1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/mm/memory.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 61da177e4SLinus Torvalds */ 71da177e4SLinus Torvalds 81da177e4SLinus Torvalds /* 91da177e4SLinus Torvalds * demand-loading started 01.12.91 - seems it is high on the list of 101da177e4SLinus Torvalds * things wanted, and it should be easy to implement. - Linus 111da177e4SLinus Torvalds */ 121da177e4SLinus Torvalds 131da177e4SLinus Torvalds /* 141da177e4SLinus Torvalds * Ok, demand-loading was easy, shared pages a little bit tricker. Shared 151da177e4SLinus Torvalds * pages started 02.12.91, seems to work. - Linus. 161da177e4SLinus Torvalds * 171da177e4SLinus Torvalds * Tested sharing by executing about 30 /bin/sh: under the old kernel it 181da177e4SLinus Torvalds * would have taken more than the 6M I have free, but it worked well as 191da177e4SLinus Torvalds * far as I could see. 201da177e4SLinus Torvalds * 211da177e4SLinus Torvalds * Also corrected some "invalidate()"s - I wasn't doing enough of them. 221da177e4SLinus Torvalds */ 231da177e4SLinus Torvalds 241da177e4SLinus Torvalds /* 251da177e4SLinus Torvalds * Real VM (paging to/from disk) started 18.12.91. Much more work and 261da177e4SLinus Torvalds * thought has to go into this. Oh, well.. 271da177e4SLinus Torvalds * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. 281da177e4SLinus Torvalds * Found it. Everything seems to work now. 291da177e4SLinus Torvalds * 20.12.91 - Ok, making the swap-device changeable like the root. 301da177e4SLinus Torvalds */ 311da177e4SLinus Torvalds 321da177e4SLinus Torvalds /* 331da177e4SLinus Torvalds * 05.04.94 - Multi-page memory management added for v1.1. 341da177e4SLinus Torvalds * Idea by Alex Bligh (alex@cconcepts.co.uk) 351da177e4SLinus Torvalds * 361da177e4SLinus Torvalds * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG 371da177e4SLinus Torvalds * (Gerhard.Wichert@pdb.siemens.de) 381da177e4SLinus Torvalds * 391da177e4SLinus Torvalds * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) 401da177e4SLinus Torvalds */ 411da177e4SLinus Torvalds 421da177e4SLinus Torvalds #include <linux/kernel_stat.h> 431da177e4SLinus Torvalds #include <linux/mm.h> 4436090defSArnd Bergmann #include <linux/mm_inline.h> 456e84f315SIngo Molnar #include <linux/sched/mm.h> 46f7ccbae4SIngo Molnar #include <linux/sched/coredump.h> 476a3827d7SIngo Molnar #include <linux/sched/numa_balancing.h> 4829930025SIngo Molnar #include <linux/sched/task.h> 491da177e4SLinus Torvalds #include <linux/hugetlb.h> 501da177e4SLinus Torvalds #include <linux/mman.h> 511da177e4SLinus Torvalds #include <linux/swap.h> 521da177e4SLinus Torvalds #include <linux/highmem.h> 531da177e4SLinus Torvalds #include <linux/pagemap.h> 545042db43SJérôme Glisse #include <linux/memremap.h> 55b073d7f8SAlexander Potapenko #include <linux/kmsan.h> 569a840895SHugh Dickins #include <linux/ksm.h> 571da177e4SLinus Torvalds #include <linux/rmap.h> 58b95f1b31SPaul Gortmaker #include <linux/export.h> 590ff92245SShailabh Nagar #include <linux/delayacct.h> 601da177e4SLinus Torvalds #include <linux/init.h> 6101c8f1c4SDan Williams #include <linux/pfn_t.h> 62edc79b2aSPeter Zijlstra #include <linux/writeback.h> 638a9f3ccdSBalbir Singh #include <linux/memcontrol.h> 64cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 653dc14741SHugh Dickins #include <linux/swapops.h> 663dc14741SHugh Dickins #include <linux/elf.h> 675a0e3ad6STejun Heo #include <linux/gfp.h> 684daae3b4SMel Gorman #include <linux/migrate.h> 692fbc57c5SAndy Shevchenko #include <linux/string.h> 70467b171aSAneesh Kumar K.V #include <linux/memory-tiers.h> 711592eef0SKirill A. Shutemov #include <linux/debugfs.h> 726b251fc9SAndrea Arcangeli #include <linux/userfaultfd_k.h> 73bc2466e4SJan Kara #include <linux/dax.h> 746b31d595SMichal Hocko #include <linux/oom.h> 7598fa15f3SAnshuman Khandual #include <linux/numa.h> 76bce617edSPeter Xu #include <linux/perf_event.h> 77bce617edSPeter Xu #include <linux/ptrace.h> 78e80d3909SJoerg Roedel #include <linux/vmalloc.h> 7933024536SHuang Ying #include <linux/sched/sysctl.h> 801da177e4SLinus Torvalds 81b3d1411bSJoel Fernandes (Google) #include <trace/events/kmem.h> 82b3d1411bSJoel Fernandes (Google) 836952b61dSAlexey Dobriyan #include <asm/io.h> 8433a709b2SDave Hansen #include <asm/mmu_context.h> 851da177e4SLinus Torvalds #include <asm/pgalloc.h> 867c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 871da177e4SLinus Torvalds #include <asm/tlb.h> 881da177e4SLinus Torvalds #include <asm/tlbflush.h> 891da177e4SLinus Torvalds 90e80d3909SJoerg Roedel #include "pgalloc-track.h" 9142b77728SJan Beulich #include "internal.h" 92014bb1deSNeilBrown #include "swap.h" 9342b77728SJan Beulich 94af27d940SArnd Bergmann #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST) 9590572890SPeter Zijlstra #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. 9675980e97SPeter Zijlstra #endif 9775980e97SPeter Zijlstra 98a9ee6cf5SMike Rapoport #ifndef CONFIG_NUMA 991da177e4SLinus Torvalds unsigned long max_mapnr; 1001da177e4SLinus Torvalds EXPORT_SYMBOL(max_mapnr); 101166f61b9STobin C Harding 102166f61b9STobin C Harding struct page *mem_map; 1031da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 1041da177e4SLinus Torvalds #endif 1051da177e4SLinus Torvalds 1065c041f5dSPeter Xu static vm_fault_t do_fault(struct vm_fault *vmf); 1072bad466cSPeter Xu static vm_fault_t do_anonymous_page(struct vm_fault *vmf); 1082bad466cSPeter Xu static bool vmf_pte_changed(struct vm_fault *vmf); 1092bad466cSPeter Xu 1102bad466cSPeter Xu /* 1112bad466cSPeter Xu * Return true if the original pte was a uffd-wp pte marker (so the pte was 1122bad466cSPeter Xu * wr-protected). 1132bad466cSPeter Xu */ 1142bad466cSPeter Xu static bool vmf_orig_pte_uffd_wp(struct vm_fault *vmf) 1152bad466cSPeter Xu { 1162bad466cSPeter Xu if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID)) 1172bad466cSPeter Xu return false; 1182bad466cSPeter Xu 1192bad466cSPeter Xu return pte_marker_uffd_wp(vmf->orig_pte); 1202bad466cSPeter Xu } 1215c041f5dSPeter Xu 1221da177e4SLinus Torvalds /* 1231da177e4SLinus Torvalds * A number of key systems in x86 including ioremap() rely on the assumption 1241da177e4SLinus Torvalds * that high_memory defines the upper bound on direct map memory, then end 1251da177e4SLinus Torvalds * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and 1261da177e4SLinus Torvalds * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL 1271da177e4SLinus Torvalds * and ZONE_HIGHMEM. 1281da177e4SLinus Torvalds */ 1291da177e4SLinus Torvalds void *high_memory; 1301da177e4SLinus Torvalds EXPORT_SYMBOL(high_memory); 1311da177e4SLinus Torvalds 13232a93233SIngo Molnar /* 13332a93233SIngo Molnar * Randomize the address space (stacks, mmaps, brk, etc.). 13432a93233SIngo Molnar * 13532a93233SIngo Molnar * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, 13632a93233SIngo Molnar * as ancient (libc5 based) binaries can segfault. ) 13732a93233SIngo Molnar */ 13832a93233SIngo Molnar int randomize_va_space __read_mostly = 13932a93233SIngo Molnar #ifdef CONFIG_COMPAT_BRK 14032a93233SIngo Molnar 1; 14132a93233SIngo Molnar #else 14232a93233SIngo Molnar 2; 14332a93233SIngo Molnar #endif 144a62eaf15SAndi Kleen 14546bdb427SWill Deacon #ifndef arch_wants_old_prefaulted_pte 14646bdb427SWill Deacon static inline bool arch_wants_old_prefaulted_pte(void) 14746bdb427SWill Deacon { 14846bdb427SWill Deacon /* 14946bdb427SWill Deacon * Transitioning a PTE from 'old' to 'young' can be expensive on 15046bdb427SWill Deacon * some architectures, even if it's performed in hardware. By 15146bdb427SWill Deacon * default, "false" means prefaulted entries will be 'young'. 15246bdb427SWill Deacon */ 15346bdb427SWill Deacon return false; 15446bdb427SWill Deacon } 15546bdb427SWill Deacon #endif 15646bdb427SWill Deacon 157a62eaf15SAndi Kleen static int __init disable_randmaps(char *s) 158a62eaf15SAndi Kleen { 159a62eaf15SAndi Kleen randomize_va_space = 0; 1609b41046cSOGAWA Hirofumi return 1; 161a62eaf15SAndi Kleen } 162a62eaf15SAndi Kleen __setup("norandmaps", disable_randmaps); 163a62eaf15SAndi Kleen 16462eede62SHugh Dickins unsigned long zero_pfn __read_mostly; 1650b70068eSArd Biesheuvel EXPORT_SYMBOL(zero_pfn); 1660b70068eSArd Biesheuvel 167166f61b9STobin C Harding unsigned long highest_memmap_pfn __read_mostly; 168166f61b9STobin C Harding 169a13ea5b7SHugh Dickins /* 170a13ea5b7SHugh Dickins * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() 171a13ea5b7SHugh Dickins */ 172a13ea5b7SHugh Dickins static int __init init_zero_pfn(void) 173a13ea5b7SHugh Dickins { 174a13ea5b7SHugh Dickins zero_pfn = page_to_pfn(ZERO_PAGE(0)); 175a13ea5b7SHugh Dickins return 0; 176a13ea5b7SHugh Dickins } 177e720e7d0SIlya Lipnitskiy early_initcall(init_zero_pfn); 178a62eaf15SAndi Kleen 179f1a79412SShakeel Butt void mm_trace_rss_stat(struct mm_struct *mm, int member) 180b3d1411bSJoel Fernandes (Google) { 181f1a79412SShakeel Butt trace_rss_stat(mm, member); 182b3d1411bSJoel Fernandes (Google) } 183d559db08SKAMEZAWA Hiroyuki 1841da177e4SLinus Torvalds /* 1851da177e4SLinus Torvalds * Note: this doesn't free the actual pages themselves. That 1861da177e4SLinus Torvalds * has been handled earlier when unmapping all the memory regions. 1871da177e4SLinus Torvalds */ 1889e1b32caSBenjamin Herrenschmidt static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, 1899e1b32caSBenjamin Herrenschmidt unsigned long addr) 1901da177e4SLinus Torvalds { 1912f569afdSMartin Schwidefsky pgtable_t token = pmd_pgtable(*pmd); 1921da177e4SLinus Torvalds pmd_clear(pmd); 1939e1b32caSBenjamin Herrenschmidt pte_free_tlb(tlb, token, addr); 194c4812909SKirill A. Shutemov mm_dec_nr_ptes(tlb->mm); 1951da177e4SLinus Torvalds } 1961da177e4SLinus Torvalds 197e0da382cSHugh Dickins static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, 198e0da382cSHugh Dickins unsigned long addr, unsigned long end, 199e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 2001da177e4SLinus Torvalds { 2011da177e4SLinus Torvalds pmd_t *pmd; 2021da177e4SLinus Torvalds unsigned long next; 203e0da382cSHugh Dickins unsigned long start; 2041da177e4SLinus Torvalds 205e0da382cSHugh Dickins start = addr; 2061da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 2071da177e4SLinus Torvalds do { 2081da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 2091da177e4SLinus Torvalds if (pmd_none_or_clear_bad(pmd)) 2101da177e4SLinus Torvalds continue; 2119e1b32caSBenjamin Herrenschmidt free_pte_range(tlb, pmd, addr); 2121da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 2131da177e4SLinus Torvalds 214e0da382cSHugh Dickins start &= PUD_MASK; 215e0da382cSHugh Dickins if (start < floor) 216e0da382cSHugh Dickins return; 217e0da382cSHugh Dickins if (ceiling) { 218e0da382cSHugh Dickins ceiling &= PUD_MASK; 219e0da382cSHugh Dickins if (!ceiling) 220e0da382cSHugh Dickins return; 2211da177e4SLinus Torvalds } 222e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 223e0da382cSHugh Dickins return; 224e0da382cSHugh Dickins 225e0da382cSHugh Dickins pmd = pmd_offset(pud, start); 226e0da382cSHugh Dickins pud_clear(pud); 2279e1b32caSBenjamin Herrenschmidt pmd_free_tlb(tlb, pmd, start); 228dc6c9a35SKirill A. Shutemov mm_dec_nr_pmds(tlb->mm); 2291da177e4SLinus Torvalds } 2301da177e4SLinus Torvalds 231c2febafcSKirill A. Shutemov static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, 232e0da382cSHugh Dickins unsigned long addr, unsigned long end, 233e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 2341da177e4SLinus Torvalds { 2351da177e4SLinus Torvalds pud_t *pud; 2361da177e4SLinus Torvalds unsigned long next; 237e0da382cSHugh Dickins unsigned long start; 2381da177e4SLinus Torvalds 239e0da382cSHugh Dickins start = addr; 240c2febafcSKirill A. Shutemov pud = pud_offset(p4d, addr); 2411da177e4SLinus Torvalds do { 2421da177e4SLinus Torvalds next = pud_addr_end(addr, end); 2431da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 2441da177e4SLinus Torvalds continue; 245e0da382cSHugh Dickins free_pmd_range(tlb, pud, addr, next, floor, ceiling); 2461da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 2471da177e4SLinus Torvalds 248c2febafcSKirill A. Shutemov start &= P4D_MASK; 249c2febafcSKirill A. Shutemov if (start < floor) 250c2febafcSKirill A. Shutemov return; 251c2febafcSKirill A. Shutemov if (ceiling) { 252c2febafcSKirill A. Shutemov ceiling &= P4D_MASK; 253c2febafcSKirill A. Shutemov if (!ceiling) 254c2febafcSKirill A. Shutemov return; 255c2febafcSKirill A. Shutemov } 256c2febafcSKirill A. Shutemov if (end - 1 > ceiling - 1) 257c2febafcSKirill A. Shutemov return; 258c2febafcSKirill A. Shutemov 259c2febafcSKirill A. Shutemov pud = pud_offset(p4d, start); 260c2febafcSKirill A. Shutemov p4d_clear(p4d); 261c2febafcSKirill A. Shutemov pud_free_tlb(tlb, pud, start); 262b4e98d9aSKirill A. Shutemov mm_dec_nr_puds(tlb->mm); 263c2febafcSKirill A. Shutemov } 264c2febafcSKirill A. Shutemov 265c2febafcSKirill A. Shutemov static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, 266c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 267c2febafcSKirill A. Shutemov unsigned long floor, unsigned long ceiling) 268c2febafcSKirill A. Shutemov { 269c2febafcSKirill A. Shutemov p4d_t *p4d; 270c2febafcSKirill A. Shutemov unsigned long next; 271c2febafcSKirill A. Shutemov unsigned long start; 272c2febafcSKirill A. Shutemov 273c2febafcSKirill A. Shutemov start = addr; 274c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, addr); 275c2febafcSKirill A. Shutemov do { 276c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 277c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(p4d)) 278c2febafcSKirill A. Shutemov continue; 279c2febafcSKirill A. Shutemov free_pud_range(tlb, p4d, addr, next, floor, ceiling); 280c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 281c2febafcSKirill A. Shutemov 282e0da382cSHugh Dickins start &= PGDIR_MASK; 283e0da382cSHugh Dickins if (start < floor) 284e0da382cSHugh Dickins return; 285e0da382cSHugh Dickins if (ceiling) { 286e0da382cSHugh Dickins ceiling &= PGDIR_MASK; 287e0da382cSHugh Dickins if (!ceiling) 288e0da382cSHugh Dickins return; 2891da177e4SLinus Torvalds } 290e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 291e0da382cSHugh Dickins return; 292e0da382cSHugh Dickins 293c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, start); 294e0da382cSHugh Dickins pgd_clear(pgd); 295c2febafcSKirill A. Shutemov p4d_free_tlb(tlb, p4d, start); 2961da177e4SLinus Torvalds } 2971da177e4SLinus Torvalds 2981da177e4SLinus Torvalds /* 299e0da382cSHugh Dickins * This function frees user-level page tables of a process. 3001da177e4SLinus Torvalds */ 30142b77728SJan Beulich void free_pgd_range(struct mmu_gather *tlb, 302e0da382cSHugh Dickins unsigned long addr, unsigned long end, 303e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 3041da177e4SLinus Torvalds { 3051da177e4SLinus Torvalds pgd_t *pgd; 3061da177e4SLinus Torvalds unsigned long next; 3071da177e4SLinus Torvalds 308e0da382cSHugh Dickins /* 309e0da382cSHugh Dickins * The next few lines have given us lots of grief... 310e0da382cSHugh Dickins * 311e0da382cSHugh Dickins * Why are we testing PMD* at this top level? Because often 312e0da382cSHugh Dickins * there will be no work to do at all, and we'd prefer not to 313e0da382cSHugh Dickins * go all the way down to the bottom just to discover that. 314e0da382cSHugh Dickins * 315e0da382cSHugh Dickins * Why all these "- 1"s? Because 0 represents both the bottom 316e0da382cSHugh Dickins * of the address space and the top of it (using -1 for the 317e0da382cSHugh Dickins * top wouldn't help much: the masks would do the wrong thing). 318e0da382cSHugh Dickins * The rule is that addr 0 and floor 0 refer to the bottom of 319e0da382cSHugh Dickins * the address space, but end 0 and ceiling 0 refer to the top 320e0da382cSHugh Dickins * Comparisons need to use "end - 1" and "ceiling - 1" (though 321e0da382cSHugh Dickins * that end 0 case should be mythical). 322e0da382cSHugh Dickins * 323e0da382cSHugh Dickins * Wherever addr is brought up or ceiling brought down, we must 324e0da382cSHugh Dickins * be careful to reject "the opposite 0" before it confuses the 325e0da382cSHugh Dickins * subsequent tests. But what about where end is brought down 326e0da382cSHugh Dickins * by PMD_SIZE below? no, end can't go down to 0 there. 327e0da382cSHugh Dickins * 328e0da382cSHugh Dickins * Whereas we round start (addr) and ceiling down, by different 329e0da382cSHugh Dickins * masks at different levels, in order to test whether a table 330e0da382cSHugh Dickins * now has no other vmas using it, so can be freed, we don't 331e0da382cSHugh Dickins * bother to round floor or end up - the tests don't need that. 332e0da382cSHugh Dickins */ 333e0da382cSHugh Dickins 334e0da382cSHugh Dickins addr &= PMD_MASK; 335e0da382cSHugh Dickins if (addr < floor) { 336e0da382cSHugh Dickins addr += PMD_SIZE; 337e0da382cSHugh Dickins if (!addr) 338e0da382cSHugh Dickins return; 339e0da382cSHugh Dickins } 340e0da382cSHugh Dickins if (ceiling) { 341e0da382cSHugh Dickins ceiling &= PMD_MASK; 342e0da382cSHugh Dickins if (!ceiling) 343e0da382cSHugh Dickins return; 344e0da382cSHugh Dickins } 345e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 346e0da382cSHugh Dickins end -= PMD_SIZE; 347e0da382cSHugh Dickins if (addr > end - 1) 348e0da382cSHugh Dickins return; 34907e32661SAneesh Kumar K.V /* 35007e32661SAneesh Kumar K.V * We add page table cache pages with PAGE_SIZE, 35107e32661SAneesh Kumar K.V * (see pte_free_tlb()), flush the tlb if we need 35207e32661SAneesh Kumar K.V */ 353ed6a7935SPeter Zijlstra tlb_change_page_size(tlb, PAGE_SIZE); 35442b77728SJan Beulich pgd = pgd_offset(tlb->mm, addr); 3551da177e4SLinus Torvalds do { 3561da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 3571da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 3581da177e4SLinus Torvalds continue; 359c2febafcSKirill A. Shutemov free_p4d_range(tlb, pgd, addr, next, floor, ceiling); 3601da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 361e0da382cSHugh Dickins } 362e0da382cSHugh Dickins 363763ecb03SLiam R. Howlett void free_pgtables(struct mmu_gather *tlb, struct maple_tree *mt, 364763ecb03SLiam R. Howlett struct vm_area_struct *vma, unsigned long floor, 36598e51a22SSuren Baghdasaryan unsigned long ceiling, bool mm_wr_locked) 366e0da382cSHugh Dickins { 367763ecb03SLiam R. Howlett MA_STATE(mas, mt, vma->vm_end, vma->vm_end); 368763ecb03SLiam R. Howlett 369763ecb03SLiam R. Howlett do { 370e0da382cSHugh Dickins unsigned long addr = vma->vm_start; 371763ecb03SLiam R. Howlett struct vm_area_struct *next; 372763ecb03SLiam R. Howlett 373763ecb03SLiam R. Howlett /* 374763ecb03SLiam R. Howlett * Note: USER_PGTABLES_CEILING may be passed as ceiling and may 375763ecb03SLiam R. Howlett * be 0. This will underflow and is okay. 376763ecb03SLiam R. Howlett */ 377763ecb03SLiam R. Howlett next = mas_find(&mas, ceiling - 1); 378e0da382cSHugh Dickins 3798f4f8c16SHugh Dickins /* 38025d9e2d1Snpiggin@suse.de * Hide vma from rmap and truncate_pagecache before freeing 38125d9e2d1Snpiggin@suse.de * pgtables 3828f4f8c16SHugh Dickins */ 38398e51a22SSuren Baghdasaryan if (mm_wr_locked) 38498e51a22SSuren Baghdasaryan vma_start_write(vma); 3855beb4930SRik van Riel unlink_anon_vmas(vma); 3868f4f8c16SHugh Dickins unlink_file_vma(vma); 3878f4f8c16SHugh Dickins 3889da61aefSDavid Gibson if (is_vm_hugetlb_page(vma)) { 3893bf5ee95SHugh Dickins hugetlb_free_pgd_range(tlb, addr, vma->vm_end, 3903bf5ee95SHugh Dickins floor, next ? next->vm_start : ceiling); 3913bf5ee95SHugh Dickins } else { 3923bf5ee95SHugh Dickins /* 3933bf5ee95SHugh Dickins * Optimization: gather nearby vmas into one call down 3943bf5ee95SHugh Dickins */ 3953bf5ee95SHugh Dickins while (next && next->vm_start <= vma->vm_end + PMD_SIZE 3964866920bSDavid Gibson && !is_vm_hugetlb_page(next)) { 397e0da382cSHugh Dickins vma = next; 398763ecb03SLiam R. Howlett next = mas_find(&mas, ceiling - 1); 39998e51a22SSuren Baghdasaryan if (mm_wr_locked) 40098e51a22SSuren Baghdasaryan vma_start_write(vma); 4015beb4930SRik van Riel unlink_anon_vmas(vma); 4028f4f8c16SHugh Dickins unlink_file_vma(vma); 403e0da382cSHugh Dickins } 4043bf5ee95SHugh Dickins free_pgd_range(tlb, addr, vma->vm_end, 405e0da382cSHugh Dickins floor, next ? next->vm_start : ceiling); 4063bf5ee95SHugh Dickins } 407e0da382cSHugh Dickins vma = next; 408763ecb03SLiam R. Howlett } while (vma); 4091da177e4SLinus Torvalds } 4101da177e4SLinus Torvalds 41103c4f204SQi Zheng void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte) 4121da177e4SLinus Torvalds { 41303c4f204SQi Zheng spinlock_t *ptl = pmd_lock(mm, pmd); 4141bb3630eSHugh Dickins 41503c4f204SQi Zheng if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 41603c4f204SQi Zheng mm_inc_nr_ptes(mm); 417362a61adSNick Piggin /* 418362a61adSNick Piggin * Ensure all pte setup (eg. pte page lock and page clearing) are 419362a61adSNick Piggin * visible before the pte is made visible to other CPUs by being 420362a61adSNick Piggin * put into page tables. 421362a61adSNick Piggin * 422362a61adSNick Piggin * The other side of the story is the pointer chasing in the page 423362a61adSNick Piggin * table walking code (when walking the page table without locking; 424362a61adSNick Piggin * ie. most of the time). Fortunately, these data accesses consist 425362a61adSNick Piggin * of a chain of data-dependent loads, meaning most CPUs (alpha 426362a61adSNick Piggin * being the notable exception) will already guarantee loads are 427362a61adSNick Piggin * seen in-order. See the alpha page table accessors for the 428bb7cdd38SWill Deacon * smp_rmb() barriers in page table walking code. 429362a61adSNick Piggin */ 430362a61adSNick Piggin smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ 431ed33b5a6SQi Zheng pmd_populate(mm, pmd, *pte); 432ed33b5a6SQi Zheng *pte = NULL; 4334b471e88SKirill A. Shutemov } 434c4088ebdSKirill A. Shutemov spin_unlock(ptl); 435ed33b5a6SQi Zheng } 436ed33b5a6SQi Zheng 437ed33b5a6SQi Zheng int __pte_alloc(struct mm_struct *mm, pmd_t *pmd) 438ed33b5a6SQi Zheng { 439ed33b5a6SQi Zheng pgtable_t new = pte_alloc_one(mm); 440ed33b5a6SQi Zheng if (!new) 441ed33b5a6SQi Zheng return -ENOMEM; 4428ac1f832SAndrea Arcangeli 44303c4f204SQi Zheng pmd_install(mm, pmd, &new); 4442f569afdSMartin Schwidefsky if (new) 4452f569afdSMartin Schwidefsky pte_free(mm, new); 4461bb3630eSHugh Dickins return 0; 4471da177e4SLinus Torvalds } 4481da177e4SLinus Torvalds 4494cf58924SJoel Fernandes (Google) int __pte_alloc_kernel(pmd_t *pmd) 4501da177e4SLinus Torvalds { 4514cf58924SJoel Fernandes (Google) pte_t *new = pte_alloc_one_kernel(&init_mm); 4521da177e4SLinus Torvalds if (!new) 4531bb3630eSHugh Dickins return -ENOMEM; 4541da177e4SLinus Torvalds 455872fec16SHugh Dickins spin_lock(&init_mm.page_table_lock); 4568ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 457ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 458872fec16SHugh Dickins pmd_populate_kernel(&init_mm, pmd, new); 4592f569afdSMartin Schwidefsky new = NULL; 4604b471e88SKirill A. Shutemov } 461872fec16SHugh Dickins spin_unlock(&init_mm.page_table_lock); 4622f569afdSMartin Schwidefsky if (new) 4632f569afdSMartin Schwidefsky pte_free_kernel(&init_mm, new); 4641bb3630eSHugh Dickins return 0; 4651da177e4SLinus Torvalds } 4661da177e4SLinus Torvalds 467d559db08SKAMEZAWA Hiroyuki static inline void init_rss_vec(int *rss) 468ae859762SHugh Dickins { 469d559db08SKAMEZAWA Hiroyuki memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); 470d559db08SKAMEZAWA Hiroyuki } 471d559db08SKAMEZAWA Hiroyuki 472d559db08SKAMEZAWA Hiroyuki static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) 473d559db08SKAMEZAWA Hiroyuki { 474d559db08SKAMEZAWA Hiroyuki int i; 475d559db08SKAMEZAWA Hiroyuki 47634e55232SKAMEZAWA Hiroyuki if (current->mm == mm) 47705af2e10SDavid Rientjes sync_mm_rss(mm); 478d559db08SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) 479d559db08SKAMEZAWA Hiroyuki if (rss[i]) 480d559db08SKAMEZAWA Hiroyuki add_mm_counter(mm, i, rss[i]); 481ae859762SHugh Dickins } 482ae859762SHugh Dickins 4831da177e4SLinus Torvalds /* 4846aab341eSLinus Torvalds * This function is called to print an error when a bad pte 4856aab341eSLinus Torvalds * is found. For example, we might have a PFN-mapped pte in 4866aab341eSLinus Torvalds * a region that doesn't allow it. 487b5810039SNick Piggin * 488b5810039SNick Piggin * The calling function must still handle the error. 489b5810039SNick Piggin */ 4903dc14741SHugh Dickins static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, 4913dc14741SHugh Dickins pte_t pte, struct page *page) 492b5810039SNick Piggin { 4933dc14741SHugh Dickins pgd_t *pgd = pgd_offset(vma->vm_mm, addr); 494c2febafcSKirill A. Shutemov p4d_t *p4d = p4d_offset(pgd, addr); 495c2febafcSKirill A. Shutemov pud_t *pud = pud_offset(p4d, addr); 4963dc14741SHugh Dickins pmd_t *pmd = pmd_offset(pud, addr); 4973dc14741SHugh Dickins struct address_space *mapping; 4983dc14741SHugh Dickins pgoff_t index; 499d936cf9bSHugh Dickins static unsigned long resume; 500d936cf9bSHugh Dickins static unsigned long nr_shown; 501d936cf9bSHugh Dickins static unsigned long nr_unshown; 502d936cf9bSHugh Dickins 503d936cf9bSHugh Dickins /* 504d936cf9bSHugh Dickins * Allow a burst of 60 reports, then keep quiet for that minute; 505d936cf9bSHugh Dickins * or allow a steady drip of one report per second. 506d936cf9bSHugh Dickins */ 507d936cf9bSHugh Dickins if (nr_shown == 60) { 508d936cf9bSHugh Dickins if (time_before(jiffies, resume)) { 509d936cf9bSHugh Dickins nr_unshown++; 510d936cf9bSHugh Dickins return; 511d936cf9bSHugh Dickins } 512d936cf9bSHugh Dickins if (nr_unshown) { 5131170532bSJoe Perches pr_alert("BUG: Bad page map: %lu messages suppressed\n", 514d936cf9bSHugh Dickins nr_unshown); 515d936cf9bSHugh Dickins nr_unshown = 0; 516d936cf9bSHugh Dickins } 517d936cf9bSHugh Dickins nr_shown = 0; 518d936cf9bSHugh Dickins } 519d936cf9bSHugh Dickins if (nr_shown++ == 0) 520d936cf9bSHugh Dickins resume = jiffies + 60 * HZ; 5213dc14741SHugh Dickins 5223dc14741SHugh Dickins mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; 5233dc14741SHugh Dickins index = linear_page_index(vma, addr); 5243dc14741SHugh Dickins 5251170532bSJoe Perches pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", 5263dc14741SHugh Dickins current->comm, 5273dc14741SHugh Dickins (long long)pte_val(pte), (long long)pmd_val(*pmd)); 528718a3821SWu Fengguang if (page) 529f0b791a3SDave Hansen dump_page(page, "bad pte"); 5306aa9b8b2SKefeng Wang pr_alert("addr:%px vm_flags:%08lx anon_vma:%px mapping:%px index:%lx\n", 5313dc14741SHugh Dickins (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); 5327e0a1265SMatthew Wilcox (Oracle) pr_alert("file:%pD fault:%ps mmap:%ps read_folio:%ps\n", 5332682582aSKonstantin Khlebnikov vma->vm_file, 5342682582aSKonstantin Khlebnikov vma->vm_ops ? vma->vm_ops->fault : NULL, 5352682582aSKonstantin Khlebnikov vma->vm_file ? vma->vm_file->f_op->mmap : NULL, 5367e0a1265SMatthew Wilcox (Oracle) mapping ? mapping->a_ops->read_folio : NULL); 537b5810039SNick Piggin dump_stack(); 538373d4d09SRusty Russell add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); 539b5810039SNick Piggin } 540b5810039SNick Piggin 541b5810039SNick Piggin /* 5427e675137SNick Piggin * vm_normal_page -- This function gets the "struct page" associated with a pte. 5436aab341eSLinus Torvalds * 5447e675137SNick Piggin * "Special" mappings do not wish to be associated with a "struct page" (either 5457e675137SNick Piggin * it doesn't exist, or it exists but they don't want to touch it). In this 5467e675137SNick Piggin * case, NULL is returned here. "Normal" mappings do have a struct page. 547b379d790SJared Hulbert * 5487e675137SNick Piggin * There are 2 broad cases. Firstly, an architecture may define a pte_special() 5497e675137SNick Piggin * pte bit, in which case this function is trivial. Secondly, an architecture 5507e675137SNick Piggin * may not have a spare pte bit, which requires a more complicated scheme, 5517e675137SNick Piggin * described below. 5527e675137SNick Piggin * 5537e675137SNick Piggin * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a 5547e675137SNick Piggin * special mapping (even if there are underlying and valid "struct pages"). 5557e675137SNick Piggin * COWed pages of a VM_PFNMAP are always normal. 5566aab341eSLinus Torvalds * 557b379d790SJared Hulbert * The way we recognize COWed pages within VM_PFNMAP mappings is through the 558b379d790SJared Hulbert * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit 5597e675137SNick Piggin * set, and the vm_pgoff will point to the first PFN mapped: thus every special 5607e675137SNick Piggin * mapping will always honor the rule 5616aab341eSLinus Torvalds * 5626aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 5636aab341eSLinus Torvalds * 5647e675137SNick Piggin * And for normal mappings this is false. 565b379d790SJared Hulbert * 5667e675137SNick Piggin * This restricts such mappings to be a linear translation from virtual address 5677e675137SNick Piggin * to pfn. To get around this restriction, we allow arbitrary mappings so long 5687e675137SNick Piggin * as the vma is not a COW mapping; in that case, we know that all ptes are 5697e675137SNick Piggin * special (because none can have been COWed). 570b379d790SJared Hulbert * 571b379d790SJared Hulbert * 5727e675137SNick Piggin * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. 5737e675137SNick Piggin * 574b379d790SJared Hulbert * VM_MIXEDMAP mappings can likewise contain memory with or without "struct 575b379d790SJared Hulbert * page" backing, however the difference is that _all_ pages with a struct 576b379d790SJared Hulbert * page (that is, those where pfn_valid is true) are refcounted and considered 577b379d790SJared Hulbert * normal pages by the VM. The disadvantage is that pages are refcounted 578b379d790SJared Hulbert * (which can be slower and simply not an option for some PFNMAP users). The 579b379d790SJared Hulbert * advantage is that we don't have to follow the strict linearity rule of 580b379d790SJared Hulbert * PFNMAP mappings in order to support COWable mappings. 581b379d790SJared Hulbert * 582ee498ed7SHugh Dickins */ 58325b2995aSChristoph Hellwig struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 58425b2995aSChristoph Hellwig pte_t pte) 585ee498ed7SHugh Dickins { 58622b31eecSHugh Dickins unsigned long pfn = pte_pfn(pte); 5877e675137SNick Piggin 58800b3a331SLaurent Dufour if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) { 589b38af472SHugh Dickins if (likely(!pte_special(pte))) 59022b31eecSHugh Dickins goto check_pfn; 591667a0a06SDavid Vrabel if (vma->vm_ops && vma->vm_ops->find_special_page) 592667a0a06SDavid Vrabel return vma->vm_ops->find_special_page(vma, addr); 593a13ea5b7SHugh Dickins if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 594a13ea5b7SHugh Dickins return NULL; 595df6ad698SJérôme Glisse if (is_zero_pfn(pfn)) 596df6ad698SJérôme Glisse return NULL; 597e1fb4a08SDave Jiang if (pte_devmap(pte)) 5983218f871SAlex Sierra /* 5993218f871SAlex Sierra * NOTE: New users of ZONE_DEVICE will not set pte_devmap() 6003218f871SAlex Sierra * and will have refcounts incremented on their struct pages 6013218f871SAlex Sierra * when they are inserted into PTEs, thus they are safe to 6023218f871SAlex Sierra * return here. Legacy ZONE_DEVICE pages that set pte_devmap() 6033218f871SAlex Sierra * do not have refcounts. Example of legacy ZONE_DEVICE is 6043218f871SAlex Sierra * MEMORY_DEVICE_FS_DAX type in pmem or virtio_fs drivers. 6053218f871SAlex Sierra */ 606e1fb4a08SDave Jiang return NULL; 607e1fb4a08SDave Jiang 60822b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 6097e675137SNick Piggin return NULL; 6107e675137SNick Piggin } 6117e675137SNick Piggin 61200b3a331SLaurent Dufour /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */ 6137e675137SNick Piggin 614b379d790SJared Hulbert if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 615b379d790SJared Hulbert if (vma->vm_flags & VM_MIXEDMAP) { 616b379d790SJared Hulbert if (!pfn_valid(pfn)) 617b379d790SJared Hulbert return NULL; 618b379d790SJared Hulbert goto out; 619b379d790SJared Hulbert } else { 6207e675137SNick Piggin unsigned long off; 6217e675137SNick Piggin off = (addr - vma->vm_start) >> PAGE_SHIFT; 6226aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 6236aab341eSLinus Torvalds return NULL; 62467121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 625fb155c16SLinus Torvalds return NULL; 6266aab341eSLinus Torvalds } 627b379d790SJared Hulbert } 6286aab341eSLinus Torvalds 629b38af472SHugh Dickins if (is_zero_pfn(pfn)) 630b38af472SHugh Dickins return NULL; 63100b3a331SLaurent Dufour 63222b31eecSHugh Dickins check_pfn: 63322b31eecSHugh Dickins if (unlikely(pfn > highest_memmap_pfn)) { 63422b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 63522b31eecSHugh Dickins return NULL; 63622b31eecSHugh Dickins } 6376aab341eSLinus Torvalds 6386aab341eSLinus Torvalds /* 6397e675137SNick Piggin * NOTE! We still have PageReserved() pages in the page tables. 6407e675137SNick Piggin * eg. VDSO mappings can cause them to exist. 6416aab341eSLinus Torvalds */ 642b379d790SJared Hulbert out: 6436aab341eSLinus Torvalds return pfn_to_page(pfn); 644ee498ed7SHugh Dickins } 645ee498ed7SHugh Dickins 646318e9342SVishal Moola (Oracle) struct folio *vm_normal_folio(struct vm_area_struct *vma, unsigned long addr, 647318e9342SVishal Moola (Oracle) pte_t pte) 648318e9342SVishal Moola (Oracle) { 649318e9342SVishal Moola (Oracle) struct page *page = vm_normal_page(vma, addr, pte); 650318e9342SVishal Moola (Oracle) 651318e9342SVishal Moola (Oracle) if (page) 652318e9342SVishal Moola (Oracle) return page_folio(page); 653318e9342SVishal Moola (Oracle) return NULL; 654318e9342SVishal Moola (Oracle) } 655318e9342SVishal Moola (Oracle) 65628093f9fSGerald Schaefer #ifdef CONFIG_TRANSPARENT_HUGEPAGE 65728093f9fSGerald Schaefer struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, 65828093f9fSGerald Schaefer pmd_t pmd) 65928093f9fSGerald Schaefer { 66028093f9fSGerald Schaefer unsigned long pfn = pmd_pfn(pmd); 66128093f9fSGerald Schaefer 66228093f9fSGerald Schaefer /* 66328093f9fSGerald Schaefer * There is no pmd_special() but there may be special pmds, e.g. 66428093f9fSGerald Schaefer * in a direct-access (dax) mapping, so let's just replicate the 66500b3a331SLaurent Dufour * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here. 66628093f9fSGerald Schaefer */ 66728093f9fSGerald Schaefer if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 66828093f9fSGerald Schaefer if (vma->vm_flags & VM_MIXEDMAP) { 66928093f9fSGerald Schaefer if (!pfn_valid(pfn)) 67028093f9fSGerald Schaefer return NULL; 67128093f9fSGerald Schaefer goto out; 67228093f9fSGerald Schaefer } else { 67328093f9fSGerald Schaefer unsigned long off; 67428093f9fSGerald Schaefer off = (addr - vma->vm_start) >> PAGE_SHIFT; 67528093f9fSGerald Schaefer if (pfn == vma->vm_pgoff + off) 67628093f9fSGerald Schaefer return NULL; 67728093f9fSGerald Schaefer if (!is_cow_mapping(vma->vm_flags)) 67828093f9fSGerald Schaefer return NULL; 67928093f9fSGerald Schaefer } 68028093f9fSGerald Schaefer } 68128093f9fSGerald Schaefer 682e1fb4a08SDave Jiang if (pmd_devmap(pmd)) 683e1fb4a08SDave Jiang return NULL; 6843cde287bSYu Zhao if (is_huge_zero_pmd(pmd)) 68528093f9fSGerald Schaefer return NULL; 68628093f9fSGerald Schaefer if (unlikely(pfn > highest_memmap_pfn)) 68728093f9fSGerald Schaefer return NULL; 68828093f9fSGerald Schaefer 68928093f9fSGerald Schaefer /* 69028093f9fSGerald Schaefer * NOTE! We still have PageReserved() pages in the page tables. 69128093f9fSGerald Schaefer * eg. VDSO mappings can cause them to exist. 69228093f9fSGerald Schaefer */ 69328093f9fSGerald Schaefer out: 69428093f9fSGerald Schaefer return pfn_to_page(pfn); 69528093f9fSGerald Schaefer } 69628093f9fSGerald Schaefer #endif 69728093f9fSGerald Schaefer 698b756a3b5SAlistair Popple static void restore_exclusive_pte(struct vm_area_struct *vma, 699b756a3b5SAlistair Popple struct page *page, unsigned long address, 700b756a3b5SAlistair Popple pte_t *ptep) 701b756a3b5SAlistair Popple { 702b756a3b5SAlistair Popple pte_t pte; 703b756a3b5SAlistair Popple swp_entry_t entry; 704b756a3b5SAlistair Popple 705b756a3b5SAlistair Popple pte = pte_mkold(mk_pte(page, READ_ONCE(vma->vm_page_prot))); 706b756a3b5SAlistair Popple if (pte_swp_soft_dirty(*ptep)) 707b756a3b5SAlistair Popple pte = pte_mksoft_dirty(pte); 708b756a3b5SAlistair Popple 709b756a3b5SAlistair Popple entry = pte_to_swp_entry(*ptep); 710b756a3b5SAlistair Popple if (pte_swp_uffd_wp(*ptep)) 711b756a3b5SAlistair Popple pte = pte_mkuffd_wp(pte); 712b756a3b5SAlistair Popple else if (is_writable_device_exclusive_entry(entry)) 713b756a3b5SAlistair Popple pte = maybe_mkwrite(pte_mkdirty(pte), vma); 714b756a3b5SAlistair Popple 7156c287605SDavid Hildenbrand VM_BUG_ON(pte_write(pte) && !(PageAnon(page) && PageAnonExclusive(page))); 7166c287605SDavid Hildenbrand 717b756a3b5SAlistair Popple /* 718b756a3b5SAlistair Popple * No need to take a page reference as one was already 719b756a3b5SAlistair Popple * created when the swap entry was made. 720b756a3b5SAlistair Popple */ 721b756a3b5SAlistair Popple if (PageAnon(page)) 722f1e2db12SDavid Hildenbrand page_add_anon_rmap(page, vma, address, RMAP_NONE); 723b756a3b5SAlistair Popple else 724b756a3b5SAlistair Popple /* 725b756a3b5SAlistair Popple * Currently device exclusive access only supports anonymous 726b756a3b5SAlistair Popple * memory so the entry shouldn't point to a filebacked page. 727b756a3b5SAlistair Popple */ 7284d8ff640SMiaohe Lin WARN_ON_ONCE(1); 729b756a3b5SAlistair Popple 7301eba86c0SPasha Tatashin set_pte_at(vma->vm_mm, address, ptep, pte); 7311eba86c0SPasha Tatashin 732b756a3b5SAlistair Popple /* 733b756a3b5SAlistair Popple * No need to invalidate - it was non-present before. However 734b756a3b5SAlistair Popple * secondary CPUs may have mappings that need invalidating. 735b756a3b5SAlistair Popple */ 736b756a3b5SAlistair Popple update_mmu_cache(vma, address, ptep); 737b756a3b5SAlistair Popple } 738b756a3b5SAlistair Popple 739b756a3b5SAlistair Popple /* 740b756a3b5SAlistair Popple * Tries to restore an exclusive pte if the page lock can be acquired without 741b756a3b5SAlistair Popple * sleeping. 742b756a3b5SAlistair Popple */ 743b756a3b5SAlistair Popple static int 744b756a3b5SAlistair Popple try_restore_exclusive_pte(pte_t *src_pte, struct vm_area_struct *vma, 745b756a3b5SAlistair Popple unsigned long addr) 746b756a3b5SAlistair Popple { 747b756a3b5SAlistair Popple swp_entry_t entry = pte_to_swp_entry(*src_pte); 748b756a3b5SAlistair Popple struct page *page = pfn_swap_entry_to_page(entry); 749b756a3b5SAlistair Popple 750b756a3b5SAlistair Popple if (trylock_page(page)) { 751b756a3b5SAlistair Popple restore_exclusive_pte(vma, page, addr, src_pte); 752b756a3b5SAlistair Popple unlock_page(page); 753b756a3b5SAlistair Popple return 0; 754b756a3b5SAlistair Popple } 755b756a3b5SAlistair Popple 756b756a3b5SAlistair Popple return -EBUSY; 757b756a3b5SAlistair Popple } 758b756a3b5SAlistair Popple 759ee498ed7SHugh Dickins /* 7601da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 7611da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 7621da177e4SLinus Torvalds * covered by this vma. 7631da177e4SLinus Torvalds */ 7641da177e4SLinus Torvalds 765df3a57d1SLinus Torvalds static unsigned long 766df3a57d1SLinus Torvalds copy_nonpresent_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 7678f34f1eaSPeter Xu pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *dst_vma, 7688f34f1eaSPeter Xu struct vm_area_struct *src_vma, unsigned long addr, int *rss) 7691da177e4SLinus Torvalds { 7708f34f1eaSPeter Xu unsigned long vm_flags = dst_vma->vm_flags; 7711da177e4SLinus Torvalds pte_t pte = *src_pte; 7721da177e4SLinus Torvalds struct page *page; 7730697212aSChristoph Lameter swp_entry_t entry = pte_to_swp_entry(pte); 7740697212aSChristoph Lameter 7752022b4d1SHugh Dickins if (likely(!non_swap_entry(entry))) { 776570a335bSHugh Dickins if (swap_duplicate(entry) < 0) 7779a5cc85cSAlistair Popple return -EIO; 778570a335bSHugh Dickins 7791da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 7801da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 7811da177e4SLinus Torvalds spin_lock(&mmlist_lock); 782f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 783f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 784f412ac08SHugh Dickins &src_mm->mmlist); 7851da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 7861da177e4SLinus Torvalds } 7871493a191SDavid Hildenbrand /* Mark the swap entry as shared. */ 7881493a191SDavid Hildenbrand if (pte_swp_exclusive(*src_pte)) { 7891493a191SDavid Hildenbrand pte = pte_swp_clear_exclusive(*src_pte); 7901493a191SDavid Hildenbrand set_pte_at(src_mm, addr, src_pte, pte); 7911493a191SDavid Hildenbrand } 792b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]++; 7932022b4d1SHugh Dickins } else if (is_migration_entry(entry)) { 794af5cdaf8SAlistair Popple page = pfn_swap_entry_to_page(entry); 7959f9f1acdSKonstantin Khlebnikov 796eca56ff9SJerome Marchand rss[mm_counter(page)]++; 7979f9f1acdSKonstantin Khlebnikov 7986c287605SDavid Hildenbrand if (!is_readable_migration_entry(entry) && 7990697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 8000697212aSChristoph Lameter /* 8016c287605SDavid Hildenbrand * COW mappings require pages in both parent and child 8026c287605SDavid Hildenbrand * to be set to read. A previously exclusive entry is 8036c287605SDavid Hildenbrand * now shared. 8040697212aSChristoph Lameter */ 8054dd845b5SAlistair Popple entry = make_readable_migration_entry( 8064dd845b5SAlistair Popple swp_offset(entry)); 8070697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 808c3d16e16SCyrill Gorcunov if (pte_swp_soft_dirty(*src_pte)) 809c3d16e16SCyrill Gorcunov pte = pte_swp_mksoft_dirty(pte); 810f45ec5ffSPeter Xu if (pte_swp_uffd_wp(*src_pte)) 811f45ec5ffSPeter Xu pte = pte_swp_mkuffd_wp(pte); 8120697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 8130697212aSChristoph Lameter } 8145042db43SJérôme Glisse } else if (is_device_private_entry(entry)) { 815af5cdaf8SAlistair Popple page = pfn_swap_entry_to_page(entry); 8165042db43SJérôme Glisse 8175042db43SJérôme Glisse /* 8185042db43SJérôme Glisse * Update rss count even for unaddressable pages, as 8195042db43SJérôme Glisse * they should treated just like normal pages in this 8205042db43SJérôme Glisse * respect. 8215042db43SJérôme Glisse * 8225042db43SJérôme Glisse * We will likely want to have some new rss counters 8235042db43SJérôme Glisse * for unaddressable pages, at some point. But for now 8245042db43SJérôme Glisse * keep things as they are. 8255042db43SJérôme Glisse */ 8265042db43SJérôme Glisse get_page(page); 8275042db43SJérôme Glisse rss[mm_counter(page)]++; 828fb3d824dSDavid Hildenbrand /* Cannot fail as these pages cannot get pinned. */ 829fb3d824dSDavid Hildenbrand BUG_ON(page_try_dup_anon_rmap(page, false, src_vma)); 8305042db43SJérôme Glisse 8315042db43SJérôme Glisse /* 8325042db43SJérôme Glisse * We do not preserve soft-dirty information, because so 8335042db43SJérôme Glisse * far, checkpoint/restore is the only feature that 8345042db43SJérôme Glisse * requires that. And checkpoint/restore does not work 8355042db43SJérôme Glisse * when a device driver is involved (you cannot easily 8365042db43SJérôme Glisse * save and restore device driver state). 8375042db43SJérôme Glisse */ 8384dd845b5SAlistair Popple if (is_writable_device_private_entry(entry) && 8395042db43SJérôme Glisse is_cow_mapping(vm_flags)) { 8404dd845b5SAlistair Popple entry = make_readable_device_private_entry( 8414dd845b5SAlistair Popple swp_offset(entry)); 8425042db43SJérôme Glisse pte = swp_entry_to_pte(entry); 843f45ec5ffSPeter Xu if (pte_swp_uffd_wp(*src_pte)) 844f45ec5ffSPeter Xu pte = pte_swp_mkuffd_wp(pte); 8455042db43SJérôme Glisse set_pte_at(src_mm, addr, src_pte, pte); 8465042db43SJérôme Glisse } 847b756a3b5SAlistair Popple } else if (is_device_exclusive_entry(entry)) { 848b756a3b5SAlistair Popple /* 849b756a3b5SAlistair Popple * Make device exclusive entries present by restoring the 850b756a3b5SAlistair Popple * original entry then copying as for a present pte. Device 851b756a3b5SAlistair Popple * exclusive entries currently only support private writable 852b756a3b5SAlistair Popple * (ie. COW) mappings. 853b756a3b5SAlistair Popple */ 854b756a3b5SAlistair Popple VM_BUG_ON(!is_cow_mapping(src_vma->vm_flags)); 855b756a3b5SAlistair Popple if (try_restore_exclusive_pte(src_pte, src_vma, addr)) 856b756a3b5SAlistair Popple return -EBUSY; 857b756a3b5SAlistair Popple return -ENOENT; 858c56d1b62SPeter Xu } else if (is_pte_marker_entry(entry)) { 8597e3ce3f8SPeter Xu if (is_swapin_error_entry(entry) || userfaultfd_wp(dst_vma)) 860c56d1b62SPeter Xu set_pte_at(dst_mm, addr, dst_pte, pte); 861c56d1b62SPeter Xu return 0; 8621da177e4SLinus Torvalds } 8638f34f1eaSPeter Xu if (!userfaultfd_wp(dst_vma)) 8648f34f1eaSPeter Xu pte = pte_swp_clear_uffd_wp(pte); 865df3a57d1SLinus Torvalds set_pte_at(dst_mm, addr, dst_pte, pte); 866df3a57d1SLinus Torvalds return 0; 8671da177e4SLinus Torvalds } 8681da177e4SLinus Torvalds 86970e806e4SPeter Xu /* 870b51ad4f8SDavid Hildenbrand * Copy a present and normal page. 87170e806e4SPeter Xu * 872b51ad4f8SDavid Hildenbrand * NOTE! The usual case is that this isn't required; 873b51ad4f8SDavid Hildenbrand * instead, the caller can just increase the page refcount 874b51ad4f8SDavid Hildenbrand * and re-use the pte the traditional way. 87570e806e4SPeter Xu * 87670e806e4SPeter Xu * And if we need a pre-allocated page but don't yet have 87770e806e4SPeter Xu * one, return a negative error to let the preallocation 87870e806e4SPeter Xu * code know so that it can do so outside the page table 87970e806e4SPeter Xu * lock. 88070e806e4SPeter Xu */ 88170e806e4SPeter Xu static inline int 882c78f4636SPeter Xu copy_present_page(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 883c78f4636SPeter Xu pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, 884edf50470SMatthew Wilcox (Oracle) struct folio **prealloc, struct page *page) 88570e806e4SPeter Xu { 886edf50470SMatthew Wilcox (Oracle) struct folio *new_folio; 887b51ad4f8SDavid Hildenbrand pte_t pte; 88870e806e4SPeter Xu 889edf50470SMatthew Wilcox (Oracle) new_folio = *prealloc; 890edf50470SMatthew Wilcox (Oracle) if (!new_folio) 89170e806e4SPeter Xu return -EAGAIN; 89270e806e4SPeter Xu 89370e806e4SPeter Xu /* 89470e806e4SPeter Xu * We have a prealloc page, all good! Take it 89570e806e4SPeter Xu * over and copy the page & arm it. 89670e806e4SPeter Xu */ 89770e806e4SPeter Xu *prealloc = NULL; 898edf50470SMatthew Wilcox (Oracle) copy_user_highpage(&new_folio->page, page, addr, src_vma); 899edf50470SMatthew Wilcox (Oracle) __folio_mark_uptodate(new_folio); 900edf50470SMatthew Wilcox (Oracle) folio_add_new_anon_rmap(new_folio, dst_vma, addr); 901edf50470SMatthew Wilcox (Oracle) folio_add_lru_vma(new_folio, dst_vma); 902edf50470SMatthew Wilcox (Oracle) rss[MM_ANONPAGES]++; 90370e806e4SPeter Xu 90470e806e4SPeter Xu /* All done, just insert the new page copy in the child */ 905edf50470SMatthew Wilcox (Oracle) pte = mk_pte(&new_folio->page, dst_vma->vm_page_prot); 906c78f4636SPeter Xu pte = maybe_mkwrite(pte_mkdirty(pte), dst_vma); 9078f34f1eaSPeter Xu if (userfaultfd_pte_wp(dst_vma, *src_pte)) 9088f34f1eaSPeter Xu /* Uffd-wp needs to be delivered to dest pte as well */ 909f1eb1bacSPeter Xu pte = pte_mkuffd_wp(pte); 910c78f4636SPeter Xu set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); 91170e806e4SPeter Xu return 0; 91270e806e4SPeter Xu } 91370e806e4SPeter Xu 91470e806e4SPeter Xu /* 91570e806e4SPeter Xu * Copy one pte. Returns 0 if succeeded, or -EAGAIN if one preallocated page 91670e806e4SPeter Xu * is required to copy this pte. 91770e806e4SPeter Xu */ 91870e806e4SPeter Xu static inline int 919c78f4636SPeter Xu copy_present_pte(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 920c78f4636SPeter Xu pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, 921edf50470SMatthew Wilcox (Oracle) struct folio **prealloc) 922df3a57d1SLinus Torvalds { 923c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 924c78f4636SPeter Xu unsigned long vm_flags = src_vma->vm_flags; 925df3a57d1SLinus Torvalds pte_t pte = *src_pte; 926df3a57d1SLinus Torvalds struct page *page; 92714ddee41SMatthew Wilcox (Oracle) struct folio *folio; 928df3a57d1SLinus Torvalds 929c78f4636SPeter Xu page = vm_normal_page(src_vma, addr, pte); 93014ddee41SMatthew Wilcox (Oracle) if (page) 93114ddee41SMatthew Wilcox (Oracle) folio = page_folio(page); 93214ddee41SMatthew Wilcox (Oracle) if (page && folio_test_anon(folio)) { 933b51ad4f8SDavid Hildenbrand /* 934b51ad4f8SDavid Hildenbrand * If this page may have been pinned by the parent process, 935b51ad4f8SDavid Hildenbrand * copy the page immediately for the child so that we'll always 936b51ad4f8SDavid Hildenbrand * guarantee the pinned page won't be randomly replaced in the 937b51ad4f8SDavid Hildenbrand * future. 938b51ad4f8SDavid Hildenbrand */ 93914ddee41SMatthew Wilcox (Oracle) folio_get(folio); 940fb3d824dSDavid Hildenbrand if (unlikely(page_try_dup_anon_rmap(page, false, src_vma))) { 941fb3d824dSDavid Hildenbrand /* Page may be pinned, we have to copy. */ 94214ddee41SMatthew Wilcox (Oracle) folio_put(folio); 943b51ad4f8SDavid Hildenbrand return copy_present_page(dst_vma, src_vma, dst_pte, src_pte, 944b51ad4f8SDavid Hildenbrand addr, rss, prealloc, page); 945fb3d824dSDavid Hildenbrand } 946edf50470SMatthew Wilcox (Oracle) rss[MM_ANONPAGES]++; 947b51ad4f8SDavid Hildenbrand } else if (page) { 94814ddee41SMatthew Wilcox (Oracle) folio_get(folio); 949fb3d824dSDavid Hildenbrand page_dup_file_rmap(page, false); 950edf50470SMatthew Wilcox (Oracle) rss[mm_counter_file(page)]++; 95170e806e4SPeter Xu } 95270e806e4SPeter Xu 9531da177e4SLinus Torvalds /* 9541da177e4SLinus Torvalds * If it's a COW mapping, write protect it both 9551da177e4SLinus Torvalds * in the parent and the child 9561da177e4SLinus Torvalds */ 9571b2de5d0SLinus Torvalds if (is_cow_mapping(vm_flags) && pte_write(pte)) { 9581da177e4SLinus Torvalds ptep_set_wrprotect(src_mm, addr, src_pte); 9593dc90795SZachary Amsden pte = pte_wrprotect(pte); 9601da177e4SLinus Torvalds } 96114ddee41SMatthew Wilcox (Oracle) VM_BUG_ON(page && folio_test_anon(folio) && PageAnonExclusive(page)); 9621da177e4SLinus Torvalds 9631da177e4SLinus Torvalds /* 9641da177e4SLinus Torvalds * If it's a shared mapping, mark it clean in 9651da177e4SLinus Torvalds * the child 9661da177e4SLinus Torvalds */ 9671da177e4SLinus Torvalds if (vm_flags & VM_SHARED) 9681da177e4SLinus Torvalds pte = pte_mkclean(pte); 9691da177e4SLinus Torvalds pte = pte_mkold(pte); 9706aab341eSLinus Torvalds 9718f34f1eaSPeter Xu if (!userfaultfd_wp(dst_vma)) 972b569a176SPeter Xu pte = pte_clear_uffd_wp(pte); 973b569a176SPeter Xu 974c78f4636SPeter Xu set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); 97570e806e4SPeter Xu return 0; 9766aab341eSLinus Torvalds } 977ae859762SHugh Dickins 978edf50470SMatthew Wilcox (Oracle) static inline struct folio *page_copy_prealloc(struct mm_struct *src_mm, 979edf50470SMatthew Wilcox (Oracle) struct vm_area_struct *vma, unsigned long addr) 98070e806e4SPeter Xu { 981edf50470SMatthew Wilcox (Oracle) struct folio *new_folio; 98270e806e4SPeter Xu 983edf50470SMatthew Wilcox (Oracle) new_folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma, addr, false); 984edf50470SMatthew Wilcox (Oracle) if (!new_folio) 98570e806e4SPeter Xu return NULL; 98670e806e4SPeter Xu 987edf50470SMatthew Wilcox (Oracle) if (mem_cgroup_charge(new_folio, src_mm, GFP_KERNEL)) { 988edf50470SMatthew Wilcox (Oracle) folio_put(new_folio); 98970e806e4SPeter Xu return NULL; 99070e806e4SPeter Xu } 991e601ded4SKefeng Wang folio_throttle_swaprate(new_folio, GFP_KERNEL); 99270e806e4SPeter Xu 993edf50470SMatthew Wilcox (Oracle) return new_folio; 9941da177e4SLinus Torvalds } 9951da177e4SLinus Torvalds 996c78f4636SPeter Xu static int 997c78f4636SPeter Xu copy_pte_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 998c78f4636SPeter Xu pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr, 999c78f4636SPeter Xu unsigned long end) 10001da177e4SLinus Torvalds { 1001c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1002c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 1003c36987e2SDaisuke Nishimura pte_t *orig_src_pte, *orig_dst_pte; 10041da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 1005c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 100670e806e4SPeter Xu int progress, ret = 0; 1007d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 1008570a335bSHugh Dickins swp_entry_t entry = (swp_entry_t){0}; 1009edf50470SMatthew Wilcox (Oracle) struct folio *prealloc = NULL; 10101da177e4SLinus Torvalds 10111da177e4SLinus Torvalds again: 101270e806e4SPeter Xu progress = 0; 1013d559db08SKAMEZAWA Hiroyuki init_rss_vec(rss); 1014d559db08SKAMEZAWA Hiroyuki 1015c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 101670e806e4SPeter Xu if (!dst_pte) { 101770e806e4SPeter Xu ret = -ENOMEM; 101870e806e4SPeter Xu goto out; 101970e806e4SPeter Xu } 1020ece0e2b6SPeter Zijlstra src_pte = pte_offset_map(src_pmd, addr); 10214c21e2f2SHugh Dickins src_ptl = pte_lockptr(src_mm, src_pmd); 1022f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 1023c36987e2SDaisuke Nishimura orig_src_pte = src_pte; 1024c36987e2SDaisuke Nishimura orig_dst_pte = dst_pte; 10256606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 10261da177e4SLinus Torvalds 10271da177e4SLinus Torvalds do { 10281da177e4SLinus Torvalds /* 10291da177e4SLinus Torvalds * We are holding two locks at this point - either of them 10301da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 10311da177e4SLinus Torvalds */ 1032e040f218SHugh Dickins if (progress >= 32) { 1033e040f218SHugh Dickins progress = 0; 1034e040f218SHugh Dickins if (need_resched() || 103595c354feSNick Piggin spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) 10361da177e4SLinus Torvalds break; 1037e040f218SHugh Dickins } 10381da177e4SLinus Torvalds if (pte_none(*src_pte)) { 10391da177e4SLinus Torvalds progress++; 10401da177e4SLinus Torvalds continue; 10411da177e4SLinus Torvalds } 104279a1971cSLinus Torvalds if (unlikely(!pte_present(*src_pte))) { 10439a5cc85cSAlistair Popple ret = copy_nonpresent_pte(dst_mm, src_mm, 104479a1971cSLinus Torvalds dst_pte, src_pte, 10458f34f1eaSPeter Xu dst_vma, src_vma, 10468f34f1eaSPeter Xu addr, rss); 10479a5cc85cSAlistair Popple if (ret == -EIO) { 10489a5cc85cSAlistair Popple entry = pte_to_swp_entry(*src_pte); 1049570a335bSHugh Dickins break; 1050b756a3b5SAlistair Popple } else if (ret == -EBUSY) { 1051b756a3b5SAlistair Popple break; 1052b756a3b5SAlistair Popple } else if (!ret) { 10531da177e4SLinus Torvalds progress += 8; 105479a1971cSLinus Torvalds continue; 105579a1971cSLinus Torvalds } 1056b756a3b5SAlistair Popple 1057b756a3b5SAlistair Popple /* 1058b756a3b5SAlistair Popple * Device exclusive entry restored, continue by copying 1059b756a3b5SAlistair Popple * the now present pte. 1060b756a3b5SAlistair Popple */ 1061b756a3b5SAlistair Popple WARN_ON_ONCE(ret != -ENOENT); 1062b756a3b5SAlistair Popple } 106370e806e4SPeter Xu /* copy_present_pte() will clear `*prealloc' if consumed */ 1064c78f4636SPeter Xu ret = copy_present_pte(dst_vma, src_vma, dst_pte, src_pte, 1065c78f4636SPeter Xu addr, rss, &prealloc); 106670e806e4SPeter Xu /* 106770e806e4SPeter Xu * If we need a pre-allocated page for this pte, drop the 106870e806e4SPeter Xu * locks, allocate, and try again. 106970e806e4SPeter Xu */ 107070e806e4SPeter Xu if (unlikely(ret == -EAGAIN)) 107170e806e4SPeter Xu break; 107270e806e4SPeter Xu if (unlikely(prealloc)) { 107370e806e4SPeter Xu /* 107470e806e4SPeter Xu * pre-alloc page cannot be reused by next time so as 107570e806e4SPeter Xu * to strictly follow mempolicy (e.g., alloc_page_vma() 107670e806e4SPeter Xu * will allocate page according to address). This 107770e806e4SPeter Xu * could only happen if one pinned pte changed. 107870e806e4SPeter Xu */ 1079edf50470SMatthew Wilcox (Oracle) folio_put(prealloc); 108070e806e4SPeter Xu prealloc = NULL; 108170e806e4SPeter Xu } 108279a1971cSLinus Torvalds progress += 8; 10831da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 10841da177e4SLinus Torvalds 10856606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 1086c74df32cSHugh Dickins spin_unlock(src_ptl); 1087ece0e2b6SPeter Zijlstra pte_unmap(orig_src_pte); 1088d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(dst_mm, rss); 1089c36987e2SDaisuke Nishimura pte_unmap_unlock(orig_dst_pte, dst_ptl); 1090c74df32cSHugh Dickins cond_resched(); 1091570a335bSHugh Dickins 10929a5cc85cSAlistair Popple if (ret == -EIO) { 10939a5cc85cSAlistair Popple VM_WARN_ON_ONCE(!entry.val); 109470e806e4SPeter Xu if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) { 109570e806e4SPeter Xu ret = -ENOMEM; 109670e806e4SPeter Xu goto out; 109770e806e4SPeter Xu } 109870e806e4SPeter Xu entry.val = 0; 1099b756a3b5SAlistair Popple } else if (ret == -EBUSY) { 1100b756a3b5SAlistair Popple goto out; 11019a5cc85cSAlistair Popple } else if (ret == -EAGAIN) { 1102c78f4636SPeter Xu prealloc = page_copy_prealloc(src_mm, src_vma, addr); 110370e806e4SPeter Xu if (!prealloc) 1104570a335bSHugh Dickins return -ENOMEM; 11059a5cc85cSAlistair Popple } else if (ret) { 11069a5cc85cSAlistair Popple VM_WARN_ON_ONCE(1); 11079a5cc85cSAlistair Popple } 11089a5cc85cSAlistair Popple 110970e806e4SPeter Xu /* We've captured and resolved the error. Reset, try again. */ 111070e806e4SPeter Xu ret = 0; 11119a5cc85cSAlistair Popple 11121da177e4SLinus Torvalds if (addr != end) 11131da177e4SLinus Torvalds goto again; 111470e806e4SPeter Xu out: 111570e806e4SPeter Xu if (unlikely(prealloc)) 1116edf50470SMatthew Wilcox (Oracle) folio_put(prealloc); 111770e806e4SPeter Xu return ret; 11181da177e4SLinus Torvalds } 11191da177e4SLinus Torvalds 1120c78f4636SPeter Xu static inline int 1121c78f4636SPeter Xu copy_pmd_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1122c78f4636SPeter Xu pud_t *dst_pud, pud_t *src_pud, unsigned long addr, 1123c78f4636SPeter Xu unsigned long end) 11241da177e4SLinus Torvalds { 1125c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1126c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 11271da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 11281da177e4SLinus Torvalds unsigned long next; 11291da177e4SLinus Torvalds 11301da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 11311da177e4SLinus Torvalds if (!dst_pmd) 11321da177e4SLinus Torvalds return -ENOMEM; 11331da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 11341da177e4SLinus Torvalds do { 11351da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 113684c3fc4eSZi Yan if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) 113784c3fc4eSZi Yan || pmd_devmap(*src_pmd)) { 113871e3aac0SAndrea Arcangeli int err; 1139c78f4636SPeter Xu VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, src_vma); 11408f34f1eaSPeter Xu err = copy_huge_pmd(dst_mm, src_mm, dst_pmd, src_pmd, 11418f34f1eaSPeter Xu addr, dst_vma, src_vma); 114271e3aac0SAndrea Arcangeli if (err == -ENOMEM) 114371e3aac0SAndrea Arcangeli return -ENOMEM; 114471e3aac0SAndrea Arcangeli if (!err) 114571e3aac0SAndrea Arcangeli continue; 114671e3aac0SAndrea Arcangeli /* fall through */ 114771e3aac0SAndrea Arcangeli } 11481da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 11491da177e4SLinus Torvalds continue; 1150c78f4636SPeter Xu if (copy_pte_range(dst_vma, src_vma, dst_pmd, src_pmd, 1151c78f4636SPeter Xu addr, next)) 11521da177e4SLinus Torvalds return -ENOMEM; 11531da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 11541da177e4SLinus Torvalds return 0; 11551da177e4SLinus Torvalds } 11561da177e4SLinus Torvalds 1157c78f4636SPeter Xu static inline int 1158c78f4636SPeter Xu copy_pud_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1159c78f4636SPeter Xu p4d_t *dst_p4d, p4d_t *src_p4d, unsigned long addr, 1160c78f4636SPeter Xu unsigned long end) 11611da177e4SLinus Torvalds { 1162c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1163c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 11641da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 11651da177e4SLinus Torvalds unsigned long next; 11661da177e4SLinus Torvalds 1167c2febafcSKirill A. Shutemov dst_pud = pud_alloc(dst_mm, dst_p4d, addr); 11681da177e4SLinus Torvalds if (!dst_pud) 11691da177e4SLinus Torvalds return -ENOMEM; 1170c2febafcSKirill A. Shutemov src_pud = pud_offset(src_p4d, addr); 11711da177e4SLinus Torvalds do { 11721da177e4SLinus Torvalds next = pud_addr_end(addr, end); 1173a00cc7d9SMatthew Wilcox if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { 1174a00cc7d9SMatthew Wilcox int err; 1175a00cc7d9SMatthew Wilcox 1176c78f4636SPeter Xu VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, src_vma); 1177a00cc7d9SMatthew Wilcox err = copy_huge_pud(dst_mm, src_mm, 1178c78f4636SPeter Xu dst_pud, src_pud, addr, src_vma); 1179a00cc7d9SMatthew Wilcox if (err == -ENOMEM) 1180a00cc7d9SMatthew Wilcox return -ENOMEM; 1181a00cc7d9SMatthew Wilcox if (!err) 1182a00cc7d9SMatthew Wilcox continue; 1183a00cc7d9SMatthew Wilcox /* fall through */ 1184a00cc7d9SMatthew Wilcox } 11851da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 11861da177e4SLinus Torvalds continue; 1187c78f4636SPeter Xu if (copy_pmd_range(dst_vma, src_vma, dst_pud, src_pud, 1188c78f4636SPeter Xu addr, next)) 11891da177e4SLinus Torvalds return -ENOMEM; 11901da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 11911da177e4SLinus Torvalds return 0; 11921da177e4SLinus Torvalds } 11931da177e4SLinus Torvalds 1194c78f4636SPeter Xu static inline int 1195c78f4636SPeter Xu copy_p4d_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1196c78f4636SPeter Xu pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long addr, 1197c78f4636SPeter Xu unsigned long end) 1198c2febafcSKirill A. Shutemov { 1199c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1200c2febafcSKirill A. Shutemov p4d_t *src_p4d, *dst_p4d; 1201c2febafcSKirill A. Shutemov unsigned long next; 1202c2febafcSKirill A. Shutemov 1203c2febafcSKirill A. Shutemov dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); 1204c2febafcSKirill A. Shutemov if (!dst_p4d) 1205c2febafcSKirill A. Shutemov return -ENOMEM; 1206c2febafcSKirill A. Shutemov src_p4d = p4d_offset(src_pgd, addr); 1207c2febafcSKirill A. Shutemov do { 1208c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 1209c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(src_p4d)) 1210c2febafcSKirill A. Shutemov continue; 1211c78f4636SPeter Xu if (copy_pud_range(dst_vma, src_vma, dst_p4d, src_p4d, 1212c78f4636SPeter Xu addr, next)) 1213c2febafcSKirill A. Shutemov return -ENOMEM; 1214c2febafcSKirill A. Shutemov } while (dst_p4d++, src_p4d++, addr = next, addr != end); 1215c2febafcSKirill A. Shutemov return 0; 1216c2febafcSKirill A. Shutemov } 1217c2febafcSKirill A. Shutemov 1218c56d1b62SPeter Xu /* 1219c56d1b62SPeter Xu * Return true if the vma needs to copy the pgtable during this fork(). Return 1220c56d1b62SPeter Xu * false when we can speed up fork() by allowing lazy page faults later until 1221c56d1b62SPeter Xu * when the child accesses the memory range. 1222c56d1b62SPeter Xu */ 1223bc70fbf2SPeter Xu static bool 1224c56d1b62SPeter Xu vma_needs_copy(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) 1225c56d1b62SPeter Xu { 1226c56d1b62SPeter Xu /* 1227c56d1b62SPeter Xu * Always copy pgtables when dst_vma has uffd-wp enabled even if it's 1228c56d1b62SPeter Xu * file-backed (e.g. shmem). Because when uffd-wp is enabled, pgtable 1229c56d1b62SPeter Xu * contains uffd-wp protection information, that's something we can't 1230c56d1b62SPeter Xu * retrieve from page cache, and skip copying will lose those info. 1231c56d1b62SPeter Xu */ 1232c56d1b62SPeter Xu if (userfaultfd_wp(dst_vma)) 1233c56d1b62SPeter Xu return true; 1234c56d1b62SPeter Xu 1235bcd51a3cSMike Kravetz if (src_vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 1236c56d1b62SPeter Xu return true; 1237c56d1b62SPeter Xu 1238c56d1b62SPeter Xu if (src_vma->anon_vma) 1239c56d1b62SPeter Xu return true; 1240c56d1b62SPeter Xu 1241c56d1b62SPeter Xu /* 1242c56d1b62SPeter Xu * Don't copy ptes where a page fault will fill them correctly. Fork 1243c56d1b62SPeter Xu * becomes much lighter when there are big shared or private readonly 1244c56d1b62SPeter Xu * mappings. The tradeoff is that copy_page_range is more efficient 1245c56d1b62SPeter Xu * than faulting. 1246c56d1b62SPeter Xu */ 1247c56d1b62SPeter Xu return false; 1248c56d1b62SPeter Xu } 1249c56d1b62SPeter Xu 1250c78f4636SPeter Xu int 1251c78f4636SPeter Xu copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) 12521da177e4SLinus Torvalds { 12531da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 12541da177e4SLinus Torvalds unsigned long next; 1255c78f4636SPeter Xu unsigned long addr = src_vma->vm_start; 1256c78f4636SPeter Xu unsigned long end = src_vma->vm_end; 1257c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1258c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 1259ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 12602ec74c3eSSagi Grimberg bool is_cow; 1261cddb8a5cSAndrea Arcangeli int ret; 12621da177e4SLinus Torvalds 1263c56d1b62SPeter Xu if (!vma_needs_copy(dst_vma, src_vma)) 1264d992895bSNick Piggin return 0; 1265d992895bSNick Piggin 1266c78f4636SPeter Xu if (is_vm_hugetlb_page(src_vma)) 1267bc70fbf2SPeter Xu return copy_hugetlb_page_range(dst_mm, src_mm, dst_vma, src_vma); 12681da177e4SLinus Torvalds 1269c78f4636SPeter Xu if (unlikely(src_vma->vm_flags & VM_PFNMAP)) { 12702ab64037Svenkatesh.pallipadi@intel.com /* 12712ab64037Svenkatesh.pallipadi@intel.com * We do not free on error cases below as remove_vma 12722ab64037Svenkatesh.pallipadi@intel.com * gets called on error from higher level routine 12732ab64037Svenkatesh.pallipadi@intel.com */ 1274c78f4636SPeter Xu ret = track_pfn_copy(src_vma); 12752ab64037Svenkatesh.pallipadi@intel.com if (ret) 12762ab64037Svenkatesh.pallipadi@intel.com return ret; 12772ab64037Svenkatesh.pallipadi@intel.com } 12782ab64037Svenkatesh.pallipadi@intel.com 1279cddb8a5cSAndrea Arcangeli /* 1280cddb8a5cSAndrea Arcangeli * We need to invalidate the secondary MMU mappings only when 1281cddb8a5cSAndrea Arcangeli * there could be a permission downgrade on the ptes of the 1282cddb8a5cSAndrea Arcangeli * parent mm. And a permission downgrade will only happen if 1283cddb8a5cSAndrea Arcangeli * is_cow_mapping() returns true. 1284cddb8a5cSAndrea Arcangeli */ 1285c78f4636SPeter Xu is_cow = is_cow_mapping(src_vma->vm_flags); 1286ac46d4f3SJérôme Glisse 1287ac46d4f3SJérôme Glisse if (is_cow) { 12887269f999SJérôme Glisse mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_PAGE, 12897d4a8be0SAlistair Popple 0, src_mm, addr, end); 1290ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 129157efa1feSJason Gunthorpe /* 129257efa1feSJason Gunthorpe * Disabling preemption is not needed for the write side, as 129357efa1feSJason Gunthorpe * the read side doesn't spin, but goes to the mmap_lock. 129457efa1feSJason Gunthorpe * 129557efa1feSJason Gunthorpe * Use the raw variant of the seqcount_t write API to avoid 129657efa1feSJason Gunthorpe * lockdep complaining about preemptibility. 129757efa1feSJason Gunthorpe */ 129857efa1feSJason Gunthorpe mmap_assert_write_locked(src_mm); 129957efa1feSJason Gunthorpe raw_write_seqcount_begin(&src_mm->write_protect_seq); 1300ac46d4f3SJérôme Glisse } 1301cddb8a5cSAndrea Arcangeli 1302cddb8a5cSAndrea Arcangeli ret = 0; 13031da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 13041da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 13051da177e4SLinus Torvalds do { 13061da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 13071da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 13081da177e4SLinus Torvalds continue; 1309c78f4636SPeter Xu if (unlikely(copy_p4d_range(dst_vma, src_vma, dst_pgd, src_pgd, 1310c78f4636SPeter Xu addr, next))) { 1311d155df53SMa Wupeng untrack_pfn_clear(dst_vma); 1312cddb8a5cSAndrea Arcangeli ret = -ENOMEM; 1313cddb8a5cSAndrea Arcangeli break; 1314cddb8a5cSAndrea Arcangeli } 13151da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 1316cddb8a5cSAndrea Arcangeli 131757efa1feSJason Gunthorpe if (is_cow) { 131857efa1feSJason Gunthorpe raw_write_seqcount_end(&src_mm->write_protect_seq); 1319ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 132057efa1feSJason Gunthorpe } 1321cddb8a5cSAndrea Arcangeli return ret; 13221da177e4SLinus Torvalds } 13231da177e4SLinus Torvalds 13245abfd71dSPeter Xu /* Whether we should zap all COWed (private) pages too */ 13255abfd71dSPeter Xu static inline bool should_zap_cows(struct zap_details *details) 13265abfd71dSPeter Xu { 13275abfd71dSPeter Xu /* By default, zap all pages */ 13285abfd71dSPeter Xu if (!details) 13295abfd71dSPeter Xu return true; 13305abfd71dSPeter Xu 13315abfd71dSPeter Xu /* Or, we zap COWed pages only if the caller wants to */ 13322e148f1eSPeter Xu return details->even_cows; 13335abfd71dSPeter Xu } 13345abfd71dSPeter Xu 13352e148f1eSPeter Xu /* Decides whether we should zap this page with the page pointer specified */ 1336254ab940SPeter Xu static inline bool should_zap_page(struct zap_details *details, struct page *page) 13373506659eSMatthew Wilcox (Oracle) { 13385abfd71dSPeter Xu /* If we can make a decision without *page.. */ 13395abfd71dSPeter Xu if (should_zap_cows(details)) 1340254ab940SPeter Xu return true; 13413506659eSMatthew Wilcox (Oracle) 13425abfd71dSPeter Xu /* E.g. the caller passes NULL for the case of a zero page */ 13435abfd71dSPeter Xu if (!page) 1344254ab940SPeter Xu return true; 13455abfd71dSPeter Xu 13462e148f1eSPeter Xu /* Otherwise we should only zap non-anon pages */ 13472e148f1eSPeter Xu return !PageAnon(page); 13483506659eSMatthew Wilcox (Oracle) } 13493506659eSMatthew Wilcox (Oracle) 1350999dad82SPeter Xu static inline bool zap_drop_file_uffd_wp(struct zap_details *details) 1351999dad82SPeter Xu { 1352999dad82SPeter Xu if (!details) 1353999dad82SPeter Xu return false; 1354999dad82SPeter Xu 1355999dad82SPeter Xu return details->zap_flags & ZAP_FLAG_DROP_MARKER; 1356999dad82SPeter Xu } 1357999dad82SPeter Xu 1358999dad82SPeter Xu /* 1359999dad82SPeter Xu * This function makes sure that we'll replace the none pte with an uffd-wp 1360999dad82SPeter Xu * swap special pte marker when necessary. Must be with the pgtable lock held. 1361999dad82SPeter Xu */ 1362999dad82SPeter Xu static inline void 1363999dad82SPeter Xu zap_install_uffd_wp_if_needed(struct vm_area_struct *vma, 1364999dad82SPeter Xu unsigned long addr, pte_t *pte, 1365999dad82SPeter Xu struct zap_details *details, pte_t pteval) 1366999dad82SPeter Xu { 13672bad466cSPeter Xu /* Zap on anonymous always means dropping everything */ 13682bad466cSPeter Xu if (vma_is_anonymous(vma)) 13692bad466cSPeter Xu return; 13702bad466cSPeter Xu 1371999dad82SPeter Xu if (zap_drop_file_uffd_wp(details)) 1372999dad82SPeter Xu return; 1373999dad82SPeter Xu 1374999dad82SPeter Xu pte_install_uffd_wp_if_needed(vma, addr, pte, pteval); 1375999dad82SPeter Xu } 1376999dad82SPeter Xu 137751c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 1378b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 13791da177e4SLinus Torvalds unsigned long addr, unsigned long end, 138097a89413SPeter Zijlstra struct zap_details *details) 13811da177e4SLinus Torvalds { 1382b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 1383d16dfc55SPeter Zijlstra int force_flush = 0; 1384d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 138597a89413SPeter Zijlstra spinlock_t *ptl; 13865f1a1907SSteven Rostedt pte_t *start_pte; 138797a89413SPeter Zijlstra pte_t *pte; 13888a5f14a2SKirill A. Shutemov swp_entry_t entry; 1389d559db08SKAMEZAWA Hiroyuki 1390ed6a7935SPeter Zijlstra tlb_change_page_size(tlb, PAGE_SIZE); 1391d16dfc55SPeter Zijlstra again: 1392e303297eSPeter Zijlstra init_rss_vec(rss); 13935f1a1907SSteven Rostedt start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 13945f1a1907SSteven Rostedt pte = start_pte; 13953ea27719SMel Gorman flush_tlb_batched_pending(mm); 13966606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 13971da177e4SLinus Torvalds do { 13981da177e4SLinus Torvalds pte_t ptent = *pte; 13998018db85SPeter Xu struct page *page; 14008018db85SPeter Xu 1401166f61b9STobin C Harding if (pte_none(ptent)) 14021da177e4SLinus Torvalds continue; 140351c6f666SRobin Holt 14047b167b68SMinchan Kim if (need_resched()) 14057b167b68SMinchan Kim break; 14067b167b68SMinchan Kim 14076f5e6b9eSHugh Dickins if (pte_present(ptent)) { 14085df397deSLinus Torvalds unsigned int delay_rmap; 14095df397deSLinus Torvalds 141025b2995aSChristoph Hellwig page = vm_normal_page(vma, addr, ptent); 1411254ab940SPeter Xu if (unlikely(!should_zap_page(details, page))) 14121da177e4SLinus Torvalds continue; 1413b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 1414a600388dSZachary Amsden tlb->fullmm); 14151da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 1416999dad82SPeter Xu zap_install_uffd_wp_if_needed(vma, addr, pte, details, 1417999dad82SPeter Xu ptent); 14181da177e4SLinus Torvalds if (unlikely(!page)) 14191da177e4SLinus Torvalds continue; 1420eca56ff9SJerome Marchand 14215df397deSLinus Torvalds delay_rmap = 0; 1422eca56ff9SJerome Marchand if (!PageAnon(page)) { 14231cf35d47SLinus Torvalds if (pte_dirty(ptent)) { 14246237bcd9SHugh Dickins set_page_dirty(page); 14255df397deSLinus Torvalds if (tlb_delay_rmap(tlb)) { 14265df397deSLinus Torvalds delay_rmap = 1; 14275df397deSLinus Torvalds force_flush = 1; 14285df397deSLinus Torvalds } 14291cf35d47SLinus Torvalds } 14308788f678SYu Zhao if (pte_young(ptent) && likely(vma_has_recency(vma))) 1431bf3f3bc5SNick Piggin mark_page_accessed(page); 14326237bcd9SHugh Dickins } 1433eca56ff9SJerome Marchand rss[mm_counter(page)]--; 14345df397deSLinus Torvalds if (!delay_rmap) { 1435cea86fe2SHugh Dickins page_remove_rmap(page, vma, false); 14363dc14741SHugh Dickins if (unlikely(page_mapcount(page) < 0)) 14373dc14741SHugh Dickins print_bad_pte(vma, addr, ptent, page); 14385df397deSLinus Torvalds } 14395df397deSLinus Torvalds if (unlikely(__tlb_remove_page(tlb, page, delay_rmap))) { 14401cf35d47SLinus Torvalds force_flush = 1; 1441ce9ec37bSWill Deacon addr += PAGE_SIZE; 1442d16dfc55SPeter Zijlstra break; 14431cf35d47SLinus Torvalds } 14441da177e4SLinus Torvalds continue; 14451da177e4SLinus Torvalds } 14465042db43SJérôme Glisse 14475042db43SJérôme Glisse entry = pte_to_swp_entry(ptent); 1448b756a3b5SAlistair Popple if (is_device_private_entry(entry) || 1449b756a3b5SAlistair Popple is_device_exclusive_entry(entry)) { 14508018db85SPeter Xu page = pfn_swap_entry_to_page(entry); 1451254ab940SPeter Xu if (unlikely(!should_zap_page(details, page))) 14525042db43SJérôme Glisse continue; 1453999dad82SPeter Xu /* 1454999dad82SPeter Xu * Both device private/exclusive mappings should only 1455999dad82SPeter Xu * work with anonymous page so far, so we don't need to 1456999dad82SPeter Xu * consider uffd-wp bit when zap. For more information, 1457999dad82SPeter Xu * see zap_install_uffd_wp_if_needed(). 1458999dad82SPeter Xu */ 1459999dad82SPeter Xu WARN_ON_ONCE(!vma_is_anonymous(vma)); 14605042db43SJérôme Glisse rss[mm_counter(page)]--; 1461b756a3b5SAlistair Popple if (is_device_private_entry(entry)) 1462cea86fe2SHugh Dickins page_remove_rmap(page, vma, false); 14635042db43SJérôme Glisse put_page(page); 14648018db85SPeter Xu } else if (!non_swap_entry(entry)) { 14655abfd71dSPeter Xu /* Genuine swap entry, hence a private anon page */ 14665abfd71dSPeter Xu if (!should_zap_cows(details)) 14671da177e4SLinus Torvalds continue; 1468b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 14698018db85SPeter Xu if (unlikely(!free_swap_and_cache(entry))) 14708018db85SPeter Xu print_bad_pte(vma, addr, ptent, NULL); 14715abfd71dSPeter Xu } else if (is_migration_entry(entry)) { 1472af5cdaf8SAlistair Popple page = pfn_swap_entry_to_page(entry); 1473254ab940SPeter Xu if (!should_zap_page(details, page)) 14745abfd71dSPeter Xu continue; 1475eca56ff9SJerome Marchand rss[mm_counter(page)]--; 1476999dad82SPeter Xu } else if (pte_marker_entry_uffd_wp(entry)) { 14772bad466cSPeter Xu /* 14782bad466cSPeter Xu * For anon: always drop the marker; for file: only 14792bad466cSPeter Xu * drop the marker if explicitly requested. 14802bad466cSPeter Xu */ 14812bad466cSPeter Xu if (!vma_is_anonymous(vma) && 14822bad466cSPeter Xu !zap_drop_file_uffd_wp(details)) 1483999dad82SPeter Xu continue; 14849f186f9eSMiaohe Lin } else if (is_hwpoison_entry(entry) || 14859f186f9eSMiaohe Lin is_swapin_error_entry(entry)) { 14865abfd71dSPeter Xu if (!should_zap_cows(details)) 14875abfd71dSPeter Xu continue; 14885abfd71dSPeter Xu } else { 14895abfd71dSPeter Xu /* We should have covered all the swap entry types */ 14905abfd71dSPeter Xu WARN_ON_ONCE(1); 14919f9f1acdSKonstantin Khlebnikov } 14929888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 1493999dad82SPeter Xu zap_install_uffd_wp_if_needed(vma, addr, pte, details, ptent); 149497a89413SPeter Zijlstra } while (pte++, addr += PAGE_SIZE, addr != end); 1495ae859762SHugh Dickins 1496d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 14976606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 149851c6f666SRobin Holt 14991cf35d47SLinus Torvalds /* Do the actual TLB flush before dropping ptl */ 15005df397deSLinus Torvalds if (force_flush) { 15011cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 15025df397deSLinus Torvalds tlb_flush_rmaps(tlb, vma); 15035df397deSLinus Torvalds } 15041cf35d47SLinus Torvalds pte_unmap_unlock(start_pte, ptl); 15051cf35d47SLinus Torvalds 15061cf35d47SLinus Torvalds /* 15071cf35d47SLinus Torvalds * If we forced a TLB flush (either due to running out of 15081cf35d47SLinus Torvalds * batch buffers or because we needed to flush dirty TLB 15091cf35d47SLinus Torvalds * entries before releasing the ptl), free the batched 15101cf35d47SLinus Torvalds * memory too. Restart if we didn't do everything. 15111cf35d47SLinus Torvalds */ 15121cf35d47SLinus Torvalds if (force_flush) { 15131cf35d47SLinus Torvalds force_flush = 0; 1514fa0aafb8SPeter Zijlstra tlb_flush_mmu(tlb); 15157b167b68SMinchan Kim } 15167b167b68SMinchan Kim 15177b167b68SMinchan Kim if (addr != end) { 15187b167b68SMinchan Kim cond_resched(); 1519d16dfc55SPeter Zijlstra goto again; 1520d16dfc55SPeter Zijlstra } 1521d16dfc55SPeter Zijlstra 152251c6f666SRobin Holt return addr; 15231da177e4SLinus Torvalds } 15241da177e4SLinus Torvalds 152551c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1526b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 15271da177e4SLinus Torvalds unsigned long addr, unsigned long end, 152897a89413SPeter Zijlstra struct zap_details *details) 15291da177e4SLinus Torvalds { 15301da177e4SLinus Torvalds pmd_t *pmd; 15311da177e4SLinus Torvalds unsigned long next; 15321da177e4SLinus Torvalds 15331da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 15341da177e4SLinus Torvalds do { 15351da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 153684c3fc4eSZi Yan if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { 153753406ed1SHugh Dickins if (next - addr != HPAGE_PMD_SIZE) 1538fd60775aSDavid Rientjes __split_huge_pmd(vma, pmd, addr, false, NULL); 153953406ed1SHugh Dickins else if (zap_huge_pmd(tlb, vma, pmd, addr)) 15401a5a9906SAndrea Arcangeli goto next; 154171e3aac0SAndrea Arcangeli /* fall through */ 15423506659eSMatthew Wilcox (Oracle) } else if (details && details->single_folio && 15433506659eSMatthew Wilcox (Oracle) folio_test_pmd_mappable(details->single_folio) && 154422061a1fSHugh Dickins next - addr == HPAGE_PMD_SIZE && pmd_none(*pmd)) { 154522061a1fSHugh Dickins spinlock_t *ptl = pmd_lock(tlb->mm, pmd); 154622061a1fSHugh Dickins /* 154722061a1fSHugh Dickins * Take and drop THP pmd lock so that we cannot return 154822061a1fSHugh Dickins * prematurely, while zap_huge_pmd() has cleared *pmd, 154922061a1fSHugh Dickins * but not yet decremented compound_mapcount(). 155022061a1fSHugh Dickins */ 155122061a1fSHugh Dickins spin_unlock(ptl); 155271e3aac0SAndrea Arcangeli } 155322061a1fSHugh Dickins 15541a5a9906SAndrea Arcangeli /* 15551a5a9906SAndrea Arcangeli * Here there can be other concurrent MADV_DONTNEED or 15561a5a9906SAndrea Arcangeli * trans huge page faults running, and if the pmd is 15571a5a9906SAndrea Arcangeli * none or trans huge it can change under us. This is 1558c1e8d7c6SMichel Lespinasse * because MADV_DONTNEED holds the mmap_lock in read 15591a5a9906SAndrea Arcangeli * mode. 15601a5a9906SAndrea Arcangeli */ 15611a5a9906SAndrea Arcangeli if (pmd_none_or_trans_huge_or_clear_bad(pmd)) 15621a5a9906SAndrea Arcangeli goto next; 156397a89413SPeter Zijlstra next = zap_pte_range(tlb, vma, pmd, addr, next, details); 15641a5a9906SAndrea Arcangeli next: 156597a89413SPeter Zijlstra cond_resched(); 156697a89413SPeter Zijlstra } while (pmd++, addr = next, addr != end); 156751c6f666SRobin Holt 156851c6f666SRobin Holt return addr; 15691da177e4SLinus Torvalds } 15701da177e4SLinus Torvalds 157151c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1572c2febafcSKirill A. Shutemov struct vm_area_struct *vma, p4d_t *p4d, 15731da177e4SLinus Torvalds unsigned long addr, unsigned long end, 157497a89413SPeter Zijlstra struct zap_details *details) 15751da177e4SLinus Torvalds { 15761da177e4SLinus Torvalds pud_t *pud; 15771da177e4SLinus Torvalds unsigned long next; 15781da177e4SLinus Torvalds 1579c2febafcSKirill A. Shutemov pud = pud_offset(p4d, addr); 15801da177e4SLinus Torvalds do { 15811da177e4SLinus Torvalds next = pud_addr_end(addr, end); 1582a00cc7d9SMatthew Wilcox if (pud_trans_huge(*pud) || pud_devmap(*pud)) { 1583a00cc7d9SMatthew Wilcox if (next - addr != HPAGE_PUD_SIZE) { 158442fc5414SMichel Lespinasse mmap_assert_locked(tlb->mm); 1585a00cc7d9SMatthew Wilcox split_huge_pud(vma, pud, addr); 1586a00cc7d9SMatthew Wilcox } else if (zap_huge_pud(tlb, vma, pud, addr)) 1587a00cc7d9SMatthew Wilcox goto next; 1588a00cc7d9SMatthew Wilcox /* fall through */ 1589a00cc7d9SMatthew Wilcox } 159097a89413SPeter Zijlstra if (pud_none_or_clear_bad(pud)) 15911da177e4SLinus Torvalds continue; 159297a89413SPeter Zijlstra next = zap_pmd_range(tlb, vma, pud, addr, next, details); 1593a00cc7d9SMatthew Wilcox next: 1594a00cc7d9SMatthew Wilcox cond_resched(); 159597a89413SPeter Zijlstra } while (pud++, addr = next, addr != end); 159651c6f666SRobin Holt 159751c6f666SRobin Holt return addr; 15981da177e4SLinus Torvalds } 15991da177e4SLinus Torvalds 1600c2febafcSKirill A. Shutemov static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, 1601c2febafcSKirill A. Shutemov struct vm_area_struct *vma, pgd_t *pgd, 1602c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 1603c2febafcSKirill A. Shutemov struct zap_details *details) 1604c2febafcSKirill A. Shutemov { 1605c2febafcSKirill A. Shutemov p4d_t *p4d; 1606c2febafcSKirill A. Shutemov unsigned long next; 1607c2febafcSKirill A. Shutemov 1608c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, addr); 1609c2febafcSKirill A. Shutemov do { 1610c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 1611c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(p4d)) 1612c2febafcSKirill A. Shutemov continue; 1613c2febafcSKirill A. Shutemov next = zap_pud_range(tlb, vma, p4d, addr, next, details); 1614c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 1615c2febafcSKirill A. Shutemov 1616c2febafcSKirill A. Shutemov return addr; 1617c2febafcSKirill A. Shutemov } 1618c2febafcSKirill A. Shutemov 1619aac45363SMichal Hocko void unmap_page_range(struct mmu_gather *tlb, 162051c6f666SRobin Holt struct vm_area_struct *vma, 16211da177e4SLinus Torvalds unsigned long addr, unsigned long end, 162297a89413SPeter Zijlstra struct zap_details *details) 16231da177e4SLinus Torvalds { 16241da177e4SLinus Torvalds pgd_t *pgd; 16251da177e4SLinus Torvalds unsigned long next; 16261da177e4SLinus Torvalds 16271da177e4SLinus Torvalds BUG_ON(addr >= end); 16281da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 16291da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 16301da177e4SLinus Torvalds do { 16311da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 163297a89413SPeter Zijlstra if (pgd_none_or_clear_bad(pgd)) 16331da177e4SLinus Torvalds continue; 1634c2febafcSKirill A. Shutemov next = zap_p4d_range(tlb, vma, pgd, addr, next, details); 163597a89413SPeter Zijlstra } while (pgd++, addr = next, addr != end); 16361da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 16371da177e4SLinus Torvalds } 16381da177e4SLinus Torvalds 1639f5cc4eefSAl Viro 1640f5cc4eefSAl Viro static void unmap_single_vma(struct mmu_gather *tlb, 16411da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 16424f74d2c8SLinus Torvalds unsigned long end_addr, 164368f48381SSuren Baghdasaryan struct zap_details *details, bool mm_wr_locked) 16441da177e4SLinus Torvalds { 1645f5cc4eefSAl Viro unsigned long start = max(vma->vm_start, start_addr); 16461da177e4SLinus Torvalds unsigned long end; 16471da177e4SLinus Torvalds 16481da177e4SLinus Torvalds if (start >= vma->vm_end) 1649f5cc4eefSAl Viro return; 16501da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 16511da177e4SLinus Torvalds if (end <= vma->vm_start) 1652f5cc4eefSAl Viro return; 16531da177e4SLinus Torvalds 1654cbc91f71SSrikar Dronamraju if (vma->vm_file) 1655cbc91f71SSrikar Dronamraju uprobe_munmap(vma, start, end); 1656cbc91f71SSrikar Dronamraju 1657b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) 165868f48381SSuren Baghdasaryan untrack_pfn(vma, 0, 0, mm_wr_locked); 16592ab64037Svenkatesh.pallipadi@intel.com 16608b2a1238SAl Viro if (start != end) { 166151c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1662a137e1ccSAndi Kleen /* 1663a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1664a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1665a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 16667aa6b4adSDavidlohr Bueso * cleanup path of mmap_region. When 1667a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 16687aa6b4adSDavidlohr Bueso * mmap_region() nullifies vma->vm_file 1669a137e1ccSAndi Kleen * before calling this function to clean up. 1670a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1671a137e1ccSAndi Kleen * safe to do nothing in this case. 1672a137e1ccSAndi Kleen */ 167324669e58SAneesh Kumar K.V if (vma->vm_file) { 167405e90bd0SPeter Xu zap_flags_t zap_flags = details ? 167505e90bd0SPeter Xu details->zap_flags : 0; 167605e90bd0SPeter Xu __unmap_hugepage_range_final(tlb, vma, start, end, 167705e90bd0SPeter Xu NULL, zap_flags); 167824669e58SAneesh Kumar K.V } 167951c6f666SRobin Holt } else 1680038c7aa1SAl Viro unmap_page_range(tlb, vma, start, end, details); 168197a89413SPeter Zijlstra } 168251c6f666SRobin Holt } 16831da177e4SLinus Torvalds 1684f5cc4eefSAl Viro /** 1685f5cc4eefSAl Viro * unmap_vmas - unmap a range of memory covered by a list of vma's 1686f5cc4eefSAl Viro * @tlb: address of the caller's struct mmu_gather 1687763ecb03SLiam R. Howlett * @mt: the maple tree 1688f5cc4eefSAl Viro * @vma: the starting vma 1689f5cc4eefSAl Viro * @start_addr: virtual address at which to start unmapping 1690f5cc4eefSAl Viro * @end_addr: virtual address at which to end unmapping 1691f5cc4eefSAl Viro * 1692f5cc4eefSAl Viro * Unmap all pages in the vma list. 1693f5cc4eefSAl Viro * 1694f5cc4eefSAl Viro * Only addresses between `start' and `end' will be unmapped. 1695f5cc4eefSAl Viro * 1696f5cc4eefSAl Viro * The VMA list must be sorted in ascending virtual address order. 1697f5cc4eefSAl Viro * 1698f5cc4eefSAl Viro * unmap_vmas() assumes that the caller will flush the whole unmapped address 1699f5cc4eefSAl Viro * range after unmap_vmas() returns. So the only responsibility here is to 1700f5cc4eefSAl Viro * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 1701f5cc4eefSAl Viro * drops the lock and schedules. 1702f5cc4eefSAl Viro */ 1703763ecb03SLiam R. Howlett void unmap_vmas(struct mmu_gather *tlb, struct maple_tree *mt, 1704f5cc4eefSAl Viro struct vm_area_struct *vma, unsigned long start_addr, 170568f48381SSuren Baghdasaryan unsigned long end_addr, bool mm_wr_locked) 1706f5cc4eefSAl Viro { 1707ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1708999dad82SPeter Xu struct zap_details details = { 170904ada095SMike Kravetz .zap_flags = ZAP_FLAG_DROP_MARKER | ZAP_FLAG_UNMAP, 1710999dad82SPeter Xu /* Careful - we need to zap private pages too! */ 1711999dad82SPeter Xu .even_cows = true, 1712999dad82SPeter Xu }; 1713763ecb03SLiam R. Howlett MA_STATE(mas, mt, vma->vm_end, vma->vm_end); 1714f5cc4eefSAl Viro 17157d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma->vm_mm, 17166f4f13e8SJérôme Glisse start_addr, end_addr); 1717ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 1718763ecb03SLiam R. Howlett do { 171968f48381SSuren Baghdasaryan unmap_single_vma(tlb, vma, start_addr, end_addr, &details, 172068f48381SSuren Baghdasaryan mm_wr_locked); 1721763ecb03SLiam R. Howlett } while ((vma = mas_find(&mas, end_addr - 1)) != NULL); 1722ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 17231da177e4SLinus Torvalds } 17241da177e4SLinus Torvalds 17251da177e4SLinus Torvalds /** 1726f5cc4eefSAl Viro * zap_page_range_single - remove user pages in a given range 1727f5cc4eefSAl Viro * @vma: vm_area_struct holding the applicable pages 1728f5cc4eefSAl Viro * @address: starting address of pages to zap 1729f5cc4eefSAl Viro * @size: number of bytes to zap 17308a5f14a2SKirill A. Shutemov * @details: details of shared cache invalidation 1731f5cc4eefSAl Viro * 1732f5cc4eefSAl Viro * The range must fit into one VMA. 1733f5cc4eefSAl Viro */ 173421b85b09SMike Kravetz void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 1735f5cc4eefSAl Viro unsigned long size, struct zap_details *details) 1736f5cc4eefSAl Viro { 173721b85b09SMike Kravetz const unsigned long end = address + size; 1738ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1739f5cc4eefSAl Viro struct mmu_gather tlb; 1740f5cc4eefSAl Viro 1741f5cc4eefSAl Viro lru_add_drain(); 17427d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma->vm_mm, 174321b85b09SMike Kravetz address, end); 174421b85b09SMike Kravetz if (is_vm_hugetlb_page(vma)) 174521b85b09SMike Kravetz adjust_range_if_pmd_sharing_possible(vma, &range.start, 174621b85b09SMike Kravetz &range.end); 1747a72afd87SWill Deacon tlb_gather_mmu(&tlb, vma->vm_mm); 1748ac46d4f3SJérôme Glisse update_hiwater_rss(vma->vm_mm); 1749ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 175021b85b09SMike Kravetz /* 175121b85b09SMike Kravetz * unmap 'address-end' not 'range.start-range.end' as range 175221b85b09SMike Kravetz * could have been expanded for hugetlb pmd sharing. 175321b85b09SMike Kravetz */ 175468f48381SSuren Baghdasaryan unmap_single_vma(&tlb, vma, address, end, details, false); 1755ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 1756ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 17571da177e4SLinus Torvalds } 17581da177e4SLinus Torvalds 1759c627f9ccSJack Steiner /** 1760c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1761c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1762c627f9ccSJack Steiner * @address: starting address of pages to zap 1763c627f9ccSJack Steiner * @size: number of bytes to zap 1764c627f9ccSJack Steiner * 1765c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1766c627f9ccSJack Steiner * 1767c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1768c627f9ccSJack Steiner * 1769c627f9ccSJack Steiner */ 177027d036e3SLeon Romanovsky void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1771c627f9ccSJack Steiner unsigned long size) 1772c627f9ccSJack Steiner { 177388a35912SMiaohe Lin if (!range_in_vma(vma, address, address + size) || 1774c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 177527d036e3SLeon Romanovsky return; 177627d036e3SLeon Romanovsky 1777f5cc4eefSAl Viro zap_page_range_single(vma, address, size, NULL); 1778c627f9ccSJack Steiner } 1779c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1780c627f9ccSJack Steiner 17818cd3984dSArjun Roy static pmd_t *walk_to_pmd(struct mm_struct *mm, unsigned long addr) 1782c9cfcddfSLinus Torvalds { 1783c2febafcSKirill A. Shutemov pgd_t *pgd; 1784c2febafcSKirill A. Shutemov p4d_t *p4d; 1785c2febafcSKirill A. Shutemov pud_t *pud; 1786c2febafcSKirill A. Shutemov pmd_t *pmd; 1787c2febafcSKirill A. Shutemov 1788c2febafcSKirill A. Shutemov pgd = pgd_offset(mm, addr); 1789c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 1790c2febafcSKirill A. Shutemov if (!p4d) 1791c2febafcSKirill A. Shutemov return NULL; 1792c2febafcSKirill A. Shutemov pud = pud_alloc(mm, p4d, addr); 1793c2febafcSKirill A. Shutemov if (!pud) 1794c2febafcSKirill A. Shutemov return NULL; 1795c2febafcSKirill A. Shutemov pmd = pmd_alloc(mm, pud, addr); 1796c2febafcSKirill A. Shutemov if (!pmd) 1797c2febafcSKirill A. Shutemov return NULL; 1798c2febafcSKirill A. Shutemov 1799f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 18008cd3984dSArjun Roy return pmd; 18018cd3984dSArjun Roy } 18028cd3984dSArjun Roy 18038cd3984dSArjun Roy pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 18048cd3984dSArjun Roy spinlock_t **ptl) 18058cd3984dSArjun Roy { 18068cd3984dSArjun Roy pmd_t *pmd = walk_to_pmd(mm, addr); 18078cd3984dSArjun Roy 18088cd3984dSArjun Roy if (!pmd) 18098cd3984dSArjun Roy return NULL; 1810c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 1811c9cfcddfSLinus Torvalds } 1812c9cfcddfSLinus Torvalds 18138efd6f5bSArjun Roy static int validate_page_before_insert(struct page *page) 18148efd6f5bSArjun Roy { 18158efd6f5bSArjun Roy if (PageAnon(page) || PageSlab(page) || page_has_type(page)) 18168efd6f5bSArjun Roy return -EINVAL; 18178efd6f5bSArjun Roy flush_dcache_page(page); 18188efd6f5bSArjun Roy return 0; 18198efd6f5bSArjun Roy } 18208efd6f5bSArjun Roy 1821cea86fe2SHugh Dickins static int insert_page_into_pte_locked(struct vm_area_struct *vma, pte_t *pte, 18228efd6f5bSArjun Roy unsigned long addr, struct page *page, pgprot_t prot) 18238efd6f5bSArjun Roy { 18248efd6f5bSArjun Roy if (!pte_none(*pte)) 18258efd6f5bSArjun Roy return -EBUSY; 18268efd6f5bSArjun Roy /* Ok, finally just insert the thing.. */ 18278efd6f5bSArjun Roy get_page(page); 1828f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, mm_counter_file(page)); 1829cea86fe2SHugh Dickins page_add_file_rmap(page, vma, false); 1830cea86fe2SHugh Dickins set_pte_at(vma->vm_mm, addr, pte, mk_pte(page, prot)); 18318efd6f5bSArjun Roy return 0; 18328efd6f5bSArjun Roy } 18338efd6f5bSArjun Roy 18341da177e4SLinus Torvalds /* 1835238f58d8SLinus Torvalds * This is the old fallback for page remapping. 1836238f58d8SLinus Torvalds * 1837238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 1838238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 1839238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 1840238f58d8SLinus Torvalds */ 1841423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 1842423bad60SNick Piggin struct page *page, pgprot_t prot) 1843238f58d8SLinus Torvalds { 1844238f58d8SLinus Torvalds int retval; 1845238f58d8SLinus Torvalds pte_t *pte; 1846238f58d8SLinus Torvalds spinlock_t *ptl; 1847238f58d8SLinus Torvalds 18488efd6f5bSArjun Roy retval = validate_page_before_insert(page); 18498efd6f5bSArjun Roy if (retval) 18505b4e655eSKAMEZAWA Hiroyuki goto out; 1851238f58d8SLinus Torvalds retval = -ENOMEM; 1852cea86fe2SHugh Dickins pte = get_locked_pte(vma->vm_mm, addr, &ptl); 1853238f58d8SLinus Torvalds if (!pte) 18545b4e655eSKAMEZAWA Hiroyuki goto out; 1855cea86fe2SHugh Dickins retval = insert_page_into_pte_locked(vma, pte, addr, page, prot); 1856238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 1857238f58d8SLinus Torvalds out: 1858238f58d8SLinus Torvalds return retval; 1859238f58d8SLinus Torvalds } 1860238f58d8SLinus Torvalds 18618cd3984dSArjun Roy #ifdef pte_index 1862cea86fe2SHugh Dickins static int insert_page_in_batch_locked(struct vm_area_struct *vma, pte_t *pte, 18638cd3984dSArjun Roy unsigned long addr, struct page *page, pgprot_t prot) 18648cd3984dSArjun Roy { 18658cd3984dSArjun Roy int err; 18668cd3984dSArjun Roy 18678cd3984dSArjun Roy if (!page_count(page)) 18688cd3984dSArjun Roy return -EINVAL; 18698cd3984dSArjun Roy err = validate_page_before_insert(page); 18707f70c2a6SArjun Roy if (err) 18717f70c2a6SArjun Roy return err; 1872cea86fe2SHugh Dickins return insert_page_into_pte_locked(vma, pte, addr, page, prot); 18738cd3984dSArjun Roy } 18748cd3984dSArjun Roy 18758cd3984dSArjun Roy /* insert_pages() amortizes the cost of spinlock operations 18768cd3984dSArjun Roy * when inserting pages in a loop. Arch *must* define pte_index. 18778cd3984dSArjun Roy */ 18788cd3984dSArjun Roy static int insert_pages(struct vm_area_struct *vma, unsigned long addr, 18798cd3984dSArjun Roy struct page **pages, unsigned long *num, pgprot_t prot) 18808cd3984dSArjun Roy { 18818cd3984dSArjun Roy pmd_t *pmd = NULL; 18827f70c2a6SArjun Roy pte_t *start_pte, *pte; 18837f70c2a6SArjun Roy spinlock_t *pte_lock; 18848cd3984dSArjun Roy struct mm_struct *const mm = vma->vm_mm; 18858cd3984dSArjun Roy unsigned long curr_page_idx = 0; 18868cd3984dSArjun Roy unsigned long remaining_pages_total = *num; 18878cd3984dSArjun Roy unsigned long pages_to_write_in_pmd; 18888cd3984dSArjun Roy int ret; 18898cd3984dSArjun Roy more: 18908cd3984dSArjun Roy ret = -EFAULT; 18918cd3984dSArjun Roy pmd = walk_to_pmd(mm, addr); 18928cd3984dSArjun Roy if (!pmd) 18938cd3984dSArjun Roy goto out; 18948cd3984dSArjun Roy 18958cd3984dSArjun Roy pages_to_write_in_pmd = min_t(unsigned long, 18968cd3984dSArjun Roy remaining_pages_total, PTRS_PER_PTE - pte_index(addr)); 18978cd3984dSArjun Roy 18988cd3984dSArjun Roy /* Allocate the PTE if necessary; takes PMD lock once only. */ 18998cd3984dSArjun Roy ret = -ENOMEM; 19008cd3984dSArjun Roy if (pte_alloc(mm, pmd)) 19018cd3984dSArjun Roy goto out; 19028cd3984dSArjun Roy 19038cd3984dSArjun Roy while (pages_to_write_in_pmd) { 19048cd3984dSArjun Roy int pte_idx = 0; 19058cd3984dSArjun Roy const int batch_size = min_t(int, pages_to_write_in_pmd, 8); 19068cd3984dSArjun Roy 19077f70c2a6SArjun Roy start_pte = pte_offset_map_lock(mm, pmd, addr, &pte_lock); 19087f70c2a6SArjun Roy for (pte = start_pte; pte_idx < batch_size; ++pte, ++pte_idx) { 1909cea86fe2SHugh Dickins int err = insert_page_in_batch_locked(vma, pte, 19108cd3984dSArjun Roy addr, pages[curr_page_idx], prot); 19118cd3984dSArjun Roy if (unlikely(err)) { 19127f70c2a6SArjun Roy pte_unmap_unlock(start_pte, pte_lock); 19138cd3984dSArjun Roy ret = err; 19148cd3984dSArjun Roy remaining_pages_total -= pte_idx; 19158cd3984dSArjun Roy goto out; 19168cd3984dSArjun Roy } 19178cd3984dSArjun Roy addr += PAGE_SIZE; 19188cd3984dSArjun Roy ++curr_page_idx; 19198cd3984dSArjun Roy } 19207f70c2a6SArjun Roy pte_unmap_unlock(start_pte, pte_lock); 19218cd3984dSArjun Roy pages_to_write_in_pmd -= batch_size; 19228cd3984dSArjun Roy remaining_pages_total -= batch_size; 19238cd3984dSArjun Roy } 19248cd3984dSArjun Roy if (remaining_pages_total) 19258cd3984dSArjun Roy goto more; 19268cd3984dSArjun Roy ret = 0; 19278cd3984dSArjun Roy out: 19288cd3984dSArjun Roy *num = remaining_pages_total; 19298cd3984dSArjun Roy return ret; 19308cd3984dSArjun Roy } 19318cd3984dSArjun Roy #endif /* ifdef pte_index */ 19328cd3984dSArjun Roy 19338cd3984dSArjun Roy /** 19348cd3984dSArjun Roy * vm_insert_pages - insert multiple pages into user vma, batching the pmd lock. 19358cd3984dSArjun Roy * @vma: user vma to map to 19368cd3984dSArjun Roy * @addr: target start user address of these pages 19378cd3984dSArjun Roy * @pages: source kernel pages 19388cd3984dSArjun Roy * @num: in: number of pages to map. out: number of pages that were *not* 19398cd3984dSArjun Roy * mapped. (0 means all pages were successfully mapped). 19408cd3984dSArjun Roy * 19418cd3984dSArjun Roy * Preferred over vm_insert_page() when inserting multiple pages. 19428cd3984dSArjun Roy * 19438cd3984dSArjun Roy * In case of error, we may have mapped a subset of the provided 19448cd3984dSArjun Roy * pages. It is the caller's responsibility to account for this case. 19458cd3984dSArjun Roy * 19468cd3984dSArjun Roy * The same restrictions apply as in vm_insert_page(). 19478cd3984dSArjun Roy */ 19488cd3984dSArjun Roy int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr, 19498cd3984dSArjun Roy struct page **pages, unsigned long *num) 19508cd3984dSArjun Roy { 19518cd3984dSArjun Roy #ifdef pte_index 19528cd3984dSArjun Roy const unsigned long end_addr = addr + (*num * PAGE_SIZE) - 1; 19538cd3984dSArjun Roy 19548cd3984dSArjun Roy if (addr < vma->vm_start || end_addr >= vma->vm_end) 19558cd3984dSArjun Roy return -EFAULT; 19568cd3984dSArjun Roy if (!(vma->vm_flags & VM_MIXEDMAP)) { 1957d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(vma->vm_mm)); 19588cd3984dSArjun Roy BUG_ON(vma->vm_flags & VM_PFNMAP); 19591c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_MIXEDMAP); 19608cd3984dSArjun Roy } 19618cd3984dSArjun Roy /* Defer page refcount checking till we're about to map that page. */ 19628cd3984dSArjun Roy return insert_pages(vma, addr, pages, num, vma->vm_page_prot); 19638cd3984dSArjun Roy #else 19648cd3984dSArjun Roy unsigned long idx = 0, pgcount = *num; 196545779b03STom Rix int err = -EINVAL; 19668cd3984dSArjun Roy 19678cd3984dSArjun Roy for (; idx < pgcount; ++idx) { 19688cd3984dSArjun Roy err = vm_insert_page(vma, addr + (PAGE_SIZE * idx), pages[idx]); 19698cd3984dSArjun Roy if (err) 19708cd3984dSArjun Roy break; 19718cd3984dSArjun Roy } 19728cd3984dSArjun Roy *num = pgcount - idx; 19738cd3984dSArjun Roy return err; 19748cd3984dSArjun Roy #endif /* ifdef pte_index */ 19758cd3984dSArjun Roy } 19768cd3984dSArjun Roy EXPORT_SYMBOL(vm_insert_pages); 19778cd3984dSArjun Roy 1978bfa5bf6dSRolf Eike Beer /** 1979bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 1980bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1981bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 1982bfa5bf6dSRolf Eike Beer * @page: source kernel page 1983bfa5bf6dSRolf Eike Beer * 1984a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 1985a145dd41SLinus Torvalds * into a user vma. 1986a145dd41SLinus Torvalds * 1987a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 1988a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 1989a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 19908dfcc9baSNick Piggin * (see split_page()). 1991a145dd41SLinus Torvalds * 1992a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 1993a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 1994a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 1995a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 1996a145dd41SLinus Torvalds * ask for a shared writable mapping! 1997a145dd41SLinus Torvalds * 1998a145dd41SLinus Torvalds * The page does not need to be reserved. 19994b6e1e37SKonstantin Khlebnikov * 20004b6e1e37SKonstantin Khlebnikov * Usually this function is called from f_op->mmap() handler 2001c1e8d7c6SMichel Lespinasse * under mm->mmap_lock write-lock, so it can change vma->vm_flags. 20024b6e1e37SKonstantin Khlebnikov * Caller must set VM_MIXEDMAP on vma if it wants to call this 20034b6e1e37SKonstantin Khlebnikov * function from other places, for example from page-fault handler. 2004a862f68aSMike Rapoport * 2005a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 2006a145dd41SLinus Torvalds */ 2007423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 2008423bad60SNick Piggin struct page *page) 2009a145dd41SLinus Torvalds { 2010a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 2011a145dd41SLinus Torvalds return -EFAULT; 2012a145dd41SLinus Torvalds if (!page_count(page)) 2013a145dd41SLinus Torvalds return -EINVAL; 20144b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & VM_MIXEDMAP)) { 2015d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(vma->vm_mm)); 20164b6e1e37SKonstantin Khlebnikov BUG_ON(vma->vm_flags & VM_PFNMAP); 20171c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_MIXEDMAP); 20184b6e1e37SKonstantin Khlebnikov } 2019423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 2020a145dd41SLinus Torvalds } 2021e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 2022a145dd41SLinus Torvalds 2023a667d745SSouptick Joarder /* 2024a667d745SSouptick Joarder * __vm_map_pages - maps range of kernel pages into user vma 2025a667d745SSouptick Joarder * @vma: user vma to map to 2026a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2027a667d745SSouptick Joarder * @num: number of pages in page array 2028a667d745SSouptick Joarder * @offset: user's requested vm_pgoff 2029a667d745SSouptick Joarder * 2030a667d745SSouptick Joarder * This allows drivers to map range of kernel pages into a user vma. 2031a667d745SSouptick Joarder * 2032a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2033a667d745SSouptick Joarder */ 2034a667d745SSouptick Joarder static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages, 2035a667d745SSouptick Joarder unsigned long num, unsigned long offset) 2036a667d745SSouptick Joarder { 2037a667d745SSouptick Joarder unsigned long count = vma_pages(vma); 2038a667d745SSouptick Joarder unsigned long uaddr = vma->vm_start; 2039a667d745SSouptick Joarder int ret, i; 2040a667d745SSouptick Joarder 2041a667d745SSouptick Joarder /* Fail if the user requested offset is beyond the end of the object */ 204296756fcbSMiguel Ojeda if (offset >= num) 2043a667d745SSouptick Joarder return -ENXIO; 2044a667d745SSouptick Joarder 2045a667d745SSouptick Joarder /* Fail if the user requested size exceeds available object size */ 2046a667d745SSouptick Joarder if (count > num - offset) 2047a667d745SSouptick Joarder return -ENXIO; 2048a667d745SSouptick Joarder 2049a667d745SSouptick Joarder for (i = 0; i < count; i++) { 2050a667d745SSouptick Joarder ret = vm_insert_page(vma, uaddr, pages[offset + i]); 2051a667d745SSouptick Joarder if (ret < 0) 2052a667d745SSouptick Joarder return ret; 2053a667d745SSouptick Joarder uaddr += PAGE_SIZE; 2054a667d745SSouptick Joarder } 2055a667d745SSouptick Joarder 2056a667d745SSouptick Joarder return 0; 2057a667d745SSouptick Joarder } 2058a667d745SSouptick Joarder 2059a667d745SSouptick Joarder /** 2060a667d745SSouptick Joarder * vm_map_pages - maps range of kernel pages starts with non zero offset 2061a667d745SSouptick Joarder * @vma: user vma to map to 2062a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2063a667d745SSouptick Joarder * @num: number of pages in page array 2064a667d745SSouptick Joarder * 2065a667d745SSouptick Joarder * Maps an object consisting of @num pages, catering for the user's 2066a667d745SSouptick Joarder * requested vm_pgoff 2067a667d745SSouptick Joarder * 2068a667d745SSouptick Joarder * If we fail to insert any page into the vma, the function will return 2069a667d745SSouptick Joarder * immediately leaving any previously inserted pages present. Callers 2070a667d745SSouptick Joarder * from the mmap handler may immediately return the error as their caller 2071a667d745SSouptick Joarder * will destroy the vma, removing any successfully inserted pages. Other 2072a667d745SSouptick Joarder * callers should make their own arrangements for calling unmap_region(). 2073a667d745SSouptick Joarder * 2074a667d745SSouptick Joarder * Context: Process context. Called by mmap handlers. 2075a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2076a667d745SSouptick Joarder */ 2077a667d745SSouptick Joarder int vm_map_pages(struct vm_area_struct *vma, struct page **pages, 2078a667d745SSouptick Joarder unsigned long num) 2079a667d745SSouptick Joarder { 2080a667d745SSouptick Joarder return __vm_map_pages(vma, pages, num, vma->vm_pgoff); 2081a667d745SSouptick Joarder } 2082a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages); 2083a667d745SSouptick Joarder 2084a667d745SSouptick Joarder /** 2085a667d745SSouptick Joarder * vm_map_pages_zero - map range of kernel pages starts with zero offset 2086a667d745SSouptick Joarder * @vma: user vma to map to 2087a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2088a667d745SSouptick Joarder * @num: number of pages in page array 2089a667d745SSouptick Joarder * 2090a667d745SSouptick Joarder * Similar to vm_map_pages(), except that it explicitly sets the offset 2091a667d745SSouptick Joarder * to 0. This function is intended for the drivers that did not consider 2092a667d745SSouptick Joarder * vm_pgoff. 2093a667d745SSouptick Joarder * 2094a667d745SSouptick Joarder * Context: Process context. Called by mmap handlers. 2095a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2096a667d745SSouptick Joarder */ 2097a667d745SSouptick Joarder int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, 2098a667d745SSouptick Joarder unsigned long num) 2099a667d745SSouptick Joarder { 2100a667d745SSouptick Joarder return __vm_map_pages(vma, pages, num, 0); 2101a667d745SSouptick Joarder } 2102a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages_zero); 2103a667d745SSouptick Joarder 21049b5a8e00SMatthew Wilcox static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2105b2770da6SRoss Zwisler pfn_t pfn, pgprot_t prot, bool mkwrite) 2106423bad60SNick Piggin { 2107423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 2108423bad60SNick Piggin pte_t *pte, entry; 2109423bad60SNick Piggin spinlock_t *ptl; 2110423bad60SNick Piggin 2111423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 2112423bad60SNick Piggin if (!pte) 21139b5a8e00SMatthew Wilcox return VM_FAULT_OOM; 2114b2770da6SRoss Zwisler if (!pte_none(*pte)) { 2115b2770da6SRoss Zwisler if (mkwrite) { 2116b2770da6SRoss Zwisler /* 2117b2770da6SRoss Zwisler * For read faults on private mappings the PFN passed 2118b2770da6SRoss Zwisler * in may not match the PFN we have mapped if the 2119b2770da6SRoss Zwisler * mapped PFN is a writeable COW page. In the mkwrite 2120b2770da6SRoss Zwisler * case we are creating a writable PTE for a shared 2121f2c57d91SJan Kara * mapping and we expect the PFNs to match. If they 2122f2c57d91SJan Kara * don't match, we are likely racing with block 2123f2c57d91SJan Kara * allocation and mapping invalidation so just skip the 2124f2c57d91SJan Kara * update. 2125b2770da6SRoss Zwisler */ 2126f2c57d91SJan Kara if (pte_pfn(*pte) != pfn_t_to_pfn(pfn)) { 2127f2c57d91SJan Kara WARN_ON_ONCE(!is_zero_pfn(pte_pfn(*pte))); 2128423bad60SNick Piggin goto out_unlock; 2129f2c57d91SJan Kara } 2130cae85cb8SJan Kara entry = pte_mkyoung(*pte); 2131cae85cb8SJan Kara entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2132cae85cb8SJan Kara if (ptep_set_access_flags(vma, addr, pte, entry, 1)) 2133cae85cb8SJan Kara update_mmu_cache(vma, addr, pte); 2134cae85cb8SJan Kara } 2135b2770da6SRoss Zwisler goto out_unlock; 2136b2770da6SRoss Zwisler } 2137423bad60SNick Piggin 2138423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 213901c8f1c4SDan Williams if (pfn_t_devmap(pfn)) 214001c8f1c4SDan Williams entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); 214101c8f1c4SDan Williams else 214201c8f1c4SDan Williams entry = pte_mkspecial(pfn_t_pte(pfn, prot)); 2143b2770da6SRoss Zwisler 2144b2770da6SRoss Zwisler if (mkwrite) { 2145b2770da6SRoss Zwisler entry = pte_mkyoung(entry); 2146b2770da6SRoss Zwisler entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2147b2770da6SRoss Zwisler } 2148b2770da6SRoss Zwisler 2149423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 21504b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 2151423bad60SNick Piggin 2152423bad60SNick Piggin out_unlock: 2153423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 21549b5a8e00SMatthew Wilcox return VM_FAULT_NOPAGE; 2155423bad60SNick Piggin } 2156423bad60SNick Piggin 2157f5e6d1d5SMatthew Wilcox /** 2158f5e6d1d5SMatthew Wilcox * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot 2159f5e6d1d5SMatthew Wilcox * @vma: user vma to map to 2160f5e6d1d5SMatthew Wilcox * @addr: target user address of this page 2161f5e6d1d5SMatthew Wilcox * @pfn: source kernel pfn 2162f5e6d1d5SMatthew Wilcox * @pgprot: pgprot flags for the inserted page 2163f5e6d1d5SMatthew Wilcox * 2164a1a0aea5SRandy Dunlap * This is exactly like vmf_insert_pfn(), except that it allows drivers 2165f5e6d1d5SMatthew Wilcox * to override pgprot on a per-page basis. 2166f5e6d1d5SMatthew Wilcox * 2167f5e6d1d5SMatthew Wilcox * This only makes sense for IO mappings, and it makes no sense for 2168f5e6d1d5SMatthew Wilcox * COW mappings. In general, using multiple vmas is preferable; 2169ae2b01f3SMatthew Wilcox * vmf_insert_pfn_prot should only be used if using multiple VMAs is 2170f5e6d1d5SMatthew Wilcox * impractical. 2171f5e6d1d5SMatthew Wilcox * 217228d8b812SLorenzo Stoakes * pgprot typically only differs from @vma->vm_page_prot when drivers set 217328d8b812SLorenzo Stoakes * caching- and encryption bits different than those of @vma->vm_page_prot, 217428d8b812SLorenzo Stoakes * because the caching- or encryption mode may not be known at mmap() time. 217528d8b812SLorenzo Stoakes * 217628d8b812SLorenzo Stoakes * This is ok as long as @vma->vm_page_prot is not used by the core vm 217728d8b812SLorenzo Stoakes * to set caching and encryption bits for those vmas (except for COW pages). 217828d8b812SLorenzo Stoakes * This is ensured by core vm only modifying these page table entries using 217928d8b812SLorenzo Stoakes * functions that don't touch caching- or encryption bits, using pte_modify() 218028d8b812SLorenzo Stoakes * if needed. (See for example mprotect()). 218128d8b812SLorenzo Stoakes * 218228d8b812SLorenzo Stoakes * Also when new page-table entries are created, this is only done using the 218328d8b812SLorenzo Stoakes * fault() callback, and never using the value of vma->vm_page_prot, 218428d8b812SLorenzo Stoakes * except for page-table entries that point to anonymous pages as the result 218528d8b812SLorenzo Stoakes * of COW. 2186574c5b3dSThomas Hellstrom * 2187ae2b01f3SMatthew Wilcox * Context: Process context. May allocate using %GFP_KERNEL. 2188f5e6d1d5SMatthew Wilcox * Return: vm_fault_t value. 2189f5e6d1d5SMatthew Wilcox */ 2190f5e6d1d5SMatthew Wilcox vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, 2191f5e6d1d5SMatthew Wilcox unsigned long pfn, pgprot_t pgprot) 2192f5e6d1d5SMatthew Wilcox { 21936d958546SMatthew Wilcox /* 21946d958546SMatthew Wilcox * Technically, architectures with pte_special can avoid all these 21956d958546SMatthew Wilcox * restrictions (same for remap_pfn_range). However we would like 21966d958546SMatthew Wilcox * consistency in testing and feature parity among all, so we should 21976d958546SMatthew Wilcox * try to keep these invariants in place for everybody. 21986d958546SMatthew Wilcox */ 21996d958546SMatthew Wilcox BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 22006d958546SMatthew Wilcox BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 22016d958546SMatthew Wilcox (VM_PFNMAP|VM_MIXEDMAP)); 22026d958546SMatthew Wilcox BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 22036d958546SMatthew Wilcox BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 22046d958546SMatthew Wilcox 22056d958546SMatthew Wilcox if (addr < vma->vm_start || addr >= vma->vm_end) 22066d958546SMatthew Wilcox return VM_FAULT_SIGBUS; 22076d958546SMatthew Wilcox 22086d958546SMatthew Wilcox if (!pfn_modify_allowed(pfn, pgprot)) 22096d958546SMatthew Wilcox return VM_FAULT_SIGBUS; 22106d958546SMatthew Wilcox 22116d958546SMatthew Wilcox track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); 22126d958546SMatthew Wilcox 22139b5a8e00SMatthew Wilcox return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, 22146d958546SMatthew Wilcox false); 2215f5e6d1d5SMatthew Wilcox } 2216f5e6d1d5SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_pfn_prot); 2217e0dc0d8fSNick Piggin 2218ae2b01f3SMatthew Wilcox /** 2219ae2b01f3SMatthew Wilcox * vmf_insert_pfn - insert single pfn into user vma 2220ae2b01f3SMatthew Wilcox * @vma: user vma to map to 2221ae2b01f3SMatthew Wilcox * @addr: target user address of this page 2222ae2b01f3SMatthew Wilcox * @pfn: source kernel pfn 2223ae2b01f3SMatthew Wilcox * 2224ae2b01f3SMatthew Wilcox * Similar to vm_insert_page, this allows drivers to insert individual pages 2225ae2b01f3SMatthew Wilcox * they've allocated into a user vma. Same comments apply. 2226ae2b01f3SMatthew Wilcox * 2227ae2b01f3SMatthew Wilcox * This function should only be called from a vm_ops->fault handler, and 2228ae2b01f3SMatthew Wilcox * in that case the handler should return the result of this function. 2229ae2b01f3SMatthew Wilcox * 2230ae2b01f3SMatthew Wilcox * vma cannot be a COW mapping. 2231ae2b01f3SMatthew Wilcox * 2232ae2b01f3SMatthew Wilcox * As this is called only for pages that do not currently exist, we 2233ae2b01f3SMatthew Wilcox * do not need to flush old virtual caches or the TLB. 2234ae2b01f3SMatthew Wilcox * 2235ae2b01f3SMatthew Wilcox * Context: Process context. May allocate using %GFP_KERNEL. 2236ae2b01f3SMatthew Wilcox * Return: vm_fault_t value. 2237ae2b01f3SMatthew Wilcox */ 2238ae2b01f3SMatthew Wilcox vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2239ae2b01f3SMatthew Wilcox unsigned long pfn) 2240ae2b01f3SMatthew Wilcox { 2241ae2b01f3SMatthew Wilcox return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); 2242ae2b01f3SMatthew Wilcox } 2243ae2b01f3SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_pfn); 2244ae2b01f3SMatthew Wilcox 2245785a3fabSDan Williams static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) 2246785a3fabSDan Williams { 2247785a3fabSDan Williams /* these checks mirror the abort conditions in vm_normal_page */ 2248785a3fabSDan Williams if (vma->vm_flags & VM_MIXEDMAP) 2249785a3fabSDan Williams return true; 2250785a3fabSDan Williams if (pfn_t_devmap(pfn)) 2251785a3fabSDan Williams return true; 2252785a3fabSDan Williams if (pfn_t_special(pfn)) 2253785a3fabSDan Williams return true; 2254785a3fabSDan Williams if (is_zero_pfn(pfn_t_to_pfn(pfn))) 2255785a3fabSDan Williams return true; 2256785a3fabSDan Williams return false; 2257785a3fabSDan Williams } 2258785a3fabSDan Williams 225979f3aa5bSMatthew Wilcox static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma, 226028d8b812SLorenzo Stoakes unsigned long addr, pfn_t pfn, bool mkwrite) 2261423bad60SNick Piggin { 226228d8b812SLorenzo Stoakes pgprot_t pgprot = vma->vm_page_prot; 226379f3aa5bSMatthew Wilcox int err; 226487744ab3SDan Williams 2265785a3fabSDan Williams BUG_ON(!vm_mixed_ok(vma, pfn)); 2266423bad60SNick Piggin 2267423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 226879f3aa5bSMatthew Wilcox return VM_FAULT_SIGBUS; 2269308a047cSBorislav Petkov 2270308a047cSBorislav Petkov track_pfn_insert(vma, &pgprot, pfn); 2271423bad60SNick Piggin 227242e4089cSAndi Kleen if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) 227379f3aa5bSMatthew Wilcox return VM_FAULT_SIGBUS; 227442e4089cSAndi Kleen 2275423bad60SNick Piggin /* 2276423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 2277423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 2278423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 227962eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 228062eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 2281423bad60SNick Piggin */ 228200b3a331SLaurent Dufour if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && 228300b3a331SLaurent Dufour !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { 2284423bad60SNick Piggin struct page *page; 2285423bad60SNick Piggin 228603fc2da6SDan Williams /* 228703fc2da6SDan Williams * At this point we are committed to insert_page() 228803fc2da6SDan Williams * regardless of whether the caller specified flags that 228903fc2da6SDan Williams * result in pfn_t_has_page() == false. 229003fc2da6SDan Williams */ 229103fc2da6SDan Williams page = pfn_to_page(pfn_t_to_pfn(pfn)); 229279f3aa5bSMatthew Wilcox err = insert_page(vma, addr, page, pgprot); 229379f3aa5bSMatthew Wilcox } else { 22949b5a8e00SMatthew Wilcox return insert_pfn(vma, addr, pfn, pgprot, mkwrite); 2295423bad60SNick Piggin } 2296b2770da6SRoss Zwisler 22975d747637SMatthew Wilcox if (err == -ENOMEM) 22985d747637SMatthew Wilcox return VM_FAULT_OOM; 22995d747637SMatthew Wilcox if (err < 0 && err != -EBUSY) 23005d747637SMatthew Wilcox return VM_FAULT_SIGBUS; 23015d747637SMatthew Wilcox 23025d747637SMatthew Wilcox return VM_FAULT_NOPAGE; 2303423bad60SNick Piggin } 230479f3aa5bSMatthew Wilcox 230579f3aa5bSMatthew Wilcox vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 230679f3aa5bSMatthew Wilcox pfn_t pfn) 230779f3aa5bSMatthew Wilcox { 230828d8b812SLorenzo Stoakes return __vm_insert_mixed(vma, addr, pfn, false); 230979f3aa5bSMatthew Wilcox } 23105d747637SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_mixed); 2311423bad60SNick Piggin 2312ab77dab4SSouptick Joarder /* 2313ab77dab4SSouptick Joarder * If the insertion of PTE failed because someone else already added a 2314ab77dab4SSouptick Joarder * different entry in the mean time, we treat that as success as we assume 2315ab77dab4SSouptick Joarder * the same entry was actually inserted. 2316ab77dab4SSouptick Joarder */ 2317ab77dab4SSouptick Joarder vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, 2318ab77dab4SSouptick Joarder unsigned long addr, pfn_t pfn) 2319b2770da6SRoss Zwisler { 232028d8b812SLorenzo Stoakes return __vm_insert_mixed(vma, addr, pfn, true); 2321b2770da6SRoss Zwisler } 2322ab77dab4SSouptick Joarder EXPORT_SYMBOL(vmf_insert_mixed_mkwrite); 2323b2770da6SRoss Zwisler 2324a145dd41SLinus Torvalds /* 23251da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 23261da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 23271da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 23281da177e4SLinus Torvalds */ 23291da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 23301da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23311da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23321da177e4SLinus Torvalds { 233390a3e375SMiaohe Lin pte_t *pte, *mapped_pte; 2334c74df32cSHugh Dickins spinlock_t *ptl; 233542e4089cSAndi Kleen int err = 0; 23361da177e4SLinus Torvalds 233790a3e375SMiaohe Lin mapped_pte = pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 23381da177e4SLinus Torvalds if (!pte) 23391da177e4SLinus Torvalds return -ENOMEM; 23406606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 23411da177e4SLinus Torvalds do { 23421da177e4SLinus Torvalds BUG_ON(!pte_none(*pte)); 234342e4089cSAndi Kleen if (!pfn_modify_allowed(pfn, prot)) { 234442e4089cSAndi Kleen err = -EACCES; 234542e4089cSAndi Kleen break; 234642e4089cSAndi Kleen } 23477e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 23481da177e4SLinus Torvalds pfn++; 23491da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 23506606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 235190a3e375SMiaohe Lin pte_unmap_unlock(mapped_pte, ptl); 235242e4089cSAndi Kleen return err; 23531da177e4SLinus Torvalds } 23541da177e4SLinus Torvalds 23551da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 23561da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23571da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23581da177e4SLinus Torvalds { 23591da177e4SLinus Torvalds pmd_t *pmd; 23601da177e4SLinus Torvalds unsigned long next; 236142e4089cSAndi Kleen int err; 23621da177e4SLinus Torvalds 23631da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 23641da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 23651da177e4SLinus Torvalds if (!pmd) 23661da177e4SLinus Torvalds return -ENOMEM; 2367f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 23681da177e4SLinus Torvalds do { 23691da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 237042e4089cSAndi Kleen err = remap_pte_range(mm, pmd, addr, next, 237142e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 237242e4089cSAndi Kleen if (err) 237342e4089cSAndi Kleen return err; 23741da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 23751da177e4SLinus Torvalds return 0; 23761da177e4SLinus Torvalds } 23771da177e4SLinus Torvalds 2378c2febafcSKirill A. Shutemov static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, 23791da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23801da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23811da177e4SLinus Torvalds { 23821da177e4SLinus Torvalds pud_t *pud; 23831da177e4SLinus Torvalds unsigned long next; 238442e4089cSAndi Kleen int err; 23851da177e4SLinus Torvalds 23861da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 2387c2febafcSKirill A. Shutemov pud = pud_alloc(mm, p4d, addr); 23881da177e4SLinus Torvalds if (!pud) 23891da177e4SLinus Torvalds return -ENOMEM; 23901da177e4SLinus Torvalds do { 23911da177e4SLinus Torvalds next = pud_addr_end(addr, end); 239242e4089cSAndi Kleen err = remap_pmd_range(mm, pud, addr, next, 239342e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 239442e4089cSAndi Kleen if (err) 239542e4089cSAndi Kleen return err; 23961da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 23971da177e4SLinus Torvalds return 0; 23981da177e4SLinus Torvalds } 23991da177e4SLinus Torvalds 2400c2febafcSKirill A. Shutemov static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, 2401c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 2402c2febafcSKirill A. Shutemov unsigned long pfn, pgprot_t prot) 2403c2febafcSKirill A. Shutemov { 2404c2febafcSKirill A. Shutemov p4d_t *p4d; 2405c2febafcSKirill A. Shutemov unsigned long next; 240642e4089cSAndi Kleen int err; 2407c2febafcSKirill A. Shutemov 2408c2febafcSKirill A. Shutemov pfn -= addr >> PAGE_SHIFT; 2409c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 2410c2febafcSKirill A. Shutemov if (!p4d) 2411c2febafcSKirill A. Shutemov return -ENOMEM; 2412c2febafcSKirill A. Shutemov do { 2413c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 241442e4089cSAndi Kleen err = remap_pud_range(mm, p4d, addr, next, 241542e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 241642e4089cSAndi Kleen if (err) 241742e4089cSAndi Kleen return err; 2418c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 2419c2febafcSKirill A. Shutemov return 0; 2420c2febafcSKirill A. Shutemov } 2421c2febafcSKirill A. Shutemov 242274ffa5a3SChristoph Hellwig /* 242374ffa5a3SChristoph Hellwig * Variant of remap_pfn_range that does not call track_pfn_remap. The caller 242474ffa5a3SChristoph Hellwig * must have pre-validated the caching bits of the pgprot_t. 2425bfa5bf6dSRolf Eike Beer */ 242674ffa5a3SChristoph Hellwig int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr, 24271da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 24281da177e4SLinus Torvalds { 24291da177e4SLinus Torvalds pgd_t *pgd; 24301da177e4SLinus Torvalds unsigned long next; 24312d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 24321da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 24331da177e4SLinus Torvalds int err; 24341da177e4SLinus Torvalds 24350c4123e3SAlex Zhang if (WARN_ON_ONCE(!PAGE_ALIGNED(addr))) 24360c4123e3SAlex Zhang return -EINVAL; 24370c4123e3SAlex Zhang 24381da177e4SLinus Torvalds /* 24391da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 24401da177e4SLinus Torvalds * rest of the world about it: 24411da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 24421da177e4SLinus Torvalds * (accesses can have side effects). 24436aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 24446aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 24456aab341eSLinus Torvalds * with them. 2446314e51b9SKonstantin Khlebnikov * VM_DONTEXPAND 2447314e51b9SKonstantin Khlebnikov * Disable vma merging and expanding with mremap(). 2448314e51b9SKonstantin Khlebnikov * VM_DONTDUMP 2449314e51b9SKonstantin Khlebnikov * Omit vma from core dump, even when VM_IO turned off. 2450fb155c16SLinus Torvalds * 2451fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 2452fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 2453fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 2454b3b9c293SKonstantin Khlebnikov * See vm_normal_page() for details. 24551da177e4SLinus Torvalds */ 2456b3b9c293SKonstantin Khlebnikov if (is_cow_mapping(vma->vm_flags)) { 2457b3b9c293SKonstantin Khlebnikov if (addr != vma->vm_start || end != vma->vm_end) 2458b3b9c293SKonstantin Khlebnikov return -EINVAL; 24596aab341eSLinus Torvalds vma->vm_pgoff = pfn; 2460b3b9c293SKonstantin Khlebnikov } 2461b3b9c293SKonstantin Khlebnikov 24621c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP); 24631da177e4SLinus Torvalds 24641da177e4SLinus Torvalds BUG_ON(addr >= end); 24651da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 24661da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 24671da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 24681da177e4SLinus Torvalds do { 24691da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 2470c2febafcSKirill A. Shutemov err = remap_p4d_range(mm, pgd, addr, next, 24711da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 24721da177e4SLinus Torvalds if (err) 247374ffa5a3SChristoph Hellwig return err; 24741da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 24752ab64037Svenkatesh.pallipadi@intel.com 247674ffa5a3SChristoph Hellwig return 0; 247774ffa5a3SChristoph Hellwig } 24782ab64037Svenkatesh.pallipadi@intel.com 247974ffa5a3SChristoph Hellwig /** 248074ffa5a3SChristoph Hellwig * remap_pfn_range - remap kernel memory to userspace 248174ffa5a3SChristoph Hellwig * @vma: user vma to map to 248274ffa5a3SChristoph Hellwig * @addr: target page aligned user address to start at 248374ffa5a3SChristoph Hellwig * @pfn: page frame number of kernel physical memory address 248474ffa5a3SChristoph Hellwig * @size: size of mapping area 248574ffa5a3SChristoph Hellwig * @prot: page protection flags for this mapping 248674ffa5a3SChristoph Hellwig * 248774ffa5a3SChristoph Hellwig * Note: this is only safe if the mm semaphore is held when called. 248874ffa5a3SChristoph Hellwig * 248974ffa5a3SChristoph Hellwig * Return: %0 on success, negative error code otherwise. 249074ffa5a3SChristoph Hellwig */ 249174ffa5a3SChristoph Hellwig int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 249274ffa5a3SChristoph Hellwig unsigned long pfn, unsigned long size, pgprot_t prot) 249374ffa5a3SChristoph Hellwig { 249474ffa5a3SChristoph Hellwig int err; 249574ffa5a3SChristoph Hellwig 249674ffa5a3SChristoph Hellwig err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); 249774ffa5a3SChristoph Hellwig if (err) 249874ffa5a3SChristoph Hellwig return -EINVAL; 249974ffa5a3SChristoph Hellwig 250074ffa5a3SChristoph Hellwig err = remap_pfn_range_notrack(vma, addr, pfn, size, prot); 250174ffa5a3SChristoph Hellwig if (err) 250268f48381SSuren Baghdasaryan untrack_pfn(vma, pfn, PAGE_ALIGN(size), true); 25031da177e4SLinus Torvalds return err; 25041da177e4SLinus Torvalds } 25051da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 25061da177e4SLinus Torvalds 2507b4cbb197SLinus Torvalds /** 2508b4cbb197SLinus Torvalds * vm_iomap_memory - remap memory to userspace 2509b4cbb197SLinus Torvalds * @vma: user vma to map to 2510abd69b9eSWang Wenhu * @start: start of the physical memory to be mapped 2511b4cbb197SLinus Torvalds * @len: size of area 2512b4cbb197SLinus Torvalds * 2513b4cbb197SLinus Torvalds * This is a simplified io_remap_pfn_range() for common driver use. The 2514b4cbb197SLinus Torvalds * driver just needs to give us the physical memory range to be mapped, 2515b4cbb197SLinus Torvalds * we'll figure out the rest from the vma information. 2516b4cbb197SLinus Torvalds * 2517b4cbb197SLinus Torvalds * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get 2518b4cbb197SLinus Torvalds * whatever write-combining details or similar. 2519a862f68aSMike Rapoport * 2520a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 2521b4cbb197SLinus Torvalds */ 2522b4cbb197SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 2523b4cbb197SLinus Torvalds { 2524b4cbb197SLinus Torvalds unsigned long vm_len, pfn, pages; 2525b4cbb197SLinus Torvalds 2526b4cbb197SLinus Torvalds /* Check that the physical memory area passed in looks valid */ 2527b4cbb197SLinus Torvalds if (start + len < start) 2528b4cbb197SLinus Torvalds return -EINVAL; 2529b4cbb197SLinus Torvalds /* 2530b4cbb197SLinus Torvalds * You *really* shouldn't map things that aren't page-aligned, 2531b4cbb197SLinus Torvalds * but we've historically allowed it because IO memory might 2532b4cbb197SLinus Torvalds * just have smaller alignment. 2533b4cbb197SLinus Torvalds */ 2534b4cbb197SLinus Torvalds len += start & ~PAGE_MASK; 2535b4cbb197SLinus Torvalds pfn = start >> PAGE_SHIFT; 2536b4cbb197SLinus Torvalds pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; 2537b4cbb197SLinus Torvalds if (pfn + pages < pfn) 2538b4cbb197SLinus Torvalds return -EINVAL; 2539b4cbb197SLinus Torvalds 2540b4cbb197SLinus Torvalds /* We start the mapping 'vm_pgoff' pages into the area */ 2541b4cbb197SLinus Torvalds if (vma->vm_pgoff > pages) 2542b4cbb197SLinus Torvalds return -EINVAL; 2543b4cbb197SLinus Torvalds pfn += vma->vm_pgoff; 2544b4cbb197SLinus Torvalds pages -= vma->vm_pgoff; 2545b4cbb197SLinus Torvalds 2546b4cbb197SLinus Torvalds /* Can we fit all of the mapping? */ 2547b4cbb197SLinus Torvalds vm_len = vma->vm_end - vma->vm_start; 2548b4cbb197SLinus Torvalds if (vm_len >> PAGE_SHIFT > pages) 2549b4cbb197SLinus Torvalds return -EINVAL; 2550b4cbb197SLinus Torvalds 2551b4cbb197SLinus Torvalds /* Ok, let it rip */ 2552b4cbb197SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 2553b4cbb197SLinus Torvalds } 2554b4cbb197SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 2555b4cbb197SLinus Torvalds 2556aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 2557aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2558e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2559e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2560aee16b3cSJeremy Fitzhardinge { 25618abb50c7SMiaohe Lin pte_t *pte, *mapped_pte; 2562be1db475SDaniel Axtens int err = 0; 25633f649ab7SKees Cook spinlock_t *ptl; 2564aee16b3cSJeremy Fitzhardinge 2565be1db475SDaniel Axtens if (create) { 25668abb50c7SMiaohe Lin mapped_pte = pte = (mm == &init_mm) ? 2567e80d3909SJoerg Roedel pte_alloc_kernel_track(pmd, addr, mask) : 2568aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 2569aee16b3cSJeremy Fitzhardinge if (!pte) 2570aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2571be1db475SDaniel Axtens } else { 25728abb50c7SMiaohe Lin mapped_pte = pte = (mm == &init_mm) ? 2573be1db475SDaniel Axtens pte_offset_kernel(pmd, addr) : 2574be1db475SDaniel Axtens pte_offset_map_lock(mm, pmd, addr, &ptl); 2575be1db475SDaniel Axtens } 2576aee16b3cSJeremy Fitzhardinge 2577aee16b3cSJeremy Fitzhardinge BUG_ON(pmd_huge(*pmd)); 2578aee16b3cSJeremy Fitzhardinge 257938e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 258038e0edb1SJeremy Fitzhardinge 2581eeb4a05fSChristoph Hellwig if (fn) { 2582aee16b3cSJeremy Fitzhardinge do { 2583be1db475SDaniel Axtens if (create || !pte_none(*pte)) { 25848b1e0f81SAnshuman Khandual err = fn(pte++, addr, data); 2585aee16b3cSJeremy Fitzhardinge if (err) 2586aee16b3cSJeremy Fitzhardinge break; 2587be1db475SDaniel Axtens } 2588c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 2589eeb4a05fSChristoph Hellwig } 2590e80d3909SJoerg Roedel *mask |= PGTBL_PTE_MODIFIED; 2591aee16b3cSJeremy Fitzhardinge 259238e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 259338e0edb1SJeremy Fitzhardinge 2594aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 25958abb50c7SMiaohe Lin pte_unmap_unlock(mapped_pte, ptl); 2596aee16b3cSJeremy Fitzhardinge return err; 2597aee16b3cSJeremy Fitzhardinge } 2598aee16b3cSJeremy Fitzhardinge 2599aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 2600aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2601e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2602e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2603aee16b3cSJeremy Fitzhardinge { 2604aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 2605aee16b3cSJeremy Fitzhardinge unsigned long next; 2606be1db475SDaniel Axtens int err = 0; 2607aee16b3cSJeremy Fitzhardinge 2608ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 2609ceb86879SAndi Kleen 2610be1db475SDaniel Axtens if (create) { 2611e80d3909SJoerg Roedel pmd = pmd_alloc_track(mm, pud, addr, mask); 2612aee16b3cSJeremy Fitzhardinge if (!pmd) 2613aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2614be1db475SDaniel Axtens } else { 2615be1db475SDaniel Axtens pmd = pmd_offset(pud, addr); 2616be1db475SDaniel Axtens } 2617aee16b3cSJeremy Fitzhardinge do { 2618aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 26190c95cba4SNicholas Piggin if (pmd_none(*pmd) && !create) 26200c95cba4SNicholas Piggin continue; 26210c95cba4SNicholas Piggin if (WARN_ON_ONCE(pmd_leaf(*pmd))) 26220c95cba4SNicholas Piggin return -EINVAL; 26230c95cba4SNicholas Piggin if (!pmd_none(*pmd) && WARN_ON_ONCE(pmd_bad(*pmd))) { 26240c95cba4SNicholas Piggin if (!create) 26250c95cba4SNicholas Piggin continue; 26260c95cba4SNicholas Piggin pmd_clear_bad(pmd); 26270c95cba4SNicholas Piggin } 26280c95cba4SNicholas Piggin err = apply_to_pte_range(mm, pmd, addr, next, 26290c95cba4SNicholas Piggin fn, data, create, mask); 2630aee16b3cSJeremy Fitzhardinge if (err) 2631aee16b3cSJeremy Fitzhardinge break; 2632aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 26330c95cba4SNicholas Piggin 2634aee16b3cSJeremy Fitzhardinge return err; 2635aee16b3cSJeremy Fitzhardinge } 2636aee16b3cSJeremy Fitzhardinge 2637c2febafcSKirill A. Shutemov static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, 2638aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2639e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2640e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2641aee16b3cSJeremy Fitzhardinge { 2642aee16b3cSJeremy Fitzhardinge pud_t *pud; 2643aee16b3cSJeremy Fitzhardinge unsigned long next; 2644be1db475SDaniel Axtens int err = 0; 2645aee16b3cSJeremy Fitzhardinge 2646be1db475SDaniel Axtens if (create) { 2647e80d3909SJoerg Roedel pud = pud_alloc_track(mm, p4d, addr, mask); 2648aee16b3cSJeremy Fitzhardinge if (!pud) 2649aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2650be1db475SDaniel Axtens } else { 2651be1db475SDaniel Axtens pud = pud_offset(p4d, addr); 2652be1db475SDaniel Axtens } 2653aee16b3cSJeremy Fitzhardinge do { 2654aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 26550c95cba4SNicholas Piggin if (pud_none(*pud) && !create) 26560c95cba4SNicholas Piggin continue; 26570c95cba4SNicholas Piggin if (WARN_ON_ONCE(pud_leaf(*pud))) 26580c95cba4SNicholas Piggin return -EINVAL; 26590c95cba4SNicholas Piggin if (!pud_none(*pud) && WARN_ON_ONCE(pud_bad(*pud))) { 26600c95cba4SNicholas Piggin if (!create) 26610c95cba4SNicholas Piggin continue; 26620c95cba4SNicholas Piggin pud_clear_bad(pud); 26630c95cba4SNicholas Piggin } 26640c95cba4SNicholas Piggin err = apply_to_pmd_range(mm, pud, addr, next, 26650c95cba4SNicholas Piggin fn, data, create, mask); 2666aee16b3cSJeremy Fitzhardinge if (err) 2667aee16b3cSJeremy Fitzhardinge break; 2668aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 26690c95cba4SNicholas Piggin 2670aee16b3cSJeremy Fitzhardinge return err; 2671aee16b3cSJeremy Fitzhardinge } 2672aee16b3cSJeremy Fitzhardinge 2673c2febafcSKirill A. Shutemov static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, 2674c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 2675e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2676e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2677c2febafcSKirill A. Shutemov { 2678c2febafcSKirill A. Shutemov p4d_t *p4d; 2679c2febafcSKirill A. Shutemov unsigned long next; 2680be1db475SDaniel Axtens int err = 0; 2681c2febafcSKirill A. Shutemov 2682be1db475SDaniel Axtens if (create) { 2683e80d3909SJoerg Roedel p4d = p4d_alloc_track(mm, pgd, addr, mask); 2684c2febafcSKirill A. Shutemov if (!p4d) 2685c2febafcSKirill A. Shutemov return -ENOMEM; 2686be1db475SDaniel Axtens } else { 2687be1db475SDaniel Axtens p4d = p4d_offset(pgd, addr); 2688be1db475SDaniel Axtens } 2689c2febafcSKirill A. Shutemov do { 2690c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 26910c95cba4SNicholas Piggin if (p4d_none(*p4d) && !create) 26920c95cba4SNicholas Piggin continue; 26930c95cba4SNicholas Piggin if (WARN_ON_ONCE(p4d_leaf(*p4d))) 26940c95cba4SNicholas Piggin return -EINVAL; 26950c95cba4SNicholas Piggin if (!p4d_none(*p4d) && WARN_ON_ONCE(p4d_bad(*p4d))) { 26960c95cba4SNicholas Piggin if (!create) 26970c95cba4SNicholas Piggin continue; 26980c95cba4SNicholas Piggin p4d_clear_bad(p4d); 26990c95cba4SNicholas Piggin } 27000c95cba4SNicholas Piggin err = apply_to_pud_range(mm, p4d, addr, next, 27010c95cba4SNicholas Piggin fn, data, create, mask); 2702c2febafcSKirill A. Shutemov if (err) 2703c2febafcSKirill A. Shutemov break; 2704c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 27050c95cba4SNicholas Piggin 2706c2febafcSKirill A. Shutemov return err; 2707c2febafcSKirill A. Shutemov } 2708c2febafcSKirill A. Shutemov 2709be1db475SDaniel Axtens static int __apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2710be1db475SDaniel Axtens unsigned long size, pte_fn_t fn, 2711be1db475SDaniel Axtens void *data, bool create) 2712be1db475SDaniel Axtens { 2713be1db475SDaniel Axtens pgd_t *pgd; 2714e80d3909SJoerg Roedel unsigned long start = addr, next; 2715be1db475SDaniel Axtens unsigned long end = addr + size; 2716e80d3909SJoerg Roedel pgtbl_mod_mask mask = 0; 2717be1db475SDaniel Axtens int err = 0; 2718be1db475SDaniel Axtens 2719be1db475SDaniel Axtens if (WARN_ON(addr >= end)) 2720be1db475SDaniel Axtens return -EINVAL; 2721be1db475SDaniel Axtens 2722be1db475SDaniel Axtens pgd = pgd_offset(mm, addr); 2723be1db475SDaniel Axtens do { 2724be1db475SDaniel Axtens next = pgd_addr_end(addr, end); 27250c95cba4SNicholas Piggin if (pgd_none(*pgd) && !create) 2726be1db475SDaniel Axtens continue; 27270c95cba4SNicholas Piggin if (WARN_ON_ONCE(pgd_leaf(*pgd))) 27280c95cba4SNicholas Piggin return -EINVAL; 27290c95cba4SNicholas Piggin if (!pgd_none(*pgd) && WARN_ON_ONCE(pgd_bad(*pgd))) { 27300c95cba4SNicholas Piggin if (!create) 27310c95cba4SNicholas Piggin continue; 27320c95cba4SNicholas Piggin pgd_clear_bad(pgd); 27330c95cba4SNicholas Piggin } 27340c95cba4SNicholas Piggin err = apply_to_p4d_range(mm, pgd, addr, next, 27350c95cba4SNicholas Piggin fn, data, create, &mask); 2736be1db475SDaniel Axtens if (err) 2737be1db475SDaniel Axtens break; 2738be1db475SDaniel Axtens } while (pgd++, addr = next, addr != end); 2739be1db475SDaniel Axtens 2740e80d3909SJoerg Roedel if (mask & ARCH_PAGE_TABLE_SYNC_MASK) 2741e80d3909SJoerg Roedel arch_sync_kernel_mappings(start, start + size); 2742e80d3909SJoerg Roedel 2743be1db475SDaniel Axtens return err; 2744be1db475SDaniel Axtens } 2745be1db475SDaniel Axtens 2746aee16b3cSJeremy Fitzhardinge /* 2747aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 2748aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 2749aee16b3cSJeremy Fitzhardinge */ 2750aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2751aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 2752aee16b3cSJeremy Fitzhardinge { 2753be1db475SDaniel Axtens return __apply_to_page_range(mm, addr, size, fn, data, true); 2754aee16b3cSJeremy Fitzhardinge } 2755aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 2756aee16b3cSJeremy Fitzhardinge 27571da177e4SLinus Torvalds /* 2758be1db475SDaniel Axtens * Scan a region of virtual memory, calling a provided function on 2759be1db475SDaniel Axtens * each leaf page table where it exists. 2760be1db475SDaniel Axtens * 2761be1db475SDaniel Axtens * Unlike apply_to_page_range, this does _not_ fill in page tables 2762be1db475SDaniel Axtens * where they are absent. 2763be1db475SDaniel Axtens */ 2764be1db475SDaniel Axtens int apply_to_existing_page_range(struct mm_struct *mm, unsigned long addr, 2765be1db475SDaniel Axtens unsigned long size, pte_fn_t fn, void *data) 2766be1db475SDaniel Axtens { 2767be1db475SDaniel Axtens return __apply_to_page_range(mm, addr, size, fn, data, false); 2768be1db475SDaniel Axtens } 2769be1db475SDaniel Axtens EXPORT_SYMBOL_GPL(apply_to_existing_page_range); 2770be1db475SDaniel Axtens 2771be1db475SDaniel Axtens /* 27729b4bdd2fSKirill A. Shutemov * handle_pte_fault chooses page fault handler according to an entry which was 27739b4bdd2fSKirill A. Shutemov * read non-atomically. Before making any commitment, on those architectures 27749b4bdd2fSKirill A. Shutemov * or configurations (e.g. i386 with PAE) which might give a mix of unmatched 27759b4bdd2fSKirill A. Shutemov * parts, do_swap_page must check under lock before unmapping the pte and 27769b4bdd2fSKirill A. Shutemov * proceeding (but do_wp_page is only called after already making such a check; 2777a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 27788f4e2101SHugh Dickins */ 27792ca99358SPeter Xu static inline int pte_unmap_same(struct vm_fault *vmf) 27808f4e2101SHugh Dickins { 27818f4e2101SHugh Dickins int same = 1; 2782923717cbSThomas Gleixner #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPTION) 27838f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 27842ca99358SPeter Xu spinlock_t *ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); 27854c21e2f2SHugh Dickins spin_lock(ptl); 27862ca99358SPeter Xu same = pte_same(*vmf->pte, vmf->orig_pte); 27874c21e2f2SHugh Dickins spin_unlock(ptl); 27888f4e2101SHugh Dickins } 27898f4e2101SHugh Dickins #endif 27902ca99358SPeter Xu pte_unmap(vmf->pte); 27912ca99358SPeter Xu vmf->pte = NULL; 27928f4e2101SHugh Dickins return same; 27938f4e2101SHugh Dickins } 27948f4e2101SHugh Dickins 2795a873dfe1STony Luck /* 2796a873dfe1STony Luck * Return: 2797a873dfe1STony Luck * 0: copied succeeded 2798a873dfe1STony Luck * -EHWPOISON: copy failed due to hwpoison in source page 2799a873dfe1STony Luck * -EAGAIN: copied failed (some other reason) 2800a873dfe1STony Luck */ 2801a873dfe1STony Luck static inline int __wp_page_copy_user(struct page *dst, struct page *src, 280283d116c5SJia He struct vm_fault *vmf) 28036aab341eSLinus Torvalds { 2804a873dfe1STony Luck int ret; 280583d116c5SJia He void *kaddr; 280683d116c5SJia He void __user *uaddr; 2807c3e5ea6eSKirill A. Shutemov bool locked = false; 280883d116c5SJia He struct vm_area_struct *vma = vmf->vma; 280983d116c5SJia He struct mm_struct *mm = vma->vm_mm; 281083d116c5SJia He unsigned long addr = vmf->address; 281183d116c5SJia He 281283d116c5SJia He if (likely(src)) { 2813d302c239STony Luck if (copy_mc_user_highpage(dst, src, addr, vma)) { 2814d302c239STony Luck memory_failure_queue(page_to_pfn(src), 0); 2815a873dfe1STony Luck return -EHWPOISON; 2816d302c239STony Luck } 2817a873dfe1STony Luck return 0; 281883d116c5SJia He } 281983d116c5SJia He 28206aab341eSLinus Torvalds /* 28216aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 28226aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 28236aab341eSLinus Torvalds * just copying from the original user address. If that 28246aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 28256aab341eSLinus Torvalds */ 282683d116c5SJia He kaddr = kmap_atomic(dst); 282783d116c5SJia He uaddr = (void __user *)(addr & PAGE_MASK); 282883d116c5SJia He 282983d116c5SJia He /* 283083d116c5SJia He * On architectures with software "accessed" bits, we would 283183d116c5SJia He * take a double page fault, so mark it accessed here. 283283d116c5SJia He */ 2833e1fd09e3SYu Zhao if (!arch_has_hw_pte_young() && !pte_young(vmf->orig_pte)) { 283483d116c5SJia He pte_t entry; 283583d116c5SJia He 283683d116c5SJia He vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); 2837c3e5ea6eSKirill A. Shutemov locked = true; 283883d116c5SJia He if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) { 283983d116c5SJia He /* 284083d116c5SJia He * Other thread has already handled the fault 28417df67697SBibo Mao * and update local tlb only 284283d116c5SJia He */ 28437df67697SBibo Mao update_mmu_tlb(vma, addr, vmf->pte); 2844a873dfe1STony Luck ret = -EAGAIN; 284583d116c5SJia He goto pte_unlock; 284683d116c5SJia He } 284783d116c5SJia He 284883d116c5SJia He entry = pte_mkyoung(vmf->orig_pte); 284983d116c5SJia He if (ptep_set_access_flags(vma, addr, vmf->pte, entry, 0)) 285083d116c5SJia He update_mmu_cache(vma, addr, vmf->pte); 285183d116c5SJia He } 28525d2a2dbbSLinus Torvalds 28535d2a2dbbSLinus Torvalds /* 28545d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 28555d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 28565d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 28575d2a2dbbSLinus Torvalds * zeroes. 28585d2a2dbbSLinus Torvalds */ 285983d116c5SJia He if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { 2860c3e5ea6eSKirill A. Shutemov if (locked) 2861c3e5ea6eSKirill A. Shutemov goto warn; 2862c3e5ea6eSKirill A. Shutemov 2863c3e5ea6eSKirill A. Shutemov /* Re-validate under PTL if the page is still mapped */ 2864c3e5ea6eSKirill A. Shutemov vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); 2865c3e5ea6eSKirill A. Shutemov locked = true; 2866c3e5ea6eSKirill A. Shutemov if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) { 28677df67697SBibo Mao /* The PTE changed under us, update local tlb */ 28687df67697SBibo Mao update_mmu_tlb(vma, addr, vmf->pte); 2869a873dfe1STony Luck ret = -EAGAIN; 2870c3e5ea6eSKirill A. Shutemov goto pte_unlock; 2871c3e5ea6eSKirill A. Shutemov } 2872c3e5ea6eSKirill A. Shutemov 2873c3e5ea6eSKirill A. Shutemov /* 2874985ba004SEthon Paul * The same page can be mapped back since last copy attempt. 2875c3e5ea6eSKirill A. Shutemov * Try to copy again under PTL. 2876c3e5ea6eSKirill A. Shutemov */ 2877c3e5ea6eSKirill A. Shutemov if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { 287883d116c5SJia He /* 287983d116c5SJia He * Give a warn in case there can be some obscure 288083d116c5SJia He * use-case 288183d116c5SJia He */ 2882c3e5ea6eSKirill A. Shutemov warn: 288383d116c5SJia He WARN_ON_ONCE(1); 28843ecb01dfSJan Beulich clear_page(kaddr); 288583d116c5SJia He } 2886c3e5ea6eSKirill A. Shutemov } 288783d116c5SJia He 2888a873dfe1STony Luck ret = 0; 288983d116c5SJia He 289083d116c5SJia He pte_unlock: 2891c3e5ea6eSKirill A. Shutemov if (locked) 289283d116c5SJia He pte_unmap_unlock(vmf->pte, vmf->ptl); 28939b04c5feSCong Wang kunmap_atomic(kaddr); 2894c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 289583d116c5SJia He 289683d116c5SJia He return ret; 28976aab341eSLinus Torvalds } 28986aab341eSLinus Torvalds 2899c20cd45eSMichal Hocko static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) 2900c20cd45eSMichal Hocko { 2901c20cd45eSMichal Hocko struct file *vm_file = vma->vm_file; 2902c20cd45eSMichal Hocko 2903c20cd45eSMichal Hocko if (vm_file) 2904c20cd45eSMichal Hocko return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; 2905c20cd45eSMichal Hocko 2906c20cd45eSMichal Hocko /* 2907c20cd45eSMichal Hocko * Special mappings (e.g. VDSO) do not have any file so fake 2908c20cd45eSMichal Hocko * a default GFP_KERNEL for them. 2909c20cd45eSMichal Hocko */ 2910c20cd45eSMichal Hocko return GFP_KERNEL; 2911c20cd45eSMichal Hocko } 2912c20cd45eSMichal Hocko 29131da177e4SLinus Torvalds /* 2914fb09a464SKirill A. Shutemov * Notify the address space that the page is about to become writable so that 2915fb09a464SKirill A. Shutemov * it can prohibit this or wait for the page to get into an appropriate state. 2916fb09a464SKirill A. Shutemov * 2917fb09a464SKirill A. Shutemov * We do this without the lock held, so that it can sleep if it needs to. 2918fb09a464SKirill A. Shutemov */ 29192b740303SSouptick Joarder static vm_fault_t do_page_mkwrite(struct vm_fault *vmf) 2920fb09a464SKirill A. Shutemov { 29212b740303SSouptick Joarder vm_fault_t ret; 292238b8cb7fSJan Kara struct page *page = vmf->page; 292338b8cb7fSJan Kara unsigned int old_flags = vmf->flags; 2924fb09a464SKirill A. Shutemov 292538b8cb7fSJan Kara vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 2926fb09a464SKirill A. Shutemov 2927dc617f29SDarrick J. Wong if (vmf->vma->vm_file && 2928dc617f29SDarrick J. Wong IS_SWAPFILE(vmf->vma->vm_file->f_mapping->host)) 2929dc617f29SDarrick J. Wong return VM_FAULT_SIGBUS; 2930dc617f29SDarrick J. Wong 293111bac800SDave Jiang ret = vmf->vma->vm_ops->page_mkwrite(vmf); 293238b8cb7fSJan Kara /* Restore original flags so that caller is not surprised */ 293338b8cb7fSJan Kara vmf->flags = old_flags; 2934fb09a464SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 2935fb09a464SKirill A. Shutemov return ret; 2936fb09a464SKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) { 2937fb09a464SKirill A. Shutemov lock_page(page); 2938fb09a464SKirill A. Shutemov if (!page->mapping) { 2939fb09a464SKirill A. Shutemov unlock_page(page); 2940fb09a464SKirill A. Shutemov return 0; /* retry */ 2941fb09a464SKirill A. Shutemov } 2942fb09a464SKirill A. Shutemov ret |= VM_FAULT_LOCKED; 2943fb09a464SKirill A. Shutemov } else 2944fb09a464SKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(page), page); 2945fb09a464SKirill A. Shutemov return ret; 2946fb09a464SKirill A. Shutemov } 2947fb09a464SKirill A. Shutemov 2948fb09a464SKirill A. Shutemov /* 294997ba0c2bSJan Kara * Handle dirtying of a page in shared file mapping on a write fault. 29504e047f89SShachar Raindel * 295197ba0c2bSJan Kara * The function expects the page to be locked and unlocks it. 29524e047f89SShachar Raindel */ 295389b15332SJohannes Weiner static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf) 29544e047f89SShachar Raindel { 295589b15332SJohannes Weiner struct vm_area_struct *vma = vmf->vma; 29564e047f89SShachar Raindel struct address_space *mapping; 295789b15332SJohannes Weiner struct page *page = vmf->page; 295897ba0c2bSJan Kara bool dirtied; 295997ba0c2bSJan Kara bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; 29604e047f89SShachar Raindel 29614e047f89SShachar Raindel dirtied = set_page_dirty(page); 29624e047f89SShachar Raindel VM_BUG_ON_PAGE(PageAnon(page), page); 296397ba0c2bSJan Kara /* 296497ba0c2bSJan Kara * Take a local copy of the address_space - page.mapping may be zeroed 296597ba0c2bSJan Kara * by truncate after unlock_page(). The address_space itself remains 296697ba0c2bSJan Kara * pinned by vma->vm_file's reference. We rely on unlock_page()'s 296797ba0c2bSJan Kara * release semantics to prevent the compiler from undoing this copying. 296897ba0c2bSJan Kara */ 296997ba0c2bSJan Kara mapping = page_rmapping(page); 29704e047f89SShachar Raindel unlock_page(page); 29714e047f89SShachar Raindel 29724e047f89SShachar Raindel if (!page_mkwrite) 29734e047f89SShachar Raindel file_update_time(vma->vm_file); 297489b15332SJohannes Weiner 297589b15332SJohannes Weiner /* 297689b15332SJohannes Weiner * Throttle page dirtying rate down to writeback speed. 297789b15332SJohannes Weiner * 297889b15332SJohannes Weiner * mapping may be NULL here because some device drivers do not 297989b15332SJohannes Weiner * set page.mapping but still dirty their pages 298089b15332SJohannes Weiner * 2981c1e8d7c6SMichel Lespinasse * Drop the mmap_lock before waiting on IO, if we can. The file 298289b15332SJohannes Weiner * is pinning the mapping, as per above. 298389b15332SJohannes Weiner */ 298489b15332SJohannes Weiner if ((dirtied || page_mkwrite) && mapping) { 298589b15332SJohannes Weiner struct file *fpin; 298689b15332SJohannes Weiner 298789b15332SJohannes Weiner fpin = maybe_unlock_mmap_for_io(vmf, NULL); 298889b15332SJohannes Weiner balance_dirty_pages_ratelimited(mapping); 298989b15332SJohannes Weiner if (fpin) { 299089b15332SJohannes Weiner fput(fpin); 2991d9272525SPeter Xu return VM_FAULT_COMPLETED; 299289b15332SJohannes Weiner } 299389b15332SJohannes Weiner } 299489b15332SJohannes Weiner 299589b15332SJohannes Weiner return 0; 29964e047f89SShachar Raindel } 29974e047f89SShachar Raindel 299897ba0c2bSJan Kara /* 29994e047f89SShachar Raindel * Handle write page faults for pages that can be reused in the current vma 30004e047f89SShachar Raindel * 30014e047f89SShachar Raindel * This can happen either due to the mapping being with the VM_SHARED flag, 30024e047f89SShachar Raindel * or due to us being the last reference standing to the page. In either 30034e047f89SShachar Raindel * case, all we need to do here is to mark the page as writable and update 30044e047f89SShachar Raindel * any related book-keeping. 30054e047f89SShachar Raindel */ 3006997dd98dSJan Kara static inline void wp_page_reuse(struct vm_fault *vmf) 300782b0f8c3SJan Kara __releases(vmf->ptl) 30084e047f89SShachar Raindel { 300982b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3010a41b70d6SJan Kara struct page *page = vmf->page; 30114e047f89SShachar Raindel pte_t entry; 30126c287605SDavid Hildenbrand 3013c89357e2SDavid Hildenbrand VM_BUG_ON(!(vmf->flags & FAULT_FLAG_WRITE)); 3014cdb281e6SQi Zheng VM_BUG_ON(page && PageAnon(page) && !PageAnonExclusive(page)); 30156c287605SDavid Hildenbrand 30164e047f89SShachar Raindel /* 30174e047f89SShachar Raindel * Clear the pages cpupid information as the existing 30184e047f89SShachar Raindel * information potentially belongs to a now completely 30194e047f89SShachar Raindel * unrelated process. 30204e047f89SShachar Raindel */ 30214e047f89SShachar Raindel if (page) 30224e047f89SShachar Raindel page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); 30234e047f89SShachar Raindel 30242994302bSJan Kara flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); 30252994302bSJan Kara entry = pte_mkyoung(vmf->orig_pte); 30264e047f89SShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 302782b0f8c3SJan Kara if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) 302882b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 302982b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3030798a6b87SPeter Xu count_vm_event(PGREUSE); 30314e047f89SShachar Raindel } 30324e047f89SShachar Raindel 30334e047f89SShachar Raindel /* 3034c89357e2SDavid Hildenbrand * Handle the case of a page which we actually need to copy to a new page, 3035c89357e2SDavid Hildenbrand * either due to COW or unsharing. 30362f38ab2cSShachar Raindel * 3037c1e8d7c6SMichel Lespinasse * Called with mmap_lock locked and the old page referenced, but 30382f38ab2cSShachar Raindel * without the ptl held. 30392f38ab2cSShachar Raindel * 30402f38ab2cSShachar Raindel * High level logic flow: 30412f38ab2cSShachar Raindel * 30422f38ab2cSShachar Raindel * - Allocate a page, copy the content of the old page to the new one. 30432f38ab2cSShachar Raindel * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. 30442f38ab2cSShachar Raindel * - Take the PTL. If the pte changed, bail out and release the allocated page 30452f38ab2cSShachar Raindel * - If the pte is still the way we remember it, update the page table and all 30462f38ab2cSShachar Raindel * relevant references. This includes dropping the reference the page-table 30472f38ab2cSShachar Raindel * held to the old page, as well as updating the rmap. 30482f38ab2cSShachar Raindel * - In any case, unlock the PTL and drop the reference we took to the old page. 30492f38ab2cSShachar Raindel */ 30502b740303SSouptick Joarder static vm_fault_t wp_page_copy(struct vm_fault *vmf) 30512f38ab2cSShachar Raindel { 3052c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 305382b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3054bae473a4SKirill A. Shutemov struct mm_struct *mm = vma->vm_mm; 305528d41a48SMatthew Wilcox (Oracle) struct folio *old_folio = NULL; 305628d41a48SMatthew Wilcox (Oracle) struct folio *new_folio = NULL; 30572f38ab2cSShachar Raindel pte_t entry; 30582f38ab2cSShachar Raindel int page_copied = 0; 3059ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 3060a873dfe1STony Luck int ret; 30612f38ab2cSShachar Raindel 3062662ce1dcSYang Yang delayacct_wpcopy_start(); 3063662ce1dcSYang Yang 306428d41a48SMatthew Wilcox (Oracle) if (vmf->page) 306528d41a48SMatthew Wilcox (Oracle) old_folio = page_folio(vmf->page); 30662f38ab2cSShachar Raindel if (unlikely(anon_vma_prepare(vma))) 30672f38ab2cSShachar Raindel goto oom; 30682f38ab2cSShachar Raindel 30692994302bSJan Kara if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { 30706bc56a4dSMatthew Wilcox (Oracle) new_folio = vma_alloc_zeroed_movable_folio(vma, vmf->address); 30716bc56a4dSMatthew Wilcox (Oracle) if (!new_folio) 30722f38ab2cSShachar Raindel goto oom; 30732f38ab2cSShachar Raindel } else { 307428d41a48SMatthew Wilcox (Oracle) new_folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma, 307528d41a48SMatthew Wilcox (Oracle) vmf->address, false); 307628d41a48SMatthew Wilcox (Oracle) if (!new_folio) 30772f38ab2cSShachar Raindel goto oom; 307883d116c5SJia He 307928d41a48SMatthew Wilcox (Oracle) ret = __wp_page_copy_user(&new_folio->page, vmf->page, vmf); 3080a873dfe1STony Luck if (ret) { 308183d116c5SJia He /* 308283d116c5SJia He * COW failed, if the fault was solved by other, 308383d116c5SJia He * it's fine. If not, userspace would re-fault on 308483d116c5SJia He * the same address and we will handle the fault 308583d116c5SJia He * from the second attempt. 3086a873dfe1STony Luck * The -EHWPOISON case will not be retried. 308783d116c5SJia He */ 308828d41a48SMatthew Wilcox (Oracle) folio_put(new_folio); 308928d41a48SMatthew Wilcox (Oracle) if (old_folio) 309028d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 3091662ce1dcSYang Yang 3092662ce1dcSYang Yang delayacct_wpcopy_end(); 3093a873dfe1STony Luck return ret == -EHWPOISON ? VM_FAULT_HWPOISON : 0; 309483d116c5SJia He } 309528d41a48SMatthew Wilcox (Oracle) kmsan_copy_page_meta(&new_folio->page, vmf->page); 30962f38ab2cSShachar Raindel } 30972f38ab2cSShachar Raindel 309828d41a48SMatthew Wilcox (Oracle) if (mem_cgroup_charge(new_folio, mm, GFP_KERNEL)) 30992f38ab2cSShachar Raindel goto oom_free_new; 31004d4f75bfSKefeng Wang folio_throttle_swaprate(new_folio, GFP_KERNEL); 31012f38ab2cSShachar Raindel 310228d41a48SMatthew Wilcox (Oracle) __folio_mark_uptodate(new_folio); 3103eb3c24f3SMel Gorman 31047d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, mm, 31056f4f13e8SJérôme Glisse vmf->address & PAGE_MASK, 3106ac46d4f3SJérôme Glisse (vmf->address & PAGE_MASK) + PAGE_SIZE); 3107ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 31082f38ab2cSShachar Raindel 31092f38ab2cSShachar Raindel /* 31102f38ab2cSShachar Raindel * Re-check the pte - we dropped the lock 31112f38ab2cSShachar Raindel */ 311282b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); 31132994302bSJan Kara if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { 311428d41a48SMatthew Wilcox (Oracle) if (old_folio) { 311528d41a48SMatthew Wilcox (Oracle) if (!folio_test_anon(old_folio)) { 311628d41a48SMatthew Wilcox (Oracle) dec_mm_counter(mm, mm_counter_file(&old_folio->page)); 3117f1a79412SShakeel Butt inc_mm_counter(mm, MM_ANONPAGES); 31182f38ab2cSShachar Raindel } 31192f38ab2cSShachar Raindel } else { 3120f1a79412SShakeel Butt inc_mm_counter(mm, MM_ANONPAGES); 31212f38ab2cSShachar Raindel } 31222994302bSJan Kara flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); 312328d41a48SMatthew Wilcox (Oracle) entry = mk_pte(&new_folio->page, vma->vm_page_prot); 312450c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 3125c89357e2SDavid Hildenbrand if (unlikely(unshare)) { 3126c89357e2SDavid Hildenbrand if (pte_soft_dirty(vmf->orig_pte)) 3127c89357e2SDavid Hildenbrand entry = pte_mksoft_dirty(entry); 3128c89357e2SDavid Hildenbrand if (pte_uffd_wp(vmf->orig_pte)) 3129c89357e2SDavid Hildenbrand entry = pte_mkuffd_wp(entry); 3130c89357e2SDavid Hildenbrand } else { 31312f38ab2cSShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 3132c89357e2SDavid Hildenbrand } 3133111fe718SNicholas Piggin 31342f38ab2cSShachar Raindel /* 31352f38ab2cSShachar Raindel * Clear the pte entry and flush it first, before updating the 3136111fe718SNicholas Piggin * pte with the new entry, to keep TLBs on different CPUs in 3137111fe718SNicholas Piggin * sync. This code used to set the new PTE then flush TLBs, but 3138111fe718SNicholas Piggin * that left a window where the new PTE could be loaded into 3139111fe718SNicholas Piggin * some TLBs while the old PTE remains in others. 31402f38ab2cSShachar Raindel */ 314182b0f8c3SJan Kara ptep_clear_flush_notify(vma, vmf->address, vmf->pte); 314228d41a48SMatthew Wilcox (Oracle) folio_add_new_anon_rmap(new_folio, vma, vmf->address); 314328d41a48SMatthew Wilcox (Oracle) folio_add_lru_vma(new_folio, vma); 31442f38ab2cSShachar Raindel /* 31452f38ab2cSShachar Raindel * We call the notify macro here because, when using secondary 31462f38ab2cSShachar Raindel * mmu page tables (such as kvm shadow page tables), we want the 31472f38ab2cSShachar Raindel * new page to be mapped directly into the secondary page table. 31482f38ab2cSShachar Raindel */ 3149c89357e2SDavid Hildenbrand BUG_ON(unshare && pte_write(entry)); 315082b0f8c3SJan Kara set_pte_at_notify(mm, vmf->address, vmf->pte, entry); 315182b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 315228d41a48SMatthew Wilcox (Oracle) if (old_folio) { 31532f38ab2cSShachar Raindel /* 31542f38ab2cSShachar Raindel * Only after switching the pte to the new page may 31552f38ab2cSShachar Raindel * we remove the mapcount here. Otherwise another 31562f38ab2cSShachar Raindel * process may come and find the rmap count decremented 31572f38ab2cSShachar Raindel * before the pte is switched to the new page, and 31582f38ab2cSShachar Raindel * "reuse" the old page writing into it while our pte 31592f38ab2cSShachar Raindel * here still points into it and can be read by other 31602f38ab2cSShachar Raindel * threads. 31612f38ab2cSShachar Raindel * 31622f38ab2cSShachar Raindel * The critical issue is to order this 31632f38ab2cSShachar Raindel * page_remove_rmap with the ptp_clear_flush above. 31642f38ab2cSShachar Raindel * Those stores are ordered by (if nothing else,) 31652f38ab2cSShachar Raindel * the barrier present in the atomic_add_negative 31662f38ab2cSShachar Raindel * in page_remove_rmap. 31672f38ab2cSShachar Raindel * 31682f38ab2cSShachar Raindel * Then the TLB flush in ptep_clear_flush ensures that 31692f38ab2cSShachar Raindel * no process can access the old page before the 31702f38ab2cSShachar Raindel * decremented mapcount is visible. And the old page 31712f38ab2cSShachar Raindel * cannot be reused until after the decremented 31722f38ab2cSShachar Raindel * mapcount is visible. So transitively, TLBs to 31732f38ab2cSShachar Raindel * old page will be flushed before it can be reused. 31742f38ab2cSShachar Raindel */ 317528d41a48SMatthew Wilcox (Oracle) page_remove_rmap(vmf->page, vma, false); 31762f38ab2cSShachar Raindel } 31772f38ab2cSShachar Raindel 31782f38ab2cSShachar Raindel /* Free the old page.. */ 317928d41a48SMatthew Wilcox (Oracle) new_folio = old_folio; 31802f38ab2cSShachar Raindel page_copied = 1; 31812f38ab2cSShachar Raindel } else { 31827df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 31832f38ab2cSShachar Raindel } 31842f38ab2cSShachar Raindel 318528d41a48SMatthew Wilcox (Oracle) if (new_folio) 318628d41a48SMatthew Wilcox (Oracle) folio_put(new_folio); 31872f38ab2cSShachar Raindel 318882b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 31894645b9feSJérôme Glisse /* 31904645b9feSJérôme Glisse * No need to double call mmu_notifier->invalidate_range() callback as 31914645b9feSJérôme Glisse * the above ptep_clear_flush_notify() did already call it. 31924645b9feSJérôme Glisse */ 3193ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_only_end(&range); 319428d41a48SMatthew Wilcox (Oracle) if (old_folio) { 3195f4c4a3f4SHuang Ying if (page_copied) 319628d41a48SMatthew Wilcox (Oracle) free_swap_cache(&old_folio->page); 319728d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 31982f38ab2cSShachar Raindel } 3199662ce1dcSYang Yang 3200662ce1dcSYang Yang delayacct_wpcopy_end(); 3201cb8d8633SDavid Hildenbrand return 0; 32022f38ab2cSShachar Raindel oom_free_new: 320328d41a48SMatthew Wilcox (Oracle) folio_put(new_folio); 32042f38ab2cSShachar Raindel oom: 320528d41a48SMatthew Wilcox (Oracle) if (old_folio) 320628d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 3207662ce1dcSYang Yang 3208662ce1dcSYang Yang delayacct_wpcopy_end(); 32092f38ab2cSShachar Raindel return VM_FAULT_OOM; 32102f38ab2cSShachar Raindel } 32112f38ab2cSShachar Raindel 321266a6197cSJan Kara /** 321366a6197cSJan Kara * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE 321466a6197cSJan Kara * writeable once the page is prepared 321566a6197cSJan Kara * 321666a6197cSJan Kara * @vmf: structure describing the fault 321766a6197cSJan Kara * 321866a6197cSJan Kara * This function handles all that is needed to finish a write page fault in a 321966a6197cSJan Kara * shared mapping due to PTE being read-only once the mapped page is prepared. 3220a862f68aSMike Rapoport * It handles locking of PTE and modifying it. 322166a6197cSJan Kara * 322266a6197cSJan Kara * The function expects the page to be locked or other protection against 322366a6197cSJan Kara * concurrent faults / writeback (such as DAX radix tree locks). 3224a862f68aSMike Rapoport * 32252797e79fSLiu Xiang * Return: %0 on success, %VM_FAULT_NOPAGE when PTE got changed before 3226a862f68aSMike Rapoport * we acquired PTE lock. 322766a6197cSJan Kara */ 32282b740303SSouptick Joarder vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf) 322966a6197cSJan Kara { 323066a6197cSJan Kara WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); 323166a6197cSJan Kara vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, 323266a6197cSJan Kara &vmf->ptl); 323366a6197cSJan Kara /* 323466a6197cSJan Kara * We might have raced with another page fault while we released the 323566a6197cSJan Kara * pte_offset_map_lock. 323666a6197cSJan Kara */ 323766a6197cSJan Kara if (!pte_same(*vmf->pte, vmf->orig_pte)) { 32387df67697SBibo Mao update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 323966a6197cSJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3240a19e2553SJan Kara return VM_FAULT_NOPAGE; 324166a6197cSJan Kara } 324266a6197cSJan Kara wp_page_reuse(vmf); 3243a19e2553SJan Kara return 0; 324466a6197cSJan Kara } 324566a6197cSJan Kara 3246dd906184SBoaz Harrosh /* 3247dd906184SBoaz Harrosh * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED 3248dd906184SBoaz Harrosh * mapping 3249dd906184SBoaz Harrosh */ 32502b740303SSouptick Joarder static vm_fault_t wp_pfn_shared(struct vm_fault *vmf) 3251dd906184SBoaz Harrosh { 325282b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3253bae473a4SKirill A. Shutemov 3254dd906184SBoaz Harrosh if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { 32552b740303SSouptick Joarder vm_fault_t ret; 3256dd906184SBoaz Harrosh 325782b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3258fe82221fSJan Kara vmf->flags |= FAULT_FLAG_MKWRITE; 325911bac800SDave Jiang ret = vma->vm_ops->pfn_mkwrite(vmf); 32602f89dc12SJan Kara if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) 3261dd906184SBoaz Harrosh return ret; 326266a6197cSJan Kara return finish_mkwrite_fault(vmf); 3263dd906184SBoaz Harrosh } 3264997dd98dSJan Kara wp_page_reuse(vmf); 3265cb8d8633SDavid Hildenbrand return 0; 3266dd906184SBoaz Harrosh } 3267dd906184SBoaz Harrosh 32682b740303SSouptick Joarder static vm_fault_t wp_page_shared(struct vm_fault *vmf) 326982b0f8c3SJan Kara __releases(vmf->ptl) 327093e478d4SShachar Raindel { 327182b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3272cb8d8633SDavid Hildenbrand vm_fault_t ret = 0; 327393e478d4SShachar Raindel 3274a41b70d6SJan Kara get_page(vmf->page); 327593e478d4SShachar Raindel 327693e478d4SShachar Raindel if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 32772b740303SSouptick Joarder vm_fault_t tmp; 327893e478d4SShachar Raindel 327982b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 328038b8cb7fSJan Kara tmp = do_page_mkwrite(vmf); 328193e478d4SShachar Raindel if (unlikely(!tmp || (tmp & 328293e478d4SShachar Raindel (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 3283a41b70d6SJan Kara put_page(vmf->page); 328493e478d4SShachar Raindel return tmp; 328593e478d4SShachar Raindel } 328666a6197cSJan Kara tmp = finish_mkwrite_fault(vmf); 3287a19e2553SJan Kara if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { 3288a41b70d6SJan Kara unlock_page(vmf->page); 3289a41b70d6SJan Kara put_page(vmf->page); 329066a6197cSJan Kara return tmp; 329193e478d4SShachar Raindel } 329266a6197cSJan Kara } else { 3293997dd98dSJan Kara wp_page_reuse(vmf); 3294997dd98dSJan Kara lock_page(vmf->page); 329593e478d4SShachar Raindel } 329689b15332SJohannes Weiner ret |= fault_dirty_shared_page(vmf); 3297997dd98dSJan Kara put_page(vmf->page); 329893e478d4SShachar Raindel 329989b15332SJohannes Weiner return ret; 330093e478d4SShachar Raindel } 330193e478d4SShachar Raindel 33022f38ab2cSShachar Raindel /* 3303c89357e2SDavid Hildenbrand * This routine handles present pages, when 3304c89357e2SDavid Hildenbrand * * users try to write to a shared page (FAULT_FLAG_WRITE) 3305c89357e2SDavid Hildenbrand * * GUP wants to take a R/O pin on a possibly shared anonymous page 3306c89357e2SDavid Hildenbrand * (FAULT_FLAG_UNSHARE) 3307c89357e2SDavid Hildenbrand * 3308c89357e2SDavid Hildenbrand * It is done by copying the page to a new address and decrementing the 3309c89357e2SDavid Hildenbrand * shared-page counter for the old page. 33101da177e4SLinus Torvalds * 33111da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 33121da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 3313c89357e2SDavid Hildenbrand * Thus, with FAULT_FLAG_WRITE, we can safely just mark it writable once we've 3314c89357e2SDavid Hildenbrand * done any necessary COW. 33151da177e4SLinus Torvalds * 3316c89357e2SDavid Hildenbrand * In case of FAULT_FLAG_WRITE, we also mark the page dirty at this point even 3317c89357e2SDavid Hildenbrand * though the page will change only once the write actually happens. This 3318c89357e2SDavid Hildenbrand * avoids a few races, and potentially makes it more efficient. 33191da177e4SLinus Torvalds * 3320c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 33218f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 3322c1e8d7c6SMichel Lespinasse * We return with mmap_lock still held, but pte unmapped and unlocked. 33231da177e4SLinus Torvalds */ 33242b740303SSouptick Joarder static vm_fault_t do_wp_page(struct vm_fault *vmf) 332582b0f8c3SJan Kara __releases(vmf->ptl) 33261da177e4SLinus Torvalds { 3327c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 332882b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3329b9086fdeSDavid Hildenbrand struct folio *folio = NULL; 33301da177e4SLinus Torvalds 3331c89357e2SDavid Hildenbrand if (likely(!unshare)) { 3332292924b2SPeter Xu if (userfaultfd_pte_wp(vma, *vmf->pte)) { 3333529b930bSAndrea Arcangeli pte_unmap_unlock(vmf->pte, vmf->ptl); 3334529b930bSAndrea Arcangeli return handle_userfault(vmf, VM_UFFD_WP); 3335529b930bSAndrea Arcangeli } 3336529b930bSAndrea Arcangeli 33376ce64428SNadav Amit /* 33386ce64428SNadav Amit * Userfaultfd write-protect can defer flushes. Ensure the TLB 33396ce64428SNadav Amit * is flushed in this case before copying. 33406ce64428SNadav Amit */ 33416ce64428SNadav Amit if (unlikely(userfaultfd_wp(vmf->vma) && 33426ce64428SNadav Amit mm_tlb_flush_pending(vmf->vma->vm_mm))) 33436ce64428SNadav Amit flush_tlb_page(vmf->vma, vmf->address); 3344c89357e2SDavid Hildenbrand } 33456ce64428SNadav Amit 3346a41b70d6SJan Kara vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); 3347c89357e2SDavid Hildenbrand 3348251b97f5SPeter Zijlstra /* 3349b9086fdeSDavid Hildenbrand * Shared mapping: we are guaranteed to have VM_WRITE and 3350b9086fdeSDavid Hildenbrand * FAULT_FLAG_WRITE set at this point. 3351b9086fdeSDavid Hildenbrand */ 3352b9086fdeSDavid Hildenbrand if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 3353b9086fdeSDavid Hildenbrand /* 335464e45507SPeter Feiner * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a 335564e45507SPeter Feiner * VM_PFNMAP VMA. 3356251b97f5SPeter Zijlstra * 3357251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 3358dd906184SBoaz Harrosh * Just mark the pages writable and/or call ops->pfn_mkwrite. 3359251b97f5SPeter Zijlstra */ 3360b9086fdeSDavid Hildenbrand if (!vmf->page) 33612994302bSJan Kara return wp_pfn_shared(vmf); 3362b9086fdeSDavid Hildenbrand return wp_page_shared(vmf); 3363251b97f5SPeter Zijlstra } 33641da177e4SLinus Torvalds 3365b9086fdeSDavid Hildenbrand if (vmf->page) 3366e4a2ed94SMatthew Wilcox (Oracle) folio = page_folio(vmf->page); 3367b9086fdeSDavid Hildenbrand 3368b9086fdeSDavid Hildenbrand /* 3369b9086fdeSDavid Hildenbrand * Private mapping: create an exclusive anonymous page copy if reuse 3370b9086fdeSDavid Hildenbrand * is impossible. We might miss VM_WRITE for FOLL_FORCE handling. 3371b9086fdeSDavid Hildenbrand */ 3372b9086fdeSDavid Hildenbrand if (folio && folio_test_anon(folio)) { 337353a05ad9SDavid Hildenbrand /* 33746c287605SDavid Hildenbrand * If the page is exclusive to this process we must reuse the 33756c287605SDavid Hildenbrand * page without further checks. 33766c287605SDavid Hildenbrand */ 3377e4a2ed94SMatthew Wilcox (Oracle) if (PageAnonExclusive(vmf->page)) 33786c287605SDavid Hildenbrand goto reuse; 33796c287605SDavid Hildenbrand 33806c287605SDavid Hildenbrand /* 3381e4a2ed94SMatthew Wilcox (Oracle) * We have to verify under folio lock: these early checks are 3382e4a2ed94SMatthew Wilcox (Oracle) * just an optimization to avoid locking the folio and freeing 338353a05ad9SDavid Hildenbrand * the swapcache if there is little hope that we can reuse. 338453a05ad9SDavid Hildenbrand * 3385e4a2ed94SMatthew Wilcox (Oracle) * KSM doesn't necessarily raise the folio refcount. 338653a05ad9SDavid Hildenbrand */ 3387e4a2ed94SMatthew Wilcox (Oracle) if (folio_test_ksm(folio) || folio_ref_count(folio) > 3) 3388d4c47097SDavid Hildenbrand goto copy; 3389e4a2ed94SMatthew Wilcox (Oracle) if (!folio_test_lru(folio)) 3390d4c47097SDavid Hildenbrand /* 3391d4c47097SDavid Hildenbrand * Note: We cannot easily detect+handle references from 3392e4a2ed94SMatthew Wilcox (Oracle) * remote LRU pagevecs or references to LRU folios. 3393d4c47097SDavid Hildenbrand */ 3394d4c47097SDavid Hildenbrand lru_add_drain(); 3395e4a2ed94SMatthew Wilcox (Oracle) if (folio_ref_count(folio) > 1 + folio_test_swapcache(folio)) 339652d1e606SKirill Tkhai goto copy; 3397e4a2ed94SMatthew Wilcox (Oracle) if (!folio_trylock(folio)) 339852d1e606SKirill Tkhai goto copy; 3399e4a2ed94SMatthew Wilcox (Oracle) if (folio_test_swapcache(folio)) 3400e4a2ed94SMatthew Wilcox (Oracle) folio_free_swap(folio); 3401e4a2ed94SMatthew Wilcox (Oracle) if (folio_test_ksm(folio) || folio_ref_count(folio) != 1) { 3402e4a2ed94SMatthew Wilcox (Oracle) folio_unlock(folio); 340309854ba9SLinus Torvalds goto copy; 340452d1e606SKirill Tkhai } 3405c44b6743SRik van Riel /* 3406e4a2ed94SMatthew Wilcox (Oracle) * Ok, we've got the only folio reference from our mapping 3407e4a2ed94SMatthew Wilcox (Oracle) * and the folio is locked, it's dark out, and we're wearing 340853a05ad9SDavid Hildenbrand * sunglasses. Hit it. 3409c44b6743SRik van Riel */ 3410e4a2ed94SMatthew Wilcox (Oracle) page_move_anon_rmap(vmf->page, vma); 3411e4a2ed94SMatthew Wilcox (Oracle) folio_unlock(folio); 34126c287605SDavid Hildenbrand reuse: 3413c89357e2SDavid Hildenbrand if (unlikely(unshare)) { 3414c89357e2SDavid Hildenbrand pte_unmap_unlock(vmf->pte, vmf->ptl); 3415c89357e2SDavid Hildenbrand return 0; 3416c89357e2SDavid Hildenbrand } 3417be068f29SLinus Torvalds wp_page_reuse(vmf); 3418cb8d8633SDavid Hildenbrand return 0; 34191da177e4SLinus Torvalds } 342052d1e606SKirill Tkhai copy: 34211da177e4SLinus Torvalds /* 34221da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 34231da177e4SLinus Torvalds */ 3424b9086fdeSDavid Hildenbrand if (folio) 3425b9086fdeSDavid Hildenbrand folio_get(folio); 342628766805SShachar Raindel 342782b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 342894bfe85bSYang Yang #ifdef CONFIG_KSM 3429b9086fdeSDavid Hildenbrand if (folio && folio_test_ksm(folio)) 343094bfe85bSYang Yang count_vm_event(COW_KSM); 343194bfe85bSYang Yang #endif 3432a41b70d6SJan Kara return wp_page_copy(vmf); 34331da177e4SLinus Torvalds } 34341da177e4SLinus Torvalds 343597a89413SPeter Zijlstra static void unmap_mapping_range_vma(struct vm_area_struct *vma, 34361da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 34371da177e4SLinus Torvalds struct zap_details *details) 34381da177e4SLinus Torvalds { 3439f5cc4eefSAl Viro zap_page_range_single(vma, start_addr, end_addr - start_addr, details); 34401da177e4SLinus Torvalds } 34411da177e4SLinus Torvalds 3442f808c13fSDavidlohr Bueso static inline void unmap_mapping_range_tree(struct rb_root_cached *root, 3443232a6a1cSPeter Xu pgoff_t first_index, 3444232a6a1cSPeter Xu pgoff_t last_index, 34451da177e4SLinus Torvalds struct zap_details *details) 34461da177e4SLinus Torvalds { 34471da177e4SLinus Torvalds struct vm_area_struct *vma; 34481da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 34491da177e4SLinus Torvalds 3450232a6a1cSPeter Xu vma_interval_tree_foreach(vma, root, first_index, last_index) { 34511da177e4SLinus Torvalds vba = vma->vm_pgoff; 3452d6e93217SLibin vea = vba + vma_pages(vma) - 1; 3453f9871da9SMiaohe Lin zba = max(first_index, vba); 3454f9871da9SMiaohe Lin zea = min(last_index, vea); 34551da177e4SLinus Torvalds 345697a89413SPeter Zijlstra unmap_mapping_range_vma(vma, 34571da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 34581da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 345997a89413SPeter Zijlstra details); 34601da177e4SLinus Torvalds } 34611da177e4SLinus Torvalds } 34621da177e4SLinus Torvalds 34631da177e4SLinus Torvalds /** 34643506659eSMatthew Wilcox (Oracle) * unmap_mapping_folio() - Unmap single folio from processes. 34653506659eSMatthew Wilcox (Oracle) * @folio: The locked folio to be unmapped. 346622061a1fSHugh Dickins * 34673506659eSMatthew Wilcox (Oracle) * Unmap this folio from any userspace process which still has it mmaped. 346822061a1fSHugh Dickins * Typically, for efficiency, the range of nearby pages has already been 346922061a1fSHugh Dickins * unmapped by unmap_mapping_pages() or unmap_mapping_range(). But once 34703506659eSMatthew Wilcox (Oracle) * truncation or invalidation holds the lock on a folio, it may find that 34713506659eSMatthew Wilcox (Oracle) * the page has been remapped again: and then uses unmap_mapping_folio() 347222061a1fSHugh Dickins * to unmap it finally. 347322061a1fSHugh Dickins */ 34743506659eSMatthew Wilcox (Oracle) void unmap_mapping_folio(struct folio *folio) 347522061a1fSHugh Dickins { 34763506659eSMatthew Wilcox (Oracle) struct address_space *mapping = folio->mapping; 347722061a1fSHugh Dickins struct zap_details details = { }; 3478232a6a1cSPeter Xu pgoff_t first_index; 3479232a6a1cSPeter Xu pgoff_t last_index; 348022061a1fSHugh Dickins 34813506659eSMatthew Wilcox (Oracle) VM_BUG_ON(!folio_test_locked(folio)); 348222061a1fSHugh Dickins 34833506659eSMatthew Wilcox (Oracle) first_index = folio->index; 34843506659eSMatthew Wilcox (Oracle) last_index = folio->index + folio_nr_pages(folio) - 1; 3485232a6a1cSPeter Xu 34862e148f1eSPeter Xu details.even_cows = false; 34873506659eSMatthew Wilcox (Oracle) details.single_folio = folio; 3488999dad82SPeter Xu details.zap_flags = ZAP_FLAG_DROP_MARKER; 348922061a1fSHugh Dickins 34902c865995SHugh Dickins i_mmap_lock_read(mapping); 349122061a1fSHugh Dickins if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) 3492232a6a1cSPeter Xu unmap_mapping_range_tree(&mapping->i_mmap, first_index, 3493232a6a1cSPeter Xu last_index, &details); 34942c865995SHugh Dickins i_mmap_unlock_read(mapping); 349522061a1fSHugh Dickins } 349622061a1fSHugh Dickins 349722061a1fSHugh Dickins /** 3498977fbdcdSMatthew Wilcox * unmap_mapping_pages() - Unmap pages from processes. 3499977fbdcdSMatthew Wilcox * @mapping: The address space containing pages to be unmapped. 3500977fbdcdSMatthew Wilcox * @start: Index of first page to be unmapped. 3501977fbdcdSMatthew Wilcox * @nr: Number of pages to be unmapped. 0 to unmap to end of file. 3502977fbdcdSMatthew Wilcox * @even_cows: Whether to unmap even private COWed pages. 3503977fbdcdSMatthew Wilcox * 3504977fbdcdSMatthew Wilcox * Unmap the pages in this address space from any userspace process which 3505977fbdcdSMatthew Wilcox * has them mmaped. Generally, you want to remove COWed pages as well when 3506977fbdcdSMatthew Wilcox * a file is being truncated, but not when invalidating pages from the page 3507977fbdcdSMatthew Wilcox * cache. 3508977fbdcdSMatthew Wilcox */ 3509977fbdcdSMatthew Wilcox void unmap_mapping_pages(struct address_space *mapping, pgoff_t start, 3510977fbdcdSMatthew Wilcox pgoff_t nr, bool even_cows) 3511977fbdcdSMatthew Wilcox { 3512977fbdcdSMatthew Wilcox struct zap_details details = { }; 3513232a6a1cSPeter Xu pgoff_t first_index = start; 3514232a6a1cSPeter Xu pgoff_t last_index = start + nr - 1; 3515977fbdcdSMatthew Wilcox 35162e148f1eSPeter Xu details.even_cows = even_cows; 3517232a6a1cSPeter Xu if (last_index < first_index) 3518232a6a1cSPeter Xu last_index = ULONG_MAX; 3519977fbdcdSMatthew Wilcox 35202c865995SHugh Dickins i_mmap_lock_read(mapping); 3521977fbdcdSMatthew Wilcox if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) 3522232a6a1cSPeter Xu unmap_mapping_range_tree(&mapping->i_mmap, first_index, 3523232a6a1cSPeter Xu last_index, &details); 35242c865995SHugh Dickins i_mmap_unlock_read(mapping); 3525977fbdcdSMatthew Wilcox } 35266e0e99d5SDavid Howells EXPORT_SYMBOL_GPL(unmap_mapping_pages); 3527977fbdcdSMatthew Wilcox 3528977fbdcdSMatthew Wilcox /** 35298a5f14a2SKirill A. Shutemov * unmap_mapping_range - unmap the portion of all mmaps in the specified 3530977fbdcdSMatthew Wilcox * address_space corresponding to the specified byte range in the underlying 35318a5f14a2SKirill A. Shutemov * file. 35328a5f14a2SKirill A. Shutemov * 35333d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 35341da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 35351da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 353625d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 35371da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 35381da177e4SLinus Torvalds * partial pages. 35391da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 35401da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 35411da177e4SLinus Torvalds * end of the file. 35421da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 35431da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 35441da177e4SLinus Torvalds */ 35451da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 35461da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 35471da177e4SLinus Torvalds { 35481da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 35491da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 35501da177e4SLinus Torvalds 35511da177e4SLinus Torvalds /* Check for overflow. */ 35521da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 35531da177e4SLinus Torvalds long long holeend = 35541da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 35551da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 35561da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 35571da177e4SLinus Torvalds } 35581da177e4SLinus Torvalds 3559977fbdcdSMatthew Wilcox unmap_mapping_pages(mapping, hba, hlen, even_cows); 35601da177e4SLinus Torvalds } 35611da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 35621da177e4SLinus Torvalds 35631da177e4SLinus Torvalds /* 3564b756a3b5SAlistair Popple * Restore a potential device exclusive pte to a working pte entry 3565b756a3b5SAlistair Popple */ 3566b756a3b5SAlistair Popple static vm_fault_t remove_device_exclusive_entry(struct vm_fault *vmf) 3567b756a3b5SAlistair Popple { 356819672a9eSMatthew Wilcox (Oracle) struct folio *folio = page_folio(vmf->page); 3569b756a3b5SAlistair Popple struct vm_area_struct *vma = vmf->vma; 3570b756a3b5SAlistair Popple struct mmu_notifier_range range; 3571b756a3b5SAlistair Popple 35727c7b9629SAlistair Popple /* 35737c7b9629SAlistair Popple * We need a reference to lock the folio because we don't hold 35747c7b9629SAlistair Popple * the PTL so a racing thread can remove the device-exclusive 35757c7b9629SAlistair Popple * entry and unmap it. If the folio is free the entry must 35767c7b9629SAlistair Popple * have been removed already. If it happens to have already 35777c7b9629SAlistair Popple * been re-allocated after being freed all we do is lock and 35787c7b9629SAlistair Popple * unlock it. 35797c7b9629SAlistair Popple */ 35807c7b9629SAlistair Popple if (!folio_try_get(folio)) 35817c7b9629SAlistair Popple return 0; 35827c7b9629SAlistair Popple 35837c7b9629SAlistair Popple if (!folio_lock_or_retry(folio, vma->vm_mm, vmf->flags)) { 35847c7b9629SAlistair Popple folio_put(folio); 3585b756a3b5SAlistair Popple return VM_FAULT_RETRY; 35867c7b9629SAlistair Popple } 35877d4a8be0SAlistair Popple mmu_notifier_range_init_owner(&range, MMU_NOTIFY_EXCLUSIVE, 0, 3588b756a3b5SAlistair Popple vma->vm_mm, vmf->address & PAGE_MASK, 3589b756a3b5SAlistair Popple (vmf->address & PAGE_MASK) + PAGE_SIZE, NULL); 3590b756a3b5SAlistair Popple mmu_notifier_invalidate_range_start(&range); 3591b756a3b5SAlistair Popple 3592b756a3b5SAlistair Popple vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 3593b756a3b5SAlistair Popple &vmf->ptl); 3594b756a3b5SAlistair Popple if (likely(pte_same(*vmf->pte, vmf->orig_pte))) 359519672a9eSMatthew Wilcox (Oracle) restore_exclusive_pte(vma, vmf->page, vmf->address, vmf->pte); 3596b756a3b5SAlistair Popple 3597b756a3b5SAlistair Popple pte_unmap_unlock(vmf->pte, vmf->ptl); 359819672a9eSMatthew Wilcox (Oracle) folio_unlock(folio); 35997c7b9629SAlistair Popple folio_put(folio); 3600b756a3b5SAlistair Popple 3601b756a3b5SAlistair Popple mmu_notifier_invalidate_range_end(&range); 3602b756a3b5SAlistair Popple return 0; 3603b756a3b5SAlistair Popple } 3604b756a3b5SAlistair Popple 3605a160e537SMatthew Wilcox (Oracle) static inline bool should_try_to_free_swap(struct folio *folio, 3606c145e0b4SDavid Hildenbrand struct vm_area_struct *vma, 3607c145e0b4SDavid Hildenbrand unsigned int fault_flags) 3608c145e0b4SDavid Hildenbrand { 3609a160e537SMatthew Wilcox (Oracle) if (!folio_test_swapcache(folio)) 3610c145e0b4SDavid Hildenbrand return false; 36119202d527SMatthew Wilcox (Oracle) if (mem_cgroup_swap_full(folio) || (vma->vm_flags & VM_LOCKED) || 3612a160e537SMatthew Wilcox (Oracle) folio_test_mlocked(folio)) 3613c145e0b4SDavid Hildenbrand return true; 3614c145e0b4SDavid Hildenbrand /* 3615c145e0b4SDavid Hildenbrand * If we want to map a page that's in the swapcache writable, we 3616c145e0b4SDavid Hildenbrand * have to detect via the refcount if we're really the exclusive 3617c145e0b4SDavid Hildenbrand * user. Try freeing the swapcache to get rid of the swapcache 3618c145e0b4SDavid Hildenbrand * reference only in case it's likely that we'll be the exlusive user. 3619c145e0b4SDavid Hildenbrand */ 3620a160e537SMatthew Wilcox (Oracle) return (fault_flags & FAULT_FLAG_WRITE) && !folio_test_ksm(folio) && 3621a160e537SMatthew Wilcox (Oracle) folio_ref_count(folio) == 2; 3622c145e0b4SDavid Hildenbrand } 3623c145e0b4SDavid Hildenbrand 36249c28a205SPeter Xu static vm_fault_t pte_marker_clear(struct vm_fault *vmf) 36259c28a205SPeter Xu { 36269c28a205SPeter Xu vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, 36279c28a205SPeter Xu vmf->address, &vmf->ptl); 36289c28a205SPeter Xu /* 36299c28a205SPeter Xu * Be careful so that we will only recover a special uffd-wp pte into a 36309c28a205SPeter Xu * none pte. Otherwise it means the pte could have changed, so retry. 36317e3ce3f8SPeter Xu * 36327e3ce3f8SPeter Xu * This should also cover the case where e.g. the pte changed 36337e3ce3f8SPeter Xu * quickly from a PTE_MARKER_UFFD_WP into PTE_MARKER_SWAPIN_ERROR. 36347e3ce3f8SPeter Xu * So is_pte_marker() check is not enough to safely drop the pte. 36359c28a205SPeter Xu */ 36367e3ce3f8SPeter Xu if (pte_same(vmf->orig_pte, *vmf->pte)) 36379c28a205SPeter Xu pte_clear(vmf->vma->vm_mm, vmf->address, vmf->pte); 36389c28a205SPeter Xu pte_unmap_unlock(vmf->pte, vmf->ptl); 36399c28a205SPeter Xu return 0; 36409c28a205SPeter Xu } 36419c28a205SPeter Xu 36422bad466cSPeter Xu static vm_fault_t do_pte_missing(struct vm_fault *vmf) 36432bad466cSPeter Xu { 36442bad466cSPeter Xu if (vma_is_anonymous(vmf->vma)) 36452bad466cSPeter Xu return do_anonymous_page(vmf); 36462bad466cSPeter Xu else 36472bad466cSPeter Xu return do_fault(vmf); 36482bad466cSPeter Xu } 36492bad466cSPeter Xu 36509c28a205SPeter Xu /* 36519c28a205SPeter Xu * This is actually a page-missing access, but with uffd-wp special pte 36529c28a205SPeter Xu * installed. It means this pte was wr-protected before being unmapped. 36539c28a205SPeter Xu */ 36549c28a205SPeter Xu static vm_fault_t pte_marker_handle_uffd_wp(struct vm_fault *vmf) 36559c28a205SPeter Xu { 36569c28a205SPeter Xu /* 36579c28a205SPeter Xu * Just in case there're leftover special ptes even after the region 36587a079ba2SPeter Xu * got unregistered - we can simply clear them. 36599c28a205SPeter Xu */ 36602bad466cSPeter Xu if (unlikely(!userfaultfd_wp(vmf->vma))) 36619c28a205SPeter Xu return pte_marker_clear(vmf); 36629c28a205SPeter Xu 36632bad466cSPeter Xu return do_pte_missing(vmf); 36649c28a205SPeter Xu } 36659c28a205SPeter Xu 36665c041f5dSPeter Xu static vm_fault_t handle_pte_marker(struct vm_fault *vmf) 36675c041f5dSPeter Xu { 36685c041f5dSPeter Xu swp_entry_t entry = pte_to_swp_entry(vmf->orig_pte); 36695c041f5dSPeter Xu unsigned long marker = pte_marker_get(entry); 36705c041f5dSPeter Xu 36715c041f5dSPeter Xu /* 3672ca92ea3dSPeter Xu * PTE markers should never be empty. If anything weird happened, 3673ca92ea3dSPeter Xu * the best thing to do is to kill the process along with its mm. 36745c041f5dSPeter Xu */ 3675ca92ea3dSPeter Xu if (WARN_ON_ONCE(!marker)) 36765c041f5dSPeter Xu return VM_FAULT_SIGBUS; 36775c041f5dSPeter Xu 367815520a3fSPeter Xu /* Higher priority than uffd-wp when data corrupted */ 367915520a3fSPeter Xu if (marker & PTE_MARKER_SWAPIN_ERROR) 368015520a3fSPeter Xu return VM_FAULT_SIGBUS; 368115520a3fSPeter Xu 36829c28a205SPeter Xu if (pte_marker_entry_uffd_wp(entry)) 36839c28a205SPeter Xu return pte_marker_handle_uffd_wp(vmf); 36849c28a205SPeter Xu 36859c28a205SPeter Xu /* This is an unknown pte marker */ 36869c28a205SPeter Xu return VM_FAULT_SIGBUS; 36875c041f5dSPeter Xu } 36885c041f5dSPeter Xu 3689b756a3b5SAlistair Popple /* 3690c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 36918f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 36929a95f3cfSPaul Cassella * We return with pte unmapped and unlocked. 36939a95f3cfSPaul Cassella * 3694c1e8d7c6SMichel Lespinasse * We return with the mmap_lock locked or unlocked in the same cases 36959a95f3cfSPaul Cassella * as does filemap_fault(). 36961da177e4SLinus Torvalds */ 36972b740303SSouptick Joarder vm_fault_t do_swap_page(struct vm_fault *vmf) 36981da177e4SLinus Torvalds { 369982b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3700d4f9565aSMatthew Wilcox (Oracle) struct folio *swapcache, *folio = NULL; 3701d4f9565aSMatthew Wilcox (Oracle) struct page *page; 37022799e775SMiaohe Lin struct swap_info_struct *si = NULL; 370314f9135dSDavid Hildenbrand rmap_t rmap_flags = RMAP_NONE; 37041493a191SDavid Hildenbrand bool exclusive = false; 370565500d23SHugh Dickins swp_entry_t entry; 37061da177e4SLinus Torvalds pte_t pte; 3707d065bd81SMichel Lespinasse int locked; 37082b740303SSouptick Joarder vm_fault_t ret = 0; 3709aae466b0SJoonsoo Kim void *shadow = NULL; 37101da177e4SLinus Torvalds 37112ca99358SPeter Xu if (!pte_unmap_same(vmf)) 37128f4e2101SHugh Dickins goto out; 371365500d23SHugh Dickins 371417c05f18SSuren Baghdasaryan if (vmf->flags & FAULT_FLAG_VMA_LOCK) { 371517c05f18SSuren Baghdasaryan ret = VM_FAULT_RETRY; 371617c05f18SSuren Baghdasaryan goto out; 371717c05f18SSuren Baghdasaryan } 371817c05f18SSuren Baghdasaryan 37192994302bSJan Kara entry = pte_to_swp_entry(vmf->orig_pte); 3720d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 37210697212aSChristoph Lameter if (is_migration_entry(entry)) { 372282b0f8c3SJan Kara migration_entry_wait(vma->vm_mm, vmf->pmd, 372382b0f8c3SJan Kara vmf->address); 3724b756a3b5SAlistair Popple } else if (is_device_exclusive_entry(entry)) { 3725b756a3b5SAlistair Popple vmf->page = pfn_swap_entry_to_page(entry); 3726b756a3b5SAlistair Popple ret = remove_device_exclusive_entry(vmf); 37275042db43SJérôme Glisse } else if (is_device_private_entry(entry)) { 3728af5cdaf8SAlistair Popple vmf->page = pfn_swap_entry_to_page(entry); 372916ce101dSAlistair Popple vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 373016ce101dSAlistair Popple vmf->address, &vmf->ptl); 373116ce101dSAlistair Popple if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { 373216ce101dSAlistair Popple spin_unlock(vmf->ptl); 373316ce101dSAlistair Popple goto out; 373416ce101dSAlistair Popple } 373516ce101dSAlistair Popple 373616ce101dSAlistair Popple /* 373716ce101dSAlistair Popple * Get a page reference while we know the page can't be 373816ce101dSAlistair Popple * freed. 373916ce101dSAlistair Popple */ 374016ce101dSAlistair Popple get_page(vmf->page); 374116ce101dSAlistair Popple pte_unmap_unlock(vmf->pte, vmf->ptl); 37424a955bedSAlistair Popple ret = vmf->page->pgmap->ops->migrate_to_ram(vmf); 374316ce101dSAlistair Popple put_page(vmf->page); 3744d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 3745d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 37465c041f5dSPeter Xu } else if (is_pte_marker_entry(entry)) { 37475c041f5dSPeter Xu ret = handle_pte_marker(vmf); 3748d1737fdbSAndi Kleen } else { 37492994302bSJan Kara print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); 3750d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 3751d1737fdbSAndi Kleen } 37520697212aSChristoph Lameter goto out; 37530697212aSChristoph Lameter } 37540bcac06fSMinchan Kim 37552799e775SMiaohe Lin /* Prevent swapoff from happening to us. */ 37562799e775SMiaohe Lin si = get_swap_device(entry); 37572799e775SMiaohe Lin if (unlikely(!si)) 37582799e775SMiaohe Lin goto out; 37590bcac06fSMinchan Kim 37605a423081SMatthew Wilcox (Oracle) folio = swap_cache_get_folio(entry, vma, vmf->address); 37615a423081SMatthew Wilcox (Oracle) if (folio) 37625a423081SMatthew Wilcox (Oracle) page = folio_file_page(folio, swp_offset(entry)); 3763d4f9565aSMatthew Wilcox (Oracle) swapcache = folio; 3764f8020772SMinchan Kim 3765d4f9565aSMatthew Wilcox (Oracle) if (!folio) { 3766a449bf58SQian Cai if (data_race(si->flags & SWP_SYNCHRONOUS_IO) && 3767eb085574SHuang Ying __swap_count(entry) == 1) { 37680bcac06fSMinchan Kim /* skip swapcache */ 376963ad4addSMatthew Wilcox (Oracle) folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, 377063ad4addSMatthew Wilcox (Oracle) vma, vmf->address, false); 377163ad4addSMatthew Wilcox (Oracle) page = &folio->page; 377263ad4addSMatthew Wilcox (Oracle) if (folio) { 377363ad4addSMatthew Wilcox (Oracle) __folio_set_locked(folio); 377463ad4addSMatthew Wilcox (Oracle) __folio_set_swapbacked(folio); 37754c6355b2SJohannes Weiner 377665995918SMatthew Wilcox (Oracle) if (mem_cgroup_swapin_charge_folio(folio, 377763ad4addSMatthew Wilcox (Oracle) vma->vm_mm, GFP_KERNEL, 377863ad4addSMatthew Wilcox (Oracle) entry)) { 3779545b1b07SMichal Hocko ret = VM_FAULT_OOM; 37804c6355b2SJohannes Weiner goto out_page; 3781545b1b07SMichal Hocko } 37820add0c77SShakeel Butt mem_cgroup_swapin_uncharge_swap(entry); 37834c6355b2SJohannes Weiner 3784aae466b0SJoonsoo Kim shadow = get_shadow_from_swap_cache(entry); 3785aae466b0SJoonsoo Kim if (shadow) 378663ad4addSMatthew Wilcox (Oracle) workingset_refault(folio, shadow); 37870076f029SJoonsoo Kim 378863ad4addSMatthew Wilcox (Oracle) folio_add_lru(folio); 37890add0c77SShakeel Butt 37900add0c77SShakeel Butt /* To provide entry to swap_readpage() */ 379163ad4addSMatthew Wilcox (Oracle) folio_set_swap_entry(folio, entry); 37925169b844SNeilBrown swap_readpage(page, true, NULL); 379363ad4addSMatthew Wilcox (Oracle) folio->private = NULL; 37940bcac06fSMinchan Kim } 3795aa8d22a1SMinchan Kim } else { 3796e9e9b7ecSMinchan Kim page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, 3797e9e9b7ecSMinchan Kim vmf); 379863ad4addSMatthew Wilcox (Oracle) if (page) 379963ad4addSMatthew Wilcox (Oracle) folio = page_folio(page); 3800d4f9565aSMatthew Wilcox (Oracle) swapcache = folio; 38010bcac06fSMinchan Kim } 38020bcac06fSMinchan Kim 3803d4f9565aSMatthew Wilcox (Oracle) if (!folio) { 38041da177e4SLinus Torvalds /* 38058f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 38068f4e2101SHugh Dickins * while we released the pte lock. 38071da177e4SLinus Torvalds */ 380882b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 380982b0f8c3SJan Kara vmf->address, &vmf->ptl); 38102994302bSJan Kara if (likely(pte_same(*vmf->pte, vmf->orig_pte))) 38111da177e4SLinus Torvalds ret = VM_FAULT_OOM; 381265500d23SHugh Dickins goto unlock; 38131da177e4SLinus Torvalds } 38141da177e4SLinus Torvalds 38151da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 38161da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 3817f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 38182262185cSRoman Gushchin count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); 3819d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 382071f72525SWu Fengguang /* 382171f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 382271f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 382371f72525SWu Fengguang */ 3824d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 38254779cb31SAndi Kleen goto out_release; 38261da177e4SLinus Torvalds } 38271da177e4SLinus Torvalds 382819672a9eSMatthew Wilcox (Oracle) locked = folio_lock_or_retry(folio, vma->vm_mm, vmf->flags); 3829e709ffd6SRik van Riel 3830d065bd81SMichel Lespinasse if (!locked) { 3831d065bd81SMichel Lespinasse ret |= VM_FAULT_RETRY; 3832d065bd81SMichel Lespinasse goto out_release; 3833d065bd81SMichel Lespinasse } 38341da177e4SLinus Torvalds 383584d60fddSDavid Hildenbrand if (swapcache) { 38364969c119SAndrea Arcangeli /* 38373b344157SMatthew Wilcox (Oracle) * Make sure folio_free_swap() or swapoff did not release the 383884d60fddSDavid Hildenbrand * swapcache from under us. The page pin, and pte_same test 383984d60fddSDavid Hildenbrand * below, are not enough to exclude that. Even if it is still 384084d60fddSDavid Hildenbrand * swapcache, we need to check that the page's swap has not 384184d60fddSDavid Hildenbrand * changed. 38424969c119SAndrea Arcangeli */ 384363ad4addSMatthew Wilcox (Oracle) if (unlikely(!folio_test_swapcache(folio) || 384484d60fddSDavid Hildenbrand page_private(page) != entry.val)) 38454969c119SAndrea Arcangeli goto out_page; 38464969c119SAndrea Arcangeli 384784d60fddSDavid Hildenbrand /* 384884d60fddSDavid Hildenbrand * KSM sometimes has to copy on read faults, for example, if 384984d60fddSDavid Hildenbrand * page->index of !PageKSM() pages would be nonlinear inside the 385084d60fddSDavid Hildenbrand * anon VMA -- PageKSM() is lost on actual swapout. 385184d60fddSDavid Hildenbrand */ 385282b0f8c3SJan Kara page = ksm_might_need_to_copy(page, vma, vmf->address); 38534969c119SAndrea Arcangeli if (unlikely(!page)) { 38545ad64688SHugh Dickins ret = VM_FAULT_OOM; 38554969c119SAndrea Arcangeli goto out_page; 38566b970599SKefeng Wang } else if (unlikely(PTR_ERR(page) == -EHWPOISON)) { 38576b970599SKefeng Wang ret = VM_FAULT_HWPOISON; 38586b970599SKefeng Wang goto out_page; 38594969c119SAndrea Arcangeli } 386063ad4addSMatthew Wilcox (Oracle) folio = page_folio(page); 3861c145e0b4SDavid Hildenbrand 3862c145e0b4SDavid Hildenbrand /* 3863c145e0b4SDavid Hildenbrand * If we want to map a page that's in the swapcache writable, we 3864c145e0b4SDavid Hildenbrand * have to detect via the refcount if we're really the exclusive 3865c145e0b4SDavid Hildenbrand * owner. Try removing the extra reference from the local LRU 3866c145e0b4SDavid Hildenbrand * pagevecs if required. 3867c145e0b4SDavid Hildenbrand */ 3868d4f9565aSMatthew Wilcox (Oracle) if ((vmf->flags & FAULT_FLAG_WRITE) && folio == swapcache && 386963ad4addSMatthew Wilcox (Oracle) !folio_test_ksm(folio) && !folio_test_lru(folio)) 3870c145e0b4SDavid Hildenbrand lru_add_drain(); 387184d60fddSDavid Hildenbrand } 38725ad64688SHugh Dickins 38734231f842SKefeng Wang folio_throttle_swaprate(folio, GFP_KERNEL); 3874073e587eSKAMEZAWA Hiroyuki 38751da177e4SLinus Torvalds /* 38768f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 38771da177e4SLinus Torvalds */ 387882b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 387982b0f8c3SJan Kara &vmf->ptl); 38802994302bSJan Kara if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) 3881b8107480SKirill Korotaev goto out_nomap; 3882b8107480SKirill Korotaev 388363ad4addSMatthew Wilcox (Oracle) if (unlikely(!folio_test_uptodate(folio))) { 3884b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 3885b8107480SKirill Korotaev goto out_nomap; 38861da177e4SLinus Torvalds } 38871da177e4SLinus Torvalds 38888c7c6e34SKAMEZAWA Hiroyuki /* 388978fbe906SDavid Hildenbrand * PG_anon_exclusive reuses PG_mappedtodisk for anon pages. A swap pte 389078fbe906SDavid Hildenbrand * must never point at an anonymous page in the swapcache that is 389178fbe906SDavid Hildenbrand * PG_anon_exclusive. Sanity check that this holds and especially, that 389278fbe906SDavid Hildenbrand * no filesystem set PG_mappedtodisk on a page in the swapcache. Sanity 389378fbe906SDavid Hildenbrand * check after taking the PT lock and making sure that nobody 389478fbe906SDavid Hildenbrand * concurrently faulted in this page and set PG_anon_exclusive. 389578fbe906SDavid Hildenbrand */ 389663ad4addSMatthew Wilcox (Oracle) BUG_ON(!folio_test_anon(folio) && folio_test_mappedtodisk(folio)); 389763ad4addSMatthew Wilcox (Oracle) BUG_ON(folio_test_anon(folio) && PageAnonExclusive(page)); 389878fbe906SDavid Hildenbrand 389978fbe906SDavid Hildenbrand /* 39001493a191SDavid Hildenbrand * Check under PT lock (to protect against concurrent fork() sharing 39011493a191SDavid Hildenbrand * the swap entry concurrently) for certainly exclusive pages. 39021493a191SDavid Hildenbrand */ 390363ad4addSMatthew Wilcox (Oracle) if (!folio_test_ksm(folio)) { 39041493a191SDavid Hildenbrand exclusive = pte_swp_exclusive(vmf->orig_pte); 3905d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache) { 39061493a191SDavid Hildenbrand /* 39071493a191SDavid Hildenbrand * We have a fresh page that is not exposed to the 39081493a191SDavid Hildenbrand * swapcache -> certainly exclusive. 39091493a191SDavid Hildenbrand */ 39101493a191SDavid Hildenbrand exclusive = true; 391163ad4addSMatthew Wilcox (Oracle) } else if (exclusive && folio_test_writeback(folio) && 3912eacde327SMiaohe Lin data_race(si->flags & SWP_STABLE_WRITES)) { 39131493a191SDavid Hildenbrand /* 39141493a191SDavid Hildenbrand * This is tricky: not all swap backends support 39151493a191SDavid Hildenbrand * concurrent page modifications while under writeback. 39161493a191SDavid Hildenbrand * 39171493a191SDavid Hildenbrand * So if we stumble over such a page in the swapcache 39181493a191SDavid Hildenbrand * we must not set the page exclusive, otherwise we can 39191493a191SDavid Hildenbrand * map it writable without further checks and modify it 39201493a191SDavid Hildenbrand * while still under writeback. 39211493a191SDavid Hildenbrand * 39221493a191SDavid Hildenbrand * For these problematic swap backends, simply drop the 39231493a191SDavid Hildenbrand * exclusive marker: this is perfectly fine as we start 39241493a191SDavid Hildenbrand * writeback only if we fully unmapped the page and 39251493a191SDavid Hildenbrand * there are no unexpected references on the page after 39261493a191SDavid Hildenbrand * unmapping succeeded. After fully unmapped, no 39271493a191SDavid Hildenbrand * further GUP references (FOLL_GET and FOLL_PIN) can 39281493a191SDavid Hildenbrand * appear, so dropping the exclusive marker and mapping 39291493a191SDavid Hildenbrand * it only R/O is fine. 39301493a191SDavid Hildenbrand */ 39311493a191SDavid Hildenbrand exclusive = false; 39321493a191SDavid Hildenbrand } 39331493a191SDavid Hildenbrand } 39341493a191SDavid Hildenbrand 39351493a191SDavid Hildenbrand /* 3936c145e0b4SDavid Hildenbrand * Remove the swap entry and conditionally try to free up the swapcache. 3937c145e0b4SDavid Hildenbrand * We're already holding a reference on the page but haven't mapped it 3938c145e0b4SDavid Hildenbrand * yet. 39398c7c6e34SKAMEZAWA Hiroyuki */ 3940c145e0b4SDavid Hildenbrand swap_free(entry); 3941a160e537SMatthew Wilcox (Oracle) if (should_try_to_free_swap(folio, vma, vmf->flags)) 3942a160e537SMatthew Wilcox (Oracle) folio_free_swap(folio); 39431da177e4SLinus Torvalds 3944f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, MM_ANONPAGES); 3945f1a79412SShakeel Butt dec_mm_counter(vma->vm_mm, MM_SWAPENTS); 39461da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 3947c145e0b4SDavid Hildenbrand 3948c145e0b4SDavid Hildenbrand /* 39491493a191SDavid Hildenbrand * Same logic as in do_wp_page(); however, optimize for pages that are 39501493a191SDavid Hildenbrand * certainly not shared either because we just allocated them without 39511493a191SDavid Hildenbrand * exposing them to the swapcache or because the swap entry indicates 39521493a191SDavid Hildenbrand * exclusivity. 3953c145e0b4SDavid Hildenbrand */ 395463ad4addSMatthew Wilcox (Oracle) if (!folio_test_ksm(folio) && 395563ad4addSMatthew Wilcox (Oracle) (exclusive || folio_ref_count(folio) == 1)) { 39566c287605SDavid Hildenbrand if (vmf->flags & FAULT_FLAG_WRITE) { 39571da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 395882b0f8c3SJan Kara vmf->flags &= ~FAULT_FLAG_WRITE; 39596c287605SDavid Hildenbrand } 396014f9135dSDavid Hildenbrand rmap_flags |= RMAP_EXCLUSIVE; 39611da177e4SLinus Torvalds } 39621da177e4SLinus Torvalds flush_icache_page(vma, page); 39632994302bSJan Kara if (pte_swp_soft_dirty(vmf->orig_pte)) 3964179ef71cSCyrill Gorcunov pte = pte_mksoft_dirty(pte); 3965f1eb1bacSPeter Xu if (pte_swp_uffd_wp(vmf->orig_pte)) 3966f45ec5ffSPeter Xu pte = pte_mkuffd_wp(pte); 39672994302bSJan Kara vmf->orig_pte = pte; 39680bcac06fSMinchan Kim 39690bcac06fSMinchan Kim /* ksm created a completely new copy */ 3970d4f9565aSMatthew Wilcox (Oracle) if (unlikely(folio != swapcache && swapcache)) { 397140f2bbf7SDavid Hildenbrand page_add_new_anon_rmap(page, vma, vmf->address); 397263ad4addSMatthew Wilcox (Oracle) folio_add_lru_vma(folio, vma); 39730bcac06fSMinchan Kim } else { 3974f1e2db12SDavid Hildenbrand page_add_anon_rmap(page, vma, vmf->address, rmap_flags); 397500501b53SJohannes Weiner } 39761da177e4SLinus Torvalds 397763ad4addSMatthew Wilcox (Oracle) VM_BUG_ON(!folio_test_anon(folio) || 397863ad4addSMatthew Wilcox (Oracle) (pte_write(pte) && !PageAnonExclusive(page))); 39791eba86c0SPasha Tatashin set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); 39801eba86c0SPasha Tatashin arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte); 39811eba86c0SPasha Tatashin 398263ad4addSMatthew Wilcox (Oracle) folio_unlock(folio); 3983d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache && swapcache) { 39844969c119SAndrea Arcangeli /* 39854969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 39864969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 39874969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 39884969c119SAndrea Arcangeli * further safety release the lock after the swap_free 39894969c119SAndrea Arcangeli * so that the swap count won't change under a 39904969c119SAndrea Arcangeli * parallel locked swapcache. 39914969c119SAndrea Arcangeli */ 3992d4f9565aSMatthew Wilcox (Oracle) folio_unlock(swapcache); 3993d4f9565aSMatthew Wilcox (Oracle) folio_put(swapcache); 39944969c119SAndrea Arcangeli } 3995c475a8abSHugh Dickins 399682b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) { 39972994302bSJan Kara ret |= do_wp_page(vmf); 399861469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 399961469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 40001da177e4SLinus Torvalds goto out; 40011da177e4SLinus Torvalds } 40021da177e4SLinus Torvalds 40031da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 400482b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 400565500d23SHugh Dickins unlock: 400682b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 40071da177e4SLinus Torvalds out: 40082799e775SMiaohe Lin if (si) 40092799e775SMiaohe Lin put_swap_device(si); 40101da177e4SLinus Torvalds return ret; 4011b8107480SKirill Korotaev out_nomap: 401282b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 4013bc43f75cSJohannes Weiner out_page: 401463ad4addSMatthew Wilcox (Oracle) folio_unlock(folio); 40154779cb31SAndi Kleen out_release: 401663ad4addSMatthew Wilcox (Oracle) folio_put(folio); 4017d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache && swapcache) { 4018d4f9565aSMatthew Wilcox (Oracle) folio_unlock(swapcache); 4019d4f9565aSMatthew Wilcox (Oracle) folio_put(swapcache); 40204969c119SAndrea Arcangeli } 40212799e775SMiaohe Lin if (si) 40222799e775SMiaohe Lin put_swap_device(si); 402365500d23SHugh Dickins return ret; 40241da177e4SLinus Torvalds } 40251da177e4SLinus Torvalds 40261da177e4SLinus Torvalds /* 4027c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 40288f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 4029c1e8d7c6SMichel Lespinasse * We return with mmap_lock still held, but pte unmapped and unlocked. 40301da177e4SLinus Torvalds */ 40312b740303SSouptick Joarder static vm_fault_t do_anonymous_page(struct vm_fault *vmf) 40321da177e4SLinus Torvalds { 40332bad466cSPeter Xu bool uffd_wp = vmf_orig_pte_uffd_wp(vmf); 403482b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 40356bc56a4dSMatthew Wilcox (Oracle) struct folio *folio; 40362b740303SSouptick Joarder vm_fault_t ret = 0; 40371da177e4SLinus Torvalds pte_t entry; 40381da177e4SLinus Torvalds 40396b7339f4SKirill A. Shutemov /* File mapping without ->vm_ops ? */ 40406b7339f4SKirill A. Shutemov if (vma->vm_flags & VM_SHARED) 40416b7339f4SKirill A. Shutemov return VM_FAULT_SIGBUS; 40426b7339f4SKirill A. Shutemov 40437267ec00SKirill A. Shutemov /* 40447267ec00SKirill A. Shutemov * Use pte_alloc() instead of pte_alloc_map(). We can't run 40457267ec00SKirill A. Shutemov * pte_offset_map() on pmds where a huge pmd might be created 40467267ec00SKirill A. Shutemov * from a different thread. 40477267ec00SKirill A. Shutemov * 40483e4e28c5SMichel Lespinasse * pte_alloc_map() is safe to use under mmap_write_lock(mm) or when 40497267ec00SKirill A. Shutemov * parallel threads are excluded by other means. 40507267ec00SKirill A. Shutemov * 40513e4e28c5SMichel Lespinasse * Here we only have mmap_read_lock(mm). 40527267ec00SKirill A. Shutemov */ 40534cf58924SJoel Fernandes (Google) if (pte_alloc(vma->vm_mm, vmf->pmd)) 40547267ec00SKirill A. Shutemov return VM_FAULT_OOM; 40557267ec00SKirill A. Shutemov 4056f9ce0be7SKirill A. Shutemov /* See comment in handle_pte_fault() */ 405782b0f8c3SJan Kara if (unlikely(pmd_trans_unstable(vmf->pmd))) 40587267ec00SKirill A. Shutemov return 0; 40597267ec00SKirill A. Shutemov 406011ac5524SLinus Torvalds /* Use the zero-page for reads */ 406182b0f8c3SJan Kara if (!(vmf->flags & FAULT_FLAG_WRITE) && 4062bae473a4SKirill A. Shutemov !mm_forbids_zeropage(vma->vm_mm)) { 406382b0f8c3SJan Kara entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), 406462eede62SHugh Dickins vma->vm_page_prot)); 406582b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 406682b0f8c3SJan Kara vmf->address, &vmf->ptl); 40672bad466cSPeter Xu if (vmf_pte_changed(vmf)) { 40687df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 4069a13ea5b7SHugh Dickins goto unlock; 40707df67697SBibo Mao } 40716b31d595SMichal Hocko ret = check_stable_address_space(vma->vm_mm); 40726b31d595SMichal Hocko if (ret) 40736b31d595SMichal Hocko goto unlock; 40746b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 40756b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 407682b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 407782b0f8c3SJan Kara return handle_userfault(vmf, VM_UFFD_MISSING); 40786b251fc9SAndrea Arcangeli } 4079a13ea5b7SHugh Dickins goto setpte; 4080a13ea5b7SHugh Dickins } 4081a13ea5b7SHugh Dickins 40821da177e4SLinus Torvalds /* Allocate our own private page. */ 40831da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 408465500d23SHugh Dickins goto oom; 40856bc56a4dSMatthew Wilcox (Oracle) folio = vma_alloc_zeroed_movable_folio(vma, vmf->address); 40866bc56a4dSMatthew Wilcox (Oracle) if (!folio) 408765500d23SHugh Dickins goto oom; 4088eb3c24f3SMel Gorman 40896bc56a4dSMatthew Wilcox (Oracle) if (mem_cgroup_charge(folio, vma->vm_mm, GFP_KERNEL)) 4090eb3c24f3SMel Gorman goto oom_free_page; 4091e2bf3e2cSKefeng Wang folio_throttle_swaprate(folio, GFP_KERNEL); 4092eb3c24f3SMel Gorman 409352f37629SMinchan Kim /* 4094cb3184deSMatthew Wilcox (Oracle) * The memory barrier inside __folio_mark_uptodate makes sure that 4095f4f5329dSWei Yang * preceding stores to the page contents become visible before 409652f37629SMinchan Kim * the set_pte_at() write. 409752f37629SMinchan Kim */ 4098cb3184deSMatthew Wilcox (Oracle) __folio_mark_uptodate(folio); 40991da177e4SLinus Torvalds 4100cb3184deSMatthew Wilcox (Oracle) entry = mk_pte(&folio->page, vma->vm_page_prot); 410150c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 41021ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 41031ac0cb5dSHugh Dickins entry = pte_mkwrite(pte_mkdirty(entry)); 41048f4e2101SHugh Dickins 410582b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 410682b0f8c3SJan Kara &vmf->ptl); 41072bad466cSPeter Xu if (vmf_pte_changed(vmf)) { 4108bce8cb3cSQi Zheng update_mmu_tlb(vma, vmf->address, vmf->pte); 41098f4e2101SHugh Dickins goto release; 41107df67697SBibo Mao } 41119ba69294SHugh Dickins 41126b31d595SMichal Hocko ret = check_stable_address_space(vma->vm_mm); 41136b31d595SMichal Hocko if (ret) 41146b31d595SMichal Hocko goto release; 41156b31d595SMichal Hocko 41166b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 41176b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 411882b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 4119cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 412082b0f8c3SJan Kara return handle_userfault(vmf, VM_UFFD_MISSING); 41216b251fc9SAndrea Arcangeli } 41226b251fc9SAndrea Arcangeli 4123f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, MM_ANONPAGES); 4124cb3184deSMatthew Wilcox (Oracle) folio_add_new_anon_rmap(folio, vma, vmf->address); 4125cb3184deSMatthew Wilcox (Oracle) folio_add_lru_vma(folio, vma); 4126a13ea5b7SHugh Dickins setpte: 41272bad466cSPeter Xu if (uffd_wp) 41282bad466cSPeter Xu entry = pte_mkuffd_wp(entry); 412982b0f8c3SJan Kara set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); 41301da177e4SLinus Torvalds 41311da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 413282b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 413365500d23SHugh Dickins unlock: 413482b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 41356b31d595SMichal Hocko return ret; 41368f4e2101SHugh Dickins release: 4137cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 41388f4e2101SHugh Dickins goto unlock; 41398a9f3ccdSBalbir Singh oom_free_page: 4140cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 414165500d23SHugh Dickins oom: 41421da177e4SLinus Torvalds return VM_FAULT_OOM; 41431da177e4SLinus Torvalds } 41441da177e4SLinus Torvalds 41459a95f3cfSPaul Cassella /* 4146c1e8d7c6SMichel Lespinasse * The mmap_lock must have been held on entry, and may have been 41479a95f3cfSPaul Cassella * released depending on flags and vma->vm_ops->fault() return value. 41489a95f3cfSPaul Cassella * See filemap_fault() and __lock_page_retry(). 41499a95f3cfSPaul Cassella */ 41502b740303SSouptick Joarder static vm_fault_t __do_fault(struct vm_fault *vmf) 41517eae74afSKirill A. Shutemov { 415282b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 41532b740303SSouptick Joarder vm_fault_t ret; 41547eae74afSKirill A. Shutemov 415563f3655fSMichal Hocko /* 415663f3655fSMichal Hocko * Preallocate pte before we take page_lock because this might lead to 415763f3655fSMichal Hocko * deadlocks for memcg reclaim which waits for pages under writeback: 415863f3655fSMichal Hocko * lock_page(A) 415963f3655fSMichal Hocko * SetPageWriteback(A) 416063f3655fSMichal Hocko * unlock_page(A) 416163f3655fSMichal Hocko * lock_page(B) 416263f3655fSMichal Hocko * lock_page(B) 4163d383807aSYanfei Xu * pte_alloc_one 416463f3655fSMichal Hocko * shrink_page_list 416563f3655fSMichal Hocko * wait_on_page_writeback(A) 416663f3655fSMichal Hocko * SetPageWriteback(B) 416763f3655fSMichal Hocko * unlock_page(B) 416863f3655fSMichal Hocko * # flush A, B to clear the writeback 416963f3655fSMichal Hocko */ 417063f3655fSMichal Hocko if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) { 4171a7069ee3SYanfei Xu vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); 417263f3655fSMichal Hocko if (!vmf->prealloc_pte) 417363f3655fSMichal Hocko return VM_FAULT_OOM; 417463f3655fSMichal Hocko } 417563f3655fSMichal Hocko 417611bac800SDave Jiang ret = vma->vm_ops->fault(vmf); 41773917048dSJan Kara if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | 4178b1aa812bSJan Kara VM_FAULT_DONE_COW))) 41797eae74afSKirill A. Shutemov return ret; 41807eae74afSKirill A. Shutemov 4181667240e0SJan Kara if (unlikely(PageHWPoison(vmf->page))) { 41823149c79fSRik van Riel struct page *page = vmf->page; 4183e53ac737SRik van Riel vm_fault_t poisonret = VM_FAULT_HWPOISON; 4184e53ac737SRik van Riel if (ret & VM_FAULT_LOCKED) { 41853149c79fSRik van Riel if (page_mapped(page)) 41863149c79fSRik van Riel unmap_mapping_pages(page_mapping(page), 41873149c79fSRik van Riel page->index, 1, false); 4188e53ac737SRik van Riel /* Retry if a clean page was removed from the cache. */ 41893149c79fSRik van Riel if (invalidate_inode_page(page)) 41903149c79fSRik van Riel poisonret = VM_FAULT_NOPAGE; 41913149c79fSRik van Riel unlock_page(page); 4192e53ac737SRik van Riel } 41933149c79fSRik van Riel put_page(page); 4194936ca80dSJan Kara vmf->page = NULL; 4195e53ac737SRik van Riel return poisonret; 41967eae74afSKirill A. Shutemov } 41977eae74afSKirill A. Shutemov 41987eae74afSKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) 4199667240e0SJan Kara lock_page(vmf->page); 42007eae74afSKirill A. Shutemov else 4201667240e0SJan Kara VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); 42027eae74afSKirill A. Shutemov 42037eae74afSKirill A. Shutemov return ret; 42047eae74afSKirill A. Shutemov } 42057eae74afSKirill A. Shutemov 4206396bcc52SMatthew Wilcox (Oracle) #ifdef CONFIG_TRANSPARENT_HUGEPAGE 420782b0f8c3SJan Kara static void deposit_prealloc_pte(struct vm_fault *vmf) 4208953c66c2SAneesh Kumar K.V { 420982b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4210953c66c2SAneesh Kumar K.V 421182b0f8c3SJan Kara pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); 4212953c66c2SAneesh Kumar K.V /* 4213953c66c2SAneesh Kumar K.V * We are going to consume the prealloc table, 4214953c66c2SAneesh Kumar K.V * count that as nr_ptes. 4215953c66c2SAneesh Kumar K.V */ 4216c4812909SKirill A. Shutemov mm_inc_nr_ptes(vma->vm_mm); 42177f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 4218953c66c2SAneesh Kumar K.V } 4219953c66c2SAneesh Kumar K.V 4220f9ce0be7SKirill A. Shutemov vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) 422110102459SKirill A. Shutemov { 422282b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 422382b0f8c3SJan Kara bool write = vmf->flags & FAULT_FLAG_WRITE; 422482b0f8c3SJan Kara unsigned long haddr = vmf->address & HPAGE_PMD_MASK; 422510102459SKirill A. Shutemov pmd_t entry; 42262b740303SSouptick Joarder int i; 4227d01ac3c3SMatthew Wilcox (Oracle) vm_fault_t ret = VM_FAULT_FALLBACK; 422810102459SKirill A. Shutemov 422910102459SKirill A. Shutemov if (!transhuge_vma_suitable(vma, haddr)) 4230d01ac3c3SMatthew Wilcox (Oracle) return ret; 423110102459SKirill A. Shutemov 423210102459SKirill A. Shutemov page = compound_head(page); 4233d01ac3c3SMatthew Wilcox (Oracle) if (compound_order(page) != HPAGE_PMD_ORDER) 4234d01ac3c3SMatthew Wilcox (Oracle) return ret; 423510102459SKirill A. Shutemov 4236953c66c2SAneesh Kumar K.V /* 4237eac96c3eSYang Shi * Just backoff if any subpage of a THP is corrupted otherwise 4238eac96c3eSYang Shi * the corrupted page may mapped by PMD silently to escape the 4239eac96c3eSYang Shi * check. This kind of THP just can be PTE mapped. Access to 4240eac96c3eSYang Shi * the corrupted subpage should trigger SIGBUS as expected. 4241eac96c3eSYang Shi */ 4242eac96c3eSYang Shi if (unlikely(PageHasHWPoisoned(page))) 4243eac96c3eSYang Shi return ret; 4244eac96c3eSYang Shi 4245eac96c3eSYang Shi /* 4246f0953a1bSIngo Molnar * Archs like ppc64 need additional space to store information 4247953c66c2SAneesh Kumar K.V * related to pte entry. Use the preallocated table for that. 4248953c66c2SAneesh Kumar K.V */ 424982b0f8c3SJan Kara if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { 42504cf58924SJoel Fernandes (Google) vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); 425182b0f8c3SJan Kara if (!vmf->prealloc_pte) 4252953c66c2SAneesh Kumar K.V return VM_FAULT_OOM; 4253953c66c2SAneesh Kumar K.V } 4254953c66c2SAneesh Kumar K.V 425582b0f8c3SJan Kara vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); 425682b0f8c3SJan Kara if (unlikely(!pmd_none(*vmf->pmd))) 425710102459SKirill A. Shutemov goto out; 425810102459SKirill A. Shutemov 425910102459SKirill A. Shutemov for (i = 0; i < HPAGE_PMD_NR; i++) 426010102459SKirill A. Shutemov flush_icache_page(vma, page + i); 426110102459SKirill A. Shutemov 426210102459SKirill A. Shutemov entry = mk_huge_pmd(page, vma->vm_page_prot); 426310102459SKirill A. Shutemov if (write) 4264f55e1014SLinus Torvalds entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); 426510102459SKirill A. Shutemov 4266fadae295SYang Shi add_mm_counter(vma->vm_mm, mm_counter_file(page), HPAGE_PMD_NR); 4267cea86fe2SHugh Dickins page_add_file_rmap(page, vma, true); 4268cea86fe2SHugh Dickins 4269953c66c2SAneesh Kumar K.V /* 4270953c66c2SAneesh Kumar K.V * deposit and withdraw with pmd lock held 4271953c66c2SAneesh Kumar K.V */ 4272953c66c2SAneesh Kumar K.V if (arch_needs_pgtable_deposit()) 427382b0f8c3SJan Kara deposit_prealloc_pte(vmf); 427410102459SKirill A. Shutemov 427582b0f8c3SJan Kara set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); 427610102459SKirill A. Shutemov 427782b0f8c3SJan Kara update_mmu_cache_pmd(vma, haddr, vmf->pmd); 427810102459SKirill A. Shutemov 427910102459SKirill A. Shutemov /* fault is handled */ 428010102459SKirill A. Shutemov ret = 0; 428195ecedcdSKirill A. Shutemov count_vm_event(THP_FILE_MAPPED); 428210102459SKirill A. Shutemov out: 428382b0f8c3SJan Kara spin_unlock(vmf->ptl); 428410102459SKirill A. Shutemov return ret; 428510102459SKirill A. Shutemov } 428610102459SKirill A. Shutemov #else 4287f9ce0be7SKirill A. Shutemov vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) 428810102459SKirill A. Shutemov { 4289f9ce0be7SKirill A. Shutemov return VM_FAULT_FALLBACK; 429010102459SKirill A. Shutemov } 429110102459SKirill A. Shutemov #endif 429210102459SKirill A. Shutemov 42939d3af4b4SWill Deacon void do_set_pte(struct vm_fault *vmf, struct page *page, unsigned long addr) 42943bb97794SKirill A. Shutemov { 429582b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 42962bad466cSPeter Xu bool uffd_wp = vmf_orig_pte_uffd_wp(vmf); 429782b0f8c3SJan Kara bool write = vmf->flags & FAULT_FLAG_WRITE; 42989d3af4b4SWill Deacon bool prefault = vmf->address != addr; 42993bb97794SKirill A. Shutemov pte_t entry; 43007267ec00SKirill A. Shutemov 43013bb97794SKirill A. Shutemov flush_icache_page(vma, page); 43023bb97794SKirill A. Shutemov entry = mk_pte(page, vma->vm_page_prot); 430346bdb427SWill Deacon 430446bdb427SWill Deacon if (prefault && arch_wants_old_prefaulted_pte()) 430546bdb427SWill Deacon entry = pte_mkold(entry); 430650c25ee9SThomas Bogendoerfer else 430750c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 430846bdb427SWill Deacon 43093bb97794SKirill A. Shutemov if (write) 43103bb97794SKirill A. Shutemov entry = maybe_mkwrite(pte_mkdirty(entry), vma); 43119c28a205SPeter Xu if (unlikely(uffd_wp)) 4312f1eb1bacSPeter Xu entry = pte_mkuffd_wp(entry); 4313bae473a4SKirill A. Shutemov /* copy-on-write page */ 4314bae473a4SKirill A. Shutemov if (write && !(vma->vm_flags & VM_SHARED)) { 4315f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, MM_ANONPAGES); 431640f2bbf7SDavid Hildenbrand page_add_new_anon_rmap(page, vma, addr); 4317b518154eSJoonsoo Kim lru_cache_add_inactive_or_unevictable(page, vma); 43183bb97794SKirill A. Shutemov } else { 4319f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, mm_counter_file(page)); 4320cea86fe2SHugh Dickins page_add_file_rmap(page, vma, false); 43213bb97794SKirill A. Shutemov } 43229d3af4b4SWill Deacon set_pte_at(vma->vm_mm, addr, vmf->pte, entry); 43233bb97794SKirill A. Shutemov } 43243bb97794SKirill A. Shutemov 4325f46f2adeSPeter Xu static bool vmf_pte_changed(struct vm_fault *vmf) 4326f46f2adeSPeter Xu { 4327f46f2adeSPeter Xu if (vmf->flags & FAULT_FLAG_ORIG_PTE_VALID) 4328f46f2adeSPeter Xu return !pte_same(*vmf->pte, vmf->orig_pte); 4329f46f2adeSPeter Xu 4330f46f2adeSPeter Xu return !pte_none(*vmf->pte); 4331f46f2adeSPeter Xu } 4332f46f2adeSPeter Xu 43339118c0cbSJan Kara /** 43349118c0cbSJan Kara * finish_fault - finish page fault once we have prepared the page to fault 43359118c0cbSJan Kara * 43369118c0cbSJan Kara * @vmf: structure describing the fault 43379118c0cbSJan Kara * 43389118c0cbSJan Kara * This function handles all that is needed to finish a page fault once the 43399118c0cbSJan Kara * page to fault in is prepared. It handles locking of PTEs, inserts PTE for 43409118c0cbSJan Kara * given page, adds reverse page mapping, handles memcg charges and LRU 4341a862f68aSMike Rapoport * addition. 43429118c0cbSJan Kara * 43439118c0cbSJan Kara * The function expects the page to be locked and on success it consumes a 43449118c0cbSJan Kara * reference of a page being mapped (for the PTE which maps it). 4345a862f68aSMike Rapoport * 4346a862f68aSMike Rapoport * Return: %0 on success, %VM_FAULT_ code in case of error. 43479118c0cbSJan Kara */ 43482b740303SSouptick Joarder vm_fault_t finish_fault(struct vm_fault *vmf) 43499118c0cbSJan Kara { 4350f9ce0be7SKirill A. Shutemov struct vm_area_struct *vma = vmf->vma; 43519118c0cbSJan Kara struct page *page; 4352f9ce0be7SKirill A. Shutemov vm_fault_t ret; 43539118c0cbSJan Kara 43549118c0cbSJan Kara /* Did we COW the page? */ 4355f9ce0be7SKirill A. Shutemov if ((vmf->flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) 43569118c0cbSJan Kara page = vmf->cow_page; 43579118c0cbSJan Kara else 43589118c0cbSJan Kara page = vmf->page; 43596b31d595SMichal Hocko 43606b31d595SMichal Hocko /* 43616b31d595SMichal Hocko * check even for read faults because we might have lost our CoWed 43626b31d595SMichal Hocko * page 43636b31d595SMichal Hocko */ 4364f9ce0be7SKirill A. Shutemov if (!(vma->vm_flags & VM_SHARED)) { 4365f9ce0be7SKirill A. Shutemov ret = check_stable_address_space(vma->vm_mm); 4366f9ce0be7SKirill A. Shutemov if (ret) 4367f9ce0be7SKirill A. Shutemov return ret; 4368f9ce0be7SKirill A. Shutemov } 4369f9ce0be7SKirill A. Shutemov 4370f9ce0be7SKirill A. Shutemov if (pmd_none(*vmf->pmd)) { 4371f9ce0be7SKirill A. Shutemov if (PageTransCompound(page)) { 4372f9ce0be7SKirill A. Shutemov ret = do_set_pmd(vmf, page); 4373f9ce0be7SKirill A. Shutemov if (ret != VM_FAULT_FALLBACK) 4374f9ce0be7SKirill A. Shutemov return ret; 4375f9ce0be7SKirill A. Shutemov } 4376f9ce0be7SKirill A. Shutemov 437703c4f204SQi Zheng if (vmf->prealloc_pte) 437803c4f204SQi Zheng pmd_install(vma->vm_mm, vmf->pmd, &vmf->prealloc_pte); 437903c4f204SQi Zheng else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd))) 4380f9ce0be7SKirill A. Shutemov return VM_FAULT_OOM; 4381f9ce0be7SKirill A. Shutemov } 4382f9ce0be7SKirill A. Shutemov 43833fe2895cSJosef Bacik /* 43843fe2895cSJosef Bacik * See comment in handle_pte_fault() for how this scenario happens, we 43853fe2895cSJosef Bacik * need to return NOPAGE so that we drop this page. 43863fe2895cSJosef Bacik */ 4387f9ce0be7SKirill A. Shutemov if (pmd_devmap_trans_unstable(vmf->pmd)) 43883fe2895cSJosef Bacik return VM_FAULT_NOPAGE; 4389f9ce0be7SKirill A. Shutemov 4390f9ce0be7SKirill A. Shutemov vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 4391f9ce0be7SKirill A. Shutemov vmf->address, &vmf->ptl); 4392f9ce0be7SKirill A. Shutemov 439370427f6eSSergei Antonov /* Re-check under ptl */ 439470427f6eSSergei Antonov if (likely(!vmf_pte_changed(vmf))) { 439570427f6eSSergei Antonov do_set_pte(vmf, page, vmf->address); 439670427f6eSSergei Antonov 439770427f6eSSergei Antonov /* no need to invalidate: a not-present page won't be cached */ 439870427f6eSSergei Antonov update_mmu_cache(vma, vmf->address, vmf->pte); 439970427f6eSSergei Antonov 440070427f6eSSergei Antonov ret = 0; 440170427f6eSSergei Antonov } else { 4402f9ce0be7SKirill A. Shutemov update_mmu_tlb(vma, vmf->address, vmf->pte); 440370427f6eSSergei Antonov ret = VM_FAULT_NOPAGE; 440470427f6eSSergei Antonov } 440570427f6eSSergei Antonov 44069118c0cbSJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 44079118c0cbSJan Kara return ret; 44089118c0cbSJan Kara } 44099118c0cbSJan Kara 441053d36a56SLorenzo Stoakes static unsigned long fault_around_pages __read_mostly = 441153d36a56SLorenzo Stoakes 65536 >> PAGE_SHIFT; 4412a9b0f861SKirill A. Shutemov 44131592eef0SKirill A. Shutemov #ifdef CONFIG_DEBUG_FS 4414a9b0f861SKirill A. Shutemov static int fault_around_bytes_get(void *data, u64 *val) 44151592eef0SKirill A. Shutemov { 441653d36a56SLorenzo Stoakes *val = fault_around_pages << PAGE_SHIFT; 44171592eef0SKirill A. Shutemov return 0; 44181592eef0SKirill A. Shutemov } 44191592eef0SKirill A. Shutemov 4420b4903d6eSAndrey Ryabinin /* 4421da391d64SWilliam Kucharski * fault_around_bytes must be rounded down to the nearest page order as it's 4422da391d64SWilliam Kucharski * what do_fault_around() expects to see. 4423b4903d6eSAndrey Ryabinin */ 4424a9b0f861SKirill A. Shutemov static int fault_around_bytes_set(void *data, u64 val) 44251592eef0SKirill A. Shutemov { 4426a9b0f861SKirill A. Shutemov if (val / PAGE_SIZE > PTRS_PER_PTE) 44271592eef0SKirill A. Shutemov return -EINVAL; 442853d36a56SLorenzo Stoakes 442953d36a56SLorenzo Stoakes /* 443053d36a56SLorenzo Stoakes * The minimum value is 1 page, however this results in no fault-around 443153d36a56SLorenzo Stoakes * at all. See should_fault_around(). 443253d36a56SLorenzo Stoakes */ 443353d36a56SLorenzo Stoakes fault_around_pages = max(rounddown_pow_of_two(val) >> PAGE_SHIFT, 1UL); 443453d36a56SLorenzo Stoakes 44351592eef0SKirill A. Shutemov return 0; 44361592eef0SKirill A. Shutemov } 44370a1345f8SYevgen Pronenko DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, 4438a9b0f861SKirill A. Shutemov fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); 44391592eef0SKirill A. Shutemov 44401592eef0SKirill A. Shutemov static int __init fault_around_debugfs(void) 44411592eef0SKirill A. Shutemov { 4442d9f7979cSGreg Kroah-Hartman debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, 4443a9b0f861SKirill A. Shutemov &fault_around_bytes_fops); 44441592eef0SKirill A. Shutemov return 0; 44451592eef0SKirill A. Shutemov } 44461592eef0SKirill A. Shutemov late_initcall(fault_around_debugfs); 44471592eef0SKirill A. Shutemov #endif 44488c6e50b0SKirill A. Shutemov 44491fdb412bSKirill A. Shutemov /* 44501fdb412bSKirill A. Shutemov * do_fault_around() tries to map few pages around the fault address. The hope 44511fdb412bSKirill A. Shutemov * is that the pages will be needed soon and this will lower the number of 44521fdb412bSKirill A. Shutemov * faults to handle. 44531fdb412bSKirill A. Shutemov * 44541fdb412bSKirill A. Shutemov * It uses vm_ops->map_pages() to map the pages, which skips the page if it's 44551fdb412bSKirill A. Shutemov * not ready to be mapped: not up-to-date, locked, etc. 44561fdb412bSKirill A. Shutemov * 44579042599eSLorenzo Stoakes * This function doesn't cross VMA or page table boundaries, in order to call 44589042599eSLorenzo Stoakes * map_pages() and acquire a PTE lock only once. 44591fdb412bSKirill A. Shutemov * 446053d36a56SLorenzo Stoakes * fault_around_pages defines how many pages we'll try to map. 4461da391d64SWilliam Kucharski * do_fault_around() expects it to be set to a power of two less than or equal 4462da391d64SWilliam Kucharski * to PTRS_PER_PTE. 44631fdb412bSKirill A. Shutemov * 4464da391d64SWilliam Kucharski * The virtual address of the area that we map is naturally aligned to 446553d36a56SLorenzo Stoakes * fault_around_pages * PAGE_SIZE rounded down to the machine page size 4466da391d64SWilliam Kucharski * (and therefore to page order). This way it's easier to guarantee 4467da391d64SWilliam Kucharski * that we don't cross page table boundaries. 44681fdb412bSKirill A. Shutemov */ 44692b740303SSouptick Joarder static vm_fault_t do_fault_around(struct vm_fault *vmf) 44708c6e50b0SKirill A. Shutemov { 447153d36a56SLorenzo Stoakes pgoff_t nr_pages = READ_ONCE(fault_around_pages); 44729042599eSLorenzo Stoakes pgoff_t pte_off = pte_index(vmf->address); 44739042599eSLorenzo Stoakes /* The page offset of vmf->address within the VMA. */ 44749042599eSLorenzo Stoakes pgoff_t vma_off = vmf->pgoff - vmf->vma->vm_pgoff; 44759042599eSLorenzo Stoakes pgoff_t from_pte, to_pte; 447658ef47efSMatthew Wilcox (Oracle) vm_fault_t ret; 44778c6e50b0SKirill A. Shutemov 44789042599eSLorenzo Stoakes /* The PTE offset of the start address, clamped to the VMA. */ 44799042599eSLorenzo Stoakes from_pte = max(ALIGN_DOWN(pte_off, nr_pages), 44809042599eSLorenzo Stoakes pte_off - min(pte_off, vma_off)); 4481aecd6f44SKirill A. Shutemov 44829042599eSLorenzo Stoakes /* The PTE offset of the end address, clamped to the VMA and PTE. */ 44839042599eSLorenzo Stoakes to_pte = min3(from_pte + nr_pages, (pgoff_t)PTRS_PER_PTE, 44849042599eSLorenzo Stoakes pte_off + vma_pages(vmf->vma) - vma_off) - 1; 44858c6e50b0SKirill A. Shutemov 448682b0f8c3SJan Kara if (pmd_none(*vmf->pmd)) { 44874cf58924SJoel Fernandes (Google) vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm); 448882b0f8c3SJan Kara if (!vmf->prealloc_pte) 4489f9ce0be7SKirill A. Shutemov return VM_FAULT_OOM; 44908c6e50b0SKirill A. Shutemov } 44918c6e50b0SKirill A. Shutemov 449258ef47efSMatthew Wilcox (Oracle) rcu_read_lock(); 449358ef47efSMatthew Wilcox (Oracle) ret = vmf->vma->vm_ops->map_pages(vmf, 44949042599eSLorenzo Stoakes vmf->pgoff + from_pte - pte_off, 44959042599eSLorenzo Stoakes vmf->pgoff + to_pte - pte_off); 449658ef47efSMatthew Wilcox (Oracle) rcu_read_unlock(); 449758ef47efSMatthew Wilcox (Oracle) 449858ef47efSMatthew Wilcox (Oracle) return ret; 44997267ec00SKirill A. Shutemov } 45007267ec00SKirill A. Shutemov 45019c28a205SPeter Xu /* Return true if we should do read fault-around, false otherwise */ 45029c28a205SPeter Xu static inline bool should_fault_around(struct vm_fault *vmf) 45039c28a205SPeter Xu { 45049c28a205SPeter Xu /* No ->map_pages? No way to fault around... */ 45059c28a205SPeter Xu if (!vmf->vma->vm_ops->map_pages) 45069c28a205SPeter Xu return false; 45079c28a205SPeter Xu 45089c28a205SPeter Xu if (uffd_disable_fault_around(vmf->vma)) 45099c28a205SPeter Xu return false; 45109c28a205SPeter Xu 451153d36a56SLorenzo Stoakes /* A single page implies no faulting 'around' at all. */ 451253d36a56SLorenzo Stoakes return fault_around_pages > 1; 45139c28a205SPeter Xu } 45149c28a205SPeter Xu 45152b740303SSouptick Joarder static vm_fault_t do_read_fault(struct vm_fault *vmf) 4516e655fb29SKirill A. Shutemov { 45172b740303SSouptick Joarder vm_fault_t ret = 0; 45188c6e50b0SKirill A. Shutemov 45198c6e50b0SKirill A. Shutemov /* 45208c6e50b0SKirill A. Shutemov * Let's call ->map_pages() first and use ->fault() as fallback 45218c6e50b0SKirill A. Shutemov * if page by the offset is not ready to be mapped (cold cache or 45228c6e50b0SKirill A. Shutemov * something). 45238c6e50b0SKirill A. Shutemov */ 45249c28a205SPeter Xu if (should_fault_around(vmf)) { 45250721ec8bSJan Kara ret = do_fault_around(vmf); 45267267ec00SKirill A. Shutemov if (ret) 45277267ec00SKirill A. Shutemov return ret; 45288c6e50b0SKirill A. Shutemov } 4529e655fb29SKirill A. Shutemov 4530936ca80dSJan Kara ret = __do_fault(vmf); 4531e655fb29SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4532e655fb29SKirill A. Shutemov return ret; 4533e655fb29SKirill A. Shutemov 45349118c0cbSJan Kara ret |= finish_fault(vmf); 4535936ca80dSJan Kara unlock_page(vmf->page); 45367267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4537936ca80dSJan Kara put_page(vmf->page); 4538e655fb29SKirill A. Shutemov return ret; 4539e655fb29SKirill A. Shutemov } 4540e655fb29SKirill A. Shutemov 45412b740303SSouptick Joarder static vm_fault_t do_cow_fault(struct vm_fault *vmf) 4542ec47c3b9SKirill A. Shutemov { 454382b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 45442b740303SSouptick Joarder vm_fault_t ret; 4545ec47c3b9SKirill A. Shutemov 4546ec47c3b9SKirill A. Shutemov if (unlikely(anon_vma_prepare(vma))) 4547ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4548ec47c3b9SKirill A. Shutemov 4549936ca80dSJan Kara vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); 4550936ca80dSJan Kara if (!vmf->cow_page) 4551ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4552ec47c3b9SKirill A. Shutemov 45538f425e4eSMatthew Wilcox (Oracle) if (mem_cgroup_charge(page_folio(vmf->cow_page), vma->vm_mm, 45548f425e4eSMatthew Wilcox (Oracle) GFP_KERNEL)) { 4555936ca80dSJan Kara put_page(vmf->cow_page); 4556ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4557ec47c3b9SKirill A. Shutemov } 455868fa572bSKefeng Wang folio_throttle_swaprate(page_folio(vmf->cow_page), GFP_KERNEL); 4559ec47c3b9SKirill A. Shutemov 4560936ca80dSJan Kara ret = __do_fault(vmf); 4561ec47c3b9SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4562ec47c3b9SKirill A. Shutemov goto uncharge_out; 45633917048dSJan Kara if (ret & VM_FAULT_DONE_COW) 45643917048dSJan Kara return ret; 4565ec47c3b9SKirill A. Shutemov 4566936ca80dSJan Kara copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); 4567936ca80dSJan Kara __SetPageUptodate(vmf->cow_page); 4568ec47c3b9SKirill A. Shutemov 45699118c0cbSJan Kara ret |= finish_fault(vmf); 4570936ca80dSJan Kara unlock_page(vmf->page); 4571936ca80dSJan Kara put_page(vmf->page); 45727267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 45737267ec00SKirill A. Shutemov goto uncharge_out; 4574ec47c3b9SKirill A. Shutemov return ret; 4575ec47c3b9SKirill A. Shutemov uncharge_out: 4576936ca80dSJan Kara put_page(vmf->cow_page); 4577ec47c3b9SKirill A. Shutemov return ret; 4578ec47c3b9SKirill A. Shutemov } 4579ec47c3b9SKirill A. Shutemov 45802b740303SSouptick Joarder static vm_fault_t do_shared_fault(struct vm_fault *vmf) 45811da177e4SLinus Torvalds { 458282b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 45832b740303SSouptick Joarder vm_fault_t ret, tmp; 45841d65f86dSKAMEZAWA Hiroyuki 4585936ca80dSJan Kara ret = __do_fault(vmf); 45867eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4587f0c6d4d2SKirill A. Shutemov return ret; 45881da177e4SLinus Torvalds 45891da177e4SLinus Torvalds /* 4590f0c6d4d2SKirill A. Shutemov * Check if the backing address space wants to know that the page is 4591f0c6d4d2SKirill A. Shutemov * about to become writable 45921da177e4SLinus Torvalds */ 4593fb09a464SKirill A. Shutemov if (vma->vm_ops->page_mkwrite) { 4594936ca80dSJan Kara unlock_page(vmf->page); 459538b8cb7fSJan Kara tmp = do_page_mkwrite(vmf); 4596fb09a464SKirill A. Shutemov if (unlikely(!tmp || 4597fb09a464SKirill A. Shutemov (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 4598936ca80dSJan Kara put_page(vmf->page); 4599f0c6d4d2SKirill A. Shutemov return tmp; 460069676147SMark Fasheh } 4601d0217ac0SNick Piggin } 4602fb09a464SKirill A. Shutemov 46039118c0cbSJan Kara ret |= finish_fault(vmf); 46047267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | 46057267ec00SKirill A. Shutemov VM_FAULT_RETRY))) { 4606936ca80dSJan Kara unlock_page(vmf->page); 4607936ca80dSJan Kara put_page(vmf->page); 4608f0c6d4d2SKirill A. Shutemov return ret; 46099637a5efSDavid Howells } 4610d00806b1SNick Piggin 461189b15332SJohannes Weiner ret |= fault_dirty_shared_page(vmf); 4612b827e496SNick Piggin return ret; 461354cb8821SNick Piggin } 4614d00806b1SNick Piggin 46159a95f3cfSPaul Cassella /* 4616c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 46179a95f3cfSPaul Cassella * but allow concurrent faults). 4618c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 46199138e47eSMatthew Wilcox (Oracle) * return value. See filemap_fault() and __folio_lock_or_retry(). 4620c1e8d7c6SMichel Lespinasse * If mmap_lock is released, vma may become invalid (for example 4621fc8efd2dSJan Stancek * by other thread calling munmap()). 46229a95f3cfSPaul Cassella */ 46232b740303SSouptick Joarder static vm_fault_t do_fault(struct vm_fault *vmf) 462454cb8821SNick Piggin { 462582b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4626fc8efd2dSJan Stancek struct mm_struct *vm_mm = vma->vm_mm; 46272b740303SSouptick Joarder vm_fault_t ret; 462854cb8821SNick Piggin 4629ff09d7ecSAneesh Kumar K.V /* 4630ff09d7ecSAneesh Kumar K.V * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND 4631ff09d7ecSAneesh Kumar K.V */ 4632ff09d7ecSAneesh Kumar K.V if (!vma->vm_ops->fault) { 4633ff09d7ecSAneesh Kumar K.V /* 4634ff09d7ecSAneesh Kumar K.V * If we find a migration pmd entry or a none pmd entry, which 4635ff09d7ecSAneesh Kumar K.V * should never happen, return SIGBUS 4636ff09d7ecSAneesh Kumar K.V */ 4637ff09d7ecSAneesh Kumar K.V if (unlikely(!pmd_present(*vmf->pmd))) 4638b0b9b3dfSHugh Dickins ret = VM_FAULT_SIGBUS; 4639ff09d7ecSAneesh Kumar K.V else { 4640ff09d7ecSAneesh Kumar K.V vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, 4641ff09d7ecSAneesh Kumar K.V vmf->pmd, 4642ff09d7ecSAneesh Kumar K.V vmf->address, 4643ff09d7ecSAneesh Kumar K.V &vmf->ptl); 4644ff09d7ecSAneesh Kumar K.V /* 4645ff09d7ecSAneesh Kumar K.V * Make sure this is not a temporary clearing of pte 4646ff09d7ecSAneesh Kumar K.V * by holding ptl and checking again. A R/M/W update 4647ff09d7ecSAneesh Kumar K.V * of pte involves: take ptl, clearing the pte so that 4648ff09d7ecSAneesh Kumar K.V * we don't have concurrent modification by hardware 4649ff09d7ecSAneesh Kumar K.V * followed by an update. 4650ff09d7ecSAneesh Kumar K.V */ 4651ff09d7ecSAneesh Kumar K.V if (unlikely(pte_none(*vmf->pte))) 4652ff09d7ecSAneesh Kumar K.V ret = VM_FAULT_SIGBUS; 4653ff09d7ecSAneesh Kumar K.V else 4654ff09d7ecSAneesh Kumar K.V ret = VM_FAULT_NOPAGE; 4655ff09d7ecSAneesh Kumar K.V 4656ff09d7ecSAneesh Kumar K.V pte_unmap_unlock(vmf->pte, vmf->ptl); 4657ff09d7ecSAneesh Kumar K.V } 4658ff09d7ecSAneesh Kumar K.V } else if (!(vmf->flags & FAULT_FLAG_WRITE)) 4659b0b9b3dfSHugh Dickins ret = do_read_fault(vmf); 4660b0b9b3dfSHugh Dickins else if (!(vma->vm_flags & VM_SHARED)) 4661b0b9b3dfSHugh Dickins ret = do_cow_fault(vmf); 4662b0b9b3dfSHugh Dickins else 4663b0b9b3dfSHugh Dickins ret = do_shared_fault(vmf); 4664b0b9b3dfSHugh Dickins 4665b0b9b3dfSHugh Dickins /* preallocated pagetable is unused: free it */ 4666b0b9b3dfSHugh Dickins if (vmf->prealloc_pte) { 4667fc8efd2dSJan Stancek pte_free(vm_mm, vmf->prealloc_pte); 46687f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 4669b0b9b3dfSHugh Dickins } 4670b0b9b3dfSHugh Dickins return ret; 467154cb8821SNick Piggin } 467254cb8821SNick Piggin 4673f4c0d836SYang Shi int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, 4674f4c0d836SYang Shi unsigned long addr, int page_nid, int *flags) 46759532fec1SMel Gorman { 46769532fec1SMel Gorman get_page(page); 46779532fec1SMel Gorman 4678fc137c0dSRaghavendra K T /* Record the current PID acceesing VMA */ 4679fc137c0dSRaghavendra K T vma_set_access_pid_bit(vma); 4680fc137c0dSRaghavendra K T 46819532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS); 468204bb2f94SRik van Riel if (page_nid == numa_node_id()) { 46839532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); 468404bb2f94SRik van Riel *flags |= TNF_FAULT_LOCAL; 468504bb2f94SRik van Riel } 46869532fec1SMel Gorman 46879532fec1SMel Gorman return mpol_misplaced(page, vma, addr); 46889532fec1SMel Gorman } 46899532fec1SMel Gorman 46902b740303SSouptick Joarder static vm_fault_t do_numa_page(struct vm_fault *vmf) 4691d10e63f2SMel Gorman { 469282b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 46934daae3b4SMel Gorman struct page *page = NULL; 469498fa15f3SAnshuman Khandual int page_nid = NUMA_NO_NODE; 46956a56ccbcSDavid Hildenbrand bool writable = false; 469690572890SPeter Zijlstra int last_cpupid; 4697cbee9f88SPeter Zijlstra int target_nid; 469804a86453SAneesh Kumar K.V pte_t pte, old_pte; 46996688cc05SPeter Zijlstra int flags = 0; 4700d10e63f2SMel Gorman 4701d10e63f2SMel Gorman /* 4702d10e63f2SMel Gorman * The "pte" at this point cannot be used safely without 4703d10e63f2SMel Gorman * validation through pte_unmap_same(). It's of NUMA type but 4704d10e63f2SMel Gorman * the pfn may be screwed if the read is non atomic. 4705d10e63f2SMel Gorman */ 470682b0f8c3SJan Kara vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); 470782b0f8c3SJan Kara spin_lock(vmf->ptl); 4708cee216a6SAneesh Kumar K.V if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { 470982b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 47104daae3b4SMel Gorman goto out; 47114daae3b4SMel Gorman } 47124daae3b4SMel Gorman 4713b99a342dSHuang Ying /* Get the normal PTE */ 4714b99a342dSHuang Ying old_pte = ptep_get(vmf->pte); 471504a86453SAneesh Kumar K.V pte = pte_modify(old_pte, vma->vm_page_prot); 4716d10e63f2SMel Gorman 47176a56ccbcSDavid Hildenbrand /* 47186a56ccbcSDavid Hildenbrand * Detect now whether the PTE could be writable; this information 47196a56ccbcSDavid Hildenbrand * is only valid while holding the PT lock. 47206a56ccbcSDavid Hildenbrand */ 47216a56ccbcSDavid Hildenbrand writable = pte_write(pte); 47226a56ccbcSDavid Hildenbrand if (!writable && vma_wants_manual_pte_write_upgrade(vma) && 47236a56ccbcSDavid Hildenbrand can_change_pte_writable(vma, vmf->address, pte)) 47246a56ccbcSDavid Hildenbrand writable = true; 47256a56ccbcSDavid Hildenbrand 472682b0f8c3SJan Kara page = vm_normal_page(vma, vmf->address, pte); 47273218f871SAlex Sierra if (!page || is_zone_device_page(page)) 4728b99a342dSHuang Ying goto out_map; 4729d10e63f2SMel Gorman 4730e81c4802SKirill A. Shutemov /* TODO: handle PTE-mapped THP */ 4731b99a342dSHuang Ying if (PageCompound(page)) 4732b99a342dSHuang Ying goto out_map; 4733e81c4802SKirill A. Shutemov 47346688cc05SPeter Zijlstra /* 4735bea66fbdSMel Gorman * Avoid grouping on RO pages in general. RO pages shouldn't hurt as 4736bea66fbdSMel Gorman * much anyway since they can be in shared cache state. This misses 4737bea66fbdSMel Gorman * the case where a mapping is writable but the process never writes 4738bea66fbdSMel Gorman * to it but pte_write gets cleared during protection updates and 4739bea66fbdSMel Gorman * pte_dirty has unpredictable behaviour between PTE scan updates, 4740bea66fbdSMel Gorman * background writeback, dirty balancing and application behaviour. 47416688cc05SPeter Zijlstra */ 47426a56ccbcSDavid Hildenbrand if (!writable) 47436688cc05SPeter Zijlstra flags |= TNF_NO_GROUP; 47446688cc05SPeter Zijlstra 4745dabe1d99SRik van Riel /* 4746dabe1d99SRik van Riel * Flag if the page is shared between multiple address spaces. This 4747dabe1d99SRik van Riel * is later used when determining whether to group tasks together 4748dabe1d99SRik van Riel */ 4749dabe1d99SRik van Riel if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) 4750dabe1d99SRik van Riel flags |= TNF_SHARED; 4751dabe1d99SRik van Riel 47528191acbdSMel Gorman page_nid = page_to_nid(page); 475333024536SHuang Ying /* 475433024536SHuang Ying * For memory tiering mode, cpupid of slow memory page is used 475533024536SHuang Ying * to record page access time. So use default value. 475633024536SHuang Ying */ 475733024536SHuang Ying if ((sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) && 475833024536SHuang Ying !node_is_toptier(page_nid)) 475933024536SHuang Ying last_cpupid = (-1 & LAST_CPUPID_MASK); 476033024536SHuang Ying else 476133024536SHuang Ying last_cpupid = page_cpupid_last(page); 476282b0f8c3SJan Kara target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, 4763bae473a4SKirill A. Shutemov &flags); 476498fa15f3SAnshuman Khandual if (target_nid == NUMA_NO_NODE) { 47654daae3b4SMel Gorman put_page(page); 4766b99a342dSHuang Ying goto out_map; 47674daae3b4SMel Gorman } 4768b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 47696a56ccbcSDavid Hildenbrand writable = false; 47704daae3b4SMel Gorman 47714daae3b4SMel Gorman /* Migrate to the requested node */ 4772bf90ac19SWang Qing if (migrate_misplaced_page(page, vma, target_nid)) { 47738191acbdSMel Gorman page_nid = target_nid; 47746688cc05SPeter Zijlstra flags |= TNF_MIGRATED; 4775b99a342dSHuang Ying } else { 4776074c2381SMel Gorman flags |= TNF_MIGRATE_FAIL; 4777b99a342dSHuang Ying vmf->pte = pte_offset_map(vmf->pmd, vmf->address); 4778b99a342dSHuang Ying spin_lock(vmf->ptl); 4779b99a342dSHuang Ying if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { 4780b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 4781b99a342dSHuang Ying goto out; 4782b99a342dSHuang Ying } 4783b99a342dSHuang Ying goto out_map; 4784b99a342dSHuang Ying } 47854daae3b4SMel Gorman 47864daae3b4SMel Gorman out: 478798fa15f3SAnshuman Khandual if (page_nid != NUMA_NO_NODE) 47886688cc05SPeter Zijlstra task_numa_fault(last_cpupid, page_nid, 1, flags); 4789d10e63f2SMel Gorman return 0; 4790b99a342dSHuang Ying out_map: 4791b99a342dSHuang Ying /* 4792b99a342dSHuang Ying * Make it present again, depending on how arch implements 4793b99a342dSHuang Ying * non-accessible ptes, some can allow access by kernel mode. 4794b99a342dSHuang Ying */ 4795b99a342dSHuang Ying old_pte = ptep_modify_prot_start(vma, vmf->address, vmf->pte); 4796b99a342dSHuang Ying pte = pte_modify(old_pte, vma->vm_page_prot); 4797b99a342dSHuang Ying pte = pte_mkyoung(pte); 47986a56ccbcSDavid Hildenbrand if (writable) 4799b99a342dSHuang Ying pte = pte_mkwrite(pte); 4800b99a342dSHuang Ying ptep_modify_prot_commit(vma, vmf->address, vmf->pte, old_pte, pte); 4801b99a342dSHuang Ying update_mmu_cache(vma, vmf->address, vmf->pte); 4802b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 4803b99a342dSHuang Ying goto out; 4804d10e63f2SMel Gorman } 4805d10e63f2SMel Gorman 48062b740303SSouptick Joarder static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf) 4807b96375f7SMatthew Wilcox { 4808f4200391SDave Jiang if (vma_is_anonymous(vmf->vma)) 480982b0f8c3SJan Kara return do_huge_pmd_anonymous_page(vmf); 4810a2d58167SDave Jiang if (vmf->vma->vm_ops->huge_fault) 4811c791ace1SDave Jiang return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); 4812b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 4813b96375f7SMatthew Wilcox } 4814b96375f7SMatthew Wilcox 4815183f24aaSGeert Uytterhoeven /* `inline' is required to avoid gcc 4.1.2 build error */ 48165db4f15cSYang Shi static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf) 4817b96375f7SMatthew Wilcox { 4818c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 4819aea06577SDavid Hildenbrand vm_fault_t ret; 4820c89357e2SDavid Hildenbrand 4821529b930bSAndrea Arcangeli if (vma_is_anonymous(vmf->vma)) { 4822c89357e2SDavid Hildenbrand if (likely(!unshare) && 4823c89357e2SDavid Hildenbrand userfaultfd_huge_pmd_wp(vmf->vma, vmf->orig_pmd)) 4824529b930bSAndrea Arcangeli return handle_userfault(vmf, VM_UFFD_WP); 48255db4f15cSYang Shi return do_huge_pmd_wp_page(vmf); 4826529b930bSAndrea Arcangeli } 4827af9e4d5fSKirill A. Shutemov 4828aea06577SDavid Hildenbrand if (vmf->vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 4829aea06577SDavid Hildenbrand if (vmf->vma->vm_ops->huge_fault) { 4830aea06577SDavid Hildenbrand ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); 4831327e9fd4SThomas Hellstrom (VMware) if (!(ret & VM_FAULT_FALLBACK)) 4832327e9fd4SThomas Hellstrom (VMware) return ret; 4833327e9fd4SThomas Hellstrom (VMware) } 4834aea06577SDavid Hildenbrand } 4835327e9fd4SThomas Hellstrom (VMware) 4836327e9fd4SThomas Hellstrom (VMware) /* COW or write-notify handled on pte level: split pmd. */ 483782b0f8c3SJan Kara __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); 4838af9e4d5fSKirill A. Shutemov 4839b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 4840b96375f7SMatthew Wilcox } 4841b96375f7SMatthew Wilcox 48422b740303SSouptick Joarder static vm_fault_t create_huge_pud(struct vm_fault *vmf) 4843a00cc7d9SMatthew Wilcox { 4844327e9fd4SThomas Hellstrom (VMware) #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ 4845327e9fd4SThomas Hellstrom (VMware) defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) 4846a00cc7d9SMatthew Wilcox /* No support for anonymous transparent PUD pages yet */ 4847a00cc7d9SMatthew Wilcox if (vma_is_anonymous(vmf->vma)) 484814c99d65SGowans, James return VM_FAULT_FALLBACK; 484914c99d65SGowans, James if (vmf->vma->vm_ops->huge_fault) 485014c99d65SGowans, James return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); 485114c99d65SGowans, James #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 485214c99d65SGowans, James return VM_FAULT_FALLBACK; 485314c99d65SGowans, James } 485414c99d65SGowans, James 485514c99d65SGowans, James static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) 485614c99d65SGowans, James { 485714c99d65SGowans, James #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ 485814c99d65SGowans, James defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) 4859aea06577SDavid Hildenbrand vm_fault_t ret; 4860aea06577SDavid Hildenbrand 486114c99d65SGowans, James /* No support for anonymous transparent PUD pages yet */ 486214c99d65SGowans, James if (vma_is_anonymous(vmf->vma)) 4863327e9fd4SThomas Hellstrom (VMware) goto split; 4864aea06577SDavid Hildenbrand if (vmf->vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 4865327e9fd4SThomas Hellstrom (VMware) if (vmf->vma->vm_ops->huge_fault) { 4866aea06577SDavid Hildenbrand ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); 4867327e9fd4SThomas Hellstrom (VMware) if (!(ret & VM_FAULT_FALLBACK)) 4868327e9fd4SThomas Hellstrom (VMware) return ret; 4869327e9fd4SThomas Hellstrom (VMware) } 4870aea06577SDavid Hildenbrand } 4871327e9fd4SThomas Hellstrom (VMware) split: 4872327e9fd4SThomas Hellstrom (VMware) /* COW or write-notify not handled on PUD level: split pud.*/ 4873327e9fd4SThomas Hellstrom (VMware) __split_huge_pud(vmf->vma, vmf->pud, vmf->address); 487414c99d65SGowans, James #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 4875a00cc7d9SMatthew Wilcox return VM_FAULT_FALLBACK; 4876a00cc7d9SMatthew Wilcox } 4877a00cc7d9SMatthew Wilcox 48781da177e4SLinus Torvalds /* 48791da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 48801da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 48811da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 48821da177e4SLinus Torvalds * 48831da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 48841da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 48851da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 48861da177e4SLinus Torvalds * 4887c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, but allow 48887267ec00SKirill A. Shutemov * concurrent faults). 48899a95f3cfSPaul Cassella * 4890c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our return value. 48919138e47eSMatthew Wilcox (Oracle) * See filemap_fault() and __folio_lock_or_retry(). 48921da177e4SLinus Torvalds */ 48932b740303SSouptick Joarder static vm_fault_t handle_pte_fault(struct vm_fault *vmf) 48941da177e4SLinus Torvalds { 48951da177e4SLinus Torvalds pte_t entry; 48961da177e4SLinus Torvalds 489782b0f8c3SJan Kara if (unlikely(pmd_none(*vmf->pmd))) { 48987267ec00SKirill A. Shutemov /* 48997267ec00SKirill A. Shutemov * Leave __pte_alloc() until later: because vm_ops->fault may 49007267ec00SKirill A. Shutemov * want to allocate huge page, and if we expose page table 49017267ec00SKirill A. Shutemov * for an instant, it will be difficult to retract from 49027267ec00SKirill A. Shutemov * concurrent faults and from rmap lookups. 49037267ec00SKirill A. Shutemov */ 490482b0f8c3SJan Kara vmf->pte = NULL; 4905f46f2adeSPeter Xu vmf->flags &= ~FAULT_FLAG_ORIG_PTE_VALID; 49067267ec00SKirill A. Shutemov } else { 4907f9ce0be7SKirill A. Shutemov /* 4908f9ce0be7SKirill A. Shutemov * If a huge pmd materialized under us just retry later. Use 4909f9ce0be7SKirill A. Shutemov * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead 4910f9ce0be7SKirill A. Shutemov * of pmd_trans_huge() to ensure the pmd didn't become 4911f9ce0be7SKirill A. Shutemov * pmd_trans_huge under us and then back to pmd_none, as a 4912f9ce0be7SKirill A. Shutemov * result of MADV_DONTNEED running immediately after a huge pmd 4913f9ce0be7SKirill A. Shutemov * fault in a different thread of this mm, in turn leading to a 4914f9ce0be7SKirill A. Shutemov * misleading pmd_trans_huge() retval. All we have to ensure is 4915f9ce0be7SKirill A. Shutemov * that it is a regular pmd that we can walk with 4916f9ce0be7SKirill A. Shutemov * pte_offset_map() and we can do that through an atomic read 4917f9ce0be7SKirill A. Shutemov * in C, which is what pmd_trans_unstable() provides. 4918f9ce0be7SKirill A. Shutemov */ 4919d0f0931dSRoss Zwisler if (pmd_devmap_trans_unstable(vmf->pmd)) 49207267ec00SKirill A. Shutemov return 0; 49217267ec00SKirill A. Shutemov /* 49227267ec00SKirill A. Shutemov * A regular pmd is established and it can't morph into a huge 49237267ec00SKirill A. Shutemov * pmd from under us anymore at this point because we hold the 4924c1e8d7c6SMichel Lespinasse * mmap_lock read mode and khugepaged takes it in write mode. 49257267ec00SKirill A. Shutemov * So now it's safe to run pte_offset_map(). 49267267ec00SKirill A. Shutemov */ 492782b0f8c3SJan Kara vmf->pte = pte_offset_map(vmf->pmd, vmf->address); 49282994302bSJan Kara vmf->orig_pte = *vmf->pte; 4929f46f2adeSPeter Xu vmf->flags |= FAULT_FLAG_ORIG_PTE_VALID; 49307267ec00SKirill A. Shutemov 4931e37c6982SChristian Borntraeger /* 4932e37c6982SChristian Borntraeger * some architectures can have larger ptes than wordsize, 49337267ec00SKirill A. Shutemov * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and 4934b03a0fe0SPaul E. McKenney * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic 4935b03a0fe0SPaul E. McKenney * accesses. The code below just needs a consistent view 4936b03a0fe0SPaul E. McKenney * for the ifs and we later double check anyway with the 49377267ec00SKirill A. Shutemov * ptl lock held. So here a barrier will do. 4938e37c6982SChristian Borntraeger */ 4939e37c6982SChristian Borntraeger barrier(); 49402994302bSJan Kara if (pte_none(vmf->orig_pte)) { 494182b0f8c3SJan Kara pte_unmap(vmf->pte); 494282b0f8c3SJan Kara vmf->pte = NULL; 49437267ec00SKirill A. Shutemov } 49447267ec00SKirill A. Shutemov } 49457267ec00SKirill A. Shutemov 49462bad466cSPeter Xu if (!vmf->pte) 49472bad466cSPeter Xu return do_pte_missing(vmf); 49487267ec00SKirill A. Shutemov 49492994302bSJan Kara if (!pte_present(vmf->orig_pte)) 49502994302bSJan Kara return do_swap_page(vmf); 49511da177e4SLinus Torvalds 49522994302bSJan Kara if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) 49532994302bSJan Kara return do_numa_page(vmf); 4954d10e63f2SMel Gorman 495582b0f8c3SJan Kara vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); 495682b0f8c3SJan Kara spin_lock(vmf->ptl); 49572994302bSJan Kara entry = vmf->orig_pte; 49587df67697SBibo Mao if (unlikely(!pte_same(*vmf->pte, entry))) { 49597df67697SBibo Mao update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 49608f4e2101SHugh Dickins goto unlock; 49617df67697SBibo Mao } 4962c89357e2SDavid Hildenbrand if (vmf->flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) { 4963f6f37321SLinus Torvalds if (!pte_write(entry)) 49642994302bSJan Kara return do_wp_page(vmf); 4965c89357e2SDavid Hildenbrand else if (likely(vmf->flags & FAULT_FLAG_WRITE)) 49661da177e4SLinus Torvalds entry = pte_mkdirty(entry); 49671da177e4SLinus Torvalds } 49681da177e4SLinus Torvalds entry = pte_mkyoung(entry); 496982b0f8c3SJan Kara if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, 497082b0f8c3SJan Kara vmf->flags & FAULT_FLAG_WRITE)) { 497182b0f8c3SJan Kara update_mmu_cache(vmf->vma, vmf->address, vmf->pte); 49721a44e149SAndrea Arcangeli } else { 4973b7333b58SYang Shi /* Skip spurious TLB flush for retried page fault */ 4974b7333b58SYang Shi if (vmf->flags & FAULT_FLAG_TRIED) 4975b7333b58SYang Shi goto unlock; 49761a44e149SAndrea Arcangeli /* 49771a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 49781a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 49791a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 49801a44e149SAndrea Arcangeli * with threads. 49811a44e149SAndrea Arcangeli */ 498282b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) 498399c29133SGerald Schaefer flush_tlb_fix_spurious_fault(vmf->vma, vmf->address, 498499c29133SGerald Schaefer vmf->pte); 49851a44e149SAndrea Arcangeli } 49868f4e2101SHugh Dickins unlock: 498782b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 498883c54070SNick Piggin return 0; 49891da177e4SLinus Torvalds } 49901da177e4SLinus Torvalds 49911da177e4SLinus Torvalds /* 49921da177e4SLinus Torvalds * By the time we get here, we already hold the mm semaphore 49939a95f3cfSPaul Cassella * 4994c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 49959138e47eSMatthew Wilcox (Oracle) * return value. See filemap_fault() and __folio_lock_or_retry(). 49961da177e4SLinus Torvalds */ 49972b740303SSouptick Joarder static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma, 49982b740303SSouptick Joarder unsigned long address, unsigned int flags) 49991da177e4SLinus Torvalds { 500082b0f8c3SJan Kara struct vm_fault vmf = { 5001bae473a4SKirill A. Shutemov .vma = vma, 50021a29d85eSJan Kara .address = address & PAGE_MASK, 5003824ddc60SNadav Amit .real_address = address, 5004bae473a4SKirill A. Shutemov .flags = flags, 50050721ec8bSJan Kara .pgoff = linear_page_index(vma, address), 5006667240e0SJan Kara .gfp_mask = __get_fault_gfp_mask(vma), 5007bae473a4SKirill A. Shutemov }; 5008dcddffd4SKirill A. Shutemov struct mm_struct *mm = vma->vm_mm; 50097da4e2cbSYang Shi unsigned long vm_flags = vma->vm_flags; 50101da177e4SLinus Torvalds pgd_t *pgd; 5011c2febafcSKirill A. Shutemov p4d_t *p4d; 50122b740303SSouptick Joarder vm_fault_t ret; 50131da177e4SLinus Torvalds 50141da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 5015c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, address); 5016c2febafcSKirill A. Shutemov if (!p4d) 5017c2febafcSKirill A. Shutemov return VM_FAULT_OOM; 5018a00cc7d9SMatthew Wilcox 5019c2febafcSKirill A. Shutemov vmf.pud = pud_alloc(mm, p4d, address); 5020a00cc7d9SMatthew Wilcox if (!vmf.pud) 5021c74df32cSHugh Dickins return VM_FAULT_OOM; 5022625110b5SThomas Hellstrom retry_pud: 50237da4e2cbSYang Shi if (pud_none(*vmf.pud) && 5024a7f4e6e4SZach O'Keefe hugepage_vma_check(vma, vm_flags, false, true, true)) { 5025a00cc7d9SMatthew Wilcox ret = create_huge_pud(&vmf); 5026a00cc7d9SMatthew Wilcox if (!(ret & VM_FAULT_FALLBACK)) 5027a00cc7d9SMatthew Wilcox return ret; 5028a00cc7d9SMatthew Wilcox } else { 5029a00cc7d9SMatthew Wilcox pud_t orig_pud = *vmf.pud; 5030a00cc7d9SMatthew Wilcox 5031a00cc7d9SMatthew Wilcox barrier(); 5032a00cc7d9SMatthew Wilcox if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { 5033a00cc7d9SMatthew Wilcox 5034c89357e2SDavid Hildenbrand /* 5035c89357e2SDavid Hildenbrand * TODO once we support anonymous PUDs: NUMA case and 5036c89357e2SDavid Hildenbrand * FAULT_FLAG_UNSHARE handling. 5037c89357e2SDavid Hildenbrand */ 5038c89357e2SDavid Hildenbrand if ((flags & FAULT_FLAG_WRITE) && !pud_write(orig_pud)) { 5039a00cc7d9SMatthew Wilcox ret = wp_huge_pud(&vmf, orig_pud); 5040a00cc7d9SMatthew Wilcox if (!(ret & VM_FAULT_FALLBACK)) 5041a00cc7d9SMatthew Wilcox return ret; 5042a00cc7d9SMatthew Wilcox } else { 5043a00cc7d9SMatthew Wilcox huge_pud_set_accessed(&vmf, orig_pud); 5044a00cc7d9SMatthew Wilcox return 0; 5045a00cc7d9SMatthew Wilcox } 5046a00cc7d9SMatthew Wilcox } 5047a00cc7d9SMatthew Wilcox } 5048a00cc7d9SMatthew Wilcox 5049a00cc7d9SMatthew Wilcox vmf.pmd = pmd_alloc(mm, vmf.pud, address); 505082b0f8c3SJan Kara if (!vmf.pmd) 5051c74df32cSHugh Dickins return VM_FAULT_OOM; 5052625110b5SThomas Hellstrom 5053625110b5SThomas Hellstrom /* Huge pud page fault raced with pmd_alloc? */ 5054625110b5SThomas Hellstrom if (pud_trans_unstable(vmf.pud)) 5055625110b5SThomas Hellstrom goto retry_pud; 5056625110b5SThomas Hellstrom 50577da4e2cbSYang Shi if (pmd_none(*vmf.pmd) && 5058a7f4e6e4SZach O'Keefe hugepage_vma_check(vma, vm_flags, false, true, true)) { 5059a2d58167SDave Jiang ret = create_huge_pmd(&vmf); 5060c0292554SKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 5061c0292554SKirill A. Shutemov return ret; 506271e3aac0SAndrea Arcangeli } else { 50635db4f15cSYang Shi vmf.orig_pmd = *vmf.pmd; 50641f1d06c3SDavid Rientjes 506571e3aac0SAndrea Arcangeli barrier(); 50665db4f15cSYang Shi if (unlikely(is_swap_pmd(vmf.orig_pmd))) { 506784c3fc4eSZi Yan VM_BUG_ON(thp_migration_supported() && 50685db4f15cSYang Shi !is_pmd_migration_entry(vmf.orig_pmd)); 50695db4f15cSYang Shi if (is_pmd_migration_entry(vmf.orig_pmd)) 507084c3fc4eSZi Yan pmd_migration_entry_wait(mm, vmf.pmd); 507184c3fc4eSZi Yan return 0; 507284c3fc4eSZi Yan } 50735db4f15cSYang Shi if (pmd_trans_huge(vmf.orig_pmd) || pmd_devmap(vmf.orig_pmd)) { 50745db4f15cSYang Shi if (pmd_protnone(vmf.orig_pmd) && vma_is_accessible(vma)) 50755db4f15cSYang Shi return do_huge_pmd_numa_page(&vmf); 5076d10e63f2SMel Gorman 5077c89357e2SDavid Hildenbrand if ((flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) && 5078c89357e2SDavid Hildenbrand !pmd_write(vmf.orig_pmd)) { 50795db4f15cSYang Shi ret = wp_huge_pmd(&vmf); 50809845cbbdSKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 50811f1d06c3SDavid Rientjes return ret; 5082a1dd450bSWill Deacon } else { 50835db4f15cSYang Shi huge_pmd_set_accessed(&vmf); 508471e3aac0SAndrea Arcangeli return 0; 508571e3aac0SAndrea Arcangeli } 508671e3aac0SAndrea Arcangeli } 50879845cbbdSKirill A. Shutemov } 508871e3aac0SAndrea Arcangeli 508982b0f8c3SJan Kara return handle_pte_fault(&vmf); 50901da177e4SLinus Torvalds } 50911da177e4SLinus Torvalds 5092bce617edSPeter Xu /** 5093f0953a1bSIngo Molnar * mm_account_fault - Do page fault accounting 5094bce617edSPeter Xu * 5095bce617edSPeter Xu * @regs: the pt_regs struct pointer. When set to NULL, will skip accounting 5096bce617edSPeter Xu * of perf event counters, but we'll still do the per-task accounting to 5097bce617edSPeter Xu * the task who triggered this page fault. 5098bce617edSPeter Xu * @address: the faulted address. 5099bce617edSPeter Xu * @flags: the fault flags. 5100bce617edSPeter Xu * @ret: the fault retcode. 5101bce617edSPeter Xu * 5102f0953a1bSIngo Molnar * This will take care of most of the page fault accounting. Meanwhile, it 5103bce617edSPeter Xu * will also include the PERF_COUNT_SW_PAGE_FAULTS_[MAJ|MIN] perf counter 5104f0953a1bSIngo Molnar * updates. However, note that the handling of PERF_COUNT_SW_PAGE_FAULTS should 5105bce617edSPeter Xu * still be in per-arch page fault handlers at the entry of page fault. 5106bce617edSPeter Xu */ 510753156443SSuren Baghdasaryan static inline void mm_account_fault(struct mm_struct *mm, struct pt_regs *regs, 5108bce617edSPeter Xu unsigned long address, unsigned int flags, 5109bce617edSPeter Xu vm_fault_t ret) 5110bce617edSPeter Xu { 5111bce617edSPeter Xu bool major; 5112bce617edSPeter Xu 511353156443SSuren Baghdasaryan /* Incomplete faults will be accounted upon completion. */ 511453156443SSuren Baghdasaryan if (ret & VM_FAULT_RETRY) 511553156443SSuren Baghdasaryan return; 511653156443SSuren Baghdasaryan 5117bce617edSPeter Xu /* 511853156443SSuren Baghdasaryan * To preserve the behavior of older kernels, PGFAULT counters record 511953156443SSuren Baghdasaryan * both successful and failed faults, as opposed to perf counters, 512053156443SSuren Baghdasaryan * which ignore failed cases. 5121bce617edSPeter Xu */ 512253156443SSuren Baghdasaryan count_vm_event(PGFAULT); 512353156443SSuren Baghdasaryan count_memcg_event_mm(mm, PGFAULT); 512453156443SSuren Baghdasaryan 512553156443SSuren Baghdasaryan /* 512653156443SSuren Baghdasaryan * Do not account for unsuccessful faults (e.g. when the address wasn't 512753156443SSuren Baghdasaryan * valid). That includes arch_vma_access_permitted() failing before 512853156443SSuren Baghdasaryan * reaching here. So this is not a "this many hardware page faults" 512953156443SSuren Baghdasaryan * counter. We should use the hw profiling for that. 513053156443SSuren Baghdasaryan */ 513153156443SSuren Baghdasaryan if (ret & VM_FAULT_ERROR) 5132bce617edSPeter Xu return; 5133bce617edSPeter Xu 5134bce617edSPeter Xu /* 5135bce617edSPeter Xu * We define the fault as a major fault when the final successful fault 5136bce617edSPeter Xu * is VM_FAULT_MAJOR, or if it retried (which implies that we couldn't 5137bce617edSPeter Xu * handle it immediately previously). 5138bce617edSPeter Xu */ 5139bce617edSPeter Xu major = (ret & VM_FAULT_MAJOR) || (flags & FAULT_FLAG_TRIED); 5140bce617edSPeter Xu 5141a2beb5f1SPeter Xu if (major) 5142a2beb5f1SPeter Xu current->maj_flt++; 5143a2beb5f1SPeter Xu else 5144a2beb5f1SPeter Xu current->min_flt++; 5145a2beb5f1SPeter Xu 5146bce617edSPeter Xu /* 5147a2beb5f1SPeter Xu * If the fault is done for GUP, regs will be NULL. We only do the 5148a2beb5f1SPeter Xu * accounting for the per thread fault counters who triggered the 5149a2beb5f1SPeter Xu * fault, and we skip the perf event updates. 5150bce617edSPeter Xu */ 5151bce617edSPeter Xu if (!regs) 5152bce617edSPeter Xu return; 5153bce617edSPeter Xu 5154a2beb5f1SPeter Xu if (major) 5155bce617edSPeter Xu perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 5156a2beb5f1SPeter Xu else 5157bce617edSPeter Xu perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 5158bce617edSPeter Xu } 5159bce617edSPeter Xu 5160ec1c86b2SYu Zhao #ifdef CONFIG_LRU_GEN 5161ec1c86b2SYu Zhao static void lru_gen_enter_fault(struct vm_area_struct *vma) 5162ec1c86b2SYu Zhao { 51638788f678SYu Zhao /* the LRU algorithm only applies to accesses with recency */ 51648788f678SYu Zhao current->in_lru_fault = vma_has_recency(vma); 5165ec1c86b2SYu Zhao } 5166ec1c86b2SYu Zhao 5167ec1c86b2SYu Zhao static void lru_gen_exit_fault(void) 5168ec1c86b2SYu Zhao { 5169ec1c86b2SYu Zhao current->in_lru_fault = false; 5170ec1c86b2SYu Zhao } 5171ec1c86b2SYu Zhao #else 5172ec1c86b2SYu Zhao static void lru_gen_enter_fault(struct vm_area_struct *vma) 5173ec1c86b2SYu Zhao { 5174ec1c86b2SYu Zhao } 5175ec1c86b2SYu Zhao 5176ec1c86b2SYu Zhao static void lru_gen_exit_fault(void) 5177ec1c86b2SYu Zhao { 5178ec1c86b2SYu Zhao } 5179ec1c86b2SYu Zhao #endif /* CONFIG_LRU_GEN */ 5180ec1c86b2SYu Zhao 5181cdc5021cSDavid Hildenbrand static vm_fault_t sanitize_fault_flags(struct vm_area_struct *vma, 5182cdc5021cSDavid Hildenbrand unsigned int *flags) 5183cdc5021cSDavid Hildenbrand { 5184cdc5021cSDavid Hildenbrand if (unlikely(*flags & FAULT_FLAG_UNSHARE)) { 5185cdc5021cSDavid Hildenbrand if (WARN_ON_ONCE(*flags & FAULT_FLAG_WRITE)) 5186cdc5021cSDavid Hildenbrand return VM_FAULT_SIGSEGV; 5187cdc5021cSDavid Hildenbrand /* 5188cdc5021cSDavid Hildenbrand * FAULT_FLAG_UNSHARE only applies to COW mappings. Let's 5189cdc5021cSDavid Hildenbrand * just treat it like an ordinary read-fault otherwise. 5190cdc5021cSDavid Hildenbrand */ 5191cdc5021cSDavid Hildenbrand if (!is_cow_mapping(vma->vm_flags)) 5192cdc5021cSDavid Hildenbrand *flags &= ~FAULT_FLAG_UNSHARE; 519379881fedSDavid Hildenbrand } else if (*flags & FAULT_FLAG_WRITE) { 519479881fedSDavid Hildenbrand /* Write faults on read-only mappings are impossible ... */ 519579881fedSDavid Hildenbrand if (WARN_ON_ONCE(!(vma->vm_flags & VM_MAYWRITE))) 519679881fedSDavid Hildenbrand return VM_FAULT_SIGSEGV; 519779881fedSDavid Hildenbrand /* ... and FOLL_FORCE only applies to COW mappings. */ 519879881fedSDavid Hildenbrand if (WARN_ON_ONCE(!(vma->vm_flags & VM_WRITE) && 519979881fedSDavid Hildenbrand !is_cow_mapping(vma->vm_flags))) 520079881fedSDavid Hildenbrand return VM_FAULT_SIGSEGV; 5201cdc5021cSDavid Hildenbrand } 5202cdc5021cSDavid Hildenbrand return 0; 5203cdc5021cSDavid Hildenbrand } 5204cdc5021cSDavid Hildenbrand 52059a95f3cfSPaul Cassella /* 52069a95f3cfSPaul Cassella * By the time we get here, we already hold the mm semaphore 52079a95f3cfSPaul Cassella * 5208c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 52099138e47eSMatthew Wilcox (Oracle) * return value. See filemap_fault() and __folio_lock_or_retry(). 52109a95f3cfSPaul Cassella */ 52112b740303SSouptick Joarder vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address, 5212bce617edSPeter Xu unsigned int flags, struct pt_regs *regs) 5213519e5247SJohannes Weiner { 521453156443SSuren Baghdasaryan /* If the fault handler drops the mmap_lock, vma may be freed */ 521553156443SSuren Baghdasaryan struct mm_struct *mm = vma->vm_mm; 52162b740303SSouptick Joarder vm_fault_t ret; 5217519e5247SJohannes Weiner 5218519e5247SJohannes Weiner __set_current_state(TASK_RUNNING); 5219519e5247SJohannes Weiner 5220cdc5021cSDavid Hildenbrand ret = sanitize_fault_flags(vma, &flags); 5221cdc5021cSDavid Hildenbrand if (ret) 522253156443SSuren Baghdasaryan goto out; 5223cdc5021cSDavid Hildenbrand 5224de0c799bSLaurent Dufour if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, 5225de0c799bSLaurent Dufour flags & FAULT_FLAG_INSTRUCTION, 522653156443SSuren Baghdasaryan flags & FAULT_FLAG_REMOTE)) { 522753156443SSuren Baghdasaryan ret = VM_FAULT_SIGSEGV; 522853156443SSuren Baghdasaryan goto out; 522953156443SSuren Baghdasaryan } 5230de0c799bSLaurent Dufour 5231519e5247SJohannes Weiner /* 5232519e5247SJohannes Weiner * Enable the memcg OOM handling for faults triggered in user 5233519e5247SJohannes Weiner * space. Kernel faults are handled more gracefully. 5234519e5247SJohannes Weiner */ 5235519e5247SJohannes Weiner if (flags & FAULT_FLAG_USER) 523629ef680aSMichal Hocko mem_cgroup_enter_user_fault(); 5237519e5247SJohannes Weiner 5238ec1c86b2SYu Zhao lru_gen_enter_fault(vma); 5239ec1c86b2SYu Zhao 5240bae473a4SKirill A. Shutemov if (unlikely(is_vm_hugetlb_page(vma))) 5241bae473a4SKirill A. Shutemov ret = hugetlb_fault(vma->vm_mm, vma, address, flags); 5242bae473a4SKirill A. Shutemov else 5243dcddffd4SKirill A. Shutemov ret = __handle_mm_fault(vma, address, flags); 5244519e5247SJohannes Weiner 5245ec1c86b2SYu Zhao lru_gen_exit_fault(); 5246ec1c86b2SYu Zhao 524749426420SJohannes Weiner if (flags & FAULT_FLAG_USER) { 524829ef680aSMichal Hocko mem_cgroup_exit_user_fault(); 524949426420SJohannes Weiner /* 525049426420SJohannes Weiner * The task may have entered a memcg OOM situation but 525149426420SJohannes Weiner * if the allocation error was handled gracefully (no 525249426420SJohannes Weiner * VM_FAULT_OOM), there is no need to kill anything. 525349426420SJohannes Weiner * Just clean up the OOM state peacefully. 525449426420SJohannes Weiner */ 525549426420SJohannes Weiner if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) 525649426420SJohannes Weiner mem_cgroup_oom_synchronize(false); 525749426420SJohannes Weiner } 525853156443SSuren Baghdasaryan out: 525953156443SSuren Baghdasaryan mm_account_fault(mm, regs, address, flags, ret); 5260bce617edSPeter Xu 5261519e5247SJohannes Weiner return ret; 5262519e5247SJohannes Weiner } 5263e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(handle_mm_fault); 5264519e5247SJohannes Weiner 5265c2508ec5SLinus Torvalds #ifdef CONFIG_LOCK_MM_AND_FIND_VMA 5266c2508ec5SLinus Torvalds #include <linux/extable.h> 5267c2508ec5SLinus Torvalds 5268c2508ec5SLinus Torvalds static inline bool get_mmap_lock_carefully(struct mm_struct *mm, struct pt_regs *regs) 5269c2508ec5SLinus Torvalds { 5270c2508ec5SLinus Torvalds /* Even if this succeeds, make it clear we *might* have slept */ 5271c2508ec5SLinus Torvalds if (likely(mmap_read_trylock(mm))) { 5272c2508ec5SLinus Torvalds might_sleep(); 5273c2508ec5SLinus Torvalds return true; 5274c2508ec5SLinus Torvalds } 5275c2508ec5SLinus Torvalds 5276c2508ec5SLinus Torvalds if (regs && !user_mode(regs)) { 5277c2508ec5SLinus Torvalds unsigned long ip = instruction_pointer(regs); 5278c2508ec5SLinus Torvalds if (!search_exception_tables(ip)) 5279c2508ec5SLinus Torvalds return false; 5280c2508ec5SLinus Torvalds } 5281c2508ec5SLinus Torvalds 5282*eda00472SLinus Torvalds return !mmap_read_lock_killable(mm); 5283c2508ec5SLinus Torvalds } 5284c2508ec5SLinus Torvalds 5285c2508ec5SLinus Torvalds static inline bool mmap_upgrade_trylock(struct mm_struct *mm) 5286c2508ec5SLinus Torvalds { 5287c2508ec5SLinus Torvalds /* 5288c2508ec5SLinus Torvalds * We don't have this operation yet. 5289c2508ec5SLinus Torvalds * 5290c2508ec5SLinus Torvalds * It should be easy enough to do: it's basically a 5291c2508ec5SLinus Torvalds * atomic_long_try_cmpxchg_acquire() 5292c2508ec5SLinus Torvalds * from RWSEM_READER_BIAS -> RWSEM_WRITER_LOCKED, but 5293c2508ec5SLinus Torvalds * it also needs the proper lockdep magic etc. 5294c2508ec5SLinus Torvalds */ 5295c2508ec5SLinus Torvalds return false; 5296c2508ec5SLinus Torvalds } 5297c2508ec5SLinus Torvalds 5298c2508ec5SLinus Torvalds static inline bool upgrade_mmap_lock_carefully(struct mm_struct *mm, struct pt_regs *regs) 5299c2508ec5SLinus Torvalds { 5300c2508ec5SLinus Torvalds mmap_read_unlock(mm); 5301c2508ec5SLinus Torvalds if (regs && !user_mode(regs)) { 5302c2508ec5SLinus Torvalds unsigned long ip = instruction_pointer(regs); 5303c2508ec5SLinus Torvalds if (!search_exception_tables(ip)) 5304c2508ec5SLinus Torvalds return false; 5305c2508ec5SLinus Torvalds } 5306*eda00472SLinus Torvalds return !mmap_write_lock_killable(mm); 5307c2508ec5SLinus Torvalds } 5308c2508ec5SLinus Torvalds 5309c2508ec5SLinus Torvalds /* 5310c2508ec5SLinus Torvalds * Helper for page fault handling. 5311c2508ec5SLinus Torvalds * 5312c2508ec5SLinus Torvalds * This is kind of equivalend to "mmap_read_lock()" followed 5313c2508ec5SLinus Torvalds * by "find_extend_vma()", except it's a lot more careful about 5314c2508ec5SLinus Torvalds * the locking (and will drop the lock on failure). 5315c2508ec5SLinus Torvalds * 5316c2508ec5SLinus Torvalds * For example, if we have a kernel bug that causes a page 5317c2508ec5SLinus Torvalds * fault, we don't want to just use mmap_read_lock() to get 5318c2508ec5SLinus Torvalds * the mm lock, because that would deadlock if the bug were 5319c2508ec5SLinus Torvalds * to happen while we're holding the mm lock for writing. 5320c2508ec5SLinus Torvalds * 5321c2508ec5SLinus Torvalds * So this checks the exception tables on kernel faults in 5322c2508ec5SLinus Torvalds * order to only do this all for instructions that are actually 5323c2508ec5SLinus Torvalds * expected to fault. 5324c2508ec5SLinus Torvalds * 5325c2508ec5SLinus Torvalds * We can also actually take the mm lock for writing if we 5326c2508ec5SLinus Torvalds * need to extend the vma, which helps the VM layer a lot. 5327c2508ec5SLinus Torvalds */ 5328c2508ec5SLinus Torvalds struct vm_area_struct *lock_mm_and_find_vma(struct mm_struct *mm, 5329c2508ec5SLinus Torvalds unsigned long addr, struct pt_regs *regs) 5330c2508ec5SLinus Torvalds { 5331c2508ec5SLinus Torvalds struct vm_area_struct *vma; 5332c2508ec5SLinus Torvalds 5333c2508ec5SLinus Torvalds if (!get_mmap_lock_carefully(mm, regs)) 5334c2508ec5SLinus Torvalds return NULL; 5335c2508ec5SLinus Torvalds 5336c2508ec5SLinus Torvalds vma = find_vma(mm, addr); 5337c2508ec5SLinus Torvalds if (likely(vma && (vma->vm_start <= addr))) 5338c2508ec5SLinus Torvalds return vma; 5339c2508ec5SLinus Torvalds 5340c2508ec5SLinus Torvalds /* 5341c2508ec5SLinus Torvalds * Well, dang. We might still be successful, but only 5342c2508ec5SLinus Torvalds * if we can extend a vma to do so. 5343c2508ec5SLinus Torvalds */ 5344c2508ec5SLinus Torvalds if (!vma || !(vma->vm_flags & VM_GROWSDOWN)) { 5345c2508ec5SLinus Torvalds mmap_read_unlock(mm); 5346c2508ec5SLinus Torvalds return NULL; 5347c2508ec5SLinus Torvalds } 5348c2508ec5SLinus Torvalds 5349c2508ec5SLinus Torvalds /* 5350c2508ec5SLinus Torvalds * We can try to upgrade the mmap lock atomically, 5351c2508ec5SLinus Torvalds * in which case we can continue to use the vma 5352c2508ec5SLinus Torvalds * we already looked up. 5353c2508ec5SLinus Torvalds * 5354c2508ec5SLinus Torvalds * Otherwise we'll have to drop the mmap lock and 5355c2508ec5SLinus Torvalds * re-take it, and also look up the vma again, 5356c2508ec5SLinus Torvalds * re-checking it. 5357c2508ec5SLinus Torvalds */ 5358c2508ec5SLinus Torvalds if (!mmap_upgrade_trylock(mm)) { 5359c2508ec5SLinus Torvalds if (!upgrade_mmap_lock_carefully(mm, regs)) 5360c2508ec5SLinus Torvalds return NULL; 5361c2508ec5SLinus Torvalds 5362c2508ec5SLinus Torvalds vma = find_vma(mm, addr); 5363c2508ec5SLinus Torvalds if (!vma) 5364c2508ec5SLinus Torvalds goto fail; 5365c2508ec5SLinus Torvalds if (vma->vm_start <= addr) 5366c2508ec5SLinus Torvalds goto success; 5367c2508ec5SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 5368c2508ec5SLinus Torvalds goto fail; 5369c2508ec5SLinus Torvalds } 5370c2508ec5SLinus Torvalds 5371c2508ec5SLinus Torvalds if (expand_stack(vma, addr)) 5372c2508ec5SLinus Torvalds goto fail; 5373c2508ec5SLinus Torvalds 5374c2508ec5SLinus Torvalds success: 5375c2508ec5SLinus Torvalds mmap_write_downgrade(mm); 5376c2508ec5SLinus Torvalds return vma; 5377c2508ec5SLinus Torvalds 5378c2508ec5SLinus Torvalds fail: 5379c2508ec5SLinus Torvalds mmap_write_unlock(mm); 5380c2508ec5SLinus Torvalds return NULL; 5381c2508ec5SLinus Torvalds } 5382c2508ec5SLinus Torvalds #endif 5383c2508ec5SLinus Torvalds 538450ee3253SSuren Baghdasaryan #ifdef CONFIG_PER_VMA_LOCK 538550ee3253SSuren Baghdasaryan /* 538650ee3253SSuren Baghdasaryan * Lookup and lock a VMA under RCU protection. Returned VMA is guaranteed to be 538750ee3253SSuren Baghdasaryan * stable and not isolated. If the VMA is not found or is being modified the 538850ee3253SSuren Baghdasaryan * function returns NULL. 538950ee3253SSuren Baghdasaryan */ 539050ee3253SSuren Baghdasaryan struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm, 539150ee3253SSuren Baghdasaryan unsigned long address) 539250ee3253SSuren Baghdasaryan { 539350ee3253SSuren Baghdasaryan MA_STATE(mas, &mm->mm_mt, address, address); 539450ee3253SSuren Baghdasaryan struct vm_area_struct *vma; 539550ee3253SSuren Baghdasaryan 539650ee3253SSuren Baghdasaryan rcu_read_lock(); 539750ee3253SSuren Baghdasaryan retry: 539850ee3253SSuren Baghdasaryan vma = mas_walk(&mas); 539950ee3253SSuren Baghdasaryan if (!vma) 540050ee3253SSuren Baghdasaryan goto inval; 540150ee3253SSuren Baghdasaryan 540250ee3253SSuren Baghdasaryan /* Only anonymous vmas are supported for now */ 540350ee3253SSuren Baghdasaryan if (!vma_is_anonymous(vma)) 540450ee3253SSuren Baghdasaryan goto inval; 540550ee3253SSuren Baghdasaryan 54062ac0af1bSSuren Baghdasaryan /* find_mergeable_anon_vma uses adjacent vmas which are not locked */ 54072ac0af1bSSuren Baghdasaryan if (!vma->anon_vma) 54082ac0af1bSSuren Baghdasaryan goto inval; 54092ac0af1bSSuren Baghdasaryan 541050ee3253SSuren Baghdasaryan if (!vma_start_read(vma)) 541150ee3253SSuren Baghdasaryan goto inval; 541250ee3253SSuren Baghdasaryan 5413444eeb17SSuren Baghdasaryan /* 5414444eeb17SSuren Baghdasaryan * Due to the possibility of userfault handler dropping mmap_lock, avoid 5415444eeb17SSuren Baghdasaryan * it for now and fall back to page fault handling under mmap_lock. 5416444eeb17SSuren Baghdasaryan */ 5417444eeb17SSuren Baghdasaryan if (userfaultfd_armed(vma)) { 5418444eeb17SSuren Baghdasaryan vma_end_read(vma); 5419444eeb17SSuren Baghdasaryan goto inval; 5420444eeb17SSuren Baghdasaryan } 5421444eeb17SSuren Baghdasaryan 542250ee3253SSuren Baghdasaryan /* Check since vm_start/vm_end might change before we lock the VMA */ 542350ee3253SSuren Baghdasaryan if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { 542450ee3253SSuren Baghdasaryan vma_end_read(vma); 542550ee3253SSuren Baghdasaryan goto inval; 542650ee3253SSuren Baghdasaryan } 542750ee3253SSuren Baghdasaryan 542850ee3253SSuren Baghdasaryan /* Check if the VMA got isolated after we found it */ 542950ee3253SSuren Baghdasaryan if (vma->detached) { 543050ee3253SSuren Baghdasaryan vma_end_read(vma); 543152f23865SSuren Baghdasaryan count_vm_vma_lock_event(VMA_LOCK_MISS); 543250ee3253SSuren Baghdasaryan /* The area was replaced with another one */ 543350ee3253SSuren Baghdasaryan goto retry; 543450ee3253SSuren Baghdasaryan } 543550ee3253SSuren Baghdasaryan 543650ee3253SSuren Baghdasaryan rcu_read_unlock(); 543750ee3253SSuren Baghdasaryan return vma; 543850ee3253SSuren Baghdasaryan inval: 543950ee3253SSuren Baghdasaryan rcu_read_unlock(); 544052f23865SSuren Baghdasaryan count_vm_vma_lock_event(VMA_LOCK_ABORT); 544150ee3253SSuren Baghdasaryan return NULL; 544250ee3253SSuren Baghdasaryan } 544350ee3253SSuren Baghdasaryan #endif /* CONFIG_PER_VMA_LOCK */ 544450ee3253SSuren Baghdasaryan 544590eceff1SKirill A. Shutemov #ifndef __PAGETABLE_P4D_FOLDED 544690eceff1SKirill A. Shutemov /* 544790eceff1SKirill A. Shutemov * Allocate p4d page table. 544890eceff1SKirill A. Shutemov * We've already handled the fast-path in-line. 544990eceff1SKirill A. Shutemov */ 545090eceff1SKirill A. Shutemov int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 545190eceff1SKirill A. Shutemov { 545290eceff1SKirill A. Shutemov p4d_t *new = p4d_alloc_one(mm, address); 545390eceff1SKirill A. Shutemov if (!new) 545490eceff1SKirill A. Shutemov return -ENOMEM; 545590eceff1SKirill A. Shutemov 545690eceff1SKirill A. Shutemov spin_lock(&mm->page_table_lock); 5457ed33b5a6SQi Zheng if (pgd_present(*pgd)) { /* Another has populated it */ 545890eceff1SKirill A. Shutemov p4d_free(mm, new); 5459ed33b5a6SQi Zheng } else { 5460ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 546190eceff1SKirill A. Shutemov pgd_populate(mm, pgd, new); 5462ed33b5a6SQi Zheng } 546390eceff1SKirill A. Shutemov spin_unlock(&mm->page_table_lock); 546490eceff1SKirill A. Shutemov return 0; 546590eceff1SKirill A. Shutemov } 546690eceff1SKirill A. Shutemov #endif /* __PAGETABLE_P4D_FOLDED */ 546790eceff1SKirill A. Shutemov 54681da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 54691da177e4SLinus Torvalds /* 54701da177e4SLinus Torvalds * Allocate page upper directory. 5471872fec16SHugh Dickins * We've already handled the fast-path in-line. 54721da177e4SLinus Torvalds */ 5473c2febafcSKirill A. Shutemov int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) 54741da177e4SLinus Torvalds { 5475c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 5476c74df32cSHugh Dickins if (!new) 54771bb3630eSHugh Dickins return -ENOMEM; 54781da177e4SLinus Torvalds 5479872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 5480b4e98d9aSKirill A. Shutemov if (!p4d_present(*p4d)) { 5481b4e98d9aSKirill A. Shutemov mm_inc_nr_puds(mm); 5482ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 5483c2febafcSKirill A. Shutemov p4d_populate(mm, p4d, new); 5484b4e98d9aSKirill A. Shutemov } else /* Another has populated it */ 5485c2febafcSKirill A. Shutemov pud_free(mm, new); 5486872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 54871bb3630eSHugh Dickins return 0; 54881da177e4SLinus Torvalds } 54891da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 54901da177e4SLinus Torvalds 54911da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 54921da177e4SLinus Torvalds /* 54931da177e4SLinus Torvalds * Allocate page middle directory. 5494872fec16SHugh Dickins * We've already handled the fast-path in-line. 54951da177e4SLinus Torvalds */ 54961bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 54971da177e4SLinus Torvalds { 5498a00cc7d9SMatthew Wilcox spinlock_t *ptl; 5499c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 5500c74df32cSHugh Dickins if (!new) 55011bb3630eSHugh Dickins return -ENOMEM; 55021da177e4SLinus Torvalds 5503a00cc7d9SMatthew Wilcox ptl = pud_lock(mm, pud); 5504dc6c9a35SKirill A. Shutemov if (!pud_present(*pud)) { 5505dc6c9a35SKirill A. Shutemov mm_inc_nr_pmds(mm); 5506ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 55071da177e4SLinus Torvalds pud_populate(mm, pud, new); 5508ed33b5a6SQi Zheng } else { /* Another has populated it */ 55095e541973SBenjamin Herrenschmidt pmd_free(mm, new); 5510ed33b5a6SQi Zheng } 5511a00cc7d9SMatthew Wilcox spin_unlock(ptl); 55121bb3630eSHugh Dickins return 0; 55131da177e4SLinus Torvalds } 55141da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 55151da177e4SLinus Torvalds 55163b6748e2SJohannes Weiner /** 55179fd6dad1SPaolo Bonzini * follow_pte - look up PTE at a user virtual address 55189fd6dad1SPaolo Bonzini * @mm: the mm_struct of the target address space 55199fd6dad1SPaolo Bonzini * @address: user virtual address 55209fd6dad1SPaolo Bonzini * @ptepp: location to store found PTE 55219fd6dad1SPaolo Bonzini * @ptlp: location to store the lock for the PTE 55229fd6dad1SPaolo Bonzini * 55239fd6dad1SPaolo Bonzini * On a successful return, the pointer to the PTE is stored in @ptepp; 55249fd6dad1SPaolo Bonzini * the corresponding lock is taken and its location is stored in @ptlp. 55259fd6dad1SPaolo Bonzini * The contents of the PTE are only stable until @ptlp is released; 55269fd6dad1SPaolo Bonzini * any further use, if any, must be protected against invalidation 55279fd6dad1SPaolo Bonzini * with MMU notifiers. 55289fd6dad1SPaolo Bonzini * 55299fd6dad1SPaolo Bonzini * Only IO mappings and raw PFN mappings are allowed. The mmap semaphore 55309fd6dad1SPaolo Bonzini * should be taken for read. 55319fd6dad1SPaolo Bonzini * 55329fd6dad1SPaolo Bonzini * KVM uses this function. While it is arguably less bad than ``follow_pfn``, 55339fd6dad1SPaolo Bonzini * it is not a good general-purpose API. 55349fd6dad1SPaolo Bonzini * 55359fd6dad1SPaolo Bonzini * Return: zero on success, -ve otherwise. 55369fd6dad1SPaolo Bonzini */ 55379fd6dad1SPaolo Bonzini int follow_pte(struct mm_struct *mm, unsigned long address, 55389fd6dad1SPaolo Bonzini pte_t **ptepp, spinlock_t **ptlp) 55399fd6dad1SPaolo Bonzini { 55400e5e64c0SMuchun Song pgd_t *pgd; 55410e5e64c0SMuchun Song p4d_t *p4d; 55420e5e64c0SMuchun Song pud_t *pud; 55430e5e64c0SMuchun Song pmd_t *pmd; 55440e5e64c0SMuchun Song pte_t *ptep; 55450e5e64c0SMuchun Song 55460e5e64c0SMuchun Song pgd = pgd_offset(mm, address); 55470e5e64c0SMuchun Song if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 55480e5e64c0SMuchun Song goto out; 55490e5e64c0SMuchun Song 55500e5e64c0SMuchun Song p4d = p4d_offset(pgd, address); 55510e5e64c0SMuchun Song if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) 55520e5e64c0SMuchun Song goto out; 55530e5e64c0SMuchun Song 55540e5e64c0SMuchun Song pud = pud_offset(p4d, address); 55550e5e64c0SMuchun Song if (pud_none(*pud) || unlikely(pud_bad(*pud))) 55560e5e64c0SMuchun Song goto out; 55570e5e64c0SMuchun Song 55580e5e64c0SMuchun Song pmd = pmd_offset(pud, address); 55590e5e64c0SMuchun Song VM_BUG_ON(pmd_trans_huge(*pmd)); 55600e5e64c0SMuchun Song 55610e5e64c0SMuchun Song if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) 55620e5e64c0SMuchun Song goto out; 55630e5e64c0SMuchun Song 55640e5e64c0SMuchun Song ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 55650e5e64c0SMuchun Song if (!pte_present(*ptep)) 55660e5e64c0SMuchun Song goto unlock; 55670e5e64c0SMuchun Song *ptepp = ptep; 55680e5e64c0SMuchun Song return 0; 55690e5e64c0SMuchun Song unlock: 55700e5e64c0SMuchun Song pte_unmap_unlock(ptep, *ptlp); 55710e5e64c0SMuchun Song out: 55720e5e64c0SMuchun Song return -EINVAL; 55739fd6dad1SPaolo Bonzini } 55749fd6dad1SPaolo Bonzini EXPORT_SYMBOL_GPL(follow_pte); 55759fd6dad1SPaolo Bonzini 55769fd6dad1SPaolo Bonzini /** 55773b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 55783b6748e2SJohannes Weiner * @vma: memory mapping 55793b6748e2SJohannes Weiner * @address: user virtual address 55803b6748e2SJohannes Weiner * @pfn: location to store found PFN 55813b6748e2SJohannes Weiner * 55823b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 55833b6748e2SJohannes Weiner * 55849fd6dad1SPaolo Bonzini * This function does not allow the caller to read the permissions 55859fd6dad1SPaolo Bonzini * of the PTE. Do not use it. 55869fd6dad1SPaolo Bonzini * 5587a862f68aSMike Rapoport * Return: zero and the pfn at @pfn on success, -ve otherwise. 55883b6748e2SJohannes Weiner */ 55893b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 55903b6748e2SJohannes Weiner unsigned long *pfn) 55913b6748e2SJohannes Weiner { 55923b6748e2SJohannes Weiner int ret = -EINVAL; 55933b6748e2SJohannes Weiner spinlock_t *ptl; 55943b6748e2SJohannes Weiner pte_t *ptep; 55953b6748e2SJohannes Weiner 55963b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 55973b6748e2SJohannes Weiner return ret; 55983b6748e2SJohannes Weiner 55999fd6dad1SPaolo Bonzini ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 56003b6748e2SJohannes Weiner if (ret) 56013b6748e2SJohannes Weiner return ret; 56023b6748e2SJohannes Weiner *pfn = pte_pfn(*ptep); 56033b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 56043b6748e2SJohannes Weiner return 0; 56053b6748e2SJohannes Weiner } 56063b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 56073b6748e2SJohannes Weiner 560828b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 5609d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 561028b2ee20SRik van Riel unsigned long address, unsigned int flags, 5611d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 561228b2ee20SRik van Riel { 561303668a4dSJohannes Weiner int ret = -EINVAL; 561428b2ee20SRik van Riel pte_t *ptep, pte; 561528b2ee20SRik van Riel spinlock_t *ptl; 561628b2ee20SRik van Riel 5617d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 5618d87fe660Svenkatesh.pallipadi@intel.com goto out; 561928b2ee20SRik van Riel 56209fd6dad1SPaolo Bonzini if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 5621d87fe660Svenkatesh.pallipadi@intel.com goto out; 562228b2ee20SRik van Riel pte = *ptep; 562303668a4dSJohannes Weiner 5624f6f37321SLinus Torvalds if ((flags & FOLL_WRITE) && !pte_write(pte)) 562528b2ee20SRik van Riel goto unlock; 562628b2ee20SRik van Riel 562728b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 562803668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 562928b2ee20SRik van Riel 563003668a4dSJohannes Weiner ret = 0; 563128b2ee20SRik van Riel unlock: 563228b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 563328b2ee20SRik van Riel out: 5634d87fe660Svenkatesh.pallipadi@intel.com return ret; 563528b2ee20SRik van Riel } 563628b2ee20SRik van Riel 563796667f8aSDaniel Vetter /** 563896667f8aSDaniel Vetter * generic_access_phys - generic implementation for iomem mmap access 563996667f8aSDaniel Vetter * @vma: the vma to access 5640f0953a1bSIngo Molnar * @addr: userspace address, not relative offset within @vma 564196667f8aSDaniel Vetter * @buf: buffer to read/write 564296667f8aSDaniel Vetter * @len: length of transfer 564396667f8aSDaniel Vetter * @write: set to FOLL_WRITE when writing, otherwise reading 564496667f8aSDaniel Vetter * 564596667f8aSDaniel Vetter * This is a generic implementation for &vm_operations_struct.access for an 564696667f8aSDaniel Vetter * iomem mapping. This callback is used by access_process_vm() when the @vma is 564796667f8aSDaniel Vetter * not page based. 564896667f8aSDaniel Vetter */ 564928b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 565028b2ee20SRik van Riel void *buf, int len, int write) 565128b2ee20SRik van Riel { 565228b2ee20SRik van Riel resource_size_t phys_addr; 565328b2ee20SRik van Riel unsigned long prot = 0; 56542bc7273bSKOSAKI Motohiro void __iomem *maddr; 565596667f8aSDaniel Vetter pte_t *ptep, pte; 565696667f8aSDaniel Vetter spinlock_t *ptl; 565796667f8aSDaniel Vetter int offset = offset_in_page(addr); 565896667f8aSDaniel Vetter int ret = -EINVAL; 565928b2ee20SRik van Riel 566096667f8aSDaniel Vetter if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 566196667f8aSDaniel Vetter return -EINVAL; 566296667f8aSDaniel Vetter 566396667f8aSDaniel Vetter retry: 5664e913a8cdSLinus Torvalds if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) 566596667f8aSDaniel Vetter return -EINVAL; 566696667f8aSDaniel Vetter pte = *ptep; 566796667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 566896667f8aSDaniel Vetter 566996667f8aSDaniel Vetter prot = pgprot_val(pte_pgprot(pte)); 567096667f8aSDaniel Vetter phys_addr = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 567196667f8aSDaniel Vetter 567296667f8aSDaniel Vetter if ((write & FOLL_WRITE) && !pte_write(pte)) 567328b2ee20SRik van Riel return -EINVAL; 567428b2ee20SRik van Riel 56759cb12d7bSGrazvydas Ignotas maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); 567624eee1e4Sjie@chenjie6@huwei.com if (!maddr) 567724eee1e4Sjie@chenjie6@huwei.com return -ENOMEM; 567824eee1e4Sjie@chenjie6@huwei.com 5679e913a8cdSLinus Torvalds if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) 568096667f8aSDaniel Vetter goto out_unmap; 568196667f8aSDaniel Vetter 568296667f8aSDaniel Vetter if (!pte_same(pte, *ptep)) { 568396667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 568496667f8aSDaniel Vetter iounmap(maddr); 568596667f8aSDaniel Vetter 568696667f8aSDaniel Vetter goto retry; 568796667f8aSDaniel Vetter } 568896667f8aSDaniel Vetter 568928b2ee20SRik van Riel if (write) 569028b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 569128b2ee20SRik van Riel else 569228b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 569396667f8aSDaniel Vetter ret = len; 569496667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 569596667f8aSDaniel Vetter out_unmap: 569628b2ee20SRik van Riel iounmap(maddr); 569728b2ee20SRik van Riel 569896667f8aSDaniel Vetter return ret; 569928b2ee20SRik van Riel } 57005a73633eSUwe Kleine-König EXPORT_SYMBOL_GPL(generic_access_phys); 570128b2ee20SRik van Riel #endif 570228b2ee20SRik van Riel 57030ec76a11SDavid Howells /* 5704d3f5ffcaSJohn Hubbard * Access another process' address space as given in mm. 57050ec76a11SDavid Howells */ 5706d3f5ffcaSJohn Hubbard int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf, 5707d3f5ffcaSJohn Hubbard int len, unsigned int gup_flags) 57080ec76a11SDavid Howells { 57090ec76a11SDavid Howells struct vm_area_struct *vma; 57100ec76a11SDavid Howells void *old_buf = buf; 5711442486ecSLorenzo Stoakes int write = gup_flags & FOLL_WRITE; 57120ec76a11SDavid Howells 5713d8ed45c5SMichel Lespinasse if (mmap_read_lock_killable(mm)) 57141e426fe2SKonstantin Khlebnikov return 0; 57151e426fe2SKonstantin Khlebnikov 5716183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 57170ec76a11SDavid Howells while (len) { 57180ec76a11SDavid Howells int bytes, ret, offset; 57190ec76a11SDavid Howells void *maddr; 572028b2ee20SRik van Riel struct page *page = NULL; 57210ec76a11SDavid Howells 572264019a2eSPeter Xu ret = get_user_pages_remote(mm, addr, 1, 57235b56d49fSLorenzo Stoakes gup_flags, &page, &vma, NULL); 572428b2ee20SRik van Riel if (ret <= 0) { 5725dbffcd03SRik van Riel #ifndef CONFIG_HAVE_IOREMAP_PROT 5726dbffcd03SRik van Riel break; 5727dbffcd03SRik van Riel #else 572828b2ee20SRik van Riel /* 572928b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 573028b2ee20SRik van Riel * we can access using slightly different code. 573128b2ee20SRik van Riel */ 57323e418f98SLiam Howlett vma = vma_lookup(mm, addr); 57333e418f98SLiam Howlett if (!vma) 57340ec76a11SDavid Howells break; 573528b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 573628b2ee20SRik van Riel ret = vma->vm_ops->access(vma, addr, buf, 573728b2ee20SRik van Riel len, write); 573828b2ee20SRik van Riel if (ret <= 0) 573928b2ee20SRik van Riel break; 574028b2ee20SRik van Riel bytes = ret; 5741dbffcd03SRik van Riel #endif 574228b2ee20SRik van Riel } else { 57430ec76a11SDavid Howells bytes = len; 57440ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 57450ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 57460ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 57470ec76a11SDavid Howells 57480ec76a11SDavid Howells maddr = kmap(page); 57490ec76a11SDavid Howells if (write) { 57500ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 57510ec76a11SDavid Howells maddr + offset, buf, bytes); 57520ec76a11SDavid Howells set_page_dirty_lock(page); 57530ec76a11SDavid Howells } else { 57540ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 57550ec76a11SDavid Howells buf, maddr + offset, bytes); 57560ec76a11SDavid Howells } 57570ec76a11SDavid Howells kunmap(page); 575809cbfeafSKirill A. Shutemov put_page(page); 575928b2ee20SRik van Riel } 57600ec76a11SDavid Howells len -= bytes; 57610ec76a11SDavid Howells buf += bytes; 57620ec76a11SDavid Howells addr += bytes; 57630ec76a11SDavid Howells } 5764d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 57650ec76a11SDavid Howells 57660ec76a11SDavid Howells return buf - old_buf; 57670ec76a11SDavid Howells } 576803252919SAndi Kleen 57695ddd36b9SStephen Wilson /** 5770ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 57715ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 57725ddd36b9SStephen Wilson * @addr: start address to access 57735ddd36b9SStephen Wilson * @buf: source or destination buffer 57745ddd36b9SStephen Wilson * @len: number of bytes to transfer 57756347e8d5SLorenzo Stoakes * @gup_flags: flags modifying lookup behaviour 57765ddd36b9SStephen Wilson * 57775ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 5778a862f68aSMike Rapoport * 5779a862f68aSMike Rapoport * Return: number of bytes copied from source to destination. 57805ddd36b9SStephen Wilson */ 57815ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 57826347e8d5SLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 57835ddd36b9SStephen Wilson { 5784d3f5ffcaSJohn Hubbard return __access_remote_vm(mm, addr, buf, len, gup_flags); 57855ddd36b9SStephen Wilson } 57865ddd36b9SStephen Wilson 578703252919SAndi Kleen /* 5788206cb636SStephen Wilson * Access another process' address space. 5789206cb636SStephen Wilson * Source/target buffer must be kernel space, 5790206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 5791206cb636SStephen Wilson */ 5792206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 5793f307ab6dSLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 5794206cb636SStephen Wilson { 5795206cb636SStephen Wilson struct mm_struct *mm; 5796206cb636SStephen Wilson int ret; 5797206cb636SStephen Wilson 5798206cb636SStephen Wilson mm = get_task_mm(tsk); 5799206cb636SStephen Wilson if (!mm) 5800206cb636SStephen Wilson return 0; 5801206cb636SStephen Wilson 5802d3f5ffcaSJohn Hubbard ret = __access_remote_vm(mm, addr, buf, len, gup_flags); 5803442486ecSLorenzo Stoakes 5804206cb636SStephen Wilson mmput(mm); 5805206cb636SStephen Wilson 5806206cb636SStephen Wilson return ret; 5807206cb636SStephen Wilson } 5808fcd35857SCatalin Marinas EXPORT_SYMBOL_GPL(access_process_vm); 5809206cb636SStephen Wilson 581003252919SAndi Kleen /* 581103252919SAndi Kleen * Print the name of a VMA. 581203252919SAndi Kleen */ 581303252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 581403252919SAndi Kleen { 581503252919SAndi Kleen struct mm_struct *mm = current->mm; 581603252919SAndi Kleen struct vm_area_struct *vma; 581703252919SAndi Kleen 5818e8bff74aSIngo Molnar /* 58190a7f682dSMichal Hocko * we might be running from an atomic context so we cannot sleep 5820e8bff74aSIngo Molnar */ 5821d8ed45c5SMichel Lespinasse if (!mmap_read_trylock(mm)) 5822e8bff74aSIngo Molnar return; 5823e8bff74aSIngo Molnar 582403252919SAndi Kleen vma = find_vma(mm, ip); 582503252919SAndi Kleen if (vma && vma->vm_file) { 582603252919SAndi Kleen struct file *f = vma->vm_file; 58270a7f682dSMichal Hocko char *buf = (char *)__get_free_page(GFP_NOWAIT); 582803252919SAndi Kleen if (buf) { 58292fbc57c5SAndy Shevchenko char *p; 583003252919SAndi Kleen 58319bf39ab2SMiklos Szeredi p = file_path(f, buf, PAGE_SIZE); 583203252919SAndi Kleen if (IS_ERR(p)) 583303252919SAndi Kleen p = "?"; 58342fbc57c5SAndy Shevchenko printk("%s%s[%lx+%lx]", prefix, kbasename(p), 583503252919SAndi Kleen vma->vm_start, 583603252919SAndi Kleen vma->vm_end - vma->vm_start); 583703252919SAndi Kleen free_page((unsigned long)buf); 583803252919SAndi Kleen } 583903252919SAndi Kleen } 5840d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 584103252919SAndi Kleen } 58423ee1afa3SNick Piggin 5843662bbcb2SMichael S. Tsirkin #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) 58449ec23531SDavid Hildenbrand void __might_fault(const char *file, int line) 58453ee1afa3SNick Piggin { 58469ec23531SDavid Hildenbrand if (pagefault_disabled()) 5847662bbcb2SMichael S. Tsirkin return; 584842a38756SThomas Gleixner __might_sleep(file, line); 58499ec23531SDavid Hildenbrand #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) 5850662bbcb2SMichael S. Tsirkin if (current->mm) 5851da1c55f1SMichel Lespinasse might_lock_read(¤t->mm->mmap_lock); 58529ec23531SDavid Hildenbrand #endif 58533ee1afa3SNick Piggin } 58549ec23531SDavid Hildenbrand EXPORT_SYMBOL(__might_fault); 58553ee1afa3SNick Piggin #endif 585647ad8475SAndrea Arcangeli 585747ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 5858c6ddfb6cSHuang Ying /* 5859c6ddfb6cSHuang Ying * Process all subpages of the specified huge page with the specified 5860c6ddfb6cSHuang Ying * operation. The target subpage will be processed last to keep its 5861c6ddfb6cSHuang Ying * cache lines hot. 5862c6ddfb6cSHuang Ying */ 58631cb9dc4bSLiu Shixin static inline int process_huge_page( 5864c6ddfb6cSHuang Ying unsigned long addr_hint, unsigned int pages_per_huge_page, 58651cb9dc4bSLiu Shixin int (*process_subpage)(unsigned long addr, int idx, void *arg), 5866c6ddfb6cSHuang Ying void *arg) 5867c6ddfb6cSHuang Ying { 58681cb9dc4bSLiu Shixin int i, n, base, l, ret; 5869c6ddfb6cSHuang Ying unsigned long addr = addr_hint & 5870c6ddfb6cSHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 5871c6ddfb6cSHuang Ying 5872c6ddfb6cSHuang Ying /* Process target subpage last to keep its cache lines hot */ 5873c6ddfb6cSHuang Ying might_sleep(); 5874c6ddfb6cSHuang Ying n = (addr_hint - addr) / PAGE_SIZE; 5875c6ddfb6cSHuang Ying if (2 * n <= pages_per_huge_page) { 5876c6ddfb6cSHuang Ying /* If target subpage in first half of huge page */ 5877c6ddfb6cSHuang Ying base = 0; 5878c6ddfb6cSHuang Ying l = n; 5879c6ddfb6cSHuang Ying /* Process subpages at the end of huge page */ 5880c6ddfb6cSHuang Ying for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { 5881c6ddfb6cSHuang Ying cond_resched(); 58821cb9dc4bSLiu Shixin ret = process_subpage(addr + i * PAGE_SIZE, i, arg); 58831cb9dc4bSLiu Shixin if (ret) 58841cb9dc4bSLiu Shixin return ret; 5885c6ddfb6cSHuang Ying } 5886c6ddfb6cSHuang Ying } else { 5887c6ddfb6cSHuang Ying /* If target subpage in second half of huge page */ 5888c6ddfb6cSHuang Ying base = pages_per_huge_page - 2 * (pages_per_huge_page - n); 5889c6ddfb6cSHuang Ying l = pages_per_huge_page - n; 5890c6ddfb6cSHuang Ying /* Process subpages at the begin of huge page */ 5891c6ddfb6cSHuang Ying for (i = 0; i < base; i++) { 5892c6ddfb6cSHuang Ying cond_resched(); 58931cb9dc4bSLiu Shixin ret = process_subpage(addr + i * PAGE_SIZE, i, arg); 58941cb9dc4bSLiu Shixin if (ret) 58951cb9dc4bSLiu Shixin return ret; 5896c6ddfb6cSHuang Ying } 5897c6ddfb6cSHuang Ying } 5898c6ddfb6cSHuang Ying /* 5899c6ddfb6cSHuang Ying * Process remaining subpages in left-right-left-right pattern 5900c6ddfb6cSHuang Ying * towards the target subpage 5901c6ddfb6cSHuang Ying */ 5902c6ddfb6cSHuang Ying for (i = 0; i < l; i++) { 5903c6ddfb6cSHuang Ying int left_idx = base + i; 5904c6ddfb6cSHuang Ying int right_idx = base + 2 * l - 1 - i; 5905c6ddfb6cSHuang Ying 5906c6ddfb6cSHuang Ying cond_resched(); 59071cb9dc4bSLiu Shixin ret = process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg); 59081cb9dc4bSLiu Shixin if (ret) 59091cb9dc4bSLiu Shixin return ret; 5910c6ddfb6cSHuang Ying cond_resched(); 59111cb9dc4bSLiu Shixin ret = process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg); 59121cb9dc4bSLiu Shixin if (ret) 59131cb9dc4bSLiu Shixin return ret; 5914c6ddfb6cSHuang Ying } 59151cb9dc4bSLiu Shixin return 0; 5916c6ddfb6cSHuang Ying } 5917c6ddfb6cSHuang Ying 591847ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 591947ad8475SAndrea Arcangeli unsigned long addr, 592047ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 592147ad8475SAndrea Arcangeli { 592247ad8475SAndrea Arcangeli int i; 592314455eabSCheng Li struct page *p; 592447ad8475SAndrea Arcangeli 592547ad8475SAndrea Arcangeli might_sleep(); 592614455eabSCheng Li for (i = 0; i < pages_per_huge_page; i++) { 592714455eabSCheng Li p = nth_page(page, i); 592847ad8475SAndrea Arcangeli cond_resched(); 592947ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 593047ad8475SAndrea Arcangeli } 593147ad8475SAndrea Arcangeli } 5932c6ddfb6cSHuang Ying 59331cb9dc4bSLiu Shixin static int clear_subpage(unsigned long addr, int idx, void *arg) 5934c6ddfb6cSHuang Ying { 5935c6ddfb6cSHuang Ying struct page *page = arg; 5936c6ddfb6cSHuang Ying 5937c6ddfb6cSHuang Ying clear_user_highpage(page + idx, addr); 59381cb9dc4bSLiu Shixin return 0; 5939c6ddfb6cSHuang Ying } 5940c6ddfb6cSHuang Ying 594147ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 5942c79b57e4SHuang Ying unsigned long addr_hint, unsigned int pages_per_huge_page) 594347ad8475SAndrea Arcangeli { 5944c79b57e4SHuang Ying unsigned long addr = addr_hint & 5945c79b57e4SHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 594647ad8475SAndrea Arcangeli 594747ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 594847ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 594947ad8475SAndrea Arcangeli return; 595047ad8475SAndrea Arcangeli } 595147ad8475SAndrea Arcangeli 5952c6ddfb6cSHuang Ying process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page); 595347ad8475SAndrea Arcangeli } 595447ad8475SAndrea Arcangeli 59551cb9dc4bSLiu Shixin static int copy_user_gigantic_page(struct folio *dst, struct folio *src, 595647ad8475SAndrea Arcangeli unsigned long addr, 595747ad8475SAndrea Arcangeli struct vm_area_struct *vma, 595847ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 595947ad8475SAndrea Arcangeli { 596047ad8475SAndrea Arcangeli int i; 5961c0e8150eSZhangPeng struct page *dst_page; 5962c0e8150eSZhangPeng struct page *src_page; 596347ad8475SAndrea Arcangeli 596414455eabSCheng Li for (i = 0; i < pages_per_huge_page; i++) { 5965c0e8150eSZhangPeng dst_page = folio_page(dst, i); 5966c0e8150eSZhangPeng src_page = folio_page(src, i); 596714455eabSCheng Li 596847ad8475SAndrea Arcangeli cond_resched(); 59691cb9dc4bSLiu Shixin if (copy_mc_user_highpage(dst_page, src_page, 59701cb9dc4bSLiu Shixin addr + i*PAGE_SIZE, vma)) { 59711cb9dc4bSLiu Shixin memory_failure_queue(page_to_pfn(src_page), 0); 59721cb9dc4bSLiu Shixin return -EHWPOISON; 597347ad8475SAndrea Arcangeli } 597447ad8475SAndrea Arcangeli } 59751cb9dc4bSLiu Shixin return 0; 59761cb9dc4bSLiu Shixin } 597747ad8475SAndrea Arcangeli 5978c9f4cd71SHuang Ying struct copy_subpage_arg { 5979c9f4cd71SHuang Ying struct page *dst; 5980c9f4cd71SHuang Ying struct page *src; 5981c9f4cd71SHuang Ying struct vm_area_struct *vma; 5982c9f4cd71SHuang Ying }; 5983c9f4cd71SHuang Ying 59841cb9dc4bSLiu Shixin static int copy_subpage(unsigned long addr, int idx, void *arg) 5985c9f4cd71SHuang Ying { 5986c9f4cd71SHuang Ying struct copy_subpage_arg *copy_arg = arg; 5987c9f4cd71SHuang Ying 59881cb9dc4bSLiu Shixin if (copy_mc_user_highpage(copy_arg->dst + idx, copy_arg->src + idx, 59891cb9dc4bSLiu Shixin addr, copy_arg->vma)) { 59901cb9dc4bSLiu Shixin memory_failure_queue(page_to_pfn(copy_arg->src + idx), 0); 59911cb9dc4bSLiu Shixin return -EHWPOISON; 59921cb9dc4bSLiu Shixin } 59931cb9dc4bSLiu Shixin return 0; 5994c9f4cd71SHuang Ying } 5995c9f4cd71SHuang Ying 59961cb9dc4bSLiu Shixin int copy_user_large_folio(struct folio *dst, struct folio *src, 5997c0e8150eSZhangPeng unsigned long addr_hint, struct vm_area_struct *vma) 599847ad8475SAndrea Arcangeli { 5999c0e8150eSZhangPeng unsigned int pages_per_huge_page = folio_nr_pages(dst); 6000c9f4cd71SHuang Ying unsigned long addr = addr_hint & 6001c9f4cd71SHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 6002c9f4cd71SHuang Ying struct copy_subpage_arg arg = { 6003c0e8150eSZhangPeng .dst = &dst->page, 6004c0e8150eSZhangPeng .src = &src->page, 6005c9f4cd71SHuang Ying .vma = vma, 6006c9f4cd71SHuang Ying }; 600747ad8475SAndrea Arcangeli 60081cb9dc4bSLiu Shixin if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) 60091cb9dc4bSLiu Shixin return copy_user_gigantic_page(dst, src, addr, vma, 601047ad8475SAndrea Arcangeli pages_per_huge_page); 601147ad8475SAndrea Arcangeli 60121cb9dc4bSLiu Shixin return process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg); 601347ad8475SAndrea Arcangeli } 6014fa4d75c1SMike Kravetz 6015e87340caSZhangPeng long copy_folio_from_user(struct folio *dst_folio, 6016fa4d75c1SMike Kravetz const void __user *usr_src, 6017810a56b9SMike Kravetz bool allow_pagefault) 6018fa4d75c1SMike Kravetz { 6019e87340caSZhangPeng void *kaddr; 6020fa4d75c1SMike Kravetz unsigned long i, rc = 0; 6021e87340caSZhangPeng unsigned int nr_pages = folio_nr_pages(dst_folio); 6022e87340caSZhangPeng unsigned long ret_val = nr_pages * PAGE_SIZE; 602314455eabSCheng Li struct page *subpage; 6024fa4d75c1SMike Kravetz 6025e87340caSZhangPeng for (i = 0; i < nr_pages; i++) { 6026e87340caSZhangPeng subpage = folio_page(dst_folio, i); 6027e87340caSZhangPeng kaddr = kmap_local_page(subpage); 60280d508c1fSZhangPeng if (!allow_pagefault) 60290d508c1fSZhangPeng pagefault_disable(); 6030e87340caSZhangPeng rc = copy_from_user(kaddr, usr_src + i * PAGE_SIZE, PAGE_SIZE); 60310d508c1fSZhangPeng if (!allow_pagefault) 60320d508c1fSZhangPeng pagefault_enable(); 6033e87340caSZhangPeng kunmap_local(kaddr); 6034fa4d75c1SMike Kravetz 6035fa4d75c1SMike Kravetz ret_val -= (PAGE_SIZE - rc); 6036fa4d75c1SMike Kravetz if (rc) 6037fa4d75c1SMike Kravetz break; 6038fa4d75c1SMike Kravetz 6039e763243cSMuchun Song flush_dcache_page(subpage); 6040e763243cSMuchun Song 6041fa4d75c1SMike Kravetz cond_resched(); 6042fa4d75c1SMike Kravetz } 6043fa4d75c1SMike Kravetz return ret_val; 6044fa4d75c1SMike Kravetz } 604547ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 604649076ec2SKirill A. Shutemov 604740b64acdSOlof Johansson #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS 6048b35f1819SKirill A. Shutemov 6049b35f1819SKirill A. Shutemov static struct kmem_cache *page_ptl_cachep; 6050b35f1819SKirill A. Shutemov 6051b35f1819SKirill A. Shutemov void __init ptlock_cache_init(void) 6052b35f1819SKirill A. Shutemov { 6053b35f1819SKirill A. Shutemov page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, 6054b35f1819SKirill A. Shutemov SLAB_PANIC, NULL); 6055b35f1819SKirill A. Shutemov } 6056b35f1819SKirill A. Shutemov 6057539edb58SPeter Zijlstra bool ptlock_alloc(struct page *page) 605849076ec2SKirill A. Shutemov { 605949076ec2SKirill A. Shutemov spinlock_t *ptl; 606049076ec2SKirill A. Shutemov 6061b35f1819SKirill A. Shutemov ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); 606249076ec2SKirill A. Shutemov if (!ptl) 606349076ec2SKirill A. Shutemov return false; 6064539edb58SPeter Zijlstra page->ptl = ptl; 606549076ec2SKirill A. Shutemov return true; 606649076ec2SKirill A. Shutemov } 606749076ec2SKirill A. Shutemov 6068539edb58SPeter Zijlstra void ptlock_free(struct page *page) 606949076ec2SKirill A. Shutemov { 6070b35f1819SKirill A. Shutemov kmem_cache_free(page_ptl_cachep, page->ptl); 607149076ec2SKirill A. Shutemov } 607249076ec2SKirill A. Shutemov #endif 6073