to_pfn(src), 0); 2973a873dfe1STony Luck return -EHWPOISON; 2974d302c239STony Luck } 2975a873dfe1STony Luck return 0; 297683d116c5SJia He } 297783d116c5SJia He 29786aab341eSLinus Torvalds /* 29796aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 29806aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 29816aab341eSLinus Torvalds * just copying from the original user address. If that 29826aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 29836aab341eSLinus Torvalds */ 298424d2613aSFabio M. De Francesco kaddr = kmap_local_page(dst); 298524d2613aSFabio M. De Francesco pagefault_disable(); 298683d116c5SJia He uaddr = (void __user *)(addr & PAGE_MASK); 298783d116c5SJia He 298883d116c5SJia He /* 298983d116c5SJia He * On architectures with software "accessed" bits, we would 299083d116c5SJia He * take a double page fault, so mark it accessed here. 299183d116c5SJia He */ 29923db82b93SHugh Dickins vmf->pte = NULL; 2993e1fd09e3SYu Zhao if (!arch_has_hw_pte_young() && !pte_young(vmf->orig_pte)) { 299483d116c5SJia He pte_t entry; 299583d116c5SJia He 299683d116c5SJia He vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); 2997c33c7948SRyan Roberts if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 299883d116c5SJia He /* 299983d116c5SJia He * Other thread has already handled the fault 30007df67697SBibo Mao * and update local tlb only 300183d116c5SJia He */ 3002a92cbb82SHugh Dickins if (vmf->pte) 30037df67697SBibo Mao update_mmu_tlb(vma, addr, vmf->pte); 3004a873dfe1STony Luck ret = -EAGAIN; 300583d116c5SJia He goto pte_unlock; 300683d116c5SJia He } 300783d116c5SJia He 300883d116c5SJia He entry = pte_mkyoung(vmf->orig_pte); 300983d116c5SJia He if (ptep_set_access_flags(vma, addr, vmf->pte, entry, 0)) 30105003a2bdSMatthew Wilcox (Oracle) update_mmu_cache_range(vmf, vma, addr, vmf->pte, 1); 301183d116c5SJia He } 30125d2a2dbbSLinus Torvalds 30135d2a2dbbSLinus Torvalds /* 30145d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 30155d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 30165d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 30175d2a2dbbSLinus Torvalds * zeroes. 30185d2a2dbbSLinus Torvalds */ 301983d116c5SJia He if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { 30203db82b93SHugh Dickins if (vmf->pte) 3021c3e5ea6eSKirill A. Shutemov goto warn; 3022c3e5ea6eSKirill A. Shutemov 3023c3e5ea6eSKirill A. Shutemov /* Re-validate under PTL if the page is still mapped */ 3024c3e5ea6eSKirill A. Shutemov vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); 3025c33c7948SRyan Roberts if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 30267df67697SBibo Mao /* The PTE changed under us, update local tlb */ 3027a92cbb82SHugh Dickins if (vmf->pte) 30287df67697SBibo Mao update_mmu_tlb(vma, addr, vmf->pte); 3029a873dfe1STony Luck ret = -EAGAIN; 3030c3e5ea6eSKirill A. Shutemov goto pte_unlock; 3031c3e5ea6eSKirill A. Shutemov } 3032c3e5ea6eSKirill A. Shutemov 3033c3e5ea6eSKirill A. Shutemov /* 3034985ba004SEthon Paul * The same page can be mapped back since last copy attempt. 3035c3e5ea6eSKirill A. Shutemov * Try to copy again under PTL. 3036c3e5ea6eSKirill A. Shutemov */ 3037c3e5ea6eSKirill A. Shutemov if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { 303883d116c5SJia He /* 303983d116c5SJia He * Give a warn in case there can be some obscure 304083d116c5SJia He * use-case 304183d116c5SJia He */ 3042c3e5ea6eSKirill A. Shutemov warn: 304383d116c5SJia He WARN_ON_ONCE(1); 30443ecb01dfSJan Beulich clear_page(kaddr); 304583d116c5SJia He } 3046c3e5ea6eSKirill A. Shutemov } 304783d116c5SJia He 3048a873dfe1STony Luck ret = 0; 304983d116c5SJia He 305083d116c5SJia He pte_unlock: 30513db82b93SHugh Dickins if (vmf->pte) 305283d116c5SJia He pte_unmap_unlock(vmf->pte, vmf->ptl); 305324d2613aSFabio M. De Francesco pagefault_enable(); 305424d2613aSFabio M. De Francesco kunmap_local(kaddr); 3055c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 305683d116c5SJia He 305783d116c5SJia He return ret; 30586aab341eSLinus Torvalds } 30596aab341eSLinus Torvalds 3060c20cd45eSMichal Hocko static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) 3061c20cd45eSMichal Hocko { 3062c20cd45eSMichal Hocko struct file *vm_file = vma->vm_file; 3063c20cd45eSMichal Hocko 3064c20cd45eSMichal Hocko if (vm_file) 3065c20cd45eSMichal Hocko return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; 3066c20cd45eSMichal Hocko 3067c20cd45eSMichal Hocko /* 3068c20cd45eSMichal Hocko * Special mappings (e.g. VDSO) do not have any file so fake 3069c20cd45eSMichal Hocko * a default GFP_KERNEL for them. 3070c20cd45eSMichal Hocko */ 3071c20cd45eSMichal Hocko return GFP_KERNEL; 3072c20cd45eSMichal Hocko } 3073c20cd45eSMichal Hocko 30741da177e4SLinus Torvalds /* 3075fb09a464SKirill A. Shutemov * Notify the address space that the page is about to become writable so that 3076fb09a464SKirill A. Shutemov * it can prohibit this or wait for the page to get into an appropriate state. 3077fb09a464SKirill A. Shutemov * 3078fb09a464SKirill A. Shutemov * We do this without the lock held, so that it can sleep if it needs to. 3079fb09a464SKirill A. Shutemov */ 308086aa6998SSidhartha Kumar static vm_fault_t do_page_mkwrite(struct vm_fault *vmf, struct folio *folio) 3081fb09a464SKirill A. Shutemov { 30822b740303SSouptick Joarder vm_fault_t ret; 308338b8cb7fSJan Kara unsigned int old_flags = vmf->flags; 3084fb09a464SKirill A. Shutemov 308538b8cb7fSJan Kara vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 3086fb09a464SKirill A. Shutemov 3087dc617f29SDarrick J. Wong if (vmf->vma->vm_file && 3088dc617f29SDarrick J. Wong IS_SWAPFILE(vmf->vma->vm_file->f_mapping->host)) 3089dc617f29SDarrick J. Wong return VM_FAULT_SIGBUS; 3090dc617f29SDarrick J. Wong 309111bac800SDave Jiang ret = vmf->vma->vm_ops->page_mkwrite(vmf); 309238b8cb7fSJan Kara /* Restore original flags so that caller is not surprised */ 309338b8cb7fSJan Kara vmf->flags = old_flags; 3094fb09a464SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 3095fb09a464SKirill A. Shutemov return ret; 3096fb09a464SKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) { 30973d243659SSidhartha Kumar folio_lock(folio); 30983d243659SSidhartha Kumar if (!folio->mapping) { 30993d243659SSidhartha Kumar folio_unlock(folio); 3100fb09a464SKirill A. Shutemov return 0; /* retry */ 3101fb09a464SKirill A. Shutemov } 3102fb09a464SKirill A. Shutemov ret |= VM_FAULT_LOCKED; 3103fb09a464SKirill A. Shutemov } else 31043d243659SSidhartha Kumar VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); 3105fb09a464SKirill A. Shutemov return ret; 3106fb09a464SKirill A. Shutemov } 3107fb09a464SKirill A. Shutemov 3108fb09a464SKirill A. Shutemov /* 310997ba0c2bSJan Kara * Handle dirtying of a page in shared file mapping on a write fault. 31104e047f89SShachar Raindel * 311197ba0c2bSJan Kara * The function expects the page to be locked and unlocks it. 31124e047f89SShachar Raindel */ 311389b15332SJohannes Weiner static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf) 31144e047f89SShachar Raindel { 311589b15332SJohannes Weiner struct vm_area_struct *vma = vmf->vma; 31164e047f89SShachar Raindel struct address_space *mapping; 311715b4919aSZhangPeng struct folio *folio = page_folio(vmf->page); 311897ba0c2bSJan Kara bool dirtied; 311997ba0c2bSJan Kara bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; 31204e047f89SShachar Raindel 312115b4919aSZhangPeng dirtied = folio_mark_dirty(folio); 312215b4919aSZhangPeng VM_BUG_ON_FOLIO(folio_test_anon(folio), folio); 312397ba0c2bSJan Kara /* 312415b4919aSZhangPeng * Take a local copy of the address_space - folio.mapping may be zeroed 312515b4919aSZhangPeng * by truncate after folio_unlock(). The address_space itself remains 312615b4919aSZhangPeng * pinned by vma->vm_file's reference. We rely on folio_unlock()'s 312797ba0c2bSJan Kara * release semantics to prevent the compiler from undoing this copying. 312897ba0c2bSJan Kara */ 312915b4919aSZhangPeng mapping = folio_raw_mapping(folio); 313015b4919aSZhangPeng folio_unlock(folio); 31314e047f89SShachar Raindel 31324e047f89SShachar Raindel if (!page_mkwrite) 31334e047f89SShachar Raindel file_update_time(vma->vm_file); 313489b15332SJohannes Weiner 313589b15332SJohannes Weiner /* 313689b15332SJohannes Weiner * Throttle page dirtying rate down to writeback speed. 313789b15332SJohannes Weiner * 313889b15332SJohannes Weiner * mapping may be NULL here because some device drivers do not 313989b15332SJohannes Weiner * set page.mapping but still dirty their pages 314089b15332SJohannes Weiner * 3141c1e8d7c6SMichel Lespinasse * Drop the mmap_lock before waiting on IO, if we can. The file 314289b15332SJohannes Weiner * is pinning the mapping, as per above. 314389b15332SJohannes Weiner */ 314489b15332SJohannes Weiner if ((dirtied || page_mkwrite) && mapping) { 314589b15332SJohannes Weiner struct file *fpin; 314689b15332SJohannes Weiner 314789b15332SJohannes Weiner fpin = maybe_unlock_mmap_for_io(vmf, NULL); 314889b15332SJohannes Weiner balance_dirty_pages_ratelimited(mapping); 314989b15332SJohannes Weiner if (fpin) { 315089b15332SJohannes Weiner fput(fpin); 3151d9272525SPeter Xu return VM_FAULT_COMPLETED; 315289b15332SJohannes Weiner } 315389b15332SJohannes Weiner } 315489b15332SJohannes Weiner 315589b15332SJohannes Weiner return 0; 31564e047f89SShachar Raindel } 31574e047f89SShachar Raindel 315897ba0c2bSJan Kara /* 31594e047f89SShachar Raindel * Handle write page faults for pages that can be reused in the current vma 31604e047f89SShachar Raindel * 31614e047f89SShachar Raindel * This can happen either due to the mapping being with the VM_SHARED flag, 31624e047f89SShachar Raindel * or due to us being the last reference standing to the page. In either 31634e047f89SShachar Raindel * case, all we need to do here is to mark the page as writable and update 31644e047f89SShachar Raindel * any related book-keeping. 31654e047f89SShachar Raindel */ 3166a86bc96bSKefeng Wang static inline void wp_page_reuse(struct vm_fault *vmf, struct folio *folio) 316782b0f8c3SJan Kara __releases(vmf->ptl) 31684e047f89SShachar Raindel { 316982b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 31704e047f89SShachar Raindel pte_t entry; 31716c287605SDavid Hildenbrand 3172c89357e2SDavid Hildenbrand VM_BUG_ON(!(vmf->flags & FAULT_FLAG_WRITE)); 31736c287605SDavid Hildenbrand 3174c2c3b514SKefeng Wang if (folio) { 3175c2c3b514SKefeng Wang VM_BUG_ON(folio_test_anon(folio) && 3176c2c3b514SKefeng Wang !PageAnonExclusive(vmf->page)); 31774e047f89SShachar Raindel /* 3178c2c3b514SKefeng Wang * Clear the folio's cpupid information as the existing 31794e047f89SShachar Raindel * information potentially belongs to a now completely 31804e047f89SShachar Raindel * unrelated process. 31814e047f89SShachar Raindel */ 3182c2c3b514SKefeng Wang folio_xchg_last_cpupid(folio, (1 << LAST_CPUPID_SHIFT) - 1); 3183c2c3b514SKefeng Wang } 31844e047f89SShachar Raindel 31852994302bSJan Kara flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); 31862994302bSJan Kara entry = pte_mkyoung(vmf->orig_pte); 31874e047f89SShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 318882b0f8c3SJan Kara if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) 31895003a2bdSMatthew Wilcox (Oracle) update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1); 319082b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3191798a6b87SPeter Xu count_vm_event(PGREUSE); 31924e047f89SShachar Raindel } 31934e047f89SShachar Raindel 31944ed43798SMatthew Wilcox (Oracle) /* 31954ed43798SMatthew Wilcox (Oracle) * We could add a bitflag somewhere, but for now, we know that all 31964ed43798SMatthew Wilcox (Oracle) * vm_ops that have a ->map_pages have been audited and don't need 31974ed43798SMatthew Wilcox (Oracle) * the mmap_lock to be held. 31984ed43798SMatthew Wilcox (Oracle) */ 31994ed43798SMatthew Wilcox (Oracle) static inline vm_fault_t vmf_can_call_fault(const struct vm_fault *vmf) 32004ed43798SMatthew Wilcox (Oracle) { 32014ed43798SMatthew Wilcox (Oracle) struct vm_area_struct *vma = vmf->vma; 32024ed43798SMatthew Wilcox (Oracle) 32034ed43798SMatthew Wilcox (Oracle) if (vma->vm_ops->map_pages || !(vmf->flags & FAULT_FLAG_VMA_LOCK)) 32044ed43798SMatthew Wilcox (Oracle) return 0; 32054ed43798SMatthew Wilcox (Oracle) vma_end_read(vma); 32064ed43798SMatthew Wilcox (Oracle) return VM_FAULT_RETRY; 32074ed43798SMatthew Wilcox (Oracle) } 32084ed43798SMatthew Wilcox (Oracle) 3209997f0ecbSVishal Moola (Oracle) vm_fault_t vmf_anon_prepare(struct vm_fault *vmf) 3210164b06f2SMatthew Wilcox (Oracle) { 3211164b06f2SMatthew Wilcox (Oracle) struct vm_area_struct *vma = vmf->vma; 3212164b06f2SMatthew Wilcox (Oracle) 3213164b06f2SMatthew Wilcox (Oracle) if (likely(vma->anon_vma)) 3214164b06f2SMatthew Wilcox (Oracle) return 0; 3215164b06f2SMatthew Wilcox (Oracle) if (vmf->flags & FAULT_FLAG_VMA_LOCK) { 3216164b06f2SMatthew Wilcox (Oracle) vma_end_read(vma); 3217164b06f2SMatthew Wilcox (Oracle) return VM_FAULT_RETRY; 3218164b06f2SMatthew Wilcox (Oracle) } 3219164b06f2SMatthew Wilcox (Oracle) if (__anon_vma_prepare(vma)) 3220164b06f2SMatthew Wilcox (Oracle) return VM_FAULT_OOM; 3221164b06f2SMatthew Wilcox (Oracle) return 0; 3222164b06f2SMatthew Wilcox (Oracle) } 3223164b06f2SMatthew Wilcox (Oracle) 32244e047f89SShachar Raindel /* 3225c89357e2SDavid Hildenbrand * Handle the case of a page which we actually need to copy to a new page, 3226c89357e2SDavid Hildenbrand * either due to COW or unsharing. 32272f38ab2cSShachar Raindel * 3228c1e8d7c6SMichel Lespinasse * Called with mmap_lock locked and the old page referenced, but 32292f38ab2cSShachar Raindel * without the ptl held. 32302f38ab2cSShachar Raindel * 32312f38ab2cSShachar Raindel * High level logic flow: 32322f38ab2cSShachar Raindel * 32332f38ab2cSShachar Raindel * - Allocate a page, copy the content of the old page to the new one. 32342f38ab2cSShachar Raindel * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. 32352f38ab2cSShachar Raindel * - Take the PTL. If the pte changed, bail out and release the allocated page 32362f38ab2cSShachar Raindel * - If the pte is still the way we remember it, update the page table and all 32372f38ab2cSShachar Raindel * relevant references. This includes dropping the reference the page-table 32382f38ab2cSShachar Raindel * held to the old page, as well as updating the rmap. 32392f38ab2cSShachar Raindel * - In any case, unlock the PTL and drop the reference we took to the old page. 32402f38ab2cSShachar Raindel */ 32412b740303SSouptick Joarder static vm_fault_t wp_page_copy(struct vm_fault *vmf) 32422f38ab2cSShachar Raindel { 3243c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 324482b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3245bae473a4SKirill A. Shutemov struct mm_struct *mm = vma->vm_mm; 324628d41a48SMatthew Wilcox (Oracle) struct folio *old_folio = NULL; 324728d41a48SMatthew Wilcox (Oracle) struct folio *new_folio = NULL; 32482f38ab2cSShachar Raindel pte_t entry; 32492f38ab2cSShachar Raindel int page_copied = 0; 3250ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 3251164b06f2SMatthew Wilcox (Oracle) vm_fault_t ret; 3252cf503cc6SKefeng Wang bool pfn_is_zero; 32532f38ab2cSShachar Raindel 3254662ce1dcSYang Yang delayacct_wpcopy_start(); 3255662ce1dcSYang Yang 325628d41a48SMatthew Wilcox (Oracle) if (vmf->page) 325728d41a48SMatthew Wilcox (Oracle) old_folio = page_folio(vmf->page); 3258164b06f2SMatthew Wilcox (Oracle) ret = vmf_anon_prepare(vmf); 3259164b06f2SMatthew Wilcox (Oracle) if (unlikely(ret)) 3260164b06f2SMatthew Wilcox (Oracle) goto out; 32612f38ab2cSShachar Raindel 3262cf503cc6SKefeng Wang pfn_is_zero = is_zero_pfn(pte_pfn(vmf->orig_pte)); 3263cf503cc6SKefeng Wang new_folio = folio_prealloc(mm, vma, vmf->address, pfn_is_zero); 32646bc56a4dSMatthew Wilcox (Oracle) if (!new_folio) 32652f38ab2cSShachar Raindel goto oom; 3266cf503cc6SKefeng Wang 3267cf503cc6SKefeng Wang if (!pfn_is_zero) { 3268164b06f2SMatthew Wilcox (Oracle) int err; 326983d116c5SJia He 3270164b06f2SMatthew Wilcox (Oracle) err = __wp_page_copy_user(&new_folio->page, vmf->page, vmf); 3271164b06f2SMatthew Wilcox (Oracle) if (err) { 327283d116c5SJia He /* 327383d116c5SJia He * COW failed, if the fault was solved by other, 327483d116c5SJia He * it's fine. If not, userspace would re-fault on 327583d116c5SJia He * the same address and we will handle the fault 327683d116c5SJia He * from the second attempt. 3277a873dfe1STony Luck * The -EHWPOISON case will not be retried. 327883d116c5SJia He */ 327928d41a48SMatthew Wilcox (Oracle) folio_put(new_folio); 328028d41a48SMatthew Wilcox (Oracle) if (old_folio) 328128d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 3282662ce1dcSYang Yang 3283662ce1dcSYang Yang delayacct_wpcopy_end(); 3284164b06f2SMatthew Wilcox (Oracle) return err == -EHWPOISON ? VM_FAULT_HWPOISON : 0; 328583d116c5SJia He } 328628d41a48SMatthew Wilcox (Oracle) kmsan_copy_page_meta(&new_folio->page, vmf->page); 32872f38ab2cSShachar Raindel } 32882f38ab2cSShachar Raindel 328928d41a48SMatthew Wilcox (Oracle) __folio_mark_uptodate(new_folio); 3290eb3c24f3SMel Gorman 32917d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, mm, 32926f4f13e8SJérôme Glisse vmf->address & PAGE_MASK, 3293ac46d4f3SJérôme Glisse (vmf->address & PAGE_MASK) + PAGE_SIZE); 3294ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 32952f38ab2cSShachar Raindel 32962f38ab2cSShachar Raindel /* 32972f38ab2cSShachar Raindel * Re-check the pte - we dropped the lock 32982f38ab2cSShachar Raindel */ 329982b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); 3300c33c7948SRyan Roberts if (likely(vmf->pte && pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 330128d41a48SMatthew Wilcox (Oracle) if (old_folio) { 330228d41a48SMatthew Wilcox (Oracle) if (!folio_test_anon(old_folio)) { 33036b27cc6cSKefeng Wang dec_mm_counter(mm, mm_counter_file(old_folio)); 3304f1a79412SShakeel Butt inc_mm_counter(mm, MM_ANONPAGES); 33052f38ab2cSShachar Raindel } 33062f38ab2cSShachar Raindel } else { 33076080d19fSxu xin ksm_might_unmap_zero_page(mm, vmf->orig_pte); 3308f1a79412SShakeel Butt inc_mm_counter(mm, MM_ANONPAGES); 33092f38ab2cSShachar Raindel } 33102994302bSJan Kara flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); 331128d41a48SMatthew Wilcox (Oracle) entry = mk_pte(&new_folio->page, vma->vm_page_prot); 331250c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 3313c89357e2SDavid Hildenbrand if (unlikely(unshare)) { 3314c89357e2SDavid Hildenbrand if (pte_soft_dirty(vmf->orig_pte)) 3315c89357e2SDavid Hildenbrand entry = pte_mksoft_dirty(entry); 3316c89357e2SDavid Hildenbrand if (pte_uffd_wp(vmf->orig_pte)) 3317c89357e2SDavid Hildenbrand entry = pte_mkuffd_wp(entry); 3318c89357e2SDavid Hildenbrand } else { 33192f38ab2cSShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 3320c89357e2SDavid Hildenbrand } 3321111fe718SNicholas Piggin 33222f38ab2cSShachar Raindel /* 33232f38ab2cSShachar Raindel * Clear the pte entry and flush it first, before updating the 3324111fe718SNicholas Piggin * pte with the new entry, to keep TLBs on different CPUs in 3325111fe718SNicholas Piggin * sync. This code used to set the new PTE then flush TLBs, but 3326111fe718SNicholas Piggin * that left a window where the new PTE could be loaded into 3327111fe718SNicholas Piggin * some TLBs while the old PTE remains in others. 33282f38ab2cSShachar Raindel */ 3329ec8832d0SAlistair Popple ptep_clear_flush(vma, vmf->address, vmf->pte); 333028d41a48SMatthew Wilcox (Oracle) folio_add_new_anon_rmap(new_folio, vma, vmf->address); 333128d41a48SMatthew Wilcox (Oracle) folio_add_lru_vma(new_folio, vma); 33322f38ab2cSShachar Raindel /* 33332f38ab2cSShachar Raindel * We call the notify macro here because, when using secondary 33342f38ab2cSShachar Raindel * mmu page tables (such as kvm shadow page tables), we want the 33352f38ab2cSShachar Raindel * new page to be mapped directly into the secondary page table. 33362f38ab2cSShachar Raindel */ 3337c89357e2SDavid Hildenbrand BUG_ON(unshare && pte_write(entry)); 333882b0f8c3SJan Kara set_pte_at_notify(mm, vmf->address, vmf->pte, entry); 33395003a2bdSMatthew Wilcox (Oracle) update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1); 334028d41a48SMatthew Wilcox (Oracle) if (old_folio) { 33412f38ab2cSShachar Raindel /* 33422f38ab2cSShachar Raindel * Only after switching the pte to the new page may 33432f38ab2cSShachar Raindel * we remove the mapcount here. Otherwise another 33442f38ab2cSShachar Raindel * process may come and find the rmap count decremented 33452f38ab2cSShachar Raindel * before the pte is switched to the new page, and 33462f38ab2cSShachar Raindel * "reuse" the old page writing into it while our pte 33472f38ab2cSShachar Raindel * here still points into it and can be read by other 33482f38ab2cSShachar Raindel * threads. 33492f38ab2cSShachar Raindel * 33502f38ab2cSShachar Raindel * The critical issue is to order this 3351c4626503SDavid Hildenbrand * folio_remove_rmap_pte() with the ptp_clear_flush 3352c4626503SDavid Hildenbrand * above. Those stores are ordered by (if nothing else,) 33532f38ab2cSShachar Raindel * the barrier present in the atomic_add_negative 3354c4626503SDavid Hildenbrand * in folio_remove_rmap_pte(); 33552f38ab2cSShachar Raindel * 33562f38ab2cSShachar Raindel * Then the TLB flush in ptep_clear_flush ensures that 33572f38ab2cSShachar Raindel * no process can access the old page before the 33582f38ab2cSShachar Raindel * decremented mapcount is visible. And the old page 33592f38ab2cSShachar Raindel * cannot be reused until after the decremented 33602f38ab2cSShachar Raindel * mapcount is visible. So transitively, TLBs to 33612f38ab2cSShachar Raindel * old page will be flushed before it can be reused. 33622f38ab2cSShachar Raindel */ 3363c4626503SDavid Hildenbrand folio_remove_rmap_pte(old_folio, vmf->page, vma); 33642f38ab2cSShachar Raindel } 33652f38ab2cSShachar Raindel 33662f38ab2cSShachar Raindel /* Free the old page.. */ 336728d41a48SMatthew Wilcox (Oracle) new_folio = old_folio; 33682f38ab2cSShachar Raindel page_copied = 1; 33693db82b93SHugh Dickins pte_unmap_unlock(vmf->pte, vmf->ptl); 33703db82b93SHugh Dickins } else if (vmf->pte) { 33717df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 33723db82b93SHugh Dickins pte_unmap_unlock(vmf->pte, vmf->ptl); 33732f38ab2cSShachar Raindel } 33742f38ab2cSShachar Raindel 3375ec8832d0SAlistair Popple mmu_notifier_invalidate_range_end(&range); 33763db82b93SHugh Dickins 33773db82b93SHugh Dickins if (new_folio) 33783db82b93SHugh Dickins folio_put(new_folio); 337928d41a48SMatthew Wilcox (Oracle) if (old_folio) { 3380f4c4a3f4SHuang Ying if (page_copied) 338163b77499SMatthew Wilcox (Oracle) free_swap_cache(old_folio); 338228d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 33832f38ab2cSShachar Raindel } 3384662ce1dcSYang Yang 3385662ce1dcSYang Yang delayacct_wpcopy_end(); 3386cb8d8633SDavid Hildenbrand return 0; 33872f38ab2cSShachar Raindel oom: 3388164b06f2SMatthew Wilcox (Oracle) ret = VM_FAULT_OOM; 3389164b06f2SMatthew Wilcox (Oracle) out: 339028d41a48SMatthew Wilcox (Oracle) if (old_folio) 339128d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 3392662ce1dcSYang Yang 3393662ce1dcSYang Yang delayacct_wpcopy_end(); 3394164b06f2SMatthew Wilcox (Oracle) return ret; 33952f38ab2cSShachar Raindel } 33962f38ab2cSShachar Raindel 339766a6197cSJan Kara /** 339866a6197cSJan Kara * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE 339966a6197cSJan Kara * writeable once the page is prepared 340066a6197cSJan Kara * 340166a6197cSJan Kara * @vmf: structure describing the fault 3402a86bc96bSKefeng Wang * @folio: the folio of vmf->page 340366a6197cSJan Kara * 340466a6197cSJan Kara * This function handles all that is needed to finish a write page fault in a 340566a6197cSJan Kara * shared mapping due to PTE being read-only once the mapped page is prepared. 3406a862f68aSMike Rapoport * It handles locking of PTE and modifying it. 340766a6197cSJan Kara * 340866a6197cSJan Kara * The function expects the page to be locked or other protection against 340966a6197cSJan Kara * concurrent faults / writeback (such as DAX radix tree locks). 3410a862f68aSMike Rapoport * 34112797e79fSLiu Xiang * Return: %0 on success, %VM_FAULT_NOPAGE when PTE got changed before 3412a862f68aSMike Rapoport * we acquired PTE lock. 341366a6197cSJan Kara */ 3414a86bc96bSKefeng Wang static vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf, struct folio *folio) 341566a6197cSJan Kara { 341666a6197cSJan Kara WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); 341766a6197cSJan Kara vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, 341866a6197cSJan Kara &vmf->ptl); 34193db82b93SHugh Dickins if (!vmf->pte) 34203db82b93SHugh Dickins return VM_FAULT_NOPAGE; 342166a6197cSJan Kara /* 342266a6197cSJan Kara * We might have raced with another page fault while we released the 342366a6197cSJan Kara * pte_offset_map_lock. 342466a6197cSJan Kara */ 3425c33c7948SRyan Roberts if (!pte_same(ptep_get(vmf->pte), vmf->orig_pte)) { 34267df67697SBibo Mao update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 342766a6197cSJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3428a19e2553SJan Kara return VM_FAULT_NOPAGE; 342966a6197cSJan Kara } 3430a86bc96bSKefeng Wang wp_page_reuse(vmf, folio); 3431a19e2553SJan Kara return 0; 343266a6197cSJan Kara } 343366a6197cSJan Kara 3434dd906184SBoaz Harrosh /* 3435dd906184SBoaz Harrosh * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED 3436dd906184SBoaz Harrosh * mapping 3437dd906184SBoaz Harrosh */ 34382b740303SSouptick Joarder static vm_fault_t wp_pfn_shared(struct vm_fault *vmf) 3439dd906184SBoaz Harrosh { 344082b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3441bae473a4SKirill A. Shutemov 3442dd906184SBoaz Harrosh if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { 34432b740303SSouptick Joarder vm_fault_t ret; 3444dd906184SBoaz Harrosh 344582b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 34464a68fef1SMatthew Wilcox (Oracle) ret = vmf_can_call_fault(vmf); 34474a68fef1SMatthew Wilcox (Oracle) if (ret) 34484a68fef1SMatthew Wilcox (Oracle) return ret; 3449063e60d8SMatthew Wilcox (Oracle) 3450fe82221fSJan Kara vmf->flags |= FAULT_FLAG_MKWRITE; 345111bac800SDave Jiang ret = vma->vm_ops->pfn_mkwrite(vmf); 34522f89dc12SJan Kara if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) 3453dd906184SBoaz Harrosh return ret; 3454a86bc96bSKefeng Wang return finish_mkwrite_fault(vmf, NULL); 3455dd906184SBoaz Harrosh } 3456a86bc96bSKefeng Wang wp_page_reuse(vmf, NULL); 3457cb8d8633SDavid Hildenbrand return 0; 3458dd906184SBoaz Harrosh } 3459dd906184SBoaz Harrosh 34605a97858bSSidhartha Kumar static vm_fault_t wp_page_shared(struct vm_fault *vmf, struct folio *folio) 346182b0f8c3SJan Kara __releases(vmf->ptl) 346293e478d4SShachar Raindel { 346382b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3464cb8d8633SDavid Hildenbrand vm_fault_t ret = 0; 346593e478d4SShachar Raindel 34665a97858bSSidhartha Kumar folio_get(folio); 346793e478d4SShachar Raindel 346893e478d4SShachar Raindel if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 34692b740303SSouptick Joarder vm_fault_t tmp; 347093e478d4SShachar Raindel 347182b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 34724a68fef1SMatthew Wilcox (Oracle) tmp = vmf_can_call_fault(vmf); 34734a68fef1SMatthew Wilcox (Oracle) if (tmp) { 3474063e60d8SMatthew Wilcox (Oracle) folio_put(folio); 34754a68fef1SMatthew Wilcox (Oracle) return tmp; 3476063e60d8SMatthew Wilcox (Oracle) } 3477063e60d8SMatthew Wilcox (Oracle) 347886aa6998SSidhartha Kumar tmp = do_page_mkwrite(vmf, folio); 347993e478d4SShachar Raindel if (unlikely(!tmp || (tmp & 348093e478d4SShachar Raindel (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 34815a97858bSSidhartha Kumar folio_put(folio); 348293e478d4SShachar Raindel return tmp; 348393e478d4SShachar Raindel } 3484a86bc96bSKefeng Wang tmp = finish_mkwrite_fault(vmf, folio); 3485a19e2553SJan Kara if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { 34865a97858bSSidhartha Kumar folio_unlock(folio); 34875a97858bSSidhartha Kumar folio_put(folio); 348866a6197cSJan Kara return tmp; 348993e478d4SShachar Raindel } 349066a6197cSJan Kara } else { 3491a86bc96bSKefeng Wang wp_page_reuse(vmf, folio); 34925a97858bSSidhartha Kumar folio_lock(folio); 349393e478d4SShachar Raindel } 349489b15332SJohannes Weiner ret |= fault_dirty_shared_page(vmf); 34955a97858bSSidhartha Kumar folio_put(folio); 349693e478d4SShachar Raindel 349789b15332SJohannes Weiner return ret; 349893e478d4SShachar Raindel } 349993e478d4SShachar Raindel 3500dec078ccSDavid Hildenbrand static bool wp_can_reuse_anon_folio(struct folio *folio, 3501dec078ccSDavid Hildenbrand struct vm_area_struct *vma) 3502dec078ccSDavid Hildenbrand { 3503dec078ccSDavid Hildenbrand /* 3504cd197c3aSBarry Song * We could currently only reuse a subpage of a large folio if no 3505cd197c3aSBarry Song * other subpages of the large folios are still mapped. However, 3506cd197c3aSBarry Song * let's just consistently not reuse subpages even if we could 3507cd197c3aSBarry Song * reuse in that scenario, and give back a large folio a bit 3508cd197c3aSBarry Song * sooner. 3509cd197c3aSBarry Song */ 3510cd197c3aSBarry Song if (folio_test_large(folio)) 3511cd197c3aSBarry Song return false; 3512cd197c3aSBarry Song 3513cd197c3aSBarry Song /* 3514dec078ccSDavid Hildenbrand * We have to verify under folio lock: these early checks are 3515dec078ccSDavid Hildenbrand * just an optimization to avoid locking the folio and freeing 3516dec078ccSDavid Hildenbrand * the swapcache if there is little hope that we can reuse. 3517dec078ccSDavid Hildenbrand * 3518dec078ccSDavid Hildenbrand * KSM doesn't necessarily raise the folio refcount. 3519dec078ccSDavid Hildenbrand */ 3520dec078ccSDavid Hildenbrand if (folio_test_ksm(folio) || folio_ref_count(folio) > 3) 3521dec078ccSDavid Hildenbrand return false; 3522dec078ccSDavid Hildenbrand if (!folio_test_lru(folio)) 3523dec078ccSDavid Hildenbrand /* 3524dec078ccSDavid Hildenbrand * We cannot easily detect+handle references from 3525dec078ccSDavid Hildenbrand * remote LRU caches or references to LRU folios. 3526dec078ccSDavid Hildenbrand */ 3527dec078ccSDavid Hildenbrand lru_add_drain(); 3528dec078ccSDavid Hildenbrand if (folio_ref_count(folio) > 1 + folio_test_swapcache(folio)) 3529dec078ccSDavid Hildenbrand return false; 3530dec078ccSDavid Hildenbrand if (!folio_trylock(folio)) 3531dec078ccSDavid Hildenbrand return false; 3532dec078ccSDavid Hildenbrand if (folio_test_swapcache(folio)) 3533dec078ccSDavid Hildenbrand folio_free_swap(folio); 3534dec078ccSDavid Hildenbrand if (folio_test_ksm(folio) || folio_ref_count(folio) != 1) { 3535dec078ccSDavid Hildenbrand folio_unlock(folio); 3536dec078ccSDavid Hildenbrand return false; 3537dec078ccSDavid Hildenbrand } 3538dec078ccSDavid Hildenbrand /* 3539dec078ccSDavid Hildenbrand * Ok, we've got the only folio reference from our mapping 3540dec078ccSDavid Hildenbrand * and the folio is locked, it's dark out, and we're wearing 3541dec078ccSDavid Hildenbrand * sunglasses. Hit it. 3542dec078ccSDavid Hildenbrand */ 3543dec078ccSDavid Hildenbrand folio_move_anon_rmap(folio, vma); 3544dec078ccSDavid Hildenbrand folio_unlock(folio); 3545dec078ccSDavid Hildenbrand return true; 3546dec078ccSDavid Hildenbrand } 3547dec078ccSDavid Hildenbrand 35482f38ab2cSShachar Raindel /* 3549c89357e2SDavid Hildenbrand * This routine handles present pages, when 3550c89357e2SDavid Hildenbrand * * users try to write to a shared page (FAULT_FLAG_WRITE) 3551c89357e2SDavid Hildenbrand * * GUP wants to take a R/O pin on a possibly shared anonymous page 3552c89357e2SDavid Hildenbrand * (FAULT_FLAG_UNSHARE) 3553c89357e2SDavid Hildenbrand * 3554c89357e2SDavid Hildenbrand * It is done by copying the page to a new address and decrementing the 3555c89357e2SDavid Hildenbrand * shared-page counter for the old page. 35561da177e4SLinus Torvalds * 35571da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 35581da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 3559c89357e2SDavid Hildenbrand * Thus, with FAULT_FLAG_WRITE, we can safely just mark it writable once we've 3560c89357e2SDavid Hildenbrand * done any necessary COW. 35611da177e4SLinus Torvalds * 3562c89357e2SDavid Hildenbrand * In case of FAULT_FLAG_WRITE, we also mark the page dirty at this point even 3563c89357e2SDavid Hildenbrand * though the page will change only once the write actually happens. This 3564c89357e2SDavid Hildenbrand * avoids a few races, and potentially makes it more efficient. 35651da177e4SLinus Torvalds * 3566c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 35678f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 3568c1e8d7c6SMichel Lespinasse * We return with mmap_lock still held, but pte unmapped and unlocked. 35691da177e4SLinus Torvalds */ 35702b740303SSouptick Joarder static vm_fault_t do_wp_page(struct vm_fault *vmf) 357182b0f8c3SJan Kara __releases(vmf->ptl) 35721da177e4SLinus Torvalds { 3573c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 357482b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3575b9086fdeSDavid Hildenbrand struct folio *folio = NULL; 3576d61ea1cbSPeter Xu pte_t pte; 35771da177e4SLinus Torvalds 3578c89357e2SDavid Hildenbrand if (likely(!unshare)) { 3579c33c7948SRyan Roberts if (userfaultfd_pte_wp(vma, ptep_get(vmf->pte))) { 3580d61ea1cbSPeter Xu if (!userfaultfd_wp_async(vma)) { 3581529b930bSAndrea Arcangeli pte_unmap_unlock(vmf->pte, vmf->ptl); 3582529b930bSAndrea Arcangeli return handle_userfault(vmf, VM_UFFD_WP); 3583529b930bSAndrea Arcangeli } 3584529b930bSAndrea Arcangeli 35856ce64428SNadav Amit /* 3586d61ea1cbSPeter Xu * Nothing needed (cache flush, TLB invalidations, 3587d61ea1cbSPeter Xu * etc.) because we're only removing the uffd-wp bit, 3588d61ea1cbSPeter Xu * which is completely invisible to the user. 3589d61ea1cbSPeter Xu */ 3590d61ea1cbSPeter Xu pte = pte_clear_uffd_wp(ptep_get(vmf->pte)); 3591d61ea1cbSPeter Xu 3592d61ea1cbSPeter Xu set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); 3593d61ea1cbSPeter Xu /* 3594d61ea1cbSPeter Xu * Update this to be prepared for following up CoW 3595d61ea1cbSPeter Xu * handling 3596d61ea1cbSPeter Xu */ 3597d61ea1cbSPeter Xu vmf->orig_pte = pte; 3598d61ea1cbSPeter Xu } 3599d61ea1cbSPeter Xu 3600d61ea1cbSPeter Xu /* 36016ce64428SNadav Amit * Userfaultfd write-protect can defer flushes. Ensure the TLB 36026ce64428SNadav Amit * is flushed in this case before copying. 36036ce64428SNadav Amit */ 36046ce64428SNadav Amit if (unlikely(userfaultfd_wp(vmf->vma) && 36056ce64428SNadav Amit mm_tlb_flush_pending(vmf->vma->vm_mm))) 36066ce64428SNadav Amit flush_tlb_page(vmf->vma, vmf->address); 3607c89357e2SDavid Hildenbrand } 36086ce64428SNadav Amit 3609a41b70d6SJan Kara vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); 3610c89357e2SDavid Hildenbrand 36115a97858bSSidhartha Kumar if (vmf->page) 36125a97858bSSidhartha Kumar folio = page_folio(vmf->page); 36135a97858bSSidhartha Kumar 3614251b97f5SPeter Zijlstra /* 3615b9086fdeSDavid Hildenbrand * Shared mapping: we are guaranteed to have VM_WRITE and 3616b9086fdeSDavid Hildenbrand * FAULT_FLAG_WRITE set at this point. 3617b9086fdeSDavid Hildenbrand */ 3618b9086fdeSDavid Hildenbrand if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 3619b9086fdeSDavid Hildenbrand /* 362064e45507SPeter Feiner * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a 362164e45507SPeter Feiner * VM_PFNMAP VMA. 3622251b97f5SPeter Zijlstra * 3623251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 3624dd906184SBoaz Harrosh * Just mark the pages writable and/or call ops->pfn_mkwrite. 3625251b97f5SPeter Zijlstra */ 3626b9086fdeSDavid Hildenbrand if (!vmf->page) 36272994302bSJan Kara return wp_pfn_shared(vmf); 36285a97858bSSidhartha Kumar return wp_page_shared(vmf, folio); 3629251b97f5SPeter Zijlstra } 36301da177e4SLinus Torvalds 3631b9086fdeSDavid Hildenbrand /* 3632b9086fdeSDavid Hildenbrand * Private mapping: create an exclusive anonymous page copy if reuse 3633b9086fdeSDavid Hildenbrand * is impossible. We might miss VM_WRITE for FOLL_FORCE handling. 363453a05ad9SDavid Hildenbrand * 3635dec078ccSDavid Hildenbrand * If we encounter a page that is marked exclusive, we must reuse 3636dec078ccSDavid Hildenbrand * the page without further checks. 363753a05ad9SDavid Hildenbrand */ 3638dec078ccSDavid Hildenbrand if (folio && folio_test_anon(folio) && 3639dec078ccSDavid Hildenbrand (PageAnonExclusive(vmf->page) || wp_can_reuse_anon_folio(folio, vma))) { 3640dec078ccSDavid Hildenbrand if (!PageAnonExclusive(vmf->page)) 36415ca43289SDavid Hildenbrand SetPageAnonExclusive(vmf->page); 3642c89357e2SDavid Hildenbrand if (unlikely(unshare)) { 3643c89357e2SDavid Hildenbrand pte_unmap_unlock(vmf->pte, vmf->ptl); 3644c89357e2SDavid Hildenbrand return 0; 3645c89357e2SDavid Hildenbrand } 3646a86bc96bSKefeng Wang wp_page_reuse(vmf, folio); 3647cb8d8633SDavid Hildenbrand return 0; 36481da177e4SLinus Torvalds } 36491da177e4SLinus Torvalds /* 36501da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 36511da177e4SLinus Torvalds */ 3652b9086fdeSDavid Hildenbrand if (folio) 3653b9086fdeSDavid Hildenbrand folio_get(folio); 365428766805SShachar Raindel 365582b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 365694bfe85bSYang Yang #ifdef CONFIG_KSM 3657b9086fdeSDavid Hildenbrand if (folio && folio_test_ksm(folio)) 365894bfe85bSYang Yang count_vm_event(COW_KSM); 365994bfe85bSYang Yang #endif 3660a41b70d6SJan Kara return wp_page_copy(vmf); 36611da177e4SLinus Torvalds } 36621da177e4SLinus Torvalds 366397a89413SPeter Zijlstra static void unmap_mapping_range_vma(struct vm_area_struct *vma, 36641da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 36651da177e4SLinus Torvalds struct zap_details *details) 36661da177e4SLinus Torvalds { 3667f5cc4eefSAl Viro zap_page_range_single(vma, start_addr, end_addr - start_addr, details); 36681da177e4SLinus Torvalds } 36691da177e4SLinus Torvalds 3670f808c13fSDavidlohr Bueso static inline void unmap_mapping_range_tree(struct rb_root_cached *root, 3671232a6a1cSPeter Xu pgoff_t first_index, 3672232a6a1cSPeter Xu pgoff_t last_index, 36731da177e4SLinus Torvalds struct zap_details *details) 36741da177e4SLinus Torvalds { 36751da177e4SLinus Torvalds struct vm_area_struct *vma; 36761da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 36771da177e4SLinus Torvalds 3678232a6a1cSPeter Xu vma_interval_tree_foreach(vma, root, first_index, last_index) { 36791da177e4SLinus Torvalds vba = vma->vm_pgoff; 3680d6e93217SLibin vea = vba + vma_pages(vma) - 1; 3681f9871da9SMiaohe Lin zba = max(first_index, vba); 3682f9871da9SMiaohe Lin zea = min(last_index, vea); 36831da177e4SLinus Torvalds 368497a89413SPeter Zijlstra unmap_mapping_range_vma(vma, 36851da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 36861da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 368797a89413SPeter Zijlstra details); 36881da177e4SLinus Torvalds } 36891da177e4SLinus Torvalds } 36901da177e4SLinus Torvalds 36911da177e4SLinus Torvalds /** 36923506659eSMatthew Wilcox (Oracle) * unmap_mapping_folio() - Unmap single folio from processes. 36933506659eSMatthew Wilcox (Oracle) * @folio: The locked folio to be unmapped. 369422061a1fSHugh Dickins * 36953506659eSMatthew Wilcox (Oracle) * Unmap this folio from any userspace process which still has it mmaped. 369622061a1fSHugh Dickins * Typically, for efficiency, the range of nearby pages has already been 369722061a1fSHugh Dickins * unmapped by unmap_mapping_pages() or unmap_mapping_range(). But once 36983506659eSMatthew Wilcox (Oracle) * truncation or invalidation holds the lock on a folio, it may find that 36993506659eSMatthew Wilcox (Oracle) * the page has been remapped again: and then uses unmap_mapping_folio() 370022061a1fSHugh Dickins * to unmap it finally. 370122061a1fSHugh Dickins */ 37023506659eSMatthew Wilcox (Oracle) void unmap_mapping_folio(struct folio *folio) 370322061a1fSHugh Dickins { 37043506659eSMatthew Wilcox (Oracle) struct address_space *mapping = folio->mapping; 370522061a1fSHugh Dickins struct zap_details details = { }; 3706232a6a1cSPeter Xu pgoff_t first_index; 3707232a6a1cSPeter Xu pgoff_t last_index; 370822061a1fSHugh Dickins 37093506659eSMatthew Wilcox (Oracle) VM_BUG_ON(!folio_test_locked(folio)); 371022061a1fSHugh Dickins 37113506659eSMatthew Wilcox (Oracle) first_index = folio->index; 371287b11f86SSidhartha Kumar last_index = folio_next_index(folio) - 1; 3713232a6a1cSPeter Xu 37142e148f1eSPeter Xu details.even_cows = false; 37153506659eSMatthew Wilcox (Oracle) details.single_folio = folio; 3716999dad82SPeter Xu details.zap_flags = ZAP_FLAG_DROP_MARKER; 371722061a1fSHugh Dickins 37182c865995SHugh Dickins i_mmap_lock_read(mapping); 371922061a1fSHugh Dickins if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) 3720232a6a1cSPeter Xu unmap_mapping_range_tree(&mapping->i_mmap, first_index, 3721232a6a1cSPeter Xu last_index, &details); 37222c865995SHugh Dickins i_mmap_unlock_read(mapping); 372322061a1fSHugh Dickins } 372422061a1fSHugh Dickins 372522061a1fSHugh Dickins /** 3726977fbdcdSMatthew Wilcox * unmap_mapping_pages() - Unmap pages from processes. 3727977fbdcdSMatthew Wilcox * @mapping: The address space containing pages to be unmapped. 3728977fbdcdSMatthew Wilcox * @start: Index of first page to be unmapped. 3729977fbdcdSMatthew Wilcox * @nr: Number of pages to be unmapped. 0 to unmap to end of file. 3730977fbdcdSMatthew Wilcox * @even_cows: Whether to unmap even private COWed pages. 3731977fbdcdSMatthew Wilcox * 3732977fbdcdSMatthew Wilcox * Unmap the pages in this address space from any userspace process which 3733977fbdcdSMatthew Wilcox * has them mmaped. Generally, you want to remove COWed pages as well when 3734977fbdcdSMatthew Wilcox * a file is being truncated, but not when invalidating pages from the page 3735977fbdcdSMatthew Wilcox * cache. 3736977fbdcdSMatthew Wilcox */ 3737977fbdcdSMatthew Wilcox void unmap_mapping_pages(struct address_space *mapping, pgoff_t start, 3738977fbdcdSMatthew Wilcox pgoff_t nr, bool even_cows) 3739977fbdcdSMatthew Wilcox { 3740977fbdcdSMatthew Wilcox struct zap_details details = { }; 3741232a6a1cSPeter Xu pgoff_t first_index = start; 3742232a6a1cSPeter Xu pgoff_t last_index = start + nr - 1; 3743977fbdcdSMatthew Wilcox 37442e148f1eSPeter Xu details.even_cows = even_cows; 3745232a6a1cSPeter Xu if (last_index < first_index) 3746232a6a1cSPeter Xu last_index = ULONG_MAX; 3747977fbdcdSMatthew Wilcox 37482c865995SHugh Dickins i_mmap_lock_read(mapping); 3749977fbdcdSMatthew Wilcox if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) 3750232a6a1cSPeter Xu unmap_mapping_range_tree(&mapping->i_mmap, first_index, 3751232a6a1cSPeter Xu last_index, &details); 37522c865995SHugh Dickins i_mmap_unlock_read(mapping); 3753977fbdcdSMatthew Wilcox } 37546e0e99d5SDavid Howells EXPORT_SYMBOL_GPL(unmap_mapping_pages); 3755977fbdcdSMatthew Wilcox 3756977fbdcdSMatthew Wilcox /** 37578a5f14a2SKirill A. Shutemov * unmap_mapping_range - unmap the portion of all mmaps in the specified 3758977fbdcdSMatthew Wilcox * address_space corresponding to the specified byte range in the underlying 37598a5f14a2SKirill A. Shutemov * file. 37608a5f14a2SKirill A. Shutemov * 37613d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 37621da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 37631da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 376425d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 37651da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 37661da177e4SLinus Torvalds * partial pages. 37671da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 37681da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 37691da177e4SLinus Torvalds * end of the file. 37701da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 37711da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 37721da177e4SLinus Torvalds */ 37731da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 37741da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 37751da177e4SLinus Torvalds { 37769eab0421SJiajun Xie pgoff_t hba = (pgoff_t)(holebegin) >> PAGE_SHIFT; 37779eab0421SJiajun Xie pgoff_t hlen = ((pgoff_t)(holelen) + PAGE_SIZE - 1) >> PAGE_SHIFT; 37781da177e4SLinus Torvalds 37791da177e4SLinus Torvalds /* Check for overflow. */ 37801da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 37811da177e4SLinus Torvalds long long holeend = 37821da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 37831da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 37841da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 37851da177e4SLinus Torvalds } 37861da177e4SLinus Torvalds 3787977fbdcdSMatthew Wilcox unmap_mapping_pages(mapping, hba, hlen, even_cows); 37881da177e4SLinus Torvalds } 37891da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 37901da177e4SLinus Torvalds 37911da177e4SLinus Torvalds /* 3792b756a3b5SAlistair Popple * Restore a potential device exclusive pte to a working pte entry 3793b756a3b5SAlistair Popple */ 3794b756a3b5SAlistair Popple static vm_fault_t remove_device_exclusive_entry(struct vm_fault *vmf) 3795b756a3b5SAlistair Popple { 379619672a9eSMatthew Wilcox (Oracle) struct folio *folio = page_folio(vmf->page); 3797b756a3b5SAlistair Popple struct vm_area_struct *vma = vmf->vma; 3798b756a3b5SAlistair Popple struct mmu_notifier_range range; 3799fdc724d6SSuren Baghdasaryan vm_fault_t ret; 3800b756a3b5SAlistair Popple 38017c7b9629SAlistair Popple /* 38027c7b9629SAlistair Popple * We need a reference to lock the folio because we don't hold 38037c7b9629SAlistair Popple * the PTL so a racing thread can remove the device-exclusive 38047c7b9629SAlistair Popple * entry and unmap it. If the folio is free the entry must 38057c7b9629SAlistair Popple * have been removed already. If it happens to have already 38067c7b9629SAlistair Popple * been re-allocated after being freed all we do is lock and 38077c7b9629SAlistair Popple * unlock it. 38087c7b9629SAlistair Popple */ 38097c7b9629SAlistair Popple if (!folio_try_get(folio)) 38107c7b9629SAlistair Popple return 0; 38117c7b9629SAlistair Popple 3812fdc724d6SSuren Baghdasaryan ret = folio_lock_or_retry(folio, vmf); 3813fdc724d6SSuren Baghdasaryan if (ret) { 38147c7b9629SAlistair Popple folio_put(folio); 3815fdc724d6SSuren Baghdasaryan return ret; 38167c7b9629SAlistair Popple } 38177d4a8be0SAlistair Popple mmu_notifier_range_init_owner(&range, MMU_NOTIFY_EXCLUSIVE, 0, 3818b756a3b5SAlistair Popple vma->vm_mm, vmf->address & PAGE_MASK, 3819b756a3b5SAlistair Popple (vmf->address & PAGE_MASK) + PAGE_SIZE, NULL); 3820b756a3b5SAlistair Popple mmu_notifier_invalidate_range_start(&range); 3821b756a3b5SAlistair Popple 3822b756a3b5SAlistair Popple vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 3823b756a3b5SAlistair Popple &vmf->ptl); 3824c33c7948SRyan Roberts if (likely(vmf->pte && pte_same(ptep_get(vmf->pte), vmf->orig_pte))) 382519672a9eSMatthew Wilcox (Oracle) restore_exclusive_pte(vma, vmf->page, vmf->address, vmf->pte); 3826b756a3b5SAlistair Popple 38273db82b93SHugh Dickins if (vmf->pte) 3828b756a3b5SAlistair Popple pte_unmap_unlock(vmf->pte, vmf->ptl); 382919672a9eSMatthew Wilcox (Oracle) folio_unlock(folio); 38307c7b9629SAlistair Popple folio_put(folio); 3831b756a3b5SAlistair Popple 3832b756a3b5SAlistair Popple mmu_notifier_invalidate_range_end(&range); 3833b756a3b5SAlistair Popple return 0; 3834b756a3b5SAlistair Popple } 3835b756a3b5SAlistair Popple 3836a160e537SMatthew Wilcox (Oracle) static inline bool should_try_to_free_swap(struct folio *folio, 3837c145e0b4SDavid Hildenbrand struct vm_area_struct *vma, 3838c145e0b4SDavid Hildenbrand unsigned int fault_flags) 3839c145e0b4SDavid Hildenbrand { 3840a160e537SMatthew Wilcox (Oracle) if (!folio_test_swapcache(folio)) 3841c145e0b4SDavid Hildenbrand return false; 38429202d527SMatthew Wilcox (Oracle) if (mem_cgroup_swap_full(folio) || (vma->vm_flags & VM_LOCKED) || 3843a160e537SMatthew Wilcox (Oracle) folio_test_mlocked(folio)) 3844c145e0b4SDavid Hildenbrand return true; 3845c145e0b4SDavid Hildenbrand /* 3846c145e0b4SDavid Hildenbrand * If we want to map a page that's in the swapcache writable, we 3847c145e0b4SDavid Hildenbrand * have to detect via the refcount if we're really the exclusive 3848c145e0b4SDavid Hildenbrand * user. Try freeing the swapcache to get rid of the swapcache 3849c145e0b4SDavid Hildenbrand * reference only in case it's likely that we'll be the exlusive user. 3850c145e0b4SDavid Hildenbrand */ 3851a160e537SMatthew Wilcox (Oracle) return (fault_flags & FAULT_FLAG_WRITE) && !folio_test_ksm(folio) && 3852a160e537SMatthew Wilcox (Oracle) folio_ref_count(folio) == 2; 3853c145e0b4SDavid Hildenbrand } 3854c145e0b4SDavid Hildenbrand 38559c28a205SPeter Xu static vm_fault_t pte_marker_clear(struct vm_fault *vmf) 38569c28a205SPeter Xu { 38579c28a205SPeter Xu vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, 38589c28a205SPeter Xu vmf->address, &vmf->ptl); 38593db82b93SHugh Dickins if (!vmf->pte) 38603db82b93SHugh Dickins return 0; 38619c28a205SPeter Xu /* 38629c28a205SPeter Xu * Be careful so that we will only recover a special uffd-wp pte into a 38639c28a205SPeter Xu * none pte. Otherwise it means the pte could have changed, so retry. 38647e3ce3f8SPeter Xu * 38657e3ce3f8SPeter Xu * This should also cover the case where e.g. the pte changed 3866af19487fSAxel Rasmussen * quickly from a PTE_MARKER_UFFD_WP into PTE_MARKER_POISONED. 38677e3ce3f8SPeter Xu * So is_pte_marker() check is not enough to safely drop the pte. 38689c28a205SPeter Xu */ 3869c33c7948SRyan Roberts if (pte_same(vmf->orig_pte, ptep_get(vmf->pte))) 38709c28a205SPeter Xu pte_clear(vmf->vma->vm_mm, vmf->address, vmf->pte); 38719c28a205SPeter Xu pte_unmap_unlock(vmf->pte, vmf->ptl); 38729c28a205SPeter Xu return 0; 38739c28a205SPeter Xu } 38749c28a205SPeter Xu 38752bad466cSPeter Xu static vm_fault_t do_pte_missing(struct vm_fault *vmf) 38762bad466cSPeter Xu { 38772bad466cSPeter Xu if (vma_is_anonymous(vmf->vma)) 38782bad466cSPeter Xu return do_anonymous_page(vmf); 38792bad466cSPeter Xu else 38802bad466cSPeter Xu return do_fault(vmf); 38812bad466cSPeter Xu } 38822bad466cSPeter Xu 38839c28a205SPeter Xu /* 38849c28a205SPeter Xu * This is actually a page-missing access, but with uffd-wp special pte 38859c28a205SPeter Xu * installed. It means this pte was wr-protected before being unmapped. 38869c28a205SPeter Xu */ 38879c28a205SPeter Xu static vm_fault_t pte_marker_handle_uffd_wp(struct vm_fault *vmf) 38889c28a205SPeter Xu { 38899c28a205SPeter Xu /* 38909c28a205SPeter Xu * Just in case there're leftover special ptes even after the region 38917a079ba2SPeter Xu * got unregistered - we can simply clear them. 38929c28a205SPeter Xu */ 38932bad466cSPeter Xu if (unlikely(!userfaultfd_wp(vmf->vma))) 38949c28a205SPeter Xu return pte_marker_clear(vmf); 38959c28a205SPeter Xu 38962bad466cSPeter Xu return do_pte_missing(vmf); 38979c28a205SPeter Xu } 38989c28a205SPeter Xu 38995c041f5dSPeter Xu static vm_fault_t handle_pte_marker(struct vm_fault *vmf) 39005c041f5dSPeter Xu { 39015c041f5dSPeter Xu swp_entry_t entry = pte_to_swp_entry(vmf->orig_pte); 39025c041f5dSPeter Xu unsigned long marker = pte_marker_get(entry); 39035c041f5dSPeter Xu 39045c041f5dSPeter Xu /* 3905ca92ea3dSPeter Xu * PTE markers should never be empty. If anything weird happened, 3906ca92ea3dSPeter Xu * the best thing to do is to kill the process along with its mm. 39075c041f5dSPeter Xu */ 3908ca92ea3dSPeter Xu if (WARN_ON_ONCE(!marker)) 39095c041f5dSPeter Xu return VM_FAULT_SIGBUS; 39105c041f5dSPeter Xu 391115520a3fSPeter Xu /* Higher priority than uffd-wp when data corrupted */ 3912af19487fSAxel Rasmussen if (marker & PTE_MARKER_POISONED) 3913af19487fSAxel Rasmussen return VM_FAULT_HWPOISON; 391415520a3fSPeter Xu 39159c28a205SPeter Xu if (pte_marker_entry_uffd_wp(entry)) 39169c28a205SPeter Xu return pte_marker_handle_uffd_wp(vmf); 39179c28a205SPeter Xu 39189c28a205SPeter Xu /* This is an unknown pte marker */ 39199c28a205SPeter Xu return VM_FAULT_SIGBUS; 39205c041f5dSPeter Xu } 39215c041f5dSPeter Xu 3922b756a3b5SAlistair Popple /* 3923c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 39248f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 39259a95f3cfSPaul Cassella * We return with pte unmapped and unlocked. 39269a95f3cfSPaul Cassella * 3927c1e8d7c6SMichel Lespinasse * We return with the mmap_lock locked or unlocked in the same cases 39289a95f3cfSPaul Cassella * as does filemap_fault(). 39291da177e4SLinus Torvalds */ 39302b740303SSouptick Joarder vm_fault_t do_swap_page(struct vm_fault *vmf) 39311da177e4SLinus Torvalds { 393282b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3933d4f9565aSMatthew Wilcox (Oracle) struct folio *swapcache, *folio = NULL; 3934d4f9565aSMatthew Wilcox (Oracle) struct page *page; 39352799e775SMiaohe Lin struct swap_info_struct *si = NULL; 393614f9135dSDavid Hildenbrand rmap_t rmap_flags = RMAP_NONE; 393713ddaf26SKairui Song bool need_clear_cache = false; 39381493a191SDavid Hildenbrand bool exclusive = false; 393965500d23SHugh Dickins swp_entry_t entry; 39401da177e4SLinus Torvalds pte_t pte; 39412b740303SSouptick Joarder vm_fault_t ret = 0; 3942aae466b0SJoonsoo Kim void *shadow = NULL; 39431da177e4SLinus Torvalds 39442ca99358SPeter Xu if (!pte_unmap_same(vmf)) 39458f4e2101SHugh Dickins goto out; 394665500d23SHugh Dickins 39472994302bSJan Kara entry = pte_to_swp_entry(vmf->orig_pte); 3948d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 39490697212aSChristoph Lameter if (is_migration_entry(entry)) { 395082b0f8c3SJan Kara migration_entry_wait(vma->vm_mm, vmf->pmd, 395182b0f8c3SJan Kara vmf->address); 3952b756a3b5SAlistair Popple } else if (is_device_exclusive_entry(entry)) { 3953b756a3b5SAlistair Popple vmf->page = pfn_swap_entry_to_page(entry); 3954b756a3b5SAlistair Popple ret = remove_device_exclusive_entry(vmf); 39555042db43SJérôme Glisse } else if (is_device_private_entry(entry)) { 39561235ccd0SSuren Baghdasaryan if (vmf->flags & FAULT_FLAG_VMA_LOCK) { 39571235ccd0SSuren Baghdasaryan /* 39581235ccd0SSuren Baghdasaryan * migrate_to_ram is not yet ready to operate 39591235ccd0SSuren Baghdasaryan * under VMA lock. 39601235ccd0SSuren Baghdasaryan */ 39611235ccd0SSuren Baghdasaryan vma_end_read(vma); 39621235ccd0SSuren Baghdasaryan ret = VM_FAULT_RETRY; 39631235ccd0SSuren Baghdasaryan goto out; 39641235ccd0SSuren Baghdasaryan } 39651235ccd0SSuren Baghdasaryan 3966af5cdaf8SAlistair Popple vmf->page = pfn_swap_entry_to_page(entry); 396716ce101dSAlistair Popple vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 396816ce101dSAlistair Popple vmf->address, &vmf->ptl); 39693db82b93SHugh Dickins if (unlikely(!vmf->pte || 3970c33c7948SRyan Roberts !pte_same(ptep_get(vmf->pte), 3971c33c7948SRyan Roberts vmf->orig_pte))) 39723b65f437SRyan Roberts goto unlock; 397316ce101dSAlistair Popple 397416ce101dSAlistair Popple /* 397516ce101dSAlistair Popple * Get a page reference while we know the page can't be 397616ce101dSAlistair Popple * freed. 397716ce101dSAlistair Popple */ 397816ce101dSAlistair Popple get_page(vmf->page); 397916ce101dSAlistair Popple pte_unmap_unlock(vmf->pte, vmf->ptl); 39804a955bedSAlistair Popple ret = vmf->page->pgmap->ops->migrate_to_ram(vmf); 398116ce101dSAlistair Popple put_page(vmf->page); 3982d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 3983d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 39845c041f5dSPeter Xu } else if (is_pte_marker_entry(entry)) { 39855c041f5dSPeter Xu ret = handle_pte_marker(vmf); 3986d1737fdbSAndi Kleen } else { 39872994302bSJan Kara print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); 3988d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 3989d1737fdbSAndi Kleen } 39900697212aSChristoph Lameter goto out; 39910697212aSChristoph Lameter } 39920bcac06fSMinchan Kim 39932799e775SMiaohe Lin /* Prevent swapoff from happening to us. */ 39942799e775SMiaohe Lin si = get_swap_device(entry); 39952799e775SMiaohe Lin if (unlikely(!si)) 39962799e775SMiaohe Lin goto out; 39970bcac06fSMinchan Kim 39985a423081SMatthew Wilcox (Oracle) folio = swap_cache_get_folio(entry, vma, vmf->address); 39995a423081SMatthew Wilcox (Oracle) if (folio) 40005a423081SMatthew Wilcox (Oracle) page = folio_file_page(folio, swp_offset(entry)); 4001d4f9565aSMatthew Wilcox (Oracle) swapcache = folio; 4002f8020772SMinchan Kim 4003d4f9565aSMatthew Wilcox (Oracle) if (!folio) { 4004a449bf58SQian Cai if (data_race(si->flags & SWP_SYNCHRONOUS_IO) && 4005eb085574SHuang Ying __swap_count(entry) == 1) { 400613ddaf26SKairui Song /* 400713ddaf26SKairui Song * Prevent parallel swapin from proceeding with 400813ddaf26SKairui Song * the cache flag. Otherwise, another thread may 400913ddaf26SKairui Song * finish swapin first, free the entry, and swapout 401013ddaf26SKairui Song * reusing the same entry. It's undetectable as 401113ddaf26SKairui Song * pte_same() returns true due to entry reuse. 401213ddaf26SKairui Song */ 401313ddaf26SKairui Song if (swapcache_prepare(entry)) { 401413ddaf26SKairui Song /* Relax a bit to prevent rapid repeated page faults */ 401513ddaf26SKairui Song schedule_timeout_uninterruptible(1); 401613ddaf26SKairui Song goto out; 401713ddaf26SKairui Song } 401813ddaf26SKairui Song need_clear_cache = true; 401913ddaf26SKairui Song 40200bcac06fSMinchan Kim /* skip swapcache */ 402163ad4addSMatthew Wilcox (Oracle) folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, 402263ad4addSMatthew Wilcox (Oracle) vma, vmf->address, false); 402363ad4addSMatthew Wilcox (Oracle) page = &folio->page; 402463ad4addSMatthew Wilcox (Oracle) if (folio) { 402563ad4addSMatthew Wilcox (Oracle) __folio_set_locked(folio); 402663ad4addSMatthew Wilcox (Oracle) __folio_set_swapbacked(folio); 40274c6355b2SJohannes Weiner 402865995918SMatthew Wilcox (Oracle) if (mem_cgroup_swapin_charge_folio(folio, 402963ad4addSMatthew Wilcox (Oracle) vma->vm_mm, GFP_KERNEL, 403063ad4addSMatthew Wilcox (Oracle) entry)) { 4031545b1b07SMichal Hocko ret = VM_FAULT_OOM; 40324c6355b2SJohannes Weiner goto out_page; 4033545b1b07SMichal Hocko } 40340add0c77SShakeel Butt mem_cgroup_swapin_uncharge_swap(entry); 40354c6355b2SJohannes Weiner 4036aae466b0SJoonsoo Kim shadow = get_shadow_from_swap_cache(entry); 4037aae466b0SJoonsoo Kim if (shadow) 403863ad4addSMatthew Wilcox (Oracle) workingset_refault(folio, shadow); 40390076f029SJoonsoo Kim 404063ad4addSMatthew Wilcox (Oracle) folio_add_lru(folio); 40410add0c77SShakeel Butt 4042c9bdf768SMatthew Wilcox (Oracle) /* To provide entry to swap_read_folio() */ 40433d2c9087SDavid Hildenbrand folio->swap = entry; 4044c9bdf768SMatthew Wilcox (Oracle) swap_read_folio(folio, true, NULL); 404563ad4addSMatthew Wilcox (Oracle) folio->private = NULL; 40460bcac06fSMinchan Kim } 4047aa8d22a1SMinchan Kim } else { 4048e9e9b7ecSMinchan Kim page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, 4049e9e9b7ecSMinchan Kim vmf); 405063ad4addSMatthew Wilcox (Oracle) if (page) 405163ad4addSMatthew Wilcox (Oracle) folio = page_folio(page); 4052d4f9565aSMatthew Wilcox (Oracle) swapcache = folio; 40530bcac06fSMinchan Kim } 40540bcac06fSMinchan Kim 4055d4f9565aSMatthew Wilcox (Oracle) if (!folio) { 40561da177e4SLinus Torvalds /* 40578f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 40588f4e2101SHugh Dickins * while we released the pte lock. 40591da177e4SLinus Torvalds */ 406082b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 406182b0f8c3SJan Kara vmf->address, &vmf->ptl); 4062c33c7948SRyan Roberts if (likely(vmf->pte && 4063c33c7948SRyan Roberts pte_same(ptep_get(vmf->pte), vmf->orig_pte))) 40641da177e4SLinus Torvalds ret = VM_FAULT_OOM; 406565500d23SHugh Dickins goto unlock; 40661da177e4SLinus Torvalds } 40671da177e4SLinus Torvalds 40681da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 40691da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 4070f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 40712262185cSRoman Gushchin count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); 4072d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 407371f72525SWu Fengguang /* 407471f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 407571f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 407671f72525SWu Fengguang */ 4077d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 40784779cb31SAndi Kleen goto out_release; 40791da177e4SLinus Torvalds } 40801da177e4SLinus Torvalds 4081fdc724d6SSuren Baghdasaryan ret |= folio_lock_or_retry(folio, vmf); 4082fdc724d6SSuren Baghdasaryan if (ret & VM_FAULT_RETRY) 4083d065bd81SMichel Lespinasse goto out_release; 40841da177e4SLinus Torvalds 408584d60fddSDavid Hildenbrand if (swapcache) { 40864969c119SAndrea Arcangeli /* 40873b344157SMatthew Wilcox (Oracle) * Make sure folio_free_swap() or swapoff did not release the 408884d60fddSDavid Hildenbrand * swapcache from under us. The page pin, and pte_same test 408984d60fddSDavid Hildenbrand * below, are not enough to exclude that. Even if it is still 409084d60fddSDavid Hildenbrand * swapcache, we need to check that the page's swap has not 409184d60fddSDavid Hildenbrand * changed. 40924969c119SAndrea Arcangeli */ 409363ad4addSMatthew Wilcox (Oracle) if (unlikely(!folio_test_swapcache(folio) || 4094cfeed8ffSDavid Hildenbrand page_swap_entry(page).val != entry.val)) 40954969c119SAndrea Arcangeli goto out_page; 40964969c119SAndrea Arcangeli 409784d60fddSDavid Hildenbrand /* 409884d60fddSDavid Hildenbrand * KSM sometimes has to copy on read faults, for example, if 409984d60fddSDavid Hildenbrand * page->index of !PageKSM() pages would be nonlinear inside the 410084d60fddSDavid Hildenbrand * anon VMA -- PageKSM() is lost on actual swapout. 410184d60fddSDavid Hildenbrand */ 410296db66d9SMatthew Wilcox (Oracle) folio = ksm_might_need_to_copy(folio, vma, vmf->address); 410396db66d9SMatthew Wilcox (Oracle) if (unlikely(!folio)) { 41045ad64688SHugh Dickins ret = VM_FAULT_OOM; 410596db66d9SMatthew Wilcox (Oracle) folio = swapcache; 41064969c119SAndrea Arcangeli goto out_page; 410796db66d9SMatthew Wilcox (Oracle) } else if (unlikely(folio == ERR_PTR(-EHWPOISON))) { 41086b970599SKefeng Wang ret = VM_FAULT_HWPOISON; 410996db66d9SMatthew Wilcox (Oracle) folio = swapcache; 41106b970599SKefeng Wang goto out_page; 41114969c119SAndrea Arcangeli } 411296db66d9SMatthew Wilcox (Oracle) if (folio != swapcache) 411396db66d9SMatthew Wilcox (Oracle) page = folio_page(folio, 0); 4114c145e0b4SDavid Hildenbrand 4115c145e0b4SDavid Hildenbrand /* 4116c145e0b4SDavid Hildenbrand * If we want to map a page that's in the swapcache writable, we 4117c145e0b4SDavid Hildenbrand * have to detect via the refcount if we're really the exclusive 4118c145e0b4SDavid Hildenbrand * owner. Try removing the extra reference from the local LRU 41191fec6890SMatthew Wilcox (Oracle) * caches if required. 4120c145e0b4SDavid Hildenbrand */ 4121d4f9565aSMatthew Wilcox (Oracle) if ((vmf->flags & FAULT_FLAG_WRITE) && folio == swapcache && 412263ad4addSMatthew Wilcox (Oracle) !folio_test_ksm(folio) && !folio_test_lru(folio)) 4123c145e0b4SDavid Hildenbrand lru_add_drain(); 412484d60fddSDavid Hildenbrand } 41255ad64688SHugh Dickins 41264231f842SKefeng Wang folio_throttle_swaprate(folio, GFP_KERNEL); 4127073e587eSKAMEZAWA Hiroyuki 41281da177e4SLinus Torvalds /* 41298f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 41301da177e4SLinus Torvalds */ 413182b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 413282b0f8c3SJan Kara &vmf->ptl); 4133c33c7948SRyan Roberts if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) 4134b8107480SKirill Korotaev goto out_nomap; 4135b8107480SKirill Korotaev 413663ad4addSMatthew Wilcox (Oracle) if (unlikely(!folio_test_uptodate(folio))) { 4137b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 4138b8107480SKirill Korotaev goto out_nomap; 41391da177e4SLinus Torvalds } 41401da177e4SLinus Torvalds 41418c7c6e34SKAMEZAWA Hiroyuki /* 414278fbe906SDavid Hildenbrand * PG_anon_exclusive reuses PG_mappedtodisk for anon pages. A swap pte 414378fbe906SDavid Hildenbrand * must never point at an anonymous page in the swapcache that is 414478fbe906SDavid Hildenbrand * PG_anon_exclusive. Sanity check that this holds and especially, that 414578fbe906SDavid Hildenbrand * no filesystem set PG_mappedtodisk on a page in the swapcache. Sanity 414678fbe906SDavid Hildenbrand * check after taking the PT lock and making sure that nobody 414778fbe906SDavid Hildenbrand * concurrently faulted in this page and set PG_anon_exclusive. 414878fbe906SDavid Hildenbrand */ 414963ad4addSMatthew Wilcox (Oracle) BUG_ON(!folio_test_anon(folio) && folio_test_mappedtodisk(folio)); 415063ad4addSMatthew Wilcox (Oracle) BUG_ON(folio_test_anon(folio) && PageAnonExclusive(page)); 415178fbe906SDavid Hildenbrand 415278fbe906SDavid Hildenbrand /* 41531493a191SDavid Hildenbrand * Check under PT lock (to protect against concurrent fork() sharing 41541493a191SDavid Hildenbrand * the swap entry concurrently) for certainly exclusive pages. 41551493a191SDavid Hildenbrand */ 415663ad4addSMatthew Wilcox (Oracle) if (!folio_test_ksm(folio)) { 41571493a191SDavid Hildenbrand exclusive = pte_swp_exclusive(vmf->orig_pte); 4158d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache) { 41591493a191SDavid Hildenbrand /* 41601493a191SDavid Hildenbrand * We have a fresh page that is not exposed to the 41611493a191SDavid Hildenbrand * swapcache -> certainly exclusive. 41621493a191SDavid Hildenbrand */ 41631493a191SDavid Hildenbrand exclusive = true; 416463ad4addSMatthew Wilcox (Oracle) } else if (exclusive && folio_test_writeback(folio) && 4165eacde327SMiaohe Lin data_race(si->flags & SWP_STABLE_WRITES)) { 41661493a191SDavid Hildenbrand /* 41671493a191SDavid Hildenbrand * This is tricky: not all swap backends support 41681493a191SDavid Hildenbrand * concurrent page modifications while under writeback. 41691493a191SDavid Hildenbrand * 41701493a191SDavid Hildenbrand * So if we stumble over such a page in the swapcache 41711493a191SDavid Hildenbrand * we must not set the page exclusive, otherwise we can 41721493a191SDavid Hildenbrand * map it writable without further checks and modify it 41731493a191SDavid Hildenbrand * while still under writeback. 41741493a191SDavid Hildenbrand * 41751493a191SDavid Hildenbrand * For these problematic swap backends, simply drop the 41761493a191SDavid Hildenbrand * exclusive marker: this is perfectly fine as we start 41771493a191SDavid Hildenbrand * writeback only if we fully unmapped the page and 41781493a191SDavid Hildenbrand * there are no unexpected references on the page after 41791493a191SDavid Hildenbrand * unmapping succeeded. After fully unmapped, no 41801493a191SDavid Hildenbrand * further GUP references (FOLL_GET and FOLL_PIN) can 41811493a191SDavid Hildenbrand * appear, so dropping the exclusive marker and mapping 41821493a191SDavid Hildenbrand * it only R/O is fine. 41831493a191SDavid Hildenbrand */ 41841493a191SDavid Hildenbrand exclusive = false; 41851493a191SDavid Hildenbrand } 41861493a191SDavid Hildenbrand } 41871493a191SDavid Hildenbrand 41881493a191SDavid Hildenbrand /* 41896dca4ac6SPeter Collingbourne * Some architectures may have to restore extra metadata to the page 41906dca4ac6SPeter Collingbourne * when reading from swap. This metadata may be indexed by swap entry 41916dca4ac6SPeter Collingbourne * so this must be called before swap_free(). 41926dca4ac6SPeter Collingbourne */ 41936dca4ac6SPeter Collingbourne arch_swap_restore(entry, folio); 41946dca4ac6SPeter Collingbourne 41956dca4ac6SPeter Collingbourne /* 4196c145e0b4SDavid Hildenbrand * Remove the swap entry and conditionally try to free up the swapcache. 4197c145e0b4SDavid Hildenbrand * We're already holding a reference on the page but haven't mapped it 4198c145e0b4SDavid Hildenbrand * yet. 41998c7c6e34SKAMEZAWA Hiroyuki */ 4200c145e0b4SDavid Hildenbrand swap_free(entry); 4201a160e537SMatthew Wilcox (Oracle) if (should_try_to_free_swap(folio, vma, vmf->flags)) 4202a160e537SMatthew Wilcox (Oracle) folio_free_swap(folio); 42031da177e4SLinus Torvalds 4204f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, MM_ANONPAGES); 4205f1a79412SShakeel Butt dec_mm_counter(vma->vm_mm, MM_SWAPENTS); 42061da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 4207c145e0b4SDavid Hildenbrand 4208c145e0b4SDavid Hildenbrand /* 42091493a191SDavid Hildenbrand * Same logic as in do_wp_page(); however, optimize for pages that are 42101493a191SDavid Hildenbrand * certainly not shared either because we just allocated them without 42111493a191SDavid Hildenbrand * exposing them to the swapcache or because the swap entry indicates 42121493a191SDavid Hildenbrand * exclusivity. 4213c145e0b4SDavid Hildenbrand */ 421463ad4addSMatthew Wilcox (Oracle) if (!folio_test_ksm(folio) && 421563ad4addSMatthew Wilcox (Oracle) (exclusive || folio_ref_count(folio) == 1)) { 42166c287605SDavid Hildenbrand if (vmf->flags & FAULT_FLAG_WRITE) { 42171da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 421882b0f8c3SJan Kara vmf->flags &= ~FAULT_FLAG_WRITE; 42196c287605SDavid Hildenbrand } 422014f9135dSDavid Hildenbrand rmap_flags |= RMAP_EXCLUSIVE; 42211da177e4SLinus Torvalds } 42221da177e4SLinus Torvalds flush_icache_page(vma, page); 42232994302bSJan Kara if (pte_swp_soft_dirty(vmf->orig_pte)) 4224179ef71cSCyrill Gorcunov pte = pte_mksoft_dirty(pte); 4225f1eb1bacSPeter Xu if (pte_swp_uffd_wp(vmf->orig_pte)) 4226f45ec5ffSPeter Xu pte = pte_mkuffd_wp(pte); 42272994302bSJan Kara vmf->orig_pte = pte; 42280bcac06fSMinchan Kim 42290bcac06fSMinchan Kim /* ksm created a completely new copy */ 4230d4f9565aSMatthew Wilcox (Oracle) if (unlikely(folio != swapcache && swapcache)) { 42312853b66bSMatthew Wilcox (Oracle) folio_add_new_anon_rmap(folio, vma, vmf->address); 423263ad4addSMatthew Wilcox (Oracle) folio_add_lru_vma(folio, vma); 42330bcac06fSMinchan Kim } else { 4234b832a354SDavid Hildenbrand folio_add_anon_rmap_pte(folio, page, vma, vmf->address, 4235b832a354SDavid Hildenbrand rmap_flags); 423600501b53SJohannes Weiner } 42371da177e4SLinus Torvalds 423863ad4addSMatthew Wilcox (Oracle) VM_BUG_ON(!folio_test_anon(folio) || 423963ad4addSMatthew Wilcox (Oracle) (pte_write(pte) && !PageAnonExclusive(page))); 42401eba86c0SPasha Tatashin set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); 42411eba86c0SPasha Tatashin arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte); 42421eba86c0SPasha Tatashin 424363ad4addSMatthew Wilcox (Oracle) folio_unlock(folio); 4244d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache && swapcache) { 42454969c119SAndrea Arcangeli /* 42464969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 42474969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 42484969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 42494969c119SAndrea Arcangeli * further safety release the lock after the swap_free 42504969c119SAndrea Arcangeli * so that the swap count won't change under a 42514969c119SAndrea Arcangeli * parallel locked swapcache. 42524969c119SAndrea Arcangeli */ 4253d4f9565aSMatthew Wilcox (Oracle) folio_unlock(swapcache); 4254d4f9565aSMatthew Wilcox (Oracle) folio_put(swapcache); 42554969c119SAndrea Arcangeli } 4256c475a8abSHugh Dickins 425782b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) { 42582994302bSJan Kara ret |= do_wp_page(vmf); 425961469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 426061469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 42611da177e4SLinus Torvalds goto out; 42621da177e4SLinus Torvalds } 42631da177e4SLinus Torvalds 42641da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 42655003a2bdSMatthew Wilcox (Oracle) update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1); 426665500d23SHugh Dickins unlock: 42673db82b93SHugh Dickins if (vmf->pte) 426882b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 42691da177e4SLinus Torvalds out: 427013ddaf26SKairui Song /* Clear the swap cache pin for direct swapin after PTL unlock */ 427113ddaf26SKairui Song if (need_clear_cache) 427213ddaf26SKairui Song swapcache_clear(si, entry); 42732799e775SMiaohe Lin if (si) 42742799e775SMiaohe Lin put_swap_device(si); 42751da177e4SLinus Torvalds return ret; 4276b8107480SKirill Korotaev out_nomap: 42773db82b93SHugh Dickins if (vmf->pte) 427882b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 4279bc43f75cSJohannes Weiner out_page: 428063ad4addSMatthew Wilcox (Oracle) folio_unlock(folio); 42814779cb31SAndi Kleen out_release: 428263ad4addSMatthew Wilcox (Oracle) folio_put(folio); 4283d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache && swapcache) { 4284d4f9565aSMatthew Wilcox (Oracle) folio_unlock(swapcache); 4285d4f9565aSMatthew Wilcox (Oracle) folio_put(swapcache); 42864969c119SAndrea Arcangeli } 428713ddaf26SKairui Song if (need_clear_cache) 428813ddaf26SKairui Song swapcache_clear(si, entry); 42892799e775SMiaohe Lin if (si) 42902799e775SMiaohe Lin put_swap_device(si); 429165500d23SHugh Dickins return ret; 42921da177e4SLinus Torvalds } 42931da177e4SLinus Torvalds 429419eaf449SRyan Roberts static bool pte_range_none(pte_t *pte, int nr_pages) 429519eaf449SRyan Roberts { 429619eaf449SRyan Roberts int i; 429719eaf449SRyan Roberts 429819eaf449SRyan Roberts for (i = 0; i < nr_pages; i++) { 429919eaf449SRyan Roberts if (!pte_none(ptep_get_lockless(pte + i))) 430019eaf449SRyan Roberts return false; 430119eaf449SRyan Roberts } 430219eaf449SRyan Roberts 430319eaf449SRyan Roberts return true; 430419eaf449SRyan Roberts } 430519eaf449SRyan Roberts 430619eaf449SRyan Roberts static struct folio *alloc_anon_folio(struct vm_fault *vmf) 430719eaf449SRyan Roberts { 430819eaf449SRyan Roberts struct vm_area_struct *vma = vmf->vma; 4309085ff35eSKefeng Wang #ifdef CONFIG_TRANSPARENT_HUGEPAGE 431019eaf449SRyan Roberts unsigned long orders; 431119eaf449SRyan Roberts struct folio *folio; 431219eaf449SRyan Roberts unsigned long addr; 431319eaf449SRyan Roberts pte_t *pte; 431419eaf449SRyan Roberts gfp_t gfp; 431519eaf449SRyan Roberts int order; 431619eaf449SRyan Roberts 431719eaf449SRyan Roberts /* 431819eaf449SRyan Roberts * If uffd is active for the vma we need per-page fault fidelity to 431919eaf449SRyan Roberts * maintain the uffd semantics. 432019eaf449SRyan Roberts */ 432119eaf449SRyan Roberts if (unlikely(userfaultfd_armed(vma))) 432219eaf449SRyan Roberts goto fallback; 432319eaf449SRyan Roberts 432419eaf449SRyan Roberts /* 432519eaf449SRyan Roberts * Get a list of all the (large) orders below PMD_ORDER that are enabled 432619eaf449SRyan Roberts * for this vma. Then filter out the orders that can't be allocated over 432719eaf449SRyan Roberts * the faulting address and still be fully contained in the vma. 432819eaf449SRyan Roberts */ 432919eaf449SRyan Roberts orders = thp_vma_allowable_orders(vma, vma->vm_flags, false, true, true, 433019eaf449SRyan Roberts BIT(PMD_ORDER) - 1); 433119eaf449SRyan Roberts orders = thp_vma_suitable_orders(vma, vmf->address, orders); 433219eaf449SRyan Roberts 433319eaf449SRyan Roberts if (!orders) 433419eaf449SRyan Roberts goto fallback; 433519eaf449SRyan Roberts 433619eaf449SRyan Roberts pte = pte_offset_map(vmf->pmd, vmf->address & PMD_MASK); 433719eaf449SRyan Roberts if (!pte) 433819eaf449SRyan Roberts return ERR_PTR(-EAGAIN); 433919eaf449SRyan Roberts 434019eaf449SRyan Roberts /* 434119eaf449SRyan Roberts * Find the highest order where the aligned range is completely 434219eaf449SRyan Roberts * pte_none(). Note that all remaining orders will be completely 434319eaf449SRyan Roberts * pte_none(). 434419eaf449SRyan Roberts */ 434519eaf449SRyan Roberts order = highest_order(orders); 434619eaf449SRyan Roberts while (orders) { 434719eaf449SRyan Roberts addr = ALIGN_DOWN(vmf->address, PAGE_SIZE << order); 434819eaf449SRyan Roberts if (pte_range_none(pte + pte_index(addr), 1 << order)) 434919eaf449SRyan Roberts break; 435019eaf449SRyan Roberts order = next_order(&orders, order); 435119eaf449SRyan Roberts } 435219eaf449SRyan Roberts 435319eaf449SRyan Roberts pte_unmap(pte); 435419eaf449SRyan Roberts 435519eaf449SRyan Roberts /* Try allocating the highest of the remaining orders. */ 435619eaf449SRyan Roberts gfp = vma_thp_gfp_mask(vma); 435719eaf449SRyan Roberts while (orders) { 435819eaf449SRyan Roberts addr = ALIGN_DOWN(vmf->address, PAGE_SIZE << order); 435919eaf449SRyan Roberts folio = vma_alloc_folio(gfp, order, vma, addr, true); 436019eaf449SRyan Roberts if (folio) { 4361085ff35eSKefeng Wang if (mem_cgroup_charge(folio, vma->vm_mm, gfp)) { 4362085ff35eSKefeng Wang folio_put(folio); 4363085ff35eSKefeng Wang goto next; 4364085ff35eSKefeng Wang } 4365085ff35eSKefeng Wang folio_throttle_swaprate(folio, gfp); 436619eaf449SRyan Roberts clear_huge_page(&folio->page, vmf->address, 1 << order); 436719eaf449SRyan Roberts return folio; 436819eaf449SRyan Roberts } 4369085ff35eSKefeng Wang next: 437019eaf449SRyan Roberts order = next_order(&orders, order); 437119eaf449SRyan Roberts } 437219eaf449SRyan Roberts 437319eaf449SRyan Roberts fallback: 437419eaf449SRyan Roberts #endif 4375085ff35eSKefeng Wang return folio_prealloc(vma->vm_mm, vma, vmf->address, true); 437619eaf449SRyan Roberts } 437719eaf449SRyan Roberts 43781da177e4SLinus Torvalds /* 4379c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 43808f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 4381c1e8d7c6SMichel Lespinasse * We return with mmap_lock still held, but pte unmapped and unlocked. 43821da177e4SLinus Torvalds */ 43832b740303SSouptick Joarder static vm_fault_t do_anonymous_page(struct vm_fault *vmf) 43841da177e4SLinus Torvalds { 43852bad466cSPeter Xu bool uffd_wp = vmf_orig_pte_uffd_wp(vmf); 438682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 438719eaf449SRyan Roberts unsigned long addr = vmf->address; 43886bc56a4dSMatthew Wilcox (Oracle) struct folio *folio; 43892b740303SSouptick Joarder vm_fault_t ret = 0; 439019eaf449SRyan Roberts int nr_pages = 1; 43911da177e4SLinus Torvalds pte_t entry; 439219eaf449SRyan Roberts int i; 43931da177e4SLinus Torvalds 43946b7339f4SKirill A. Shutemov /* File mapping without ->vm_ops ? */ 43956b7339f4SKirill A. Shutemov if (vma->vm_flags & VM_SHARED) 43966b7339f4SKirill A. Shutemov return VM_FAULT_SIGBUS; 43976b7339f4SKirill A. Shutemov 43987267ec00SKirill A. Shutemov /* 43993db82b93SHugh Dickins * Use pte_alloc() instead of pte_alloc_map(), so that OOM can 44003db82b93SHugh Dickins * be distinguished from a transient failure of pte_offset_map(). 44017267ec00SKirill A. Shutemov */ 44024cf58924SJoel Fernandes (Google) if (pte_alloc(vma->vm_mm, vmf->pmd)) 44037267ec00SKirill A. Shutemov return VM_FAULT_OOM; 44047267ec00SKirill A. Shutemov 440511ac5524SLinus Torvalds /* Use the zero-page for reads */ 440682b0f8c3SJan Kara if (!(vmf->flags & FAULT_FLAG_WRITE) && 4407bae473a4SKirill A. Shutemov !mm_forbids_zeropage(vma->vm_mm)) { 440882b0f8c3SJan Kara entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), 440962eede62SHugh Dickins vma->vm_page_prot)); 441082b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 441182b0f8c3SJan Kara vmf->address, &vmf->ptl); 44123db82b93SHugh Dickins if (!vmf->pte) 44133db82b93SHugh Dickins goto unlock; 44142bad466cSPeter Xu if (vmf_pte_changed(vmf)) { 44157df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 4416a13ea5b7SHugh Dickins goto unlock; 44177df67697SBibo Mao } 44186b31d595SMichal Hocko ret = check_stable_address_space(vma->vm_mm); 44196b31d595SMichal Hocko if (ret) 44206b31d595SMichal Hocko goto unlock; 44216b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 44226b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 442382b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 442482b0f8c3SJan Kara return handle_userfault(vmf, VM_UFFD_MISSING); 44256b251fc9SAndrea Arcangeli } 4426a13ea5b7SHugh Dickins goto setpte; 4427a13ea5b7SHugh Dickins } 4428a13ea5b7SHugh Dickins 44291da177e4SLinus Torvalds /* Allocate our own private page. */ 44301da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 443165500d23SHugh Dickins goto oom; 443219eaf449SRyan Roberts /* Returns NULL on OOM or ERR_PTR(-EAGAIN) if we must retry the fault */ 443319eaf449SRyan Roberts folio = alloc_anon_folio(vmf); 443419eaf449SRyan Roberts if (IS_ERR(folio)) 443519eaf449SRyan Roberts return 0; 44366bc56a4dSMatthew Wilcox (Oracle) if (!folio) 443765500d23SHugh Dickins goto oom; 4438eb3c24f3SMel Gorman 443919eaf449SRyan Roberts nr_pages = folio_nr_pages(folio); 444019eaf449SRyan Roberts addr = ALIGN_DOWN(vmf->address, nr_pages * PAGE_SIZE); 444119eaf449SRyan Roberts 444252f37629SMinchan Kim /* 4443cb3184deSMatthew Wilcox (Oracle) * The memory barrier inside __folio_mark_uptodate makes sure that 4444f4f5329dSWei Yang * preceding stores to the page contents become visible before 444552f37629SMinchan Kim * the set_pte_at() write. 444652f37629SMinchan Kim */ 4447cb3184deSMatthew Wilcox (Oracle) __folio_mark_uptodate(folio); 44481da177e4SLinus Torvalds 4449cb3184deSMatthew Wilcox (Oracle) entry = mk_pte(&folio->page, vma->vm_page_prot); 445050c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 44511ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 4452161e393cSRick Edgecombe entry = pte_mkwrite(pte_mkdirty(entry), vma); 44538f4e2101SHugh Dickins 445419eaf449SRyan Roberts vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); 44553db82b93SHugh Dickins if (!vmf->pte) 44563db82b93SHugh Dickins goto release; 445719eaf449SRyan Roberts if (nr_pages == 1 && vmf_pte_changed(vmf)) { 445819eaf449SRyan Roberts update_mmu_tlb(vma, addr, vmf->pte); 445919eaf449SRyan Roberts goto release; 446019eaf449SRyan Roberts } else if (nr_pages > 1 && !pte_range_none(vmf->pte, nr_pages)) { 446119eaf449SRyan Roberts for (i = 0; i < nr_pages; i++) 446219eaf449SRyan Roberts update_mmu_tlb(vma, addr + PAGE_SIZE * i, vmf->pte + i); 44638f4e2101SHugh Dickins goto release; 44647df67697SBibo Mao } 44659ba69294SHugh Dickins 44666b31d595SMichal Hocko ret = check_stable_address_space(vma->vm_mm); 44676b31d595SMichal Hocko if (ret) 44686b31d595SMichal Hocko goto release; 44696b31d595SMichal Hocko 44706b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 44716b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 447282b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 4473cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 447482b0f8c3SJan Kara return handle_userfault(vmf, VM_UFFD_MISSING); 44756b251fc9SAndrea Arcangeli } 44766b251fc9SAndrea Arcangeli 447719eaf449SRyan Roberts folio_ref_add(folio, nr_pages - 1); 447819eaf449SRyan Roberts add_mm_counter(vma->vm_mm, MM_ANONPAGES, nr_pages); 447919eaf449SRyan Roberts folio_add_new_anon_rmap(folio, vma, addr); 4480cb3184deSMatthew Wilcox (Oracle) folio_add_lru_vma(folio, vma); 4481a13ea5b7SHugh Dickins setpte: 44822bad466cSPeter Xu if (uffd_wp) 44832bad466cSPeter Xu entry = pte_mkuffd_wp(entry); 448419eaf449SRyan Roberts set_ptes(vma->vm_mm, addr, vmf->pte, entry, nr_pages); 44851da177e4SLinus Torvalds 44861da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 448719eaf449SRyan Roberts update_mmu_cache_range(vmf, vma, addr, vmf->pte, nr_pages); 448865500d23SHugh Dickins unlock: 44893db82b93SHugh Dickins if (vmf->pte) 449082b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 44916b31d595SMichal Hocko return ret; 44928f4e2101SHugh Dickins release: 4493cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 44948f4e2101SHugh Dickins goto unlock; 449565500d23SHugh Dickins oom: 44961da177e4SLinus Torvalds return VM_FAULT_OOM; 44971da177e4SLinus Torvalds } 44981da177e4SLinus Torvalds 44999a95f3cfSPaul Cassella /* 4500c1e8d7c6SMichel Lespinasse * The mmap_lock must have been held on entry, and may have been 45019a95f3cfSPaul Cassella * released depending on flags and vma->vm_ops->fault() return value. 45029a95f3cfSPaul Cassella * See filemap_fault() and __lock_page_retry(). 45039a95f3cfSPaul Cassella */ 45042b740303SSouptick Joarder static vm_fault_t __do_fault(struct vm_fault *vmf) 45057eae74afSKirill A. Shutemov { 450682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 450701d1e0e6SMatthew Wilcox (Oracle) struct folio *folio; 45082b740303SSouptick Joarder vm_fault_t ret; 45097eae74afSKirill A. Shutemov 451063f3655fSMichal Hocko /* 451163f3655fSMichal Hocko * Preallocate pte before we take page_lock because this might lead to 451263f3655fSMichal Hocko * deadlocks for memcg reclaim which waits for pages under writeback: 451363f3655fSMichal Hocko * lock_page(A) 451463f3655fSMichal Hocko * SetPageWriteback(A) 451563f3655fSMichal Hocko * unlock_page(A) 451663f3655fSMichal Hocko * lock_page(B) 451763f3655fSMichal Hocko * lock_page(B) 4518d383807aSYanfei Xu * pte_alloc_one 451963f3655fSMichal Hocko * shrink_page_list 452063f3655fSMichal Hocko * wait_on_page_writeback(A) 452163f3655fSMichal Hocko * SetPageWriteback(B) 452263f3655fSMichal Hocko * unlock_page(B) 452363f3655fSMichal Hocko * # flush A, B to clear the writeback 452463f3655fSMichal Hocko */ 452563f3655fSMichal Hocko if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) { 4526a7069ee3SYanfei Xu vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); 452763f3655fSMichal Hocko if (!vmf->prealloc_pte) 452863f3655fSMichal Hocko return VM_FAULT_OOM; 452963f3655fSMichal Hocko } 453063f3655fSMichal Hocko 453111bac800SDave Jiang ret = vma->vm_ops->fault(vmf); 45323917048dSJan Kara if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | 4533b1aa812bSJan Kara VM_FAULT_DONE_COW))) 45347eae74afSKirill A. Shutemov return ret; 45357eae74afSKirill A. Shutemov 453601d1e0e6SMatthew Wilcox (Oracle) folio = page_folio(vmf->page); 4537667240e0SJan Kara if (unlikely(PageHWPoison(vmf->page))) { 4538e53ac737SRik van Riel vm_fault_t poisonret = VM_FAULT_HWPOISON; 4539e53ac737SRik van Riel if (ret & VM_FAULT_LOCKED) { 454001d1e0e6SMatthew Wilcox (Oracle) if (page_mapped(vmf->page)) 454101d1e0e6SMatthew Wilcox (Oracle) unmap_mapping_folio(folio); 454201d1e0e6SMatthew Wilcox (Oracle) /* Retry if a clean folio was removed from the cache. */ 454301d1e0e6SMatthew Wilcox (Oracle) if (mapping_evict_folio(folio->mapping, folio)) 45443149c79fSRik van Riel poisonret = VM_FAULT_NOPAGE; 454501d1e0e6SMatthew Wilcox (Oracle) folio_unlock(folio); 4546e53ac737SRik van Riel } 454701d1e0e6SMatthew Wilcox (Oracle) folio_put(folio); 4548936ca80dSJan Kara vmf->page = NULL; 4549e53ac737SRik van Riel return poisonret; 45507eae74afSKirill A. Shutemov } 45517eae74afSKirill A. Shutemov 45527eae74afSKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) 455301d1e0e6SMatthew Wilcox (Oracle) folio_lock(folio); 45547eae74afSKirill A. Shutemov else 455501d1e0e6SMatthew Wilcox (Oracle) VM_BUG_ON_PAGE(!folio_test_locked(folio), vmf->page); 45567eae74afSKirill A. Shutemov 45577eae74afSKirill A. Shutemov return ret; 45587eae74afSKirill A. Shutemov } 45597eae74afSKirill A. Shutemov 4560396bcc52SMatthew Wilcox (Oracle) #ifdef CONFIG_TRANSPARENT_HUGEPAGE 456182b0f8c3SJan Kara static void deposit_prealloc_pte(struct vm_fault *vmf) 4562953c66c2SAneesh Kumar K.V { 456382b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4564953c66c2SAneesh Kumar K.V 456582b0f8c3SJan Kara pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); 4566953c66c2SAneesh Kumar K.V /* 4567953c66c2SAneesh Kumar K.V * We are going to consume the prealloc table, 4568953c66c2SAneesh Kumar K.V * count that as nr_ptes. 4569953c66c2SAneesh Kumar K.V */ 4570c4812909SKirill A. Shutemov mm_inc_nr_ptes(vma->vm_mm); 45717f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 4572953c66c2SAneesh Kumar K.V } 4573953c66c2SAneesh Kumar K.V 4574f9ce0be7SKirill A. Shutemov vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) 457510102459SKirill A. Shutemov { 4576ef37b2eaSDavid Hildenbrand struct folio *folio = page_folio(page); 457782b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 457882b0f8c3SJan Kara bool write = vmf->flags & FAULT_FLAG_WRITE; 457982b0f8c3SJan Kara unsigned long haddr = vmf->address & HPAGE_PMD_MASK; 458010102459SKirill A. Shutemov pmd_t entry; 4581d01ac3c3SMatthew Wilcox (Oracle) vm_fault_t ret = VM_FAULT_FALLBACK; 458210102459SKirill A. Shutemov 45833485b883SRyan Roberts if (!thp_vma_suitable_order(vma, haddr, PMD_ORDER)) 4584d01ac3c3SMatthew Wilcox (Oracle) return ret; 458510102459SKirill A. Shutemov 4586ef37b2eaSDavid Hildenbrand if (page != &folio->page || folio_order(folio) != HPAGE_PMD_ORDER) 4587d01ac3c3SMatthew Wilcox (Oracle) return ret; 458810102459SKirill A. Shutemov 4589953c66c2SAneesh Kumar K.V /* 4590eac96c3eSYang Shi * Just backoff if any subpage of a THP is corrupted otherwise 4591eac96c3eSYang Shi * the corrupted page may mapped by PMD silently to escape the 4592eac96c3eSYang Shi * check. This kind of THP just can be PTE mapped. Access to 4593eac96c3eSYang Shi * the corrupted subpage should trigger SIGBUS as expected. 4594eac96c3eSYang Shi */ 4595ef37b2eaSDavid Hildenbrand if (unlikely(folio_test_has_hwpoisoned(folio))) 4596eac96c3eSYang Shi return ret; 4597eac96c3eSYang Shi 4598eac96c3eSYang Shi /* 4599f0953a1bSIngo Molnar * Archs like ppc64 need additional space to store information 4600953c66c2SAneesh Kumar K.V * related to pte entry. Use the preallocated table for that. 4601953c66c2SAneesh Kumar K.V */ 460282b0f8c3SJan Kara if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { 46034cf58924SJoel Fernandes (Google) vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); 460482b0f8c3SJan Kara if (!vmf->prealloc_pte) 4605953c66c2SAneesh Kumar K.V return VM_FAULT_OOM; 4606953c66c2SAneesh Kumar K.V } 4607953c66c2SAneesh Kumar K.V 460882b0f8c3SJan Kara vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); 460982b0f8c3SJan Kara if (unlikely(!pmd_none(*vmf->pmd))) 461010102459SKirill A. Shutemov goto out; 461110102459SKirill A. Shutemov 46129f1f5b60SMatthew Wilcox (Oracle) flush_icache_pages(vma, page, HPAGE_PMD_NR); 461310102459SKirill A. Shutemov 461410102459SKirill A. Shutemov entry = mk_huge_pmd(page, vma->vm_page_prot); 461510102459SKirill A. Shutemov if (write) 4616f55e1014SLinus Torvalds entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); 461710102459SKirill A. Shutemov 46186b27cc6cSKefeng Wang add_mm_counter(vma->vm_mm, mm_counter_file(folio), HPAGE_PMD_NR); 4619ef37b2eaSDavid Hildenbrand folio_add_file_rmap_pmd(folio, page, vma); 4620cea86fe2SHugh Dickins 4621953c66c2SAneesh Kumar K.V /* 4622953c66c2SAneesh Kumar K.V * deposit and withdraw with pmd lock held 4623953c66c2SAneesh Kumar K.V */ 4624953c66c2SAneesh Kumar K.V if (arch_needs_pgtable_deposit()) 462582b0f8c3SJan Kara deposit_prealloc_pte(vmf); 462610102459SKirill A. Shutemov 462782b0f8c3SJan Kara set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); 462810102459SKirill A. Shutemov 462982b0f8c3SJan Kara update_mmu_cache_pmd(vma, haddr, vmf->pmd); 463010102459SKirill A. Shutemov 463110102459SKirill A. Shutemov /* fault is handled */ 463210102459SKirill A. Shutemov ret = 0; 463395ecedcdSKirill A. Shutemov count_vm_event(THP_FILE_MAPPED); 463410102459SKirill A. Shutemov out: 463582b0f8c3SJan Kara spin_unlock(vmf->ptl); 463610102459SKirill A. Shutemov return ret; 463710102459SKirill A. Shutemov } 463810102459SKirill A. Shutemov #else 4639f9ce0be7SKirill A. Shutemov vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) 464010102459SKirill A. Shutemov { 4641f9ce0be7SKirill A. Shutemov return VM_FAULT_FALLBACK; 464210102459SKirill A. Shutemov } 464310102459SKirill A. Shutemov #endif 464410102459SKirill A. Shutemov 46453bd786f7SYin Fengwei /** 46463bd786f7SYin Fengwei * set_pte_range - Set a range of PTEs to point to pages in a folio. 46473bd786f7SYin Fengwei * @vmf: Fault decription. 46483bd786f7SYin Fengwei * @folio: The folio that contains @page. 46493bd786f7SYin Fengwei * @page: The first page to create a PTE for. 46503bd786f7SYin Fengwei * @nr: The number of PTEs to create. 46513bd786f7SYin Fengwei * @addr: The first address to create a PTE for. 46523bd786f7SYin Fengwei */ 46533bd786f7SYin Fengwei void set_pte_range(struct vm_fault *vmf, struct folio *folio, 46543bd786f7SYin Fengwei struct page *page, unsigned int nr, unsigned long addr) 46553bb97794SKirill A. Shutemov { 465682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 46572bad466cSPeter Xu bool uffd_wp = vmf_orig_pte_uffd_wp(vmf); 465882b0f8c3SJan Kara bool write = vmf->flags & FAULT_FLAG_WRITE; 46593bd786f7SYin Fengwei bool prefault = in_range(vmf->address, addr, nr * PAGE_SIZE); 46603bb97794SKirill A. Shutemov pte_t entry; 46617267ec00SKirill A. Shutemov 46623bd786f7SYin Fengwei flush_icache_pages(vma, page, nr); 46633bb97794SKirill A. Shutemov entry = mk_pte(page, vma->vm_page_prot); 466446bdb427SWill Deacon 466546bdb427SWill Deacon if (prefault && arch_wants_old_prefaulted_pte()) 466646bdb427SWill Deacon entry = pte_mkold(entry); 466750c25ee9SThomas Bogendoerfer else 466850c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 466946bdb427SWill Deacon 46703bb97794SKirill A. Shutemov if (write) 46713bb97794SKirill A. Shutemov entry = maybe_mkwrite(pte_mkdirty(entry), vma); 46729c28a205SPeter Xu if (unlikely(uffd_wp)) 4673f1eb1bacSPeter Xu entry = pte_mkuffd_wp(entry); 4674bae473a4SKirill A. Shutemov /* copy-on-write page */ 4675bae473a4SKirill A. Shutemov if (write && !(vma->vm_flags & VM_SHARED)) { 46763bd786f7SYin Fengwei add_mm_counter(vma->vm_mm, MM_ANONPAGES, nr); 46773bd786f7SYin Fengwei VM_BUG_ON_FOLIO(nr != 1, folio); 46783bd786f7SYin Fengwei folio_add_new_anon_rmap(folio, vma, addr); 46793bd786f7SYin Fengwei folio_add_lru_vma(folio, vma); 46803bb97794SKirill A. Shutemov } else { 46816b27cc6cSKefeng Wang add_mm_counter(vma->vm_mm, mm_counter_file(folio), nr); 468268f03208SDavid Hildenbrand folio_add_file_rmap_ptes(folio, page, nr, vma); 46833bb97794SKirill A. Shutemov } 46843bd786f7SYin Fengwei set_ptes(vma->vm_mm, addr, vmf->pte, entry, nr); 46853bd786f7SYin Fengwei 46863bd786f7SYin Fengwei /* no need to invalidate: a not-present page won't be cached */ 46873bd786f7SYin Fengwei update_mmu_cache_range(vmf, vma, addr, vmf->pte, nr); 46883bb97794SKirill A. Shutemov } 46893bb97794SKirill A. Shutemov 4690f46f2adeSPeter Xu static bool vmf_pte_changed(struct vm_fault *vmf) 4691f46f2adeSPeter Xu { 4692f46f2adeSPeter Xu if (vmf->flags & FAULT_FLAG_ORIG_PTE_VALID) 4693c33c7948SRyan Roberts return !pte_same(ptep_get(vmf->pte), vmf->orig_pte); 4694f46f2adeSPeter Xu 4695c33c7948SRyan Roberts return !pte_none(ptep_get(vmf->pte)); 4696f46f2adeSPeter Xu } 4697f46f2adeSPeter Xu 46989118c0cbSJan Kara /** 46999118c0cbSJan Kara * finish_fault - finish page fault once we have prepared the page to fault 47009118c0cbSJan Kara * 47019118c0cbSJan Kara * @vmf: structure describing the fault 47029118c0cbSJan Kara * 47039118c0cbSJan Kara * This function handles all that is needed to finish a page fault once the 47049118c0cbSJan Kara * page to fault in is prepared. It handles locking of PTEs, inserts PTE for 47059118c0cbSJan Kara * given page, adds reverse page mapping, handles memcg charges and LRU 4706a862f68aSMike Rapoport * addition. 47079118c0cbSJan Kara * 47089118c0cbSJan Kara * The function expects the page to be locked and on success it consumes a 47099118c0cbSJan Kara * reference of a page being mapped (for the PTE which maps it). 4710a862f68aSMike Rapoport * 4711a862f68aSMike Rapoport * Return: %0 on success, %VM_FAULT_ code in case of error. 47129118c0cbSJan Kara */ 47132b740303SSouptick Joarder vm_fault_t finish_fault(struct vm_fault *vmf) 47149118c0cbSJan Kara { 4715f9ce0be7SKirill A. Shutemov struct vm_area_struct *vma = vmf->vma; 47169118c0cbSJan Kara struct page *page; 4717f9ce0be7SKirill A. Shutemov vm_fault_t ret; 47189118c0cbSJan Kara 47199118c0cbSJan Kara /* Did we COW the page? */ 4720f9ce0be7SKirill A. Shutemov if ((vmf->flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) 47219118c0cbSJan Kara page = vmf->cow_page; 47229118c0cbSJan Kara else 47239118c0cbSJan Kara page = vmf->page; 47246b31d595SMichal Hocko 47256b31d595SMichal Hocko /* 47266b31d595SMichal Hocko * check even for read faults because we might have lost our CoWed 47276b31d595SMichal Hocko * page 47286b31d595SMichal Hocko */ 4729f9ce0be7SKirill A. Shutemov if (!(vma->vm_flags & VM_SHARED)) { 4730f9ce0be7SKirill A. Shutemov ret = check_stable_address_space(vma->vm_mm); 4731f9ce0be7SKirill A. Shutemov if (ret) 4732f9ce0be7SKirill A. Shutemov return ret; 4733f9ce0be7SKirill A. Shutemov } 4734f9ce0be7SKirill A. Shutemov 4735f9ce0be7SKirill A. Shutemov if (pmd_none(*vmf->pmd)) { 4736f9ce0be7SKirill A. Shutemov if (PageTransCompound(page)) { 4737f9ce0be7SKirill A. Shutemov ret = do_set_pmd(vmf, page); 4738f9ce0be7SKirill A. Shutemov if (ret != VM_FAULT_FALLBACK) 4739f9ce0be7SKirill A. Shutemov return ret; 4740f9ce0be7SKirill A. Shutemov } 4741f9ce0be7SKirill A. Shutemov 474203c4f204SQi Zheng if (vmf->prealloc_pte) 474303c4f204SQi Zheng pmd_install(vma->vm_mm, vmf->pmd, &vmf->prealloc_pte); 474403c4f204SQi Zheng else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd))) 4745f9ce0be7SKirill A. Shutemov return VM_FAULT_OOM; 4746f9ce0be7SKirill A. Shutemov } 4747f9ce0be7SKirill A. Shutemov 4748f9ce0be7SKirill A. Shutemov vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 4749f9ce0be7SKirill A. Shutemov vmf->address, &vmf->ptl); 47503db82b93SHugh Dickins if (!vmf->pte) 47513db82b93SHugh Dickins return VM_FAULT_NOPAGE; 4752f9ce0be7SKirill A. Shutemov 475370427f6eSSergei Antonov /* Re-check under ptl */ 475470427f6eSSergei Antonov if (likely(!vmf_pte_changed(vmf))) { 47553bd786f7SYin Fengwei struct folio *folio = page_folio(page); 475670427f6eSSergei Antonov 47573bd786f7SYin Fengwei set_pte_range(vmf, folio, page, 1, vmf->address); 475870427f6eSSergei Antonov ret = 0; 475970427f6eSSergei Antonov } else { 4760f9ce0be7SKirill A. Shutemov update_mmu_tlb(vma, vmf->address, vmf->pte); 476170427f6eSSergei Antonov ret = VM_FAULT_NOPAGE; 476270427f6eSSergei Antonov } 476370427f6eSSergei Antonov 47649118c0cbSJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 47659118c0cbSJan Kara return ret; 47669118c0cbSJan Kara } 47679118c0cbSJan Kara 476853d36a56SLorenzo Stoakes static unsigned long fault_around_pages __read_mostly = 476953d36a56SLorenzo Stoakes 65536 >> PAGE_SHIFT; 4770a9b0f861SKirill A. Shutemov 47711592eef0SKirill A. Shutemov #ifdef CONFIG_DEBUG_FS 4772a9b0f861SKirill A. Shutemov static int fault_around_bytes_get(void *data, u64 *val) 47731592eef0SKirill A. Shutemov { 477453d36a56SLorenzo Stoakes *val = fault_around_pages << PAGE_SHIFT; 47751592eef0SKirill A. Shutemov return 0; 47761592eef0SKirill A. Shutemov } 47771592eef0SKirill A. Shutemov 4778b4903d6eSAndrey Ryabinin /* 4779da391d64SWilliam Kucharski * fault_around_bytes must be rounded down to the nearest page order as it's 4780da391d64SWilliam Kucharski * what do_fault_around() expects to see. 4781b4903d6eSAndrey Ryabinin */ 4782a9b0f861SKirill A. Shutemov static int fault_around_bytes_set(void *data, u64 val) 47831592eef0SKirill A. Shutemov { 4784a9b0f861SKirill A. Shutemov if (val / PAGE_SIZE > PTRS_PER_PTE) 47851592eef0SKirill A. Shutemov return -EINVAL; 478653d36a56SLorenzo Stoakes 478753d36a56SLorenzo Stoakes /* 478853d36a56SLorenzo Stoakes * The minimum value is 1 page, however this results in no fault-around 478953d36a56SLorenzo Stoakes * at all. See should_fault_around(). 479053d36a56SLorenzo Stoakes */ 47915aa598a7SKefeng Wang val = max(val, PAGE_SIZE); 47925aa598a7SKefeng Wang fault_around_pages = rounddown_pow_of_two(val) >> PAGE_SHIFT; 479353d36a56SLorenzo Stoakes 47941592eef0SKirill A. Shutemov return 0; 47951592eef0SKirill A. Shutemov } 47960a1345f8SYevgen Pronenko DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, 4797a9b0f861SKirill A. Shutemov fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); 47981592eef0SKirill A. Shutemov 47991592eef0SKirill A. Shutemov static int __init fault_around_debugfs(void) 48001592eef0SKirill A. Shutemov { 4801d9f7979cSGreg Kroah-Hartman debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, 4802a9b0f861SKirill A. Shutemov &fault_around_bytes_fops); 48031592eef0SKirill A. Shutemov return 0; 48041592eef0SKirill A. Shutemov } 48051592eef0SKirill A. Shutemov late_initcall(fault_around_debugfs); 48061592eef0SKirill A. Shutemov #endif 48078c6e50b0SKirill A. Shutemov 48081fdb412bSKirill A. Shutemov /* 48091fdb412bSKirill A. Shutemov * do_fault_around() tries to map few pages around the fault address. The hope 48101fdb412bSKirill A. Shutemov * is that the pages will be needed soon and this will lower the number of 48111fdb412bSKirill A. Shutemov * faults to handle. 48121fdb412bSKirill A. Shutemov * 48131fdb412bSKirill A. Shutemov * It uses vm_ops->map_pages() to map the pages, which skips the page if it's 48141fdb412bSKirill A. Shutemov * not ready to be mapped: not up-to-date, locked, etc. 48151fdb412bSKirill A. Shutemov * 48169042599eSLorenzo Stoakes * This function doesn't cross VMA or page table boundaries, in order to call 48179042599eSLorenzo Stoakes * map_pages() and acquire a PTE lock only once. 48181fdb412bSKirill A. Shutemov * 481953d36a56SLorenzo Stoakes * fault_around_pages defines how many pages we'll try to map. 4820da391d64SWilliam Kucharski * do_fault_around() expects it to be set to a power of two less than or equal 4821da391d64SWilliam Kucharski * to PTRS_PER_PTE. 48221fdb412bSKirill A. Shutemov * 4823da391d64SWilliam Kucharski * The virtual address of the area that we map is naturally aligned to 482453d36a56SLorenzo Stoakes * fault_around_pages * PAGE_SIZE rounded down to the machine page size 4825da391d64SWilliam Kucharski * (and therefore to page order). This way it's easier to guarantee 4826da391d64SWilliam Kucharski * that we don't cross page table boundaries. 48271fdb412bSKirill A. Shutemov */ 48282b740303SSouptick Joarder static vm_fault_t do_fault_around(struct vm_fault *vmf) 48298c6e50b0SKirill A. Shutemov { 483053d36a56SLorenzo Stoakes pgoff_t nr_pages = READ_ONCE(fault_around_pages); 48319042599eSLorenzo Stoakes pgoff_t pte_off = pte_index(vmf->address); 48329042599eSLorenzo Stoakes /* The page offset of vmf->address within the VMA. */ 48339042599eSLorenzo Stoakes pgoff_t vma_off = vmf->pgoff - vmf->vma->vm_pgoff; 48349042599eSLorenzo Stoakes pgoff_t from_pte, to_pte; 483558ef47efSMatthew Wilcox (Oracle) vm_fault_t ret; 48368c6e50b0SKirill A. Shutemov 48379042599eSLorenzo Stoakes /* The PTE offset of the start address, clamped to the VMA. */ 48389042599eSLorenzo Stoakes from_pte = max(ALIGN_DOWN(pte_off, nr_pages), 48399042599eSLorenzo Stoakes pte_off - min(pte_off, vma_off)); 4840aecd6f44SKirill A. Shutemov 48419042599eSLorenzo Stoakes /* The PTE offset of the end address, clamped to the VMA and PTE. */ 48429042599eSLorenzo Stoakes to_pte = min3(from_pte + nr_pages, (pgoff_t)PTRS_PER_PTE, 48439042599eSLorenzo Stoakes pte_off + vma_pages(vmf->vma) - vma_off) - 1; 48448c6e50b0SKirill A. Shutemov 484582b0f8c3SJan Kara if (pmd_none(*vmf->pmd)) { 48464cf58924SJoel Fernandes (Google) vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm); 484782b0f8c3SJan Kara if (!vmf->prealloc_pte) 4848f9ce0be7SKirill A. Shutemov return VM_FAULT_OOM; 48498c6e50b0SKirill A. Shutemov } 48508c6e50b0SKirill A. Shutemov 485158ef47efSMatthew Wilcox (Oracle) rcu_read_lock(); 485258ef47efSMatthew Wilcox (Oracle) ret = vmf->vma->vm_ops->map_pages(vmf, 48539042599eSLorenzo Stoakes vmf->pgoff + from_pte - pte_off, 48549042599eSLorenzo Stoakes vmf->pgoff + to_pte - pte_off); 485558ef47efSMatthew Wilcox (Oracle) rcu_read_unlock(); 485658ef47efSMatthew Wilcox (Oracle) 485758ef47efSMatthew Wilcox (Oracle) return ret; 48587267ec00SKirill A. Shutemov } 48597267ec00SKirill A. Shutemov 48609c28a205SPeter Xu /* Return true if we should do read fault-around, false otherwise */ 48619c28a205SPeter Xu static inline bool should_fault_around(struct vm_fault *vmf) 48629c28a205SPeter Xu { 48639c28a205SPeter Xu /* No ->map_pages? No way to fault around... */ 48649c28a205SPeter Xu if (!vmf->vma->vm_ops->map_pages) 48659c28a205SPeter Xu return false; 48669c28a205SPeter Xu 48679c28a205SPeter Xu if (uffd_disable_fault_around(vmf->vma)) 48689c28a205SPeter Xu return false; 48699c28a205SPeter Xu 487053d36a56SLorenzo Stoakes /* A single page implies no faulting 'around' at all. */ 487153d36a56SLorenzo Stoakes return fault_around_pages > 1; 48729c28a205SPeter Xu } 48739c28a205SPeter Xu 48742b740303SSouptick Joarder static vm_fault_t do_read_fault(struct vm_fault *vmf) 4875e655fb29SKirill A. Shutemov { 48762b740303SSouptick Joarder vm_fault_t ret = 0; 487722d1e68fSSidhartha Kumar struct folio *folio; 48788c6e50b0SKirill A. Shutemov 48798c6e50b0SKirill A. Shutemov /* 48808c6e50b0SKirill A. Shutemov * Let's call ->map_pages() first and use ->fault() as fallback 48818c6e50b0SKirill A. Shutemov * if page by the offset is not ready to be mapped (cold cache or 48828c6e50b0SKirill A. Shutemov * something). 48838c6e50b0SKirill A. Shutemov */ 48849c28a205SPeter Xu if (should_fault_around(vmf)) { 48850721ec8bSJan Kara ret = do_fault_around(vmf); 48867267ec00SKirill A. Shutemov if (ret) 48877267ec00SKirill A. Shutemov return ret; 48888c6e50b0SKirill A. Shutemov } 4889e655fb29SKirill A. Shutemov 489012214ebaSMatthew Wilcox (Oracle) ret = vmf_can_call_fault(vmf); 489112214ebaSMatthew Wilcox (Oracle) if (ret) 489212214ebaSMatthew Wilcox (Oracle) return ret; 4893f5617ffeSMatthew Wilcox (Oracle) 4894936ca80dSJan Kara ret = __do_fault(vmf); 4895e655fb29SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4896e655fb29SKirill A. Shutemov return ret; 4897e655fb29SKirill A. Shutemov 48989118c0cbSJan Kara ret |= finish_fault(vmf); 489922d1e68fSSidhartha Kumar folio = page_folio(vmf->page); 490022d1e68fSSidhartha Kumar folio_unlock(folio); 49017267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 490222d1e68fSSidhartha Kumar folio_put(folio); 4903e655fb29SKirill A. Shutemov return ret; 4904e655fb29SKirill A. Shutemov } 4905e655fb29SKirill A. Shutemov 49062b740303SSouptick Joarder static vm_fault_t do_cow_fault(struct vm_fault *vmf) 4907ec47c3b9SKirill A. Shutemov { 490882b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4909e4621e70SKefeng Wang struct folio *folio; 49102b740303SSouptick Joarder vm_fault_t ret; 4911ec47c3b9SKirill A. Shutemov 49124de8c93aSMatthew Wilcox (Oracle) ret = vmf_can_call_fault(vmf); 49134de8c93aSMatthew Wilcox (Oracle) if (!ret) 49144de8c93aSMatthew Wilcox (Oracle) ret = vmf_anon_prepare(vmf); 49154de8c93aSMatthew Wilcox (Oracle) if (ret) 49164de8c93aSMatthew Wilcox (Oracle) return ret; 4917ec47c3b9SKirill A. Shutemov 4918e4621e70SKefeng Wang folio = folio_prealloc(vma->vm_mm, vma, vmf->address, false); 4919e4621e70SKefeng Wang if (!folio) 4920ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4921ec47c3b9SKirill A. Shutemov 4922e4621e70SKefeng Wang vmf->cow_page = &folio->page; 4923ec47c3b9SKirill A. Shutemov 4924936ca80dSJan Kara ret = __do_fault(vmf); 4925ec47c3b9SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4926ec47c3b9SKirill A. Shutemov goto uncharge_out; 49273917048dSJan Kara if (ret & VM_FAULT_DONE_COW) 49283917048dSJan Kara return ret; 4929ec47c3b9SKirill A. Shutemov 4930936ca80dSJan Kara copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); 4931e4621e70SKefeng Wang __folio_mark_uptodate(folio); 4932ec47c3b9SKirill A. Shutemov 49339118c0cbSJan Kara ret |= finish_fault(vmf); 4934936ca80dSJan Kara unlock_page(vmf->page); 4935936ca80dSJan Kara put_page(vmf->page); 49367267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 49377267ec00SKirill A. Shutemov goto uncharge_out; 4938ec47c3b9SKirill A. Shutemov return ret; 4939ec47c3b9SKirill A. Shutemov uncharge_out: 4940e4621e70SKefeng Wang folio_put(folio); 4941ec47c3b9SKirill A. Shutemov return ret; 4942ec47c3b9SKirill A. Shutemov } 4943ec47c3b9SKirill A. Shutemov 49442b740303SSouptick Joarder static vm_fault_t do_shared_fault(struct vm_fault *vmf) 49451da177e4SLinus Torvalds { 494682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 49472b740303SSouptick Joarder vm_fault_t ret, tmp; 49486f609b7eSSidhartha Kumar struct folio *folio; 49491d65f86dSKAMEZAWA Hiroyuki 49504ed43798SMatthew Wilcox (Oracle) ret = vmf_can_call_fault(vmf); 49514ed43798SMatthew Wilcox (Oracle) if (ret) 49524ed43798SMatthew Wilcox (Oracle) return ret; 495354cb8821SNick Piggin 4954936ca80dSJan Kara ret = __do_fault(vmf); 49557eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4956f0c6d4d2SKirill A. Shutemov return ret; 49571da177e4SLinus Torvalds 49586f609b7eSSidhartha Kumar folio = page_folio(vmf->page); 49596f609b7eSSidhartha Kumar 49601da177e4SLinus Torvalds /* 4961f0c6d4d2SKirill A. Shutemov * Check if the backing address space wants to know that the page is 4962f0c6d4d2SKirill A. Shutemov * about to become writable 49631da177e4SLinus Torvalds */ 4964fb09a464SKirill A. Shutemov if (vma->vm_ops->page_mkwrite) { 49656f609b7eSSidhartha Kumar folio_unlock(folio); 496686aa6998SSidhartha Kumar tmp = do_page_mkwrite(vmf, folio); 4967fb09a464SKirill A. Shutemov if (unlikely(!tmp || 4968fb09a464SKirill A. Shutemov (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 49696f609b7eSSidhartha Kumar folio_put(folio); 4970f0c6d4d2SKirill A. Shutemov return tmp; 497169676147SMark Fasheh } 4972d0217ac0SNick Piggin } 4973fb09a464SKirill A. Shutemov 49749118c0cbSJan Kara ret |= finish_fault(vmf); 49757267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | 49767267ec00SKirill A. Shutemov VM_FAULT_RETRY))) { 49776f609b7eSSidhartha Kumar folio_unlock(folio); 49786f609b7eSSidhartha Kumar folio_put(folio); 4979f0c6d4d2SKirill A. Shutemov return ret; 49809637a5efSDavid Howells } 4981d00806b1SNick Piggin 498289b15332SJohannes Weiner ret |= fault_dirty_shared_page(vmf); 4983b827e496SNick Piggin return ret; 498454cb8821SNick Piggin } 4985d00806b1SNick Piggin 49869a95f3cfSPaul Cassella /* 4987c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 49889a95f3cfSPaul Cassella * but allow concurrent faults). 4989c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 49909138e47eSMatthew Wilcox (Oracle) * return value. See filemap_fault() and __folio_lock_or_retry(). 4991c1e8d7c6SMichel Lespinasse * If mmap_lock is released, vma may become invalid (for example 4992fc8efd2dSJan Stancek * by other thread calling munmap()). 49939a95f3cfSPaul Cassella */ 49942b740303SSouptick Joarder static vm_fault_t do_fault(struct vm_fault *vmf) 499554cb8821SNick Piggin { 499682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4997fc8efd2dSJan Stancek struct mm_struct *vm_mm = vma->vm_mm; 49982b740303SSouptick Joarder vm_fault_t ret; 499954cb8821SNick Piggin 5000ff09d7ecSAneesh Kumar K.V /* 5001ff09d7ecSAneesh Kumar K.V * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND 5002ff09d7ecSAneesh Kumar K.V */ 5003ff09d7ecSAneesh Kumar K.V if (!vma->vm_ops->fault) { 50043db82b93SHugh Dickins vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, 50053db82b93SHugh Dickins vmf->address, &vmf->ptl); 50063db82b93SHugh Dickins if (unlikely(!vmf->pte)) 5007b0b9b3dfSHugh Dickins ret = VM_FAULT_SIGBUS; 5008ff09d7ecSAneesh Kumar K.V else { 5009ff09d7ecSAneesh Kumar K.V /* 5010ff09d7ecSAneesh Kumar K.V * Make sure this is not a temporary clearing of pte 5011ff09d7ecSAneesh Kumar K.V * by holding ptl and checking again. A R/M/W update 5012ff09d7ecSAneesh Kumar K.V * of pte involves: take ptl, clearing the pte so that 5013ff09d7ecSAneesh Kumar K.V * we don't have concurrent modification by hardware 5014ff09d7ecSAneesh Kumar K.V * followed by an update. 5015ff09d7ecSAneesh Kumar K.V */ 5016c33c7948SRyan Roberts if (unlikely(pte_none(ptep_get(vmf->pte)))) 5017ff09d7ecSAneesh Kumar K.V ret = VM_FAULT_SIGBUS; 5018ff09d7ecSAneesh Kumar K.V else 5019ff09d7ecSAneesh Kumar K.V ret = VM_FAULT_NOPAGE; 5020ff09d7ecSAneesh Kumar K.V 5021ff09d7ecSAneesh Kumar K.V pte_unmap_unlock(vmf->pte, vmf->ptl); 5022ff09d7ecSAneesh Kumar K.V } 5023ff09d7ecSAneesh Kumar K.V } else if (!(vmf->flags & FAULT_FLAG_WRITE)) 5024b0b9b3dfSHugh Dickins ret = do_read_fault(vmf); 5025b0b9b3dfSHugh Dickins else if (!(vma->vm_flags & VM_SHARED)) 5026b0b9b3dfSHugh Dickins ret = do_cow_fault(vmf); 5027b0b9b3dfSHugh Dickins else 5028b0b9b3dfSHugh Dickins ret = do_shared_fault(vmf); 5029b0b9b3dfSHugh Dickins 5030b0b9b3dfSHugh Dickins /* preallocated pagetable is unused: free it */ 5031b0b9b3dfSHugh Dickins if (vmf->prealloc_pte) { 5032fc8efd2dSJan Stancek pte_free(vm_mm, vmf->prealloc_pte); 50337f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 5034b0b9b3dfSHugh Dickins } 5035b0b9b3dfSHugh Dickins return ret; 503654cb8821SNick Piggin } 503754cb8821SNick Piggin 5038cda6d936SKefeng Wang int numa_migrate_prep(struct folio *folio, struct vm_area_struct *vma, 5039f4c0d836SYang Shi unsigned long addr, int page_nid, int *flags) 50409532fec1SMel Gorman { 5041cda6d936SKefeng Wang folio_get(folio); 50429532fec1SMel Gorman 5043fc137c0dSRaghavendra K T /* Record the current PID acceesing VMA */ 5044fc137c0dSRaghavendra K T vma_set_access_pid_bit(vma); 5045fc137c0dSRaghavendra K T 50469532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS); 504704bb2f94SRik van Riel if (page_nid == numa_node_id()) { 50489532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); 504904bb2f94SRik van Riel *flags |= TNF_FAULT_LOCAL; 505004bb2f94SRik van Riel } 50519532fec1SMel Gorman 505275c70128SKefeng Wang return mpol_misplaced(folio, vma, addr); 50539532fec1SMel Gorman } 50549532fec1SMel Gorman 50552b740303SSouptick Joarder static vm_fault_t do_numa_page(struct vm_fault *vmf) 5056d10e63f2SMel Gorman { 505782b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 50586695cf68SKefeng Wang struct folio *folio = NULL; 50596695cf68SKefeng Wang int nid = NUMA_NO_NODE; 50606a56ccbcSDavid Hildenbrand bool writable = false; 506190572890SPeter Zijlstra int last_cpupid; 5062cbee9f88SPeter Zijlstra int target_nid; 506304a86453SAneesh Kumar K.V pte_t pte, old_pte; 50646688cc05SPeter Zijlstra int flags = 0; 5065d10e63f2SMel Gorman 5066d10e63f2SMel Gorman /* 50676c1b748eSJohn Hubbard * The pte cannot be used safely until we verify, while holding the page 50686c1b748eSJohn Hubbard * table lock, that its contents have not changed during fault handling. 5069d10e63f2SMel Gorman */ 507082b0f8c3SJan Kara spin_lock(vmf->ptl); 50716c1b748eSJohn Hubbard /* Read the live PTE from the page tables: */ 50726c1b748eSJohn Hubbard old_pte = ptep_get(vmf->pte); 50736c1b748eSJohn Hubbard 50746c1b748eSJohn Hubbard if (unlikely(!pte_same(old_pte, vmf->orig_pte))) { 507582b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 50764daae3b4SMel Gorman goto out; 50774daae3b4SMel Gorman } 50784daae3b4SMel Gorman 507904a86453SAneesh Kumar K.V pte = pte_modify(old_pte, vma->vm_page_prot); 5080d10e63f2SMel Gorman 50816a56ccbcSDavid Hildenbrand /* 50826a56ccbcSDavid Hildenbrand * Detect now whether the PTE could be writable; this information 50836a56ccbcSDavid Hildenbrand * is only valid while holding the PT lock. 50846a56ccbcSDavid Hildenbrand */ 50856a56ccbcSDavid Hildenbrand writable = pte_write(pte); 50866a56ccbcSDavid Hildenbrand if (!writable && vma_wants_manual_pte_write_upgrade(vma) && 50876a56ccbcSDavid Hildenbrand can_change_pte_writable(vma, vmf->address, pte)) 50886a56ccbcSDavid Hildenbrand writable = true; 50896a56ccbcSDavid Hildenbrand 50906695cf68SKefeng Wang folio = vm_normal_folio(vma, vmf->address, pte); 50916695cf68SKefeng Wang if (!folio || folio_is_zone_device(folio)) 5092b99a342dSHuang Ying goto out_map; 5093d10e63f2SMel Gorman 5094e81c4802SKirill A. Shutemov /* TODO: handle PTE-mapped THP */ 50956695cf68SKefeng Wang if (folio_test_large(folio)) 5096b99a342dSHuang Ying goto out_map; 5097e81c4802SKirill A. Shutemov 50986688cc05SPeter Zijlstra /* 5099bea66fbdSMel Gorman * Avoid grouping on RO pages in general. RO pages shouldn't hurt as 5100bea66fbdSMel Gorman * much anyway since they can be in shared cache state. This misses 5101bea66fbdSMel Gorman * the case where a mapping is writable but the process never writes 5102bea66fbdSMel Gorman * to it but pte_write gets cleared during protection updates and 5103bea66fbdSMel Gorman * pte_dirty has unpredictable behaviour between PTE scan updates, 5104bea66fbdSMel Gorman * background writeback, dirty balancing and application behaviour. 51056688cc05SPeter Zijlstra */ 51066a56ccbcSDavid Hildenbrand if (!writable) 51076688cc05SPeter Zijlstra flags |= TNF_NO_GROUP; 51086688cc05SPeter Zijlstra 5109dabe1d99SRik van Riel /* 51106695cf68SKefeng Wang * Flag if the folio is shared between multiple address spaces. This 5111dabe1d99SRik van Riel * is later used when determining whether to group tasks together 5112dabe1d99SRik van Riel */ 51136695cf68SKefeng Wang if (folio_estimated_sharers(folio) > 1 && (vma->vm_flags & VM_SHARED)) 5114dabe1d99SRik van Riel flags |= TNF_SHARED; 5115dabe1d99SRik van Riel 51166695cf68SKefeng Wang nid = folio_nid(folio); 511733024536SHuang Ying /* 511833024536SHuang Ying * For memory tiering mode, cpupid of slow memory page is used 511933024536SHuang Ying * to record page access time. So use default value. 512033024536SHuang Ying */ 512133024536SHuang Ying if ((sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) && 51226695cf68SKefeng Wang !node_is_toptier(nid)) 512333024536SHuang Ying last_cpupid = (-1 & LAST_CPUPID_MASK); 512433024536SHuang Ying else 512567b33e3fSKefeng Wang last_cpupid = folio_last_cpupid(folio); 5126cda6d936SKefeng Wang target_nid = numa_migrate_prep(folio, vma, vmf->address, nid, &flags); 512798fa15f3SAnshuman Khandual if (target_nid == NUMA_NO_NODE) { 51286695cf68SKefeng Wang folio_put(folio); 5129b99a342dSHuang Ying goto out_map; 51304daae3b4SMel Gorman } 5131b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 51326a56ccbcSDavid Hildenbrand writable = false; 51334daae3b4SMel Gorman 51344daae3b4SMel Gorman /* Migrate to the requested node */ 51356695cf68SKefeng Wang if (migrate_misplaced_folio(folio, vma, target_nid)) { 51366695cf68SKefeng Wang nid = target_nid; 51376688cc05SPeter Zijlstra flags |= TNF_MIGRATED; 5138b99a342dSHuang Ying } else { 5139074c2381SMel Gorman flags |= TNF_MIGRATE_FAIL; 5140c7ad0880SHugh Dickins vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 5141c7ad0880SHugh Dickins vmf->address, &vmf->ptl); 5142c7ad0880SHugh Dickins if (unlikely(!vmf->pte)) 5143c7ad0880SHugh Dickins goto out; 5144c33c7948SRyan Roberts if (unlikely(!pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 5145b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 5146b99a342dSHuang Ying goto out; 5147b99a342dSHuang Ying } 5148b99a342dSHuang Ying goto out_map; 5149b99a342dSHuang Ying } 51504daae3b4SMel Gorman 51514daae3b4SMel Gorman out: 51526695cf68SKefeng Wang if (nid != NUMA_NO_NODE) 51536695cf68SKefeng Wang task_numa_fault(last_cpupid, nid, 1, flags); 5154d10e63f2SMel Gorman return 0; 5155b99a342dSHuang Ying out_map: 5156b99a342dSHuang Ying /* 5157b99a342dSHuang Ying * Make it present again, depending on how arch implements 5158b99a342dSHuang Ying * non-accessible ptes, some can allow access by kernel mode. 5159b99a342dSHuang Ying */ 5160b99a342dSHuang Ying old_pte = ptep_modify_prot_start(vma, vmf->address, vmf->pte); 5161b99a342dSHuang Ying pte = pte_modify(old_pte, vma->vm_page_prot); 5162b99a342dSHuang Ying pte = pte_mkyoung(pte); 51636a56ccbcSDavid Hildenbrand if (writable) 5164161e393cSRick Edgecombe pte = pte_mkwrite(pte, vma); 5165b99a342dSHuang Ying ptep_modify_prot_commit(vma, vmf->address, vmf->pte, old_pte, pte); 51665003a2bdSMatthew Wilcox (Oracle) update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1); 5167b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 5168b99a342dSHuang Ying goto out; 5169d10e63f2SMel Gorman } 5170d10e63f2SMel Gorman 51712b740303SSouptick Joarder static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf) 5172b96375f7SMatthew Wilcox { 51738f5fd0e1SMatthew Wilcox (Oracle) struct vm_area_struct *vma = vmf->vma; 51748f5fd0e1SMatthew Wilcox (Oracle) if (vma_is_anonymous(vma)) 517582b0f8c3SJan Kara return do_huge_pmd_anonymous_page(vmf); 517640d49a3cSMatthew Wilcox (Oracle) if (vma->vm_ops->huge_fault) 51771d024e7aSMatthew Wilcox (Oracle) return vma->vm_ops->huge_fault(vmf, PMD_ORDER); 5178b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 5179b96375f7SMatthew Wilcox } 5180b96375f7SMatthew Wilcox 5181183f24aaSGeert Uytterhoeven /* `inline' is required to avoid gcc 4.1.2 build error */ 51825db4f15cSYang Shi static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf) 5183b96375f7SMatthew Wilcox { 51848f5fd0e1SMatthew Wilcox (Oracle) struct vm_area_struct *vma = vmf->vma; 5185c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 5186aea06577SDavid Hildenbrand vm_fault_t ret; 5187c89357e2SDavid Hildenbrand 51888f5fd0e1SMatthew Wilcox (Oracle) if (vma_is_anonymous(vma)) { 5189c89357e2SDavid Hildenbrand if (likely(!unshare) && 5190d61ea1cbSPeter Xu userfaultfd_huge_pmd_wp(vma, vmf->orig_pmd)) { 5191d61ea1cbSPeter Xu if (userfaultfd_wp_async(vmf->vma)) 5192d61ea1cbSPeter Xu goto split; 5193529b930bSAndrea Arcangeli return handle_userfault(vmf, VM_UFFD_WP); 5194d61ea1cbSPeter Xu } 51955db4f15cSYang Shi return do_huge_pmd_wp_page(vmf); 5196529b930bSAndrea Arcangeli } 5197af9e4d5fSKirill A. Shutemov 51988f5fd0e1SMatthew Wilcox (Oracle) if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 51998f5fd0e1SMatthew Wilcox (Oracle) if (vma->vm_ops->huge_fault) { 52001d024e7aSMatthew Wilcox (Oracle) ret = vma->vm_ops->huge_fault(vmf, PMD_ORDER); 5201327e9fd4SThomas Hellstrom (VMware) if (!(ret & VM_FAULT_FALLBACK)) 5202327e9fd4SThomas Hellstrom (VMware) return ret; 5203327e9fd4SThomas Hellstrom (VMware) } 5204aea06577SDavid Hildenbrand } 5205327e9fd4SThomas Hellstrom (VMware) 5206d61ea1cbSPeter Xu split: 5207327e9fd4SThomas Hellstrom (VMware) /* COW or write-notify handled on pte level: split pmd. */ 52088f5fd0e1SMatthew Wilcox (Oracle) __split_huge_pmd(vma, vmf->pmd, vmf->address, false, NULL); 5209af9e4d5fSKirill A. Shutemov 5210b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 5211b96375f7SMatthew Wilcox } 5212b96375f7SMatthew Wilcox 52132b740303SSouptick Joarder static vm_fault_t create_huge_pud(struct vm_fault *vmf) 5214a00cc7d9SMatthew Wilcox { 5215327e9fd4SThomas Hellstrom (VMware) #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ 5216327e9fd4SThomas Hellstrom (VMware) defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) 5217c4fd825eSMatthew Wilcox (Oracle) struct vm_area_struct *vma = vmf->vma; 5218a00cc7d9SMatthew Wilcox /* No support for anonymous transparent PUD pages yet */ 5219c4fd825eSMatthew Wilcox (Oracle) if (vma_is_anonymous(vma)) 522014c99d65SGowans, James return VM_FAULT_FALLBACK; 522140d49a3cSMatthew Wilcox (Oracle) if (vma->vm_ops->huge_fault) 52221d024e7aSMatthew Wilcox (Oracle) return vma->vm_ops->huge_fault(vmf, PUD_ORDER); 522314c99d65SGowans, James #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 522414c99d65SGowans, James return VM_FAULT_FALLBACK; 522514c99d65SGowans, James } 522614c99d65SGowans, James 522714c99d65SGowans, James static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) 522814c99d65SGowans, James { 522914c99d65SGowans, James #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ 523014c99d65SGowans, James defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) 5231c4fd825eSMatthew Wilcox (Oracle) struct vm_area_struct *vma = vmf->vma; 5232aea06577SDavid Hildenbrand vm_fault_t ret; 5233aea06577SDavid Hildenbrand 523414c99d65SGowans, James /* No support for anonymous transparent PUD pages yet */ 5235c4fd825eSMatthew Wilcox (Oracle) if (vma_is_anonymous(vma)) 5236327e9fd4SThomas Hellstrom (VMware) goto split; 5237c4fd825eSMatthew Wilcox (Oracle) if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 5238c4fd825eSMatthew Wilcox (Oracle) if (vma->vm_ops->huge_fault) { 52391d024e7aSMatthew Wilcox (Oracle) ret = vma->vm_ops->huge_fault(vmf, PUD_ORDER); 5240327e9fd4SThomas Hellstrom (VMware) if (!(ret & VM_FAULT_FALLBACK)) 5241327e9fd4SThomas Hellstrom (VMware) return ret; 5242327e9fd4SThomas Hellstrom (VMware) } 5243aea06577SDavid Hildenbrand } 5244327e9fd4SThomas Hellstrom (VMware) split: 5245327e9fd4SThomas Hellstrom (VMware) /* COW or write-notify not handled on PUD level: split pud.*/ 5246c4fd825eSMatthew Wilcox (Oracle) __split_huge_pud(vma, vmf->pud, vmf->address); 524714c99d65SGowans, James #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 5248a00cc7d9SMatthew Wilcox return VM_FAULT_FALLBACK; 5249a00cc7d9SMatthew Wilcox } 5250a00cc7d9SMatthew Wilcox 52511da177e4SLinus Torvalds /* 52521da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 52531da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 52541da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 52551da177e4SLinus Torvalds * 52561da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 52571da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 52581da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 52591da177e4SLinus Torvalds * 5260c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, but allow 52617267ec00SKirill A. Shutemov * concurrent faults). 52629a95f3cfSPaul Cassella * 5263c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our return value. 52649138e47eSMatthew Wilcox (Oracle) * See filemap_fault() and __folio_lock_or_retry(). 52651da177e4SLinus Torvalds */ 52662b740303SSouptick Joarder static vm_fault_t handle_pte_fault(struct vm_fault *vmf) 52671da177e4SLinus Torvalds { 52681da177e4SLinus Torvalds pte_t entry; 52691da177e4SLinus Torvalds 527082b0f8c3SJan Kara if (unlikely(pmd_none(*vmf->pmd))) { 52717267ec00SKirill A. Shutemov /* 52727267ec00SKirill A. Shutemov * Leave __pte_alloc() until later: because vm_ops->fault may 52737267ec00SKirill A. Shutemov * want to allocate huge page, and if we expose page table 52747267ec00SKirill A. Shutemov * for an instant, it will be difficult to retract from 52757267ec00SKirill A. Shutemov * concurrent faults and from rmap lookups. 52767267ec00SKirill A. Shutemov */ 527782b0f8c3SJan Kara vmf->pte = NULL; 5278f46f2adeSPeter Xu vmf->flags &= ~FAULT_FLAG_ORIG_PTE_VALID; 52797267ec00SKirill A. Shutemov } else { 5280f9ce0be7SKirill A. Shutemov /* 52817267ec00SKirill A. Shutemov * A regular pmd is established and it can't morph into a huge 5282c7ad0880SHugh Dickins * pmd by anon khugepaged, since that takes mmap_lock in write 5283c7ad0880SHugh Dickins * mode; but shmem or file collapse to THP could still morph 5284c7ad0880SHugh Dickins * it into a huge pmd: just retry later if so. 52857267ec00SKirill A. Shutemov */ 5286c7ad0880SHugh Dickins vmf->pte = pte_offset_map_nolock(vmf->vma->vm_mm, vmf->pmd, 5287c7ad0880SHugh Dickins vmf->address, &vmf->ptl); 5288c7ad0880SHugh Dickins if (unlikely(!vmf->pte)) 5289c7ad0880SHugh Dickins return 0; 529026e1a0c3SHugh Dickins vmf->orig_pte = ptep_get_lockless(vmf->pte); 5291f46f2adeSPeter Xu vmf->flags |= FAULT_FLAG_ORIG_PTE_VALID; 52927267ec00SKirill A. Shutemov 52932994302bSJan Kara if (pte_none(vmf->orig_pte)) { 529482b0f8c3SJan Kara pte_unmap(vmf->pte); 529582b0f8c3SJan Kara vmf->pte = NULL; 52967267ec00SKirill A. Shutemov } 52977267ec00SKirill A. Shutemov } 52987267ec00SKirill A. Shutemov 52992bad466cSPeter Xu if (!vmf->pte) 53002bad466cSPeter Xu return do_pte_missing(vmf); 53017267ec00SKirill A. Shutemov 53022994302bSJan Kara if (!pte_present(vmf->orig_pte)) 53032994302bSJan Kara return do_swap_page(vmf); 53041da177e4SLinus Torvalds 53052994302bSJan Kara if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) 53062994302bSJan Kara return do_numa_page(vmf); 5307d10e63f2SMel Gorman 530882b0f8c3SJan Kara spin_lock(vmf->ptl); 53092994302bSJan Kara entry = vmf->orig_pte; 5310c33c7948SRyan Roberts if (unlikely(!pte_same(ptep_get(vmf->pte), entry))) { 53117df67697SBibo Mao update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 53128f4e2101SHugh Dickins goto unlock; 53137df67697SBibo Mao } 5314c89357e2SDavid Hildenbrand if (vmf->flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) { 5315f6f37321SLinus Torvalds if (!pte_write(entry)) 53162994302bSJan Kara return do_wp_page(vmf); 5317c89357e2SDavid Hildenbrand else if (likely(vmf->flags & FAULT_FLAG_WRITE)) 53181da177e4SLinus Torvalds entry = pte_mkdirty(entry); 53191da177e4SLinus Torvalds } 53201da177e4SLinus Torvalds entry = pte_mkyoung(entry); 532182b0f8c3SJan Kara if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, 532282b0f8c3SJan Kara vmf->flags & FAULT_FLAG_WRITE)) { 53235003a2bdSMatthew Wilcox (Oracle) update_mmu_cache_range(vmf, vmf->vma, vmf->address, 53245003a2bdSMatthew Wilcox (Oracle) vmf->pte, 1); 53251a44e149SAndrea Arcangeli } else { 5326b7333b58SYang Shi /* Skip spurious TLB flush for retried page fault */ 5327b7333b58SYang Shi if (vmf->flags & FAULT_FLAG_TRIED) 5328b7333b58SYang Shi goto unlock; 53291a44e149SAndrea Arcangeli /* 53301a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 53311a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 53321a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 53331a44e149SAndrea Arcangeli * with threads. 53341a44e149SAndrea Arcangeli */ 533582b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) 533699c29133SGerald Schaefer flush_tlb_fix_spurious_fault(vmf->vma, vmf->address, 533799c29133SGerald Schaefer vmf->pte); 53381a44e149SAndrea Arcangeli } 53398f4e2101SHugh Dickins unlock: 534082b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 534183c54070SNick Piggin return 0; 53421da177e4SLinus Torvalds } 53431da177e4SLinus Torvalds 53441da177e4SLinus Torvalds /* 53454ec31152SMatthew Wilcox (Oracle) * On entry, we hold either the VMA lock or the mmap_lock 53464ec31152SMatthew Wilcox (Oracle) * (FAULT_FLAG_VMA_LOCK tells you which). If VM_FAULT_RETRY is set in 53474ec31152SMatthew Wilcox (Oracle) * the result, the mmap_lock is not held on exit. See filemap_fault() 53484ec31152SMatthew Wilcox (Oracle) * and __folio_lock_or_retry(). 53491da177e4SLinus Torvalds */ 53502b740303SSouptick Joarder static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma, 53512b740303SSouptick Joarder unsigned long address, unsigned int flags) 53521da177e4SLinus Torvalds { 535382b0f8c3SJan Kara struct vm_fault vmf = { 5354bae473a4SKirill A. Shutemov .vma = vma, 53551a29d85eSJan Kara .address = address & PAGE_MASK, 5356824ddc60SNadav Amit .real_address = address, 5357bae473a4SKirill A. Shutemov .flags = flags, 53580721ec8bSJan Kara .pgoff = linear_page_index(vma, address), 5359667240e0SJan Kara .gfp_mask = __get_fault_gfp_mask(vma), 5360bae473a4SKirill A. Shutemov }; 5361dcddffd4SKirill A. Shutemov struct mm_struct *mm = vma->vm_mm; 53627da4e2cbSYang Shi unsigned long vm_flags = vma->vm_flags; 53631da177e4SLinus Torvalds pgd_t *pgd; 5364c2febafcSKirill A. Shutemov p4d_t *p4d; 53652b740303SSouptick Joarder vm_fault_t ret; 53661da177e4SLinus Torvalds 53671da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 5368c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, address); 5369c2febafcSKirill A. Shutemov if (!p4d) 5370c2febafcSKirill A. Shutemov return VM_FAULT_OOM; 5371a00cc7d9SMatthew Wilcox 5372c2febafcSKirill A. Shutemov vmf.pud = pud_alloc(mm, p4d, address); 5373a00cc7d9SMatthew Wilcox if (!vmf.pud) 5374c74df32cSHugh Dickins return VM_FAULT_OOM; 5375625110b5SThomas Hellstrom retry_pud: 53767da4e2cbSYang Shi if (pud_none(*vmf.pud) && 53773485b883SRyan Roberts thp_vma_allowable_order(vma, vm_flags, false, true, true, PUD_ORDER)) { 5378a00cc7d9SMatthew Wilcox ret = create_huge_pud(&vmf); 5379a00cc7d9SMatthew Wilcox if (!(ret & VM_FAULT_FALLBACK)) 5380a00cc7d9SMatthew Wilcox return ret; 5381a00cc7d9SMatthew Wilcox } else { 5382a00cc7d9SMatthew Wilcox pud_t orig_pud = *vmf.pud; 5383a00cc7d9SMatthew Wilcox 5384a00cc7d9SMatthew Wilcox barrier(); 5385a00cc7d9SMatthew Wilcox if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { 5386a00cc7d9SMatthew Wilcox 5387c89357e2SDavid Hildenbrand /* 5388c89357e2SDavid Hildenbrand * TODO once we support anonymous PUDs: NUMA case and 5389c89357e2SDavid Hildenbrand * FAULT_FLAG_UNSHARE handling. 5390c89357e2SDavid Hildenbrand */ 5391c89357e2SDavid Hildenbrand if ((flags & FAULT_FLAG_WRITE) && !pud_write(orig_pud)) { 5392a00cc7d9SMatthew Wilcox ret = wp_huge_pud(&vmf, orig_pud); 5393a00cc7d9SMatthew Wilcox if (!(ret & VM_FAULT_FALLBACK)) 5394a00cc7d9SMatthew Wilcox return ret; 5395a00cc7d9SMatthew Wilcox } else { 5396a00cc7d9SMatthew Wilcox huge_pud_set_accessed(&vmf, orig_pud); 5397a00cc7d9SMatthew Wilcox return 0; 5398a00cc7d9SMatthew Wilcox } 5399a00cc7d9SMatthew Wilcox } 5400a00cc7d9SMatthew Wilcox } 5401a00cc7d9SMatthew Wilcox 5402a00cc7d9SMatthew Wilcox vmf.pmd = pmd_alloc(mm, vmf.pud, address); 540382b0f8c3SJan Kara if (!vmf.pmd) 5404c74df32cSHugh Dickins return VM_FAULT_OOM; 5405625110b5SThomas Hellstrom 5406625110b5SThomas Hellstrom /* Huge pud page fault raced with pmd_alloc? */ 5407625110b5SThomas Hellstrom if (pud_trans_unstable(vmf.pud)) 5408625110b5SThomas Hellstrom goto retry_pud; 5409625110b5SThomas Hellstrom 54107da4e2cbSYang Shi if (pmd_none(*vmf.pmd) && 54113485b883SRyan Roberts thp_vma_allowable_order(vma, vm_flags, false, true, true, PMD_ORDER)) { 5412a2d58167SDave Jiang ret = create_huge_pmd(&vmf); 5413c0292554SKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 5414c0292554SKirill A. Shutemov return ret; 541571e3aac0SAndrea Arcangeli } else { 541626e1a0c3SHugh Dickins vmf.orig_pmd = pmdp_get_lockless(vmf.pmd); 54171f1d06c3SDavid Rientjes 54185db4f15cSYang Shi if (unlikely(is_swap_pmd(vmf.orig_pmd))) { 541984c3fc4eSZi Yan VM_BUG_ON(thp_migration_supported() && 54205db4f15cSYang Shi !is_pmd_migration_entry(vmf.orig_pmd)); 54215db4f15cSYang Shi if (is_pmd_migration_entry(vmf.orig_pmd)) 542284c3fc4eSZi Yan pmd_migration_entry_wait(mm, vmf.pmd); 542384c3fc4eSZi Yan return 0; 542484c3fc4eSZi Yan } 54255db4f15cSYang Shi if (pmd_trans_huge(vmf.orig_pmd) || pmd_devmap(vmf.orig_pmd)) { 54265db4f15cSYang Shi if (pmd_protnone(vmf.orig_pmd) && vma_is_accessible(vma)) 54275db4f15cSYang Shi return do_huge_pmd_numa_page(&vmf); 5428d10e63f2SMel Gorman 5429c89357e2SDavid Hildenbrand if ((flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) && 5430c89357e2SDavid Hildenbrand !pmd_write(vmf.orig_pmd)) { 54315db4f15cSYang Shi ret = wp_huge_pmd(&vmf); 54329845cbbdSKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 54331f1d06c3SDavid Rientjes return ret; 5434a1dd450bSWill Deacon } else { 54355db4f15cSYang Shi huge_pmd_set_accessed(&vmf); 543671e3aac0SAndrea Arcangeli return 0; 543771e3aac0SAndrea Arcangeli } 543871e3aac0SAndrea Arcangeli } 54399845cbbdSKirill A. Shutemov } 544071e3aac0SAndrea Arcangeli 544182b0f8c3SJan Kara return handle_pte_fault(&vmf); 54421da177e4SLinus Torvalds } 54431da177e4SLinus Torvalds 5444bce617edSPeter Xu /** 5445f0953a1bSIngo Molnar * mm_account_fault - Do page fault accounting 5446809ef83cSYang Li * @mm: mm from which memcg should be extracted. It can be NULL. 5447bce617edSPeter Xu * @regs: the pt_regs struct pointer. When set to NULL, will skip accounting 5448bce617edSPeter Xu * of perf event counters, but we'll still do the per-task accounting to 5449bce617edSPeter Xu * the task who triggered this page fault. 5450bce617edSPeter Xu * @address: the faulted address. 5451bce617edSPeter Xu * @flags: the fault flags. 5452bce617edSPeter Xu * @ret: the fault retcode. 5453bce617edSPeter Xu * 5454f0953a1bSIngo Molnar * This will take care of most of the page fault accounting. Meanwhile, it 5455bce617edSPeter Xu * will also include the PERF_COUNT_SW_PAGE_FAULTS_[MAJ|MIN] perf counter 5456f0953a1bSIngo Molnar * updates. However, note that the handling of PERF_COUNT_SW_PAGE_FAULTS should 5457bce617edSPeter Xu * still be in per-arch page fault handlers at the entry of page fault. 5458bce617edSPeter Xu */ 545953156443SSuren Baghdasaryan static inline void mm_account_fault(struct mm_struct *mm, struct pt_regs *regs, 5460bce617edSPeter Xu unsigned long address, unsigned int flags, 5461bce617edSPeter Xu vm_fault_t ret) 5462bce617edSPeter Xu { 5463bce617edSPeter Xu bool major; 5464bce617edSPeter Xu 546553156443SSuren Baghdasaryan /* Incomplete faults will be accounted upon completion. */ 546653156443SSuren Baghdasaryan if (ret & VM_FAULT_RETRY) 546753156443SSuren Baghdasaryan return; 546853156443SSuren Baghdasaryan 5469bce617edSPeter Xu /* 547053156443SSuren Baghdasaryan * To preserve the behavior of older kernels, PGFAULT counters record 547153156443SSuren Baghdasaryan * both successful and failed faults, as opposed to perf counters, 547253156443SSuren Baghdasaryan * which ignore failed cases. 5473bce617edSPeter Xu */ 547453156443SSuren Baghdasaryan count_vm_event(PGFAULT); 547553156443SSuren Baghdasaryan count_memcg_event_mm(mm, PGFAULT); 547653156443SSuren Baghdasaryan 547753156443SSuren Baghdasaryan /* 547853156443SSuren Baghdasaryan * Do not account for unsuccessful faults (e.g. when the address wasn't 547953156443SSuren Baghdasaryan * valid). That includes arch_vma_access_permitted() failing before 548053156443SSuren Baghdasaryan * reaching here. So this is not a "this many hardware page faults" 548153156443SSuren Baghdasaryan * counter. We should use the hw profiling for that. 548253156443SSuren Baghdasaryan */ 548353156443SSuren Baghdasaryan if (ret & VM_FAULT_ERROR) 5484bce617edSPeter Xu return; 5485bce617edSPeter Xu 5486bce617edSPeter Xu /* 5487bce617edSPeter Xu * We define the fault as a major fault when the final successful fault 5488bce617edSPeter Xu * is VM_FAULT_MAJOR, or if it retried (which implies that we couldn't 5489bce617edSPeter Xu * handle it immediately previously). 5490bce617edSPeter Xu */ 5491bce617edSPeter Xu major = (ret & VM_FAULT_MAJOR) || (flags & FAULT_FLAG_TRIED); 5492bce617edSPeter Xu 5493a2beb5f1SPeter Xu if (major) 5494a2beb5f1SPeter Xu current->maj_flt++; 5495a2beb5f1SPeter Xu else 5496a2beb5f1SPeter Xu current->min_flt++; 5497a2beb5f1SPeter Xu 5498bce617edSPeter Xu /* 5499a2beb5f1SPeter Xu * If the fault is done for GUP, regs will be NULL. We only do the 5500a2beb5f1SPeter Xu * accounting for the per thread fault counters who triggered the 5501a2beb5f1SPeter Xu * fault, and we skip the perf event updates. 5502bce617edSPeter Xu */ 5503bce617edSPeter Xu if (!regs) 5504bce617edSPeter Xu return; 5505bce617edSPeter Xu 5506a2beb5f1SPeter Xu if (major) 5507bce617edSPeter Xu perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 5508a2beb5f1SPeter Xu else 5509bce617edSPeter Xu perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 5510bce617edSPeter Xu } 5511bce617edSPeter Xu 5512ec1c86b2SYu Zhao #ifdef CONFIG_LRU_GEN 5513ec1c86b2SYu Zhao static void lru_gen_enter_fault(struct vm_area_struct *vma) 5514ec1c86b2SYu Zhao { 55158788f678SYu Zhao /* the LRU algorithm only applies to accesses with recency */ 55168788f678SYu Zhao current->in_lru_fault = vma_has_recency(vma); 5517ec1c86b2SYu Zhao } 5518ec1c86b2SYu Zhao 5519ec1c86b2SYu Zhao static void lru_gen_exit_fault(void) 5520ec1c86b2SYu Zhao { 5521ec1c86b2SYu Zhao current->in_lru_fault = false; 5522ec1c86b2SYu Zhao } 5523ec1c86b2SYu Zhao #else 5524ec1c86b2SYu Zhao static void lru_gen_enter_fault(struct vm_area_struct *vma) 5525ec1c86b2SYu Zhao { 5526ec1c86b2SYu Zhao } 5527ec1c86b2SYu Zhao 5528ec1c86b2SYu Zhao static void lru_gen_exit_fault(void) 5529ec1c86b2SYu Zhao { 5530ec1c86b2SYu Zhao } 5531ec1c86b2SYu Zhao #endif /* CONFIG_LRU_GEN */ 5532ec1c86b2SYu Zhao 5533cdc5021cSDavid Hildenbrand static vm_fault_t sanitize_fault_flags(struct vm_area_struct *vma, 5534cdc5021cSDavid Hildenbrand unsigned int *flags) 5535cdc5021cSDavid Hildenbrand { 5536cdc5021cSDavid Hildenbrand if (unlikely(*flags & FAULT_FLAG_UNSHARE)) { 5537cdc5021cSDavid Hildenbrand if (WARN_ON_ONCE(*flags & FAULT_FLAG_WRITE)) 5538cdc5021cSDavid Hildenbrand return VM_FAULT_SIGSEGV; 5539cdc5021cSDavid Hildenbrand /* 5540cdc5021cSDavid Hildenbrand * FAULT_FLAG_UNSHARE only applies to COW mappings. Let's 5541cdc5021cSDavid Hildenbrand * just treat it like an ordinary read-fault otherwise. 5542cdc5021cSDavid Hildenbrand */ 5543cdc5021cSDavid Hildenbrand if (!is_cow_mapping(vma->vm_flags)) 5544cdc5021cSDavid Hildenbrand *flags &= ~FAULT_FLAG_UNSHARE; 554579881fedSDavid Hildenbrand } else if (*flags & FAULT_FLAG_WRITE) { 554679881fedSDavid Hildenbrand /* Write faults on read-only mappings are impossible ... */ 554779881fedSDavid Hildenbrand if (WARN_ON_ONCE(!(vma->vm_flags & VM_MAYWRITE))) 554879881fedSDavid Hildenbrand return VM_FAULT_SIGSEGV; 554979881fedSDavid Hildenbrand /* ... and FOLL_FORCE only applies to COW mappings. */ 555079881fedSDavid Hildenbrand if (WARN_ON_ONCE(!(vma->vm_flags & VM_WRITE) && 555179881fedSDavid Hildenbrand !is_cow_mapping(vma->vm_flags))) 555279881fedSDavid Hildenbrand return VM_FAULT_SIGSEGV; 5553cdc5021cSDavid Hildenbrand } 55544089eef0SSuren Baghdasaryan #ifdef CONFIG_PER_VMA_LOCK 55554089eef0SSuren Baghdasaryan /* 55564089eef0SSuren Baghdasaryan * Per-VMA locks can't be used with FAULT_FLAG_RETRY_NOWAIT because of 55574089eef0SSuren Baghdasaryan * the assumption that lock is dropped on VM_FAULT_RETRY. 55584089eef0SSuren Baghdasaryan */ 55594089eef0SSuren Baghdasaryan if (WARN_ON_ONCE((*flags & 55604089eef0SSuren Baghdasaryan (FAULT_FLAG_VMA_LOCK | FAULT_FLAG_RETRY_NOWAIT)) == 55614089eef0SSuren Baghdasaryan (FAULT_FLAG_VMA_LOCK | FAULT_FLAG_RETRY_NOWAIT))) 55624089eef0SSuren Baghdasaryan return VM_FAULT_SIGSEGV; 55634089eef0SSuren Baghdasaryan #endif 55644089eef0SSuren Baghdasaryan 5565cdc5021cSDavid Hildenbrand return 0; 5566cdc5021cSDavid Hildenbrand } 5567cdc5021cSDavid Hildenbrand 55689a95f3cfSPaul Cassella /* 55699a95f3cfSPaul Cassella * By the time we get here, we already hold the mm semaphore 55709a95f3cfSPaul Cassella * 5571c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 55729138e47eSMatthew Wilcox (Oracle) * return value. See filemap_fault() and __folio_lock_or_retry(). 55739a95f3cfSPaul Cassella */ 55742b740303SSouptick Joarder vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address, 5575bce617edSPeter Xu unsigned int flags, struct pt_regs *regs) 5576519e5247SJohannes Weiner { 557753156443SSuren Baghdasaryan /* If the fault handler drops the mmap_lock, vma may be freed */ 557853156443SSuren Baghdasaryan struct mm_struct *mm = vma->vm_mm; 55792b740303SSouptick Joarder vm_fault_t ret; 5580519e5247SJohannes Weiner 5581519e5247SJohannes Weiner __set_current_state(TASK_RUNNING); 5582519e5247SJohannes Weiner 5583cdc5021cSDavid Hildenbrand ret = sanitize_fault_flags(vma, &flags); 5584cdc5021cSDavid Hildenbrand if (ret) 558553156443SSuren Baghdasaryan goto out; 5586cdc5021cSDavid Hildenbrand 5587de0c799bSLaurent Dufour if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, 5588de0c799bSLaurent Dufour flags & FAULT_FLAG_INSTRUCTION, 558953156443SSuren Baghdasaryan flags & FAULT_FLAG_REMOTE)) { 559053156443SSuren Baghdasaryan ret = VM_FAULT_SIGSEGV; 559153156443SSuren Baghdasaryan goto out; 559253156443SSuren Baghdasaryan } 5593de0c799bSLaurent Dufour 5594519e5247SJohannes Weiner /* 5595519e5247SJohannes Weiner * Enable the memcg OOM handling for faults triggered in user 5596519e5247SJohannes Weiner * space. Kernel faults are handled more gracefully. 5597519e5247SJohannes Weiner */ 5598519e5247SJohannes Weiner if (flags & FAULT_FLAG_USER) 559929ef680aSMichal Hocko mem_cgroup_enter_user_fault(); 5600519e5247SJohannes Weiner 5601ec1c86b2SYu Zhao lru_gen_enter_fault(vma); 5602ec1c86b2SYu Zhao 5603bae473a4SKirill A. Shutemov if (unlikely(is_vm_hugetlb_page(vma))) 5604bae473a4SKirill A. Shutemov ret = hugetlb_fault(vma->vm_mm, vma, address, flags); 5605bae473a4SKirill A. Shutemov else 5606dcddffd4SKirill A. Shutemov ret = __handle_mm_fault(vma, address, flags); 5607519e5247SJohannes Weiner 5608ec1c86b2SYu Zhao lru_gen_exit_fault(); 5609ec1c86b2SYu Zhao 561049426420SJohannes Weiner if (flags & FAULT_FLAG_USER) { 561129ef680aSMichal Hocko mem_cgroup_exit_user_fault(); 561249426420SJohannes Weiner /* 561349426420SJohannes Weiner * The task may have entered a memcg OOM situation but 561449426420SJohannes Weiner * if the allocation error was handled gracefully (no 561549426420SJohannes Weiner * VM_FAULT_OOM), there is no need to kill anything. 561649426420SJohannes Weiner * Just clean up the OOM state peacefully. 561749426420SJohannes Weiner */ 561849426420SJohannes Weiner if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) 561949426420SJohannes Weiner mem_cgroup_oom_synchronize(false); 562049426420SJohannes Weiner } 562153156443SSuren Baghdasaryan out: 562253156443SSuren Baghdasaryan mm_account_fault(mm, regs, address, flags, ret); 5623bce617edSPeter Xu 5624519e5247SJohannes Weiner return ret; 5625519e5247SJohannes Weiner } 5626e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(handle_mm_fault); 5627519e5247SJohannes Weiner 5628c2508ec5SLinus Torvalds #ifdef CONFIG_LOCK_MM_AND_FIND_VMA 5629c2508ec5SLinus Torvalds #include <linux/extable.h> 5630c2508ec5SLinus Torvalds 5631c2508ec5SLinus Torvalds static inline bool get_mmap_lock_carefully(struct mm_struct *mm, struct pt_regs *regs) 5632c2508ec5SLinus Torvalds { 56334542057eSLinus Torvalds if (likely(mmap_read_trylock(mm))) 5634c2508ec5SLinus Torvalds return true; 5635c2508ec5SLinus Torvalds 5636c2508ec5SLinus Torvalds if (regs && !user_mode(regs)) { 56378fa50708SJiaxun Yang unsigned long ip = exception_ip(regs); 5638c2508ec5SLinus Torvalds if (!search_exception_tables(ip)) 5639c2508ec5SLinus Torvalds return false; 5640c2508ec5SLinus Torvalds } 5641c2508ec5SLinus Torvalds 5642eda00472SLinus Torvalds return !mmap_read_lock_killable(mm); 5643c2508ec5SLinus Torvalds } 5644c2508ec5SLinus Torvalds 5645c2508ec5SLinus Torvalds static inline bool mmap_upgrade_trylock(struct mm_struct *mm) 5646c2508ec5SLinus Torvalds { 5647c2508ec5SLinus Torvalds /* 5648c2508ec5SLinus Torvalds * We don't have this operation yet. 5649c2508ec5SLinus Torvalds * 5650c2508ec5SLinus Torvalds * It should be easy enough to do: it's basically a 5651c2508ec5SLinus Torvalds * atomic_long_try_cmpxchg_acquire() 5652c2508ec5SLinus Torvalds * from RWSEM_READER_BIAS -> RWSEM_WRITER_LOCKED, but 5653c2508ec5SLinus Torvalds * it also needs the proper lockdep magic etc. 5654c2508ec5SLinus Torvalds */ 5655c2508ec5SLinus Torvalds return false; 5656c2508ec5SLinus Torvalds } 5657c2508ec5SLinus Torvalds 5658c2508ec5SLinus Torvalds static inline bool upgrade_mmap_lock_carefully(struct mm_struct *mm, struct pt_regs *regs) 5659c2508ec5SLinus Torvalds { 5660c2508ec5SLinus Torvalds mmap_read_unlock(mm); 5661c2508ec5SLinus Torvalds if (regs && !user_mode(regs)) { 56628fa50708SJiaxun Yang unsigned long ip = exception_ip(regs); 5663c2508ec5SLinus Torvalds if (!search_exception_tables(ip)) 5664c2508ec5SLinus Torvalds return false; 5665c2508ec5SLinus Torvalds } 5666eda00472SLinus Torvalds return !mmap_write_lock_killable(mm); 5667c2508ec5SLinus Torvalds } 5668c2508ec5SLinus Torvalds 5669c2508ec5SLinus Torvalds /* 5670c2508ec5SLinus Torvalds * Helper for page fault handling. 5671c2508ec5SLinus Torvalds * 5672c2508ec5SLinus Torvalds * This is kind of equivalend to "mmap_read_lock()" followed 5673c2508ec5SLinus Torvalds * by "find_extend_vma()", except it's a lot more careful about 5674c2508ec5SLinus Torvalds * the locking (and will drop the lock on failure). 5675c2508ec5SLinus Torvalds * 5676c2508ec5SLinus Torvalds * For example, if we have a kernel bug that causes a page 5677c2508ec5SLinus Torvalds * fault, we don't want to just use mmap_read_lock() to get 5678c2508ec5SLinus Torvalds * the mm lock, because that would deadlock if the bug were 5679c2508ec5SLinus Torvalds * to happen while we're holding the mm lock for writing. 5680c2508ec5SLinus Torvalds * 5681c2508ec5SLinus Torvalds * So this checks the exception tables on kernel faults in 5682c2508ec5SLinus Torvalds * order to only do this all for instructions that are actually 5683c2508ec5SLinus Torvalds * expected to fault. 5684c2508ec5SLinus Torvalds * 5685c2508ec5SLinus Torvalds * We can also actually take the mm lock for writing if we 5686c2508ec5SLinus Torvalds * need to extend the vma, which helps the VM layer a lot. 5687c2508ec5SLinus Torvalds */ 5688c2508ec5SLinus Torvalds struct vm_area_struct *lock_mm_and_find_vma(struct mm_struct *mm, 5689c2508ec5SLinus Torvalds unsigned long addr, struct pt_regs *regs) 5690c2508ec5SLinus Torvalds { 5691c2508ec5SLinus Torvalds struct vm_area_struct *vma; 5692c2508ec5SLinus Torvalds 5693c2508ec5SLinus Torvalds if (!get_mmap_lock_carefully(mm, regs)) 5694c2508ec5SLinus Torvalds return NULL; 5695c2508ec5SLinus Torvalds 5696c2508ec5SLinus Torvalds vma = find_vma(mm, addr); 5697c2508ec5SLinus Torvalds if (likely(vma && (vma->vm_start <= addr))) 5698c2508ec5SLinus Torvalds return vma; 5699c2508ec5SLinus Torvalds 5700c2508ec5SLinus Torvalds /* 5701c2508ec5SLinus Torvalds * Well, dang. We might still be successful, but only 5702c2508ec5SLinus Torvalds * if we can extend a vma to do so. 5703c2508ec5SLinus Torvalds */ 5704c2508ec5SLinus Torvalds if (!vma || !(vma->vm_flags & VM_GROWSDOWN)) { 5705c2508ec5SLinus Torvalds mmap_read_unlock(mm); 5706c2508ec5SLinus Torvalds return NULL; 5707c2508ec5SLinus Torvalds } 5708c2508ec5SLinus Torvalds 5709c2508ec5SLinus Torvalds /* 5710c2508ec5SLinus Torvalds * We can try to upgrade the mmap lock atomically, 5711c2508ec5SLinus Torvalds * in which case we can continue to use the vma 5712c2508ec5SLinus Torvalds * we already looked up. 5713c2508ec5SLinus Torvalds * 5714c2508ec5SLinus Torvalds * Otherwise we'll have to drop the mmap lock and 5715c2508ec5SLinus Torvalds * re-take it, and also look up the vma again, 5716c2508ec5SLinus Torvalds * re-checking it. 5717c2508ec5SLinus Torvalds */ 5718c2508ec5SLinus Torvalds if (!mmap_upgrade_trylock(mm)) { 5719c2508ec5SLinus Torvalds if (!upgrade_mmap_lock_carefully(mm, regs)) 5720c2508ec5SLinus Torvalds return NULL; 5721c2508ec5SLinus Torvalds 5722c2508ec5SLinus Torvalds vma = find_vma(mm, addr); 5723c2508ec5SLinus Torvalds if (!vma) 5724c2508ec5SLinus Torvalds goto fail; 5725c2508ec5SLinus Torvalds if (vma->vm_start <= addr) 5726c2508ec5SLinus Torvalds goto success; 5727c2508ec5SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 5728c2508ec5SLinus Torvalds goto fail; 5729c2508ec5SLinus Torvalds } 5730c2508ec5SLinus Torvalds 57318d7071afSLinus Torvalds if (expand_stack_locked(vma, addr)) 5732c2508ec5SLinus Torvalds goto fail; 5733c2508ec5SLinus Torvalds 5734c2508ec5SLinus Torvalds success: 5735c2508ec5SLinus Torvalds mmap_write_downgrade(mm); 5736c2508ec5SLinus Torvalds return vma; 5737c2508ec5SLinus Torvalds 5738c2508ec5SLinus Torvalds fail: 5739c2508ec5SLinus Torvalds mmap_write_unlock(mm); 5740c2508ec5SLinus Torvalds return NULL; 5741c2508ec5SLinus Torvalds } 5742c2508ec5SLinus Torvalds #endif 5743c2508ec5SLinus Torvalds 574450ee3253SSuren Baghdasaryan #ifdef CONFIG_PER_VMA_LOCK 574550ee3253SSuren Baghdasaryan /* 574650ee3253SSuren Baghdasaryan * Lookup and lock a VMA under RCU protection. Returned VMA is guaranteed to be 574750ee3253SSuren Baghdasaryan * stable and not isolated. If the VMA is not found or is being modified the 574850ee3253SSuren Baghdasaryan * function returns NULL. 574950ee3253SSuren Baghdasaryan */ 575050ee3253SSuren Baghdasaryan struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm, 575150ee3253SSuren Baghdasaryan unsigned long address) 575250ee3253SSuren Baghdasaryan { 575350ee3253SSuren Baghdasaryan MA_STATE(mas, &mm->mm_mt, address, address); 575450ee3253SSuren Baghdasaryan struct vm_area_struct *vma; 575550ee3253SSuren Baghdasaryan 575650ee3253SSuren Baghdasaryan rcu_read_lock(); 575750ee3253SSuren Baghdasaryan retry: 575850ee3253SSuren Baghdasaryan vma = mas_walk(&mas); 575950ee3253SSuren Baghdasaryan if (!vma) 576050ee3253SSuren Baghdasaryan goto inval; 576150ee3253SSuren Baghdasaryan 576250ee3253SSuren Baghdasaryan if (!vma_start_read(vma)) 576350ee3253SSuren Baghdasaryan goto inval; 576450ee3253SSuren Baghdasaryan 5765444eeb17SSuren Baghdasaryan /* 5766657b5146SJann Horn * find_mergeable_anon_vma uses adjacent vmas which are not locked. 5767657b5146SJann Horn * This check must happen after vma_start_read(); otherwise, a 5768657b5146SJann Horn * concurrent mremap() with MREMAP_DONTUNMAP could dissociate the VMA 5769657b5146SJann Horn * from its anon_vma. 5770657b5146SJann Horn */ 577129a22b9eSSuren Baghdasaryan if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) 5772657b5146SJann Horn goto inval_end_read; 5773444eeb17SSuren Baghdasaryan 577450ee3253SSuren Baghdasaryan /* Check since vm_start/vm_end might change before we lock the VMA */ 5775657b5146SJann Horn if (unlikely(address < vma->vm_start || address >= vma->vm_end)) 5776657b5146SJann Horn goto inval_end_read; 577750ee3253SSuren Baghdasaryan 577850ee3253SSuren Baghdasaryan /* Check if the VMA got isolated after we found it */ 577950ee3253SSuren Baghdasaryan if (vma->detached) { 578050ee3253SSuren Baghdasaryan vma_end_read(vma); 578152f23865SSuren Baghdasaryan count_vm_vma_lock_event(VMA_LOCK_MISS); 578250ee3253SSuren Baghdasaryan /* The area was replaced with another one */ 578350ee3253SSuren Baghdasaryan goto retry; 578450ee3253SSuren Baghdasaryan } 578550ee3253SSuren Baghdasaryan 578650ee3253SSuren Baghdasaryan rcu_read_unlock(); 578750ee3253SSuren Baghdasaryan return vma; 5788657b5146SJann Horn 5789657b5146SJann Horn inval_end_read: 5790657b5146SJann Horn vma_end_read(vma); 579150ee3253SSuren Baghdasaryan inval: 579250ee3253SSuren Baghdasaryan rcu_read_unlock(); 579352f23865SSuren Baghdasaryan count_vm_vma_lock_event(VMA_LOCK_ABORT); 579450ee3253SSuren Baghdasaryan return NULL; 579550ee3253SSuren Baghdasaryan } 579650ee3253SSuren Baghdasaryan #endif /* CONFIG_PER_VMA_LOCK */ 579750ee3253SSuren Baghdasaryan 579890eceff1SKirill A. Shutemov #ifndef __PAGETABLE_P4D_FOLDED 579990eceff1SKirill A. Shutemov /* 580090eceff1SKirill A. Shutemov * Allocate p4d page table. 580190eceff1SKirill A. Shutemov * We've already handled the fast-path in-line. 580290eceff1SKirill A. Shutemov */ 580390eceff1SKirill A. Shutemov int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 580490eceff1SKirill A. Shutemov { 580590eceff1SKirill A. Shutemov p4d_t *new = p4d_alloc_one(mm, address); 580690eceff1SKirill A. Shutemov if (!new) 580790eceff1SKirill A. Shutemov return -ENOMEM; 580890eceff1SKirill A. Shutemov 580990eceff1SKirill A. Shutemov spin_lock(&mm->page_table_lock); 5810ed33b5a6SQi Zheng if (pgd_present(*pgd)) { /* Another has populated it */ 581190eceff1SKirill A. Shutemov p4d_free(mm, new); 5812ed33b5a6SQi Zheng } else { 5813ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 581490eceff1SKirill A. Shutemov pgd_populate(mm, pgd, new); 5815ed33b5a6SQi Zheng } 581690eceff1SKirill A. Shutemov spin_unlock(&mm->page_table_lock); 581790eceff1SKirill A. Shutemov return 0; 581890eceff1SKirill A. Shutemov } 581990eceff1SKirill A. Shutemov #endif /* __PAGETABLE_P4D_FOLDED */ 582090eceff1SKirill A. Shutemov 58211da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 58221da177e4SLinus Torvalds /* 58231da177e4SLinus Torvalds * Allocate page upper directory. 5824872fec16SHugh Dickins * We've already handled the fast-path in-line. 58251da177e4SLinus Torvalds */ 5826c2febafcSKirill A. Shutemov int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) 58271da177e4SLinus Torvalds { 5828c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 5829c74df32cSHugh Dickins if (!new) 58301bb3630eSHugh Dickins return -ENOMEM; 58311da177e4SLinus Torvalds 5832872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 5833b4e98d9aSKirill A. Shutemov if (!p4d_present(*p4d)) { 5834b4e98d9aSKirill A. Shutemov mm_inc_nr_puds(mm); 5835ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 5836c2febafcSKirill A. Shutemov p4d_populate(mm, p4d, new); 5837b4e98d9aSKirill A. Shutemov } else /* Another has populated it */ 5838c2febafcSKirill A. Shutemov pud_free(mm, new); 5839872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 58401bb3630eSHugh Dickins return 0; 58411da177e4SLinus Torvalds } 58421da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 58431da177e4SLinus Torvalds 58441da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 58451da177e4SLinus Torvalds /* 58461da177e4SLinus Torvalds * Allocate page middle directory. 5847872fec16SHugh Dickins * We've already handled the fast-path in-line. 58481da177e4SLinus Torvalds */ 58491bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 58501da177e4SLinus Torvalds { 5851a00cc7d9SMatthew Wilcox spinlock_t *ptl; 5852c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 5853c74df32cSHugh Dickins if (!new) 58541bb3630eSHugh Dickins return -ENOMEM; 58551da177e4SLinus Torvalds 5856a00cc7d9SMatthew Wilcox ptl = pud_lock(mm, pud); 5857dc6c9a35SKirill A. Shutemov if (!pud_present(*pud)) { 5858dc6c9a35SKirill A. Shutemov mm_inc_nr_pmds(mm); 5859ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 58601da177e4SLinus Torvalds pud_populate(mm, pud, new); 5861ed33b5a6SQi Zheng } else { /* Another has populated it */ 58625e541973SBenjamin Herrenschmidt pmd_free(mm, new); 5863ed33b5a6SQi Zheng } 5864a00cc7d9SMatthew Wilcox spin_unlock(ptl); 58651bb3630eSHugh Dickins return 0; 58661da177e4SLinus Torvalds } 58671da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 58681da177e4SLinus Torvalds 58693b6748e2SJohannes Weiner /** 58709fd6dad1SPaolo Bonzini * follow_pte - look up PTE at a user virtual address 58719fd6dad1SPaolo Bonzini * @mm: the mm_struct of the target address space 58729fd6dad1SPaolo Bonzini * @address: user virtual address 58739fd6dad1SPaolo Bonzini * @ptepp: location to store found PTE 58749fd6dad1SPaolo Bonzini * @ptlp: location to store the lock for the PTE 58759fd6dad1SPaolo Bonzini * 58769fd6dad1SPaolo Bonzini * On a successful return, the pointer to the PTE is stored in @ptepp; 58779fd6dad1SPaolo Bonzini * the corresponding lock is taken and its location is stored in @ptlp. 58789fd6dad1SPaolo Bonzini * The contents of the PTE are only stable until @ptlp is released; 58799fd6dad1SPaolo Bonzini * any further use, if any, must be protected against invalidation 58809fd6dad1SPaolo Bonzini * with MMU notifiers. 58819fd6dad1SPaolo Bonzini * 58829fd6dad1SPaolo Bonzini * Only IO mappings and raw PFN mappings are allowed. The mmap semaphore 58839fd6dad1SPaolo Bonzini * should be taken for read. 58849fd6dad1SPaolo Bonzini * 58859fd6dad1SPaolo Bonzini * KVM uses this function. While it is arguably less bad than ``follow_pfn``, 58869fd6dad1SPaolo Bonzini * it is not a good general-purpose API. 58879fd6dad1SPaolo Bonzini * 58889fd6dad1SPaolo Bonzini * Return: zero on success, -ve otherwise. 58899fd6dad1SPaolo Bonzini */ 58909fd6dad1SPaolo Bonzini int follow_pte(struct mm_struct *mm, unsigned long address, 58919fd6dad1SPaolo Bonzini pte_t **ptepp, spinlock_t **ptlp) 58929fd6dad1SPaolo Bonzini { 58930e5e64c0SMuchun Song pgd_t *pgd; 58940e5e64c0SMuchun Song p4d_t *p4d; 58950e5e64c0SMuchun Song pud_t *pud; 58960e5e64c0SMuchun Song pmd_t *pmd; 58970e5e64c0SMuchun Song pte_t *ptep; 58980e5e64c0SMuchun Song 58990e5e64c0SMuchun Song pgd = pgd_offset(mm, address); 59000e5e64c0SMuchun Song if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 59010e5e64c0SMuchun Song goto out; 59020e5e64c0SMuchun Song 59030e5e64c0SMuchun Song p4d = p4d_offset(pgd, address); 59040e5e64c0SMuchun Song if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) 59050e5e64c0SMuchun Song goto out; 59060e5e64c0SMuchun Song 59070e5e64c0SMuchun Song pud = pud_offset(p4d, address); 59080e5e64c0SMuchun Song if (pud_none(*pud) || unlikely(pud_bad(*pud))) 59090e5e64c0SMuchun Song goto out; 59100e5e64c0SMuchun Song 59110e5e64c0SMuchun Song pmd = pmd_offset(pud, address); 59120e5e64c0SMuchun Song VM_BUG_ON(pmd_trans_huge(*pmd)); 59130e5e64c0SMuchun Song 59140e5e64c0SMuchun Song ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 59153db82b93SHugh Dickins if (!ptep) 59163db82b93SHugh Dickins goto out; 5917c33c7948SRyan Roberts if (!pte_present(ptep_get(ptep))) 59180e5e64c0SMuchun Song goto unlock; 59190e5e64c0SMuchun Song *ptepp = ptep; 59200e5e64c0SMuchun Song return 0; 59210e5e64c0SMuchun Song unlock: 59220e5e64c0SMuchun Song pte_unmap_unlock(ptep, *ptlp); 59230e5e64c0SMuchun Song out: 59240e5e64c0SMuchun Song return -EINVAL; 59259fd6dad1SPaolo Bonzini } 59269fd6dad1SPaolo Bonzini EXPORT_SYMBOL_GPL(follow_pte); 59279fd6dad1SPaolo Bonzini 59289fd6dad1SPaolo Bonzini /** 59293b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 59303b6748e2SJohannes Weiner * @vma: memory mapping 59313b6748e2SJohannes Weiner * @address: user virtual address 59323b6748e2SJohannes Weiner * @pfn: location to store found PFN 59333b6748e2SJohannes Weiner * 59343b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 59353b6748e2SJohannes Weiner * 59369fd6dad1SPaolo Bonzini * This function does not allow the caller to read the permissions 59379fd6dad1SPaolo Bonzini * of the PTE. Do not use it. 59389fd6dad1SPaolo Bonzini * 5939a862f68aSMike Rapoport * Return: zero and the pfn at @pfn on success, -ve otherwise. 59403b6748e2SJohannes Weiner */ 59413b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 59423b6748e2SJohannes Weiner unsigned long *pfn) 59433b6748e2SJohannes Weiner { 59443b6748e2SJohannes Weiner int ret = -EINVAL; 59453b6748e2SJohannes Weiner spinlock_t *ptl; 59463b6748e2SJohannes Weiner pte_t *ptep; 59473b6748e2SJohannes Weiner 59483b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 59493b6748e2SJohannes Weiner return ret; 59503b6748e2SJohannes Weiner 59519fd6dad1SPaolo Bonzini ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 59523b6748e2SJohannes Weiner if (ret) 59533b6748e2SJohannes Weiner return ret; 5954c33c7948SRyan Roberts *pfn = pte_pfn(ptep_get(ptep)); 59553b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 59563b6748e2SJohannes Weiner return 0; 59573b6748e2SJohannes Weiner } 59583b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 59593b6748e2SJohannes Weiner 596028b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 5961d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 596228b2ee20SRik van Riel unsigned long address, unsigned int flags, 5963d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 596428b2ee20SRik van Riel { 596503668a4dSJohannes Weiner int ret = -EINVAL; 596628b2ee20SRik van Riel pte_t *ptep, pte; 596728b2ee20SRik van Riel spinlock_t *ptl; 596828b2ee20SRik van Riel 5969d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 5970d87fe660Svenkatesh.pallipadi@intel.com goto out; 597128b2ee20SRik van Riel 59729fd6dad1SPaolo Bonzini if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 5973d87fe660Svenkatesh.pallipadi@intel.com goto out; 5974c33c7948SRyan Roberts pte = ptep_get(ptep); 597503668a4dSJohannes Weiner 5976*04c35ab3SDavid Hildenbrand /* Never return PFNs of anon folios in COW mappings. */ 5977*04c35ab3SDavid Hildenbrand if (vm_normal_folio(vma, address, pte)) 5978*04c35ab3SDavid Hildenbrand goto unlock; 5979*04c35ab3SDavid Hildenbrand 5980f6f37321SLinus Torvalds if ((flags & FOLL_WRITE) && !pte_write(pte)) 598128b2ee20SRik van Riel goto unlock; 598228b2ee20SRik van Riel 598328b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 598403668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 598528b2ee20SRik van Riel 598603668a4dSJohannes Weiner ret = 0; 598728b2ee20SRik van Riel unlock: 598828b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 598928b2ee20SRik van Riel out: 5990d87fe660Svenkatesh.pallipadi@intel.com return ret; 599128b2ee20SRik van Riel } 599228b2ee20SRik van Riel 599396667f8aSDaniel Vetter /** 599496667f8aSDaniel Vetter * generic_access_phys - generic implementation for iomem mmap access 599596667f8aSDaniel Vetter * @vma: the vma to access 5996f0953a1bSIngo Molnar * @addr: userspace address, not relative offset within @vma 599796667f8aSDaniel Vetter * @buf: buffer to read/write 599896667f8aSDaniel Vetter * @len: length of transfer 599996667f8aSDaniel Vetter * @write: set to FOLL_WRITE when writing, otherwise reading 600096667f8aSDaniel Vetter * 600196667f8aSDaniel Vetter * This is a generic implementation for &vm_operations_struct.access for an 600296667f8aSDaniel Vetter * iomem mapping. This callback is used by access_process_vm() when the @vma is 600396667f8aSDaniel Vetter * not page based. 600496667f8aSDaniel Vetter */ 600528b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 600628b2ee20SRik van Riel void *buf, int len, int write) 600728b2ee20SRik van Riel { 600828b2ee20SRik van Riel resource_size_t phys_addr; 600928b2ee20SRik van Riel unsigned long prot = 0; 60102bc7273bSKOSAKI Motohiro void __iomem *maddr; 601196667f8aSDaniel Vetter pte_t *ptep, pte; 601296667f8aSDaniel Vetter spinlock_t *ptl; 601396667f8aSDaniel Vetter int offset = offset_in_page(addr); 601496667f8aSDaniel Vetter int ret = -EINVAL; 601528b2ee20SRik van Riel 601696667f8aSDaniel Vetter if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 601796667f8aSDaniel Vetter return -EINVAL; 601896667f8aSDaniel Vetter 601996667f8aSDaniel Vetter retry: 6020e913a8cdSLinus Torvalds if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) 602196667f8aSDaniel Vetter return -EINVAL; 6022c33c7948SRyan Roberts pte = ptep_get(ptep); 602396667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 602496667f8aSDaniel Vetter 602596667f8aSDaniel Vetter prot = pgprot_val(pte_pgprot(pte)); 602696667f8aSDaniel Vetter phys_addr = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 602796667f8aSDaniel Vetter 602896667f8aSDaniel Vetter if ((write & FOLL_WRITE) && !pte_write(pte)) 602928b2ee20SRik van Riel return -EINVAL; 603028b2ee20SRik van Riel 60319cb12d7bSGrazvydas Ignotas maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); 603224eee1e4Sjie@chenjie6@huwei.com if (!maddr) 603324eee1e4Sjie@chenjie6@huwei.com return -ENOMEM; 603424eee1e4Sjie@chenjie6@huwei.com 6035e913a8cdSLinus Torvalds if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) 603696667f8aSDaniel Vetter goto out_unmap; 603796667f8aSDaniel Vetter 6038c33c7948SRyan Roberts if (!pte_same(pte, ptep_get(ptep))) { 603996667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 604096667f8aSDaniel Vetter iounmap(maddr); 604196667f8aSDaniel Vetter 604296667f8aSDaniel Vetter goto retry; 604396667f8aSDaniel Vetter } 604496667f8aSDaniel Vetter 604528b2ee20SRik van Riel if (write) 604628b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 604728b2ee20SRik van Riel else 604828b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 604996667f8aSDaniel Vetter ret = len; 605096667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 605196667f8aSDaniel Vetter out_unmap: 605228b2ee20SRik van Riel iounmap(maddr); 605328b2ee20SRik van Riel 605496667f8aSDaniel Vetter return ret; 605528b2ee20SRik van Riel } 60565a73633eSUwe Kleine-König EXPORT_SYMBOL_GPL(generic_access_phys); 605728b2ee20SRik van Riel #endif 605828b2ee20SRik van Riel 60590ec76a11SDavid Howells /* 6060d3f5ffcaSJohn Hubbard * Access another process' address space as given in mm. 60610ec76a11SDavid Howells */ 6062c43cfa42SLorenzo Stoakes static int __access_remote_vm(struct mm_struct *mm, unsigned long addr, 6063c43cfa42SLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 60640ec76a11SDavid Howells { 60650ec76a11SDavid Howells void *old_buf = buf; 6066442486ecSLorenzo Stoakes int write = gup_flags & FOLL_WRITE; 60670ec76a11SDavid Howells 6068d8ed45c5SMichel Lespinasse if (mmap_read_lock_killable(mm)) 60691e426fe2SKonstantin Khlebnikov return 0; 60701e426fe2SKonstantin Khlebnikov 607122883973SKirill A. Shutemov /* Untag the address before looking up the VMA */ 607222883973SKirill A. Shutemov addr = untagged_addr_remote(mm, addr); 607322883973SKirill A. Shutemov 6074eee9c708SLinus Torvalds /* Avoid triggering the temporary warning in __get_user_pages */ 6075eee9c708SLinus Torvalds if (!vma_lookup(mm, addr) && !expand_stack(mm, addr)) 6076eee9c708SLinus Torvalds return 0; 6077eee9c708SLinus Torvalds 6078183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 60790ec76a11SDavid Howells while (len) { 6080ca5e8632SLorenzo Stoakes int bytes, offset; 60810ec76a11SDavid Howells void *maddr; 6082ca5e8632SLorenzo Stoakes struct vm_area_struct *vma = NULL; 6083ca5e8632SLorenzo Stoakes struct page *page = get_user_page_vma_remote(mm, addr, 6084ca5e8632SLorenzo Stoakes gup_flags, &vma); 60850ec76a11SDavid Howells 60866a1960b8SLorenzo Stoakes if (IS_ERR(page)) { 60878d7071afSLinus Torvalds /* We might need to expand the stack to access it */ 60888d7071afSLinus Torvalds vma = vma_lookup(mm, addr); 60898d7071afSLinus Torvalds if (!vma) { 60908d7071afSLinus Torvalds vma = expand_stack(mm, addr); 60919471f1f2SLinus Torvalds 60929471f1f2SLinus Torvalds /* mmap_lock was dropped on failure */ 60938d7071afSLinus Torvalds if (!vma) 60949471f1f2SLinus Torvalds return buf - old_buf; 60959471f1f2SLinus Torvalds 60969471f1f2SLinus Torvalds /* Try again if stack expansion worked */ 60979471f1f2SLinus Torvalds continue; 60988d7071afSLinus Torvalds } 60998d7071afSLinus Torvalds 610028b2ee20SRik van Riel /* 610128b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 610228b2ee20SRik van Riel * we can access using slightly different code. 610328b2ee20SRik van Riel */ 61049471f1f2SLinus Torvalds bytes = 0; 61059471f1f2SLinus Torvalds #ifdef CONFIG_HAVE_IOREMAP_PROT 610628b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 61079471f1f2SLinus Torvalds bytes = vma->vm_ops->access(vma, addr, buf, 610828b2ee20SRik van Riel len, write); 6109dbffcd03SRik van Riel #endif 61109471f1f2SLinus Torvalds if (bytes <= 0) 61119471f1f2SLinus Torvalds break; 611228b2ee20SRik van Riel } else { 61130ec76a11SDavid Howells bytes = len; 61140ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 61150ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 61160ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 61170ec76a11SDavid Howells 6118f7ef5fe7SFabio M. De Francesco maddr = kmap_local_page(page); 61190ec76a11SDavid Howells if (write) { 61200ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 61210ec76a11SDavid Howells maddr + offset, buf, bytes); 61220ec76a11SDavid Howells set_page_dirty_lock(page); 61230ec76a11SDavid Howells } else { 61240ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 61250ec76a11SDavid Howells buf, maddr + offset, bytes); 61260ec76a11SDavid Howells } 6127f7ef5fe7SFabio M. De Francesco unmap_and_put_page(page, maddr); 612828b2ee20SRik van Riel } 61290ec76a11SDavid Howells len -= bytes; 61300ec76a11SDavid Howells buf += bytes; 61310ec76a11SDavid Howells addr += bytes; 61320ec76a11SDavid Howells } 6133d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 61340ec76a11SDavid Howells 61350ec76a11SDavid Howells return buf - old_buf; 61360ec76a11SDavid Howells } 613703252919SAndi Kleen 61385ddd36b9SStephen Wilson /** 6139ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 61405ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 61415ddd36b9SStephen Wilson * @addr: start address to access 61425ddd36b9SStephen Wilson * @buf: source or destination buffer 61435ddd36b9SStephen Wilson * @len: number of bytes to transfer 61446347e8d5SLorenzo Stoakes * @gup_flags: flags modifying lookup behaviour 61455ddd36b9SStephen Wilson * 61465ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 6147a862f68aSMike Rapoport * 6148a862f68aSMike Rapoport * Return: number of bytes copied from source to destination. 61495ddd36b9SStephen Wilson */ 61505ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 61516347e8d5SLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 61525ddd36b9SStephen Wilson { 6153d3f5ffcaSJohn Hubbard return __access_remote_vm(mm, addr, buf, len, gup_flags); 61545ddd36b9SStephen Wilson } 61555ddd36b9SStephen Wilson 615603252919SAndi Kleen /* 6157206cb636SStephen Wilson * Access another process' address space. 6158206cb636SStephen Wilson * Source/target buffer must be kernel space, 6159206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 6160206cb636SStephen Wilson */ 6161206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 6162f307ab6dSLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 6163206cb636SStephen Wilson { 6164206cb636SStephen Wilson struct mm_struct *mm; 6165206cb636SStephen Wilson int ret; 6166206cb636SStephen Wilson 6167206cb636SStephen Wilson mm = get_task_mm(tsk); 6168206cb636SStephen Wilson if (!mm) 6169206cb636SStephen Wilson return 0; 6170206cb636SStephen Wilson 6171d3f5ffcaSJohn Hubbard ret = __access_remote_vm(mm, addr, buf, len, gup_flags); 6172442486ecSLorenzo Stoakes 6173206cb636SStephen Wilson mmput(mm); 6174206cb636SStephen Wilson 6175206cb636SStephen Wilson return ret; 6176206cb636SStephen Wilson } 6177fcd35857SCatalin Marinas EXPORT_SYMBOL_GPL(access_process_vm); 6178206cb636SStephen Wilson 617903252919SAndi Kleen /* 618003252919SAndi Kleen * Print the name of a VMA. 618103252919SAndi Kleen */ 618203252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 618303252919SAndi Kleen { 618403252919SAndi Kleen struct mm_struct *mm = current->mm; 618503252919SAndi Kleen struct vm_area_struct *vma; 618603252919SAndi Kleen 6187e8bff74aSIngo Molnar /* 61880a7f682dSMichal Hocko * we might be running from an atomic context so we cannot sleep 6189e8bff74aSIngo Molnar */ 6190d8ed45c5SMichel Lespinasse if (!mmap_read_trylock(mm)) 6191e8bff74aSIngo Molnar return; 6192e8bff74aSIngo Molnar 619303252919SAndi Kleen vma = find_vma(mm, ip); 619403252919SAndi Kleen if (vma && vma->vm_file) { 619503252919SAndi Kleen struct file *f = vma->vm_file; 61960a7f682dSMichal Hocko char *buf = (char *)__get_free_page(GFP_NOWAIT); 619703252919SAndi Kleen if (buf) { 61982fbc57c5SAndy Shevchenko char *p; 619903252919SAndi Kleen 62009bf39ab2SMiklos Szeredi p = file_path(f, buf, PAGE_SIZE); 620103252919SAndi Kleen if (IS_ERR(p)) 620203252919SAndi Kleen p = "?"; 62032fbc57c5SAndy Shevchenko printk("%s%s[%lx+%lx]", prefix, kbasename(p), 620403252919SAndi Kleen vma->vm_start, 620503252919SAndi Kleen vma->vm_end - vma->vm_start); 620603252919SAndi Kleen free_page((unsigned long)buf); 620703252919SAndi Kleen } 620803252919SAndi Kleen } 6209d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 621003252919SAndi Kleen } 62113ee1afa3SNick Piggin 6212662bbcb2SMichael S. Tsirkin #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) 62139ec23531SDavid Hildenbrand void __might_fault(const char *file, int line) 62143ee1afa3SNick Piggin { 62159ec23531SDavid Hildenbrand if (pagefault_disabled()) 6216662bbcb2SMichael S. Tsirkin return; 621742a38756SThomas Gleixner __might_sleep(file, line); 62189ec23531SDavid Hildenbrand #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) 6219662bbcb2SMichael S. Tsirkin if (current->mm) 6220da1c55f1SMichel Lespinasse might_lock_read(¤t->mm->mmap_lock); 62219ec23531SDavid Hildenbrand #endif 62223ee1afa3SNick Piggin } 62239ec23531SDavid Hildenbrand EXPORT_SYMBOL(__might_fault); 62243ee1afa3SNick Piggin #endif 622547ad8475SAndrea Arcangeli 622647ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 6227c6ddfb6cSHuang Ying /* 6228c6ddfb6cSHuang Ying * Process all subpages of the specified huge page with the specified 6229c6ddfb6cSHuang Ying * operation. The target subpage will be processed last to keep its 6230c6ddfb6cSHuang Ying * cache lines hot. 6231c6ddfb6cSHuang Ying */ 62321cb9dc4bSLiu Shixin static inline int process_huge_page( 6233c6ddfb6cSHuang Ying unsigned long addr_hint, unsigned int pages_per_huge_page, 62341cb9dc4bSLiu Shixin int (*process_subpage)(unsigned long addr, int idx, void *arg), 6235c6ddfb6cSHuang Ying void *arg) 6236c6ddfb6cSHuang Ying { 62371cb9dc4bSLiu Shixin int i, n, base, l, ret; 6238c6ddfb6cSHuang Ying unsigned long addr = addr_hint & 6239c6ddfb6cSHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 6240c6ddfb6cSHuang Ying 6241c6ddfb6cSHuang Ying /* Process target subpage last to keep its cache lines hot */ 6242c6ddfb6cSHuang Ying might_sleep(); 6243c6ddfb6cSHuang Ying n = (addr_hint - addr) / PAGE_SIZE; 6244c6ddfb6cSHuang Ying if (2 * n <= pages_per_huge_page) { 6245c6ddfb6cSHuang Ying /* If target subpage in first half of huge page */ 6246c6ddfb6cSHuang Ying base = 0; 6247c6ddfb6cSHuang Ying l = n; 6248c6ddfb6cSHuang Ying /* Process subpages at the end of huge page */ 6249c6ddfb6cSHuang Ying for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { 6250c6ddfb6cSHuang Ying cond_resched(); 62511cb9dc4bSLiu Shixin ret = process_subpage(addr + i * PAGE_SIZE, i, arg); 62521cb9dc4bSLiu Shixin if (ret) 62531cb9dc4bSLiu Shixin return ret; 6254c6ddfb6cSHuang Ying } 6255c6ddfb6cSHuang Ying } else { 6256c6ddfb6cSHuang Ying /* If target subpage in second half of huge page */ 6257c6ddfb6cSHuang Ying base = pages_per_huge_page - 2 * (pages_per_huge_page - n); 6258c6ddfb6cSHuang Ying l = pages_per_huge_page - n; 6259c6ddfb6cSHuang Ying /* Process subpages at the begin of huge page */ 6260c6ddfb6cSHuang Ying for (i = 0; i < base; i++) { 6261c6ddfb6cSHuang Ying cond_resched(); 62621cb9dc4bSLiu Shixin ret = process_subpage(addr + i * PAGE_SIZE, i, arg); 62631cb9dc4bSLiu Shixin if (ret) 62641cb9dc4bSLiu Shixin return ret; 6265c6ddfb6cSHuang Ying } 6266c6ddfb6cSHuang Ying } 6267c6ddfb6cSHuang Ying /* 6268c6ddfb6cSHuang Ying * Process remaining subpages in left-right-left-right pattern 6269c6ddfb6cSHuang Ying * towards the target subpage 6270c6ddfb6cSHuang Ying */ 6271c6ddfb6cSHuang Ying for (i = 0; i < l; i++) { 6272c6ddfb6cSHuang Ying int left_idx = base + i; 6273c6ddfb6cSHuang Ying int right_idx = base + 2 * l - 1 - i; 6274c6ddfb6cSHuang Ying 6275c6ddfb6cSHuang Ying cond_resched(); 62761cb9dc4bSLiu Shixin ret = process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg); 62771cb9dc4bSLiu Shixin if (ret) 62781cb9dc4bSLiu Shixin return ret; 6279c6ddfb6cSHuang Ying cond_resched(); 62801cb9dc4bSLiu Shixin ret = process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg); 62811cb9dc4bSLiu Shixin if (ret) 62821cb9dc4bSLiu Shixin return ret; 6283c6ddfb6cSHuang Ying } 62841cb9dc4bSLiu Shixin return 0; 6285c6ddfb6cSHuang Ying } 6286c6ddfb6cSHuang Ying 628747ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 628847ad8475SAndrea Arcangeli unsigned long addr, 628947ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 629047ad8475SAndrea Arcangeli { 629147ad8475SAndrea Arcangeli int i; 629214455eabSCheng Li struct page *p; 629347ad8475SAndrea Arcangeli 629447ad8475SAndrea Arcangeli might_sleep(); 629514455eabSCheng Li for (i = 0; i < pages_per_huge_page; i++) { 629614455eabSCheng Li p = nth_page(page, i); 629747ad8475SAndrea Arcangeli cond_resched(); 629847ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 629947ad8475SAndrea Arcangeli } 630047ad8475SAndrea Arcangeli } 6301c6ddfb6cSHuang Ying 63021cb9dc4bSLiu Shixin static int clear_subpage(unsigned long addr, int idx, void *arg) 6303c6ddfb6cSHuang Ying { 6304c6ddfb6cSHuang Ying struct page *page = arg; 6305c6ddfb6cSHuang Ying 630621fff064SKefeng Wang clear_user_highpage(nth_page(page, idx), addr); 63071cb9dc4bSLiu Shixin return 0; 6308c6ddfb6cSHuang Ying } 6309c6ddfb6cSHuang Ying 631047ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 6311c79b57e4SHuang Ying unsigned long addr_hint, unsigned int pages_per_huge_page) 631247ad8475SAndrea Arcangeli { 6313c79b57e4SHuang Ying unsigned long addr = addr_hint & 6314c79b57e4SHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 631547ad8475SAndrea Arcangeli 631647ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 631747ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 631847ad8475SAndrea Arcangeli return; 631947ad8475SAndrea Arcangeli } 632047ad8475SAndrea Arcangeli 6321c6ddfb6cSHuang Ying process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page); 632247ad8475SAndrea Arcangeli } 632347ad8475SAndrea Arcangeli 63241cb9dc4bSLiu Shixin static int copy_user_gigantic_page(struct folio *dst, struct folio *src, 632547ad8475SAndrea Arcangeli unsigned long addr, 632647ad8475SAndrea Arcangeli struct vm_area_struct *vma, 632747ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 632847ad8475SAndrea Arcangeli { 632947ad8475SAndrea Arcangeli int i; 6330c0e8150eSZhangPeng struct page *dst_page; 6331c0e8150eSZhangPeng struct page *src_page; 633247ad8475SAndrea Arcangeli 633314455eabSCheng Li for (i = 0; i < pages_per_huge_page; i++) { 6334c0e8150eSZhangPeng dst_page = folio_page(dst, i); 6335c0e8150eSZhangPeng src_page = folio_page(src, i); 633614455eabSCheng Li 633747ad8475SAndrea Arcangeli cond_resched(); 63381cb9dc4bSLiu Shixin if (copy_mc_user_highpage(dst_page, src_page, 63391cb9dc4bSLiu Shixin addr + i*PAGE_SIZE, vma)) { 63401cb9dc4bSLiu Shixin memory_failure_queue(page_to_pfn(src_page), 0); 63411cb9dc4bSLiu Shixin return -EHWPOISON; 634247ad8475SAndrea Arcangeli } 634347ad8475SAndrea Arcangeli } 63441cb9dc4bSLiu Shixin return 0; 63451cb9dc4bSLiu Shixin } 634647ad8475SAndrea Arcangeli 6347c9f4cd71SHuang Ying struct copy_subpage_arg { 6348c9f4cd71SHuang Ying struct page *dst; 6349c9f4cd71SHuang Ying struct page *src; 6350c9f4cd71SHuang Ying struct vm_area_struct *vma; 6351c9f4cd71SHuang Ying }; 6352c9f4cd71SHuang Ying 63531cb9dc4bSLiu Shixin static int copy_subpage(unsigned long addr, int idx, void *arg) 6354c9f4cd71SHuang Ying { 6355c9f4cd71SHuang Ying struct copy_subpage_arg *copy_arg = arg; 635621fff064SKefeng Wang struct page *dst = nth_page(copy_arg->dst, idx); 635721fff064SKefeng Wang struct page *src = nth_page(copy_arg->src, idx); 6358c9f4cd71SHuang Ying 635921fff064SKefeng Wang if (copy_mc_user_highpage(dst, src, addr, copy_arg->vma)) { 636021fff064SKefeng Wang memory_failure_queue(page_to_pfn(src), 0); 63611cb9dc4bSLiu Shixin return -EHWPOISON; 63621cb9dc4bSLiu Shixin } 63631cb9dc4bSLiu Shixin return 0; 6364c9f4cd71SHuang Ying } 6365c9f4cd71SHuang Ying 63661cb9dc4bSLiu Shixin int copy_user_large_folio(struct folio *dst, struct folio *src, 6367c0e8150eSZhangPeng unsigned long addr_hint, struct vm_area_struct *vma) 636847ad8475SAndrea Arcangeli { 6369c0e8150eSZhangPeng unsigned int pages_per_huge_page = folio_nr_pages(dst); 6370c9f4cd71SHuang Ying unsigned long addr = addr_hint & 6371c9f4cd71SHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 6372c9f4cd71SHuang Ying struct copy_subpage_arg arg = { 6373c0e8150eSZhangPeng .dst = &dst->page, 6374c0e8150eSZhangPeng .src = &src->page, 6375c9f4cd71SHuang Ying .vma = vma, 6376c9f4cd71SHuang Ying }; 637747ad8475SAndrea Arcangeli 63781cb9dc4bSLiu Shixin if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) 63791cb9dc4bSLiu Shixin return copy_user_gigantic_page(dst, src, addr, vma, 638047ad8475SAndrea Arcangeli pages_per_huge_page); 638147ad8475SAndrea Arcangeli 63821cb9dc4bSLiu Shixin return process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg); 638347ad8475SAndrea Arcangeli } 6384fa4d75c1SMike Kravetz 6385e87340caSZhangPeng long copy_folio_from_user(struct folio *dst_folio, 6386fa4d75c1SMike Kravetz const void __user *usr_src, 6387810a56b9SMike Kravetz bool allow_pagefault) 6388fa4d75c1SMike Kravetz { 6389e87340caSZhangPeng void *kaddr; 6390fa4d75c1SMike Kravetz unsigned long i, rc = 0; 6391e87340caSZhangPeng unsigned int nr_pages = folio_nr_pages(dst_folio); 6392e87340caSZhangPeng unsigned long ret_val = nr_pages * PAGE_SIZE; 639314455eabSCheng Li struct page *subpage; 6394fa4d75c1SMike Kravetz 6395e87340caSZhangPeng for (i = 0; i < nr_pages; i++) { 6396e87340caSZhangPeng subpage = folio_page(dst_folio, i); 6397e87340caSZhangPeng kaddr = kmap_local_page(subpage); 63980d508c1fSZhangPeng if (!allow_pagefault) 63990d508c1fSZhangPeng pagefault_disable(); 6400e87340caSZhangPeng rc = copy_from_user(kaddr, usr_src + i * PAGE_SIZE, PAGE_SIZE); 64010d508c1fSZhangPeng if (!allow_pagefault) 64020d508c1fSZhangPeng pagefault_enable(); 6403e87340caSZhangPeng kunmap_local(kaddr); 6404fa4d75c1SMike Kravetz 6405fa4d75c1SMike Kravetz ret_val -= (PAGE_SIZE - rc); 6406fa4d75c1SMike Kravetz if (rc) 6407fa4d75c1SMike Kravetz break; 6408fa4d75c1SMike Kravetz 6409e763243cSMuchun Song flush_dcache_page(subpage); 6410e763243cSMuchun Song 6411fa4d75c1SMike Kravetz cond_resched(); 6412fa4d75c1SMike Kravetz } 6413fa4d75c1SMike Kravetz return ret_val; 6414fa4d75c1SMike Kravetz } 641547ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 641649076ec2SKirill A. Shutemov 641740b64acdSOlof Johansson #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS 6418b35f1819SKirill A. Shutemov 6419b35f1819SKirill A. Shutemov static struct kmem_cache *page_ptl_cachep; 6420b35f1819SKirill A. Shutemov 6421b35f1819SKirill A. Shutemov void __init ptlock_cache_init(void) 6422b35f1819SKirill A. Shutemov { 6423b35f1819SKirill A. Shutemov page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, 6424b35f1819SKirill A. Shutemov SLAB_PANIC, NULL); 6425b35f1819SKirill A. Shutemov } 6426b35f1819SKirill A. Shutemov 6427f5ecca06SVishal Moola (Oracle) bool ptlock_alloc(struct ptdesc *ptdesc) 642849076ec2SKirill A. Shutemov { 642949076ec2SKirill A. Shutemov spinlock_t *ptl; 643049076ec2SKirill A. Shutemov 6431b35f1819SKirill A. Shutemov ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); 643249076ec2SKirill A. Shutemov if (!ptl) 643349076ec2SKirill A. Shutemov return false; 6434f5ecca06SVishal Moola (Oracle) ptdesc->ptl = ptl; 643549076ec2SKirill A. Shutemov return true; 643649076ec2SKirill A. Shutemov } 643749076ec2SKirill A. Shutemov 64386ed1b8a0SVishal Moola (Oracle) void ptlock_free(struct ptdesc *ptdesc) 643949076ec2SKirill A. Shutemov { 64406ed1b8a0SVishal Moola (Oracle) kmem_cache_free(page_ptl_cachep, ptdesc->ptl); 644149076ec2SKirill A. Shutemov } 644249076ec2SKirill A. Shutemov #endif 6443