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 933*23ed1908SDavid Hildenbrand static inline void __copy_present_pte(struct vm_area_struct *dst_vma, 934*23ed1908SDavid Hildenbrand struct vm_area_struct *src_vma, pte_t *dst_pte, pte_t *src_pte, 935*23ed1908SDavid Hildenbrand pte_t pte, unsigned long addr) 936*23ed1908SDavid Hildenbrand { 937*23ed1908SDavid Hildenbrand struct mm_struct *src_mm = src_vma->vm_mm; 938*23ed1908SDavid Hildenbrand 939*23ed1908SDavid Hildenbrand /* If it's a COW mapping, write protect it both processes. */ 940*23ed1908SDavid Hildenbrand if (is_cow_mapping(src_vma->vm_flags) && pte_write(pte)) { 941*23ed1908SDavid Hildenbrand ptep_set_wrprotect(src_mm, addr, src_pte); 942*23ed1908SDavid Hildenbrand pte = pte_wrprotect(pte); 943*23ed1908SDavid Hildenbrand } 944*23ed1908SDavid Hildenbrand 945*23ed1908SDavid Hildenbrand /* If it's a shared mapping, mark it clean in the child. */ 946*23ed1908SDavid Hildenbrand if (src_vma->vm_flags & VM_SHARED) 947*23ed1908SDavid Hildenbrand pte = pte_mkclean(pte); 948*23ed1908SDavid Hildenbrand pte = pte_mkold(pte); 949*23ed1908SDavid Hildenbrand 950*23ed1908SDavid Hildenbrand if (!userfaultfd_wp(dst_vma)) 951*23ed1908SDavid Hildenbrand pte = pte_clear_uffd_wp(pte); 952*23ed1908SDavid Hildenbrand 953*23ed1908SDavid Hildenbrand set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); 954*23ed1908SDavid Hildenbrand } 955*23ed1908SDavid Hildenbrand 95670e806e4SPeter Xu /* 95770e806e4SPeter Xu * Copy one pte. Returns 0 if succeeded, or -EAGAIN if one preallocated page 95870e806e4SPeter Xu * is required to copy this pte. 95970e806e4SPeter Xu */ 96070e806e4SPeter Xu static inline int 961c78f4636SPeter Xu copy_present_pte(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 962c78f4636SPeter Xu pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, 963edf50470SMatthew Wilcox (Oracle) struct folio **prealloc) 964df3a57d1SLinus Torvalds { 965c33c7948SRyan Roberts pte_t pte = ptep_get(src_pte); 966df3a57d1SLinus Torvalds struct page *page; 96714ddee41SMatthew Wilcox (Oracle) struct folio *folio; 968df3a57d1SLinus Torvalds 969c78f4636SPeter Xu page = vm_normal_page(src_vma, addr, pte); 970*23ed1908SDavid Hildenbrand if (unlikely(!page)) 971*23ed1908SDavid Hildenbrand goto copy_pte; 972*23ed1908SDavid Hildenbrand 97314ddee41SMatthew Wilcox (Oracle) folio = page_folio(page); 974*23ed1908SDavid Hildenbrand folio_get(folio); 975*23ed1908SDavid Hildenbrand if (folio_test_anon(folio)) { 976b51ad4f8SDavid Hildenbrand /* 977b51ad4f8SDavid Hildenbrand * If this page may have been pinned by the parent process, 978b51ad4f8SDavid Hildenbrand * copy the page immediately for the child so that we'll always 979b51ad4f8SDavid Hildenbrand * guarantee the pinned page won't be randomly replaced in the 980b51ad4f8SDavid Hildenbrand * future. 981b51ad4f8SDavid Hildenbrand */ 98208e7795eSDavid Hildenbrand if (unlikely(folio_try_dup_anon_rmap_pte(folio, page, src_vma))) { 983fb3d824dSDavid Hildenbrand /* Page may be pinned, we have to copy. */ 98414ddee41SMatthew Wilcox (Oracle) folio_put(folio); 985b51ad4f8SDavid Hildenbrand return copy_present_page(dst_vma, src_vma, dst_pte, src_pte, 986b51ad4f8SDavid Hildenbrand addr, rss, prealloc, page); 987fb3d824dSDavid Hildenbrand } 988edf50470SMatthew Wilcox (Oracle) rss[MM_ANONPAGES]++; 989*23ed1908SDavid Hildenbrand VM_WARN_ON_FOLIO(PageAnonExclusive(page), folio); 990*23ed1908SDavid Hildenbrand } else { 991d8ef5e31SDavid Hildenbrand folio_dup_file_rmap_pte(folio, page); 9926b27cc6cSKefeng Wang rss[mm_counter_file(folio)]++; 99370e806e4SPeter Xu } 99470e806e4SPeter Xu 995*23ed1908SDavid Hildenbrand copy_pte: 996*23ed1908SDavid Hildenbrand __copy_present_pte(dst_vma, src_vma, dst_pte, src_pte, pte, addr); 99770e806e4SPeter Xu return 0; 9986aab341eSLinus Torvalds } 999ae859762SHugh Dickins 1000294de6d8SKefeng Wang static inline struct folio *folio_prealloc(struct mm_struct *src_mm, 1001294de6d8SKefeng Wang struct vm_area_struct *vma, unsigned long addr, bool need_zero) 100270e806e4SPeter Xu { 1003edf50470SMatthew Wilcox (Oracle) struct folio *new_folio; 100470e806e4SPeter Xu 1005294de6d8SKefeng Wang if (need_zero) 1006294de6d8SKefeng Wang new_folio = vma_alloc_zeroed_movable_folio(vma, addr); 1007294de6d8SKefeng Wang else 1008294de6d8SKefeng Wang new_folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma, 1009294de6d8SKefeng Wang addr, false); 1010294de6d8SKefeng Wang 1011edf50470SMatthew Wilcox (Oracle) if (!new_folio) 101270e806e4SPeter Xu return NULL; 101370e806e4SPeter Xu 1014edf50470SMatthew Wilcox (Oracle) if (mem_cgroup_charge(new_folio, src_mm, GFP_KERNEL)) { 1015edf50470SMatthew Wilcox (Oracle) folio_put(new_folio); 101670e806e4SPeter Xu return NULL; 101770e806e4SPeter Xu } 1018e601ded4SKefeng Wang folio_throttle_swaprate(new_folio, GFP_KERNEL); 101970e806e4SPeter Xu 1020edf50470SMatthew Wilcox (Oracle) return new_folio; 10211da177e4SLinus Torvalds } 10221da177e4SLinus Torvalds 1023c78f4636SPeter Xu static int 1024c78f4636SPeter Xu copy_pte_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1025c78f4636SPeter Xu pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr, 1026c78f4636SPeter Xu unsigned long end) 10271da177e4SLinus Torvalds { 1028c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1029c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 1030c36987e2SDaisuke Nishimura pte_t *orig_src_pte, *orig_dst_pte; 10311da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 1032c33c7948SRyan Roberts pte_t ptent; 1033c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 103470e806e4SPeter Xu int progress, ret = 0; 1035d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 1036570a335bSHugh Dickins swp_entry_t entry = (swp_entry_t){0}; 1037edf50470SMatthew Wilcox (Oracle) struct folio *prealloc = NULL; 10381da177e4SLinus Torvalds 10391da177e4SLinus Torvalds again: 104070e806e4SPeter Xu progress = 0; 1041d559db08SKAMEZAWA Hiroyuki init_rss_vec(rss); 1042d559db08SKAMEZAWA Hiroyuki 10433db82b93SHugh Dickins /* 10443db82b93SHugh Dickins * copy_pmd_range()'s prior pmd_none_or_clear_bad(src_pmd), and the 10453db82b93SHugh Dickins * error handling here, assume that exclusive mmap_lock on dst and src 10463db82b93SHugh Dickins * protects anon from unexpected THP transitions; with shmem and file 10473db82b93SHugh Dickins * protected by mmap_lock-less collapse skipping areas with anon_vma 10483db82b93SHugh Dickins * (whereas vma_needs_copy() skips areas without anon_vma). A rework 10493db82b93SHugh Dickins * can remove such assumptions later, but this is good enough for now. 10503db82b93SHugh Dickins */ 1051c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 105270e806e4SPeter Xu if (!dst_pte) { 105370e806e4SPeter Xu ret = -ENOMEM; 105470e806e4SPeter Xu goto out; 105570e806e4SPeter Xu } 10563db82b93SHugh Dickins src_pte = pte_offset_map_nolock(src_mm, src_pmd, addr, &src_ptl); 10573db82b93SHugh Dickins if (!src_pte) { 10583db82b93SHugh Dickins pte_unmap_unlock(dst_pte, dst_ptl); 10593db82b93SHugh Dickins /* ret == 0 */ 10603db82b93SHugh Dickins goto out; 10613db82b93SHugh Dickins } 1062f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 1063c36987e2SDaisuke Nishimura orig_src_pte = src_pte; 1064c36987e2SDaisuke Nishimura orig_dst_pte = dst_pte; 10656606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 10661da177e4SLinus Torvalds 10671da177e4SLinus Torvalds do { 10681da177e4SLinus Torvalds /* 10691da177e4SLinus Torvalds * We are holding two locks at this point - either of them 10701da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 10711da177e4SLinus Torvalds */ 1072e040f218SHugh Dickins if (progress >= 32) { 1073e040f218SHugh Dickins progress = 0; 1074e040f218SHugh Dickins if (need_resched() || 107595c354feSNick Piggin spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) 10761da177e4SLinus Torvalds break; 1077e040f218SHugh Dickins } 1078c33c7948SRyan Roberts ptent = ptep_get(src_pte); 1079c33c7948SRyan Roberts if (pte_none(ptent)) { 10801da177e4SLinus Torvalds progress++; 10811da177e4SLinus Torvalds continue; 10821da177e4SLinus Torvalds } 1083c33c7948SRyan Roberts if (unlikely(!pte_present(ptent))) { 10849a5cc85cSAlistair Popple ret = copy_nonpresent_pte(dst_mm, src_mm, 108579a1971cSLinus Torvalds dst_pte, src_pte, 10868f34f1eaSPeter Xu dst_vma, src_vma, 10878f34f1eaSPeter Xu addr, rss); 10889a5cc85cSAlistair Popple if (ret == -EIO) { 1089c33c7948SRyan Roberts entry = pte_to_swp_entry(ptep_get(src_pte)); 1090570a335bSHugh Dickins break; 1091b756a3b5SAlistair Popple } else if (ret == -EBUSY) { 1092b756a3b5SAlistair Popple break; 1093b756a3b5SAlistair Popple } else if (!ret) { 10941da177e4SLinus Torvalds progress += 8; 109579a1971cSLinus Torvalds continue; 109679a1971cSLinus Torvalds } 1097b756a3b5SAlistair Popple 1098b756a3b5SAlistair Popple /* 1099b756a3b5SAlistair Popple * Device exclusive entry restored, continue by copying 1100b756a3b5SAlistair Popple * the now present pte. 1101b756a3b5SAlistair Popple */ 1102b756a3b5SAlistair Popple WARN_ON_ONCE(ret != -ENOENT); 1103b756a3b5SAlistair Popple } 110470e806e4SPeter Xu /* copy_present_pte() will clear `*prealloc' if consumed */ 1105c78f4636SPeter Xu ret = copy_present_pte(dst_vma, src_vma, dst_pte, src_pte, 1106c78f4636SPeter Xu addr, rss, &prealloc); 110770e806e4SPeter Xu /* 110870e806e4SPeter Xu * If we need a pre-allocated page for this pte, drop the 110970e806e4SPeter Xu * locks, allocate, and try again. 111070e806e4SPeter Xu */ 111170e806e4SPeter Xu if (unlikely(ret == -EAGAIN)) 111270e806e4SPeter Xu break; 111370e806e4SPeter Xu if (unlikely(prealloc)) { 111470e806e4SPeter Xu /* 111570e806e4SPeter Xu * pre-alloc page cannot be reused by next time so as 111670e806e4SPeter Xu * to strictly follow mempolicy (e.g., alloc_page_vma() 111770e806e4SPeter Xu * will allocate page according to address). This 111870e806e4SPeter Xu * could only happen if one pinned pte changed. 111970e806e4SPeter Xu */ 1120edf50470SMatthew Wilcox (Oracle) folio_put(prealloc); 112170e806e4SPeter Xu prealloc = NULL; 112270e806e4SPeter Xu } 112379a1971cSLinus Torvalds progress += 8; 11241da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 11251da177e4SLinus Torvalds 11266606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 11273db82b93SHugh Dickins pte_unmap_unlock(orig_src_pte, src_ptl); 1128d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(dst_mm, rss); 1129c36987e2SDaisuke Nishimura pte_unmap_unlock(orig_dst_pte, dst_ptl); 1130c74df32cSHugh Dickins cond_resched(); 1131570a335bSHugh Dickins 11329a5cc85cSAlistair Popple if (ret == -EIO) { 11339a5cc85cSAlistair Popple VM_WARN_ON_ONCE(!entry.val); 113470e806e4SPeter Xu if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) { 113570e806e4SPeter Xu ret = -ENOMEM; 113670e806e4SPeter Xu goto out; 113770e806e4SPeter Xu } 113870e806e4SPeter Xu entry.val = 0; 1139b756a3b5SAlistair Popple } else if (ret == -EBUSY) { 1140b756a3b5SAlistair Popple goto out; 11419a5cc85cSAlistair Popple } else if (ret == -EAGAIN) { 1142294de6d8SKefeng Wang prealloc = folio_prealloc(src_mm, src_vma, addr, false); 114370e806e4SPeter Xu if (!prealloc) 1144570a335bSHugh Dickins return -ENOMEM; 11459a5cc85cSAlistair Popple } else if (ret) { 11469a5cc85cSAlistair Popple VM_WARN_ON_ONCE(1); 11479a5cc85cSAlistair Popple } 11489a5cc85cSAlistair Popple 114970e806e4SPeter Xu /* We've captured and resolved the error. Reset, try again. */ 115070e806e4SPeter Xu ret = 0; 11519a5cc85cSAlistair Popple 11521da177e4SLinus Torvalds if (addr != end) 11531da177e4SLinus Torvalds goto again; 115470e806e4SPeter Xu out: 115570e806e4SPeter Xu if (unlikely(prealloc)) 1156edf50470SMatthew Wilcox (Oracle) folio_put(prealloc); 115770e806e4SPeter Xu return ret; 11581da177e4SLinus Torvalds } 11591da177e4SLinus Torvalds 1160c78f4636SPeter Xu static inline int 1161c78f4636SPeter Xu copy_pmd_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1162c78f4636SPeter Xu pud_t *dst_pud, pud_t *src_pud, unsigned long addr, 1163c78f4636SPeter Xu unsigned long end) 11641da177e4SLinus Torvalds { 1165c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1166c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 11671da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 11681da177e4SLinus Torvalds unsigned long next; 11691da177e4SLinus Torvalds 11701da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 11711da177e4SLinus Torvalds if (!dst_pmd) 11721da177e4SLinus Torvalds return -ENOMEM; 11731da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 11741da177e4SLinus Torvalds do { 11751da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 117684c3fc4eSZi Yan if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) 117784c3fc4eSZi Yan || pmd_devmap(*src_pmd)) { 117871e3aac0SAndrea Arcangeli int err; 1179c78f4636SPeter Xu VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, src_vma); 11808f34f1eaSPeter Xu err = copy_huge_pmd(dst_mm, src_mm, dst_pmd, src_pmd, 11818f34f1eaSPeter Xu addr, dst_vma, src_vma); 118271e3aac0SAndrea Arcangeli if (err == -ENOMEM) 118371e3aac0SAndrea Arcangeli return -ENOMEM; 118471e3aac0SAndrea Arcangeli if (!err) 118571e3aac0SAndrea Arcangeli continue; 118671e3aac0SAndrea Arcangeli /* fall through */ 118771e3aac0SAndrea Arcangeli } 11881da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 11891da177e4SLinus Torvalds continue; 1190c78f4636SPeter Xu if (copy_pte_range(dst_vma, src_vma, dst_pmd, src_pmd, 1191c78f4636SPeter Xu addr, next)) 11921da177e4SLinus Torvalds return -ENOMEM; 11931da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 11941da177e4SLinus Torvalds return 0; 11951da177e4SLinus Torvalds } 11961da177e4SLinus Torvalds 1197c78f4636SPeter Xu static inline int 1198c78f4636SPeter Xu copy_pud_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1199c78f4636SPeter Xu p4d_t *dst_p4d, p4d_t *src_p4d, unsigned long addr, 1200c78f4636SPeter Xu unsigned long end) 12011da177e4SLinus Torvalds { 1202c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1203c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 12041da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 12051da177e4SLinus Torvalds unsigned long next; 12061da177e4SLinus Torvalds 1207c2febafcSKirill A. Shutemov dst_pud = pud_alloc(dst_mm, dst_p4d, addr); 12081da177e4SLinus Torvalds if (!dst_pud) 12091da177e4SLinus Torvalds return -ENOMEM; 1210c2febafcSKirill A. Shutemov src_pud = pud_offset(src_p4d, addr); 12111da177e4SLinus Torvalds do { 12121da177e4SLinus Torvalds next = pud_addr_end(addr, end); 1213a00cc7d9SMatthew Wilcox if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { 1214a00cc7d9SMatthew Wilcox int err; 1215a00cc7d9SMatthew Wilcox 1216c78f4636SPeter Xu VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, src_vma); 1217a00cc7d9SMatthew Wilcox err = copy_huge_pud(dst_mm, src_mm, 1218c78f4636SPeter Xu dst_pud, src_pud, addr, src_vma); 1219a00cc7d9SMatthew Wilcox if (err == -ENOMEM) 1220a00cc7d9SMatthew Wilcox return -ENOMEM; 1221a00cc7d9SMatthew Wilcox if (!err) 1222a00cc7d9SMatthew Wilcox continue; 1223a00cc7d9SMatthew Wilcox /* fall through */ 1224a00cc7d9SMatthew Wilcox } 12251da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 12261da177e4SLinus Torvalds continue; 1227c78f4636SPeter Xu if (copy_pmd_range(dst_vma, src_vma, dst_pud, src_pud, 1228c78f4636SPeter Xu addr, next)) 12291da177e4SLinus Torvalds return -ENOMEM; 12301da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 12311da177e4SLinus Torvalds return 0; 12321da177e4SLinus Torvalds } 12331da177e4SLinus Torvalds 1234c78f4636SPeter Xu static inline int 1235c78f4636SPeter Xu copy_p4d_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1236c78f4636SPeter Xu pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long addr, 1237c78f4636SPeter Xu unsigned long end) 1238c2febafcSKirill A. Shutemov { 1239c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1240c2febafcSKirill A. Shutemov p4d_t *src_p4d, *dst_p4d; 1241c2febafcSKirill A. Shutemov unsigned long next; 1242c2febafcSKirill A. Shutemov 1243c2febafcSKirill A. Shutemov dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); 1244c2febafcSKirill A. Shutemov if (!dst_p4d) 1245c2febafcSKirill A. Shutemov return -ENOMEM; 1246c2febafcSKirill A. Shutemov src_p4d = p4d_offset(src_pgd, addr); 1247c2febafcSKirill A. Shutemov do { 1248c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 1249c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(src_p4d)) 1250c2febafcSKirill A. Shutemov continue; 1251c78f4636SPeter Xu if (copy_pud_range(dst_vma, src_vma, dst_p4d, src_p4d, 1252c78f4636SPeter Xu addr, next)) 1253c2febafcSKirill A. Shutemov return -ENOMEM; 1254c2febafcSKirill A. Shutemov } while (dst_p4d++, src_p4d++, addr = next, addr != end); 1255c2febafcSKirill A. Shutemov return 0; 1256c2febafcSKirill A. Shutemov } 1257c2febafcSKirill A. Shutemov 1258c56d1b62SPeter Xu /* 1259c56d1b62SPeter Xu * Return true if the vma needs to copy the pgtable during this fork(). Return 1260c56d1b62SPeter Xu * false when we can speed up fork() by allowing lazy page faults later until 1261c56d1b62SPeter Xu * when the child accesses the memory range. 1262c56d1b62SPeter Xu */ 1263bc70fbf2SPeter Xu static bool 1264c56d1b62SPeter Xu vma_needs_copy(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) 1265c56d1b62SPeter Xu { 1266c56d1b62SPeter Xu /* 1267c56d1b62SPeter Xu * Always copy pgtables when dst_vma has uffd-wp enabled even if it's 1268c56d1b62SPeter Xu * file-backed (e.g. shmem). Because when uffd-wp is enabled, pgtable 1269c56d1b62SPeter Xu * contains uffd-wp protection information, that's something we can't 1270c56d1b62SPeter Xu * retrieve from page cache, and skip copying will lose those info. 1271c56d1b62SPeter Xu */ 1272c56d1b62SPeter Xu if (userfaultfd_wp(dst_vma)) 1273c56d1b62SPeter Xu return true; 1274c56d1b62SPeter Xu 1275bcd51a3cSMike Kravetz if (src_vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 1276c56d1b62SPeter Xu return true; 1277c56d1b62SPeter Xu 1278c56d1b62SPeter Xu if (src_vma->anon_vma) 1279c56d1b62SPeter Xu return true; 1280c56d1b62SPeter Xu 1281c56d1b62SPeter Xu /* 1282c56d1b62SPeter Xu * Don't copy ptes where a page fault will fill them correctly. Fork 1283c56d1b62SPeter Xu * becomes much lighter when there are big shared or private readonly 1284c56d1b62SPeter Xu * mappings. The tradeoff is that copy_page_range is more efficient 1285c56d1b62SPeter Xu * than faulting. 1286c56d1b62SPeter Xu */ 1287c56d1b62SPeter Xu return false; 1288c56d1b62SPeter Xu } 1289c56d1b62SPeter Xu 1290c78f4636SPeter Xu int 1291c78f4636SPeter Xu copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) 12921da177e4SLinus Torvalds { 12931da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 12941da177e4SLinus Torvalds unsigned long next; 1295c78f4636SPeter Xu unsigned long addr = src_vma->vm_start; 1296c78f4636SPeter Xu unsigned long end = src_vma->vm_end; 1297c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1298c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 1299ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 13002ec74c3eSSagi Grimberg bool is_cow; 1301cddb8a5cSAndrea Arcangeli int ret; 13021da177e4SLinus Torvalds 1303c56d1b62SPeter Xu if (!vma_needs_copy(dst_vma, src_vma)) 1304d992895bSNick Piggin return 0; 1305d992895bSNick Piggin 1306c78f4636SPeter Xu if (is_vm_hugetlb_page(src_vma)) 1307bc70fbf2SPeter Xu return copy_hugetlb_page_range(dst_mm, src_mm, dst_vma, src_vma); 13081da177e4SLinus Torvalds 1309c78f4636SPeter Xu if (unlikely(src_vma->vm_flags & VM_PFNMAP)) { 13102ab64037Svenkatesh.pallipadi@intel.com /* 13112ab64037Svenkatesh.pallipadi@intel.com * We do not free on error cases below as remove_vma 13122ab64037Svenkatesh.pallipadi@intel.com * gets called on error from higher level routine 13132ab64037Svenkatesh.pallipadi@intel.com */ 1314c78f4636SPeter Xu ret = track_pfn_copy(src_vma); 13152ab64037Svenkatesh.pallipadi@intel.com if (ret) 13162ab64037Svenkatesh.pallipadi@intel.com return ret; 13172ab64037Svenkatesh.pallipadi@intel.com } 13182ab64037Svenkatesh.pallipadi@intel.com 1319cddb8a5cSAndrea Arcangeli /* 1320cddb8a5cSAndrea Arcangeli * We need to invalidate the secondary MMU mappings only when 1321cddb8a5cSAndrea Arcangeli * there could be a permission downgrade on the ptes of the 1322cddb8a5cSAndrea Arcangeli * parent mm. And a permission downgrade will only happen if 1323cddb8a5cSAndrea Arcangeli * is_cow_mapping() returns true. 1324cddb8a5cSAndrea Arcangeli */ 1325c78f4636SPeter Xu is_cow = is_cow_mapping(src_vma->vm_flags); 1326ac46d4f3SJérôme Glisse 1327ac46d4f3SJérôme Glisse if (is_cow) { 13287269f999SJérôme Glisse mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_PAGE, 13297d4a8be0SAlistair Popple 0, src_mm, addr, end); 1330ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 133157efa1feSJason Gunthorpe /* 133257efa1feSJason Gunthorpe * Disabling preemption is not needed for the write side, as 133357efa1feSJason Gunthorpe * the read side doesn't spin, but goes to the mmap_lock. 133457efa1feSJason Gunthorpe * 133557efa1feSJason Gunthorpe * Use the raw variant of the seqcount_t write API to avoid 133657efa1feSJason Gunthorpe * lockdep complaining about preemptibility. 133757efa1feSJason Gunthorpe */ 1338e727bfd5SSuren Baghdasaryan vma_assert_write_locked(src_vma); 133957efa1feSJason Gunthorpe raw_write_seqcount_begin(&src_mm->write_protect_seq); 1340ac46d4f3SJérôme Glisse } 1341cddb8a5cSAndrea Arcangeli 1342cddb8a5cSAndrea Arcangeli ret = 0; 13431da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 13441da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 13451da177e4SLinus Torvalds do { 13461da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 13471da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 13481da177e4SLinus Torvalds continue; 1349c78f4636SPeter Xu if (unlikely(copy_p4d_range(dst_vma, src_vma, dst_pgd, src_pgd, 1350c78f4636SPeter Xu addr, next))) { 1351d155df53SMa Wupeng untrack_pfn_clear(dst_vma); 1352cddb8a5cSAndrea Arcangeli ret = -ENOMEM; 1353cddb8a5cSAndrea Arcangeli break; 1354cddb8a5cSAndrea Arcangeli } 13551da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 1356cddb8a5cSAndrea Arcangeli 135757efa1feSJason Gunthorpe if (is_cow) { 135857efa1feSJason Gunthorpe raw_write_seqcount_end(&src_mm->write_protect_seq); 1359ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 136057efa1feSJason Gunthorpe } 1361cddb8a5cSAndrea Arcangeli return ret; 13621da177e4SLinus Torvalds } 13631da177e4SLinus Torvalds 13645abfd71dSPeter Xu /* Whether we should zap all COWed (private) pages too */ 13655abfd71dSPeter Xu static inline bool should_zap_cows(struct zap_details *details) 13665abfd71dSPeter Xu { 13675abfd71dSPeter Xu /* By default, zap all pages */ 13685abfd71dSPeter Xu if (!details) 13695abfd71dSPeter Xu return true; 13705abfd71dSPeter Xu 13715abfd71dSPeter Xu /* Or, we zap COWed pages only if the caller wants to */ 13722e148f1eSPeter Xu return details->even_cows; 13735abfd71dSPeter Xu } 13745abfd71dSPeter Xu 1375eabafaaaSKefeng Wang /* Decides whether we should zap this folio with the folio pointer specified */ 1376eabafaaaSKefeng Wang static inline bool should_zap_folio(struct zap_details *details, 1377eabafaaaSKefeng Wang struct folio *folio) 13783506659eSMatthew Wilcox (Oracle) { 1379eabafaaaSKefeng Wang /* If we can make a decision without *folio.. */ 13805abfd71dSPeter Xu if (should_zap_cows(details)) 1381254ab940SPeter Xu return true; 13823506659eSMatthew Wilcox (Oracle) 1383eabafaaaSKefeng Wang /* E.g. the caller passes NULL for the case of a zero folio */ 1384eabafaaaSKefeng Wang if (!folio) 1385254ab940SPeter Xu return true; 13865abfd71dSPeter Xu 1387eabafaaaSKefeng Wang /* Otherwise we should only zap non-anon folios */ 1388eabafaaaSKefeng Wang return !folio_test_anon(folio); 13893506659eSMatthew Wilcox (Oracle) } 13903506659eSMatthew Wilcox (Oracle) 1391999dad82SPeter Xu static inline bool zap_drop_file_uffd_wp(struct zap_details *details) 1392999dad82SPeter Xu { 1393999dad82SPeter Xu if (!details) 1394999dad82SPeter Xu return false; 1395999dad82SPeter Xu 1396999dad82SPeter Xu return details->zap_flags & ZAP_FLAG_DROP_MARKER; 1397999dad82SPeter Xu } 1398999dad82SPeter Xu 1399999dad82SPeter Xu /* 1400999dad82SPeter Xu * This function makes sure that we'll replace the none pte with an uffd-wp 1401999dad82SPeter Xu * swap special pte marker when necessary. Must be with the pgtable lock held. 1402999dad82SPeter Xu */ 1403999dad82SPeter Xu static inline void 1404999dad82SPeter Xu zap_install_uffd_wp_if_needed(struct vm_area_struct *vma, 1405999dad82SPeter Xu unsigned long addr, pte_t *pte, 1406999dad82SPeter Xu struct zap_details *details, pte_t pteval) 1407999dad82SPeter Xu { 14082bad466cSPeter Xu /* Zap on anonymous always means dropping everything */ 14092bad466cSPeter Xu if (vma_is_anonymous(vma)) 14102bad466cSPeter Xu return; 14112bad466cSPeter Xu 1412999dad82SPeter Xu if (zap_drop_file_uffd_wp(details)) 1413999dad82SPeter Xu return; 1414999dad82SPeter Xu 1415999dad82SPeter Xu pte_install_uffd_wp_if_needed(vma, addr, pte, pteval); 1416999dad82SPeter Xu } 1417999dad82SPeter Xu 141851c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 1419b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 14201da177e4SLinus Torvalds unsigned long addr, unsigned long end, 142197a89413SPeter Zijlstra struct zap_details *details) 14221da177e4SLinus Torvalds { 1423b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 1424d16dfc55SPeter Zijlstra int force_flush = 0; 1425d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 142697a89413SPeter Zijlstra spinlock_t *ptl; 14275f1a1907SSteven Rostedt pte_t *start_pte; 142897a89413SPeter Zijlstra pte_t *pte; 14298a5f14a2SKirill A. Shutemov swp_entry_t entry; 1430d559db08SKAMEZAWA Hiroyuki 1431ed6a7935SPeter Zijlstra tlb_change_page_size(tlb, PAGE_SIZE); 1432e303297eSPeter Zijlstra init_rss_vec(rss); 14333db82b93SHugh Dickins start_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 14343db82b93SHugh Dickins if (!pte) 14353db82b93SHugh Dickins return addr; 14363db82b93SHugh Dickins 14373ea27719SMel Gorman flush_tlb_batched_pending(mm); 14386606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 14391da177e4SLinus Torvalds do { 1440c33c7948SRyan Roberts pte_t ptent = ptep_get(pte); 1441eabafaaaSKefeng Wang struct folio *folio = NULL; 14428018db85SPeter Xu struct page *page; 14438018db85SPeter Xu 1444166f61b9STobin C Harding if (pte_none(ptent)) 14451da177e4SLinus Torvalds continue; 144651c6f666SRobin Holt 14477b167b68SMinchan Kim if (need_resched()) 14487b167b68SMinchan Kim break; 14497b167b68SMinchan Kim 14506f5e6b9eSHugh Dickins if (pte_present(ptent)) { 14515df397deSLinus Torvalds unsigned int delay_rmap; 14525df397deSLinus Torvalds 145325b2995aSChristoph Hellwig page = vm_normal_page(vma, addr, ptent); 1454eabafaaaSKefeng Wang if (page) 1455eabafaaaSKefeng Wang folio = page_folio(page); 1456eabafaaaSKefeng Wang 1457eabafaaaSKefeng Wang if (unlikely(!should_zap_folio(details, folio))) 14581da177e4SLinus Torvalds continue; 1459b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 1460a600388dSZachary Amsden tlb->fullmm); 1461e5136e87SRick Edgecombe arch_check_zapped_pte(vma, ptent); 14621da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 1463999dad82SPeter Xu zap_install_uffd_wp_if_needed(vma, addr, pte, details, 1464999dad82SPeter Xu ptent); 1465e2942062Sxu xin if (unlikely(!page)) { 14666080d19fSxu xin ksm_might_unmap_zero_page(mm, ptent); 14671da177e4SLinus Torvalds continue; 1468e2942062Sxu xin } 1469eca56ff9SJerome Marchand 14705df397deSLinus Torvalds delay_rmap = 0; 1471c4626503SDavid Hildenbrand if (!folio_test_anon(folio)) { 14721cf35d47SLinus Torvalds if (pte_dirty(ptent)) { 1473e4e3df29SDavid Hildenbrand folio_mark_dirty(folio); 14745df397deSLinus Torvalds if (tlb_delay_rmap(tlb)) { 14755df397deSLinus Torvalds delay_rmap = 1; 14765df397deSLinus Torvalds force_flush = 1; 14775df397deSLinus Torvalds } 14781cf35d47SLinus Torvalds } 14798788f678SYu Zhao if (pte_young(ptent) && likely(vma_has_recency(vma))) 1480c4626503SDavid Hildenbrand folio_mark_accessed(folio); 14816237bcd9SHugh Dickins } 1482a23f517bSKefeng Wang rss[mm_counter(folio)]--; 14835df397deSLinus Torvalds if (!delay_rmap) { 1484c4626503SDavid Hildenbrand folio_remove_rmap_pte(folio, page, vma); 14853dc14741SHugh Dickins if (unlikely(page_mapcount(page) < 0)) 14863dc14741SHugh Dickins print_bad_pte(vma, addr, ptent, page); 14875df397deSLinus Torvalds } 14885df397deSLinus Torvalds if (unlikely(__tlb_remove_page(tlb, page, delay_rmap))) { 14891cf35d47SLinus Torvalds force_flush = 1; 1490ce9ec37bSWill Deacon addr += PAGE_SIZE; 1491d16dfc55SPeter Zijlstra break; 14921cf35d47SLinus Torvalds } 14931da177e4SLinus Torvalds continue; 14941da177e4SLinus Torvalds } 14955042db43SJérôme Glisse 14965042db43SJérôme Glisse entry = pte_to_swp_entry(ptent); 1497b756a3b5SAlistair Popple if (is_device_private_entry(entry) || 1498b756a3b5SAlistair Popple is_device_exclusive_entry(entry)) { 14998018db85SPeter Xu page = pfn_swap_entry_to_page(entry); 1500c4626503SDavid Hildenbrand folio = page_folio(page); 1501eabafaaaSKefeng Wang if (unlikely(!should_zap_folio(details, folio))) 15025042db43SJérôme Glisse continue; 1503999dad82SPeter Xu /* 1504999dad82SPeter Xu * Both device private/exclusive mappings should only 1505999dad82SPeter Xu * work with anonymous page so far, so we don't need to 1506999dad82SPeter Xu * consider uffd-wp bit when zap. For more information, 1507999dad82SPeter Xu * see zap_install_uffd_wp_if_needed(). 1508999dad82SPeter Xu */ 1509999dad82SPeter Xu WARN_ON_ONCE(!vma_is_anonymous(vma)); 1510a23f517bSKefeng Wang rss[mm_counter(folio)]--; 1511b756a3b5SAlistair Popple if (is_device_private_entry(entry)) 1512c4626503SDavid Hildenbrand folio_remove_rmap_pte(folio, page, vma); 1513c4626503SDavid Hildenbrand folio_put(folio); 15148018db85SPeter Xu } else if (!non_swap_entry(entry)) { 15155abfd71dSPeter Xu /* Genuine swap entry, hence a private anon page */ 15165abfd71dSPeter Xu if (!should_zap_cows(details)) 15171da177e4SLinus Torvalds continue; 1518b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 15198018db85SPeter Xu if (unlikely(!free_swap_and_cache(entry))) 15208018db85SPeter Xu print_bad_pte(vma, addr, ptent, NULL); 15215abfd71dSPeter Xu } else if (is_migration_entry(entry)) { 1522eabafaaaSKefeng Wang folio = pfn_swap_entry_folio(entry); 1523eabafaaaSKefeng Wang if (!should_zap_folio(details, folio)) 15245abfd71dSPeter Xu continue; 1525a23f517bSKefeng Wang rss[mm_counter(folio)]--; 1526999dad82SPeter Xu } else if (pte_marker_entry_uffd_wp(entry)) { 15272bad466cSPeter Xu /* 15282bad466cSPeter Xu * For anon: always drop the marker; for file: only 15292bad466cSPeter Xu * drop the marker if explicitly requested. 15302bad466cSPeter Xu */ 15312bad466cSPeter Xu if (!vma_is_anonymous(vma) && 15322bad466cSPeter Xu !zap_drop_file_uffd_wp(details)) 1533999dad82SPeter Xu continue; 15349f186f9eSMiaohe Lin } else if (is_hwpoison_entry(entry) || 1535af19487fSAxel Rasmussen is_poisoned_swp_entry(entry)) { 15365abfd71dSPeter Xu if (!should_zap_cows(details)) 15375abfd71dSPeter Xu continue; 15385abfd71dSPeter Xu } else { 15395abfd71dSPeter Xu /* We should have covered all the swap entry types */ 1540727d16f1SAndrew Morton pr_alert("unrecognized swap entry 0x%lx\n", entry.val); 15415abfd71dSPeter Xu WARN_ON_ONCE(1); 15429f9f1acdSKonstantin Khlebnikov } 15439888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 1544999dad82SPeter Xu zap_install_uffd_wp_if_needed(vma, addr, pte, details, ptent); 154597a89413SPeter Zijlstra } while (pte++, addr += PAGE_SIZE, addr != end); 1546ae859762SHugh Dickins 1547d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 15486606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 154951c6f666SRobin Holt 15501cf35d47SLinus Torvalds /* Do the actual TLB flush before dropping ptl */ 15515df397deSLinus Torvalds if (force_flush) { 15521cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 15535df397deSLinus Torvalds tlb_flush_rmaps(tlb, vma); 15545df397deSLinus Torvalds } 15551cf35d47SLinus Torvalds pte_unmap_unlock(start_pte, ptl); 15561cf35d47SLinus Torvalds 15571cf35d47SLinus Torvalds /* 15581cf35d47SLinus Torvalds * If we forced a TLB flush (either due to running out of 15591cf35d47SLinus Torvalds * batch buffers or because we needed to flush dirty TLB 15601cf35d47SLinus Torvalds * entries before releasing the ptl), free the batched 15613db82b93SHugh Dickins * memory too. Come back again if we didn't do everything. 15621cf35d47SLinus Torvalds */ 15633db82b93SHugh Dickins if (force_flush) 1564fa0aafb8SPeter Zijlstra tlb_flush_mmu(tlb); 1565d16dfc55SPeter Zijlstra 156651c6f666SRobin Holt return addr; 15671da177e4SLinus Torvalds } 15681da177e4SLinus Torvalds 156951c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1570b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 15711da177e4SLinus Torvalds unsigned long addr, unsigned long end, 157297a89413SPeter Zijlstra struct zap_details *details) 15731da177e4SLinus Torvalds { 15741da177e4SLinus Torvalds pmd_t *pmd; 15751da177e4SLinus Torvalds unsigned long next; 15761da177e4SLinus Torvalds 15771da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 15781da177e4SLinus Torvalds do { 15791da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 158084c3fc4eSZi Yan if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { 158153406ed1SHugh Dickins if (next - addr != HPAGE_PMD_SIZE) 1582fd60775aSDavid Rientjes __split_huge_pmd(vma, pmd, addr, false, NULL); 15833db82b93SHugh Dickins else if (zap_huge_pmd(tlb, vma, pmd, addr)) { 15843db82b93SHugh Dickins addr = next; 15853db82b93SHugh Dickins continue; 15863db82b93SHugh Dickins } 158771e3aac0SAndrea Arcangeli /* fall through */ 15883506659eSMatthew Wilcox (Oracle) } else if (details && details->single_folio && 15893506659eSMatthew Wilcox (Oracle) folio_test_pmd_mappable(details->single_folio) && 159022061a1fSHugh Dickins next - addr == HPAGE_PMD_SIZE && pmd_none(*pmd)) { 159122061a1fSHugh Dickins spinlock_t *ptl = pmd_lock(tlb->mm, pmd); 159222061a1fSHugh Dickins /* 159322061a1fSHugh Dickins * Take and drop THP pmd lock so that we cannot return 159422061a1fSHugh Dickins * prematurely, while zap_huge_pmd() has cleared *pmd, 159522061a1fSHugh Dickins * but not yet decremented compound_mapcount(). 159622061a1fSHugh Dickins */ 159722061a1fSHugh Dickins spin_unlock(ptl); 159871e3aac0SAndrea Arcangeli } 15993db82b93SHugh Dickins if (pmd_none(*pmd)) { 16003db82b93SHugh Dickins addr = next; 16013db82b93SHugh Dickins continue; 16023db82b93SHugh Dickins } 16033db82b93SHugh Dickins addr = zap_pte_range(tlb, vma, pmd, addr, next, details); 16043db82b93SHugh Dickins if (addr != next) 16053db82b93SHugh Dickins pmd--; 16063db82b93SHugh Dickins } while (pmd++, cond_resched(), addr != end); 160751c6f666SRobin Holt 160851c6f666SRobin Holt return addr; 16091da177e4SLinus Torvalds } 16101da177e4SLinus Torvalds 161151c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1612c2febafcSKirill A. Shutemov struct vm_area_struct *vma, p4d_t *p4d, 16131da177e4SLinus Torvalds unsigned long addr, unsigned long end, 161497a89413SPeter Zijlstra struct zap_details *details) 16151da177e4SLinus Torvalds { 16161da177e4SLinus Torvalds pud_t *pud; 16171da177e4SLinus Torvalds unsigned long next; 16181da177e4SLinus Torvalds 1619c2febafcSKirill A. Shutemov pud = pud_offset(p4d, addr); 16201da177e4SLinus Torvalds do { 16211da177e4SLinus Torvalds next = pud_addr_end(addr, end); 1622a00cc7d9SMatthew Wilcox if (pud_trans_huge(*pud) || pud_devmap(*pud)) { 1623a00cc7d9SMatthew Wilcox if (next - addr != HPAGE_PUD_SIZE) { 162442fc5414SMichel Lespinasse mmap_assert_locked(tlb->mm); 1625a00cc7d9SMatthew Wilcox split_huge_pud(vma, pud, addr); 1626a00cc7d9SMatthew Wilcox } else if (zap_huge_pud(tlb, vma, pud, addr)) 1627a00cc7d9SMatthew Wilcox goto next; 1628a00cc7d9SMatthew Wilcox /* fall through */ 1629a00cc7d9SMatthew Wilcox } 163097a89413SPeter Zijlstra if (pud_none_or_clear_bad(pud)) 16311da177e4SLinus Torvalds continue; 163297a89413SPeter Zijlstra next = zap_pmd_range(tlb, vma, pud, addr, next, details); 1633a00cc7d9SMatthew Wilcox next: 1634a00cc7d9SMatthew Wilcox cond_resched(); 163597a89413SPeter Zijlstra } while (pud++, addr = next, addr != end); 163651c6f666SRobin Holt 163751c6f666SRobin Holt return addr; 16381da177e4SLinus Torvalds } 16391da177e4SLinus Torvalds 1640c2febafcSKirill A. Shutemov static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, 1641c2febafcSKirill A. Shutemov struct vm_area_struct *vma, pgd_t *pgd, 1642c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 1643c2febafcSKirill A. Shutemov struct zap_details *details) 1644c2febafcSKirill A. Shutemov { 1645c2febafcSKirill A. Shutemov p4d_t *p4d; 1646c2febafcSKirill A. Shutemov unsigned long next; 1647c2febafcSKirill A. Shutemov 1648c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, addr); 1649c2febafcSKirill A. Shutemov do { 1650c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 1651c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(p4d)) 1652c2febafcSKirill A. Shutemov continue; 1653c2febafcSKirill A. Shutemov next = zap_pud_range(tlb, vma, p4d, addr, next, details); 1654c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 1655c2febafcSKirill A. Shutemov 1656c2febafcSKirill A. Shutemov return addr; 1657c2febafcSKirill A. Shutemov } 1658c2febafcSKirill A. Shutemov 1659aac45363SMichal Hocko void unmap_page_range(struct mmu_gather *tlb, 166051c6f666SRobin Holt struct vm_area_struct *vma, 16611da177e4SLinus Torvalds unsigned long addr, unsigned long end, 166297a89413SPeter Zijlstra struct zap_details *details) 16631da177e4SLinus Torvalds { 16641da177e4SLinus Torvalds pgd_t *pgd; 16651da177e4SLinus Torvalds unsigned long next; 16661da177e4SLinus Torvalds 16671da177e4SLinus Torvalds BUG_ON(addr >= end); 16681da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 16691da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 16701da177e4SLinus Torvalds do { 16711da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 167297a89413SPeter Zijlstra if (pgd_none_or_clear_bad(pgd)) 16731da177e4SLinus Torvalds continue; 1674c2febafcSKirill A. Shutemov next = zap_p4d_range(tlb, vma, pgd, addr, next, details); 167597a89413SPeter Zijlstra } while (pgd++, addr = next, addr != end); 16761da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 16771da177e4SLinus Torvalds } 16781da177e4SLinus Torvalds 1679f5cc4eefSAl Viro 1680f5cc4eefSAl Viro static void unmap_single_vma(struct mmu_gather *tlb, 16811da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 16824f74d2c8SLinus Torvalds unsigned long end_addr, 168368f48381SSuren Baghdasaryan struct zap_details *details, bool mm_wr_locked) 16841da177e4SLinus Torvalds { 1685f5cc4eefSAl Viro unsigned long start = max(vma->vm_start, start_addr); 16861da177e4SLinus Torvalds unsigned long end; 16871da177e4SLinus Torvalds 16881da177e4SLinus Torvalds if (start >= vma->vm_end) 1689f5cc4eefSAl Viro return; 16901da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 16911da177e4SLinus Torvalds if (end <= vma->vm_start) 1692f5cc4eefSAl Viro return; 16931da177e4SLinus Torvalds 1694cbc91f71SSrikar Dronamraju if (vma->vm_file) 1695cbc91f71SSrikar Dronamraju uprobe_munmap(vma, start, end); 1696cbc91f71SSrikar Dronamraju 1697b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) 169868f48381SSuren Baghdasaryan untrack_pfn(vma, 0, 0, mm_wr_locked); 16992ab64037Svenkatesh.pallipadi@intel.com 17008b2a1238SAl Viro if (start != end) { 170151c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1702a137e1ccSAndi Kleen /* 1703a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1704a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1705a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 17067aa6b4adSDavidlohr Bueso * cleanup path of mmap_region. When 1707a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 17087aa6b4adSDavidlohr Bueso * mmap_region() nullifies vma->vm_file 1709a137e1ccSAndi Kleen * before calling this function to clean up. 1710a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1711a137e1ccSAndi Kleen * safe to do nothing in this case. 1712a137e1ccSAndi Kleen */ 171324669e58SAneesh Kumar K.V if (vma->vm_file) { 171405e90bd0SPeter Xu zap_flags_t zap_flags = details ? 171505e90bd0SPeter Xu details->zap_flags : 0; 17162820b0f0SRik van Riel __unmap_hugepage_range(tlb, vma, start, end, 171705e90bd0SPeter Xu NULL, zap_flags); 171824669e58SAneesh Kumar K.V } 171951c6f666SRobin Holt } else 1720038c7aa1SAl Viro unmap_page_range(tlb, vma, start, end, details); 172197a89413SPeter Zijlstra } 172251c6f666SRobin Holt } 17231da177e4SLinus Torvalds 1724f5cc4eefSAl Viro /** 1725f5cc4eefSAl Viro * unmap_vmas - unmap a range of memory covered by a list of vma's 1726f5cc4eefSAl Viro * @tlb: address of the caller's struct mmu_gather 17276e412203SYang Li * @mas: the maple state 1728f5cc4eefSAl Viro * @vma: the starting vma 1729f5cc4eefSAl Viro * @start_addr: virtual address at which to start unmapping 1730f5cc4eefSAl Viro * @end_addr: virtual address at which to end unmapping 17316e412203SYang Li * @tree_end: The maximum index to check 1732809ef83cSYang Li * @mm_wr_locked: lock flag 1733f5cc4eefSAl Viro * 1734f5cc4eefSAl Viro * Unmap all pages in the vma list. 1735f5cc4eefSAl Viro * 1736f5cc4eefSAl Viro * Only addresses between `start' and `end' will be unmapped. 1737f5cc4eefSAl Viro * 1738f5cc4eefSAl Viro * The VMA list must be sorted in ascending virtual address order. 1739f5cc4eefSAl Viro * 1740f5cc4eefSAl Viro * unmap_vmas() assumes that the caller will flush the whole unmapped address 1741f5cc4eefSAl Viro * range after unmap_vmas() returns. So the only responsibility here is to 1742f5cc4eefSAl Viro * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 1743f5cc4eefSAl Viro * drops the lock and schedules. 1744f5cc4eefSAl Viro */ 1745fd892593SLiam R. Howlett void unmap_vmas(struct mmu_gather *tlb, struct ma_state *mas, 1746f5cc4eefSAl Viro struct vm_area_struct *vma, unsigned long start_addr, 1747fd892593SLiam R. Howlett unsigned long end_addr, unsigned long tree_end, 1748fd892593SLiam R. Howlett bool mm_wr_locked) 1749f5cc4eefSAl Viro { 1750ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1751999dad82SPeter Xu struct zap_details details = { 175204ada095SMike Kravetz .zap_flags = ZAP_FLAG_DROP_MARKER | ZAP_FLAG_UNMAP, 1753999dad82SPeter Xu /* Careful - we need to zap private pages too! */ 1754999dad82SPeter Xu .even_cows = true, 1755999dad82SPeter Xu }; 1756f5cc4eefSAl Viro 17577d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma->vm_mm, 17586f4f13e8SJérôme Glisse start_addr, end_addr); 1759ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 1760763ecb03SLiam R. Howlett do { 17612820b0f0SRik van Riel unsigned long start = start_addr; 17622820b0f0SRik van Riel unsigned long end = end_addr; 17632820b0f0SRik van Riel hugetlb_zap_begin(vma, &start, &end); 17642820b0f0SRik van Riel unmap_single_vma(tlb, vma, start, end, &details, 176568f48381SSuren Baghdasaryan mm_wr_locked); 17662820b0f0SRik van Riel hugetlb_zap_end(vma, &details); 1767d2406291SPeng Zhang vma = mas_find(mas, tree_end - 1); 1768d2406291SPeng Zhang } while (vma && likely(!xa_is_zero(vma))); 1769ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 17701da177e4SLinus Torvalds } 17711da177e4SLinus Torvalds 17721da177e4SLinus Torvalds /** 1773f5cc4eefSAl Viro * zap_page_range_single - remove user pages in a given range 1774f5cc4eefSAl Viro * @vma: vm_area_struct holding the applicable pages 1775f5cc4eefSAl Viro * @address: starting address of pages to zap 1776f5cc4eefSAl Viro * @size: number of bytes to zap 17778a5f14a2SKirill A. Shutemov * @details: details of shared cache invalidation 1778f5cc4eefSAl Viro * 1779f5cc4eefSAl Viro * The range must fit into one VMA. 1780f5cc4eefSAl Viro */ 178121b85b09SMike Kravetz void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 1782f5cc4eefSAl Viro unsigned long size, struct zap_details *details) 1783f5cc4eefSAl Viro { 178421b85b09SMike Kravetz const unsigned long end = address + size; 1785ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1786f5cc4eefSAl Viro struct mmu_gather tlb; 1787f5cc4eefSAl Viro 1788f5cc4eefSAl Viro lru_add_drain(); 17897d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma->vm_mm, 179021b85b09SMike Kravetz address, end); 17912820b0f0SRik van Riel hugetlb_zap_begin(vma, &range.start, &range.end); 1792a72afd87SWill Deacon tlb_gather_mmu(&tlb, vma->vm_mm); 1793ac46d4f3SJérôme Glisse update_hiwater_rss(vma->vm_mm); 1794ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 179521b85b09SMike Kravetz /* 179621b85b09SMike Kravetz * unmap 'address-end' not 'range.start-range.end' as range 179721b85b09SMike Kravetz * could have been expanded for hugetlb pmd sharing. 179821b85b09SMike Kravetz */ 179968f48381SSuren Baghdasaryan unmap_single_vma(&tlb, vma, address, end, details, false); 1800ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 1801ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 18022820b0f0SRik van Riel hugetlb_zap_end(vma, details); 18031da177e4SLinus Torvalds } 18041da177e4SLinus Torvalds 1805c627f9ccSJack Steiner /** 1806c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1807c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1808c627f9ccSJack Steiner * @address: starting address of pages to zap 1809c627f9ccSJack Steiner * @size: number of bytes to zap 1810c627f9ccSJack Steiner * 1811c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1812c627f9ccSJack Steiner * 1813c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1814c627f9ccSJack Steiner * 1815c627f9ccSJack Steiner */ 181627d036e3SLeon Romanovsky void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1817c627f9ccSJack Steiner unsigned long size) 1818c627f9ccSJack Steiner { 181988a35912SMiaohe Lin if (!range_in_vma(vma, address, address + size) || 1820c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 182127d036e3SLeon Romanovsky return; 182227d036e3SLeon Romanovsky 1823f5cc4eefSAl Viro zap_page_range_single(vma, address, size, NULL); 1824c627f9ccSJack Steiner } 1825c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1826c627f9ccSJack Steiner 18278cd3984dSArjun Roy static pmd_t *walk_to_pmd(struct mm_struct *mm, unsigned long addr) 1828c9cfcddfSLinus Torvalds { 1829c2febafcSKirill A. Shutemov pgd_t *pgd; 1830c2febafcSKirill A. Shutemov p4d_t *p4d; 1831c2febafcSKirill A. Shutemov pud_t *pud; 1832c2febafcSKirill A. Shutemov pmd_t *pmd; 1833c2febafcSKirill A. Shutemov 1834c2febafcSKirill A. Shutemov pgd = pgd_offset(mm, addr); 1835c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 1836c2febafcSKirill A. Shutemov if (!p4d) 1837c2febafcSKirill A. Shutemov return NULL; 1838c2febafcSKirill A. Shutemov pud = pud_alloc(mm, p4d, addr); 1839c2febafcSKirill A. Shutemov if (!pud) 1840c2febafcSKirill A. Shutemov return NULL; 1841c2febafcSKirill A. Shutemov pmd = pmd_alloc(mm, pud, addr); 1842c2febafcSKirill A. Shutemov if (!pmd) 1843c2febafcSKirill A. Shutemov return NULL; 1844c2febafcSKirill A. Shutemov 1845f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 18468cd3984dSArjun Roy return pmd; 18478cd3984dSArjun Roy } 18488cd3984dSArjun Roy 18498cd3984dSArjun Roy pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 18508cd3984dSArjun Roy spinlock_t **ptl) 18518cd3984dSArjun Roy { 18528cd3984dSArjun Roy pmd_t *pmd = walk_to_pmd(mm, addr); 18538cd3984dSArjun Roy 18548cd3984dSArjun Roy if (!pmd) 18558cd3984dSArjun Roy return NULL; 1856c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 1857c9cfcddfSLinus Torvalds } 1858c9cfcddfSLinus Torvalds 18598efd6f5bSArjun Roy static int validate_page_before_insert(struct page *page) 18608efd6f5bSArjun Roy { 1861f8b6187dSKefeng Wang struct folio *folio = page_folio(page); 1862f8b6187dSKefeng Wang 1863f8b6187dSKefeng Wang if (folio_test_anon(folio) || folio_test_slab(folio) || 1864f8b6187dSKefeng Wang page_has_type(page)) 18658efd6f5bSArjun Roy return -EINVAL; 1866f8b6187dSKefeng Wang flush_dcache_folio(folio); 18678efd6f5bSArjun Roy return 0; 18688efd6f5bSArjun Roy } 18698efd6f5bSArjun Roy 1870cea86fe2SHugh Dickins static int insert_page_into_pte_locked(struct vm_area_struct *vma, pte_t *pte, 18718efd6f5bSArjun Roy unsigned long addr, struct page *page, pgprot_t prot) 18728efd6f5bSArjun Roy { 1873ef37b2eaSDavid Hildenbrand struct folio *folio = page_folio(page); 1874ef37b2eaSDavid Hildenbrand 1875c33c7948SRyan Roberts if (!pte_none(ptep_get(pte))) 18768efd6f5bSArjun Roy return -EBUSY; 18778efd6f5bSArjun Roy /* Ok, finally just insert the thing.. */ 1878ef37b2eaSDavid Hildenbrand folio_get(folio); 18796b27cc6cSKefeng Wang inc_mm_counter(vma->vm_mm, mm_counter_file(folio)); 1880ef37b2eaSDavid Hildenbrand folio_add_file_rmap_pte(folio, page, vma); 1881cea86fe2SHugh Dickins set_pte_at(vma->vm_mm, addr, pte, mk_pte(page, prot)); 18828efd6f5bSArjun Roy return 0; 18838efd6f5bSArjun Roy } 18848efd6f5bSArjun Roy 18851da177e4SLinus Torvalds /* 1886238f58d8SLinus Torvalds * This is the old fallback for page remapping. 1887238f58d8SLinus Torvalds * 1888238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 1889238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 1890238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 1891238f58d8SLinus Torvalds */ 1892423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 1893423bad60SNick Piggin struct page *page, pgprot_t prot) 1894238f58d8SLinus Torvalds { 1895238f58d8SLinus Torvalds int retval; 1896238f58d8SLinus Torvalds pte_t *pte; 1897238f58d8SLinus Torvalds spinlock_t *ptl; 1898238f58d8SLinus Torvalds 18998efd6f5bSArjun Roy retval = validate_page_before_insert(page); 19008efd6f5bSArjun Roy if (retval) 19015b4e655eSKAMEZAWA Hiroyuki goto out; 1902238f58d8SLinus Torvalds retval = -ENOMEM; 1903cea86fe2SHugh Dickins pte = get_locked_pte(vma->vm_mm, addr, &ptl); 1904238f58d8SLinus Torvalds if (!pte) 19055b4e655eSKAMEZAWA Hiroyuki goto out; 1906cea86fe2SHugh Dickins retval = insert_page_into_pte_locked(vma, pte, addr, page, prot); 1907238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 1908238f58d8SLinus Torvalds out: 1909238f58d8SLinus Torvalds return retval; 1910238f58d8SLinus Torvalds } 1911238f58d8SLinus Torvalds 1912cea86fe2SHugh Dickins static int insert_page_in_batch_locked(struct vm_area_struct *vma, pte_t *pte, 19138cd3984dSArjun Roy unsigned long addr, struct page *page, pgprot_t prot) 19148cd3984dSArjun Roy { 19158cd3984dSArjun Roy int err; 19168cd3984dSArjun Roy 19178cd3984dSArjun Roy if (!page_count(page)) 19188cd3984dSArjun Roy return -EINVAL; 19198cd3984dSArjun Roy err = validate_page_before_insert(page); 19207f70c2a6SArjun Roy if (err) 19217f70c2a6SArjun Roy return err; 1922cea86fe2SHugh Dickins return insert_page_into_pte_locked(vma, pte, addr, page, prot); 19238cd3984dSArjun Roy } 19248cd3984dSArjun Roy 19258cd3984dSArjun Roy /* insert_pages() amortizes the cost of spinlock operations 1926bb7dbaafSMatthew Wilcox (Oracle) * when inserting pages in a loop. 19278cd3984dSArjun Roy */ 19288cd3984dSArjun Roy static int insert_pages(struct vm_area_struct *vma, unsigned long addr, 19298cd3984dSArjun Roy struct page **pages, unsigned long *num, pgprot_t prot) 19308cd3984dSArjun Roy { 19318cd3984dSArjun Roy pmd_t *pmd = NULL; 19327f70c2a6SArjun Roy pte_t *start_pte, *pte; 19337f70c2a6SArjun Roy spinlock_t *pte_lock; 19348cd3984dSArjun Roy struct mm_struct *const mm = vma->vm_mm; 19358cd3984dSArjun Roy unsigned long curr_page_idx = 0; 19368cd3984dSArjun Roy unsigned long remaining_pages_total = *num; 19378cd3984dSArjun Roy unsigned long pages_to_write_in_pmd; 19388cd3984dSArjun Roy int ret; 19398cd3984dSArjun Roy more: 19408cd3984dSArjun Roy ret = -EFAULT; 19418cd3984dSArjun Roy pmd = walk_to_pmd(mm, addr); 19428cd3984dSArjun Roy if (!pmd) 19438cd3984dSArjun Roy goto out; 19448cd3984dSArjun Roy 19458cd3984dSArjun Roy pages_to_write_in_pmd = min_t(unsigned long, 19468cd3984dSArjun Roy remaining_pages_total, PTRS_PER_PTE - pte_index(addr)); 19478cd3984dSArjun Roy 19488cd3984dSArjun Roy /* Allocate the PTE if necessary; takes PMD lock once only. */ 19498cd3984dSArjun Roy ret = -ENOMEM; 19508cd3984dSArjun Roy if (pte_alloc(mm, pmd)) 19518cd3984dSArjun Roy goto out; 19528cd3984dSArjun Roy 19538cd3984dSArjun Roy while (pages_to_write_in_pmd) { 19548cd3984dSArjun Roy int pte_idx = 0; 19558cd3984dSArjun Roy const int batch_size = min_t(int, pages_to_write_in_pmd, 8); 19568cd3984dSArjun Roy 19577f70c2a6SArjun Roy start_pte = pte_offset_map_lock(mm, pmd, addr, &pte_lock); 19583db82b93SHugh Dickins if (!start_pte) { 19593db82b93SHugh Dickins ret = -EFAULT; 19603db82b93SHugh Dickins goto out; 19613db82b93SHugh Dickins } 19627f70c2a6SArjun Roy for (pte = start_pte; pte_idx < batch_size; ++pte, ++pte_idx) { 1963cea86fe2SHugh Dickins int err = insert_page_in_batch_locked(vma, pte, 19648cd3984dSArjun Roy addr, pages[curr_page_idx], prot); 19658cd3984dSArjun Roy if (unlikely(err)) { 19667f70c2a6SArjun Roy pte_unmap_unlock(start_pte, pte_lock); 19678cd3984dSArjun Roy ret = err; 19688cd3984dSArjun Roy remaining_pages_total -= pte_idx; 19698cd3984dSArjun Roy goto out; 19708cd3984dSArjun Roy } 19718cd3984dSArjun Roy addr += PAGE_SIZE; 19728cd3984dSArjun Roy ++curr_page_idx; 19738cd3984dSArjun Roy } 19747f70c2a6SArjun Roy pte_unmap_unlock(start_pte, pte_lock); 19758cd3984dSArjun Roy pages_to_write_in_pmd -= batch_size; 19768cd3984dSArjun Roy remaining_pages_total -= batch_size; 19778cd3984dSArjun Roy } 19788cd3984dSArjun Roy if (remaining_pages_total) 19798cd3984dSArjun Roy goto more; 19808cd3984dSArjun Roy ret = 0; 19818cd3984dSArjun Roy out: 19828cd3984dSArjun Roy *num = remaining_pages_total; 19838cd3984dSArjun Roy return ret; 19848cd3984dSArjun Roy } 19858cd3984dSArjun Roy 19868cd3984dSArjun Roy /** 19878cd3984dSArjun Roy * vm_insert_pages - insert multiple pages into user vma, batching the pmd lock. 19888cd3984dSArjun Roy * @vma: user vma to map to 19898cd3984dSArjun Roy * @addr: target start user address of these pages 19908cd3984dSArjun Roy * @pages: source kernel pages 19918cd3984dSArjun Roy * @num: in: number of pages to map. out: number of pages that were *not* 19928cd3984dSArjun Roy * mapped. (0 means all pages were successfully mapped). 19938cd3984dSArjun Roy * 19948cd3984dSArjun Roy * Preferred over vm_insert_page() when inserting multiple pages. 19958cd3984dSArjun Roy * 19968cd3984dSArjun Roy * In case of error, we may have mapped a subset of the provided 19978cd3984dSArjun Roy * pages. It is the caller's responsibility to account for this case. 19988cd3984dSArjun Roy * 19998cd3984dSArjun Roy * The same restrictions apply as in vm_insert_page(). 20008cd3984dSArjun Roy */ 20018cd3984dSArjun Roy int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr, 20028cd3984dSArjun Roy struct page **pages, unsigned long *num) 20038cd3984dSArjun Roy { 20048cd3984dSArjun Roy const unsigned long end_addr = addr + (*num * PAGE_SIZE) - 1; 20058cd3984dSArjun Roy 20068cd3984dSArjun Roy if (addr < vma->vm_start || end_addr >= vma->vm_end) 20078cd3984dSArjun Roy return -EFAULT; 20088cd3984dSArjun Roy if (!(vma->vm_flags & VM_MIXEDMAP)) { 2009d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(vma->vm_mm)); 20108cd3984dSArjun Roy BUG_ON(vma->vm_flags & VM_PFNMAP); 20111c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_MIXEDMAP); 20128cd3984dSArjun Roy } 20138cd3984dSArjun Roy /* Defer page refcount checking till we're about to map that page. */ 20148cd3984dSArjun Roy return insert_pages(vma, addr, pages, num, vma->vm_page_prot); 20158cd3984dSArjun Roy } 20168cd3984dSArjun Roy EXPORT_SYMBOL(vm_insert_pages); 20178cd3984dSArjun Roy 2018bfa5bf6dSRolf Eike Beer /** 2019bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 2020bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 2021bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 2022bfa5bf6dSRolf Eike Beer * @page: source kernel page 2023bfa5bf6dSRolf Eike Beer * 2024a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 2025a145dd41SLinus Torvalds * into a user vma. 2026a145dd41SLinus Torvalds * 2027a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 2028a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 2029a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 20308dfcc9baSNick Piggin * (see split_page()). 2031a145dd41SLinus Torvalds * 2032a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 2033a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 2034a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 2035a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 2036a145dd41SLinus Torvalds * ask for a shared writable mapping! 2037a145dd41SLinus Torvalds * 2038a145dd41SLinus Torvalds * The page does not need to be reserved. 20394b6e1e37SKonstantin Khlebnikov * 20404b6e1e37SKonstantin Khlebnikov * Usually this function is called from f_op->mmap() handler 2041c1e8d7c6SMichel Lespinasse * under mm->mmap_lock write-lock, so it can change vma->vm_flags. 20424b6e1e37SKonstantin Khlebnikov * Caller must set VM_MIXEDMAP on vma if it wants to call this 20434b6e1e37SKonstantin Khlebnikov * function from other places, for example from page-fault handler. 2044a862f68aSMike Rapoport * 2045a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 2046a145dd41SLinus Torvalds */ 2047423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 2048423bad60SNick Piggin struct page *page) 2049a145dd41SLinus Torvalds { 2050a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 2051a145dd41SLinus Torvalds return -EFAULT; 2052a145dd41SLinus Torvalds if (!page_count(page)) 2053a145dd41SLinus Torvalds return -EINVAL; 20544b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & VM_MIXEDMAP)) { 2055d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(vma->vm_mm)); 20564b6e1e37SKonstantin Khlebnikov BUG_ON(vma->vm_flags & VM_PFNMAP); 20571c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_MIXEDMAP); 20584b6e1e37SKonstantin Khlebnikov } 2059423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 2060a145dd41SLinus Torvalds } 2061e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 2062a145dd41SLinus Torvalds 2063a667d745SSouptick Joarder /* 2064a667d745SSouptick Joarder * __vm_map_pages - maps range of kernel pages into user vma 2065a667d745SSouptick Joarder * @vma: user vma to map to 2066a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2067a667d745SSouptick Joarder * @num: number of pages in page array 2068a667d745SSouptick Joarder * @offset: user's requested vm_pgoff 2069a667d745SSouptick Joarder * 2070a667d745SSouptick Joarder * This allows drivers to map range of kernel pages into a user vma. 2071a667d745SSouptick Joarder * 2072a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2073a667d745SSouptick Joarder */ 2074a667d745SSouptick Joarder static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages, 2075a667d745SSouptick Joarder unsigned long num, unsigned long offset) 2076a667d745SSouptick Joarder { 2077a667d745SSouptick Joarder unsigned long count = vma_pages(vma); 2078a667d745SSouptick Joarder unsigned long uaddr = vma->vm_start; 2079a667d745SSouptick Joarder int ret, i; 2080a667d745SSouptick Joarder 2081a667d745SSouptick Joarder /* Fail if the user requested offset is beyond the end of the object */ 208296756fcbSMiguel Ojeda if (offset >= num) 2083a667d745SSouptick Joarder return -ENXIO; 2084a667d745SSouptick Joarder 2085a667d745SSouptick Joarder /* Fail if the user requested size exceeds available object size */ 2086a667d745SSouptick Joarder if (count > num - offset) 2087a667d745SSouptick Joarder return -ENXIO; 2088a667d745SSouptick Joarder 2089a667d745SSouptick Joarder for (i = 0; i < count; i++) { 2090a667d745SSouptick Joarder ret = vm_insert_page(vma, uaddr, pages[offset + i]); 2091a667d745SSouptick Joarder if (ret < 0) 2092a667d745SSouptick Joarder return ret; 2093a667d745SSouptick Joarder uaddr += PAGE_SIZE; 2094a667d745SSouptick Joarder } 2095a667d745SSouptick Joarder 2096a667d745SSouptick Joarder return 0; 2097a667d745SSouptick Joarder } 2098a667d745SSouptick Joarder 2099a667d745SSouptick Joarder /** 2100a667d745SSouptick Joarder * vm_map_pages - maps range of kernel pages starts with non zero offset 2101a667d745SSouptick Joarder * @vma: user vma to map to 2102a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2103a667d745SSouptick Joarder * @num: number of pages in page array 2104a667d745SSouptick Joarder * 2105a667d745SSouptick Joarder * Maps an object consisting of @num pages, catering for the user's 2106a667d745SSouptick Joarder * requested vm_pgoff 2107a667d745SSouptick Joarder * 2108a667d745SSouptick Joarder * If we fail to insert any page into the vma, the function will return 2109a667d745SSouptick Joarder * immediately leaving any previously inserted pages present. Callers 2110a667d745SSouptick Joarder * from the mmap handler may immediately return the error as their caller 2111a667d745SSouptick Joarder * will destroy the vma, removing any successfully inserted pages. Other 2112a667d745SSouptick Joarder * callers should make their own arrangements for calling unmap_region(). 2113a667d745SSouptick Joarder * 2114a667d745SSouptick Joarder * Context: Process context. Called by mmap handlers. 2115a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2116a667d745SSouptick Joarder */ 2117a667d745SSouptick Joarder int vm_map_pages(struct vm_area_struct *vma, struct page **pages, 2118a667d745SSouptick Joarder unsigned long num) 2119a667d745SSouptick Joarder { 2120a667d745SSouptick Joarder return __vm_map_pages(vma, pages, num, vma->vm_pgoff); 2121a667d745SSouptick Joarder } 2122a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages); 2123a667d745SSouptick Joarder 2124a667d745SSouptick Joarder /** 2125a667d745SSouptick Joarder * vm_map_pages_zero - map range of kernel pages starts with zero offset 2126a667d745SSouptick Joarder * @vma: user vma to map to 2127a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2128a667d745SSouptick Joarder * @num: number of pages in page array 2129a667d745SSouptick Joarder * 2130a667d745SSouptick Joarder * Similar to vm_map_pages(), except that it explicitly sets the offset 2131a667d745SSouptick Joarder * to 0. This function is intended for the drivers that did not consider 2132a667d745SSouptick Joarder * vm_pgoff. 2133a667d745SSouptick Joarder * 2134a667d745SSouptick Joarder * Context: Process context. Called by mmap handlers. 2135a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2136a667d745SSouptick Joarder */ 2137a667d745SSouptick Joarder int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, 2138a667d745SSouptick Joarder unsigned long num) 2139a667d745SSouptick Joarder { 2140a667d745SSouptick Joarder return __vm_map_pages(vma, pages, num, 0); 2141a667d745SSouptick Joarder } 2142a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages_zero); 2143a667d745SSouptick Joarder 21449b5a8e00SMatthew Wilcox static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2145b2770da6SRoss Zwisler pfn_t pfn, pgprot_t prot, bool mkwrite) 2146423bad60SNick Piggin { 2147423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 2148423bad60SNick Piggin pte_t *pte, entry; 2149423bad60SNick Piggin spinlock_t *ptl; 2150423bad60SNick Piggin 2151423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 2152423bad60SNick Piggin if (!pte) 21539b5a8e00SMatthew Wilcox return VM_FAULT_OOM; 2154c33c7948SRyan Roberts entry = ptep_get(pte); 2155c33c7948SRyan Roberts if (!pte_none(entry)) { 2156b2770da6SRoss Zwisler if (mkwrite) { 2157b2770da6SRoss Zwisler /* 2158b2770da6SRoss Zwisler * For read faults on private mappings the PFN passed 2159b2770da6SRoss Zwisler * in may not match the PFN we have mapped if the 2160b2770da6SRoss Zwisler * mapped PFN is a writeable COW page. In the mkwrite 2161b2770da6SRoss Zwisler * case we are creating a writable PTE for a shared 2162f2c57d91SJan Kara * mapping and we expect the PFNs to match. If they 2163f2c57d91SJan Kara * don't match, we are likely racing with block 2164f2c57d91SJan Kara * allocation and mapping invalidation so just skip the 2165f2c57d91SJan Kara * update. 2166b2770da6SRoss Zwisler */ 2167c33c7948SRyan Roberts if (pte_pfn(entry) != pfn_t_to_pfn(pfn)) { 2168c33c7948SRyan Roberts WARN_ON_ONCE(!is_zero_pfn(pte_pfn(entry))); 2169423bad60SNick Piggin goto out_unlock; 2170f2c57d91SJan Kara } 2171c33c7948SRyan Roberts entry = pte_mkyoung(entry); 2172cae85cb8SJan Kara entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2173cae85cb8SJan Kara if (ptep_set_access_flags(vma, addr, pte, entry, 1)) 2174cae85cb8SJan Kara update_mmu_cache(vma, addr, pte); 2175cae85cb8SJan Kara } 2176b2770da6SRoss Zwisler goto out_unlock; 2177b2770da6SRoss Zwisler } 2178423bad60SNick Piggin 2179423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 218001c8f1c4SDan Williams if (pfn_t_devmap(pfn)) 218101c8f1c4SDan Williams entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); 218201c8f1c4SDan Williams else 218301c8f1c4SDan Williams entry = pte_mkspecial(pfn_t_pte(pfn, prot)); 2184b2770da6SRoss Zwisler 2185b2770da6SRoss Zwisler if (mkwrite) { 2186b2770da6SRoss Zwisler entry = pte_mkyoung(entry); 2187b2770da6SRoss Zwisler entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2188b2770da6SRoss Zwisler } 2189b2770da6SRoss Zwisler 2190423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 21914b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 2192423bad60SNick Piggin 2193423bad60SNick Piggin out_unlock: 2194423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 21959b5a8e00SMatthew Wilcox return VM_FAULT_NOPAGE; 2196423bad60SNick Piggin } 2197423bad60SNick Piggin 2198f5e6d1d5SMatthew Wilcox /** 2199f5e6d1d5SMatthew Wilcox * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot 2200f5e6d1d5SMatthew Wilcox * @vma: user vma to map to 2201f5e6d1d5SMatthew Wilcox * @addr: target user address of this page 2202f5e6d1d5SMatthew Wilcox * @pfn: source kernel pfn 2203f5e6d1d5SMatthew Wilcox * @pgprot: pgprot flags for the inserted page 2204f5e6d1d5SMatthew Wilcox * 2205a1a0aea5SRandy Dunlap * This is exactly like vmf_insert_pfn(), except that it allows drivers 2206f5e6d1d5SMatthew Wilcox * to override pgprot on a per-page basis. 2207f5e6d1d5SMatthew Wilcox * 2208f5e6d1d5SMatthew Wilcox * This only makes sense for IO mappings, and it makes no sense for 2209f5e6d1d5SMatthew Wilcox * COW mappings. In general, using multiple vmas is preferable; 2210ae2b01f3SMatthew Wilcox * vmf_insert_pfn_prot should only be used if using multiple VMAs is 2211f5e6d1d5SMatthew Wilcox * impractical. 2212f5e6d1d5SMatthew Wilcox * 221328d8b812SLorenzo Stoakes * pgprot typically only differs from @vma->vm_page_prot when drivers set 221428d8b812SLorenzo Stoakes * caching- and encryption bits different than those of @vma->vm_page_prot, 221528d8b812SLorenzo Stoakes * because the caching- or encryption mode may not be known at mmap() time. 221628d8b812SLorenzo Stoakes * 221728d8b812SLorenzo Stoakes * This is ok as long as @vma->vm_page_prot is not used by the core vm 221828d8b812SLorenzo Stoakes * to set caching and encryption bits for those vmas (except for COW pages). 221928d8b812SLorenzo Stoakes * This is ensured by core vm only modifying these page table entries using 222028d8b812SLorenzo Stoakes * functions that don't touch caching- or encryption bits, using pte_modify() 222128d8b812SLorenzo Stoakes * if needed. (See for example mprotect()). 222228d8b812SLorenzo Stoakes * 222328d8b812SLorenzo Stoakes * Also when new page-table entries are created, this is only done using the 222428d8b812SLorenzo Stoakes * fault() callback, and never using the value of vma->vm_page_prot, 222528d8b812SLorenzo Stoakes * except for page-table entries that point to anonymous pages as the result 222628d8b812SLorenzo Stoakes * of COW. 2227574c5b3dSThomas Hellstrom * 2228ae2b01f3SMatthew Wilcox * Context: Process context. May allocate using %GFP_KERNEL. 2229f5e6d1d5SMatthew Wilcox * Return: vm_fault_t value. 2230f5e6d1d5SMatthew Wilcox */ 2231f5e6d1d5SMatthew Wilcox vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, 2232f5e6d1d5SMatthew Wilcox unsigned long pfn, pgprot_t pgprot) 2233f5e6d1d5SMatthew Wilcox { 22346d958546SMatthew Wilcox /* 22356d958546SMatthew Wilcox * Technically, architectures with pte_special can avoid all these 22366d958546SMatthew Wilcox * restrictions (same for remap_pfn_range). However we would like 22376d958546SMatthew Wilcox * consistency in testing and feature parity among all, so we should 22386d958546SMatthew Wilcox * try to keep these invariants in place for everybody. 22396d958546SMatthew Wilcox */ 22406d958546SMatthew Wilcox BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 22416d958546SMatthew Wilcox BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 22426d958546SMatthew Wilcox (VM_PFNMAP|VM_MIXEDMAP)); 22436d958546SMatthew Wilcox BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 22446d958546SMatthew Wilcox BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 22456d958546SMatthew Wilcox 22466d958546SMatthew Wilcox if (addr < vma->vm_start || addr >= vma->vm_end) 22476d958546SMatthew Wilcox return VM_FAULT_SIGBUS; 22486d958546SMatthew Wilcox 22496d958546SMatthew Wilcox if (!pfn_modify_allowed(pfn, pgprot)) 22506d958546SMatthew Wilcox return VM_FAULT_SIGBUS; 22516d958546SMatthew Wilcox 22526d958546SMatthew Wilcox track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); 22536d958546SMatthew Wilcox 22549b5a8e00SMatthew Wilcox return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, 22556d958546SMatthew Wilcox false); 2256f5e6d1d5SMatthew Wilcox } 2257f5e6d1d5SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_pfn_prot); 2258e0dc0d8fSNick Piggin 2259ae2b01f3SMatthew Wilcox /** 2260ae2b01f3SMatthew Wilcox * vmf_insert_pfn - insert single pfn into user vma 2261ae2b01f3SMatthew Wilcox * @vma: user vma to map to 2262ae2b01f3SMatthew Wilcox * @addr: target user address of this page 2263ae2b01f3SMatthew Wilcox * @pfn: source kernel pfn 2264ae2b01f3SMatthew Wilcox * 2265ae2b01f3SMatthew Wilcox * Similar to vm_insert_page, this allows drivers to insert individual pages 2266ae2b01f3SMatthew Wilcox * they've allocated into a user vma. Same comments apply. 2267ae2b01f3SMatthew Wilcox * 2268ae2b01f3SMatthew Wilcox * This function should only be called from a vm_ops->fault handler, and 2269ae2b01f3SMatthew Wilcox * in that case the handler should return the result of this function. 2270ae2b01f3SMatthew Wilcox * 2271ae2b01f3SMatthew Wilcox * vma cannot be a COW mapping. 2272ae2b01f3SMatthew Wilcox * 2273ae2b01f3SMatthew Wilcox * As this is called only for pages that do not currently exist, we 2274ae2b01f3SMatthew Wilcox * do not need to flush old virtual caches or the TLB. 2275ae2b01f3SMatthew Wilcox * 2276ae2b01f3SMatthew Wilcox * Context: Process context. May allocate using %GFP_KERNEL. 2277ae2b01f3SMatthew Wilcox * Return: vm_fault_t value. 2278ae2b01f3SMatthew Wilcox */ 2279ae2b01f3SMatthew Wilcox vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2280ae2b01f3SMatthew Wilcox unsigned long pfn) 2281ae2b01f3SMatthew Wilcox { 2282ae2b01f3SMatthew Wilcox return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); 2283ae2b01f3SMatthew Wilcox } 2284ae2b01f3SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_pfn); 2285ae2b01f3SMatthew Wilcox 2286785a3fabSDan Williams static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) 2287785a3fabSDan Williams { 2288785a3fabSDan Williams /* these checks mirror the abort conditions in vm_normal_page */ 2289785a3fabSDan Williams if (vma->vm_flags & VM_MIXEDMAP) 2290785a3fabSDan Williams return true; 2291785a3fabSDan Williams if (pfn_t_devmap(pfn)) 2292785a3fabSDan Williams return true; 2293785a3fabSDan Williams if (pfn_t_special(pfn)) 2294785a3fabSDan Williams return true; 2295785a3fabSDan Williams if (is_zero_pfn(pfn_t_to_pfn(pfn))) 2296785a3fabSDan Williams return true; 2297785a3fabSDan Williams return false; 2298785a3fabSDan Williams } 2299785a3fabSDan Williams 230079f3aa5bSMatthew Wilcox static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma, 230128d8b812SLorenzo Stoakes unsigned long addr, pfn_t pfn, bool mkwrite) 2302423bad60SNick Piggin { 230328d8b812SLorenzo Stoakes pgprot_t pgprot = vma->vm_page_prot; 230479f3aa5bSMatthew Wilcox int err; 230587744ab3SDan Williams 2306785a3fabSDan Williams BUG_ON(!vm_mixed_ok(vma, pfn)); 2307423bad60SNick Piggin 2308423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 230979f3aa5bSMatthew Wilcox return VM_FAULT_SIGBUS; 2310308a047cSBorislav Petkov 2311308a047cSBorislav Petkov track_pfn_insert(vma, &pgprot, pfn); 2312423bad60SNick Piggin 231342e4089cSAndi Kleen if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) 231479f3aa5bSMatthew Wilcox return VM_FAULT_SIGBUS; 231542e4089cSAndi Kleen 2316423bad60SNick Piggin /* 2317423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 2318423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 2319423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 232062eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 232162eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 2322423bad60SNick Piggin */ 232300b3a331SLaurent Dufour if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && 232400b3a331SLaurent Dufour !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { 2325423bad60SNick Piggin struct page *page; 2326423bad60SNick Piggin 232703fc2da6SDan Williams /* 232803fc2da6SDan Williams * At this point we are committed to insert_page() 232903fc2da6SDan Williams * regardless of whether the caller specified flags that 233003fc2da6SDan Williams * result in pfn_t_has_page() == false. 233103fc2da6SDan Williams */ 233203fc2da6SDan Williams page = pfn_to_page(pfn_t_to_pfn(pfn)); 233379f3aa5bSMatthew Wilcox err = insert_page(vma, addr, page, pgprot); 233479f3aa5bSMatthew Wilcox } else { 23359b5a8e00SMatthew Wilcox return insert_pfn(vma, addr, pfn, pgprot, mkwrite); 2336423bad60SNick Piggin } 2337b2770da6SRoss Zwisler 23385d747637SMatthew Wilcox if (err == -ENOMEM) 23395d747637SMatthew Wilcox return VM_FAULT_OOM; 23405d747637SMatthew Wilcox if (err < 0 && err != -EBUSY) 23415d747637SMatthew Wilcox return VM_FAULT_SIGBUS; 23425d747637SMatthew Wilcox 23435d747637SMatthew Wilcox return VM_FAULT_NOPAGE; 2344423bad60SNick Piggin } 234579f3aa5bSMatthew Wilcox 234679f3aa5bSMatthew Wilcox vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 234779f3aa5bSMatthew Wilcox pfn_t pfn) 234879f3aa5bSMatthew Wilcox { 234928d8b812SLorenzo Stoakes return __vm_insert_mixed(vma, addr, pfn, false); 235079f3aa5bSMatthew Wilcox } 23515d747637SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_mixed); 2352423bad60SNick Piggin 2353ab77dab4SSouptick Joarder /* 2354ab77dab4SSouptick Joarder * If the insertion of PTE failed because someone else already added a 2355ab77dab4SSouptick Joarder * different entry in the mean time, we treat that as success as we assume 2356ab77dab4SSouptick Joarder * the same entry was actually inserted. 2357ab77dab4SSouptick Joarder */ 2358ab77dab4SSouptick Joarder vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, 2359ab77dab4SSouptick Joarder unsigned long addr, pfn_t pfn) 2360b2770da6SRoss Zwisler { 236128d8b812SLorenzo Stoakes return __vm_insert_mixed(vma, addr, pfn, true); 2362b2770da6SRoss Zwisler } 2363ab77dab4SSouptick Joarder EXPORT_SYMBOL(vmf_insert_mixed_mkwrite); 2364b2770da6SRoss Zwisler 2365a145dd41SLinus Torvalds /* 23661da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 23671da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 23681da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 23691da177e4SLinus Torvalds */ 23701da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 23711da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23721da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23731da177e4SLinus Torvalds { 237490a3e375SMiaohe Lin pte_t *pte, *mapped_pte; 2375c74df32cSHugh Dickins spinlock_t *ptl; 237642e4089cSAndi Kleen int err = 0; 23771da177e4SLinus Torvalds 237890a3e375SMiaohe Lin mapped_pte = pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 23791da177e4SLinus Torvalds if (!pte) 23801da177e4SLinus Torvalds return -ENOMEM; 23816606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 23821da177e4SLinus Torvalds do { 2383c33c7948SRyan Roberts BUG_ON(!pte_none(ptep_get(pte))); 238442e4089cSAndi Kleen if (!pfn_modify_allowed(pfn, prot)) { 238542e4089cSAndi Kleen err = -EACCES; 238642e4089cSAndi Kleen break; 238742e4089cSAndi Kleen } 23887e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 23891da177e4SLinus Torvalds pfn++; 23901da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 23916606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 239290a3e375SMiaohe Lin pte_unmap_unlock(mapped_pte, ptl); 239342e4089cSAndi Kleen return err; 23941da177e4SLinus Torvalds } 23951da177e4SLinus Torvalds 23961da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 23971da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23981da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23991da177e4SLinus Torvalds { 24001da177e4SLinus Torvalds pmd_t *pmd; 24011da177e4SLinus Torvalds unsigned long next; 240242e4089cSAndi Kleen int err; 24031da177e4SLinus Torvalds 24041da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 24051da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 24061da177e4SLinus Torvalds if (!pmd) 24071da177e4SLinus Torvalds return -ENOMEM; 2408f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 24091da177e4SLinus Torvalds do { 24101da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 241142e4089cSAndi Kleen err = remap_pte_range(mm, pmd, addr, next, 241242e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 241342e4089cSAndi Kleen if (err) 241442e4089cSAndi Kleen return err; 24151da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 24161da177e4SLinus Torvalds return 0; 24171da177e4SLinus Torvalds } 24181da177e4SLinus Torvalds 2419c2febafcSKirill A. Shutemov static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, 24201da177e4SLinus Torvalds unsigned long addr, unsigned long end, 24211da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 24221da177e4SLinus Torvalds { 24231da177e4SLinus Torvalds pud_t *pud; 24241da177e4SLinus Torvalds unsigned long next; 242542e4089cSAndi Kleen int err; 24261da177e4SLinus Torvalds 24271da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 2428c2febafcSKirill A. Shutemov pud = pud_alloc(mm, p4d, addr); 24291da177e4SLinus Torvalds if (!pud) 24301da177e4SLinus Torvalds return -ENOMEM; 24311da177e4SLinus Torvalds do { 24321da177e4SLinus Torvalds next = pud_addr_end(addr, end); 243342e4089cSAndi Kleen err = remap_pmd_range(mm, pud, addr, next, 243442e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 243542e4089cSAndi Kleen if (err) 243642e4089cSAndi Kleen return err; 24371da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 24381da177e4SLinus Torvalds return 0; 24391da177e4SLinus Torvalds } 24401da177e4SLinus Torvalds 2441c2febafcSKirill A. Shutemov static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, 2442c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 2443c2febafcSKirill A. Shutemov unsigned long pfn, pgprot_t prot) 2444c2febafcSKirill A. Shutemov { 2445c2febafcSKirill A. Shutemov p4d_t *p4d; 2446c2febafcSKirill A. Shutemov unsigned long next; 244742e4089cSAndi Kleen int err; 2448c2febafcSKirill A. Shutemov 2449c2febafcSKirill A. Shutemov pfn -= addr >> PAGE_SHIFT; 2450c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 2451c2febafcSKirill A. Shutemov if (!p4d) 2452c2febafcSKirill A. Shutemov return -ENOMEM; 2453c2febafcSKirill A. Shutemov do { 2454c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 245542e4089cSAndi Kleen err = remap_pud_range(mm, p4d, addr, next, 245642e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 245742e4089cSAndi Kleen if (err) 245842e4089cSAndi Kleen return err; 2459c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 2460c2febafcSKirill A. Shutemov return 0; 2461c2febafcSKirill A. Shutemov } 2462c2febafcSKirill A. Shutemov 246374ffa5a3SChristoph Hellwig /* 246474ffa5a3SChristoph Hellwig * Variant of remap_pfn_range that does not call track_pfn_remap. The caller 246574ffa5a3SChristoph Hellwig * must have pre-validated the caching bits of the pgprot_t. 2466bfa5bf6dSRolf Eike Beer */ 246774ffa5a3SChristoph Hellwig int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr, 24681da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 24691da177e4SLinus Torvalds { 24701da177e4SLinus Torvalds pgd_t *pgd; 24711da177e4SLinus Torvalds unsigned long next; 24722d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 24731da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 24741da177e4SLinus Torvalds int err; 24751da177e4SLinus Torvalds 24760c4123e3SAlex Zhang if (WARN_ON_ONCE(!PAGE_ALIGNED(addr))) 24770c4123e3SAlex Zhang return -EINVAL; 24780c4123e3SAlex Zhang 24791da177e4SLinus Torvalds /* 24801da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 24811da177e4SLinus Torvalds * rest of the world about it: 24821da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 24831da177e4SLinus Torvalds * (accesses can have side effects). 24846aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 24856aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 24866aab341eSLinus Torvalds * with them. 2487314e51b9SKonstantin Khlebnikov * VM_DONTEXPAND 2488314e51b9SKonstantin Khlebnikov * Disable vma merging and expanding with mremap(). 2489314e51b9SKonstantin Khlebnikov * VM_DONTDUMP 2490314e51b9SKonstantin Khlebnikov * Omit vma from core dump, even when VM_IO turned off. 2491fb155c16SLinus Torvalds * 2492fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 2493fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 2494fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 2495b3b9c293SKonstantin Khlebnikov * See vm_normal_page() for details. 24961da177e4SLinus Torvalds */ 2497b3b9c293SKonstantin Khlebnikov if (is_cow_mapping(vma->vm_flags)) { 2498b3b9c293SKonstantin Khlebnikov if (addr != vma->vm_start || end != vma->vm_end) 2499b3b9c293SKonstantin Khlebnikov return -EINVAL; 25006aab341eSLinus Torvalds vma->vm_pgoff = pfn; 2501b3b9c293SKonstantin Khlebnikov } 2502b3b9c293SKonstantin Khlebnikov 25031c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP); 25041da177e4SLinus Torvalds 25051da177e4SLinus Torvalds BUG_ON(addr >= end); 25061da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 25071da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 25081da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 25091da177e4SLinus Torvalds do { 25101da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 2511c2febafcSKirill A. Shutemov err = remap_p4d_range(mm, pgd, addr, next, 25121da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 25131da177e4SLinus Torvalds if (err) 251474ffa5a3SChristoph Hellwig return err; 25151da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 25162ab64037Svenkatesh.pallipadi@intel.com 251774ffa5a3SChristoph Hellwig return 0; 251874ffa5a3SChristoph Hellwig } 25192ab64037Svenkatesh.pallipadi@intel.com 252074ffa5a3SChristoph Hellwig /** 252174ffa5a3SChristoph Hellwig * remap_pfn_range - remap kernel memory to userspace 252274ffa5a3SChristoph Hellwig * @vma: user vma to map to 252374ffa5a3SChristoph Hellwig * @addr: target page aligned user address to start at 252474ffa5a3SChristoph Hellwig * @pfn: page frame number of kernel physical memory address 252574ffa5a3SChristoph Hellwig * @size: size of mapping area 252674ffa5a3SChristoph Hellwig * @prot: page protection flags for this mapping 252774ffa5a3SChristoph Hellwig * 252874ffa5a3SChristoph Hellwig * Note: this is only safe if the mm semaphore is held when called. 252974ffa5a3SChristoph Hellwig * 253074ffa5a3SChristoph Hellwig * Return: %0 on success, negative error code otherwise. 253174ffa5a3SChristoph Hellwig */ 253274ffa5a3SChristoph Hellwig int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 253374ffa5a3SChristoph Hellwig unsigned long pfn, unsigned long size, pgprot_t prot) 253474ffa5a3SChristoph Hellwig { 253574ffa5a3SChristoph Hellwig int err; 253674ffa5a3SChristoph Hellwig 253774ffa5a3SChristoph Hellwig err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); 253874ffa5a3SChristoph Hellwig if (err) 253974ffa5a3SChristoph Hellwig return -EINVAL; 254074ffa5a3SChristoph Hellwig 254174ffa5a3SChristoph Hellwig err = remap_pfn_range_notrack(vma, addr, pfn, size, prot); 254274ffa5a3SChristoph Hellwig if (err) 254368f48381SSuren Baghdasaryan untrack_pfn(vma, pfn, PAGE_ALIGN(size), true); 25441da177e4SLinus Torvalds return err; 25451da177e4SLinus Torvalds } 25461da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 25471da177e4SLinus Torvalds 2548b4cbb197SLinus Torvalds /** 2549b4cbb197SLinus Torvalds * vm_iomap_memory - remap memory to userspace 2550b4cbb197SLinus Torvalds * @vma: user vma to map to 2551abd69b9eSWang Wenhu * @start: start of the physical memory to be mapped 2552b4cbb197SLinus Torvalds * @len: size of area 2553b4cbb197SLinus Torvalds * 2554b4cbb197SLinus Torvalds * This is a simplified io_remap_pfn_range() for common driver use. The 2555b4cbb197SLinus Torvalds * driver just needs to give us the physical memory range to be mapped, 2556b4cbb197SLinus Torvalds * we'll figure out the rest from the vma information. 2557b4cbb197SLinus Torvalds * 2558b4cbb197SLinus Torvalds * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get 2559b4cbb197SLinus Torvalds * whatever write-combining details or similar. 2560a862f68aSMike Rapoport * 2561a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 2562b4cbb197SLinus Torvalds */ 2563b4cbb197SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 2564b4cbb197SLinus Torvalds { 2565b4cbb197SLinus Torvalds unsigned long vm_len, pfn, pages; 2566b4cbb197SLinus Torvalds 2567b4cbb197SLinus Torvalds /* Check that the physical memory area passed in looks valid */ 2568b4cbb197SLinus Torvalds if (start + len < start) 2569b4cbb197SLinus Torvalds return -EINVAL; 2570b4cbb197SLinus Torvalds /* 2571b4cbb197SLinus Torvalds * You *really* shouldn't map things that aren't page-aligned, 2572b4cbb197SLinus Torvalds * but we've historically allowed it because IO memory might 2573b4cbb197SLinus Torvalds * just have smaller alignment. 2574b4cbb197SLinus Torvalds */ 2575b4cbb197SLinus Torvalds len += start & ~PAGE_MASK; 2576b4cbb197SLinus Torvalds pfn = start >> PAGE_SHIFT; 2577b4cbb197SLinus Torvalds pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; 2578b4cbb197SLinus Torvalds if (pfn + pages < pfn) 2579b4cbb197SLinus Torvalds return -EINVAL; 2580b4cbb197SLinus Torvalds 2581b4cbb197SLinus Torvalds /* We start the mapping 'vm_pgoff' pages into the area */ 2582b4cbb197SLinus Torvalds if (vma->vm_pgoff > pages) 2583b4cbb197SLinus Torvalds return -EINVAL; 2584b4cbb197SLinus Torvalds pfn += vma->vm_pgoff; 2585b4cbb197SLinus Torvalds pages -= vma->vm_pgoff; 2586b4cbb197SLinus Torvalds 2587b4cbb197SLinus Torvalds /* Can we fit all of the mapping? */ 2588b4cbb197SLinus Torvalds vm_len = vma->vm_end - vma->vm_start; 2589b4cbb197SLinus Torvalds if (vm_len >> PAGE_SHIFT > pages) 2590b4cbb197SLinus Torvalds return -EINVAL; 2591b4cbb197SLinus Torvalds 2592b4cbb197SLinus Torvalds /* Ok, let it rip */ 2593b4cbb197SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 2594b4cbb197SLinus Torvalds } 2595b4cbb197SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 2596b4cbb197SLinus Torvalds 2597aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 2598aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2599e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2600e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2601aee16b3cSJeremy Fitzhardinge { 26028abb50c7SMiaohe Lin pte_t *pte, *mapped_pte; 2603be1db475SDaniel Axtens int err = 0; 26043f649ab7SKees Cook spinlock_t *ptl; 2605aee16b3cSJeremy Fitzhardinge 2606be1db475SDaniel Axtens if (create) { 26078abb50c7SMiaohe Lin mapped_pte = pte = (mm == &init_mm) ? 2608e80d3909SJoerg Roedel pte_alloc_kernel_track(pmd, addr, mask) : 2609aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 2610aee16b3cSJeremy Fitzhardinge if (!pte) 2611aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2612be1db475SDaniel Axtens } else { 26138abb50c7SMiaohe Lin mapped_pte = pte = (mm == &init_mm) ? 2614be1db475SDaniel Axtens pte_offset_kernel(pmd, addr) : 2615be1db475SDaniel Axtens pte_offset_map_lock(mm, pmd, addr, &ptl); 26163db82b93SHugh Dickins if (!pte) 26173db82b93SHugh Dickins return -EINVAL; 2618be1db475SDaniel Axtens } 2619aee16b3cSJeremy Fitzhardinge 262038e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 262138e0edb1SJeremy Fitzhardinge 2622eeb4a05fSChristoph Hellwig if (fn) { 2623aee16b3cSJeremy Fitzhardinge do { 2624c33c7948SRyan Roberts if (create || !pte_none(ptep_get(pte))) { 26258b1e0f81SAnshuman Khandual err = fn(pte++, addr, data); 2626aee16b3cSJeremy Fitzhardinge if (err) 2627aee16b3cSJeremy Fitzhardinge break; 2628be1db475SDaniel Axtens } 2629c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 2630eeb4a05fSChristoph Hellwig } 2631e80d3909SJoerg Roedel *mask |= PGTBL_PTE_MODIFIED; 2632aee16b3cSJeremy Fitzhardinge 263338e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 263438e0edb1SJeremy Fitzhardinge 2635aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 26368abb50c7SMiaohe Lin pte_unmap_unlock(mapped_pte, ptl); 2637aee16b3cSJeremy Fitzhardinge return err; 2638aee16b3cSJeremy Fitzhardinge } 2639aee16b3cSJeremy Fitzhardinge 2640aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 2641aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2642e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2643e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2644aee16b3cSJeremy Fitzhardinge { 2645aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 2646aee16b3cSJeremy Fitzhardinge unsigned long next; 2647be1db475SDaniel Axtens int err = 0; 2648aee16b3cSJeremy Fitzhardinge 2649ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 2650ceb86879SAndi Kleen 2651be1db475SDaniel Axtens if (create) { 2652e80d3909SJoerg Roedel pmd = pmd_alloc_track(mm, pud, addr, mask); 2653aee16b3cSJeremy Fitzhardinge if (!pmd) 2654aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2655be1db475SDaniel Axtens } else { 2656be1db475SDaniel Axtens pmd = pmd_offset(pud, addr); 2657be1db475SDaniel Axtens } 2658aee16b3cSJeremy Fitzhardinge do { 2659aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 26600c95cba4SNicholas Piggin if (pmd_none(*pmd) && !create) 26610c95cba4SNicholas Piggin continue; 26620c95cba4SNicholas Piggin if (WARN_ON_ONCE(pmd_leaf(*pmd))) 26630c95cba4SNicholas Piggin return -EINVAL; 26640c95cba4SNicholas Piggin if (!pmd_none(*pmd) && WARN_ON_ONCE(pmd_bad(*pmd))) { 26650c95cba4SNicholas Piggin if (!create) 26660c95cba4SNicholas Piggin continue; 26670c95cba4SNicholas Piggin pmd_clear_bad(pmd); 26680c95cba4SNicholas Piggin } 26690c95cba4SNicholas Piggin err = apply_to_pte_range(mm, pmd, addr, next, 26700c95cba4SNicholas Piggin fn, data, create, mask); 2671aee16b3cSJeremy Fitzhardinge if (err) 2672aee16b3cSJeremy Fitzhardinge break; 2673aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 26740c95cba4SNicholas Piggin 2675aee16b3cSJeremy Fitzhardinge return err; 2676aee16b3cSJeremy Fitzhardinge } 2677aee16b3cSJeremy Fitzhardinge 2678c2febafcSKirill A. Shutemov static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, 2679aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2680e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2681e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2682aee16b3cSJeremy Fitzhardinge { 2683aee16b3cSJeremy Fitzhardinge pud_t *pud; 2684aee16b3cSJeremy Fitzhardinge unsigned long next; 2685be1db475SDaniel Axtens int err = 0; 2686aee16b3cSJeremy Fitzhardinge 2687be1db475SDaniel Axtens if (create) { 2688e80d3909SJoerg Roedel pud = pud_alloc_track(mm, p4d, addr, mask); 2689aee16b3cSJeremy Fitzhardinge if (!pud) 2690aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2691be1db475SDaniel Axtens } else { 2692be1db475SDaniel Axtens pud = pud_offset(p4d, addr); 2693be1db475SDaniel Axtens } 2694aee16b3cSJeremy Fitzhardinge do { 2695aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 26960c95cba4SNicholas Piggin if (pud_none(*pud) && !create) 26970c95cba4SNicholas Piggin continue; 26980c95cba4SNicholas Piggin if (WARN_ON_ONCE(pud_leaf(*pud))) 26990c95cba4SNicholas Piggin return -EINVAL; 27000c95cba4SNicholas Piggin if (!pud_none(*pud) && WARN_ON_ONCE(pud_bad(*pud))) { 27010c95cba4SNicholas Piggin if (!create) 27020c95cba4SNicholas Piggin continue; 27030c95cba4SNicholas Piggin pud_clear_bad(pud); 27040c95cba4SNicholas Piggin } 27050c95cba4SNicholas Piggin err = apply_to_pmd_range(mm, pud, addr, next, 27060c95cba4SNicholas Piggin fn, data, create, mask); 2707aee16b3cSJeremy Fitzhardinge if (err) 2708aee16b3cSJeremy Fitzhardinge break; 2709aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 27100c95cba4SNicholas Piggin 2711aee16b3cSJeremy Fitzhardinge return err; 2712aee16b3cSJeremy Fitzhardinge } 2713aee16b3cSJeremy Fitzhardinge 2714c2febafcSKirill A. Shutemov static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, 2715c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 2716e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2717e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2718c2febafcSKirill A. Shutemov { 2719c2febafcSKirill A. Shutemov p4d_t *p4d; 2720c2febafcSKirill A. Shutemov unsigned long next; 2721be1db475SDaniel Axtens int err = 0; 2722c2febafcSKirill A. Shutemov 2723be1db475SDaniel Axtens if (create) { 2724e80d3909SJoerg Roedel p4d = p4d_alloc_track(mm, pgd, addr, mask); 2725c2febafcSKirill A. Shutemov if (!p4d) 2726c2febafcSKirill A. Shutemov return -ENOMEM; 2727be1db475SDaniel Axtens } else { 2728be1db475SDaniel Axtens p4d = p4d_offset(pgd, addr); 2729be1db475SDaniel Axtens } 2730c2febafcSKirill A. Shutemov do { 2731c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 27320c95cba4SNicholas Piggin if (p4d_none(*p4d) && !create) 27330c95cba4SNicholas Piggin continue; 27340c95cba4SNicholas Piggin if (WARN_ON_ONCE(p4d_leaf(*p4d))) 27350c95cba4SNicholas Piggin return -EINVAL; 27360c95cba4SNicholas Piggin if (!p4d_none(*p4d) && WARN_ON_ONCE(p4d_bad(*p4d))) { 27370c95cba4SNicholas Piggin if (!create) 27380c95cba4SNicholas Piggin continue; 27390c95cba4SNicholas Piggin p4d_clear_bad(p4d); 27400c95cba4SNicholas Piggin } 27410c95cba4SNicholas Piggin err = apply_to_pud_range(mm, p4d, addr, next, 27420c95cba4SNicholas Piggin fn, data, create, mask); 2743c2febafcSKirill A. Shutemov if (err) 2744c2febafcSKirill A. Shutemov break; 2745c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 27460c95cba4SNicholas Piggin 2747c2febafcSKirill A. Shutemov return err; 2748c2febafcSKirill A. Shutemov } 2749c2febafcSKirill A. Shutemov 2750be1db475SDaniel Axtens static int __apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2751be1db475SDaniel Axtens unsigned long size, pte_fn_t fn, 2752be1db475SDaniel Axtens void *data, bool create) 2753be1db475SDaniel Axtens { 2754be1db475SDaniel Axtens pgd_t *pgd; 2755e80d3909SJoerg Roedel unsigned long start = addr, next; 2756be1db475SDaniel Axtens unsigned long end = addr + size; 2757e80d3909SJoerg Roedel pgtbl_mod_mask mask = 0; 2758be1db475SDaniel Axtens int err = 0; 2759be1db475SDaniel Axtens 2760be1db475SDaniel Axtens if (WARN_ON(addr >= end)) 2761be1db475SDaniel Axtens return -EINVAL; 2762be1db475SDaniel Axtens 2763be1db475SDaniel Axtens pgd = pgd_offset(mm, addr); 2764be1db475SDaniel Axtens do { 2765be1db475SDaniel Axtens next = pgd_addr_end(addr, end); 27660c95cba4SNicholas Piggin if (pgd_none(*pgd) && !create) 2767be1db475SDaniel Axtens continue; 27680c95cba4SNicholas Piggin if (WARN_ON_ONCE(pgd_leaf(*pgd))) 27690c95cba4SNicholas Piggin return -EINVAL; 27700c95cba4SNicholas Piggin if (!pgd_none(*pgd) && WARN_ON_ONCE(pgd_bad(*pgd))) { 27710c95cba4SNicholas Piggin if (!create) 27720c95cba4SNicholas Piggin continue; 27730c95cba4SNicholas Piggin pgd_clear_bad(pgd); 27740c95cba4SNicholas Piggin } 27750c95cba4SNicholas Piggin err = apply_to_p4d_range(mm, pgd, addr, next, 27760c95cba4SNicholas Piggin fn, data, create, &mask); 2777be1db475SDaniel Axtens if (err) 2778be1db475SDaniel Axtens break; 2779be1db475SDaniel Axtens } while (pgd++, addr = next, addr != end); 2780be1db475SDaniel Axtens 2781e80d3909SJoerg Roedel if (mask & ARCH_PAGE_TABLE_SYNC_MASK) 2782e80d3909SJoerg Roedel arch_sync_kernel_mappings(start, start + size); 2783e80d3909SJoerg Roedel 2784be1db475SDaniel Axtens return err; 2785be1db475SDaniel Axtens } 2786be1db475SDaniel Axtens 2787aee16b3cSJeremy Fitzhardinge /* 2788aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 2789aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 2790aee16b3cSJeremy Fitzhardinge */ 2791aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2792aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 2793aee16b3cSJeremy Fitzhardinge { 2794be1db475SDaniel Axtens return __apply_to_page_range(mm, addr, size, fn, data, true); 2795aee16b3cSJeremy Fitzhardinge } 2796aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 2797aee16b3cSJeremy Fitzhardinge 27981da177e4SLinus Torvalds /* 2799be1db475SDaniel Axtens * Scan a region of virtual memory, calling a provided function on 2800be1db475SDaniel Axtens * each leaf page table where it exists. 2801be1db475SDaniel Axtens * 2802be1db475SDaniel Axtens * Unlike apply_to_page_range, this does _not_ fill in page tables 2803be1db475SDaniel Axtens * where they are absent. 2804be1db475SDaniel Axtens */ 2805be1db475SDaniel Axtens int apply_to_existing_page_range(struct mm_struct *mm, unsigned long addr, 2806be1db475SDaniel Axtens unsigned long size, pte_fn_t fn, void *data) 2807be1db475SDaniel Axtens { 2808be1db475SDaniel Axtens return __apply_to_page_range(mm, addr, size, fn, data, false); 2809be1db475SDaniel Axtens } 2810be1db475SDaniel Axtens EXPORT_SYMBOL_GPL(apply_to_existing_page_range); 2811be1db475SDaniel Axtens 2812be1db475SDaniel Axtens /* 28139b4bdd2fSKirill A. Shutemov * handle_pte_fault chooses page fault handler according to an entry which was 28149b4bdd2fSKirill A. Shutemov * read non-atomically. Before making any commitment, on those architectures 28159b4bdd2fSKirill A. Shutemov * or configurations (e.g. i386 with PAE) which might give a mix of unmatched 28169b4bdd2fSKirill A. Shutemov * parts, do_swap_page must check under lock before unmapping the pte and 28179b4bdd2fSKirill A. Shutemov * proceeding (but do_wp_page is only called after already making such a check; 2818a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 28198f4e2101SHugh Dickins */ 28202ca99358SPeter Xu static inline int pte_unmap_same(struct vm_fault *vmf) 28218f4e2101SHugh Dickins { 28228f4e2101SHugh Dickins int same = 1; 2823923717cbSThomas Gleixner #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPTION) 28248f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 2825c7ad0880SHugh Dickins spin_lock(vmf->ptl); 2826c33c7948SRyan Roberts same = pte_same(ptep_get(vmf->pte), vmf->orig_pte); 2827c7ad0880SHugh Dickins spin_unlock(vmf->ptl); 28288f4e2101SHugh Dickins } 28298f4e2101SHugh Dickins #endif 28302ca99358SPeter Xu pte_unmap(vmf->pte); 28312ca99358SPeter Xu vmf->pte = NULL; 28328f4e2101SHugh Dickins return same; 28338f4e2101SHugh Dickins } 28348f4e2101SHugh Dickins 2835a873dfe1STony Luck /* 2836a873dfe1STony Luck * Return: 2837a873dfe1STony Luck * 0: copied succeeded 2838a873dfe1STony Luck * -EHWPOISON: copy failed due to hwpoison in source page 2839a873dfe1STony Luck * -EAGAIN: copied failed (some other reason) 2840a873dfe1STony Luck */ 2841a873dfe1STony Luck static inline int __wp_page_copy_user(struct page *dst, struct page *src, 284283d116c5SJia He struct vm_fault *vmf) 28436aab341eSLinus Torvalds { 2844a873dfe1STony Luck int ret; 284583d116c5SJia He void *kaddr; 284683d116c5SJia He void __user *uaddr; 284783d116c5SJia He struct vm_area_struct *vma = vmf->vma; 284883d116c5SJia He struct mm_struct *mm = vma->vm_mm; 284983d116c5SJia He unsigned long addr = vmf->address; 285083d116c5SJia He 285183d116c5SJia He if (likely(src)) { 2852d302c239STony Luck if (copy_mc_user_highpage(dst, src, addr, vma)) { 2853d302c239STony Luck memory_failure_queue(page_to_pfn(src), 0); 2854a873dfe1STony Luck return -EHWPOISON; 2855d302c239STony Luck } 2856a873dfe1STony Luck return 0; 285783d116c5SJia He } 285883d116c5SJia He 28596aab341eSLinus Torvalds /* 28606aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 28616aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 28626aab341eSLinus Torvalds * just copying from the original user address. If that 28636aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 28646aab341eSLinus Torvalds */ 286524d2613aSFabio M. De Francesco kaddr = kmap_local_page(dst); 286624d2613aSFabio M. De Francesco pagefault_disable(); 286783d116c5SJia He uaddr = (void __user *)(addr & PAGE_MASK); 286883d116c5SJia He 286983d116c5SJia He /* 287083d116c5SJia He * On architectures with software "accessed" bits, we would 287183d116c5SJia He * take a double page fault, so mark it accessed here. 287283d116c5SJia He */ 28733db82b93SHugh Dickins vmf->pte = NULL; 2874e1fd09e3SYu Zhao if (!arch_has_hw_pte_young() && !pte_young(vmf->orig_pte)) { 287583d116c5SJia He pte_t entry; 287683d116c5SJia He 287783d116c5SJia He vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); 2878c33c7948SRyan Roberts if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 287983d116c5SJia He /* 288083d116c5SJia He * Other thread has already handled the fault 28817df67697SBibo Mao * and update local tlb only 288283d116c5SJia He */ 2883a92cbb82SHugh Dickins if (vmf->pte) 28847df67697SBibo Mao update_mmu_tlb(vma, addr, vmf->pte); 2885a873dfe1STony Luck ret = -EAGAIN; 288683d116c5SJia He goto pte_unlock; 288783d116c5SJia He } 288883d116c5SJia He 288983d116c5SJia He entry = pte_mkyoung(vmf->orig_pte); 289083d116c5SJia He if (ptep_set_access_flags(vma, addr, vmf->pte, entry, 0)) 28915003a2bdSMatthew Wilcox (Oracle) update_mmu_cache_range(vmf, vma, addr, vmf->pte, 1); 289283d116c5SJia He } 28935d2a2dbbSLinus Torvalds 28945d2a2dbbSLinus Torvalds /* 28955d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 28965d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 28975d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 28985d2a2dbbSLinus Torvalds * zeroes. 28995d2a2dbbSLinus Torvalds */ 290083d116c5SJia He if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { 29013db82b93SHugh Dickins if (vmf->pte) 2902c3e5ea6eSKirill A. Shutemov goto warn; 2903c3e5ea6eSKirill A. Shutemov 2904c3e5ea6eSKirill A. Shutemov /* Re-validate under PTL if the page is still mapped */ 2905c3e5ea6eSKirill A. Shutemov vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); 2906c33c7948SRyan Roberts if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 29077df67697SBibo Mao /* The PTE changed under us, update local tlb */ 2908a92cbb82SHugh Dickins if (vmf->pte) 29097df67697SBibo Mao update_mmu_tlb(vma, addr, vmf->pte); 2910a873dfe1STony Luck ret = -EAGAIN; 2911c3e5ea6eSKirill A. Shutemov goto pte_unlock; 2912c3e5ea6eSKirill A. Shutemov } 2913c3e5ea6eSKirill A. Shutemov 2914c3e5ea6eSKirill A. Shutemov /* 2915985ba004SEthon Paul * The same page can be mapped back since last copy attempt. 2916c3e5ea6eSKirill A. Shutemov * Try to copy again under PTL. 2917c3e5ea6eSKirill A. Shutemov */ 2918c3e5ea6eSKirill A. Shutemov if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { 291983d116c5SJia He /* 292083d116c5SJia He * Give a warn in case there can be some obscure 292183d116c5SJia He * use-case 292283d116c5SJia He */ 2923c3e5ea6eSKirill A. Shutemov warn: 292483d116c5SJia He WARN_ON_ONCE(1); 29253ecb01dfSJan Beulich clear_page(kaddr); 292683d116c5SJia He } 2927c3e5ea6eSKirill A. Shutemov } 292883d116c5SJia He 2929a873dfe1STony Luck ret = 0; 293083d116c5SJia He 293183d116c5SJia He pte_unlock: 29323db82b93SHugh Dickins if (vmf->pte) 293383d116c5SJia He pte_unmap_unlock(vmf->pte, vmf->ptl); 293424d2613aSFabio M. De Francesco pagefault_enable(); 293524d2613aSFabio M. De Francesco kunmap_local(kaddr); 2936c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 293783d116c5SJia He 293883d116c5SJia He return ret; 29396aab341eSLinus Torvalds } 29406aab341eSLinus Torvalds 2941c20cd45eSMichal Hocko static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) 2942c20cd45eSMichal Hocko { 2943c20cd45eSMichal Hocko struct file *vm_file = vma->vm_file; 2944c20cd45eSMichal Hocko 2945c20cd45eSMichal Hocko if (vm_file) 2946c20cd45eSMichal Hocko return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; 2947c20cd45eSMichal Hocko 2948c20cd45eSMichal Hocko /* 2949c20cd45eSMichal Hocko * Special mappings (e.g. VDSO) do not have any file so fake 2950c20cd45eSMichal Hocko * a default GFP_KERNEL for them. 2951c20cd45eSMichal Hocko */ 2952c20cd45eSMichal Hocko return GFP_KERNEL; 2953c20cd45eSMichal Hocko } 2954c20cd45eSMichal Hocko 29551da177e4SLinus Torvalds /* 2956fb09a464SKirill A. Shutemov * Notify the address space that the page is about to become writable so that 2957fb09a464SKirill A. Shutemov * it can prohibit this or wait for the page to get into an appropriate state. 2958fb09a464SKirill A. Shutemov * 2959fb09a464SKirill A. Shutemov * We do this without the lock held, so that it can sleep if it needs to. 2960fb09a464SKirill A. Shutemov */ 296186aa6998SSidhartha Kumar static vm_fault_t do_page_mkwrite(struct vm_fault *vmf, struct folio *folio) 2962fb09a464SKirill A. Shutemov { 29632b740303SSouptick Joarder vm_fault_t ret; 296438b8cb7fSJan Kara unsigned int old_flags = vmf->flags; 2965fb09a464SKirill A. Shutemov 296638b8cb7fSJan Kara vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 2967fb09a464SKirill A. Shutemov 2968dc617f29SDarrick J. Wong if (vmf->vma->vm_file && 2969dc617f29SDarrick J. Wong IS_SWAPFILE(vmf->vma->vm_file->f_mapping->host)) 2970dc617f29SDarrick J. Wong return VM_FAULT_SIGBUS; 2971dc617f29SDarrick J. Wong 297211bac800SDave Jiang ret = vmf->vma->vm_ops->page_mkwrite(vmf); 297338b8cb7fSJan Kara /* Restore original flags so that caller is not surprised */ 297438b8cb7fSJan Kara vmf->flags = old_flags; 2975fb09a464SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 2976fb09a464SKirill A. Shutemov return ret; 2977fb09a464SKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) { 29783d243659SSidhartha Kumar folio_lock(folio); 29793d243659SSidhartha Kumar if (!folio->mapping) { 29803d243659SSidhartha Kumar folio_unlock(folio); 2981fb09a464SKirill A. Shutemov return 0; /* retry */ 2982fb09a464SKirill A. Shutemov } 2983fb09a464SKirill A. Shutemov ret |= VM_FAULT_LOCKED; 2984fb09a464SKirill A. Shutemov } else 29853d243659SSidhartha Kumar VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); 2986fb09a464SKirill A. Shutemov return ret; 2987fb09a464SKirill A. Shutemov } 2988fb09a464SKirill A. Shutemov 2989fb09a464SKirill A. Shutemov /* 299097ba0c2bSJan Kara * Handle dirtying of a page in shared file mapping on a write fault. 29914e047f89SShachar Raindel * 299297ba0c2bSJan Kara * The function expects the page to be locked and unlocks it. 29934e047f89SShachar Raindel */ 299489b15332SJohannes Weiner static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf) 29954e047f89SShachar Raindel { 299689b15332SJohannes Weiner struct vm_area_struct *vma = vmf->vma; 29974e047f89SShachar Raindel struct address_space *mapping; 299815b4919aSZhangPeng struct folio *folio = page_folio(vmf->page); 299997ba0c2bSJan Kara bool dirtied; 300097ba0c2bSJan Kara bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; 30014e047f89SShachar Raindel 300215b4919aSZhangPeng dirtied = folio_mark_dirty(folio); 300315b4919aSZhangPeng VM_BUG_ON_FOLIO(folio_test_anon(folio), folio); 300497ba0c2bSJan Kara /* 300515b4919aSZhangPeng * Take a local copy of the address_space - folio.mapping may be zeroed 300615b4919aSZhangPeng * by truncate after folio_unlock(). The address_space itself remains 300715b4919aSZhangPeng * pinned by vma->vm_file's reference. We rely on folio_unlock()'s 300897ba0c2bSJan Kara * release semantics to prevent the compiler from undoing this copying. 300997ba0c2bSJan Kara */ 301015b4919aSZhangPeng mapping = folio_raw_mapping(folio); 301115b4919aSZhangPeng folio_unlock(folio); 30124e047f89SShachar Raindel 30134e047f89SShachar Raindel if (!page_mkwrite) 30144e047f89SShachar Raindel file_update_time(vma->vm_file); 301589b15332SJohannes Weiner 301689b15332SJohannes Weiner /* 301789b15332SJohannes Weiner * Throttle page dirtying rate down to writeback speed. 301889b15332SJohannes Weiner * 301989b15332SJohannes Weiner * mapping may be NULL here because some device drivers do not 302089b15332SJohannes Weiner * set page.mapping but still dirty their pages 302189b15332SJohannes Weiner * 3022c1e8d7c6SMichel Lespinasse * Drop the mmap_lock before waiting on IO, if we can. The file 302389b15332SJohannes Weiner * is pinning the mapping, as per above. 302489b15332SJohannes Weiner */ 302589b15332SJohannes Weiner if ((dirtied || page_mkwrite) && mapping) { 302689b15332SJohannes Weiner struct file *fpin; 302789b15332SJohannes Weiner 302889b15332SJohannes Weiner fpin = maybe_unlock_mmap_for_io(vmf, NULL); 302989b15332SJohannes Weiner balance_dirty_pages_ratelimited(mapping); 303089b15332SJohannes Weiner if (fpin) { 303189b15332SJohannes Weiner fput(fpin); 3032d9272525SPeter Xu return VM_FAULT_COMPLETED; 303389b15332SJohannes Weiner } 303489b15332SJohannes Weiner } 303589b15332SJohannes Weiner 303689b15332SJohannes Weiner return 0; 30374e047f89SShachar Raindel } 30384e047f89SShachar Raindel 303997ba0c2bSJan Kara /* 30404e047f89SShachar Raindel * Handle write page faults for pages that can be reused in the current vma 30414e047f89SShachar Raindel * 30424e047f89SShachar Raindel * This can happen either due to the mapping being with the VM_SHARED flag, 30434e047f89SShachar Raindel * or due to us being the last reference standing to the page. In either 30444e047f89SShachar Raindel * case, all we need to do here is to mark the page as writable and update 30454e047f89SShachar Raindel * any related book-keeping. 30464e047f89SShachar Raindel */ 3047a86bc96bSKefeng Wang static inline void wp_page_reuse(struct vm_fault *vmf, struct folio *folio) 304882b0f8c3SJan Kara __releases(vmf->ptl) 30494e047f89SShachar Raindel { 305082b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 30514e047f89SShachar Raindel pte_t entry; 30526c287605SDavid Hildenbrand 3053c89357e2SDavid Hildenbrand VM_BUG_ON(!(vmf->flags & FAULT_FLAG_WRITE)); 30546c287605SDavid Hildenbrand 3055c2c3b514SKefeng Wang if (folio) { 3056c2c3b514SKefeng Wang VM_BUG_ON(folio_test_anon(folio) && 3057c2c3b514SKefeng Wang !PageAnonExclusive(vmf->page)); 30584e047f89SShachar Raindel /* 3059c2c3b514SKefeng Wang * Clear the folio's cpupid information as the existing 30604e047f89SShachar Raindel * information potentially belongs to a now completely 30614e047f89SShachar Raindel * unrelated process. 30624e047f89SShachar Raindel */ 3063c2c3b514SKefeng Wang folio_xchg_last_cpupid(folio, (1 << LAST_CPUPID_SHIFT) - 1); 3064c2c3b514SKefeng Wang } 30654e047f89SShachar Raindel 30662994302bSJan Kara flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); 30672994302bSJan Kara entry = pte_mkyoung(vmf->orig_pte); 30684e047f89SShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 306982b0f8c3SJan Kara if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) 30705003a2bdSMatthew Wilcox (Oracle) update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1); 307182b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3072798a6b87SPeter Xu count_vm_event(PGREUSE); 30734e047f89SShachar Raindel } 30744e047f89SShachar Raindel 30754ed43798SMatthew Wilcox (Oracle) /* 30764ed43798SMatthew Wilcox (Oracle) * We could add a bitflag somewhere, but for now, we know that all 30774ed43798SMatthew Wilcox (Oracle) * vm_ops that have a ->map_pages have been audited and don't need 30784ed43798SMatthew Wilcox (Oracle) * the mmap_lock to be held. 30794ed43798SMatthew Wilcox (Oracle) */ 30804ed43798SMatthew Wilcox (Oracle) static inline vm_fault_t vmf_can_call_fault(const struct vm_fault *vmf) 30814ed43798SMatthew Wilcox (Oracle) { 30824ed43798SMatthew Wilcox (Oracle) struct vm_area_struct *vma = vmf->vma; 30834ed43798SMatthew Wilcox (Oracle) 30844ed43798SMatthew Wilcox (Oracle) if (vma->vm_ops->map_pages || !(vmf->flags & FAULT_FLAG_VMA_LOCK)) 30854ed43798SMatthew Wilcox (Oracle) return 0; 30864ed43798SMatthew Wilcox (Oracle) vma_end_read(vma); 30874ed43798SMatthew Wilcox (Oracle) return VM_FAULT_RETRY; 30884ed43798SMatthew Wilcox (Oracle) } 30894ed43798SMatthew Wilcox (Oracle) 3090164b06f2SMatthew Wilcox (Oracle) static vm_fault_t vmf_anon_prepare(struct vm_fault *vmf) 3091164b06f2SMatthew Wilcox (Oracle) { 3092164b06f2SMatthew Wilcox (Oracle) struct vm_area_struct *vma = vmf->vma; 3093164b06f2SMatthew Wilcox (Oracle) 3094164b06f2SMatthew Wilcox (Oracle) if (likely(vma->anon_vma)) 3095164b06f2SMatthew Wilcox (Oracle) return 0; 3096164b06f2SMatthew Wilcox (Oracle) if (vmf->flags & FAULT_FLAG_VMA_LOCK) { 3097164b06f2SMatthew Wilcox (Oracle) vma_end_read(vma); 3098164b06f2SMatthew Wilcox (Oracle) return VM_FAULT_RETRY; 3099164b06f2SMatthew Wilcox (Oracle) } 3100164b06f2SMatthew Wilcox (Oracle) if (__anon_vma_prepare(vma)) 3101164b06f2SMatthew Wilcox (Oracle) return VM_FAULT_OOM; 3102164b06f2SMatthew Wilcox (Oracle) return 0; 3103164b06f2SMatthew Wilcox (Oracle) } 3104164b06f2SMatthew Wilcox (Oracle) 31054e047f89SShachar Raindel /* 3106c89357e2SDavid Hildenbrand * Handle the case of a page which we actually need to copy to a new page, 3107c89357e2SDavid Hildenbrand * either due to COW or unsharing. 31082f38ab2cSShachar Raindel * 3109c1e8d7c6SMichel Lespinasse * Called with mmap_lock locked and the old page referenced, but 31102f38ab2cSShachar Raindel * without the ptl held. 31112f38ab2cSShachar Raindel * 31122f38ab2cSShachar Raindel * High level logic flow: 31132f38ab2cSShachar Raindel * 31142f38ab2cSShachar Raindel * - Allocate a page, copy the content of the old page to the new one. 31152f38ab2cSShachar Raindel * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. 31162f38ab2cSShachar Raindel * - Take the PTL. If the pte changed, bail out and release the allocated page 31172f38ab2cSShachar Raindel * - If the pte is still the way we remember it, update the page table and all 31182f38ab2cSShachar Raindel * relevant references. This includes dropping the reference the page-table 31192f38ab2cSShachar Raindel * held to the old page, as well as updating the rmap. 31202f38ab2cSShachar Raindel * - In any case, unlock the PTL and drop the reference we took to the old page. 31212f38ab2cSShachar Raindel */ 31222b740303SSouptick Joarder static vm_fault_t wp_page_copy(struct vm_fault *vmf) 31232f38ab2cSShachar Raindel { 3124c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 312582b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3126bae473a4SKirill A. Shutemov struct mm_struct *mm = vma->vm_mm; 312728d41a48SMatthew Wilcox (Oracle) struct folio *old_folio = NULL; 312828d41a48SMatthew Wilcox (Oracle) struct folio *new_folio = NULL; 31292f38ab2cSShachar Raindel pte_t entry; 31302f38ab2cSShachar Raindel int page_copied = 0; 3131ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 3132164b06f2SMatthew Wilcox (Oracle) vm_fault_t ret; 3133cf503cc6SKefeng Wang bool pfn_is_zero; 31342f38ab2cSShachar Raindel 3135662ce1dcSYang Yang delayacct_wpcopy_start(); 3136662ce1dcSYang Yang 313728d41a48SMatthew Wilcox (Oracle) if (vmf->page) 313828d41a48SMatthew Wilcox (Oracle) old_folio = page_folio(vmf->page); 3139164b06f2SMatthew Wilcox (Oracle) ret = vmf_anon_prepare(vmf); 3140164b06f2SMatthew Wilcox (Oracle) if (unlikely(ret)) 3141164b06f2SMatthew Wilcox (Oracle) goto out; 31422f38ab2cSShachar Raindel 3143cf503cc6SKefeng Wang pfn_is_zero = is_zero_pfn(pte_pfn(vmf->orig_pte)); 3144cf503cc6SKefeng Wang new_folio = folio_prealloc(mm, vma, vmf->address, pfn_is_zero); 31456bc56a4dSMatthew Wilcox (Oracle) if (!new_folio) 31462f38ab2cSShachar Raindel goto oom; 3147cf503cc6SKefeng Wang 3148cf503cc6SKefeng Wang if (!pfn_is_zero) { 3149164b06f2SMatthew Wilcox (Oracle) int err; 315083d116c5SJia He 3151164b06f2SMatthew Wilcox (Oracle) err = __wp_page_copy_user(&new_folio->page, vmf->page, vmf); 3152164b06f2SMatthew Wilcox (Oracle) if (err) { 315383d116c5SJia He /* 315483d116c5SJia He * COW failed, if the fault was solved by other, 315583d116c5SJia He * it's fine. If not, userspace would re-fault on 315683d116c5SJia He * the same address and we will handle the fault 315783d116c5SJia He * from the second attempt. 3158a873dfe1STony Luck * The -EHWPOISON case will not be retried. 315983d116c5SJia He */ 316028d41a48SMatthew Wilcox (Oracle) folio_put(new_folio); 316128d41a48SMatthew Wilcox (Oracle) if (old_folio) 316228d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 3163662ce1dcSYang Yang 3164662ce1dcSYang Yang delayacct_wpcopy_end(); 3165164b06f2SMatthew Wilcox (Oracle) return err == -EHWPOISON ? VM_FAULT_HWPOISON : 0; 316683d116c5SJia He } 316728d41a48SMatthew Wilcox (Oracle) kmsan_copy_page_meta(&new_folio->page, vmf->page); 31682f38ab2cSShachar Raindel } 31692f38ab2cSShachar Raindel 317028d41a48SMatthew Wilcox (Oracle) __folio_mark_uptodate(new_folio); 3171eb3c24f3SMel Gorman 31727d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, mm, 31736f4f13e8SJérôme Glisse vmf->address & PAGE_MASK, 3174ac46d4f3SJérôme Glisse (vmf->address & PAGE_MASK) + PAGE_SIZE); 3175ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 31762f38ab2cSShachar Raindel 31772f38ab2cSShachar Raindel /* 31782f38ab2cSShachar Raindel * Re-check the pte - we dropped the lock 31792f38ab2cSShachar Raindel */ 318082b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); 3181c33c7948SRyan Roberts if (likely(vmf->pte && pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 318228d41a48SMatthew Wilcox (Oracle) if (old_folio) { 318328d41a48SMatthew Wilcox (Oracle) if (!folio_test_anon(old_folio)) { 31846b27cc6cSKefeng Wang dec_mm_counter(mm, mm_counter_file(old_folio)); 3185f1a79412SShakeel Butt inc_mm_counter(mm, MM_ANONPAGES); 31862f38ab2cSShachar Raindel } 31872f38ab2cSShachar Raindel } else { 31886080d19fSxu xin ksm_might_unmap_zero_page(mm, vmf->orig_pte); 3189f1a79412SShakeel Butt inc_mm_counter(mm, MM_ANONPAGES); 31902f38ab2cSShachar Raindel } 31912994302bSJan Kara flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); 319228d41a48SMatthew Wilcox (Oracle) entry = mk_pte(&new_folio->page, vma->vm_page_prot); 319350c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 3194c89357e2SDavid Hildenbrand if (unlikely(unshare)) { 3195c89357e2SDavid Hildenbrand if (pte_soft_dirty(vmf->orig_pte)) 3196c89357e2SDavid Hildenbrand entry = pte_mksoft_dirty(entry); 3197c89357e2SDavid Hildenbrand if (pte_uffd_wp(vmf->orig_pte)) 3198c89357e2SDavid Hildenbrand entry = pte_mkuffd_wp(entry); 3199c89357e2SDavid Hildenbrand } else { 32002f38ab2cSShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 3201c89357e2SDavid Hildenbrand } 3202111fe718SNicholas Piggin 32032f38ab2cSShachar Raindel /* 32042f38ab2cSShachar Raindel * Clear the pte entry and flush it first, before updating the 3205111fe718SNicholas Piggin * pte with the new entry, to keep TLBs on different CPUs in 3206111fe718SNicholas Piggin * sync. This code used to set the new PTE then flush TLBs, but 3207111fe718SNicholas Piggin * that left a window where the new PTE could be loaded into 3208111fe718SNicholas Piggin * some TLBs while the old PTE remains in others. 32092f38ab2cSShachar Raindel */ 3210ec8832d0SAlistair Popple ptep_clear_flush(vma, vmf->address, vmf->pte); 321128d41a48SMatthew Wilcox (Oracle) folio_add_new_anon_rmap(new_folio, vma, vmf->address); 321228d41a48SMatthew Wilcox (Oracle) folio_add_lru_vma(new_folio, vma); 32132f38ab2cSShachar Raindel /* 32142f38ab2cSShachar Raindel * We call the notify macro here because, when using secondary 32152f38ab2cSShachar Raindel * mmu page tables (such as kvm shadow page tables), we want the 32162f38ab2cSShachar Raindel * new page to be mapped directly into the secondary page table. 32172f38ab2cSShachar Raindel */ 3218c89357e2SDavid Hildenbrand BUG_ON(unshare && pte_write(entry)); 321982b0f8c3SJan Kara set_pte_at_notify(mm, vmf->address, vmf->pte, entry); 32205003a2bdSMatthew Wilcox (Oracle) update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1); 322128d41a48SMatthew Wilcox (Oracle) if (old_folio) { 32222f38ab2cSShachar Raindel /* 32232f38ab2cSShachar Raindel * Only after switching the pte to the new page may 32242f38ab2cSShachar Raindel * we remove the mapcount here. Otherwise another 32252f38ab2cSShachar Raindel * process may come and find the rmap count decremented 32262f38ab2cSShachar Raindel * before the pte is switched to the new page, and 32272f38ab2cSShachar Raindel * "reuse" the old page writing into it while our pte 32282f38ab2cSShachar Raindel * here still points into it and can be read by other 32292f38ab2cSShachar Raindel * threads. 32302f38ab2cSShachar Raindel * 32312f38ab2cSShachar Raindel * The critical issue is to order this 3232c4626503SDavid Hildenbrand * folio_remove_rmap_pte() with the ptp_clear_flush 3233c4626503SDavid Hildenbrand * above. Those stores are ordered by (if nothing else,) 32342f38ab2cSShachar Raindel * the barrier present in the atomic_add_negative 3235c4626503SDavid Hildenbrand * in folio_remove_rmap_pte(); 32362f38ab2cSShachar Raindel * 32372f38ab2cSShachar Raindel * Then the TLB flush in ptep_clear_flush ensures that 32382f38ab2cSShachar Raindel * no process can access the old page before the 32392f38ab2cSShachar Raindel * decremented mapcount is visible. And the old page 32402f38ab2cSShachar Raindel * cannot be reused until after the decremented 32412f38ab2cSShachar Raindel * mapcount is visible. So transitively, TLBs to 32422f38ab2cSShachar Raindel * old page will be flushed before it can be reused. 32432f38ab2cSShachar Raindel */ 3244c4626503SDavid Hildenbrand folio_remove_rmap_pte(old_folio, vmf->page, vma); 32452f38ab2cSShachar Raindel } 32462f38ab2cSShachar Raindel 32472f38ab2cSShachar Raindel /* Free the old page.. */ 324828d41a48SMatthew Wilcox (Oracle) new_folio = old_folio; 32492f38ab2cSShachar Raindel page_copied = 1; 32503db82b93SHugh Dickins pte_unmap_unlock(vmf->pte, vmf->ptl); 32513db82b93SHugh Dickins } else if (vmf->pte) { 32527df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 32533db82b93SHugh Dickins pte_unmap_unlock(vmf->pte, vmf->ptl); 32542f38ab2cSShachar Raindel } 32552f38ab2cSShachar Raindel 3256ec8832d0SAlistair Popple mmu_notifier_invalidate_range_end(&range); 32573db82b93SHugh Dickins 32583db82b93SHugh Dickins if (new_folio) 32593db82b93SHugh Dickins folio_put(new_folio); 326028d41a48SMatthew Wilcox (Oracle) if (old_folio) { 3261f4c4a3f4SHuang Ying if (page_copied) 326228d41a48SMatthew Wilcox (Oracle) free_swap_cache(&old_folio->page); 326328d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 32642f38ab2cSShachar Raindel } 3265662ce1dcSYang Yang 3266662ce1dcSYang Yang delayacct_wpcopy_end(); 3267cb8d8633SDavid Hildenbrand return 0; 32682f38ab2cSShachar Raindel oom: 3269164b06f2SMatthew Wilcox (Oracle) ret = VM_FAULT_OOM; 3270164b06f2SMatthew Wilcox (Oracle) out: 327128d41a48SMatthew Wilcox (Oracle) if (old_folio) 327228d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 3273662ce1dcSYang Yang 3274662ce1dcSYang Yang delayacct_wpcopy_end(); 3275164b06f2SMatthew Wilcox (Oracle) return ret; 32762f38ab2cSShachar Raindel } 32772f38ab2cSShachar Raindel 327866a6197cSJan Kara /** 327966a6197cSJan Kara * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE 328066a6197cSJan Kara * writeable once the page is prepared 328166a6197cSJan Kara * 328266a6197cSJan Kara * @vmf: structure describing the fault 3283a86bc96bSKefeng Wang * @folio: the folio of vmf->page 328466a6197cSJan Kara * 328566a6197cSJan Kara * This function handles all that is needed to finish a write page fault in a 328666a6197cSJan Kara * shared mapping due to PTE being read-only once the mapped page is prepared. 3287a862f68aSMike Rapoport * It handles locking of PTE and modifying it. 328866a6197cSJan Kara * 328966a6197cSJan Kara * The function expects the page to be locked or other protection against 329066a6197cSJan Kara * concurrent faults / writeback (such as DAX radix tree locks). 3291a862f68aSMike Rapoport * 32922797e79fSLiu Xiang * Return: %0 on success, %VM_FAULT_NOPAGE when PTE got changed before 3293a862f68aSMike Rapoport * we acquired PTE lock. 329466a6197cSJan Kara */ 3295a86bc96bSKefeng Wang static vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf, struct folio *folio) 329666a6197cSJan Kara { 329766a6197cSJan Kara WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); 329866a6197cSJan Kara vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, 329966a6197cSJan Kara &vmf->ptl); 33003db82b93SHugh Dickins if (!vmf->pte) 33013db82b93SHugh Dickins return VM_FAULT_NOPAGE; 330266a6197cSJan Kara /* 330366a6197cSJan Kara * We might have raced with another page fault while we released the 330466a6197cSJan Kara * pte_offset_map_lock. 330566a6197cSJan Kara */ 3306c33c7948SRyan Roberts if (!pte_same(ptep_get(vmf->pte), vmf->orig_pte)) { 33077df67697SBibo Mao update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 330866a6197cSJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3309a19e2553SJan Kara return VM_FAULT_NOPAGE; 331066a6197cSJan Kara } 3311a86bc96bSKefeng Wang wp_page_reuse(vmf, folio); 3312a19e2553SJan Kara return 0; 331366a6197cSJan Kara } 331466a6197cSJan Kara 3315dd906184SBoaz Harrosh /* 3316dd906184SBoaz Harrosh * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED 3317dd906184SBoaz Harrosh * mapping 3318dd906184SBoaz Harrosh */ 33192b740303SSouptick Joarder static vm_fault_t wp_pfn_shared(struct vm_fault *vmf) 3320dd906184SBoaz Harrosh { 332182b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3322bae473a4SKirill A. Shutemov 3323dd906184SBoaz Harrosh if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { 33242b740303SSouptick Joarder vm_fault_t ret; 3325dd906184SBoaz Harrosh 332682b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 33274a68fef1SMatthew Wilcox (Oracle) ret = vmf_can_call_fault(vmf); 33284a68fef1SMatthew Wilcox (Oracle) if (ret) 33294a68fef1SMatthew Wilcox (Oracle) return ret; 3330063e60d8SMatthew Wilcox (Oracle) 3331fe82221fSJan Kara vmf->flags |= FAULT_FLAG_MKWRITE; 333211bac800SDave Jiang ret = vma->vm_ops->pfn_mkwrite(vmf); 33332f89dc12SJan Kara if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) 3334dd906184SBoaz Harrosh return ret; 3335a86bc96bSKefeng Wang return finish_mkwrite_fault(vmf, NULL); 3336dd906184SBoaz Harrosh } 3337a86bc96bSKefeng Wang wp_page_reuse(vmf, NULL); 3338cb8d8633SDavid Hildenbrand return 0; 3339dd906184SBoaz Harrosh } 3340dd906184SBoaz Harrosh 33415a97858bSSidhartha Kumar static vm_fault_t wp_page_shared(struct vm_fault *vmf, struct folio *folio) 334282b0f8c3SJan Kara __releases(vmf->ptl) 334393e478d4SShachar Raindel { 334482b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3345cb8d8633SDavid Hildenbrand vm_fault_t ret = 0; 334693e478d4SShachar Raindel 33475a97858bSSidhartha Kumar folio_get(folio); 334893e478d4SShachar Raindel 334993e478d4SShachar Raindel if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 33502b740303SSouptick Joarder vm_fault_t tmp; 335193e478d4SShachar Raindel 335282b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 33534a68fef1SMatthew Wilcox (Oracle) tmp = vmf_can_call_fault(vmf); 33544a68fef1SMatthew Wilcox (Oracle) if (tmp) { 3355063e60d8SMatthew Wilcox (Oracle) folio_put(folio); 33564a68fef1SMatthew Wilcox (Oracle) return tmp; 3357063e60d8SMatthew Wilcox (Oracle) } 3358063e60d8SMatthew Wilcox (Oracle) 335986aa6998SSidhartha Kumar tmp = do_page_mkwrite(vmf, folio); 336093e478d4SShachar Raindel if (unlikely(!tmp || (tmp & 336193e478d4SShachar Raindel (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 33625a97858bSSidhartha Kumar folio_put(folio); 336393e478d4SShachar Raindel return tmp; 336493e478d4SShachar Raindel } 3365a86bc96bSKefeng Wang tmp = finish_mkwrite_fault(vmf, folio); 3366a19e2553SJan Kara if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { 33675a97858bSSidhartha Kumar folio_unlock(folio); 33685a97858bSSidhartha Kumar folio_put(folio); 336966a6197cSJan Kara return tmp; 337093e478d4SShachar Raindel } 337166a6197cSJan Kara } else { 3372a86bc96bSKefeng Wang wp_page_reuse(vmf, folio); 33735a97858bSSidhartha Kumar folio_lock(folio); 337493e478d4SShachar Raindel } 337589b15332SJohannes Weiner ret |= fault_dirty_shared_page(vmf); 33765a97858bSSidhartha Kumar folio_put(folio); 337793e478d4SShachar Raindel 337889b15332SJohannes Weiner return ret; 337993e478d4SShachar Raindel } 338093e478d4SShachar Raindel 3381dec078ccSDavid Hildenbrand static bool wp_can_reuse_anon_folio(struct folio *folio, 3382dec078ccSDavid Hildenbrand struct vm_area_struct *vma) 3383dec078ccSDavid Hildenbrand { 3384dec078ccSDavid Hildenbrand /* 3385dec078ccSDavid Hildenbrand * We have to verify under folio lock: these early checks are 3386dec078ccSDavid Hildenbrand * just an optimization to avoid locking the folio and freeing 3387dec078ccSDavid Hildenbrand * the swapcache if there is little hope that we can reuse. 3388dec078ccSDavid Hildenbrand * 3389dec078ccSDavid Hildenbrand * KSM doesn't necessarily raise the folio refcount. 3390dec078ccSDavid Hildenbrand */ 3391dec078ccSDavid Hildenbrand if (folio_test_ksm(folio) || folio_ref_count(folio) > 3) 3392dec078ccSDavid Hildenbrand return false; 3393dec078ccSDavid Hildenbrand if (!folio_test_lru(folio)) 3394dec078ccSDavid Hildenbrand /* 3395dec078ccSDavid Hildenbrand * We cannot easily detect+handle references from 3396dec078ccSDavid Hildenbrand * remote LRU caches or references to LRU folios. 3397dec078ccSDavid Hildenbrand */ 3398dec078ccSDavid Hildenbrand lru_add_drain(); 3399dec078ccSDavid Hildenbrand if (folio_ref_count(folio) > 1 + folio_test_swapcache(folio)) 3400dec078ccSDavid Hildenbrand return false; 3401dec078ccSDavid Hildenbrand if (!folio_trylock(folio)) 3402dec078ccSDavid Hildenbrand return false; 3403dec078ccSDavid Hildenbrand if (folio_test_swapcache(folio)) 3404dec078ccSDavid Hildenbrand folio_free_swap(folio); 3405dec078ccSDavid Hildenbrand if (folio_test_ksm(folio) || folio_ref_count(folio) != 1) { 3406dec078ccSDavid Hildenbrand folio_unlock(folio); 3407dec078ccSDavid Hildenbrand return false; 3408dec078ccSDavid Hildenbrand } 3409dec078ccSDavid Hildenbrand /* 3410dec078ccSDavid Hildenbrand * Ok, we've got the only folio reference from our mapping 3411dec078ccSDavid Hildenbrand * and the folio is locked, it's dark out, and we're wearing 3412dec078ccSDavid Hildenbrand * sunglasses. Hit it. 3413dec078ccSDavid Hildenbrand */ 3414dec078ccSDavid Hildenbrand folio_move_anon_rmap(folio, vma); 3415dec078ccSDavid Hildenbrand folio_unlock(folio); 3416dec078ccSDavid Hildenbrand return true; 3417dec078ccSDavid Hildenbrand } 3418dec078ccSDavid Hildenbrand 34192f38ab2cSShachar Raindel /* 3420c89357e2SDavid Hildenbrand * This routine handles present pages, when 3421c89357e2SDavid Hildenbrand * * users try to write to a shared page (FAULT_FLAG_WRITE) 3422c89357e2SDavid Hildenbrand * * GUP wants to take a R/O pin on a possibly shared anonymous page 3423c89357e2SDavid Hildenbrand * (FAULT_FLAG_UNSHARE) 3424c89357e2SDavid Hildenbrand * 3425c89357e2SDavid Hildenbrand * It is done by copying the page to a new address and decrementing the 3426c89357e2SDavid Hildenbrand * shared-page counter for the old page. 34271da177e4SLinus Torvalds * 34281da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 34291da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 3430c89357e2SDavid Hildenbrand * Thus, with FAULT_FLAG_WRITE, we can safely just mark it writable once we've 3431c89357e2SDavid Hildenbrand * done any necessary COW. 34321da177e4SLinus Torvalds * 3433c89357e2SDavid Hildenbrand * In case of FAULT_FLAG_WRITE, we also mark the page dirty at this point even 3434c89357e2SDavid Hildenbrand * though the page will change only once the write actually happens. This 3435c89357e2SDavid Hildenbrand * avoids a few races, and potentially makes it more efficient. 34361da177e4SLinus Torvalds * 3437c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 34388f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 3439c1e8d7c6SMichel Lespinasse * We return with mmap_lock still held, but pte unmapped and unlocked. 34401da177e4SLinus Torvalds */ 34412b740303SSouptick Joarder static vm_fault_t do_wp_page(struct vm_fault *vmf) 344282b0f8c3SJan Kara __releases(vmf->ptl) 34431da177e4SLinus Torvalds { 3444c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 344582b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3446b9086fdeSDavid Hildenbrand struct folio *folio = NULL; 3447d61ea1cbSPeter Xu pte_t pte; 34481da177e4SLinus Torvalds 3449c89357e2SDavid Hildenbrand if (likely(!unshare)) { 3450c33c7948SRyan Roberts if (userfaultfd_pte_wp(vma, ptep_get(vmf->pte))) { 3451d61ea1cbSPeter Xu if (!userfaultfd_wp_async(vma)) { 3452529b930bSAndrea Arcangeli pte_unmap_unlock(vmf->pte, vmf->ptl); 3453529b930bSAndrea Arcangeli return handle_userfault(vmf, VM_UFFD_WP); 3454529b930bSAndrea Arcangeli } 3455529b930bSAndrea Arcangeli 34566ce64428SNadav Amit /* 3457d61ea1cbSPeter Xu * Nothing needed (cache flush, TLB invalidations, 3458d61ea1cbSPeter Xu * etc.) because we're only removing the uffd-wp bit, 3459d61ea1cbSPeter Xu * which is completely invisible to the user. 3460d61ea1cbSPeter Xu */ 3461d61ea1cbSPeter Xu pte = pte_clear_uffd_wp(ptep_get(vmf->pte)); 3462d61ea1cbSPeter Xu 3463d61ea1cbSPeter Xu set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); 3464d61ea1cbSPeter Xu /* 3465d61ea1cbSPeter Xu * Update this to be prepared for following up CoW 3466d61ea1cbSPeter Xu * handling 3467d61ea1cbSPeter Xu */ 3468d61ea1cbSPeter Xu vmf->orig_pte = pte; 3469d61ea1cbSPeter Xu } 3470d61ea1cbSPeter Xu 3471d61ea1cbSPeter Xu /* 34726ce64428SNadav Amit * Userfaultfd write-protect can defer flushes. Ensure the TLB 34736ce64428SNadav Amit * is flushed in this case before copying. 34746ce64428SNadav Amit */ 34756ce64428SNadav Amit if (unlikely(userfaultfd_wp(vmf->vma) && 34766ce64428SNadav Amit mm_tlb_flush_pending(vmf->vma->vm_mm))) 34776ce64428SNadav Amit flush_tlb_page(vmf->vma, vmf->address); 3478c89357e2SDavid Hildenbrand } 34796ce64428SNadav Amit 3480a41b70d6SJan Kara vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); 3481c89357e2SDavid Hildenbrand 34825a97858bSSidhartha Kumar if (vmf->page) 34835a97858bSSidhartha Kumar folio = page_folio(vmf->page); 34845a97858bSSidhartha Kumar 3485251b97f5SPeter Zijlstra /* 3486b9086fdeSDavid Hildenbrand * Shared mapping: we are guaranteed to have VM_WRITE and 3487b9086fdeSDavid Hildenbrand * FAULT_FLAG_WRITE set at this point. 3488b9086fdeSDavid Hildenbrand */ 3489b9086fdeSDavid Hildenbrand if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 3490b9086fdeSDavid Hildenbrand /* 349164e45507SPeter Feiner * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a 349264e45507SPeter Feiner * VM_PFNMAP VMA. 3493251b97f5SPeter Zijlstra * 3494251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 3495dd906184SBoaz Harrosh * Just mark the pages writable and/or call ops->pfn_mkwrite. 3496251b97f5SPeter Zijlstra */ 3497b9086fdeSDavid Hildenbrand if (!vmf->page) 34982994302bSJan Kara return wp_pfn_shared(vmf); 34995a97858bSSidhartha Kumar return wp_page_shared(vmf, folio); 3500251b97f5SPeter Zijlstra } 35011da177e4SLinus Torvalds 3502b9086fdeSDavid Hildenbrand /* 3503b9086fdeSDavid Hildenbrand * Private mapping: create an exclusive anonymous page copy if reuse 3504b9086fdeSDavid Hildenbrand * is impossible. We might miss VM_WRITE for FOLL_FORCE handling. 350553a05ad9SDavid Hildenbrand * 3506dec078ccSDavid Hildenbrand * If we encounter a page that is marked exclusive, we must reuse 3507dec078ccSDavid Hildenbrand * the page without further checks. 350853a05ad9SDavid Hildenbrand */ 3509dec078ccSDavid Hildenbrand if (folio && folio_test_anon(folio) && 3510dec078ccSDavid Hildenbrand (PageAnonExclusive(vmf->page) || wp_can_reuse_anon_folio(folio, vma))) { 3511dec078ccSDavid Hildenbrand if (!PageAnonExclusive(vmf->page)) 35125ca43289SDavid Hildenbrand SetPageAnonExclusive(vmf->page); 3513c89357e2SDavid Hildenbrand if (unlikely(unshare)) { 3514c89357e2SDavid Hildenbrand pte_unmap_unlock(vmf->pte, vmf->ptl); 3515c89357e2SDavid Hildenbrand return 0; 3516c89357e2SDavid Hildenbrand } 3517a86bc96bSKefeng Wang wp_page_reuse(vmf, folio); 3518cb8d8633SDavid Hildenbrand return 0; 35191da177e4SLinus Torvalds } 35201da177e4SLinus Torvalds /* 35211da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 35221da177e4SLinus Torvalds */ 3523b9086fdeSDavid Hildenbrand if (folio) 3524b9086fdeSDavid Hildenbrand folio_get(folio); 352528766805SShachar Raindel 352682b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 352794bfe85bSYang Yang #ifdef CONFIG_KSM 3528b9086fdeSDavid Hildenbrand if (folio && folio_test_ksm(folio)) 352994bfe85bSYang Yang count_vm_event(COW_KSM); 353094bfe85bSYang Yang #endif 3531a41b70d6SJan Kara return wp_page_copy(vmf); 35321da177e4SLinus Torvalds } 35331da177e4SLinus Torvalds 353497a89413SPeter Zijlstra static void unmap_mapping_range_vma(struct vm_area_struct *vma, 35351da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 35361da177e4SLinus Torvalds struct zap_details *details) 35371da177e4SLinus Torvalds { 3538f5cc4eefSAl Viro zap_page_range_single(vma, start_addr, end_addr - start_addr, details); 35391da177e4SLinus Torvalds } 35401da177e4SLinus Torvalds 3541f808c13fSDavidlohr Bueso static inline void unmap_mapping_range_tree(struct rb_root_cached *root, 3542232a6a1cSPeter Xu pgoff_t first_index, 3543232a6a1cSPeter Xu pgoff_t last_index, 35441da177e4SLinus Torvalds struct zap_details *details) 35451da177e4SLinus Torvalds { 35461da177e4SLinus Torvalds struct vm_area_struct *vma; 35471da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 35481da177e4SLinus Torvalds 3549232a6a1cSPeter Xu vma_interval_tree_foreach(vma, root, first_index, last_index) { 35501da177e4SLinus Torvalds vba = vma->vm_pgoff; 3551d6e93217SLibin vea = vba + vma_pages(vma) - 1; 3552f9871da9SMiaohe Lin zba = max(first_index, vba); 3553f9871da9SMiaohe Lin zea = min(last_index, vea); 35541da177e4SLinus Torvalds 355597a89413SPeter Zijlstra unmap_mapping_range_vma(vma, 35561da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 35571da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 355897a89413SPeter Zijlstra details); 35591da177e4SLinus Torvalds } 35601da177e4SLinus Torvalds } 35611da177e4SLinus Torvalds 35621da177e4SLinus Torvalds /** 35633506659eSMatthew Wilcox (Oracle) * unmap_mapping_folio() - Unmap single folio from processes. 35643506659eSMatthew Wilcox (Oracle) * @folio: The locked folio to be unmapped. 356522061a1fSHugh Dickins * 35663506659eSMatthew Wilcox (Oracle) * Unmap this folio from any userspace process which still has it mmaped. 356722061a1fSHugh Dickins * Typically, for efficiency, the range of nearby pages has already been 356822061a1fSHugh Dickins * unmapped by unmap_mapping_pages() or unmap_mapping_range(). But once 35693506659eSMatthew Wilcox (Oracle) * truncation or invalidation holds the lock on a folio, it may find that 35703506659eSMatthew Wilcox (Oracle) * the page has been remapped again: and then uses unmap_mapping_folio() 357122061a1fSHugh Dickins * to unmap it finally. 357222061a1fSHugh Dickins */ 35733506659eSMatthew Wilcox (Oracle) void unmap_mapping_folio(struct folio *folio) 357422061a1fSHugh Dickins { 35753506659eSMatthew Wilcox (Oracle) struct address_space *mapping = folio->mapping; 357622061a1fSHugh Dickins struct zap_details details = { }; 3577232a6a1cSPeter Xu pgoff_t first_index; 3578232a6a1cSPeter Xu pgoff_t last_index; 357922061a1fSHugh Dickins 35803506659eSMatthew Wilcox (Oracle) VM_BUG_ON(!folio_test_locked(folio)); 358122061a1fSHugh Dickins 35823506659eSMatthew Wilcox (Oracle) first_index = folio->index; 358387b11f86SSidhartha Kumar last_index = folio_next_index(folio) - 1; 3584232a6a1cSPeter Xu 35852e148f1eSPeter Xu details.even_cows = false; 35863506659eSMatthew Wilcox (Oracle) details.single_folio = folio; 3587999dad82SPeter Xu details.zap_flags = ZAP_FLAG_DROP_MARKER; 358822061a1fSHugh Dickins 35892c865995SHugh Dickins i_mmap_lock_read(mapping); 359022061a1fSHugh Dickins if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) 3591232a6a1cSPeter Xu unmap_mapping_range_tree(&mapping->i_mmap, first_index, 3592232a6a1cSPeter Xu last_index, &details); 35932c865995SHugh Dickins i_mmap_unlock_read(mapping); 359422061a1fSHugh Dickins } 359522061a1fSHugh Dickins 359622061a1fSHugh Dickins /** 3597977fbdcdSMatthew Wilcox * unmap_mapping_pages() - Unmap pages from processes. 3598977fbdcdSMatthew Wilcox * @mapping: The address space containing pages to be unmapped. 3599977fbdcdSMatthew Wilcox * @start: Index of first page to be unmapped. 3600977fbdcdSMatthew Wilcox * @nr: Number of pages to be unmapped. 0 to unmap to end of file. 3601977fbdcdSMatthew Wilcox * @even_cows: Whether to unmap even private COWed pages. 3602977fbdcdSMatthew Wilcox * 3603977fbdcdSMatthew Wilcox * Unmap the pages in this address space from any userspace process which 3604977fbdcdSMatthew Wilcox * has them mmaped. Generally, you want to remove COWed pages as well when 3605977fbdcdSMatthew Wilcox * a file is being truncated, but not when invalidating pages from the page 3606977fbdcdSMatthew Wilcox * cache. 3607977fbdcdSMatthew Wilcox */ 3608977fbdcdSMatthew Wilcox void unmap_mapping_pages(struct address_space *mapping, pgoff_t start, 3609977fbdcdSMatthew Wilcox pgoff_t nr, bool even_cows) 3610977fbdcdSMatthew Wilcox { 3611977fbdcdSMatthew Wilcox struct zap_details details = { }; 3612232a6a1cSPeter Xu pgoff_t first_index = start; 3613232a6a1cSPeter Xu pgoff_t last_index = start + nr - 1; 3614977fbdcdSMatthew Wilcox 36152e148f1eSPeter Xu details.even_cows = even_cows; 3616232a6a1cSPeter Xu if (last_index < first_index) 3617232a6a1cSPeter Xu last_index = ULONG_MAX; 3618977fbdcdSMatthew Wilcox 36192c865995SHugh Dickins i_mmap_lock_read(mapping); 3620977fbdcdSMatthew Wilcox if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) 3621232a6a1cSPeter Xu unmap_mapping_range_tree(&mapping->i_mmap, first_index, 3622232a6a1cSPeter Xu last_index, &details); 36232c865995SHugh Dickins i_mmap_unlock_read(mapping); 3624977fbdcdSMatthew Wilcox } 36256e0e99d5SDavid Howells EXPORT_SYMBOL_GPL(unmap_mapping_pages); 3626977fbdcdSMatthew Wilcox 3627977fbdcdSMatthew Wilcox /** 36288a5f14a2SKirill A. Shutemov * unmap_mapping_range - unmap the portion of all mmaps in the specified 3629977fbdcdSMatthew Wilcox * address_space corresponding to the specified byte range in the underlying 36308a5f14a2SKirill A. Shutemov * file. 36318a5f14a2SKirill A. Shutemov * 36323d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 36331da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 36341da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 363525d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 36361da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 36371da177e4SLinus Torvalds * partial pages. 36381da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 36391da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 36401da177e4SLinus Torvalds * end of the file. 36411da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 36421da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 36431da177e4SLinus Torvalds */ 36441da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 36451da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 36461da177e4SLinus Torvalds { 36479eab0421SJiajun Xie pgoff_t hba = (pgoff_t)(holebegin) >> PAGE_SHIFT; 36489eab0421SJiajun Xie pgoff_t hlen = ((pgoff_t)(holelen) + PAGE_SIZE - 1) >> PAGE_SHIFT; 36491da177e4SLinus Torvalds 36501da177e4SLinus Torvalds /* Check for overflow. */ 36511da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 36521da177e4SLinus Torvalds long long holeend = 36531da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 36541da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 36551da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 36561da177e4SLinus Torvalds } 36571da177e4SLinus Torvalds 3658977fbdcdSMatthew Wilcox unmap_mapping_pages(mapping, hba, hlen, even_cows); 36591da177e4SLinus Torvalds } 36601da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 36611da177e4SLinus Torvalds 36621da177e4SLinus Torvalds /* 3663b756a3b5SAlistair Popple * Restore a potential device exclusive pte to a working pte entry 3664b756a3b5SAlistair Popple */ 3665b756a3b5SAlistair Popple static vm_fault_t remove_device_exclusive_entry(struct vm_fault *vmf) 3666b756a3b5SAlistair Popple { 366719672a9eSMatthew Wilcox (Oracle) struct folio *folio = page_folio(vmf->page); 3668b756a3b5SAlistair Popple struct vm_area_struct *vma = vmf->vma; 3669b756a3b5SAlistair Popple struct mmu_notifier_range range; 3670fdc724d6SSuren Baghdasaryan vm_fault_t ret; 3671b756a3b5SAlistair Popple 36727c7b9629SAlistair Popple /* 36737c7b9629SAlistair Popple * We need a reference to lock the folio because we don't hold 36747c7b9629SAlistair Popple * the PTL so a racing thread can remove the device-exclusive 36757c7b9629SAlistair Popple * entry and unmap it. If the folio is free the entry must 36767c7b9629SAlistair Popple * have been removed already. If it happens to have already 36777c7b9629SAlistair Popple * been re-allocated after being freed all we do is lock and 36787c7b9629SAlistair Popple * unlock it. 36797c7b9629SAlistair Popple */ 36807c7b9629SAlistair Popple if (!folio_try_get(folio)) 36817c7b9629SAlistair Popple return 0; 36827c7b9629SAlistair Popple 3683fdc724d6SSuren Baghdasaryan ret = folio_lock_or_retry(folio, vmf); 3684fdc724d6SSuren Baghdasaryan if (ret) { 36857c7b9629SAlistair Popple folio_put(folio); 3686fdc724d6SSuren Baghdasaryan return ret; 36877c7b9629SAlistair Popple } 36887d4a8be0SAlistair Popple mmu_notifier_range_init_owner(&range, MMU_NOTIFY_EXCLUSIVE, 0, 3689b756a3b5SAlistair Popple vma->vm_mm, vmf->address & PAGE_MASK, 3690b756a3b5SAlistair Popple (vmf->address & PAGE_MASK) + PAGE_SIZE, NULL); 3691b756a3b5SAlistair Popple mmu_notifier_invalidate_range_start(&range); 3692b756a3b5SAlistair Popple 3693b756a3b5SAlistair Popple vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 3694b756a3b5SAlistair Popple &vmf->ptl); 3695c33c7948SRyan Roberts if (likely(vmf->pte && pte_same(ptep_get(vmf->pte), vmf->orig_pte))) 369619672a9eSMatthew Wilcox (Oracle) restore_exclusive_pte(vma, vmf->page, vmf->address, vmf->pte); 3697b756a3b5SAlistair Popple 36983db82b93SHugh Dickins if (vmf->pte) 3699b756a3b5SAlistair Popple pte_unmap_unlock(vmf->pte, vmf->ptl); 370019672a9eSMatthew Wilcox (Oracle) folio_unlock(folio); 37017c7b9629SAlistair Popple folio_put(folio); 3702b756a3b5SAlistair Popple 3703b756a3b5SAlistair Popple mmu_notifier_invalidate_range_end(&range); 3704b756a3b5SAlistair Popple return 0; 3705b756a3b5SAlistair Popple } 3706b756a3b5SAlistair Popple 3707a160e537SMatthew Wilcox (Oracle) static inline bool should_try_to_free_swap(struct folio *folio, 3708c145e0b4SDavid Hildenbrand struct vm_area_struct *vma, 3709c145e0b4SDavid Hildenbrand unsigned int fault_flags) 3710c145e0b4SDavid Hildenbrand { 3711a160e537SMatthew Wilcox (Oracle) if (!folio_test_swapcache(folio)) 3712c145e0b4SDavid Hildenbrand return false; 37139202d527SMatthew Wilcox (Oracle) if (mem_cgroup_swap_full(folio) || (vma->vm_flags & VM_LOCKED) || 3714a160e537SMatthew Wilcox (Oracle) folio_test_mlocked(folio)) 3715c145e0b4SDavid Hildenbrand return true; 3716c145e0b4SDavid Hildenbrand /* 3717c145e0b4SDavid Hildenbrand * If we want to map a page that's in the swapcache writable, we 3718c145e0b4SDavid Hildenbrand * have to detect via the refcount if we're really the exclusive 3719c145e0b4SDavid Hildenbrand * user. Try freeing the swapcache to get rid of the swapcache 3720c145e0b4SDavid Hildenbrand * reference only in case it's likely that we'll be the exlusive user. 3721c145e0b4SDavid Hildenbrand */ 3722a160e537SMatthew Wilcox (Oracle) return (fault_flags & FAULT_FLAG_WRITE) && !folio_test_ksm(folio) && 3723a160e537SMatthew Wilcox (Oracle) folio_ref_count(folio) == 2; 3724c145e0b4SDavid Hildenbrand } 3725c145e0b4SDavid Hildenbrand 37269c28a205SPeter Xu static vm_fault_t pte_marker_clear(struct vm_fault *vmf) 37279c28a205SPeter Xu { 37289c28a205SPeter Xu vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, 37299c28a205SPeter Xu vmf->address, &vmf->ptl); 37303db82b93SHugh Dickins if (!vmf->pte) 37313db82b93SHugh Dickins return 0; 37329c28a205SPeter Xu /* 37339c28a205SPeter Xu * Be careful so that we will only recover a special uffd-wp pte into a 37349c28a205SPeter Xu * none pte. Otherwise it means the pte could have changed, so retry. 37357e3ce3f8SPeter Xu * 37367e3ce3f8SPeter Xu * This should also cover the case where e.g. the pte changed 3737af19487fSAxel Rasmussen * quickly from a PTE_MARKER_UFFD_WP into PTE_MARKER_POISONED. 37387e3ce3f8SPeter Xu * So is_pte_marker() check is not enough to safely drop the pte. 37399c28a205SPeter Xu */ 3740c33c7948SRyan Roberts if (pte_same(vmf->orig_pte, ptep_get(vmf->pte))) 37419c28a205SPeter Xu pte_clear(vmf->vma->vm_mm, vmf->address, vmf->pte); 37429c28a205SPeter Xu pte_unmap_unlock(vmf->pte, vmf->ptl); 37439c28a205SPeter Xu return 0; 37449c28a205SPeter Xu } 37459c28a205SPeter Xu 37462bad466cSPeter Xu static vm_fault_t do_pte_missing(struct vm_fault *vmf) 37472bad466cSPeter Xu { 37482bad466cSPeter Xu if (vma_is_anonymous(vmf->vma)) 37492bad466cSPeter Xu return do_anonymous_page(vmf); 37502bad466cSPeter Xu else 37512bad466cSPeter Xu return do_fault(vmf); 37522bad466cSPeter Xu } 37532bad466cSPeter Xu 37549c28a205SPeter Xu /* 37559c28a205SPeter Xu * This is actually a page-missing access, but with uffd-wp special pte 37569c28a205SPeter Xu * installed. It means this pte was wr-protected before being unmapped. 37579c28a205SPeter Xu */ 37589c28a205SPeter Xu static vm_fault_t pte_marker_handle_uffd_wp(struct vm_fault *vmf) 37599c28a205SPeter Xu { 37609c28a205SPeter Xu /* 37619c28a205SPeter Xu * Just in case there're leftover special ptes even after the region 37627a079ba2SPeter Xu * got unregistered - we can simply clear them. 37639c28a205SPeter Xu */ 37642bad466cSPeter Xu if (unlikely(!userfaultfd_wp(vmf->vma))) 37659c28a205SPeter Xu return pte_marker_clear(vmf); 37669c28a205SPeter Xu 37672bad466cSPeter Xu return do_pte_missing(vmf); 37689c28a205SPeter Xu } 37699c28a205SPeter Xu 37705c041f5dSPeter Xu static vm_fault_t handle_pte_marker(struct vm_fault *vmf) 37715c041f5dSPeter Xu { 37725c041f5dSPeter Xu swp_entry_t entry = pte_to_swp_entry(vmf->orig_pte); 37735c041f5dSPeter Xu unsigned long marker = pte_marker_get(entry); 37745c041f5dSPeter Xu 37755c041f5dSPeter Xu /* 3776ca92ea3dSPeter Xu * PTE markers should never be empty. If anything weird happened, 3777ca92ea3dSPeter Xu * the best thing to do is to kill the process along with its mm. 37785c041f5dSPeter Xu */ 3779ca92ea3dSPeter Xu if (WARN_ON_ONCE(!marker)) 37805c041f5dSPeter Xu return VM_FAULT_SIGBUS; 37815c041f5dSPeter Xu 378215520a3fSPeter Xu /* Higher priority than uffd-wp when data corrupted */ 3783af19487fSAxel Rasmussen if (marker & PTE_MARKER_POISONED) 3784af19487fSAxel Rasmussen return VM_FAULT_HWPOISON; 378515520a3fSPeter Xu 37869c28a205SPeter Xu if (pte_marker_entry_uffd_wp(entry)) 37879c28a205SPeter Xu return pte_marker_handle_uffd_wp(vmf); 37889c28a205SPeter Xu 37899c28a205SPeter Xu /* This is an unknown pte marker */ 37909c28a205SPeter Xu return VM_FAULT_SIGBUS; 37915c041f5dSPeter Xu } 37925c041f5dSPeter Xu 3793b756a3b5SAlistair Popple /* 3794c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 37958f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 37969a95f3cfSPaul Cassella * We return with pte unmapped and unlocked. 37979a95f3cfSPaul Cassella * 3798c1e8d7c6SMichel Lespinasse * We return with the mmap_lock locked or unlocked in the same cases 37999a95f3cfSPaul Cassella * as does filemap_fault(). 38001da177e4SLinus Torvalds */ 38012b740303SSouptick Joarder vm_fault_t do_swap_page(struct vm_fault *vmf) 38021da177e4SLinus Torvalds { 380382b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3804d4f9565aSMatthew Wilcox (Oracle) struct folio *swapcache, *folio = NULL; 3805d4f9565aSMatthew Wilcox (Oracle) struct page *page; 38062799e775SMiaohe Lin struct swap_info_struct *si = NULL; 380714f9135dSDavid Hildenbrand rmap_t rmap_flags = RMAP_NONE; 380813ddaf26SKairui Song bool need_clear_cache = false; 38091493a191SDavid Hildenbrand bool exclusive = false; 381065500d23SHugh Dickins swp_entry_t entry; 38111da177e4SLinus Torvalds pte_t pte; 38122b740303SSouptick Joarder vm_fault_t ret = 0; 3813aae466b0SJoonsoo Kim void *shadow = NULL; 38141da177e4SLinus Torvalds 38152ca99358SPeter Xu if (!pte_unmap_same(vmf)) 38168f4e2101SHugh Dickins goto out; 381765500d23SHugh Dickins 38182994302bSJan Kara entry = pte_to_swp_entry(vmf->orig_pte); 3819d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 38200697212aSChristoph Lameter if (is_migration_entry(entry)) { 382182b0f8c3SJan Kara migration_entry_wait(vma->vm_mm, vmf->pmd, 382282b0f8c3SJan Kara vmf->address); 3823b756a3b5SAlistair Popple } else if (is_device_exclusive_entry(entry)) { 3824b756a3b5SAlistair Popple vmf->page = pfn_swap_entry_to_page(entry); 3825b756a3b5SAlistair Popple ret = remove_device_exclusive_entry(vmf); 38265042db43SJérôme Glisse } else if (is_device_private_entry(entry)) { 38271235ccd0SSuren Baghdasaryan if (vmf->flags & FAULT_FLAG_VMA_LOCK) { 38281235ccd0SSuren Baghdasaryan /* 38291235ccd0SSuren Baghdasaryan * migrate_to_ram is not yet ready to operate 38301235ccd0SSuren Baghdasaryan * under VMA lock. 38311235ccd0SSuren Baghdasaryan */ 38321235ccd0SSuren Baghdasaryan vma_end_read(vma); 38331235ccd0SSuren Baghdasaryan ret = VM_FAULT_RETRY; 38341235ccd0SSuren Baghdasaryan goto out; 38351235ccd0SSuren Baghdasaryan } 38361235ccd0SSuren Baghdasaryan 3837af5cdaf8SAlistair Popple vmf->page = pfn_swap_entry_to_page(entry); 383816ce101dSAlistair Popple vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 383916ce101dSAlistair Popple vmf->address, &vmf->ptl); 38403db82b93SHugh Dickins if (unlikely(!vmf->pte || 3841c33c7948SRyan Roberts !pte_same(ptep_get(vmf->pte), 3842c33c7948SRyan Roberts vmf->orig_pte))) 38433b65f437SRyan Roberts goto unlock; 384416ce101dSAlistair Popple 384516ce101dSAlistair Popple /* 384616ce101dSAlistair Popple * Get a page reference while we know the page can't be 384716ce101dSAlistair Popple * freed. 384816ce101dSAlistair Popple */ 384916ce101dSAlistair Popple get_page(vmf->page); 385016ce101dSAlistair Popple pte_unmap_unlock(vmf->pte, vmf->ptl); 38514a955bedSAlistair Popple ret = vmf->page->pgmap->ops->migrate_to_ram(vmf); 385216ce101dSAlistair Popple put_page(vmf->page); 3853d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 3854d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 38555c041f5dSPeter Xu } else if (is_pte_marker_entry(entry)) { 38565c041f5dSPeter Xu ret = handle_pte_marker(vmf); 3857d1737fdbSAndi Kleen } else { 38582994302bSJan Kara print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); 3859d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 3860d1737fdbSAndi Kleen } 38610697212aSChristoph Lameter goto out; 38620697212aSChristoph Lameter } 38630bcac06fSMinchan Kim 38642799e775SMiaohe Lin /* Prevent swapoff from happening to us. */ 38652799e775SMiaohe Lin si = get_swap_device(entry); 38662799e775SMiaohe Lin if (unlikely(!si)) 38672799e775SMiaohe Lin goto out; 38680bcac06fSMinchan Kim 38695a423081SMatthew Wilcox (Oracle) folio = swap_cache_get_folio(entry, vma, vmf->address); 38705a423081SMatthew Wilcox (Oracle) if (folio) 38715a423081SMatthew Wilcox (Oracle) page = folio_file_page(folio, swp_offset(entry)); 3872d4f9565aSMatthew Wilcox (Oracle) swapcache = folio; 3873f8020772SMinchan Kim 3874d4f9565aSMatthew Wilcox (Oracle) if (!folio) { 3875a449bf58SQian Cai if (data_race(si->flags & SWP_SYNCHRONOUS_IO) && 3876eb085574SHuang Ying __swap_count(entry) == 1) { 387713ddaf26SKairui Song /* 387813ddaf26SKairui Song * Prevent parallel swapin from proceeding with 387913ddaf26SKairui Song * the cache flag. Otherwise, another thread may 388013ddaf26SKairui Song * finish swapin first, free the entry, and swapout 388113ddaf26SKairui Song * reusing the same entry. It's undetectable as 388213ddaf26SKairui Song * pte_same() returns true due to entry reuse. 388313ddaf26SKairui Song */ 388413ddaf26SKairui Song if (swapcache_prepare(entry)) { 388513ddaf26SKairui Song /* Relax a bit to prevent rapid repeated page faults */ 388613ddaf26SKairui Song schedule_timeout_uninterruptible(1); 388713ddaf26SKairui Song goto out; 388813ddaf26SKairui Song } 388913ddaf26SKairui Song need_clear_cache = true; 389013ddaf26SKairui Song 38910bcac06fSMinchan Kim /* skip swapcache */ 389263ad4addSMatthew Wilcox (Oracle) folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, 389363ad4addSMatthew Wilcox (Oracle) vma, vmf->address, false); 389463ad4addSMatthew Wilcox (Oracle) page = &folio->page; 389563ad4addSMatthew Wilcox (Oracle) if (folio) { 389663ad4addSMatthew Wilcox (Oracle) __folio_set_locked(folio); 389763ad4addSMatthew Wilcox (Oracle) __folio_set_swapbacked(folio); 38984c6355b2SJohannes Weiner 389965995918SMatthew Wilcox (Oracle) if (mem_cgroup_swapin_charge_folio(folio, 390063ad4addSMatthew Wilcox (Oracle) vma->vm_mm, GFP_KERNEL, 390163ad4addSMatthew Wilcox (Oracle) entry)) { 3902545b1b07SMichal Hocko ret = VM_FAULT_OOM; 39034c6355b2SJohannes Weiner goto out_page; 3904545b1b07SMichal Hocko } 39050add0c77SShakeel Butt mem_cgroup_swapin_uncharge_swap(entry); 39064c6355b2SJohannes Weiner 3907aae466b0SJoonsoo Kim shadow = get_shadow_from_swap_cache(entry); 3908aae466b0SJoonsoo Kim if (shadow) 390963ad4addSMatthew Wilcox (Oracle) workingset_refault(folio, shadow); 39100076f029SJoonsoo Kim 391163ad4addSMatthew Wilcox (Oracle) folio_add_lru(folio); 39120add0c77SShakeel Butt 3913c9bdf768SMatthew Wilcox (Oracle) /* To provide entry to swap_read_folio() */ 39143d2c9087SDavid Hildenbrand folio->swap = entry; 3915c9bdf768SMatthew Wilcox (Oracle) swap_read_folio(folio, true, NULL); 391663ad4addSMatthew Wilcox (Oracle) folio->private = NULL; 39170bcac06fSMinchan Kim } 3918aa8d22a1SMinchan Kim } else { 3919e9e9b7ecSMinchan Kim page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, 3920e9e9b7ecSMinchan Kim vmf); 392163ad4addSMatthew Wilcox (Oracle) if (page) 392263ad4addSMatthew Wilcox (Oracle) folio = page_folio(page); 3923d4f9565aSMatthew Wilcox (Oracle) swapcache = folio; 39240bcac06fSMinchan Kim } 39250bcac06fSMinchan Kim 3926d4f9565aSMatthew Wilcox (Oracle) if (!folio) { 39271da177e4SLinus Torvalds /* 39288f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 39298f4e2101SHugh Dickins * while we released the pte lock. 39301da177e4SLinus Torvalds */ 393182b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 393282b0f8c3SJan Kara vmf->address, &vmf->ptl); 3933c33c7948SRyan Roberts if (likely(vmf->pte && 3934c33c7948SRyan Roberts pte_same(ptep_get(vmf->pte), vmf->orig_pte))) 39351da177e4SLinus Torvalds ret = VM_FAULT_OOM; 393665500d23SHugh Dickins goto unlock; 39371da177e4SLinus Torvalds } 39381da177e4SLinus Torvalds 39391da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 39401da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 3941f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 39422262185cSRoman Gushchin count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); 3943d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 394471f72525SWu Fengguang /* 394571f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 394671f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 394771f72525SWu Fengguang */ 3948d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 39494779cb31SAndi Kleen goto out_release; 39501da177e4SLinus Torvalds } 39511da177e4SLinus Torvalds 3952fdc724d6SSuren Baghdasaryan ret |= folio_lock_or_retry(folio, vmf); 3953fdc724d6SSuren Baghdasaryan if (ret & VM_FAULT_RETRY) 3954d065bd81SMichel Lespinasse goto out_release; 39551da177e4SLinus Torvalds 395684d60fddSDavid Hildenbrand if (swapcache) { 39574969c119SAndrea Arcangeli /* 39583b344157SMatthew Wilcox (Oracle) * Make sure folio_free_swap() or swapoff did not release the 395984d60fddSDavid Hildenbrand * swapcache from under us. The page pin, and pte_same test 396084d60fddSDavid Hildenbrand * below, are not enough to exclude that. Even if it is still 396184d60fddSDavid Hildenbrand * swapcache, we need to check that the page's swap has not 396284d60fddSDavid Hildenbrand * changed. 39634969c119SAndrea Arcangeli */ 396463ad4addSMatthew Wilcox (Oracle) if (unlikely(!folio_test_swapcache(folio) || 3965cfeed8ffSDavid Hildenbrand page_swap_entry(page).val != entry.val)) 39664969c119SAndrea Arcangeli goto out_page; 39674969c119SAndrea Arcangeli 396884d60fddSDavid Hildenbrand /* 396984d60fddSDavid Hildenbrand * KSM sometimes has to copy on read faults, for example, if 397084d60fddSDavid Hildenbrand * page->index of !PageKSM() pages would be nonlinear inside the 397184d60fddSDavid Hildenbrand * anon VMA -- PageKSM() is lost on actual swapout. 397284d60fddSDavid Hildenbrand */ 397396db66d9SMatthew Wilcox (Oracle) folio = ksm_might_need_to_copy(folio, vma, vmf->address); 397496db66d9SMatthew Wilcox (Oracle) if (unlikely(!folio)) { 39755ad64688SHugh Dickins ret = VM_FAULT_OOM; 397696db66d9SMatthew Wilcox (Oracle) folio = swapcache; 39774969c119SAndrea Arcangeli goto out_page; 397896db66d9SMatthew Wilcox (Oracle) } else if (unlikely(folio == ERR_PTR(-EHWPOISON))) { 39796b970599SKefeng Wang ret = VM_FAULT_HWPOISON; 398096db66d9SMatthew Wilcox (Oracle) folio = swapcache; 39816b970599SKefeng Wang goto out_page; 39824969c119SAndrea Arcangeli } 398396db66d9SMatthew Wilcox (Oracle) if (folio != swapcache) 398496db66d9SMatthew Wilcox (Oracle) page = folio_page(folio, 0); 3985c145e0b4SDavid Hildenbrand 3986c145e0b4SDavid Hildenbrand /* 3987c145e0b4SDavid Hildenbrand * If we want to map a page that's in the swapcache writable, we 3988c145e0b4SDavid Hildenbrand * have to detect via the refcount if we're really the exclusive 3989c145e0b4SDavid Hildenbrand * owner. Try removing the extra reference from the local LRU 39901fec6890SMatthew Wilcox (Oracle) * caches if required. 3991c145e0b4SDavid Hildenbrand */ 3992d4f9565aSMatthew Wilcox (Oracle) if ((vmf->flags & FAULT_FLAG_WRITE) && folio == swapcache && 399363ad4addSMatthew Wilcox (Oracle) !folio_test_ksm(folio) && !folio_test_lru(folio)) 3994c145e0b4SDavid Hildenbrand lru_add_drain(); 399584d60fddSDavid Hildenbrand } 39965ad64688SHugh Dickins 39974231f842SKefeng Wang folio_throttle_swaprate(folio, GFP_KERNEL); 3998073e587eSKAMEZAWA Hiroyuki 39991da177e4SLinus Torvalds /* 40008f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 40011da177e4SLinus Torvalds */ 400282b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 400382b0f8c3SJan Kara &vmf->ptl); 4004c33c7948SRyan Roberts if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) 4005b8107480SKirill Korotaev goto out_nomap; 4006b8107480SKirill Korotaev 400763ad4addSMatthew Wilcox (Oracle) if (unlikely(!folio_test_uptodate(folio))) { 4008b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 4009b8107480SKirill Korotaev goto out_nomap; 40101da177e4SLinus Torvalds } 40111da177e4SLinus Torvalds 40128c7c6e34SKAMEZAWA Hiroyuki /* 401378fbe906SDavid Hildenbrand * PG_anon_exclusive reuses PG_mappedtodisk for anon pages. A swap pte 401478fbe906SDavid Hildenbrand * must never point at an anonymous page in the swapcache that is 401578fbe906SDavid Hildenbrand * PG_anon_exclusive. Sanity check that this holds and especially, that 401678fbe906SDavid Hildenbrand * no filesystem set PG_mappedtodisk on a page in the swapcache. Sanity 401778fbe906SDavid Hildenbrand * check after taking the PT lock and making sure that nobody 401878fbe906SDavid Hildenbrand * concurrently faulted in this page and set PG_anon_exclusive. 401978fbe906SDavid Hildenbrand */ 402063ad4addSMatthew Wilcox (Oracle) BUG_ON(!folio_test_anon(folio) && folio_test_mappedtodisk(folio)); 402163ad4addSMatthew Wilcox (Oracle) BUG_ON(folio_test_anon(folio) && PageAnonExclusive(page)); 402278fbe906SDavid Hildenbrand 402378fbe906SDavid Hildenbrand /* 40241493a191SDavid Hildenbrand * Check under PT lock (to protect against concurrent fork() sharing 40251493a191SDavid Hildenbrand * the swap entry concurrently) for certainly exclusive pages. 40261493a191SDavid Hildenbrand */ 402763ad4addSMatthew Wilcox (Oracle) if (!folio_test_ksm(folio)) { 40281493a191SDavid Hildenbrand exclusive = pte_swp_exclusive(vmf->orig_pte); 4029d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache) { 40301493a191SDavid Hildenbrand /* 40311493a191SDavid Hildenbrand * We have a fresh page that is not exposed to the 40321493a191SDavid Hildenbrand * swapcache -> certainly exclusive. 40331493a191SDavid Hildenbrand */ 40341493a191SDavid Hildenbrand exclusive = true; 403563ad4addSMatthew Wilcox (Oracle) } else if (exclusive && folio_test_writeback(folio) && 4036eacde327SMiaohe Lin data_race(si->flags & SWP_STABLE_WRITES)) { 40371493a191SDavid Hildenbrand /* 40381493a191SDavid Hildenbrand * This is tricky: not all swap backends support 40391493a191SDavid Hildenbrand * concurrent page modifications while under writeback. 40401493a191SDavid Hildenbrand * 40411493a191SDavid Hildenbrand * So if we stumble over such a page in the swapcache 40421493a191SDavid Hildenbrand * we must not set the page exclusive, otherwise we can 40431493a191SDavid Hildenbrand * map it writable without further checks and modify it 40441493a191SDavid Hildenbrand * while still under writeback. 40451493a191SDavid Hildenbrand * 40461493a191SDavid Hildenbrand * For these problematic swap backends, simply drop the 40471493a191SDavid Hildenbrand * exclusive marker: this is perfectly fine as we start 40481493a191SDavid Hildenbrand * writeback only if we fully unmapped the page and 40491493a191SDavid Hildenbrand * there are no unexpected references on the page after 40501493a191SDavid Hildenbrand * unmapping succeeded. After fully unmapped, no 40511493a191SDavid Hildenbrand * further GUP references (FOLL_GET and FOLL_PIN) can 40521493a191SDavid Hildenbrand * appear, so dropping the exclusive marker and mapping 40531493a191SDavid Hildenbrand * it only R/O is fine. 40541493a191SDavid Hildenbrand */ 40551493a191SDavid Hildenbrand exclusive = false; 40561493a191SDavid Hildenbrand } 40571493a191SDavid Hildenbrand } 40581493a191SDavid Hildenbrand 40591493a191SDavid Hildenbrand /* 40606dca4ac6SPeter Collingbourne * Some architectures may have to restore extra metadata to the page 40616dca4ac6SPeter Collingbourne * when reading from swap. This metadata may be indexed by swap entry 40626dca4ac6SPeter Collingbourne * so this must be called before swap_free(). 40636dca4ac6SPeter Collingbourne */ 40646dca4ac6SPeter Collingbourne arch_swap_restore(entry, folio); 40656dca4ac6SPeter Collingbourne 40666dca4ac6SPeter Collingbourne /* 4067c145e0b4SDavid Hildenbrand * Remove the swap entry and conditionally try to free up the swapcache. 4068c145e0b4SDavid Hildenbrand * We're already holding a reference on the page but haven't mapped it 4069c145e0b4SDavid Hildenbrand * yet. 40708c7c6e34SKAMEZAWA Hiroyuki */ 4071c145e0b4SDavid Hildenbrand swap_free(entry); 4072a160e537SMatthew Wilcox (Oracle) if (should_try_to_free_swap(folio, vma, vmf->flags)) 4073a160e537SMatthew Wilcox (Oracle) folio_free_swap(folio); 40741da177e4SLinus Torvalds 4075f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, MM_ANONPAGES); 4076f1a79412SShakeel Butt dec_mm_counter(vma->vm_mm, MM_SWAPENTS); 40771da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 4078c145e0b4SDavid Hildenbrand 4079c145e0b4SDavid Hildenbrand /* 40801493a191SDavid Hildenbrand * Same logic as in do_wp_page(); however, optimize for pages that are 40811493a191SDavid Hildenbrand * certainly not shared either because we just allocated them without 40821493a191SDavid Hildenbrand * exposing them to the swapcache or because the swap entry indicates 40831493a191SDavid Hildenbrand * exclusivity. 4084c145e0b4SDavid Hildenbrand */ 408563ad4addSMatthew Wilcox (Oracle) if (!folio_test_ksm(folio) && 408663ad4addSMatthew Wilcox (Oracle) (exclusive || folio_ref_count(folio) == 1)) { 40876c287605SDavid Hildenbrand if (vmf->flags & FAULT_FLAG_WRITE) { 40881da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 408982b0f8c3SJan Kara vmf->flags &= ~FAULT_FLAG_WRITE; 40906c287605SDavid Hildenbrand } 409114f9135dSDavid Hildenbrand rmap_flags |= RMAP_EXCLUSIVE; 40921da177e4SLinus Torvalds } 40931da177e4SLinus Torvalds flush_icache_page(vma, page); 40942994302bSJan Kara if (pte_swp_soft_dirty(vmf->orig_pte)) 4095179ef71cSCyrill Gorcunov pte = pte_mksoft_dirty(pte); 4096f1eb1bacSPeter Xu if (pte_swp_uffd_wp(vmf->orig_pte)) 4097f45ec5ffSPeter Xu pte = pte_mkuffd_wp(pte); 40982994302bSJan Kara vmf->orig_pte = pte; 40990bcac06fSMinchan Kim 41000bcac06fSMinchan Kim /* ksm created a completely new copy */ 4101d4f9565aSMatthew Wilcox (Oracle) if (unlikely(folio != swapcache && swapcache)) { 41022853b66bSMatthew Wilcox (Oracle) folio_add_new_anon_rmap(folio, vma, vmf->address); 410363ad4addSMatthew Wilcox (Oracle) folio_add_lru_vma(folio, vma); 41040bcac06fSMinchan Kim } else { 4105b832a354SDavid Hildenbrand folio_add_anon_rmap_pte(folio, page, vma, vmf->address, 4106b832a354SDavid Hildenbrand rmap_flags); 410700501b53SJohannes Weiner } 41081da177e4SLinus Torvalds 410963ad4addSMatthew Wilcox (Oracle) VM_BUG_ON(!folio_test_anon(folio) || 411063ad4addSMatthew Wilcox (Oracle) (pte_write(pte) && !PageAnonExclusive(page))); 41111eba86c0SPasha Tatashin set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); 41121eba86c0SPasha Tatashin arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte); 41131eba86c0SPasha Tatashin 411463ad4addSMatthew Wilcox (Oracle) folio_unlock(folio); 4115d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache && swapcache) { 41164969c119SAndrea Arcangeli /* 41174969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 41184969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 41194969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 41204969c119SAndrea Arcangeli * further safety release the lock after the swap_free 41214969c119SAndrea Arcangeli * so that the swap count won't change under a 41224969c119SAndrea Arcangeli * parallel locked swapcache. 41234969c119SAndrea Arcangeli */ 4124d4f9565aSMatthew Wilcox (Oracle) folio_unlock(swapcache); 4125d4f9565aSMatthew Wilcox (Oracle) folio_put(swapcache); 41264969c119SAndrea Arcangeli } 4127c475a8abSHugh Dickins 412882b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) { 41292994302bSJan Kara ret |= do_wp_page(vmf); 413061469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 413161469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 41321da177e4SLinus Torvalds goto out; 41331da177e4SLinus Torvalds } 41341da177e4SLinus Torvalds 41351da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 41365003a2bdSMatthew Wilcox (Oracle) update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1); 413765500d23SHugh Dickins unlock: 41383db82b93SHugh Dickins if (vmf->pte) 413982b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 41401da177e4SLinus Torvalds out: 414113ddaf26SKairui Song /* Clear the swap cache pin for direct swapin after PTL unlock */ 414213ddaf26SKairui Song if (need_clear_cache) 414313ddaf26SKairui Song swapcache_clear(si, entry); 41442799e775SMiaohe Lin if (si) 41452799e775SMiaohe Lin put_swap_device(si); 41461da177e4SLinus Torvalds return ret; 4147b8107480SKirill Korotaev out_nomap: 41483db82b93SHugh Dickins if (vmf->pte) 414982b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 4150bc43f75cSJohannes Weiner out_page: 415163ad4addSMatthew Wilcox (Oracle) folio_unlock(folio); 41524779cb31SAndi Kleen out_release: 415363ad4addSMatthew Wilcox (Oracle) folio_put(folio); 4154d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache && swapcache) { 4155d4f9565aSMatthew Wilcox (Oracle) folio_unlock(swapcache); 4156d4f9565aSMatthew Wilcox (Oracle) folio_put(swapcache); 41574969c119SAndrea Arcangeli } 415813ddaf26SKairui Song if (need_clear_cache) 415913ddaf26SKairui Song swapcache_clear(si, entry); 41602799e775SMiaohe Lin if (si) 41612799e775SMiaohe Lin put_swap_device(si); 416265500d23SHugh Dickins return ret; 41631da177e4SLinus Torvalds } 41641da177e4SLinus Torvalds 416519eaf449SRyan Roberts static bool pte_range_none(pte_t *pte, int nr_pages) 416619eaf449SRyan Roberts { 416719eaf449SRyan Roberts int i; 416819eaf449SRyan Roberts 416919eaf449SRyan Roberts for (i = 0; i < nr_pages; i++) { 417019eaf449SRyan Roberts if (!pte_none(ptep_get_lockless(pte + i))) 417119eaf449SRyan Roberts return false; 417219eaf449SRyan Roberts } 417319eaf449SRyan Roberts 417419eaf449SRyan Roberts return true; 417519eaf449SRyan Roberts } 417619eaf449SRyan Roberts 417719eaf449SRyan Roberts static struct folio *alloc_anon_folio(struct vm_fault *vmf) 417819eaf449SRyan Roberts { 417919eaf449SRyan Roberts struct vm_area_struct *vma = vmf->vma; 4180085ff35eSKefeng Wang #ifdef CONFIG_TRANSPARENT_HUGEPAGE 418119eaf449SRyan Roberts unsigned long orders; 418219eaf449SRyan Roberts struct folio *folio; 418319eaf449SRyan Roberts unsigned long addr; 418419eaf449SRyan Roberts pte_t *pte; 418519eaf449SRyan Roberts gfp_t gfp; 418619eaf449SRyan Roberts int order; 418719eaf449SRyan Roberts 418819eaf449SRyan Roberts /* 418919eaf449SRyan Roberts * If uffd is active for the vma we need per-page fault fidelity to 419019eaf449SRyan Roberts * maintain the uffd semantics. 419119eaf449SRyan Roberts */ 419219eaf449SRyan Roberts if (unlikely(userfaultfd_armed(vma))) 419319eaf449SRyan Roberts goto fallback; 419419eaf449SRyan Roberts 419519eaf449SRyan Roberts /* 419619eaf449SRyan Roberts * Get a list of all the (large) orders below PMD_ORDER that are enabled 419719eaf449SRyan Roberts * for this vma. Then filter out the orders that can't be allocated over 419819eaf449SRyan Roberts * the faulting address and still be fully contained in the vma. 419919eaf449SRyan Roberts */ 420019eaf449SRyan Roberts orders = thp_vma_allowable_orders(vma, vma->vm_flags, false, true, true, 420119eaf449SRyan Roberts BIT(PMD_ORDER) - 1); 420219eaf449SRyan Roberts orders = thp_vma_suitable_orders(vma, vmf->address, orders); 420319eaf449SRyan Roberts 420419eaf449SRyan Roberts if (!orders) 420519eaf449SRyan Roberts goto fallback; 420619eaf449SRyan Roberts 420719eaf449SRyan Roberts pte = pte_offset_map(vmf->pmd, vmf->address & PMD_MASK); 420819eaf449SRyan Roberts if (!pte) 420919eaf449SRyan Roberts return ERR_PTR(-EAGAIN); 421019eaf449SRyan Roberts 421119eaf449SRyan Roberts /* 421219eaf449SRyan Roberts * Find the highest order where the aligned range is completely 421319eaf449SRyan Roberts * pte_none(). Note that all remaining orders will be completely 421419eaf449SRyan Roberts * pte_none(). 421519eaf449SRyan Roberts */ 421619eaf449SRyan Roberts order = highest_order(orders); 421719eaf449SRyan Roberts while (orders) { 421819eaf449SRyan Roberts addr = ALIGN_DOWN(vmf->address, PAGE_SIZE << order); 421919eaf449SRyan Roberts if (pte_range_none(pte + pte_index(addr), 1 << order)) 422019eaf449SRyan Roberts break; 422119eaf449SRyan Roberts order = next_order(&orders, order); 422219eaf449SRyan Roberts } 422319eaf449SRyan Roberts 422419eaf449SRyan Roberts pte_unmap(pte); 422519eaf449SRyan Roberts 422619eaf449SRyan Roberts /* Try allocating the highest of the remaining orders. */ 422719eaf449SRyan Roberts gfp = vma_thp_gfp_mask(vma); 422819eaf449SRyan Roberts while (orders) { 422919eaf449SRyan Roberts addr = ALIGN_DOWN(vmf->address, PAGE_SIZE << order); 423019eaf449SRyan Roberts folio = vma_alloc_folio(gfp, order, vma, addr, true); 423119eaf449SRyan Roberts if (folio) { 4232085ff35eSKefeng Wang if (mem_cgroup_charge(folio, vma->vm_mm, gfp)) { 4233085ff35eSKefeng Wang folio_put(folio); 4234085ff35eSKefeng Wang goto next; 4235085ff35eSKefeng Wang } 4236085ff35eSKefeng Wang folio_throttle_swaprate(folio, gfp); 423719eaf449SRyan Roberts clear_huge_page(&folio->page, vmf->address, 1 << order); 423819eaf449SRyan Roberts return folio; 423919eaf449SRyan Roberts } 4240085ff35eSKefeng Wang next: 424119eaf449SRyan Roberts order = next_order(&orders, order); 424219eaf449SRyan Roberts } 424319eaf449SRyan Roberts 424419eaf449SRyan Roberts fallback: 424519eaf449SRyan Roberts #endif 4246085ff35eSKefeng Wang return folio_prealloc(vma->vm_mm, vma, vmf->address, true); 424719eaf449SRyan Roberts } 424819eaf449SRyan Roberts 42491da177e4SLinus Torvalds /* 4250c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 42518f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 4252c1e8d7c6SMichel Lespinasse * We return with mmap_lock still held, but pte unmapped and unlocked. 42531da177e4SLinus Torvalds */ 42542b740303SSouptick Joarder static vm_fault_t do_anonymous_page(struct vm_fault *vmf) 42551da177e4SLinus Torvalds { 42562bad466cSPeter Xu bool uffd_wp = vmf_orig_pte_uffd_wp(vmf); 425782b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 425819eaf449SRyan Roberts unsigned long addr = vmf->address; 42596bc56a4dSMatthew Wilcox (Oracle) struct folio *folio; 42602b740303SSouptick Joarder vm_fault_t ret = 0; 426119eaf449SRyan Roberts int nr_pages = 1; 42621da177e4SLinus Torvalds pte_t entry; 426319eaf449SRyan Roberts int i; 42641da177e4SLinus Torvalds 42656b7339f4SKirill A. Shutemov /* File mapping without ->vm_ops ? */ 42666b7339f4SKirill A. Shutemov if (vma->vm_flags & VM_SHARED) 42676b7339f4SKirill A. Shutemov return VM_FAULT_SIGBUS; 42686b7339f4SKirill A. Shutemov 42697267ec00SKirill A. Shutemov /* 42703db82b93SHugh Dickins * Use pte_alloc() instead of pte_alloc_map(), so that OOM can 42713db82b93SHugh Dickins * be distinguished from a transient failure of pte_offset_map(). 42727267ec00SKirill A. Shutemov */ 42734cf58924SJoel Fernandes (Google) if (pte_alloc(vma->vm_mm, vmf->pmd)) 42747267ec00SKirill A. Shutemov return VM_FAULT_OOM; 42757267ec00SKirill A. Shutemov 427611ac5524SLinus Torvalds /* Use the zero-page for reads */ 427782b0f8c3SJan Kara if (!(vmf->flags & FAULT_FLAG_WRITE) && 4278bae473a4SKirill A. Shutemov !mm_forbids_zeropage(vma->vm_mm)) { 427982b0f8c3SJan Kara entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), 428062eede62SHugh Dickins vma->vm_page_prot)); 428182b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 428282b0f8c3SJan Kara vmf->address, &vmf->ptl); 42833db82b93SHugh Dickins if (!vmf->pte) 42843db82b93SHugh Dickins goto unlock; 42852bad466cSPeter Xu if (vmf_pte_changed(vmf)) { 42867df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 4287a13ea5b7SHugh Dickins goto unlock; 42887df67697SBibo Mao } 42896b31d595SMichal Hocko ret = check_stable_address_space(vma->vm_mm); 42906b31d595SMichal Hocko if (ret) 42916b31d595SMichal Hocko goto unlock; 42926b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 42936b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 429482b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 429582b0f8c3SJan Kara return handle_userfault(vmf, VM_UFFD_MISSING); 42966b251fc9SAndrea Arcangeli } 4297a13ea5b7SHugh Dickins goto setpte; 4298a13ea5b7SHugh Dickins } 4299a13ea5b7SHugh Dickins 43001da177e4SLinus Torvalds /* Allocate our own private page. */ 43011da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 430265500d23SHugh Dickins goto oom; 430319eaf449SRyan Roberts /* Returns NULL on OOM or ERR_PTR(-EAGAIN) if we must retry the fault */ 430419eaf449SRyan Roberts folio = alloc_anon_folio(vmf); 430519eaf449SRyan Roberts if (IS_ERR(folio)) 430619eaf449SRyan Roberts return 0; 43076bc56a4dSMatthew Wilcox (Oracle) if (!folio) 430865500d23SHugh Dickins goto oom; 4309eb3c24f3SMel Gorman 431019eaf449SRyan Roberts nr_pages = folio_nr_pages(folio); 431119eaf449SRyan Roberts addr = ALIGN_DOWN(vmf->address, nr_pages * PAGE_SIZE); 431219eaf449SRyan Roberts 431352f37629SMinchan Kim /* 4314cb3184deSMatthew Wilcox (Oracle) * The memory barrier inside __folio_mark_uptodate makes sure that 4315f4f5329dSWei Yang * preceding stores to the page contents become visible before 431652f37629SMinchan Kim * the set_pte_at() write. 431752f37629SMinchan Kim */ 4318cb3184deSMatthew Wilcox (Oracle) __folio_mark_uptodate(folio); 43191da177e4SLinus Torvalds 4320cb3184deSMatthew Wilcox (Oracle) entry = mk_pte(&folio->page, vma->vm_page_prot); 432150c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 43221ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 4323161e393cSRick Edgecombe entry = pte_mkwrite(pte_mkdirty(entry), vma); 43248f4e2101SHugh Dickins 432519eaf449SRyan Roberts vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); 43263db82b93SHugh Dickins if (!vmf->pte) 43273db82b93SHugh Dickins goto release; 432819eaf449SRyan Roberts if (nr_pages == 1 && vmf_pte_changed(vmf)) { 432919eaf449SRyan Roberts update_mmu_tlb(vma, addr, vmf->pte); 433019eaf449SRyan Roberts goto release; 433119eaf449SRyan Roberts } else if (nr_pages > 1 && !pte_range_none(vmf->pte, nr_pages)) { 433219eaf449SRyan Roberts for (i = 0; i < nr_pages; i++) 433319eaf449SRyan Roberts update_mmu_tlb(vma, addr + PAGE_SIZE * i, vmf->pte + i); 43348f4e2101SHugh Dickins goto release; 43357df67697SBibo Mao } 43369ba69294SHugh Dickins 43376b31d595SMichal Hocko ret = check_stable_address_space(vma->vm_mm); 43386b31d595SMichal Hocko if (ret) 43396b31d595SMichal Hocko goto release; 43406b31d595SMichal Hocko 43416b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 43426b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 434382b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 4344cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 434582b0f8c3SJan Kara return handle_userfault(vmf, VM_UFFD_MISSING); 43466b251fc9SAndrea Arcangeli } 43476b251fc9SAndrea Arcangeli 434819eaf449SRyan Roberts folio_ref_add(folio, nr_pages - 1); 434919eaf449SRyan Roberts add_mm_counter(vma->vm_mm, MM_ANONPAGES, nr_pages); 435019eaf449SRyan Roberts folio_add_new_anon_rmap(folio, vma, addr); 4351cb3184deSMatthew Wilcox (Oracle) folio_add_lru_vma(folio, vma); 4352a13ea5b7SHugh Dickins setpte: 43532bad466cSPeter Xu if (uffd_wp) 43542bad466cSPeter Xu entry = pte_mkuffd_wp(entry); 435519eaf449SRyan Roberts set_ptes(vma->vm_mm, addr, vmf->pte, entry, nr_pages); 43561da177e4SLinus Torvalds 43571da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 435819eaf449SRyan Roberts update_mmu_cache_range(vmf, vma, addr, vmf->pte, nr_pages); 435965500d23SHugh Dickins unlock: 43603db82b93SHugh Dickins if (vmf->pte) 436182b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 43626b31d595SMichal Hocko return ret; 43638f4e2101SHugh Dickins release: 4364cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 43658f4e2101SHugh Dickins goto unlock; 436665500d23SHugh Dickins oom: 43671da177e4SLinus Torvalds return VM_FAULT_OOM; 43681da177e4SLinus Torvalds } 43691da177e4SLinus Torvalds 43709a95f3cfSPaul Cassella /* 4371c1e8d7c6SMichel Lespinasse * The mmap_lock must have been held on entry, and may have been 43729a95f3cfSPaul Cassella * released depending on flags and vma->vm_ops->fault() return value. 43739a95f3cfSPaul Cassella * See filemap_fault() and __lock_page_retry(). 43749a95f3cfSPaul Cassella */ 43752b740303SSouptick Joarder static vm_fault_t __do_fault(struct vm_fault *vmf) 43767eae74afSKirill A. Shutemov { 437782b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 437801d1e0e6SMatthew Wilcox (Oracle) struct folio *folio; 43792b740303SSouptick Joarder vm_fault_t ret; 43807eae74afSKirill A. Shutemov 438163f3655fSMichal Hocko /* 438263f3655fSMichal Hocko * Preallocate pte before we take page_lock because this might lead to 438363f3655fSMichal Hocko * deadlocks for memcg reclaim which waits for pages under writeback: 438463f3655fSMichal Hocko * lock_page(A) 438563f3655fSMichal Hocko * SetPageWriteback(A) 438663f3655fSMichal Hocko * unlock_page(A) 438763f3655fSMichal Hocko * lock_page(B) 438863f3655fSMichal Hocko * lock_page(B) 4389d383807aSYanfei Xu * pte_alloc_one 439063f3655fSMichal Hocko * shrink_page_list 439163f3655fSMichal Hocko * wait_on_page_writeback(A) 439263f3655fSMichal Hocko * SetPageWriteback(B) 439363f3655fSMichal Hocko * unlock_page(B) 439463f3655fSMichal Hocko * # flush A, B to clear the writeback 439563f3655fSMichal Hocko */ 439663f3655fSMichal Hocko if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) { 4397a7069ee3SYanfei Xu vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); 439863f3655fSMichal Hocko if (!vmf->prealloc_pte) 439963f3655fSMichal Hocko return VM_FAULT_OOM; 440063f3655fSMichal Hocko } 440163f3655fSMichal Hocko 440211bac800SDave Jiang ret = vma->vm_ops->fault(vmf); 44033917048dSJan Kara if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | 4404b1aa812bSJan Kara VM_FAULT_DONE_COW))) 44057eae74afSKirill A. Shutemov return ret; 44067eae74afSKirill A. Shutemov 440701d1e0e6SMatthew Wilcox (Oracle) folio = page_folio(vmf->page); 4408667240e0SJan Kara if (unlikely(PageHWPoison(vmf->page))) { 4409e53ac737SRik van Riel vm_fault_t poisonret = VM_FAULT_HWPOISON; 4410e53ac737SRik van Riel if (ret & VM_FAULT_LOCKED) { 441101d1e0e6SMatthew Wilcox (Oracle) if (page_mapped(vmf->page)) 441201d1e0e6SMatthew Wilcox (Oracle) unmap_mapping_folio(folio); 441301d1e0e6SMatthew Wilcox (Oracle) /* Retry if a clean folio was removed from the cache. */ 441401d1e0e6SMatthew Wilcox (Oracle) if (mapping_evict_folio(folio->mapping, folio)) 44153149c79fSRik van Riel poisonret = VM_FAULT_NOPAGE; 441601d1e0e6SMatthew Wilcox (Oracle) folio_unlock(folio); 4417e53ac737SRik van Riel } 441801d1e0e6SMatthew Wilcox (Oracle) folio_put(folio); 4419936ca80dSJan Kara vmf->page = NULL; 4420e53ac737SRik van Riel return poisonret; 44217eae74afSKirill A. Shutemov } 44227eae74afSKirill A. Shutemov 44237eae74afSKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) 442401d1e0e6SMatthew Wilcox (Oracle) folio_lock(folio); 44257eae74afSKirill A. Shutemov else 442601d1e0e6SMatthew Wilcox (Oracle) VM_BUG_ON_PAGE(!folio_test_locked(folio), vmf->page); 44277eae74afSKirill A. Shutemov 44287eae74afSKirill A. Shutemov return ret; 44297eae74afSKirill A. Shutemov } 44307eae74afSKirill A. Shutemov 4431396bcc52SMatthew Wilcox (Oracle) #ifdef CONFIG_TRANSPARENT_HUGEPAGE 443282b0f8c3SJan Kara static void deposit_prealloc_pte(struct vm_fault *vmf) 4433953c66c2SAneesh Kumar K.V { 443482b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4435953c66c2SAneesh Kumar K.V 443682b0f8c3SJan Kara pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); 4437953c66c2SAneesh Kumar K.V /* 4438953c66c2SAneesh Kumar K.V * We are going to consume the prealloc table, 4439953c66c2SAneesh Kumar K.V * count that as nr_ptes. 4440953c66c2SAneesh Kumar K.V */ 4441c4812909SKirill A. Shutemov mm_inc_nr_ptes(vma->vm_mm); 44427f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 4443953c66c2SAneesh Kumar K.V } 4444953c66c2SAneesh Kumar K.V 4445f9ce0be7SKirill A. Shutemov vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) 444610102459SKirill A. Shutemov { 4447ef37b2eaSDavid Hildenbrand struct folio *folio = page_folio(page); 444882b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 444982b0f8c3SJan Kara bool write = vmf->flags & FAULT_FLAG_WRITE; 445082b0f8c3SJan Kara unsigned long haddr = vmf->address & HPAGE_PMD_MASK; 445110102459SKirill A. Shutemov pmd_t entry; 4452d01ac3c3SMatthew Wilcox (Oracle) vm_fault_t ret = VM_FAULT_FALLBACK; 445310102459SKirill A. Shutemov 44543485b883SRyan Roberts if (!thp_vma_suitable_order(vma, haddr, PMD_ORDER)) 4455d01ac3c3SMatthew Wilcox (Oracle) return ret; 445610102459SKirill A. Shutemov 4457ef37b2eaSDavid Hildenbrand if (page != &folio->page || folio_order(folio) != HPAGE_PMD_ORDER) 4458d01ac3c3SMatthew Wilcox (Oracle) return ret; 445910102459SKirill A. Shutemov 4460953c66c2SAneesh Kumar K.V /* 4461eac96c3eSYang Shi * Just backoff if any subpage of a THP is corrupted otherwise 4462eac96c3eSYang Shi * the corrupted page may mapped by PMD silently to escape the 4463eac96c3eSYang Shi * check. This kind of THP just can be PTE mapped. Access to 4464eac96c3eSYang Shi * the corrupted subpage should trigger SIGBUS as expected. 4465eac96c3eSYang Shi */ 4466ef37b2eaSDavid Hildenbrand if (unlikely(folio_test_has_hwpoisoned(folio))) 4467eac96c3eSYang Shi return ret; 4468eac96c3eSYang Shi 4469eac96c3eSYang Shi /* 4470f0953a1bSIngo Molnar * Archs like ppc64 need additional space to store information 4471953c66c2SAneesh Kumar K.V * related to pte entry. Use the preallocated table for that. 4472953c66c2SAneesh Kumar K.V */ 447382b0f8c3SJan Kara if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { 44744cf58924SJoel Fernandes (Google) vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); 447582b0f8c3SJan Kara if (!vmf->prealloc_pte) 4476953c66c2SAneesh Kumar K.V return VM_FAULT_OOM; 4477953c66c2SAneesh Kumar K.V } 4478953c66c2SAneesh Kumar K.V 447982b0f8c3SJan Kara vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); 448082b0f8c3SJan Kara if (unlikely(!pmd_none(*vmf->pmd))) 448110102459SKirill A. Shutemov goto out; 448210102459SKirill A. Shutemov 44839f1f5b60SMatthew Wilcox (Oracle) flush_icache_pages(vma, page, HPAGE_PMD_NR); 448410102459SKirill A. Shutemov 448510102459SKirill A. Shutemov entry = mk_huge_pmd(page, vma->vm_page_prot); 448610102459SKirill A. Shutemov if (write) 4487f55e1014SLinus Torvalds entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); 448810102459SKirill A. Shutemov 44896b27cc6cSKefeng Wang add_mm_counter(vma->vm_mm, mm_counter_file(folio), HPAGE_PMD_NR); 4490ef37b2eaSDavid Hildenbrand folio_add_file_rmap_pmd(folio, page, vma); 4491cea86fe2SHugh Dickins 4492953c66c2SAneesh Kumar K.V /* 4493953c66c2SAneesh Kumar K.V * deposit and withdraw with pmd lock held 4494953c66c2SAneesh Kumar K.V */ 4495953c66c2SAneesh Kumar K.V if (arch_needs_pgtable_deposit()) 449682b0f8c3SJan Kara deposit_prealloc_pte(vmf); 449710102459SKirill A. Shutemov 449882b0f8c3SJan Kara set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); 449910102459SKirill A. Shutemov 450082b0f8c3SJan Kara update_mmu_cache_pmd(vma, haddr, vmf->pmd); 450110102459SKirill A. Shutemov 450210102459SKirill A. Shutemov /* fault is handled */ 450310102459SKirill A. Shutemov ret = 0; 450495ecedcdSKirill A. Shutemov count_vm_event(THP_FILE_MAPPED); 450510102459SKirill A. Shutemov out: 450682b0f8c3SJan Kara spin_unlock(vmf->ptl); 450710102459SKirill A. Shutemov return ret; 450810102459SKirill A. Shutemov } 450910102459SKirill A. Shutemov #else 4510f9ce0be7SKirill A. Shutemov vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) 451110102459SKirill A. Shutemov { 4512f9ce0be7SKirill A. Shutemov return VM_FAULT_FALLBACK; 451310102459SKirill A. Shutemov } 451410102459SKirill A. Shutemov #endif 451510102459SKirill A. Shutemov 45163bd786f7SYin Fengwei /** 45173bd786f7SYin Fengwei * set_pte_range - Set a range of PTEs to point to pages in a folio. 45183bd786f7SYin Fengwei * @vmf: Fault decription. 45193bd786f7SYin Fengwei * @folio: The folio that contains @page. 45203bd786f7SYin Fengwei * @page: The first page to create a PTE for. 45213bd786f7SYin Fengwei * @nr: The number of PTEs to create. 45223bd786f7SYin Fengwei * @addr: The first address to create a PTE for. 45233bd786f7SYin Fengwei */ 45243bd786f7SYin Fengwei void set_pte_range(struct vm_fault *vmf, struct folio *folio, 45253bd786f7SYin Fengwei struct page *page, unsigned int nr, unsigned long addr) 45263bb97794SKirill A. Shutemov { 452782b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 45282bad466cSPeter Xu bool uffd_wp = vmf_orig_pte_uffd_wp(vmf); 452982b0f8c3SJan Kara bool write = vmf->flags & FAULT_FLAG_WRITE; 45303bd786f7SYin Fengwei bool prefault = in_range(vmf->address, addr, nr * PAGE_SIZE); 45313bb97794SKirill A. Shutemov pte_t entry; 45327267ec00SKirill A. Shutemov 45333bd786f7SYin Fengwei flush_icache_pages(vma, page, nr); 45343bb97794SKirill A. Shutemov entry = mk_pte(page, vma->vm_page_prot); 453546bdb427SWill Deacon 453646bdb427SWill Deacon if (prefault && arch_wants_old_prefaulted_pte()) 453746bdb427SWill Deacon entry = pte_mkold(entry); 453850c25ee9SThomas Bogendoerfer else 453950c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 454046bdb427SWill Deacon 45413bb97794SKirill A. Shutemov if (write) 45423bb97794SKirill A. Shutemov entry = maybe_mkwrite(pte_mkdirty(entry), vma); 45439c28a205SPeter Xu if (unlikely(uffd_wp)) 4544f1eb1bacSPeter Xu entry = pte_mkuffd_wp(entry); 4545bae473a4SKirill A. Shutemov /* copy-on-write page */ 4546bae473a4SKirill A. Shutemov if (write && !(vma->vm_flags & VM_SHARED)) { 45473bd786f7SYin Fengwei add_mm_counter(vma->vm_mm, MM_ANONPAGES, nr); 45483bd786f7SYin Fengwei VM_BUG_ON_FOLIO(nr != 1, folio); 45493bd786f7SYin Fengwei folio_add_new_anon_rmap(folio, vma, addr); 45503bd786f7SYin Fengwei folio_add_lru_vma(folio, vma); 45513bb97794SKirill A. Shutemov } else { 45526b27cc6cSKefeng Wang add_mm_counter(vma->vm_mm, mm_counter_file(folio), nr); 455368f03208SDavid Hildenbrand folio_add_file_rmap_ptes(folio, page, nr, vma); 45543bb97794SKirill A. Shutemov } 45553bd786f7SYin Fengwei set_ptes(vma->vm_mm, addr, vmf->pte, entry, nr); 45563bd786f7SYin Fengwei 45573bd786f7SYin Fengwei /* no need to invalidate: a not-present page won't be cached */ 45583bd786f7SYin Fengwei update_mmu_cache_range(vmf, vma, addr, vmf->pte, nr); 45593bb97794SKirill A. Shutemov } 45603bb97794SKirill A. Shutemov 4561f46f2adeSPeter Xu static bool vmf_pte_changed(struct vm_fault *vmf) 4562f46f2adeSPeter Xu { 4563f46f2adeSPeter Xu if (vmf->flags & FAULT_FLAG_ORIG_PTE_VALID) 4564c33c7948SRyan Roberts return !pte_same(ptep_get(vmf->pte), vmf->orig_pte); 4565f46f2adeSPeter Xu 4566c33c7948SRyan Roberts return !pte_none(ptep_get(vmf->pte)); 4567f46f2adeSPeter Xu } 4568f46f2adeSPeter Xu 45699118c0cbSJan Kara /** 45709118c0cbSJan Kara * finish_fault - finish page fault once we have prepared the page to fault 45719118c0cbSJan Kara * 45729118c0cbSJan Kara * @vmf: structure describing the fault 45739118c0cbSJan Kara * 45749118c0cbSJan Kara * This function handles all that is needed to finish a page fault once the 45759118c0cbSJan Kara * page to fault in is prepared. It handles locking of PTEs, inserts PTE for 45769118c0cbSJan Kara * given page, adds reverse page mapping, handles memcg charges and LRU 4577a862f68aSMike Rapoport * addition. 45789118c0cbSJan Kara * 45799118c0cbSJan Kara * The function expects the page to be locked and on success it consumes a 45809118c0cbSJan Kara * reference of a page being mapped (for the PTE which maps it). 4581a862f68aSMike Rapoport * 4582a862f68aSMike Rapoport * Return: %0 on success, %VM_FAULT_ code in case of error. 45839118c0cbSJan Kara */ 45842b740303SSouptick Joarder vm_fault_t finish_fault(struct vm_fault *vmf) 45859118c0cbSJan Kara { 4586f9ce0be7SKirill A. Shutemov struct vm_area_struct *vma = vmf->vma; 45879118c0cbSJan Kara struct page *page; 4588f9ce0be7SKirill A. Shutemov vm_fault_t ret; 45899118c0cbSJan Kara 45909118c0cbSJan Kara /* Did we COW the page? */ 4591f9ce0be7SKirill A. Shutemov if ((vmf->flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) 45929118c0cbSJan Kara page = vmf->cow_page; 45939118c0cbSJan Kara else 45949118c0cbSJan Kara page = vmf->page; 45956b31d595SMichal Hocko 45966b31d595SMichal Hocko /* 45976b31d595SMichal Hocko * check even for read faults because we might have lost our CoWed 45986b31d595SMichal Hocko * page 45996b31d595SMichal Hocko */ 4600f9ce0be7SKirill A. Shutemov if (!(vma->vm_flags & VM_SHARED)) { 4601f9ce0be7SKirill A. Shutemov ret = check_stable_address_space(vma->vm_mm); 4602f9ce0be7SKirill A. Shutemov if (ret) 4603f9ce0be7SKirill A. Shutemov return ret; 4604f9ce0be7SKirill A. Shutemov } 4605f9ce0be7SKirill A. Shutemov 4606f9ce0be7SKirill A. Shutemov if (pmd_none(*vmf->pmd)) { 4607f9ce0be7SKirill A. Shutemov if (PageTransCompound(page)) { 4608f9ce0be7SKirill A. Shutemov ret = do_set_pmd(vmf, page); 4609f9ce0be7SKirill A. Shutemov if (ret != VM_FAULT_FALLBACK) 4610f9ce0be7SKirill A. Shutemov return ret; 4611f9ce0be7SKirill A. Shutemov } 4612f9ce0be7SKirill A. Shutemov 461303c4f204SQi Zheng if (vmf->prealloc_pte) 461403c4f204SQi Zheng pmd_install(vma->vm_mm, vmf->pmd, &vmf->prealloc_pte); 461503c4f204SQi Zheng else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd))) 4616f9ce0be7SKirill A. Shutemov return VM_FAULT_OOM; 4617f9ce0be7SKirill A. Shutemov } 4618f9ce0be7SKirill A. Shutemov 4619f9ce0be7SKirill A. Shutemov vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 4620f9ce0be7SKirill A. Shutemov vmf->address, &vmf->ptl); 46213db82b93SHugh Dickins if (!vmf->pte) 46223db82b93SHugh Dickins return VM_FAULT_NOPAGE; 4623f9ce0be7SKirill A. Shutemov 462470427f6eSSergei Antonov /* Re-check under ptl */ 462570427f6eSSergei Antonov if (likely(!vmf_pte_changed(vmf))) { 46263bd786f7SYin Fengwei struct folio *folio = page_folio(page); 462770427f6eSSergei Antonov 46283bd786f7SYin Fengwei set_pte_range(vmf, folio, page, 1, vmf->address); 462970427f6eSSergei Antonov ret = 0; 463070427f6eSSergei Antonov } else { 4631f9ce0be7SKirill A. Shutemov update_mmu_tlb(vma, vmf->address, vmf->pte); 463270427f6eSSergei Antonov ret = VM_FAULT_NOPAGE; 463370427f6eSSergei Antonov } 463470427f6eSSergei Antonov 46359118c0cbSJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 46369118c0cbSJan Kara return ret; 46379118c0cbSJan Kara } 46389118c0cbSJan Kara 463953d36a56SLorenzo Stoakes static unsigned long fault_around_pages __read_mostly = 464053d36a56SLorenzo Stoakes 65536 >> PAGE_SHIFT; 4641a9b0f861SKirill A. Shutemov 46421592eef0SKirill A. Shutemov #ifdef CONFIG_DEBUG_FS 4643a9b0f861SKirill A. Shutemov static int fault_around_bytes_get(void *data, u64 *val) 46441592eef0SKirill A. Shutemov { 464553d36a56SLorenzo Stoakes *val = fault_around_pages << PAGE_SHIFT; 46461592eef0SKirill A. Shutemov return 0; 46471592eef0SKirill A. Shutemov } 46481592eef0SKirill A. Shutemov 4649b4903d6eSAndrey Ryabinin /* 4650da391d64SWilliam Kucharski * fault_around_bytes must be rounded down to the nearest page order as it's 4651da391d64SWilliam Kucharski * what do_fault_around() expects to see. 4652b4903d6eSAndrey Ryabinin */ 4653a9b0f861SKirill A. Shutemov static int fault_around_bytes_set(void *data, u64 val) 46541592eef0SKirill A. Shutemov { 4655a9b0f861SKirill A. Shutemov if (val / PAGE_SIZE > PTRS_PER_PTE) 46561592eef0SKirill A. Shutemov return -EINVAL; 465753d36a56SLorenzo Stoakes 465853d36a56SLorenzo Stoakes /* 465953d36a56SLorenzo Stoakes * The minimum value is 1 page, however this results in no fault-around 466053d36a56SLorenzo Stoakes * at all. See should_fault_around(). 466153d36a56SLorenzo Stoakes */ 466253d36a56SLorenzo Stoakes fault_around_pages = max(rounddown_pow_of_two(val) >> PAGE_SHIFT, 1UL); 466353d36a56SLorenzo Stoakes 46641592eef0SKirill A. Shutemov return 0; 46651592eef0SKirill A. Shutemov } 46660a1345f8SYevgen Pronenko DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, 4667a9b0f861SKirill A. Shutemov fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); 46681592eef0SKirill A. Shutemov 46691592eef0SKirill A. Shutemov static int __init fault_around_debugfs(void) 46701592eef0SKirill A. Shutemov { 4671d9f7979cSGreg Kroah-Hartman debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, 4672a9b0f861SKirill A. Shutemov &fault_around_bytes_fops); 46731592eef0SKirill A. Shutemov return 0; 46741592eef0SKirill A. Shutemov } 46751592eef0SKirill A. Shutemov late_initcall(fault_around_debugfs); 46761592eef0SKirill A. Shutemov #endif 46778c6e50b0SKirill A. Shutemov 46781fdb412bSKirill A. Shutemov /* 46791fdb412bSKirill A. Shutemov * do_fault_around() tries to map few pages around the fault address. The hope 46801fdb412bSKirill A. Shutemov * is that the pages will be needed soon and this will lower the number of 46811fdb412bSKirill A. Shutemov * faults to handle. 46821fdb412bSKirill A. Shutemov * 46831fdb412bSKirill A. Shutemov * It uses vm_ops->map_pages() to map the pages, which skips the page if it's 46841fdb412bSKirill A. Shutemov * not ready to be mapped: not up-to-date, locked, etc. 46851fdb412bSKirill A. Shutemov * 46869042599eSLorenzo Stoakes * This function doesn't cross VMA or page table boundaries, in order to call 46879042599eSLorenzo Stoakes * map_pages() and acquire a PTE lock only once. 46881fdb412bSKirill A. Shutemov * 468953d36a56SLorenzo Stoakes * fault_around_pages defines how many pages we'll try to map. 4690da391d64SWilliam Kucharski * do_fault_around() expects it to be set to a power of two less than or equal 4691da391d64SWilliam Kucharski * to PTRS_PER_PTE. 46921fdb412bSKirill A. Shutemov * 4693da391d64SWilliam Kucharski * The virtual address of the area that we map is naturally aligned to 469453d36a56SLorenzo Stoakes * fault_around_pages * PAGE_SIZE rounded down to the machine page size 4695da391d64SWilliam Kucharski * (and therefore to page order). This way it's easier to guarantee 4696da391d64SWilliam Kucharski * that we don't cross page table boundaries. 46971fdb412bSKirill A. Shutemov */ 46982b740303SSouptick Joarder static vm_fault_t do_fault_around(struct vm_fault *vmf) 46998c6e50b0SKirill A. Shutemov { 470053d36a56SLorenzo Stoakes pgoff_t nr_pages = READ_ONCE(fault_around_pages); 47019042599eSLorenzo Stoakes pgoff_t pte_off = pte_index(vmf->address); 47029042599eSLorenzo Stoakes /* The page offset of vmf->address within the VMA. */ 47039042599eSLorenzo Stoakes pgoff_t vma_off = vmf->pgoff - vmf->vma->vm_pgoff; 47049042599eSLorenzo Stoakes pgoff_t from_pte, to_pte; 470558ef47efSMatthew Wilcox (Oracle) vm_fault_t ret; 47068c6e50b0SKirill A. Shutemov 47079042599eSLorenzo Stoakes /* The PTE offset of the start address, clamped to the VMA. */ 47089042599eSLorenzo Stoakes from_pte = max(ALIGN_DOWN(pte_off, nr_pages), 47099042599eSLorenzo Stoakes pte_off - min(pte_off, vma_off)); 4710aecd6f44SKirill A. Shutemov 47119042599eSLorenzo Stoakes /* The PTE offset of the end address, clamped to the VMA and PTE. */ 47129042599eSLorenzo Stoakes to_pte = min3(from_pte + nr_pages, (pgoff_t)PTRS_PER_PTE, 47139042599eSLorenzo Stoakes pte_off + vma_pages(vmf->vma) - vma_off) - 1; 47148c6e50b0SKirill A. Shutemov 471582b0f8c3SJan Kara if (pmd_none(*vmf->pmd)) { 47164cf58924SJoel Fernandes (Google) vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm); 471782b0f8c3SJan Kara if (!vmf->prealloc_pte) 4718f9ce0be7SKirill A. Shutemov return VM_FAULT_OOM; 47198c6e50b0SKirill A. Shutemov } 47208c6e50b0SKirill A. Shutemov 472158ef47efSMatthew Wilcox (Oracle) rcu_read_lock(); 472258ef47efSMatthew Wilcox (Oracle) ret = vmf->vma->vm_ops->map_pages(vmf, 47239042599eSLorenzo Stoakes vmf->pgoff + from_pte - pte_off, 47249042599eSLorenzo Stoakes vmf->pgoff + to_pte - pte_off); 472558ef47efSMatthew Wilcox (Oracle) rcu_read_unlock(); 472658ef47efSMatthew Wilcox (Oracle) 472758ef47efSMatthew Wilcox (Oracle) return ret; 47287267ec00SKirill A. Shutemov } 47297267ec00SKirill A. Shutemov 47309c28a205SPeter Xu /* Return true if we should do read fault-around, false otherwise */ 47319c28a205SPeter Xu static inline bool should_fault_around(struct vm_fault *vmf) 47329c28a205SPeter Xu { 47339c28a205SPeter Xu /* No ->map_pages? No way to fault around... */ 47349c28a205SPeter Xu if (!vmf->vma->vm_ops->map_pages) 47359c28a205SPeter Xu return false; 47369c28a205SPeter Xu 47379c28a205SPeter Xu if (uffd_disable_fault_around(vmf->vma)) 47389c28a205SPeter Xu return false; 47399c28a205SPeter Xu 474053d36a56SLorenzo Stoakes /* A single page implies no faulting 'around' at all. */ 474153d36a56SLorenzo Stoakes return fault_around_pages > 1; 47429c28a205SPeter Xu } 47439c28a205SPeter Xu 47442b740303SSouptick Joarder static vm_fault_t do_read_fault(struct vm_fault *vmf) 4745e655fb29SKirill A. Shutemov { 47462b740303SSouptick Joarder vm_fault_t ret = 0; 474722d1e68fSSidhartha Kumar struct folio *folio; 47488c6e50b0SKirill A. Shutemov 47498c6e50b0SKirill A. Shutemov /* 47508c6e50b0SKirill A. Shutemov * Let's call ->map_pages() first and use ->fault() as fallback 47518c6e50b0SKirill A. Shutemov * if page by the offset is not ready to be mapped (cold cache or 47528c6e50b0SKirill A. Shutemov * something). 47538c6e50b0SKirill A. Shutemov */ 47549c28a205SPeter Xu if (should_fault_around(vmf)) { 47550721ec8bSJan Kara ret = do_fault_around(vmf); 47567267ec00SKirill A. Shutemov if (ret) 47577267ec00SKirill A. Shutemov return ret; 47588c6e50b0SKirill A. Shutemov } 4759e655fb29SKirill A. Shutemov 476012214ebaSMatthew Wilcox (Oracle) ret = vmf_can_call_fault(vmf); 476112214ebaSMatthew Wilcox (Oracle) if (ret) 476212214ebaSMatthew Wilcox (Oracle) return ret; 4763f5617ffeSMatthew Wilcox (Oracle) 4764936ca80dSJan Kara ret = __do_fault(vmf); 4765e655fb29SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4766e655fb29SKirill A. Shutemov return ret; 4767e655fb29SKirill A. Shutemov 47689118c0cbSJan Kara ret |= finish_fault(vmf); 476922d1e68fSSidhartha Kumar folio = page_folio(vmf->page); 477022d1e68fSSidhartha Kumar folio_unlock(folio); 47717267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 477222d1e68fSSidhartha Kumar folio_put(folio); 4773e655fb29SKirill A. Shutemov return ret; 4774e655fb29SKirill A. Shutemov } 4775e655fb29SKirill A. Shutemov 47762b740303SSouptick Joarder static vm_fault_t do_cow_fault(struct vm_fault *vmf) 4777ec47c3b9SKirill A. Shutemov { 477882b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4779e4621e70SKefeng Wang struct folio *folio; 47802b740303SSouptick Joarder vm_fault_t ret; 4781ec47c3b9SKirill A. Shutemov 47824de8c93aSMatthew Wilcox (Oracle) ret = vmf_can_call_fault(vmf); 47834de8c93aSMatthew Wilcox (Oracle) if (!ret) 47844de8c93aSMatthew Wilcox (Oracle) ret = vmf_anon_prepare(vmf); 47854de8c93aSMatthew Wilcox (Oracle) if (ret) 47864de8c93aSMatthew Wilcox (Oracle) return ret; 4787ec47c3b9SKirill A. Shutemov 4788e4621e70SKefeng Wang folio = folio_prealloc(vma->vm_mm, vma, vmf->address, false); 4789e4621e70SKefeng Wang if (!folio) 4790ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4791ec47c3b9SKirill A. Shutemov 4792e4621e70SKefeng Wang vmf->cow_page = &folio->page; 4793ec47c3b9SKirill A. Shutemov 4794936ca80dSJan Kara ret = __do_fault(vmf); 4795ec47c3b9SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4796ec47c3b9SKirill A. Shutemov goto uncharge_out; 47973917048dSJan Kara if (ret & VM_FAULT_DONE_COW) 47983917048dSJan Kara return ret; 4799ec47c3b9SKirill A. Shutemov 4800936ca80dSJan Kara copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); 4801e4621e70SKefeng Wang __folio_mark_uptodate(folio); 4802ec47c3b9SKirill A. Shutemov 48039118c0cbSJan Kara ret |= finish_fault(vmf); 4804936ca80dSJan Kara unlock_page(vmf->page); 4805936ca80dSJan Kara put_page(vmf->page); 48067267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 48077267ec00SKirill A. Shutemov goto uncharge_out; 4808ec47c3b9SKirill A. Shutemov return ret; 4809ec47c3b9SKirill A. Shutemov uncharge_out: 4810e4621e70SKefeng Wang folio_put(folio); 4811ec47c3b9SKirill A. Shutemov return ret; 4812ec47c3b9SKirill A. Shutemov } 4813ec47c3b9SKirill A. Shutemov 48142b740303SSouptick Joarder static vm_fault_t do_shared_fault(struct vm_fault *vmf) 48151da177e4SLinus Torvalds { 481682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 48172b740303SSouptick Joarder vm_fault_t ret, tmp; 48186f609b7eSSidhartha Kumar struct folio *folio; 48191d65f86dSKAMEZAWA Hiroyuki 48204ed43798SMatthew Wilcox (Oracle) ret = vmf_can_call_fault(vmf); 48214ed43798SMatthew Wilcox (Oracle) if (ret) 48224ed43798SMatthew Wilcox (Oracle) return ret; 482354cb8821SNick Piggin 4824936ca80dSJan Kara ret = __do_fault(vmf); 48257eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4826f0c6d4d2SKirill A. Shutemov return ret; 48271da177e4SLinus Torvalds 48286f609b7eSSidhartha Kumar folio = page_folio(vmf->page); 48296f609b7eSSidhartha Kumar 48301da177e4SLinus Torvalds /* 4831f0c6d4d2SKirill A. Shutemov * Check if the backing address space wants to know that the page is 4832f0c6d4d2SKirill A. Shutemov * about to become writable 48331da177e4SLinus Torvalds */ 4834fb09a464SKirill A. Shutemov if (vma->vm_ops->page_mkwrite) { 48356f609b7eSSidhartha Kumar folio_unlock(folio); 483686aa6998SSidhartha Kumar tmp = do_page_mkwrite(vmf, folio); 4837fb09a464SKirill A. Shutemov if (unlikely(!tmp || 4838fb09a464SKirill A. Shutemov (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 48396f609b7eSSidhartha Kumar folio_put(folio); 4840f0c6d4d2SKirill A. Shutemov return tmp; 484169676147SMark Fasheh } 4842d0217ac0SNick Piggin } 4843fb09a464SKirill A. Shutemov 48449118c0cbSJan Kara ret |= finish_fault(vmf); 48457267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | 48467267ec00SKirill A. Shutemov VM_FAULT_RETRY))) { 48476f609b7eSSidhartha Kumar folio_unlock(folio); 48486f609b7eSSidhartha Kumar folio_put(folio); 4849f0c6d4d2SKirill A. Shutemov return ret; 48509637a5efSDavid Howells } 4851d00806b1SNick Piggin 485289b15332SJohannes Weiner ret |= fault_dirty_shared_page(vmf); 4853b827e496SNick Piggin return ret; 485454cb8821SNick Piggin } 4855d00806b1SNick Piggin 48569a95f3cfSPaul Cassella /* 4857c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 48589a95f3cfSPaul Cassella * but allow concurrent faults). 4859c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 48609138e47eSMatthew Wilcox (Oracle) * return value. See filemap_fault() and __folio_lock_or_retry(). 4861c1e8d7c6SMichel Lespinasse * If mmap_lock is released, vma may become invalid (for example 4862fc8efd2dSJan Stancek * by other thread calling munmap()). 48639a95f3cfSPaul Cassella */ 48642b740303SSouptick Joarder static vm_fault_t do_fault(struct vm_fault *vmf) 486554cb8821SNick Piggin { 486682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4867fc8efd2dSJan Stancek struct mm_struct *vm_mm = vma->vm_mm; 48682b740303SSouptick Joarder vm_fault_t ret; 486954cb8821SNick Piggin 4870ff09d7ecSAneesh Kumar K.V /* 4871ff09d7ecSAneesh Kumar K.V * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND 4872ff09d7ecSAneesh Kumar K.V */ 4873ff09d7ecSAneesh Kumar K.V if (!vma->vm_ops->fault) { 48743db82b93SHugh Dickins vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, 48753db82b93SHugh Dickins vmf->address, &vmf->ptl); 48763db82b93SHugh Dickins if (unlikely(!vmf->pte)) 4877b0b9b3dfSHugh Dickins ret = VM_FAULT_SIGBUS; 4878ff09d7ecSAneesh Kumar K.V else { 4879ff09d7ecSAneesh Kumar K.V /* 4880ff09d7ecSAneesh Kumar K.V * Make sure this is not a temporary clearing of pte 4881ff09d7ecSAneesh Kumar K.V * by holding ptl and checking again. A R/M/W update 4882ff09d7ecSAneesh Kumar K.V * of pte involves: take ptl, clearing the pte so that 4883ff09d7ecSAneesh Kumar K.V * we don't have concurrent modification by hardware 4884ff09d7ecSAneesh Kumar K.V * followed by an update. 4885ff09d7ecSAneesh Kumar K.V */ 4886c33c7948SRyan Roberts if (unlikely(pte_none(ptep_get(vmf->pte)))) 4887ff09d7ecSAneesh Kumar K.V ret = VM_FAULT_SIGBUS; 4888ff09d7ecSAneesh Kumar K.V else 4889ff09d7ecSAneesh Kumar K.V ret = VM_FAULT_NOPAGE; 4890ff09d7ecSAneesh Kumar K.V 4891ff09d7ecSAneesh Kumar K.V pte_unmap_unlock(vmf->pte, vmf->ptl); 4892ff09d7ecSAneesh Kumar K.V } 4893ff09d7ecSAneesh Kumar K.V } else if (!(vmf->flags & FAULT_FLAG_WRITE)) 4894b0b9b3dfSHugh Dickins ret = do_read_fault(vmf); 4895b0b9b3dfSHugh Dickins else if (!(vma->vm_flags & VM_SHARED)) 4896b0b9b3dfSHugh Dickins ret = do_cow_fault(vmf); 4897b0b9b3dfSHugh Dickins else 4898b0b9b3dfSHugh Dickins ret = do_shared_fault(vmf); 4899b0b9b3dfSHugh Dickins 4900b0b9b3dfSHugh Dickins /* preallocated pagetable is unused: free it */ 4901b0b9b3dfSHugh Dickins if (vmf->prealloc_pte) { 4902fc8efd2dSJan Stancek pte_free(vm_mm, vmf->prealloc_pte); 49037f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 4904b0b9b3dfSHugh Dickins } 4905b0b9b3dfSHugh Dickins return ret; 490654cb8821SNick Piggin } 490754cb8821SNick Piggin 4908cda6d936SKefeng Wang int numa_migrate_prep(struct folio *folio, struct vm_area_struct *vma, 4909f4c0d836SYang Shi unsigned long addr, int page_nid, int *flags) 49109532fec1SMel Gorman { 4911cda6d936SKefeng Wang folio_get(folio); 49129532fec1SMel Gorman 4913fc137c0dSRaghavendra K T /* Record the current PID acceesing VMA */ 4914fc137c0dSRaghavendra K T vma_set_access_pid_bit(vma); 4915fc137c0dSRaghavendra K T 49169532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS); 491704bb2f94SRik van Riel if (page_nid == numa_node_id()) { 49189532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); 491904bb2f94SRik van Riel *flags |= TNF_FAULT_LOCAL; 492004bb2f94SRik van Riel } 49219532fec1SMel Gorman 492275c70128SKefeng Wang return mpol_misplaced(folio, vma, addr); 49239532fec1SMel Gorman } 49249532fec1SMel Gorman 49252b740303SSouptick Joarder static vm_fault_t do_numa_page(struct vm_fault *vmf) 4926d10e63f2SMel Gorman { 492782b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 49286695cf68SKefeng Wang struct folio *folio = NULL; 49296695cf68SKefeng Wang int nid = NUMA_NO_NODE; 49306a56ccbcSDavid Hildenbrand bool writable = false; 493190572890SPeter Zijlstra int last_cpupid; 4932cbee9f88SPeter Zijlstra int target_nid; 493304a86453SAneesh Kumar K.V pte_t pte, old_pte; 49346688cc05SPeter Zijlstra int flags = 0; 4935d10e63f2SMel Gorman 4936d10e63f2SMel Gorman /* 4937d10e63f2SMel Gorman * The "pte" at this point cannot be used safely without 4938d10e63f2SMel Gorman * validation through pte_unmap_same(). It's of NUMA type but 4939d10e63f2SMel Gorman * the pfn may be screwed if the read is non atomic. 4940d10e63f2SMel Gorman */ 494182b0f8c3SJan Kara spin_lock(vmf->ptl); 4942c33c7948SRyan Roberts if (unlikely(!pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 494382b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 49444daae3b4SMel Gorman goto out; 49454daae3b4SMel Gorman } 49464daae3b4SMel Gorman 4947b99a342dSHuang Ying /* Get the normal PTE */ 4948b99a342dSHuang Ying old_pte = ptep_get(vmf->pte); 494904a86453SAneesh Kumar K.V pte = pte_modify(old_pte, vma->vm_page_prot); 4950d10e63f2SMel Gorman 49516a56ccbcSDavid Hildenbrand /* 49526a56ccbcSDavid Hildenbrand * Detect now whether the PTE could be writable; this information 49536a56ccbcSDavid Hildenbrand * is only valid while holding the PT lock. 49546a56ccbcSDavid Hildenbrand */ 49556a56ccbcSDavid Hildenbrand writable = pte_write(pte); 49566a56ccbcSDavid Hildenbrand if (!writable && vma_wants_manual_pte_write_upgrade(vma) && 49576a56ccbcSDavid Hildenbrand can_change_pte_writable(vma, vmf->address, pte)) 49586a56ccbcSDavid Hildenbrand writable = true; 49596a56ccbcSDavid Hildenbrand 49606695cf68SKefeng Wang folio = vm_normal_folio(vma, vmf->address, pte); 49616695cf68SKefeng Wang if (!folio || folio_is_zone_device(folio)) 4962b99a342dSHuang Ying goto out_map; 4963d10e63f2SMel Gorman 4964e81c4802SKirill A. Shutemov /* TODO: handle PTE-mapped THP */ 49656695cf68SKefeng Wang if (folio_test_large(folio)) 4966b99a342dSHuang Ying goto out_map; 4967e81c4802SKirill A. Shutemov 49686688cc05SPeter Zijlstra /* 4969bea66fbdSMel Gorman * Avoid grouping on RO pages in general. RO pages shouldn't hurt as 4970bea66fbdSMel Gorman * much anyway since they can be in shared cache state. This misses 4971bea66fbdSMel Gorman * the case where a mapping is writable but the process never writes 4972bea66fbdSMel Gorman * to it but pte_write gets cleared during protection updates and 4973bea66fbdSMel Gorman * pte_dirty has unpredictable behaviour between PTE scan updates, 4974bea66fbdSMel Gorman * background writeback, dirty balancing and application behaviour. 49756688cc05SPeter Zijlstra */ 49766a56ccbcSDavid Hildenbrand if (!writable) 49776688cc05SPeter Zijlstra flags |= TNF_NO_GROUP; 49786688cc05SPeter Zijlstra 4979dabe1d99SRik van Riel /* 49806695cf68SKefeng Wang * Flag if the folio is shared between multiple address spaces. This 4981dabe1d99SRik van Riel * is later used when determining whether to group tasks together 4982dabe1d99SRik van Riel */ 49836695cf68SKefeng Wang if (folio_estimated_sharers(folio) > 1 && (vma->vm_flags & VM_SHARED)) 4984dabe1d99SRik van Riel flags |= TNF_SHARED; 4985dabe1d99SRik van Riel 49866695cf68SKefeng Wang nid = folio_nid(folio); 498733024536SHuang Ying /* 498833024536SHuang Ying * For memory tiering mode, cpupid of slow memory page is used 498933024536SHuang Ying * to record page access time. So use default value. 499033024536SHuang Ying */ 499133024536SHuang Ying if ((sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) && 49926695cf68SKefeng Wang !node_is_toptier(nid)) 499333024536SHuang Ying last_cpupid = (-1 & LAST_CPUPID_MASK); 499433024536SHuang Ying else 499567b33e3fSKefeng Wang last_cpupid = folio_last_cpupid(folio); 4996cda6d936SKefeng Wang target_nid = numa_migrate_prep(folio, vma, vmf->address, nid, &flags); 499798fa15f3SAnshuman Khandual if (target_nid == NUMA_NO_NODE) { 49986695cf68SKefeng Wang folio_put(folio); 4999b99a342dSHuang Ying goto out_map; 50004daae3b4SMel Gorman } 5001b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 50026a56ccbcSDavid Hildenbrand writable = false; 50034daae3b4SMel Gorman 50044daae3b4SMel Gorman /* Migrate to the requested node */ 50056695cf68SKefeng Wang if (migrate_misplaced_folio(folio, vma, target_nid)) { 50066695cf68SKefeng Wang nid = target_nid; 50076688cc05SPeter Zijlstra flags |= TNF_MIGRATED; 5008b99a342dSHuang Ying } else { 5009074c2381SMel Gorman flags |= TNF_MIGRATE_FAIL; 5010c7ad0880SHugh Dickins vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 5011c7ad0880SHugh Dickins vmf->address, &vmf->ptl); 5012c7ad0880SHugh Dickins if (unlikely(!vmf->pte)) 5013c7ad0880SHugh Dickins goto out; 5014c33c7948SRyan Roberts if (unlikely(!pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 5015b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 5016b99a342dSHuang Ying goto out; 5017b99a342dSHuang Ying } 5018b99a342dSHuang Ying goto out_map; 5019b99a342dSHuang Ying } 50204daae3b4SMel Gorman 50214daae3b4SMel Gorman out: 50226695cf68SKefeng Wang if (nid != NUMA_NO_NODE) 50236695cf68SKefeng Wang task_numa_fault(last_cpupid, nid, 1, flags); 5024d10e63f2SMel Gorman return 0; 5025b99a342dSHuang Ying out_map: 5026b99a342dSHuang Ying /* 5027b99a342dSHuang Ying * Make it present again, depending on how arch implements 5028b99a342dSHuang Ying * non-accessible ptes, some can allow access by kernel mode. 5029b99a342dSHuang Ying */ 5030b99a342dSHuang Ying old_pte = ptep_modify_prot_start(vma, vmf->address, vmf->pte); 5031b99a342dSHuang Ying pte = pte_modify(old_pte, vma->vm_page_prot); 5032b99a342dSHuang Ying pte = pte_mkyoung(pte); 50336a56ccbcSDavid Hildenbrand if (writable) 5034161e393cSRick Edgecombe pte = pte_mkwrite(pte, vma); 5035b99a342dSHuang Ying ptep_modify_prot_commit(vma, vmf->address, vmf->pte, old_pte, pte); 50365003a2bdSMatthew Wilcox (Oracle) update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1); 5037b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 5038b99a342dSHuang Ying goto out; 5039d10e63f2SMel Gorman } 5040d10e63f2SMel Gorman 50412b740303SSouptick Joarder static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf) 5042b96375f7SMatthew Wilcox { 50438f5fd0e1SMatthew Wilcox (Oracle) struct vm_area_struct *vma = vmf->vma; 50448f5fd0e1SMatthew Wilcox (Oracle) if (vma_is_anonymous(vma)) 504582b0f8c3SJan Kara return do_huge_pmd_anonymous_page(vmf); 504640d49a3cSMatthew Wilcox (Oracle) if (vma->vm_ops->huge_fault) 50471d024e7aSMatthew Wilcox (Oracle) return vma->vm_ops->huge_fault(vmf, PMD_ORDER); 5048b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 5049b96375f7SMatthew Wilcox } 5050b96375f7SMatthew Wilcox 5051183f24aaSGeert Uytterhoeven /* `inline' is required to avoid gcc 4.1.2 build error */ 50525db4f15cSYang Shi static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf) 5053b96375f7SMatthew Wilcox { 50548f5fd0e1SMatthew Wilcox (Oracle) struct vm_area_struct *vma = vmf->vma; 5055c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 5056aea06577SDavid Hildenbrand vm_fault_t ret; 5057c89357e2SDavid Hildenbrand 50588f5fd0e1SMatthew Wilcox (Oracle) if (vma_is_anonymous(vma)) { 5059c89357e2SDavid Hildenbrand if (likely(!unshare) && 5060d61ea1cbSPeter Xu userfaultfd_huge_pmd_wp(vma, vmf->orig_pmd)) { 5061d61ea1cbSPeter Xu if (userfaultfd_wp_async(vmf->vma)) 5062d61ea1cbSPeter Xu goto split; 5063529b930bSAndrea Arcangeli return handle_userfault(vmf, VM_UFFD_WP); 5064d61ea1cbSPeter Xu } 50655db4f15cSYang Shi return do_huge_pmd_wp_page(vmf); 5066529b930bSAndrea Arcangeli } 5067af9e4d5fSKirill A. Shutemov 50688f5fd0e1SMatthew Wilcox (Oracle) if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 50698f5fd0e1SMatthew Wilcox (Oracle) if (vma->vm_ops->huge_fault) { 50701d024e7aSMatthew Wilcox (Oracle) ret = vma->vm_ops->huge_fault(vmf, PMD_ORDER); 5071327e9fd4SThomas Hellstrom (VMware) if (!(ret & VM_FAULT_FALLBACK)) 5072327e9fd4SThomas Hellstrom (VMware) return ret; 5073327e9fd4SThomas Hellstrom (VMware) } 5074aea06577SDavid Hildenbrand } 5075327e9fd4SThomas Hellstrom (VMware) 5076d61ea1cbSPeter Xu split: 5077327e9fd4SThomas Hellstrom (VMware) /* COW or write-notify handled on pte level: split pmd. */ 50788f5fd0e1SMatthew Wilcox (Oracle) __split_huge_pmd(vma, vmf->pmd, vmf->address, false, NULL); 5079af9e4d5fSKirill A. Shutemov 5080b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 5081b96375f7SMatthew Wilcox } 5082b96375f7SMatthew Wilcox 50832b740303SSouptick Joarder static vm_fault_t create_huge_pud(struct vm_fault *vmf) 5084a00cc7d9SMatthew Wilcox { 5085327e9fd4SThomas Hellstrom (VMware) #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ 5086327e9fd4SThomas Hellstrom (VMware) defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) 5087c4fd825eSMatthew Wilcox (Oracle) struct vm_area_struct *vma = vmf->vma; 5088a00cc7d9SMatthew Wilcox /* No support for anonymous transparent PUD pages yet */ 5089c4fd825eSMatthew Wilcox (Oracle) if (vma_is_anonymous(vma)) 509014c99d65SGowans, James return VM_FAULT_FALLBACK; 509140d49a3cSMatthew Wilcox (Oracle) if (vma->vm_ops->huge_fault) 50921d024e7aSMatthew Wilcox (Oracle) return vma->vm_ops->huge_fault(vmf, PUD_ORDER); 509314c99d65SGowans, James #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 509414c99d65SGowans, James return VM_FAULT_FALLBACK; 509514c99d65SGowans, James } 509614c99d65SGowans, James 509714c99d65SGowans, James static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) 509814c99d65SGowans, James { 509914c99d65SGowans, James #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ 510014c99d65SGowans, James defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) 5101c4fd825eSMatthew Wilcox (Oracle) struct vm_area_struct *vma = vmf->vma; 5102aea06577SDavid Hildenbrand vm_fault_t ret; 5103aea06577SDavid Hildenbrand 510414c99d65SGowans, James /* No support for anonymous transparent PUD pages yet */ 5105c4fd825eSMatthew Wilcox (Oracle) if (vma_is_anonymous(vma)) 5106327e9fd4SThomas Hellstrom (VMware) goto split; 5107c4fd825eSMatthew Wilcox (Oracle) if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 5108c4fd825eSMatthew Wilcox (Oracle) if (vma->vm_ops->huge_fault) { 51091d024e7aSMatthew Wilcox (Oracle) ret = vma->vm_ops->huge_fault(vmf, PUD_ORDER); 5110327e9fd4SThomas Hellstrom (VMware) if (!(ret & VM_FAULT_FALLBACK)) 5111327e9fd4SThomas Hellstrom (VMware) return ret; 5112327e9fd4SThomas Hellstrom (VMware) } 5113aea06577SDavid Hildenbrand } 5114327e9fd4SThomas Hellstrom (VMware) split: 5115327e9fd4SThomas Hellstrom (VMware) /* COW or write-notify not handled on PUD level: split pud.*/ 5116c4fd825eSMatthew Wilcox (Oracle) __split_huge_pud(vma, vmf->pud, vmf->address); 511714c99d65SGowans, James #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 5118a00cc7d9SMatthew Wilcox return VM_FAULT_FALLBACK; 5119a00cc7d9SMatthew Wilcox } 5120a00cc7d9SMatthew Wilcox 51211da177e4SLinus Torvalds /* 51221da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 51231da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 51241da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 51251da177e4SLinus Torvalds * 51261da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 51271da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 51281da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 51291da177e4SLinus Torvalds * 5130c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, but allow 51317267ec00SKirill A. Shutemov * concurrent faults). 51329a95f3cfSPaul Cassella * 5133c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our return value. 51349138e47eSMatthew Wilcox (Oracle) * See filemap_fault() and __folio_lock_or_retry(). 51351da177e4SLinus Torvalds */ 51362b740303SSouptick Joarder static vm_fault_t handle_pte_fault(struct vm_fault *vmf) 51371da177e4SLinus Torvalds { 51381da177e4SLinus Torvalds pte_t entry; 51391da177e4SLinus Torvalds 514082b0f8c3SJan Kara if (unlikely(pmd_none(*vmf->pmd))) { 51417267ec00SKirill A. Shutemov /* 51427267ec00SKirill A. Shutemov * Leave __pte_alloc() until later: because vm_ops->fault may 51437267ec00SKirill A. Shutemov * want to allocate huge page, and if we expose page table 51447267ec00SKirill A. Shutemov * for an instant, it will be difficult to retract from 51457267ec00SKirill A. Shutemov * concurrent faults and from rmap lookups. 51467267ec00SKirill A. Shutemov */ 514782b0f8c3SJan Kara vmf->pte = NULL; 5148f46f2adeSPeter Xu vmf->flags &= ~FAULT_FLAG_ORIG_PTE_VALID; 51497267ec00SKirill A. Shutemov } else { 5150f9ce0be7SKirill A. Shutemov /* 51517267ec00SKirill A. Shutemov * A regular pmd is established and it can't morph into a huge 5152c7ad0880SHugh Dickins * pmd by anon khugepaged, since that takes mmap_lock in write 5153c7ad0880SHugh Dickins * mode; but shmem or file collapse to THP could still morph 5154c7ad0880SHugh Dickins * it into a huge pmd: just retry later if so. 51557267ec00SKirill A. Shutemov */ 5156c7ad0880SHugh Dickins vmf->pte = pte_offset_map_nolock(vmf->vma->vm_mm, vmf->pmd, 5157c7ad0880SHugh Dickins vmf->address, &vmf->ptl); 5158c7ad0880SHugh Dickins if (unlikely(!vmf->pte)) 5159c7ad0880SHugh Dickins return 0; 516026e1a0c3SHugh Dickins vmf->orig_pte = ptep_get_lockless(vmf->pte); 5161f46f2adeSPeter Xu vmf->flags |= FAULT_FLAG_ORIG_PTE_VALID; 51627267ec00SKirill A. Shutemov 51632994302bSJan Kara if (pte_none(vmf->orig_pte)) { 516482b0f8c3SJan Kara pte_unmap(vmf->pte); 516582b0f8c3SJan Kara vmf->pte = NULL; 51667267ec00SKirill A. Shutemov } 51677267ec00SKirill A. Shutemov } 51687267ec00SKirill A. Shutemov 51692bad466cSPeter Xu if (!vmf->pte) 51702bad466cSPeter Xu return do_pte_missing(vmf); 51717267ec00SKirill A. Shutemov 51722994302bSJan Kara if (!pte_present(vmf->orig_pte)) 51732994302bSJan Kara return do_swap_page(vmf); 51741da177e4SLinus Torvalds 51752994302bSJan Kara if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) 51762994302bSJan Kara return do_numa_page(vmf); 5177d10e63f2SMel Gorman 517882b0f8c3SJan Kara spin_lock(vmf->ptl); 51792994302bSJan Kara entry = vmf->orig_pte; 5180c33c7948SRyan Roberts if (unlikely(!pte_same(ptep_get(vmf->pte), entry))) { 51817df67697SBibo Mao update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 51828f4e2101SHugh Dickins goto unlock; 51837df67697SBibo Mao } 5184c89357e2SDavid Hildenbrand if (vmf->flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) { 5185f6f37321SLinus Torvalds if (!pte_write(entry)) 51862994302bSJan Kara return do_wp_page(vmf); 5187c89357e2SDavid Hildenbrand else if (likely(vmf->flags & FAULT_FLAG_WRITE)) 51881da177e4SLinus Torvalds entry = pte_mkdirty(entry); 51891da177e4SLinus Torvalds } 51901da177e4SLinus Torvalds entry = pte_mkyoung(entry); 519182b0f8c3SJan Kara if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, 519282b0f8c3SJan Kara vmf->flags & FAULT_FLAG_WRITE)) { 51935003a2bdSMatthew Wilcox (Oracle) update_mmu_cache_range(vmf, vmf->vma, vmf->address, 51945003a2bdSMatthew Wilcox (Oracle) vmf->pte, 1); 51951a44e149SAndrea Arcangeli } else { 5196b7333b58SYang Shi /* Skip spurious TLB flush for retried page fault */ 5197b7333b58SYang Shi if (vmf->flags & FAULT_FLAG_TRIED) 5198b7333b58SYang Shi goto unlock; 51991a44e149SAndrea Arcangeli /* 52001a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 52011a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 52021a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 52031a44e149SAndrea Arcangeli * with threads. 52041a44e149SAndrea Arcangeli */ 520582b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) 520699c29133SGerald Schaefer flush_tlb_fix_spurious_fault(vmf->vma, vmf->address, 520799c29133SGerald Schaefer vmf->pte); 52081a44e149SAndrea Arcangeli } 52098f4e2101SHugh Dickins unlock: 521082b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 521183c54070SNick Piggin return 0; 52121da177e4SLinus Torvalds } 52131da177e4SLinus Torvalds 52141da177e4SLinus Torvalds /* 52154ec31152SMatthew Wilcox (Oracle) * On entry, we hold either the VMA lock or the mmap_lock 52164ec31152SMatthew Wilcox (Oracle) * (FAULT_FLAG_VMA_LOCK tells you which). If VM_FAULT_RETRY is set in 52174ec31152SMatthew Wilcox (Oracle) * the result, the mmap_lock is not held on exit. See filemap_fault() 52184ec31152SMatthew Wilcox (Oracle) * and __folio_lock_or_retry(). 52191da177e4SLinus Torvalds */ 52202b740303SSouptick Joarder static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma, 52212b740303SSouptick Joarder unsigned long address, unsigned int flags) 52221da177e4SLinus Torvalds { 522382b0f8c3SJan Kara struct vm_fault vmf = { 5224bae473a4SKirill A. Shutemov .vma = vma, 52251a29d85eSJan Kara .address = address & PAGE_MASK, 5226824ddc60SNadav Amit .real_address = address, 5227bae473a4SKirill A. Shutemov .flags = flags, 52280721ec8bSJan Kara .pgoff = linear_page_index(vma, address), 5229667240e0SJan Kara .gfp_mask = __get_fault_gfp_mask(vma), 5230bae473a4SKirill A. Shutemov }; 5231dcddffd4SKirill A. Shutemov struct mm_struct *mm = vma->vm_mm; 52327da4e2cbSYang Shi unsigned long vm_flags = vma->vm_flags; 52331da177e4SLinus Torvalds pgd_t *pgd; 5234c2febafcSKirill A. Shutemov p4d_t *p4d; 52352b740303SSouptick Joarder vm_fault_t ret; 52361da177e4SLinus Torvalds 52371da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 5238c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, address); 5239c2febafcSKirill A. Shutemov if (!p4d) 5240c2febafcSKirill A. Shutemov return VM_FAULT_OOM; 5241a00cc7d9SMatthew Wilcox 5242c2febafcSKirill A. Shutemov vmf.pud = pud_alloc(mm, p4d, address); 5243a00cc7d9SMatthew Wilcox if (!vmf.pud) 5244c74df32cSHugh Dickins return VM_FAULT_OOM; 5245625110b5SThomas Hellstrom retry_pud: 52467da4e2cbSYang Shi if (pud_none(*vmf.pud) && 52473485b883SRyan Roberts thp_vma_allowable_order(vma, vm_flags, false, true, true, PUD_ORDER)) { 5248a00cc7d9SMatthew Wilcox ret = create_huge_pud(&vmf); 5249a00cc7d9SMatthew Wilcox if (!(ret & VM_FAULT_FALLBACK)) 5250a00cc7d9SMatthew Wilcox return ret; 5251a00cc7d9SMatthew Wilcox } else { 5252a00cc7d9SMatthew Wilcox pud_t orig_pud = *vmf.pud; 5253a00cc7d9SMatthew Wilcox 5254a00cc7d9SMatthew Wilcox barrier(); 5255a00cc7d9SMatthew Wilcox if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { 5256a00cc7d9SMatthew Wilcox 5257c89357e2SDavid Hildenbrand /* 5258c89357e2SDavid Hildenbrand * TODO once we support anonymous PUDs: NUMA case and 5259c89357e2SDavid Hildenbrand * FAULT_FLAG_UNSHARE handling. 5260c89357e2SDavid Hildenbrand */ 5261c89357e2SDavid Hildenbrand if ((flags & FAULT_FLAG_WRITE) && !pud_write(orig_pud)) { 5262a00cc7d9SMatthew Wilcox ret = wp_huge_pud(&vmf, orig_pud); 5263a00cc7d9SMatthew Wilcox if (!(ret & VM_FAULT_FALLBACK)) 5264a00cc7d9SMatthew Wilcox return ret; 5265a00cc7d9SMatthew Wilcox } else { 5266a00cc7d9SMatthew Wilcox huge_pud_set_accessed(&vmf, orig_pud); 5267a00cc7d9SMatthew Wilcox return 0; 5268a00cc7d9SMatthew Wilcox } 5269a00cc7d9SMatthew Wilcox } 5270a00cc7d9SMatthew Wilcox } 5271a00cc7d9SMatthew Wilcox 5272a00cc7d9SMatthew Wilcox vmf.pmd = pmd_alloc(mm, vmf.pud, address); 527382b0f8c3SJan Kara if (!vmf.pmd) 5274c74df32cSHugh Dickins return VM_FAULT_OOM; 5275625110b5SThomas Hellstrom 5276625110b5SThomas Hellstrom /* Huge pud page fault raced with pmd_alloc? */ 5277625110b5SThomas Hellstrom if (pud_trans_unstable(vmf.pud)) 5278625110b5SThomas Hellstrom goto retry_pud; 5279625110b5SThomas Hellstrom 52807da4e2cbSYang Shi if (pmd_none(*vmf.pmd) && 52813485b883SRyan Roberts thp_vma_allowable_order(vma, vm_flags, false, true, true, PMD_ORDER)) { 5282a2d58167SDave Jiang ret = create_huge_pmd(&vmf); 5283c0292554SKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 5284c0292554SKirill A. Shutemov return ret; 528571e3aac0SAndrea Arcangeli } else { 528626e1a0c3SHugh Dickins vmf.orig_pmd = pmdp_get_lockless(vmf.pmd); 52871f1d06c3SDavid Rientjes 52885db4f15cSYang Shi if (unlikely(is_swap_pmd(vmf.orig_pmd))) { 528984c3fc4eSZi Yan VM_BUG_ON(thp_migration_supported() && 52905db4f15cSYang Shi !is_pmd_migration_entry(vmf.orig_pmd)); 52915db4f15cSYang Shi if (is_pmd_migration_entry(vmf.orig_pmd)) 529284c3fc4eSZi Yan pmd_migration_entry_wait(mm, vmf.pmd); 529384c3fc4eSZi Yan return 0; 529484c3fc4eSZi Yan } 52955db4f15cSYang Shi if (pmd_trans_huge(vmf.orig_pmd) || pmd_devmap(vmf.orig_pmd)) { 52965db4f15cSYang Shi if (pmd_protnone(vmf.orig_pmd) && vma_is_accessible(vma)) 52975db4f15cSYang Shi return do_huge_pmd_numa_page(&vmf); 5298d10e63f2SMel Gorman 5299c89357e2SDavid Hildenbrand if ((flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) && 5300c89357e2SDavid Hildenbrand !pmd_write(vmf.orig_pmd)) { 53015db4f15cSYang Shi ret = wp_huge_pmd(&vmf); 53029845cbbdSKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 53031f1d06c3SDavid Rientjes return ret; 5304a1dd450bSWill Deacon } else { 53055db4f15cSYang Shi huge_pmd_set_accessed(&vmf); 530671e3aac0SAndrea Arcangeli return 0; 530771e3aac0SAndrea Arcangeli } 530871e3aac0SAndrea Arcangeli } 53099845cbbdSKirill A. Shutemov } 531071e3aac0SAndrea Arcangeli 531182b0f8c3SJan Kara return handle_pte_fault(&vmf); 53121da177e4SLinus Torvalds } 53131da177e4SLinus Torvalds 5314bce617edSPeter Xu /** 5315f0953a1bSIngo Molnar * mm_account_fault - Do page fault accounting 5316809ef83cSYang Li * @mm: mm from which memcg should be extracted. It can be NULL. 5317bce617edSPeter Xu * @regs: the pt_regs struct pointer. When set to NULL, will skip accounting 5318bce617edSPeter Xu * of perf event counters, but we'll still do the per-task accounting to 5319bce617edSPeter Xu * the task who triggered this page fault. 5320bce617edSPeter Xu * @address: the faulted address. 5321bce617edSPeter Xu * @flags: the fault flags. 5322bce617edSPeter Xu * @ret: the fault retcode. 5323bce617edSPeter Xu * 5324f0953a1bSIngo Molnar * This will take care of most of the page fault accounting. Meanwhile, it 5325bce617edSPeter Xu * will also include the PERF_COUNT_SW_PAGE_FAULTS_[MAJ|MIN] perf counter 5326f0953a1bSIngo Molnar * updates. However, note that the handling of PERF_COUNT_SW_PAGE_FAULTS should 5327bce617edSPeter Xu * still be in per-arch page fault handlers at the entry of page fault. 5328bce617edSPeter Xu */ 532953156443SSuren Baghdasaryan static inline void mm_account_fault(struct mm_struct *mm, struct pt_regs *regs, 5330bce617edSPeter Xu unsigned long address, unsigned int flags, 5331bce617edSPeter Xu vm_fault_t ret) 5332bce617edSPeter Xu { 5333bce617edSPeter Xu bool major; 5334bce617edSPeter Xu 533553156443SSuren Baghdasaryan /* Incomplete faults will be accounted upon completion. */ 533653156443SSuren Baghdasaryan if (ret & VM_FAULT_RETRY) 533753156443SSuren Baghdasaryan return; 533853156443SSuren Baghdasaryan 5339bce617edSPeter Xu /* 534053156443SSuren Baghdasaryan * To preserve the behavior of older kernels, PGFAULT counters record 534153156443SSuren Baghdasaryan * both successful and failed faults, as opposed to perf counters, 534253156443SSuren Baghdasaryan * which ignore failed cases. 5343bce617edSPeter Xu */ 534453156443SSuren Baghdasaryan count_vm_event(PGFAULT); 534553156443SSuren Baghdasaryan count_memcg_event_mm(mm, PGFAULT); 534653156443SSuren Baghdasaryan 534753156443SSuren Baghdasaryan /* 534853156443SSuren Baghdasaryan * Do not account for unsuccessful faults (e.g. when the address wasn't 534953156443SSuren Baghdasaryan * valid). That includes arch_vma_access_permitted() failing before 535053156443SSuren Baghdasaryan * reaching here. So this is not a "this many hardware page faults" 535153156443SSuren Baghdasaryan * counter. We should use the hw profiling for that. 535253156443SSuren Baghdasaryan */ 535353156443SSuren Baghdasaryan if (ret & VM_FAULT_ERROR) 5354bce617edSPeter Xu return; 5355bce617edSPeter Xu 5356bce617edSPeter Xu /* 5357bce617edSPeter Xu * We define the fault as a major fault when the final successful fault 5358bce617edSPeter Xu * is VM_FAULT_MAJOR, or if it retried (which implies that we couldn't 5359bce617edSPeter Xu * handle it immediately previously). 5360bce617edSPeter Xu */ 5361bce617edSPeter Xu major = (ret & VM_FAULT_MAJOR) || (flags & FAULT_FLAG_TRIED); 5362bce617edSPeter Xu 5363a2beb5f1SPeter Xu if (major) 5364a2beb5f1SPeter Xu current->maj_flt++; 5365a2beb5f1SPeter Xu else 5366a2beb5f1SPeter Xu current->min_flt++; 5367a2beb5f1SPeter Xu 5368bce617edSPeter Xu /* 5369a2beb5f1SPeter Xu * If the fault is done for GUP, regs will be NULL. We only do the 5370a2beb5f1SPeter Xu * accounting for the per thread fault counters who triggered the 5371a2beb5f1SPeter Xu * fault, and we skip the perf event updates. 5372bce617edSPeter Xu */ 5373bce617edSPeter Xu if (!regs) 5374bce617edSPeter Xu return; 5375bce617edSPeter Xu 5376a2beb5f1SPeter Xu if (major) 5377bce617edSPeter Xu perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 5378a2beb5f1SPeter Xu else 5379bce617edSPeter Xu perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 5380bce617edSPeter Xu } 5381bce617edSPeter Xu 5382ec1c86b2SYu Zhao #ifdef CONFIG_LRU_GEN 5383ec1c86b2SYu Zhao static void lru_gen_enter_fault(struct vm_area_struct *vma) 5384ec1c86b2SYu Zhao { 53858788f678SYu Zhao /* the LRU algorithm only applies to accesses with recency */ 53868788f678SYu Zhao current->in_lru_fault = vma_has_recency(vma); 5387ec1c86b2SYu Zhao } 5388ec1c86b2SYu Zhao 5389ec1c86b2SYu Zhao static void lru_gen_exit_fault(void) 5390ec1c86b2SYu Zhao { 5391ec1c86b2SYu Zhao current->in_lru_fault = false; 5392ec1c86b2SYu Zhao } 5393ec1c86b2SYu Zhao #else 5394ec1c86b2SYu Zhao static void lru_gen_enter_fault(struct vm_area_struct *vma) 5395ec1c86b2SYu Zhao { 5396ec1c86b2SYu Zhao } 5397ec1c86b2SYu Zhao 5398ec1c86b2SYu Zhao static void lru_gen_exit_fault(void) 5399ec1c86b2SYu Zhao { 5400ec1c86b2SYu Zhao } 5401ec1c86b2SYu Zhao #endif /* CONFIG_LRU_GEN */ 5402ec1c86b2SYu Zhao 5403cdc5021cSDavid Hildenbrand static vm_fault_t sanitize_fault_flags(struct vm_area_struct *vma, 5404cdc5021cSDavid Hildenbrand unsigned int *flags) 5405cdc5021cSDavid Hildenbrand { 5406cdc5021cSDavid Hildenbrand if (unlikely(*flags & FAULT_FLAG_UNSHARE)) { 5407cdc5021cSDavid Hildenbrand if (WARN_ON_ONCE(*flags & FAULT_FLAG_WRITE)) 5408cdc5021cSDavid Hildenbrand return VM_FAULT_SIGSEGV; 5409cdc5021cSDavid Hildenbrand /* 5410cdc5021cSDavid Hildenbrand * FAULT_FLAG_UNSHARE only applies to COW mappings. Let's 5411cdc5021cSDavid Hildenbrand * just treat it like an ordinary read-fault otherwise. 5412cdc5021cSDavid Hildenbrand */ 5413cdc5021cSDavid Hildenbrand if (!is_cow_mapping(vma->vm_flags)) 5414cdc5021cSDavid Hildenbrand *flags &= ~FAULT_FLAG_UNSHARE; 541579881fedSDavid Hildenbrand } else if (*flags & FAULT_FLAG_WRITE) { 541679881fedSDavid Hildenbrand /* Write faults on read-only mappings are impossible ... */ 541779881fedSDavid Hildenbrand if (WARN_ON_ONCE(!(vma->vm_flags & VM_MAYWRITE))) 541879881fedSDavid Hildenbrand return VM_FAULT_SIGSEGV; 541979881fedSDavid Hildenbrand /* ... and FOLL_FORCE only applies to COW mappings. */ 542079881fedSDavid Hildenbrand if (WARN_ON_ONCE(!(vma->vm_flags & VM_WRITE) && 542179881fedSDavid Hildenbrand !is_cow_mapping(vma->vm_flags))) 542279881fedSDavid Hildenbrand return VM_FAULT_SIGSEGV; 5423cdc5021cSDavid Hildenbrand } 54244089eef0SSuren Baghdasaryan #ifdef CONFIG_PER_VMA_LOCK 54254089eef0SSuren Baghdasaryan /* 54264089eef0SSuren Baghdasaryan * Per-VMA locks can't be used with FAULT_FLAG_RETRY_NOWAIT because of 54274089eef0SSuren Baghdasaryan * the assumption that lock is dropped on VM_FAULT_RETRY. 54284089eef0SSuren Baghdasaryan */ 54294089eef0SSuren Baghdasaryan if (WARN_ON_ONCE((*flags & 54304089eef0SSuren Baghdasaryan (FAULT_FLAG_VMA_LOCK | FAULT_FLAG_RETRY_NOWAIT)) == 54314089eef0SSuren Baghdasaryan (FAULT_FLAG_VMA_LOCK | FAULT_FLAG_RETRY_NOWAIT))) 54324089eef0SSuren Baghdasaryan return VM_FAULT_SIGSEGV; 54334089eef0SSuren Baghdasaryan #endif 54344089eef0SSuren Baghdasaryan 5435cdc5021cSDavid Hildenbrand return 0; 5436cdc5021cSDavid Hildenbrand } 5437cdc5021cSDavid Hildenbrand 54389a95f3cfSPaul Cassella /* 54399a95f3cfSPaul Cassella * By the time we get here, we already hold the mm semaphore 54409a95f3cfSPaul Cassella * 5441c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 54429138e47eSMatthew Wilcox (Oracle) * return value. See filemap_fault() and __folio_lock_or_retry(). 54439a95f3cfSPaul Cassella */ 54442b740303SSouptick Joarder vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address, 5445bce617edSPeter Xu unsigned int flags, struct pt_regs *regs) 5446519e5247SJohannes Weiner { 544753156443SSuren Baghdasaryan /* If the fault handler drops the mmap_lock, vma may be freed */ 544853156443SSuren Baghdasaryan struct mm_struct *mm = vma->vm_mm; 54492b740303SSouptick Joarder vm_fault_t ret; 5450519e5247SJohannes Weiner 5451519e5247SJohannes Weiner __set_current_state(TASK_RUNNING); 5452519e5247SJohannes Weiner 5453cdc5021cSDavid Hildenbrand ret = sanitize_fault_flags(vma, &flags); 5454cdc5021cSDavid Hildenbrand if (ret) 545553156443SSuren Baghdasaryan goto out; 5456cdc5021cSDavid Hildenbrand 5457de0c799bSLaurent Dufour if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, 5458de0c799bSLaurent Dufour flags & FAULT_FLAG_INSTRUCTION, 545953156443SSuren Baghdasaryan flags & FAULT_FLAG_REMOTE)) { 546053156443SSuren Baghdasaryan ret = VM_FAULT_SIGSEGV; 546153156443SSuren Baghdasaryan goto out; 546253156443SSuren Baghdasaryan } 5463de0c799bSLaurent Dufour 5464519e5247SJohannes Weiner /* 5465519e5247SJohannes Weiner * Enable the memcg OOM handling for faults triggered in user 5466519e5247SJohannes Weiner * space. Kernel faults are handled more gracefully. 5467519e5247SJohannes Weiner */ 5468519e5247SJohannes Weiner if (flags & FAULT_FLAG_USER) 546929ef680aSMichal Hocko mem_cgroup_enter_user_fault(); 5470519e5247SJohannes Weiner 5471ec1c86b2SYu Zhao lru_gen_enter_fault(vma); 5472ec1c86b2SYu Zhao 5473bae473a4SKirill A. Shutemov if (unlikely(is_vm_hugetlb_page(vma))) 5474bae473a4SKirill A. Shutemov ret = hugetlb_fault(vma->vm_mm, vma, address, flags); 5475bae473a4SKirill A. Shutemov else 5476dcddffd4SKirill A. Shutemov ret = __handle_mm_fault(vma, address, flags); 5477519e5247SJohannes Weiner 5478ec1c86b2SYu Zhao lru_gen_exit_fault(); 5479ec1c86b2SYu Zhao 548049426420SJohannes Weiner if (flags & FAULT_FLAG_USER) { 548129ef680aSMichal Hocko mem_cgroup_exit_user_fault(); 548249426420SJohannes Weiner /* 548349426420SJohannes Weiner * The task may have entered a memcg OOM situation but 548449426420SJohannes Weiner * if the allocation error was handled gracefully (no 548549426420SJohannes Weiner * VM_FAULT_OOM), there is no need to kill anything. 548649426420SJohannes Weiner * Just clean up the OOM state peacefully. 548749426420SJohannes Weiner */ 548849426420SJohannes Weiner if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) 548949426420SJohannes Weiner mem_cgroup_oom_synchronize(false); 549049426420SJohannes Weiner } 549153156443SSuren Baghdasaryan out: 549253156443SSuren Baghdasaryan mm_account_fault(mm, regs, address, flags, ret); 5493bce617edSPeter Xu 5494519e5247SJohannes Weiner return ret; 5495519e5247SJohannes Weiner } 5496e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(handle_mm_fault); 5497519e5247SJohannes Weiner 5498c2508ec5SLinus Torvalds #ifdef CONFIG_LOCK_MM_AND_FIND_VMA 5499c2508ec5SLinus Torvalds #include <linux/extable.h> 5500c2508ec5SLinus Torvalds 5501c2508ec5SLinus Torvalds static inline bool get_mmap_lock_carefully(struct mm_struct *mm, struct pt_regs *regs) 5502c2508ec5SLinus Torvalds { 55034542057eSLinus Torvalds if (likely(mmap_read_trylock(mm))) 5504c2508ec5SLinus Torvalds return true; 5505c2508ec5SLinus Torvalds 5506c2508ec5SLinus Torvalds if (regs && !user_mode(regs)) { 55078fa50708SJiaxun Yang unsigned long ip = exception_ip(regs); 5508c2508ec5SLinus Torvalds if (!search_exception_tables(ip)) 5509c2508ec5SLinus Torvalds return false; 5510c2508ec5SLinus Torvalds } 5511c2508ec5SLinus Torvalds 5512eda00472SLinus Torvalds return !mmap_read_lock_killable(mm); 5513c2508ec5SLinus Torvalds } 5514c2508ec5SLinus Torvalds 5515c2508ec5SLinus Torvalds static inline bool mmap_upgrade_trylock(struct mm_struct *mm) 5516c2508ec5SLinus Torvalds { 5517c2508ec5SLinus Torvalds /* 5518c2508ec5SLinus Torvalds * We don't have this operation yet. 5519c2508ec5SLinus Torvalds * 5520c2508ec5SLinus Torvalds * It should be easy enough to do: it's basically a 5521c2508ec5SLinus Torvalds * atomic_long_try_cmpxchg_acquire() 5522c2508ec5SLinus Torvalds * from RWSEM_READER_BIAS -> RWSEM_WRITER_LOCKED, but 5523c2508ec5SLinus Torvalds * it also needs the proper lockdep magic etc. 5524c2508ec5SLinus Torvalds */ 5525c2508ec5SLinus Torvalds return false; 5526c2508ec5SLinus Torvalds } 5527c2508ec5SLinus Torvalds 5528c2508ec5SLinus Torvalds static inline bool upgrade_mmap_lock_carefully(struct mm_struct *mm, struct pt_regs *regs) 5529c2508ec5SLinus Torvalds { 5530c2508ec5SLinus Torvalds mmap_read_unlock(mm); 5531c2508ec5SLinus Torvalds if (regs && !user_mode(regs)) { 55328fa50708SJiaxun Yang unsigned long ip = exception_ip(regs); 5533c2508ec5SLinus Torvalds if (!search_exception_tables(ip)) 5534c2508ec5SLinus Torvalds return false; 5535c2508ec5SLinus Torvalds } 5536eda00472SLinus Torvalds return !mmap_write_lock_killable(mm); 5537c2508ec5SLinus Torvalds } 5538c2508ec5SLinus Torvalds 5539c2508ec5SLinus Torvalds /* 5540c2508ec5SLinus Torvalds * Helper for page fault handling. 5541c2508ec5SLinus Torvalds * 5542c2508ec5SLinus Torvalds * This is kind of equivalend to "mmap_read_lock()" followed 5543c2508ec5SLinus Torvalds * by "find_extend_vma()", except it's a lot more careful about 5544c2508ec5SLinus Torvalds * the locking (and will drop the lock on failure). 5545c2508ec5SLinus Torvalds * 5546c2508ec5SLinus Torvalds * For example, if we have a kernel bug that causes a page 5547c2508ec5SLinus Torvalds * fault, we don't want to just use mmap_read_lock() to get 5548c2508ec5SLinus Torvalds * the mm lock, because that would deadlock if the bug were 5549c2508ec5SLinus Torvalds * to happen while we're holding the mm lock for writing. 5550c2508ec5SLinus Torvalds * 5551c2508ec5SLinus Torvalds * So this checks the exception tables on kernel faults in 5552c2508ec5SLinus Torvalds * order to only do this all for instructions that are actually 5553c2508ec5SLinus Torvalds * expected to fault. 5554c2508ec5SLinus Torvalds * 5555c2508ec5SLinus Torvalds * We can also actually take the mm lock for writing if we 5556c2508ec5SLinus Torvalds * need to extend the vma, which helps the VM layer a lot. 5557c2508ec5SLinus Torvalds */ 5558c2508ec5SLinus Torvalds struct vm_area_struct *lock_mm_and_find_vma(struct mm_struct *mm, 5559c2508ec5SLinus Torvalds unsigned long addr, struct pt_regs *regs) 5560c2508ec5SLinus Torvalds { 5561c2508ec5SLinus Torvalds struct vm_area_struct *vma; 5562c2508ec5SLinus Torvalds 5563c2508ec5SLinus Torvalds if (!get_mmap_lock_carefully(mm, regs)) 5564c2508ec5SLinus Torvalds return NULL; 5565c2508ec5SLinus Torvalds 5566c2508ec5SLinus Torvalds vma = find_vma(mm, addr); 5567c2508ec5SLinus Torvalds if (likely(vma && (vma->vm_start <= addr))) 5568c2508ec5SLinus Torvalds return vma; 5569c2508ec5SLinus Torvalds 5570c2508ec5SLinus Torvalds /* 5571c2508ec5SLinus Torvalds * Well, dang. We might still be successful, but only 5572c2508ec5SLinus Torvalds * if we can extend a vma to do so. 5573c2508ec5SLinus Torvalds */ 5574c2508ec5SLinus Torvalds if (!vma || !(vma->vm_flags & VM_GROWSDOWN)) { 5575c2508ec5SLinus Torvalds mmap_read_unlock(mm); 5576c2508ec5SLinus Torvalds return NULL; 5577c2508ec5SLinus Torvalds } 5578c2508ec5SLinus Torvalds 5579c2508ec5SLinus Torvalds /* 5580c2508ec5SLinus Torvalds * We can try to upgrade the mmap lock atomically, 5581c2508ec5SLinus Torvalds * in which case we can continue to use the vma 5582c2508ec5SLinus Torvalds * we already looked up. 5583c2508ec5SLinus Torvalds * 5584c2508ec5SLinus Torvalds * Otherwise we'll have to drop the mmap lock and 5585c2508ec5SLinus Torvalds * re-take it, and also look up the vma again, 5586c2508ec5SLinus Torvalds * re-checking it. 5587c2508ec5SLinus Torvalds */ 5588c2508ec5SLinus Torvalds if (!mmap_upgrade_trylock(mm)) { 5589c2508ec5SLinus Torvalds if (!upgrade_mmap_lock_carefully(mm, regs)) 5590c2508ec5SLinus Torvalds return NULL; 5591c2508ec5SLinus Torvalds 5592c2508ec5SLinus Torvalds vma = find_vma(mm, addr); 5593c2508ec5SLinus Torvalds if (!vma) 5594c2508ec5SLinus Torvalds goto fail; 5595c2508ec5SLinus Torvalds if (vma->vm_start <= addr) 5596c2508ec5SLinus Torvalds goto success; 5597c2508ec5SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 5598c2508ec5SLinus Torvalds goto fail; 5599c2508ec5SLinus Torvalds } 5600c2508ec5SLinus Torvalds 56018d7071afSLinus Torvalds if (expand_stack_locked(vma, addr)) 5602c2508ec5SLinus Torvalds goto fail; 5603c2508ec5SLinus Torvalds 5604c2508ec5SLinus Torvalds success: 5605c2508ec5SLinus Torvalds mmap_write_downgrade(mm); 5606c2508ec5SLinus Torvalds return vma; 5607c2508ec5SLinus Torvalds 5608c2508ec5SLinus Torvalds fail: 5609c2508ec5SLinus Torvalds mmap_write_unlock(mm); 5610c2508ec5SLinus Torvalds return NULL; 5611c2508ec5SLinus Torvalds } 5612c2508ec5SLinus Torvalds #endif 5613c2508ec5SLinus Torvalds 561450ee3253SSuren Baghdasaryan #ifdef CONFIG_PER_VMA_LOCK 561550ee3253SSuren Baghdasaryan /* 561650ee3253SSuren Baghdasaryan * Lookup and lock a VMA under RCU protection. Returned VMA is guaranteed to be 561750ee3253SSuren Baghdasaryan * stable and not isolated. If the VMA is not found or is being modified the 561850ee3253SSuren Baghdasaryan * function returns NULL. 561950ee3253SSuren Baghdasaryan */ 562050ee3253SSuren Baghdasaryan struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm, 562150ee3253SSuren Baghdasaryan unsigned long address) 562250ee3253SSuren Baghdasaryan { 562350ee3253SSuren Baghdasaryan MA_STATE(mas, &mm->mm_mt, address, address); 562450ee3253SSuren Baghdasaryan struct vm_area_struct *vma; 562550ee3253SSuren Baghdasaryan 562650ee3253SSuren Baghdasaryan rcu_read_lock(); 562750ee3253SSuren Baghdasaryan retry: 562850ee3253SSuren Baghdasaryan vma = mas_walk(&mas); 562950ee3253SSuren Baghdasaryan if (!vma) 563050ee3253SSuren Baghdasaryan goto inval; 563150ee3253SSuren Baghdasaryan 563250ee3253SSuren Baghdasaryan if (!vma_start_read(vma)) 563350ee3253SSuren Baghdasaryan goto inval; 563450ee3253SSuren Baghdasaryan 5635444eeb17SSuren Baghdasaryan /* 5636657b5146SJann Horn * find_mergeable_anon_vma uses adjacent vmas which are not locked. 5637657b5146SJann Horn * This check must happen after vma_start_read(); otherwise, a 5638657b5146SJann Horn * concurrent mremap() with MREMAP_DONTUNMAP could dissociate the VMA 5639657b5146SJann Horn * from its anon_vma. 5640657b5146SJann Horn */ 564129a22b9eSSuren Baghdasaryan if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) 5642657b5146SJann Horn goto inval_end_read; 5643444eeb17SSuren Baghdasaryan 564450ee3253SSuren Baghdasaryan /* Check since vm_start/vm_end might change before we lock the VMA */ 5645657b5146SJann Horn if (unlikely(address < vma->vm_start || address >= vma->vm_end)) 5646657b5146SJann Horn goto inval_end_read; 564750ee3253SSuren Baghdasaryan 564850ee3253SSuren Baghdasaryan /* Check if the VMA got isolated after we found it */ 564950ee3253SSuren Baghdasaryan if (vma->detached) { 565050ee3253SSuren Baghdasaryan vma_end_read(vma); 565152f23865SSuren Baghdasaryan count_vm_vma_lock_event(VMA_LOCK_MISS); 565250ee3253SSuren Baghdasaryan /* The area was replaced with another one */ 565350ee3253SSuren Baghdasaryan goto retry; 565450ee3253SSuren Baghdasaryan } 565550ee3253SSuren Baghdasaryan 565650ee3253SSuren Baghdasaryan rcu_read_unlock(); 565750ee3253SSuren Baghdasaryan return vma; 5658657b5146SJann Horn 5659657b5146SJann Horn inval_end_read: 5660657b5146SJann Horn vma_end_read(vma); 566150ee3253SSuren Baghdasaryan inval: 566250ee3253SSuren Baghdasaryan rcu_read_unlock(); 566352f23865SSuren Baghdasaryan count_vm_vma_lock_event(VMA_LOCK_ABORT); 566450ee3253SSuren Baghdasaryan return NULL; 566550ee3253SSuren Baghdasaryan } 566650ee3253SSuren Baghdasaryan #endif /* CONFIG_PER_VMA_LOCK */ 566750ee3253SSuren Baghdasaryan 566890eceff1SKirill A. Shutemov #ifndef __PAGETABLE_P4D_FOLDED 566990eceff1SKirill A. Shutemov /* 567090eceff1SKirill A. Shutemov * Allocate p4d page table. 567190eceff1SKirill A. Shutemov * We've already handled the fast-path in-line. 567290eceff1SKirill A. Shutemov */ 567390eceff1SKirill A. Shutemov int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 567490eceff1SKirill A. Shutemov { 567590eceff1SKirill A. Shutemov p4d_t *new = p4d_alloc_one(mm, address); 567690eceff1SKirill A. Shutemov if (!new) 567790eceff1SKirill A. Shutemov return -ENOMEM; 567890eceff1SKirill A. Shutemov 567990eceff1SKirill A. Shutemov spin_lock(&mm->page_table_lock); 5680ed33b5a6SQi Zheng if (pgd_present(*pgd)) { /* Another has populated it */ 568190eceff1SKirill A. Shutemov p4d_free(mm, new); 5682ed33b5a6SQi Zheng } else { 5683ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 568490eceff1SKirill A. Shutemov pgd_populate(mm, pgd, new); 5685ed33b5a6SQi Zheng } 568690eceff1SKirill A. Shutemov spin_unlock(&mm->page_table_lock); 568790eceff1SKirill A. Shutemov return 0; 568890eceff1SKirill A. Shutemov } 568990eceff1SKirill A. Shutemov #endif /* __PAGETABLE_P4D_FOLDED */ 569090eceff1SKirill A. Shutemov 56911da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 56921da177e4SLinus Torvalds /* 56931da177e4SLinus Torvalds * Allocate page upper directory. 5694872fec16SHugh Dickins * We've already handled the fast-path in-line. 56951da177e4SLinus Torvalds */ 5696c2febafcSKirill A. Shutemov int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) 56971da177e4SLinus Torvalds { 5698c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 5699c74df32cSHugh Dickins if (!new) 57001bb3630eSHugh Dickins return -ENOMEM; 57011da177e4SLinus Torvalds 5702872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 5703b4e98d9aSKirill A. Shutemov if (!p4d_present(*p4d)) { 5704b4e98d9aSKirill A. Shutemov mm_inc_nr_puds(mm); 5705ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 5706c2febafcSKirill A. Shutemov p4d_populate(mm, p4d, new); 5707b4e98d9aSKirill A. Shutemov } else /* Another has populated it */ 5708c2febafcSKirill A. Shutemov pud_free(mm, new); 5709872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 57101bb3630eSHugh Dickins return 0; 57111da177e4SLinus Torvalds } 57121da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 57131da177e4SLinus Torvalds 57141da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 57151da177e4SLinus Torvalds /* 57161da177e4SLinus Torvalds * Allocate page middle directory. 5717872fec16SHugh Dickins * We've already handled the fast-path in-line. 57181da177e4SLinus Torvalds */ 57191bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 57201da177e4SLinus Torvalds { 5721a00cc7d9SMatthew Wilcox spinlock_t *ptl; 5722c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 5723c74df32cSHugh Dickins if (!new) 57241bb3630eSHugh Dickins return -ENOMEM; 57251da177e4SLinus Torvalds 5726a00cc7d9SMatthew Wilcox ptl = pud_lock(mm, pud); 5727dc6c9a35SKirill A. Shutemov if (!pud_present(*pud)) { 5728dc6c9a35SKirill A. Shutemov mm_inc_nr_pmds(mm); 5729ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 57301da177e4SLinus Torvalds pud_populate(mm, pud, new); 5731ed33b5a6SQi Zheng } else { /* Another has populated it */ 57325e541973SBenjamin Herrenschmidt pmd_free(mm, new); 5733ed33b5a6SQi Zheng } 5734a00cc7d9SMatthew Wilcox spin_unlock(ptl); 57351bb3630eSHugh Dickins return 0; 57361da177e4SLinus Torvalds } 57371da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 57381da177e4SLinus Torvalds 57393b6748e2SJohannes Weiner /** 57409fd6dad1SPaolo Bonzini * follow_pte - look up PTE at a user virtual address 57419fd6dad1SPaolo Bonzini * @mm: the mm_struct of the target address space 57429fd6dad1SPaolo Bonzini * @address: user virtual address 57439fd6dad1SPaolo Bonzini * @ptepp: location to store found PTE 57449fd6dad1SPaolo Bonzini * @ptlp: location to store the lock for the PTE 57459fd6dad1SPaolo Bonzini * 57469fd6dad1SPaolo Bonzini * On a successful return, the pointer to the PTE is stored in @ptepp; 57479fd6dad1SPaolo Bonzini * the corresponding lock is taken and its location is stored in @ptlp. 57489fd6dad1SPaolo Bonzini * The contents of the PTE are only stable until @ptlp is released; 57499fd6dad1SPaolo Bonzini * any further use, if any, must be protected against invalidation 57509fd6dad1SPaolo Bonzini * with MMU notifiers. 57519fd6dad1SPaolo Bonzini * 57529fd6dad1SPaolo Bonzini * Only IO mappings and raw PFN mappings are allowed. The mmap semaphore 57539fd6dad1SPaolo Bonzini * should be taken for read. 57549fd6dad1SPaolo Bonzini * 57559fd6dad1SPaolo Bonzini * KVM uses this function. While it is arguably less bad than ``follow_pfn``, 57569fd6dad1SPaolo Bonzini * it is not a good general-purpose API. 57579fd6dad1SPaolo Bonzini * 57589fd6dad1SPaolo Bonzini * Return: zero on success, -ve otherwise. 57599fd6dad1SPaolo Bonzini */ 57609fd6dad1SPaolo Bonzini int follow_pte(struct mm_struct *mm, unsigned long address, 57619fd6dad1SPaolo Bonzini pte_t **ptepp, spinlock_t **ptlp) 57629fd6dad1SPaolo Bonzini { 57630e5e64c0SMuchun Song pgd_t *pgd; 57640e5e64c0SMuchun Song p4d_t *p4d; 57650e5e64c0SMuchun Song pud_t *pud; 57660e5e64c0SMuchun Song pmd_t *pmd; 57670e5e64c0SMuchun Song pte_t *ptep; 57680e5e64c0SMuchun Song 57690e5e64c0SMuchun Song pgd = pgd_offset(mm, address); 57700e5e64c0SMuchun Song if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 57710e5e64c0SMuchun Song goto out; 57720e5e64c0SMuchun Song 57730e5e64c0SMuchun Song p4d = p4d_offset(pgd, address); 57740e5e64c0SMuchun Song if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) 57750e5e64c0SMuchun Song goto out; 57760e5e64c0SMuchun Song 57770e5e64c0SMuchun Song pud = pud_offset(p4d, address); 57780e5e64c0SMuchun Song if (pud_none(*pud) || unlikely(pud_bad(*pud))) 57790e5e64c0SMuchun Song goto out; 57800e5e64c0SMuchun Song 57810e5e64c0SMuchun Song pmd = pmd_offset(pud, address); 57820e5e64c0SMuchun Song VM_BUG_ON(pmd_trans_huge(*pmd)); 57830e5e64c0SMuchun Song 57840e5e64c0SMuchun Song ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 57853db82b93SHugh Dickins if (!ptep) 57863db82b93SHugh Dickins goto out; 5787c33c7948SRyan Roberts if (!pte_present(ptep_get(ptep))) 57880e5e64c0SMuchun Song goto unlock; 57890e5e64c0SMuchun Song *ptepp = ptep; 57900e5e64c0SMuchun Song return 0; 57910e5e64c0SMuchun Song unlock: 57920e5e64c0SMuchun Song pte_unmap_unlock(ptep, *ptlp); 57930e5e64c0SMuchun Song out: 57940e5e64c0SMuchun Song return -EINVAL; 57959fd6dad1SPaolo Bonzini } 57969fd6dad1SPaolo Bonzini EXPORT_SYMBOL_GPL(follow_pte); 57979fd6dad1SPaolo Bonzini 57989fd6dad1SPaolo Bonzini /** 57993b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 58003b6748e2SJohannes Weiner * @vma: memory mapping 58013b6748e2SJohannes Weiner * @address: user virtual address 58023b6748e2SJohannes Weiner * @pfn: location to store found PFN 58033b6748e2SJohannes Weiner * 58043b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 58053b6748e2SJohannes Weiner * 58069fd6dad1SPaolo Bonzini * This function does not allow the caller to read the permissions 58079fd6dad1SPaolo Bonzini * of the PTE. Do not use it. 58089fd6dad1SPaolo Bonzini * 5809a862f68aSMike Rapoport * Return: zero and the pfn at @pfn on success, -ve otherwise. 58103b6748e2SJohannes Weiner */ 58113b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 58123b6748e2SJohannes Weiner unsigned long *pfn) 58133b6748e2SJohannes Weiner { 58143b6748e2SJohannes Weiner int ret = -EINVAL; 58153b6748e2SJohannes Weiner spinlock_t *ptl; 58163b6748e2SJohannes Weiner pte_t *ptep; 58173b6748e2SJohannes Weiner 58183b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 58193b6748e2SJohannes Weiner return ret; 58203b6748e2SJohannes Weiner 58219fd6dad1SPaolo Bonzini ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 58223b6748e2SJohannes Weiner if (ret) 58233b6748e2SJohannes Weiner return ret; 5824c33c7948SRyan Roberts *pfn = pte_pfn(ptep_get(ptep)); 58253b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 58263b6748e2SJohannes Weiner return 0; 58273b6748e2SJohannes Weiner } 58283b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 58293b6748e2SJohannes Weiner 583028b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 5831d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 583228b2ee20SRik van Riel unsigned long address, unsigned int flags, 5833d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 583428b2ee20SRik van Riel { 583503668a4dSJohannes Weiner int ret = -EINVAL; 583628b2ee20SRik van Riel pte_t *ptep, pte; 583728b2ee20SRik van Riel spinlock_t *ptl; 583828b2ee20SRik van Riel 5839d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 5840d87fe660Svenkatesh.pallipadi@intel.com goto out; 584128b2ee20SRik van Riel 58429fd6dad1SPaolo Bonzini if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 5843d87fe660Svenkatesh.pallipadi@intel.com goto out; 5844c33c7948SRyan Roberts pte = ptep_get(ptep); 584503668a4dSJohannes Weiner 5846f6f37321SLinus Torvalds if ((flags & FOLL_WRITE) && !pte_write(pte)) 584728b2ee20SRik van Riel goto unlock; 584828b2ee20SRik van Riel 584928b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 585003668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 585128b2ee20SRik van Riel 585203668a4dSJohannes Weiner ret = 0; 585328b2ee20SRik van Riel unlock: 585428b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 585528b2ee20SRik van Riel out: 5856d87fe660Svenkatesh.pallipadi@intel.com return ret; 585728b2ee20SRik van Riel } 585828b2ee20SRik van Riel 585996667f8aSDaniel Vetter /** 586096667f8aSDaniel Vetter * generic_access_phys - generic implementation for iomem mmap access 586196667f8aSDaniel Vetter * @vma: the vma to access 5862f0953a1bSIngo Molnar * @addr: userspace address, not relative offset within @vma 586396667f8aSDaniel Vetter * @buf: buffer to read/write 586496667f8aSDaniel Vetter * @len: length of transfer 586596667f8aSDaniel Vetter * @write: set to FOLL_WRITE when writing, otherwise reading 586696667f8aSDaniel Vetter * 586796667f8aSDaniel Vetter * This is a generic implementation for &vm_operations_struct.access for an 586896667f8aSDaniel Vetter * iomem mapping. This callback is used by access_process_vm() when the @vma is 586996667f8aSDaniel Vetter * not page based. 587096667f8aSDaniel Vetter */ 587128b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 587228b2ee20SRik van Riel void *buf, int len, int write) 587328b2ee20SRik van Riel { 587428b2ee20SRik van Riel resource_size_t phys_addr; 587528b2ee20SRik van Riel unsigned long prot = 0; 58762bc7273bSKOSAKI Motohiro void __iomem *maddr; 587796667f8aSDaniel Vetter pte_t *ptep, pte; 587896667f8aSDaniel Vetter spinlock_t *ptl; 587996667f8aSDaniel Vetter int offset = offset_in_page(addr); 588096667f8aSDaniel Vetter int ret = -EINVAL; 588128b2ee20SRik van Riel 588296667f8aSDaniel Vetter if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 588396667f8aSDaniel Vetter return -EINVAL; 588496667f8aSDaniel Vetter 588596667f8aSDaniel Vetter retry: 5886e913a8cdSLinus Torvalds if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) 588796667f8aSDaniel Vetter return -EINVAL; 5888c33c7948SRyan Roberts pte = ptep_get(ptep); 588996667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 589096667f8aSDaniel Vetter 589196667f8aSDaniel Vetter prot = pgprot_val(pte_pgprot(pte)); 589296667f8aSDaniel Vetter phys_addr = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 589396667f8aSDaniel Vetter 589496667f8aSDaniel Vetter if ((write & FOLL_WRITE) && !pte_write(pte)) 589528b2ee20SRik van Riel return -EINVAL; 589628b2ee20SRik van Riel 58979cb12d7bSGrazvydas Ignotas maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); 589824eee1e4Sjie@chenjie6@huwei.com if (!maddr) 589924eee1e4Sjie@chenjie6@huwei.com return -ENOMEM; 590024eee1e4Sjie@chenjie6@huwei.com 5901e913a8cdSLinus Torvalds if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) 590296667f8aSDaniel Vetter goto out_unmap; 590396667f8aSDaniel Vetter 5904c33c7948SRyan Roberts if (!pte_same(pte, ptep_get(ptep))) { 590596667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 590696667f8aSDaniel Vetter iounmap(maddr); 590796667f8aSDaniel Vetter 590896667f8aSDaniel Vetter goto retry; 590996667f8aSDaniel Vetter } 591096667f8aSDaniel Vetter 591128b2ee20SRik van Riel if (write) 591228b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 591328b2ee20SRik van Riel else 591428b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 591596667f8aSDaniel Vetter ret = len; 591696667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 591796667f8aSDaniel Vetter out_unmap: 591828b2ee20SRik van Riel iounmap(maddr); 591928b2ee20SRik van Riel 592096667f8aSDaniel Vetter return ret; 592128b2ee20SRik van Riel } 59225a73633eSUwe Kleine-König EXPORT_SYMBOL_GPL(generic_access_phys); 592328b2ee20SRik van Riel #endif 592428b2ee20SRik van Riel 59250ec76a11SDavid Howells /* 5926d3f5ffcaSJohn Hubbard * Access another process' address space as given in mm. 59270ec76a11SDavid Howells */ 5928c43cfa42SLorenzo Stoakes static int __access_remote_vm(struct mm_struct *mm, unsigned long addr, 5929c43cfa42SLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 59300ec76a11SDavid Howells { 59310ec76a11SDavid Howells void *old_buf = buf; 5932442486ecSLorenzo Stoakes int write = gup_flags & FOLL_WRITE; 59330ec76a11SDavid Howells 5934d8ed45c5SMichel Lespinasse if (mmap_read_lock_killable(mm)) 59351e426fe2SKonstantin Khlebnikov return 0; 59361e426fe2SKonstantin Khlebnikov 593722883973SKirill A. Shutemov /* Untag the address before looking up the VMA */ 593822883973SKirill A. Shutemov addr = untagged_addr_remote(mm, addr); 593922883973SKirill A. Shutemov 5940eee9c708SLinus Torvalds /* Avoid triggering the temporary warning in __get_user_pages */ 5941eee9c708SLinus Torvalds if (!vma_lookup(mm, addr) && !expand_stack(mm, addr)) 5942eee9c708SLinus Torvalds return 0; 5943eee9c708SLinus Torvalds 5944183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 59450ec76a11SDavid Howells while (len) { 5946ca5e8632SLorenzo Stoakes int bytes, offset; 59470ec76a11SDavid Howells void *maddr; 5948ca5e8632SLorenzo Stoakes struct vm_area_struct *vma = NULL; 5949ca5e8632SLorenzo Stoakes struct page *page = get_user_page_vma_remote(mm, addr, 5950ca5e8632SLorenzo Stoakes gup_flags, &vma); 59510ec76a11SDavid Howells 59526a1960b8SLorenzo Stoakes if (IS_ERR(page)) { 59538d7071afSLinus Torvalds /* We might need to expand the stack to access it */ 59548d7071afSLinus Torvalds vma = vma_lookup(mm, addr); 59558d7071afSLinus Torvalds if (!vma) { 59568d7071afSLinus Torvalds vma = expand_stack(mm, addr); 59579471f1f2SLinus Torvalds 59589471f1f2SLinus Torvalds /* mmap_lock was dropped on failure */ 59598d7071afSLinus Torvalds if (!vma) 59609471f1f2SLinus Torvalds return buf - old_buf; 59619471f1f2SLinus Torvalds 59629471f1f2SLinus Torvalds /* Try again if stack expansion worked */ 59639471f1f2SLinus Torvalds continue; 59648d7071afSLinus Torvalds } 59658d7071afSLinus Torvalds 596628b2ee20SRik van Riel /* 596728b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 596828b2ee20SRik van Riel * we can access using slightly different code. 596928b2ee20SRik van Riel */ 59709471f1f2SLinus Torvalds bytes = 0; 59719471f1f2SLinus Torvalds #ifdef CONFIG_HAVE_IOREMAP_PROT 597228b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 59739471f1f2SLinus Torvalds bytes = vma->vm_ops->access(vma, addr, buf, 597428b2ee20SRik van Riel len, write); 5975dbffcd03SRik van Riel #endif 59769471f1f2SLinus Torvalds if (bytes <= 0) 59779471f1f2SLinus Torvalds break; 597828b2ee20SRik van Riel } else { 59790ec76a11SDavid Howells bytes = len; 59800ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 59810ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 59820ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 59830ec76a11SDavid Howells 5984f7ef5fe7SFabio M. De Francesco maddr = kmap_local_page(page); 59850ec76a11SDavid Howells if (write) { 59860ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 59870ec76a11SDavid Howells maddr + offset, buf, bytes); 59880ec76a11SDavid Howells set_page_dirty_lock(page); 59890ec76a11SDavid Howells } else { 59900ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 59910ec76a11SDavid Howells buf, maddr + offset, bytes); 59920ec76a11SDavid Howells } 5993f7ef5fe7SFabio M. De Francesco unmap_and_put_page(page, maddr); 599428b2ee20SRik van Riel } 59950ec76a11SDavid Howells len -= bytes; 59960ec76a11SDavid Howells buf += bytes; 59970ec76a11SDavid Howells addr += bytes; 59980ec76a11SDavid Howells } 5999d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 60000ec76a11SDavid Howells 60010ec76a11SDavid Howells return buf - old_buf; 60020ec76a11SDavid Howells } 600303252919SAndi Kleen 60045ddd36b9SStephen Wilson /** 6005ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 60065ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 60075ddd36b9SStephen Wilson * @addr: start address to access 60085ddd36b9SStephen Wilson * @buf: source or destination buffer 60095ddd36b9SStephen Wilson * @len: number of bytes to transfer 60106347e8d5SLorenzo Stoakes * @gup_flags: flags modifying lookup behaviour 60115ddd36b9SStephen Wilson * 60125ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 6013a862f68aSMike Rapoport * 6014a862f68aSMike Rapoport * Return: number of bytes copied from source to destination. 60155ddd36b9SStephen Wilson */ 60165ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 60176347e8d5SLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 60185ddd36b9SStephen Wilson { 6019d3f5ffcaSJohn Hubbard return __access_remote_vm(mm, addr, buf, len, gup_flags); 60205ddd36b9SStephen Wilson } 60215ddd36b9SStephen Wilson 602203252919SAndi Kleen /* 6023206cb636SStephen Wilson * Access another process' address space. 6024206cb636SStephen Wilson * Source/target buffer must be kernel space, 6025206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 6026206cb636SStephen Wilson */ 6027206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 6028f307ab6dSLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 6029206cb636SStephen Wilson { 6030206cb636SStephen Wilson struct mm_struct *mm; 6031206cb636SStephen Wilson int ret; 6032206cb636SStephen Wilson 6033206cb636SStephen Wilson mm = get_task_mm(tsk); 6034206cb636SStephen Wilson if (!mm) 6035206cb636SStephen Wilson return 0; 6036206cb636SStephen Wilson 6037d3f5ffcaSJohn Hubbard ret = __access_remote_vm(mm, addr, buf, len, gup_flags); 6038442486ecSLorenzo Stoakes 6039206cb636SStephen Wilson mmput(mm); 6040206cb636SStephen Wilson 6041206cb636SStephen Wilson return ret; 6042206cb636SStephen Wilson } 6043fcd35857SCatalin Marinas EXPORT_SYMBOL_GPL(access_process_vm); 6044206cb636SStephen Wilson 604503252919SAndi Kleen /* 604603252919SAndi Kleen * Print the name of a VMA. 604703252919SAndi Kleen */ 604803252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 604903252919SAndi Kleen { 605003252919SAndi Kleen struct mm_struct *mm = current->mm; 605103252919SAndi Kleen struct vm_area_struct *vma; 605203252919SAndi Kleen 6053e8bff74aSIngo Molnar /* 60540a7f682dSMichal Hocko * we might be running from an atomic context so we cannot sleep 6055e8bff74aSIngo Molnar */ 6056d8ed45c5SMichel Lespinasse if (!mmap_read_trylock(mm)) 6057e8bff74aSIngo Molnar return; 6058e8bff74aSIngo Molnar 605903252919SAndi Kleen vma = find_vma(mm, ip); 606003252919SAndi Kleen if (vma && vma->vm_file) { 606103252919SAndi Kleen struct file *f = vma->vm_file; 60620a7f682dSMichal Hocko char *buf = (char *)__get_free_page(GFP_NOWAIT); 606303252919SAndi Kleen if (buf) { 60642fbc57c5SAndy Shevchenko char *p; 606503252919SAndi Kleen 60669bf39ab2SMiklos Szeredi p = file_path(f, buf, PAGE_SIZE); 606703252919SAndi Kleen if (IS_ERR(p)) 606803252919SAndi Kleen p = "?"; 60692fbc57c5SAndy Shevchenko printk("%s%s[%lx+%lx]", prefix, kbasename(p), 607003252919SAndi Kleen vma->vm_start, 607103252919SAndi Kleen vma->vm_end - vma->vm_start); 607203252919SAndi Kleen free_page((unsigned long)buf); 607303252919SAndi Kleen } 607403252919SAndi Kleen } 6075d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 607603252919SAndi Kleen } 60773ee1afa3SNick Piggin 6078662bbcb2SMichael S. Tsirkin #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) 60799ec23531SDavid Hildenbrand void __might_fault(const char *file, int line) 60803ee1afa3SNick Piggin { 60819ec23531SDavid Hildenbrand if (pagefault_disabled()) 6082662bbcb2SMichael S. Tsirkin return; 608342a38756SThomas Gleixner __might_sleep(file, line); 60849ec23531SDavid Hildenbrand #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) 6085662bbcb2SMichael S. Tsirkin if (current->mm) 6086da1c55f1SMichel Lespinasse might_lock_read(¤t->mm->mmap_lock); 60879ec23531SDavid Hildenbrand #endif 60883ee1afa3SNick Piggin } 60899ec23531SDavid Hildenbrand EXPORT_SYMBOL(__might_fault); 60903ee1afa3SNick Piggin #endif 609147ad8475SAndrea Arcangeli 609247ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 6093c6ddfb6cSHuang Ying /* 6094c6ddfb6cSHuang Ying * Process all subpages of the specified huge page with the specified 6095c6ddfb6cSHuang Ying * operation. The target subpage will be processed last to keep its 6096c6ddfb6cSHuang Ying * cache lines hot. 6097c6ddfb6cSHuang Ying */ 60981cb9dc4bSLiu Shixin static inline int process_huge_page( 6099c6ddfb6cSHuang Ying unsigned long addr_hint, unsigned int pages_per_huge_page, 61001cb9dc4bSLiu Shixin int (*process_subpage)(unsigned long addr, int idx, void *arg), 6101c6ddfb6cSHuang Ying void *arg) 6102c6ddfb6cSHuang Ying { 61031cb9dc4bSLiu Shixin int i, n, base, l, ret; 6104c6ddfb6cSHuang Ying unsigned long addr = addr_hint & 6105c6ddfb6cSHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 6106c6ddfb6cSHuang Ying 6107c6ddfb6cSHuang Ying /* Process target subpage last to keep its cache lines hot */ 6108c6ddfb6cSHuang Ying might_sleep(); 6109c6ddfb6cSHuang Ying n = (addr_hint - addr) / PAGE_SIZE; 6110c6ddfb6cSHuang Ying if (2 * n <= pages_per_huge_page) { 6111c6ddfb6cSHuang Ying /* If target subpage in first half of huge page */ 6112c6ddfb6cSHuang Ying base = 0; 6113c6ddfb6cSHuang Ying l = n; 6114c6ddfb6cSHuang Ying /* Process subpages at the end of huge page */ 6115c6ddfb6cSHuang Ying for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { 6116c6ddfb6cSHuang Ying cond_resched(); 61171cb9dc4bSLiu Shixin ret = process_subpage(addr + i * PAGE_SIZE, i, arg); 61181cb9dc4bSLiu Shixin if (ret) 61191cb9dc4bSLiu Shixin return ret; 6120c6ddfb6cSHuang Ying } 6121c6ddfb6cSHuang Ying } else { 6122c6ddfb6cSHuang Ying /* If target subpage in second half of huge page */ 6123c6ddfb6cSHuang Ying base = pages_per_huge_page - 2 * (pages_per_huge_page - n); 6124c6ddfb6cSHuang Ying l = pages_per_huge_page - n; 6125c6ddfb6cSHuang Ying /* Process subpages at the begin of huge page */ 6126c6ddfb6cSHuang Ying for (i = 0; i < base; i++) { 6127c6ddfb6cSHuang Ying cond_resched(); 61281cb9dc4bSLiu Shixin ret = process_subpage(addr + i * PAGE_SIZE, i, arg); 61291cb9dc4bSLiu Shixin if (ret) 61301cb9dc4bSLiu Shixin return ret; 6131c6ddfb6cSHuang Ying } 6132c6ddfb6cSHuang Ying } 6133c6ddfb6cSHuang Ying /* 6134c6ddfb6cSHuang Ying * Process remaining subpages in left-right-left-right pattern 6135c6ddfb6cSHuang Ying * towards the target subpage 6136c6ddfb6cSHuang Ying */ 6137c6ddfb6cSHuang Ying for (i = 0; i < l; i++) { 6138c6ddfb6cSHuang Ying int left_idx = base + i; 6139c6ddfb6cSHuang Ying int right_idx = base + 2 * l - 1 - i; 6140c6ddfb6cSHuang Ying 6141c6ddfb6cSHuang Ying cond_resched(); 61421cb9dc4bSLiu Shixin ret = process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg); 61431cb9dc4bSLiu Shixin if (ret) 61441cb9dc4bSLiu Shixin return ret; 6145c6ddfb6cSHuang Ying cond_resched(); 61461cb9dc4bSLiu Shixin ret = process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg); 61471cb9dc4bSLiu Shixin if (ret) 61481cb9dc4bSLiu Shixin return ret; 6149c6ddfb6cSHuang Ying } 61501cb9dc4bSLiu Shixin return 0; 6151c6ddfb6cSHuang Ying } 6152c6ddfb6cSHuang Ying 615347ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 615447ad8475SAndrea Arcangeli unsigned long addr, 615547ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 615647ad8475SAndrea Arcangeli { 615747ad8475SAndrea Arcangeli int i; 615814455eabSCheng Li struct page *p; 615947ad8475SAndrea Arcangeli 616047ad8475SAndrea Arcangeli might_sleep(); 616114455eabSCheng Li for (i = 0; i < pages_per_huge_page; i++) { 616214455eabSCheng Li p = nth_page(page, i); 616347ad8475SAndrea Arcangeli cond_resched(); 616447ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 616547ad8475SAndrea Arcangeli } 616647ad8475SAndrea Arcangeli } 6167c6ddfb6cSHuang Ying 61681cb9dc4bSLiu Shixin static int clear_subpage(unsigned long addr, int idx, void *arg) 6169c6ddfb6cSHuang Ying { 6170c6ddfb6cSHuang Ying struct page *page = arg; 6171c6ddfb6cSHuang Ying 617221fff064SKefeng Wang clear_user_highpage(nth_page(page, idx), addr); 61731cb9dc4bSLiu Shixin return 0; 6174c6ddfb6cSHuang Ying } 6175c6ddfb6cSHuang Ying 617647ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 6177c79b57e4SHuang Ying unsigned long addr_hint, unsigned int pages_per_huge_page) 617847ad8475SAndrea Arcangeli { 6179c79b57e4SHuang Ying unsigned long addr = addr_hint & 6180c79b57e4SHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 618147ad8475SAndrea Arcangeli 618247ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 618347ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 618447ad8475SAndrea Arcangeli return; 618547ad8475SAndrea Arcangeli } 618647ad8475SAndrea Arcangeli 6187c6ddfb6cSHuang Ying process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page); 618847ad8475SAndrea Arcangeli } 618947ad8475SAndrea Arcangeli 61901cb9dc4bSLiu Shixin static int copy_user_gigantic_page(struct folio *dst, struct folio *src, 619147ad8475SAndrea Arcangeli unsigned long addr, 619247ad8475SAndrea Arcangeli struct vm_area_struct *vma, 619347ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 619447ad8475SAndrea Arcangeli { 619547ad8475SAndrea Arcangeli int i; 6196c0e8150eSZhangPeng struct page *dst_page; 6197c0e8150eSZhangPeng struct page *src_page; 619847ad8475SAndrea Arcangeli 619914455eabSCheng Li for (i = 0; i < pages_per_huge_page; i++) { 6200c0e8150eSZhangPeng dst_page = folio_page(dst, i); 6201c0e8150eSZhangPeng src_page = folio_page(src, i); 620214455eabSCheng Li 620347ad8475SAndrea Arcangeli cond_resched(); 62041cb9dc4bSLiu Shixin if (copy_mc_user_highpage(dst_page, src_page, 62051cb9dc4bSLiu Shixin addr + i*PAGE_SIZE, vma)) { 62061cb9dc4bSLiu Shixin memory_failure_queue(page_to_pfn(src_page), 0); 62071cb9dc4bSLiu Shixin return -EHWPOISON; 620847ad8475SAndrea Arcangeli } 620947ad8475SAndrea Arcangeli } 62101cb9dc4bSLiu Shixin return 0; 62111cb9dc4bSLiu Shixin } 621247ad8475SAndrea Arcangeli 6213c9f4cd71SHuang Ying struct copy_subpage_arg { 6214c9f4cd71SHuang Ying struct page *dst; 6215c9f4cd71SHuang Ying struct page *src; 6216c9f4cd71SHuang Ying struct vm_area_struct *vma; 6217c9f4cd71SHuang Ying }; 6218c9f4cd71SHuang Ying 62191cb9dc4bSLiu Shixin static int copy_subpage(unsigned long addr, int idx, void *arg) 6220c9f4cd71SHuang Ying { 6221c9f4cd71SHuang Ying struct copy_subpage_arg *copy_arg = arg; 622221fff064SKefeng Wang struct page *dst = nth_page(copy_arg->dst, idx); 622321fff064SKefeng Wang struct page *src = nth_page(copy_arg->src, idx); 6224c9f4cd71SHuang Ying 622521fff064SKefeng Wang if (copy_mc_user_highpage(dst, src, addr, copy_arg->vma)) { 622621fff064SKefeng Wang memory_failure_queue(page_to_pfn(src), 0); 62271cb9dc4bSLiu Shixin return -EHWPOISON; 62281cb9dc4bSLiu Shixin } 62291cb9dc4bSLiu Shixin return 0; 6230c9f4cd71SHuang Ying } 6231c9f4cd71SHuang Ying 62321cb9dc4bSLiu Shixin int copy_user_large_folio(struct folio *dst, struct folio *src, 6233c0e8150eSZhangPeng unsigned long addr_hint, struct vm_area_struct *vma) 623447ad8475SAndrea Arcangeli { 6235c0e8150eSZhangPeng unsigned int pages_per_huge_page = folio_nr_pages(dst); 6236c9f4cd71SHuang Ying unsigned long addr = addr_hint & 6237c9f4cd71SHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 6238c9f4cd71SHuang Ying struct copy_subpage_arg arg = { 6239c0e8150eSZhangPeng .dst = &dst->page, 6240c0e8150eSZhangPeng .src = &src->page, 6241c9f4cd71SHuang Ying .vma = vma, 6242c9f4cd71SHuang Ying }; 624347ad8475SAndrea Arcangeli 62441cb9dc4bSLiu Shixin if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) 62451cb9dc4bSLiu Shixin return copy_user_gigantic_page(dst, src, addr, vma, 624647ad8475SAndrea Arcangeli pages_per_huge_page); 624747ad8475SAndrea Arcangeli 62481cb9dc4bSLiu Shixin return process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg); 624947ad8475SAndrea Arcangeli } 6250fa4d75c1SMike Kravetz 6251e87340caSZhangPeng long copy_folio_from_user(struct folio *dst_folio, 6252fa4d75c1SMike Kravetz const void __user *usr_src, 6253810a56b9SMike Kravetz bool allow_pagefault) 6254fa4d75c1SMike Kravetz { 6255e87340caSZhangPeng void *kaddr; 6256fa4d75c1SMike Kravetz unsigned long i, rc = 0; 6257e87340caSZhangPeng unsigned int nr_pages = folio_nr_pages(dst_folio); 6258e87340caSZhangPeng unsigned long ret_val = nr_pages * PAGE_SIZE; 625914455eabSCheng Li struct page *subpage; 6260fa4d75c1SMike Kravetz 6261e87340caSZhangPeng for (i = 0; i < nr_pages; i++) { 6262e87340caSZhangPeng subpage = folio_page(dst_folio, i); 6263e87340caSZhangPeng kaddr = kmap_local_page(subpage); 62640d508c1fSZhangPeng if (!allow_pagefault) 62650d508c1fSZhangPeng pagefault_disable(); 6266e87340caSZhangPeng rc = copy_from_user(kaddr, usr_src + i * PAGE_SIZE, PAGE_SIZE); 62670d508c1fSZhangPeng if (!allow_pagefault) 62680d508c1fSZhangPeng pagefault_enable(); 6269e87340caSZhangPeng kunmap_local(kaddr); 6270fa4d75c1SMike Kravetz 6271fa4d75c1SMike Kravetz ret_val -= (PAGE_SIZE - rc); 6272fa4d75c1SMike Kravetz if (rc) 6273fa4d75c1SMike Kravetz break; 6274fa4d75c1SMike Kravetz 6275e763243cSMuchun Song flush_dcache_page(subpage); 6276e763243cSMuchun Song 6277fa4d75c1SMike Kravetz cond_resched(); 6278fa4d75c1SMike Kravetz } 6279fa4d75c1SMike Kravetz return ret_val; 6280fa4d75c1SMike Kravetz } 628147ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 628249076ec2SKirill A. Shutemov 628340b64acdSOlof Johansson #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS 6284b35f1819SKirill A. Shutemov 6285b35f1819SKirill A. Shutemov static struct kmem_cache *page_ptl_cachep; 6286b35f1819SKirill A. Shutemov 6287b35f1819SKirill A. Shutemov void __init ptlock_cache_init(void) 6288b35f1819SKirill A. Shutemov { 6289b35f1819SKirill A. Shutemov page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, 6290b35f1819SKirill A. Shutemov SLAB_PANIC, NULL); 6291b35f1819SKirill A. Shutemov } 6292b35f1819SKirill A. Shutemov 6293f5ecca06SVishal Moola (Oracle) bool ptlock_alloc(struct ptdesc *ptdesc) 629449076ec2SKirill A. Shutemov { 629549076ec2SKirill A. Shutemov spinlock_t *ptl; 629649076ec2SKirill A. Shutemov 6297b35f1819SKirill A. Shutemov ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); 629849076ec2SKirill A. Shutemov if (!ptl) 629949076ec2SKirill A. Shutemov return false; 6300f5ecca06SVishal Moola (Oracle) ptdesc->ptl = ptl; 630149076ec2SKirill A. Shutemov return true; 630249076ec2SKirill A. Shutemov } 630349076ec2SKirill A. Shutemov 63046ed1b8a0SVishal Moola (Oracle) void ptlock_free(struct ptdesc *ptdesc) 630549076ec2SKirill A. Shutemov { 63066ed1b8a0SVishal Moola (Oracle) kmem_cache_free(page_ptl_cachep, ptdesc->ptl); 630749076ec2SKirill A. Shutemov } 630849076ec2SKirill A. Shutemov #endif 6309