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> 807a7f0946SArjun Roy #include <linux/net_mm.h> 811da177e4SLinus Torvalds 82b3d1411bSJoel Fernandes (Google) #include <trace/events/kmem.h> 83b3d1411bSJoel Fernandes (Google) 846952b61dSAlexey Dobriyan #include <asm/io.h> 8533a709b2SDave Hansen #include <asm/mmu_context.h> 861da177e4SLinus Torvalds #include <asm/pgalloc.h> 877c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 881da177e4SLinus Torvalds #include <asm/tlb.h> 891da177e4SLinus Torvalds #include <asm/tlbflush.h> 901da177e4SLinus Torvalds 91e80d3909SJoerg Roedel #include "pgalloc-track.h" 9242b77728SJan Beulich #include "internal.h" 93014bb1deSNeilBrown #include "swap.h" 9442b77728SJan Beulich 95af27d940SArnd Bergmann #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST) 9690572890SPeter Zijlstra #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. 9775980e97SPeter Zijlstra #endif 9875980e97SPeter Zijlstra 99a9ee6cf5SMike Rapoport #ifndef CONFIG_NUMA 1001da177e4SLinus Torvalds unsigned long max_mapnr; 1011da177e4SLinus Torvalds EXPORT_SYMBOL(max_mapnr); 102166f61b9STobin C Harding 103166f61b9STobin C Harding struct page *mem_map; 1041da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 1051da177e4SLinus Torvalds #endif 1061da177e4SLinus Torvalds 1075c041f5dSPeter Xu static vm_fault_t do_fault(struct vm_fault *vmf); 1082bad466cSPeter Xu static vm_fault_t do_anonymous_page(struct vm_fault *vmf); 1092bad466cSPeter Xu static bool vmf_pte_changed(struct vm_fault *vmf); 1102bad466cSPeter Xu 1112bad466cSPeter Xu /* 1122bad466cSPeter Xu * Return true if the original pte was a uffd-wp pte marker (so the pte was 1132bad466cSPeter Xu * wr-protected). 1142bad466cSPeter Xu */ 1152bad466cSPeter Xu static bool vmf_orig_pte_uffd_wp(struct vm_fault *vmf) 1162bad466cSPeter Xu { 1172bad466cSPeter Xu if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID)) 1182bad466cSPeter Xu return false; 1192bad466cSPeter Xu 1202bad466cSPeter Xu return pte_marker_uffd_wp(vmf->orig_pte); 1212bad466cSPeter Xu } 1225c041f5dSPeter Xu 1231da177e4SLinus Torvalds /* 1241da177e4SLinus Torvalds * A number of key systems in x86 including ioremap() rely on the assumption 1251da177e4SLinus Torvalds * that high_memory defines the upper bound on direct map memory, then end 1261da177e4SLinus Torvalds * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and 1271da177e4SLinus Torvalds * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL 1281da177e4SLinus Torvalds * and ZONE_HIGHMEM. 1291da177e4SLinus Torvalds */ 1301da177e4SLinus Torvalds void *high_memory; 1311da177e4SLinus Torvalds EXPORT_SYMBOL(high_memory); 1321da177e4SLinus Torvalds 13332a93233SIngo Molnar /* 13432a93233SIngo Molnar * Randomize the address space (stacks, mmaps, brk, etc.). 13532a93233SIngo Molnar * 13632a93233SIngo Molnar * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, 13732a93233SIngo Molnar * as ancient (libc5 based) binaries can segfault. ) 13832a93233SIngo Molnar */ 13932a93233SIngo Molnar int randomize_va_space __read_mostly = 14032a93233SIngo Molnar #ifdef CONFIG_COMPAT_BRK 14132a93233SIngo Molnar 1; 14232a93233SIngo Molnar #else 14332a93233SIngo Molnar 2; 14432a93233SIngo Molnar #endif 145a62eaf15SAndi Kleen 14646bdb427SWill Deacon #ifndef arch_wants_old_prefaulted_pte 14746bdb427SWill Deacon static inline bool arch_wants_old_prefaulted_pte(void) 14846bdb427SWill Deacon { 14946bdb427SWill Deacon /* 15046bdb427SWill Deacon * Transitioning a PTE from 'old' to 'young' can be expensive on 15146bdb427SWill Deacon * some architectures, even if it's performed in hardware. By 15246bdb427SWill Deacon * default, "false" means prefaulted entries will be 'young'. 15346bdb427SWill Deacon */ 15446bdb427SWill Deacon return false; 15546bdb427SWill Deacon } 15646bdb427SWill Deacon #endif 15746bdb427SWill Deacon 158a62eaf15SAndi Kleen static int __init disable_randmaps(char *s) 159a62eaf15SAndi Kleen { 160a62eaf15SAndi Kleen randomize_va_space = 0; 1619b41046cSOGAWA Hirofumi return 1; 162a62eaf15SAndi Kleen } 163a62eaf15SAndi Kleen __setup("norandmaps", disable_randmaps); 164a62eaf15SAndi Kleen 16562eede62SHugh Dickins unsigned long zero_pfn __read_mostly; 1660b70068eSArd Biesheuvel EXPORT_SYMBOL(zero_pfn); 1670b70068eSArd Biesheuvel 168166f61b9STobin C Harding unsigned long highest_memmap_pfn __read_mostly; 169166f61b9STobin C Harding 170a13ea5b7SHugh Dickins /* 171a13ea5b7SHugh Dickins * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() 172a13ea5b7SHugh Dickins */ 173a13ea5b7SHugh Dickins static int __init init_zero_pfn(void) 174a13ea5b7SHugh Dickins { 175a13ea5b7SHugh Dickins zero_pfn = page_to_pfn(ZERO_PAGE(0)); 176a13ea5b7SHugh Dickins return 0; 177a13ea5b7SHugh Dickins } 178e720e7d0SIlya Lipnitskiy early_initcall(init_zero_pfn); 179a62eaf15SAndi Kleen 180f1a79412SShakeel Butt void mm_trace_rss_stat(struct mm_struct *mm, int member) 181b3d1411bSJoel Fernandes (Google) { 182f1a79412SShakeel Butt trace_rss_stat(mm, member); 183b3d1411bSJoel Fernandes (Google) } 184d559db08SKAMEZAWA Hiroyuki 1851da177e4SLinus Torvalds /* 1861da177e4SLinus Torvalds * Note: this doesn't free the actual pages themselves. That 1871da177e4SLinus Torvalds * has been handled earlier when unmapping all the memory regions. 1881da177e4SLinus Torvalds */ 1899e1b32caSBenjamin Herrenschmidt static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, 1909e1b32caSBenjamin Herrenschmidt unsigned long addr) 1911da177e4SLinus Torvalds { 1922f569afdSMartin Schwidefsky pgtable_t token = pmd_pgtable(*pmd); 1931da177e4SLinus Torvalds pmd_clear(pmd); 1949e1b32caSBenjamin Herrenschmidt pte_free_tlb(tlb, token, addr); 195c4812909SKirill A. Shutemov mm_dec_nr_ptes(tlb->mm); 1961da177e4SLinus Torvalds } 1971da177e4SLinus Torvalds 198e0da382cSHugh Dickins static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, 199e0da382cSHugh Dickins unsigned long addr, unsigned long end, 200e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 2011da177e4SLinus Torvalds { 2021da177e4SLinus Torvalds pmd_t *pmd; 2031da177e4SLinus Torvalds unsigned long next; 204e0da382cSHugh Dickins unsigned long start; 2051da177e4SLinus Torvalds 206e0da382cSHugh Dickins start = addr; 2071da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 2081da177e4SLinus Torvalds do { 2091da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 2101da177e4SLinus Torvalds if (pmd_none_or_clear_bad(pmd)) 2111da177e4SLinus Torvalds continue; 2129e1b32caSBenjamin Herrenschmidt free_pte_range(tlb, pmd, addr); 2131da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 2141da177e4SLinus Torvalds 215e0da382cSHugh Dickins start &= PUD_MASK; 216e0da382cSHugh Dickins if (start < floor) 217e0da382cSHugh Dickins return; 218e0da382cSHugh Dickins if (ceiling) { 219e0da382cSHugh Dickins ceiling &= PUD_MASK; 220e0da382cSHugh Dickins if (!ceiling) 221e0da382cSHugh Dickins return; 2221da177e4SLinus Torvalds } 223e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 224e0da382cSHugh Dickins return; 225e0da382cSHugh Dickins 226e0da382cSHugh Dickins pmd = pmd_offset(pud, start); 227e0da382cSHugh Dickins pud_clear(pud); 2289e1b32caSBenjamin Herrenschmidt pmd_free_tlb(tlb, pmd, start); 229dc6c9a35SKirill A. Shutemov mm_dec_nr_pmds(tlb->mm); 2301da177e4SLinus Torvalds } 2311da177e4SLinus Torvalds 232c2febafcSKirill A. Shutemov static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, 233e0da382cSHugh Dickins unsigned long addr, unsigned long end, 234e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 2351da177e4SLinus Torvalds { 2361da177e4SLinus Torvalds pud_t *pud; 2371da177e4SLinus Torvalds unsigned long next; 238e0da382cSHugh Dickins unsigned long start; 2391da177e4SLinus Torvalds 240e0da382cSHugh Dickins start = addr; 241c2febafcSKirill A. Shutemov pud = pud_offset(p4d, addr); 2421da177e4SLinus Torvalds do { 2431da177e4SLinus Torvalds next = pud_addr_end(addr, end); 2441da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 2451da177e4SLinus Torvalds continue; 246e0da382cSHugh Dickins free_pmd_range(tlb, pud, addr, next, floor, ceiling); 2471da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 2481da177e4SLinus Torvalds 249c2febafcSKirill A. Shutemov start &= P4D_MASK; 250c2febafcSKirill A. Shutemov if (start < floor) 251c2febafcSKirill A. Shutemov return; 252c2febafcSKirill A. Shutemov if (ceiling) { 253c2febafcSKirill A. Shutemov ceiling &= P4D_MASK; 254c2febafcSKirill A. Shutemov if (!ceiling) 255c2febafcSKirill A. Shutemov return; 256c2febafcSKirill A. Shutemov } 257c2febafcSKirill A. Shutemov if (end - 1 > ceiling - 1) 258c2febafcSKirill A. Shutemov return; 259c2febafcSKirill A. Shutemov 260c2febafcSKirill A. Shutemov pud = pud_offset(p4d, start); 261c2febafcSKirill A. Shutemov p4d_clear(p4d); 262c2febafcSKirill A. Shutemov pud_free_tlb(tlb, pud, start); 263b4e98d9aSKirill A. Shutemov mm_dec_nr_puds(tlb->mm); 264c2febafcSKirill A. Shutemov } 265c2febafcSKirill A. Shutemov 266c2febafcSKirill A. Shutemov static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, 267c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 268c2febafcSKirill A. Shutemov unsigned long floor, unsigned long ceiling) 269c2febafcSKirill A. Shutemov { 270c2febafcSKirill A. Shutemov p4d_t *p4d; 271c2febafcSKirill A. Shutemov unsigned long next; 272c2febafcSKirill A. Shutemov unsigned long start; 273c2febafcSKirill A. Shutemov 274c2febafcSKirill A. Shutemov start = addr; 275c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, addr); 276c2febafcSKirill A. Shutemov do { 277c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 278c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(p4d)) 279c2febafcSKirill A. Shutemov continue; 280c2febafcSKirill A. Shutemov free_pud_range(tlb, p4d, addr, next, floor, ceiling); 281c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 282c2febafcSKirill A. Shutemov 283e0da382cSHugh Dickins start &= PGDIR_MASK; 284e0da382cSHugh Dickins if (start < floor) 285e0da382cSHugh Dickins return; 286e0da382cSHugh Dickins if (ceiling) { 287e0da382cSHugh Dickins ceiling &= PGDIR_MASK; 288e0da382cSHugh Dickins if (!ceiling) 289e0da382cSHugh Dickins return; 2901da177e4SLinus Torvalds } 291e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 292e0da382cSHugh Dickins return; 293e0da382cSHugh Dickins 294c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, start); 295e0da382cSHugh Dickins pgd_clear(pgd); 296c2febafcSKirill A. Shutemov p4d_free_tlb(tlb, p4d, start); 2971da177e4SLinus Torvalds } 2981da177e4SLinus Torvalds 2991da177e4SLinus Torvalds /* 300e0da382cSHugh Dickins * This function frees user-level page tables of a process. 3011da177e4SLinus Torvalds */ 30242b77728SJan Beulich void free_pgd_range(struct mmu_gather *tlb, 303e0da382cSHugh Dickins unsigned long addr, unsigned long end, 304e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 3051da177e4SLinus Torvalds { 3061da177e4SLinus Torvalds pgd_t *pgd; 3071da177e4SLinus Torvalds unsigned long next; 3081da177e4SLinus Torvalds 309e0da382cSHugh Dickins /* 310e0da382cSHugh Dickins * The next few lines have given us lots of grief... 311e0da382cSHugh Dickins * 312e0da382cSHugh Dickins * Why are we testing PMD* at this top level? Because often 313e0da382cSHugh Dickins * there will be no work to do at all, and we'd prefer not to 314e0da382cSHugh Dickins * go all the way down to the bottom just to discover that. 315e0da382cSHugh Dickins * 316e0da382cSHugh Dickins * Why all these "- 1"s? Because 0 represents both the bottom 317e0da382cSHugh Dickins * of the address space and the top of it (using -1 for the 318e0da382cSHugh Dickins * top wouldn't help much: the masks would do the wrong thing). 319e0da382cSHugh Dickins * The rule is that addr 0 and floor 0 refer to the bottom of 320e0da382cSHugh Dickins * the address space, but end 0 and ceiling 0 refer to the top 321e0da382cSHugh Dickins * Comparisons need to use "end - 1" and "ceiling - 1" (though 322e0da382cSHugh Dickins * that end 0 case should be mythical). 323e0da382cSHugh Dickins * 324e0da382cSHugh Dickins * Wherever addr is brought up or ceiling brought down, we must 325e0da382cSHugh Dickins * be careful to reject "the opposite 0" before it confuses the 326e0da382cSHugh Dickins * subsequent tests. But what about where end is brought down 327e0da382cSHugh Dickins * by PMD_SIZE below? no, end can't go down to 0 there. 328e0da382cSHugh Dickins * 329e0da382cSHugh Dickins * Whereas we round start (addr) and ceiling down, by different 330e0da382cSHugh Dickins * masks at different levels, in order to test whether a table 331e0da382cSHugh Dickins * now has no other vmas using it, so can be freed, we don't 332e0da382cSHugh Dickins * bother to round floor or end up - the tests don't need that. 333e0da382cSHugh Dickins */ 334e0da382cSHugh Dickins 335e0da382cSHugh Dickins addr &= PMD_MASK; 336e0da382cSHugh Dickins if (addr < floor) { 337e0da382cSHugh Dickins addr += PMD_SIZE; 338e0da382cSHugh Dickins if (!addr) 339e0da382cSHugh Dickins return; 340e0da382cSHugh Dickins } 341e0da382cSHugh Dickins if (ceiling) { 342e0da382cSHugh Dickins ceiling &= PMD_MASK; 343e0da382cSHugh Dickins if (!ceiling) 344e0da382cSHugh Dickins return; 345e0da382cSHugh Dickins } 346e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 347e0da382cSHugh Dickins end -= PMD_SIZE; 348e0da382cSHugh Dickins if (addr > end - 1) 349e0da382cSHugh Dickins return; 35007e32661SAneesh Kumar K.V /* 35107e32661SAneesh Kumar K.V * We add page table cache pages with PAGE_SIZE, 35207e32661SAneesh Kumar K.V * (see pte_free_tlb()), flush the tlb if we need 35307e32661SAneesh Kumar K.V */ 354ed6a7935SPeter Zijlstra tlb_change_page_size(tlb, PAGE_SIZE); 35542b77728SJan Beulich pgd = pgd_offset(tlb->mm, addr); 3561da177e4SLinus Torvalds do { 3571da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 3581da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 3591da177e4SLinus Torvalds continue; 360c2febafcSKirill A. Shutemov free_p4d_range(tlb, pgd, addr, next, floor, ceiling); 3611da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 362e0da382cSHugh Dickins } 363e0da382cSHugh Dickins 364763ecb03SLiam R. Howlett void free_pgtables(struct mmu_gather *tlb, struct maple_tree *mt, 365763ecb03SLiam R. Howlett struct vm_area_struct *vma, unsigned long floor, 36698e51a22SSuren Baghdasaryan unsigned long ceiling, bool mm_wr_locked) 367e0da382cSHugh Dickins { 368763ecb03SLiam R. Howlett MA_STATE(mas, mt, vma->vm_end, vma->vm_end); 369763ecb03SLiam R. Howlett 370763ecb03SLiam R. Howlett do { 371e0da382cSHugh Dickins unsigned long addr = vma->vm_start; 372763ecb03SLiam R. Howlett struct vm_area_struct *next; 373763ecb03SLiam R. Howlett 374763ecb03SLiam R. Howlett /* 375763ecb03SLiam R. Howlett * Note: USER_PGTABLES_CEILING may be passed as ceiling and may 376763ecb03SLiam R. Howlett * be 0. This will underflow and is okay. 377763ecb03SLiam R. Howlett */ 378763ecb03SLiam R. Howlett next = mas_find(&mas, ceiling - 1); 379e0da382cSHugh Dickins 3808f4f8c16SHugh Dickins /* 38125d9e2d1Snpiggin@suse.de * Hide vma from rmap and truncate_pagecache before freeing 38225d9e2d1Snpiggin@suse.de * pgtables 3838f4f8c16SHugh Dickins */ 38498e51a22SSuren Baghdasaryan if (mm_wr_locked) 38598e51a22SSuren Baghdasaryan vma_start_write(vma); 3865beb4930SRik van Riel unlink_anon_vmas(vma); 3878f4f8c16SHugh Dickins unlink_file_vma(vma); 3888f4f8c16SHugh Dickins 3899da61aefSDavid Gibson if (is_vm_hugetlb_page(vma)) { 3903bf5ee95SHugh Dickins hugetlb_free_pgd_range(tlb, addr, vma->vm_end, 3913bf5ee95SHugh Dickins floor, next ? next->vm_start : ceiling); 3923bf5ee95SHugh Dickins } else { 3933bf5ee95SHugh Dickins /* 3943bf5ee95SHugh Dickins * Optimization: gather nearby vmas into one call down 3953bf5ee95SHugh Dickins */ 3963bf5ee95SHugh Dickins while (next && next->vm_start <= vma->vm_end + PMD_SIZE 3974866920bSDavid Gibson && !is_vm_hugetlb_page(next)) { 398e0da382cSHugh Dickins vma = next; 399763ecb03SLiam R. Howlett next = mas_find(&mas, ceiling - 1); 40098e51a22SSuren Baghdasaryan if (mm_wr_locked) 40198e51a22SSuren Baghdasaryan vma_start_write(vma); 4025beb4930SRik van Riel unlink_anon_vmas(vma); 4038f4f8c16SHugh Dickins unlink_file_vma(vma); 404e0da382cSHugh Dickins } 4053bf5ee95SHugh Dickins free_pgd_range(tlb, addr, vma->vm_end, 406e0da382cSHugh Dickins floor, next ? next->vm_start : ceiling); 4073bf5ee95SHugh Dickins } 408e0da382cSHugh Dickins vma = next; 409763ecb03SLiam R. Howlett } while (vma); 4101da177e4SLinus Torvalds } 4111da177e4SLinus Torvalds 41203c4f204SQi Zheng void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte) 4131da177e4SLinus Torvalds { 41403c4f204SQi Zheng spinlock_t *ptl = pmd_lock(mm, pmd); 4151bb3630eSHugh Dickins 41603c4f204SQi Zheng if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 41703c4f204SQi Zheng mm_inc_nr_ptes(mm); 418362a61adSNick Piggin /* 419362a61adSNick Piggin * Ensure all pte setup (eg. pte page lock and page clearing) are 420362a61adSNick Piggin * visible before the pte is made visible to other CPUs by being 421362a61adSNick Piggin * put into page tables. 422362a61adSNick Piggin * 423362a61adSNick Piggin * The other side of the story is the pointer chasing in the page 424362a61adSNick Piggin * table walking code (when walking the page table without locking; 425362a61adSNick Piggin * ie. most of the time). Fortunately, these data accesses consist 426362a61adSNick Piggin * of a chain of data-dependent loads, meaning most CPUs (alpha 427362a61adSNick Piggin * being the notable exception) will already guarantee loads are 428362a61adSNick Piggin * seen in-order. See the alpha page table accessors for the 429bb7cdd38SWill Deacon * smp_rmb() barriers in page table walking code. 430362a61adSNick Piggin */ 431362a61adSNick Piggin smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ 432ed33b5a6SQi Zheng pmd_populate(mm, pmd, *pte); 433ed33b5a6SQi Zheng *pte = NULL; 4344b471e88SKirill A. Shutemov } 435c4088ebdSKirill A. Shutemov spin_unlock(ptl); 436ed33b5a6SQi Zheng } 437ed33b5a6SQi Zheng 438ed33b5a6SQi Zheng int __pte_alloc(struct mm_struct *mm, pmd_t *pmd) 439ed33b5a6SQi Zheng { 440ed33b5a6SQi Zheng pgtable_t new = pte_alloc_one(mm); 441ed33b5a6SQi Zheng if (!new) 442ed33b5a6SQi Zheng return -ENOMEM; 4438ac1f832SAndrea Arcangeli 44403c4f204SQi Zheng pmd_install(mm, pmd, &new); 4452f569afdSMartin Schwidefsky if (new) 4462f569afdSMartin Schwidefsky pte_free(mm, new); 4471bb3630eSHugh Dickins return 0; 4481da177e4SLinus Torvalds } 4491da177e4SLinus Torvalds 4504cf58924SJoel Fernandes (Google) int __pte_alloc_kernel(pmd_t *pmd) 4511da177e4SLinus Torvalds { 4524cf58924SJoel Fernandes (Google) pte_t *new = pte_alloc_one_kernel(&init_mm); 4531da177e4SLinus Torvalds if (!new) 4541bb3630eSHugh Dickins return -ENOMEM; 4551da177e4SLinus Torvalds 456872fec16SHugh Dickins spin_lock(&init_mm.page_table_lock); 4578ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 458ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 459872fec16SHugh Dickins pmd_populate_kernel(&init_mm, pmd, new); 4602f569afdSMartin Schwidefsky new = NULL; 4614b471e88SKirill A. Shutemov } 462872fec16SHugh Dickins spin_unlock(&init_mm.page_table_lock); 4632f569afdSMartin Schwidefsky if (new) 4642f569afdSMartin Schwidefsky pte_free_kernel(&init_mm, new); 4651bb3630eSHugh Dickins return 0; 4661da177e4SLinus Torvalds } 4671da177e4SLinus Torvalds 468d559db08SKAMEZAWA Hiroyuki static inline void init_rss_vec(int *rss) 469ae859762SHugh Dickins { 470d559db08SKAMEZAWA Hiroyuki memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); 471d559db08SKAMEZAWA Hiroyuki } 472d559db08SKAMEZAWA Hiroyuki 473d559db08SKAMEZAWA Hiroyuki static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) 474d559db08SKAMEZAWA Hiroyuki { 475d559db08SKAMEZAWA Hiroyuki int i; 476d559db08SKAMEZAWA Hiroyuki 47734e55232SKAMEZAWA Hiroyuki if (current->mm == mm) 47805af2e10SDavid Rientjes sync_mm_rss(mm); 479d559db08SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) 480d559db08SKAMEZAWA Hiroyuki if (rss[i]) 481d559db08SKAMEZAWA Hiroyuki add_mm_counter(mm, i, rss[i]); 482ae859762SHugh Dickins } 483ae859762SHugh Dickins 4841da177e4SLinus Torvalds /* 4856aab341eSLinus Torvalds * This function is called to print an error when a bad pte 4866aab341eSLinus Torvalds * is found. For example, we might have a PFN-mapped pte in 4876aab341eSLinus Torvalds * a region that doesn't allow it. 488b5810039SNick Piggin * 489b5810039SNick Piggin * The calling function must still handle the error. 490b5810039SNick Piggin */ 4913dc14741SHugh Dickins static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, 4923dc14741SHugh Dickins pte_t pte, struct page *page) 493b5810039SNick Piggin { 4943dc14741SHugh Dickins pgd_t *pgd = pgd_offset(vma->vm_mm, addr); 495c2febafcSKirill A. Shutemov p4d_t *p4d = p4d_offset(pgd, addr); 496c2febafcSKirill A. Shutemov pud_t *pud = pud_offset(p4d, addr); 4973dc14741SHugh Dickins pmd_t *pmd = pmd_offset(pud, addr); 4983dc14741SHugh Dickins struct address_space *mapping; 4993dc14741SHugh Dickins pgoff_t index; 500d936cf9bSHugh Dickins static unsigned long resume; 501d936cf9bSHugh Dickins static unsigned long nr_shown; 502d936cf9bSHugh Dickins static unsigned long nr_unshown; 503d936cf9bSHugh Dickins 504d936cf9bSHugh Dickins /* 505d936cf9bSHugh Dickins * Allow a burst of 60 reports, then keep quiet for that minute; 506d936cf9bSHugh Dickins * or allow a steady drip of one report per second. 507d936cf9bSHugh Dickins */ 508d936cf9bSHugh Dickins if (nr_shown == 60) { 509d936cf9bSHugh Dickins if (time_before(jiffies, resume)) { 510d936cf9bSHugh Dickins nr_unshown++; 511d936cf9bSHugh Dickins return; 512d936cf9bSHugh Dickins } 513d936cf9bSHugh Dickins if (nr_unshown) { 5141170532bSJoe Perches pr_alert("BUG: Bad page map: %lu messages suppressed\n", 515d936cf9bSHugh Dickins nr_unshown); 516d936cf9bSHugh Dickins nr_unshown = 0; 517d936cf9bSHugh Dickins } 518d936cf9bSHugh Dickins nr_shown = 0; 519d936cf9bSHugh Dickins } 520d936cf9bSHugh Dickins if (nr_shown++ == 0) 521d936cf9bSHugh Dickins resume = jiffies + 60 * HZ; 5223dc14741SHugh Dickins 5233dc14741SHugh Dickins mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; 5243dc14741SHugh Dickins index = linear_page_index(vma, addr); 5253dc14741SHugh Dickins 5261170532bSJoe Perches pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", 5273dc14741SHugh Dickins current->comm, 5283dc14741SHugh Dickins (long long)pte_val(pte), (long long)pmd_val(*pmd)); 529718a3821SWu Fengguang if (page) 530f0b791a3SDave Hansen dump_page(page, "bad pte"); 5316aa9b8b2SKefeng Wang pr_alert("addr:%px vm_flags:%08lx anon_vma:%px mapping:%px index:%lx\n", 5323dc14741SHugh Dickins (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); 5337e0a1265SMatthew Wilcox (Oracle) pr_alert("file:%pD fault:%ps mmap:%ps read_folio:%ps\n", 5342682582aSKonstantin Khlebnikov vma->vm_file, 5352682582aSKonstantin Khlebnikov vma->vm_ops ? vma->vm_ops->fault : NULL, 5362682582aSKonstantin Khlebnikov vma->vm_file ? vma->vm_file->f_op->mmap : NULL, 5377e0a1265SMatthew Wilcox (Oracle) mapping ? mapping->a_ops->read_folio : NULL); 538b5810039SNick Piggin dump_stack(); 539373d4d09SRusty Russell add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); 540b5810039SNick Piggin } 541b5810039SNick Piggin 542b5810039SNick Piggin /* 5437e675137SNick Piggin * vm_normal_page -- This function gets the "struct page" associated with a pte. 5446aab341eSLinus Torvalds * 5457e675137SNick Piggin * "Special" mappings do not wish to be associated with a "struct page" (either 5467e675137SNick Piggin * it doesn't exist, or it exists but they don't want to touch it). In this 5477e675137SNick Piggin * case, NULL is returned here. "Normal" mappings do have a struct page. 548b379d790SJared Hulbert * 5497e675137SNick Piggin * There are 2 broad cases. Firstly, an architecture may define a pte_special() 5507e675137SNick Piggin * pte bit, in which case this function is trivial. Secondly, an architecture 5517e675137SNick Piggin * may not have a spare pte bit, which requires a more complicated scheme, 5527e675137SNick Piggin * described below. 5537e675137SNick Piggin * 5547e675137SNick Piggin * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a 5557e675137SNick Piggin * special mapping (even if there are underlying and valid "struct pages"). 5567e675137SNick Piggin * COWed pages of a VM_PFNMAP are always normal. 5576aab341eSLinus Torvalds * 558b379d790SJared Hulbert * The way we recognize COWed pages within VM_PFNMAP mappings is through the 559b379d790SJared Hulbert * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit 5607e675137SNick Piggin * set, and the vm_pgoff will point to the first PFN mapped: thus every special 5617e675137SNick Piggin * mapping will always honor the rule 5626aab341eSLinus Torvalds * 5636aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 5646aab341eSLinus Torvalds * 5657e675137SNick Piggin * And for normal mappings this is false. 566b379d790SJared Hulbert * 5677e675137SNick Piggin * This restricts such mappings to be a linear translation from virtual address 5687e675137SNick Piggin * to pfn. To get around this restriction, we allow arbitrary mappings so long 5697e675137SNick Piggin * as the vma is not a COW mapping; in that case, we know that all ptes are 5707e675137SNick Piggin * special (because none can have been COWed). 571b379d790SJared Hulbert * 572b379d790SJared Hulbert * 5737e675137SNick Piggin * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. 5747e675137SNick Piggin * 575b379d790SJared Hulbert * VM_MIXEDMAP mappings can likewise contain memory with or without "struct 576b379d790SJared Hulbert * page" backing, however the difference is that _all_ pages with a struct 577b379d790SJared Hulbert * page (that is, those where pfn_valid is true) are refcounted and considered 578b379d790SJared Hulbert * normal pages by the VM. The disadvantage is that pages are refcounted 579b379d790SJared Hulbert * (which can be slower and simply not an option for some PFNMAP users). The 580b379d790SJared Hulbert * advantage is that we don't have to follow the strict linearity rule of 581b379d790SJared Hulbert * PFNMAP mappings in order to support COWable mappings. 582b379d790SJared Hulbert * 583ee498ed7SHugh Dickins */ 58425b2995aSChristoph Hellwig struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 58525b2995aSChristoph Hellwig pte_t pte) 586ee498ed7SHugh Dickins { 58722b31eecSHugh Dickins unsigned long pfn = pte_pfn(pte); 5887e675137SNick Piggin 58900b3a331SLaurent Dufour if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) { 590b38af472SHugh Dickins if (likely(!pte_special(pte))) 59122b31eecSHugh Dickins goto check_pfn; 592667a0a06SDavid Vrabel if (vma->vm_ops && vma->vm_ops->find_special_page) 593667a0a06SDavid Vrabel return vma->vm_ops->find_special_page(vma, addr); 594a13ea5b7SHugh Dickins if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 595a13ea5b7SHugh Dickins return NULL; 596df6ad698SJérôme Glisse if (is_zero_pfn(pfn)) 597df6ad698SJérôme Glisse return NULL; 598e1fb4a08SDave Jiang if (pte_devmap(pte)) 5993218f871SAlex Sierra /* 6003218f871SAlex Sierra * NOTE: New users of ZONE_DEVICE will not set pte_devmap() 6013218f871SAlex Sierra * and will have refcounts incremented on their struct pages 6023218f871SAlex Sierra * when they are inserted into PTEs, thus they are safe to 6033218f871SAlex Sierra * return here. Legacy ZONE_DEVICE pages that set pte_devmap() 6043218f871SAlex Sierra * do not have refcounts. Example of legacy ZONE_DEVICE is 6053218f871SAlex Sierra * MEMORY_DEVICE_FS_DAX type in pmem or virtio_fs drivers. 6063218f871SAlex Sierra */ 607e1fb4a08SDave Jiang return NULL; 608e1fb4a08SDave Jiang 60922b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 6107e675137SNick Piggin return NULL; 6117e675137SNick Piggin } 6127e675137SNick Piggin 61300b3a331SLaurent Dufour /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */ 6147e675137SNick Piggin 615b379d790SJared Hulbert if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 616b379d790SJared Hulbert if (vma->vm_flags & VM_MIXEDMAP) { 617b379d790SJared Hulbert if (!pfn_valid(pfn)) 618b379d790SJared Hulbert return NULL; 619b379d790SJared Hulbert goto out; 620b379d790SJared Hulbert } else { 6217e675137SNick Piggin unsigned long off; 6227e675137SNick Piggin off = (addr - vma->vm_start) >> PAGE_SHIFT; 6236aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 6246aab341eSLinus Torvalds return NULL; 62567121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 626fb155c16SLinus Torvalds return NULL; 6276aab341eSLinus Torvalds } 628b379d790SJared Hulbert } 6296aab341eSLinus Torvalds 630b38af472SHugh Dickins if (is_zero_pfn(pfn)) 631b38af472SHugh Dickins return NULL; 63200b3a331SLaurent Dufour 63322b31eecSHugh Dickins check_pfn: 63422b31eecSHugh Dickins if (unlikely(pfn > highest_memmap_pfn)) { 63522b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 63622b31eecSHugh Dickins return NULL; 63722b31eecSHugh Dickins } 6386aab341eSLinus Torvalds 6396aab341eSLinus Torvalds /* 6407e675137SNick Piggin * NOTE! We still have PageReserved() pages in the page tables. 6417e675137SNick Piggin * eg. VDSO mappings can cause them to exist. 6426aab341eSLinus Torvalds */ 643b379d790SJared Hulbert out: 6446aab341eSLinus Torvalds return pfn_to_page(pfn); 645ee498ed7SHugh Dickins } 646ee498ed7SHugh Dickins 647318e9342SVishal Moola (Oracle) struct folio *vm_normal_folio(struct vm_area_struct *vma, unsigned long addr, 648318e9342SVishal Moola (Oracle) pte_t pte) 649318e9342SVishal Moola (Oracle) { 650318e9342SVishal Moola (Oracle) struct page *page = vm_normal_page(vma, addr, pte); 651318e9342SVishal Moola (Oracle) 652318e9342SVishal Moola (Oracle) if (page) 653318e9342SVishal Moola (Oracle) return page_folio(page); 654318e9342SVishal Moola (Oracle) return NULL; 655318e9342SVishal Moola (Oracle) } 656318e9342SVishal Moola (Oracle) 65728093f9fSGerald Schaefer #ifdef CONFIG_TRANSPARENT_HUGEPAGE 65828093f9fSGerald Schaefer struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, 65928093f9fSGerald Schaefer pmd_t pmd) 66028093f9fSGerald Schaefer { 66128093f9fSGerald Schaefer unsigned long pfn = pmd_pfn(pmd); 66228093f9fSGerald Schaefer 66328093f9fSGerald Schaefer /* 66428093f9fSGerald Schaefer * There is no pmd_special() but there may be special pmds, e.g. 66528093f9fSGerald Schaefer * in a direct-access (dax) mapping, so let's just replicate the 66600b3a331SLaurent Dufour * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here. 66728093f9fSGerald Schaefer */ 66828093f9fSGerald Schaefer if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 66928093f9fSGerald Schaefer if (vma->vm_flags & VM_MIXEDMAP) { 67028093f9fSGerald Schaefer if (!pfn_valid(pfn)) 67128093f9fSGerald Schaefer return NULL; 67228093f9fSGerald Schaefer goto out; 67328093f9fSGerald Schaefer } else { 67428093f9fSGerald Schaefer unsigned long off; 67528093f9fSGerald Schaefer off = (addr - vma->vm_start) >> PAGE_SHIFT; 67628093f9fSGerald Schaefer if (pfn == vma->vm_pgoff + off) 67728093f9fSGerald Schaefer return NULL; 67828093f9fSGerald Schaefer if (!is_cow_mapping(vma->vm_flags)) 67928093f9fSGerald Schaefer return NULL; 68028093f9fSGerald Schaefer } 68128093f9fSGerald Schaefer } 68228093f9fSGerald Schaefer 683e1fb4a08SDave Jiang if (pmd_devmap(pmd)) 684e1fb4a08SDave Jiang return NULL; 6853cde287bSYu Zhao if (is_huge_zero_pmd(pmd)) 68628093f9fSGerald Schaefer return NULL; 68728093f9fSGerald Schaefer if (unlikely(pfn > highest_memmap_pfn)) 68828093f9fSGerald Schaefer return NULL; 68928093f9fSGerald Schaefer 69028093f9fSGerald Schaefer /* 69128093f9fSGerald Schaefer * NOTE! We still have PageReserved() pages in the page tables. 69228093f9fSGerald Schaefer * eg. VDSO mappings can cause them to exist. 69328093f9fSGerald Schaefer */ 69428093f9fSGerald Schaefer out: 69528093f9fSGerald Schaefer return pfn_to_page(pfn); 69628093f9fSGerald Schaefer } 69728093f9fSGerald Schaefer #endif 69828093f9fSGerald Schaefer 699b756a3b5SAlistair Popple static void restore_exclusive_pte(struct vm_area_struct *vma, 700b756a3b5SAlistair Popple struct page *page, unsigned long address, 701b756a3b5SAlistair Popple pte_t *ptep) 702b756a3b5SAlistair Popple { 703c33c7948SRyan Roberts pte_t orig_pte; 704b756a3b5SAlistair Popple pte_t pte; 705b756a3b5SAlistair Popple swp_entry_t entry; 706b756a3b5SAlistair Popple 707c33c7948SRyan Roberts orig_pte = ptep_get(ptep); 708b756a3b5SAlistair Popple pte = pte_mkold(mk_pte(page, READ_ONCE(vma->vm_page_prot))); 709c33c7948SRyan Roberts if (pte_swp_soft_dirty(orig_pte)) 710b756a3b5SAlistair Popple pte = pte_mksoft_dirty(pte); 711b756a3b5SAlistair Popple 712c33c7948SRyan Roberts entry = pte_to_swp_entry(orig_pte); 713c33c7948SRyan Roberts if (pte_swp_uffd_wp(orig_pte)) 714b756a3b5SAlistair Popple pte = pte_mkuffd_wp(pte); 715b756a3b5SAlistair Popple else if (is_writable_device_exclusive_entry(entry)) 716b756a3b5SAlistair Popple pte = maybe_mkwrite(pte_mkdirty(pte), vma); 717b756a3b5SAlistair Popple 7186c287605SDavid Hildenbrand VM_BUG_ON(pte_write(pte) && !(PageAnon(page) && PageAnonExclusive(page))); 7196c287605SDavid Hildenbrand 720b756a3b5SAlistair Popple /* 721b756a3b5SAlistair Popple * No need to take a page reference as one was already 722b756a3b5SAlistair Popple * created when the swap entry was made. 723b756a3b5SAlistair Popple */ 724b756a3b5SAlistair Popple if (PageAnon(page)) 725f1e2db12SDavid Hildenbrand page_add_anon_rmap(page, vma, address, RMAP_NONE); 726b756a3b5SAlistair Popple else 727b756a3b5SAlistair Popple /* 728b756a3b5SAlistair Popple * Currently device exclusive access only supports anonymous 729b756a3b5SAlistair Popple * memory so the entry shouldn't point to a filebacked page. 730b756a3b5SAlistair Popple */ 7314d8ff640SMiaohe Lin WARN_ON_ONCE(1); 732b756a3b5SAlistair Popple 7331eba86c0SPasha Tatashin set_pte_at(vma->vm_mm, address, ptep, pte); 7341eba86c0SPasha Tatashin 735b756a3b5SAlistair Popple /* 736b756a3b5SAlistair Popple * No need to invalidate - it was non-present before. However 737b756a3b5SAlistair Popple * secondary CPUs may have mappings that need invalidating. 738b756a3b5SAlistair Popple */ 739b756a3b5SAlistair Popple update_mmu_cache(vma, address, ptep); 740b756a3b5SAlistair Popple } 741b756a3b5SAlistair Popple 742b756a3b5SAlistair Popple /* 743b756a3b5SAlistair Popple * Tries to restore an exclusive pte if the page lock can be acquired without 744b756a3b5SAlistair Popple * sleeping. 745b756a3b5SAlistair Popple */ 746b756a3b5SAlistair Popple static int 747b756a3b5SAlistair Popple try_restore_exclusive_pte(pte_t *src_pte, struct vm_area_struct *vma, 748b756a3b5SAlistair Popple unsigned long addr) 749b756a3b5SAlistair Popple { 750c33c7948SRyan Roberts swp_entry_t entry = pte_to_swp_entry(ptep_get(src_pte)); 751b756a3b5SAlistair Popple struct page *page = pfn_swap_entry_to_page(entry); 752b756a3b5SAlistair Popple 753b756a3b5SAlistair Popple if (trylock_page(page)) { 754b756a3b5SAlistair Popple restore_exclusive_pte(vma, page, addr, src_pte); 755b756a3b5SAlistair Popple unlock_page(page); 756b756a3b5SAlistair Popple return 0; 757b756a3b5SAlistair Popple } 758b756a3b5SAlistair Popple 759b756a3b5SAlistair Popple return -EBUSY; 760b756a3b5SAlistair Popple } 761b756a3b5SAlistair Popple 762ee498ed7SHugh Dickins /* 7631da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 7641da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 7651da177e4SLinus Torvalds * covered by this vma. 7661da177e4SLinus Torvalds */ 7671da177e4SLinus Torvalds 768df3a57d1SLinus Torvalds static unsigned long 769df3a57d1SLinus Torvalds copy_nonpresent_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 7708f34f1eaSPeter Xu pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *dst_vma, 7718f34f1eaSPeter Xu struct vm_area_struct *src_vma, unsigned long addr, int *rss) 7721da177e4SLinus Torvalds { 7738f34f1eaSPeter Xu unsigned long vm_flags = dst_vma->vm_flags; 774c33c7948SRyan Roberts pte_t orig_pte = ptep_get(src_pte); 775c33c7948SRyan Roberts pte_t pte = orig_pte; 7761da177e4SLinus Torvalds struct page *page; 777c33c7948SRyan Roberts swp_entry_t entry = pte_to_swp_entry(orig_pte); 7780697212aSChristoph Lameter 7792022b4d1SHugh Dickins if (likely(!non_swap_entry(entry))) { 780570a335bSHugh Dickins if (swap_duplicate(entry) < 0) 7819a5cc85cSAlistair Popple return -EIO; 782570a335bSHugh Dickins 7831da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 7841da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 7851da177e4SLinus Torvalds spin_lock(&mmlist_lock); 786f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 787f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 788f412ac08SHugh Dickins &src_mm->mmlist); 7891da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 7901da177e4SLinus Torvalds } 7911493a191SDavid Hildenbrand /* Mark the swap entry as shared. */ 792c33c7948SRyan Roberts if (pte_swp_exclusive(orig_pte)) { 793c33c7948SRyan Roberts pte = pte_swp_clear_exclusive(orig_pte); 7941493a191SDavid Hildenbrand set_pte_at(src_mm, addr, src_pte, pte); 7951493a191SDavid Hildenbrand } 796b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]++; 7972022b4d1SHugh Dickins } else if (is_migration_entry(entry)) { 798af5cdaf8SAlistair Popple page = pfn_swap_entry_to_page(entry); 7999f9f1acdSKonstantin Khlebnikov 800eca56ff9SJerome Marchand rss[mm_counter(page)]++; 8019f9f1acdSKonstantin Khlebnikov 8026c287605SDavid Hildenbrand if (!is_readable_migration_entry(entry) && 8030697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 8040697212aSChristoph Lameter /* 8056c287605SDavid Hildenbrand * COW mappings require pages in both parent and child 8066c287605SDavid Hildenbrand * to be set to read. A previously exclusive entry is 8076c287605SDavid Hildenbrand * now shared. 8080697212aSChristoph Lameter */ 8094dd845b5SAlistair Popple entry = make_readable_migration_entry( 8104dd845b5SAlistair Popple swp_offset(entry)); 8110697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 812c33c7948SRyan Roberts if (pte_swp_soft_dirty(orig_pte)) 813c3d16e16SCyrill Gorcunov pte = pte_swp_mksoft_dirty(pte); 814c33c7948SRyan Roberts if (pte_swp_uffd_wp(orig_pte)) 815f45ec5ffSPeter Xu pte = pte_swp_mkuffd_wp(pte); 8160697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 8170697212aSChristoph Lameter } 8185042db43SJérôme Glisse } else if (is_device_private_entry(entry)) { 819af5cdaf8SAlistair Popple page = pfn_swap_entry_to_page(entry); 8205042db43SJérôme Glisse 8215042db43SJérôme Glisse /* 8225042db43SJérôme Glisse * Update rss count even for unaddressable pages, as 8235042db43SJérôme Glisse * they should treated just like normal pages in this 8245042db43SJérôme Glisse * respect. 8255042db43SJérôme Glisse * 8265042db43SJérôme Glisse * We will likely want to have some new rss counters 8275042db43SJérôme Glisse * for unaddressable pages, at some point. But for now 8285042db43SJérôme Glisse * keep things as they are. 8295042db43SJérôme Glisse */ 8305042db43SJérôme Glisse get_page(page); 8315042db43SJérôme Glisse rss[mm_counter(page)]++; 832fb3d824dSDavid Hildenbrand /* Cannot fail as these pages cannot get pinned. */ 833fb3d824dSDavid Hildenbrand BUG_ON(page_try_dup_anon_rmap(page, false, src_vma)); 8345042db43SJérôme Glisse 8355042db43SJérôme Glisse /* 8365042db43SJérôme Glisse * We do not preserve soft-dirty information, because so 8375042db43SJérôme Glisse * far, checkpoint/restore is the only feature that 8385042db43SJérôme Glisse * requires that. And checkpoint/restore does not work 8395042db43SJérôme Glisse * when a device driver is involved (you cannot easily 8405042db43SJérôme Glisse * save and restore device driver state). 8415042db43SJérôme Glisse */ 8424dd845b5SAlistair Popple if (is_writable_device_private_entry(entry) && 8435042db43SJérôme Glisse is_cow_mapping(vm_flags)) { 8444dd845b5SAlistair Popple entry = make_readable_device_private_entry( 8454dd845b5SAlistair Popple swp_offset(entry)); 8465042db43SJérôme Glisse pte = swp_entry_to_pte(entry); 847c33c7948SRyan Roberts if (pte_swp_uffd_wp(orig_pte)) 848f45ec5ffSPeter Xu pte = pte_swp_mkuffd_wp(pte); 8495042db43SJérôme Glisse set_pte_at(src_mm, addr, src_pte, pte); 8505042db43SJérôme Glisse } 851b756a3b5SAlistair Popple } else if (is_device_exclusive_entry(entry)) { 852b756a3b5SAlistair Popple /* 853b756a3b5SAlistair Popple * Make device exclusive entries present by restoring the 854b756a3b5SAlistair Popple * original entry then copying as for a present pte. Device 855b756a3b5SAlistair Popple * exclusive entries currently only support private writable 856b756a3b5SAlistair Popple * (ie. COW) mappings. 857b756a3b5SAlistair Popple */ 858b756a3b5SAlistair Popple VM_BUG_ON(!is_cow_mapping(src_vma->vm_flags)); 859b756a3b5SAlistair Popple if (try_restore_exclusive_pte(src_pte, src_vma, addr)) 860b756a3b5SAlistair Popple return -EBUSY; 861b756a3b5SAlistair Popple return -ENOENT; 862c56d1b62SPeter Xu } else if (is_pte_marker_entry(entry)) { 8637e3ce3f8SPeter Xu if (is_swapin_error_entry(entry) || userfaultfd_wp(dst_vma)) 864c56d1b62SPeter Xu set_pte_at(dst_mm, addr, dst_pte, pte); 865c56d1b62SPeter Xu return 0; 8661da177e4SLinus Torvalds } 8678f34f1eaSPeter Xu if (!userfaultfd_wp(dst_vma)) 8688f34f1eaSPeter Xu pte = pte_swp_clear_uffd_wp(pte); 869df3a57d1SLinus Torvalds set_pte_at(dst_mm, addr, dst_pte, pte); 870df3a57d1SLinus Torvalds return 0; 8711da177e4SLinus Torvalds } 8721da177e4SLinus Torvalds 87370e806e4SPeter Xu /* 874b51ad4f8SDavid Hildenbrand * Copy a present and normal page. 87570e806e4SPeter Xu * 876b51ad4f8SDavid Hildenbrand * NOTE! The usual case is that this isn't required; 877b51ad4f8SDavid Hildenbrand * instead, the caller can just increase the page refcount 878b51ad4f8SDavid Hildenbrand * and re-use the pte the traditional way. 87970e806e4SPeter Xu * 88070e806e4SPeter Xu * And if we need a pre-allocated page but don't yet have 88170e806e4SPeter Xu * one, return a negative error to let the preallocation 88270e806e4SPeter Xu * code know so that it can do so outside the page table 88370e806e4SPeter Xu * lock. 88470e806e4SPeter Xu */ 88570e806e4SPeter Xu static inline int 886c78f4636SPeter Xu copy_present_page(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 887c78f4636SPeter Xu pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, 888edf50470SMatthew Wilcox (Oracle) struct folio **prealloc, struct page *page) 88970e806e4SPeter Xu { 890edf50470SMatthew Wilcox (Oracle) struct folio *new_folio; 891b51ad4f8SDavid Hildenbrand pte_t pte; 89270e806e4SPeter Xu 893edf50470SMatthew Wilcox (Oracle) new_folio = *prealloc; 894edf50470SMatthew Wilcox (Oracle) if (!new_folio) 89570e806e4SPeter Xu return -EAGAIN; 89670e806e4SPeter Xu 89770e806e4SPeter Xu /* 89870e806e4SPeter Xu * We have a prealloc page, all good! Take it 89970e806e4SPeter Xu * over and copy the page & arm it. 90070e806e4SPeter Xu */ 90170e806e4SPeter Xu *prealloc = NULL; 902edf50470SMatthew Wilcox (Oracle) copy_user_highpage(&new_folio->page, page, addr, src_vma); 903edf50470SMatthew Wilcox (Oracle) __folio_mark_uptodate(new_folio); 904edf50470SMatthew Wilcox (Oracle) folio_add_new_anon_rmap(new_folio, dst_vma, addr); 905edf50470SMatthew Wilcox (Oracle) folio_add_lru_vma(new_folio, dst_vma); 906edf50470SMatthew Wilcox (Oracle) rss[MM_ANONPAGES]++; 90770e806e4SPeter Xu 90870e806e4SPeter Xu /* All done, just insert the new page copy in the child */ 909edf50470SMatthew Wilcox (Oracle) pte = mk_pte(&new_folio->page, dst_vma->vm_page_prot); 910c78f4636SPeter Xu pte = maybe_mkwrite(pte_mkdirty(pte), dst_vma); 911c33c7948SRyan Roberts if (userfaultfd_pte_wp(dst_vma, ptep_get(src_pte))) 9128f34f1eaSPeter Xu /* Uffd-wp needs to be delivered to dest pte as well */ 913f1eb1bacSPeter Xu pte = pte_mkuffd_wp(pte); 914c78f4636SPeter Xu set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); 91570e806e4SPeter Xu return 0; 91670e806e4SPeter Xu } 91770e806e4SPeter Xu 91870e806e4SPeter Xu /* 91970e806e4SPeter Xu * Copy one pte. Returns 0 if succeeded, or -EAGAIN if one preallocated page 92070e806e4SPeter Xu * is required to copy this pte. 92170e806e4SPeter Xu */ 92270e806e4SPeter Xu static inline int 923c78f4636SPeter Xu copy_present_pte(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 924c78f4636SPeter Xu pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, 925edf50470SMatthew Wilcox (Oracle) struct folio **prealloc) 926df3a57d1SLinus Torvalds { 927c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 928c78f4636SPeter Xu unsigned long vm_flags = src_vma->vm_flags; 929c33c7948SRyan Roberts pte_t pte = ptep_get(src_pte); 930df3a57d1SLinus Torvalds struct page *page; 93114ddee41SMatthew Wilcox (Oracle) struct folio *folio; 932df3a57d1SLinus Torvalds 933c78f4636SPeter Xu page = vm_normal_page(src_vma, addr, pte); 93414ddee41SMatthew Wilcox (Oracle) if (page) 93514ddee41SMatthew Wilcox (Oracle) folio = page_folio(page); 93614ddee41SMatthew Wilcox (Oracle) if (page && folio_test_anon(folio)) { 937b51ad4f8SDavid Hildenbrand /* 938b51ad4f8SDavid Hildenbrand * If this page may have been pinned by the parent process, 939b51ad4f8SDavid Hildenbrand * copy the page immediately for the child so that we'll always 940b51ad4f8SDavid Hildenbrand * guarantee the pinned page won't be randomly replaced in the 941b51ad4f8SDavid Hildenbrand * future. 942b51ad4f8SDavid Hildenbrand */ 94314ddee41SMatthew Wilcox (Oracle) folio_get(folio); 944fb3d824dSDavid Hildenbrand if (unlikely(page_try_dup_anon_rmap(page, false, src_vma))) { 945fb3d824dSDavid Hildenbrand /* Page may be pinned, we have to copy. */ 94614ddee41SMatthew Wilcox (Oracle) folio_put(folio); 947b51ad4f8SDavid Hildenbrand return copy_present_page(dst_vma, src_vma, dst_pte, src_pte, 948b51ad4f8SDavid Hildenbrand addr, rss, prealloc, page); 949fb3d824dSDavid Hildenbrand } 950edf50470SMatthew Wilcox (Oracle) rss[MM_ANONPAGES]++; 951b51ad4f8SDavid Hildenbrand } else if (page) { 95214ddee41SMatthew Wilcox (Oracle) folio_get(folio); 953fb3d824dSDavid Hildenbrand page_dup_file_rmap(page, false); 954edf50470SMatthew Wilcox (Oracle) rss[mm_counter_file(page)]++; 95570e806e4SPeter Xu } 95670e806e4SPeter Xu 9571da177e4SLinus Torvalds /* 9581da177e4SLinus Torvalds * If it's a COW mapping, write protect it both 9591da177e4SLinus Torvalds * in the parent and the child 9601da177e4SLinus Torvalds */ 9611b2de5d0SLinus Torvalds if (is_cow_mapping(vm_flags) && pte_write(pte)) { 9621da177e4SLinus Torvalds ptep_set_wrprotect(src_mm, addr, src_pte); 9633dc90795SZachary Amsden pte = pte_wrprotect(pte); 9641da177e4SLinus Torvalds } 96514ddee41SMatthew Wilcox (Oracle) VM_BUG_ON(page && folio_test_anon(folio) && PageAnonExclusive(page)); 9661da177e4SLinus Torvalds 9671da177e4SLinus Torvalds /* 9681da177e4SLinus Torvalds * If it's a shared mapping, mark it clean in 9691da177e4SLinus Torvalds * the child 9701da177e4SLinus Torvalds */ 9711da177e4SLinus Torvalds if (vm_flags & VM_SHARED) 9721da177e4SLinus Torvalds pte = pte_mkclean(pte); 9731da177e4SLinus Torvalds pte = pte_mkold(pte); 9746aab341eSLinus Torvalds 9758f34f1eaSPeter Xu if (!userfaultfd_wp(dst_vma)) 976b569a176SPeter Xu pte = pte_clear_uffd_wp(pte); 977b569a176SPeter Xu 978c78f4636SPeter Xu set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); 97970e806e4SPeter Xu return 0; 9806aab341eSLinus Torvalds } 981ae859762SHugh Dickins 982edf50470SMatthew Wilcox (Oracle) static inline struct folio *page_copy_prealloc(struct mm_struct *src_mm, 983edf50470SMatthew Wilcox (Oracle) struct vm_area_struct *vma, unsigned long addr) 98470e806e4SPeter Xu { 985edf50470SMatthew Wilcox (Oracle) struct folio *new_folio; 98670e806e4SPeter Xu 987edf50470SMatthew Wilcox (Oracle) new_folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma, addr, false); 988edf50470SMatthew Wilcox (Oracle) if (!new_folio) 98970e806e4SPeter Xu return NULL; 99070e806e4SPeter Xu 991edf50470SMatthew Wilcox (Oracle) if (mem_cgroup_charge(new_folio, src_mm, GFP_KERNEL)) { 992edf50470SMatthew Wilcox (Oracle) folio_put(new_folio); 99370e806e4SPeter Xu return NULL; 99470e806e4SPeter Xu } 995e601ded4SKefeng Wang folio_throttle_swaprate(new_folio, GFP_KERNEL); 99670e806e4SPeter Xu 997edf50470SMatthew Wilcox (Oracle) return new_folio; 9981da177e4SLinus Torvalds } 9991da177e4SLinus Torvalds 1000c78f4636SPeter Xu static int 1001c78f4636SPeter Xu copy_pte_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1002c78f4636SPeter Xu pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr, 1003c78f4636SPeter Xu unsigned long end) 10041da177e4SLinus Torvalds { 1005c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1006c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 1007c36987e2SDaisuke Nishimura pte_t *orig_src_pte, *orig_dst_pte; 10081da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 1009c33c7948SRyan Roberts pte_t ptent; 1010c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 101170e806e4SPeter Xu int progress, ret = 0; 1012d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 1013570a335bSHugh Dickins swp_entry_t entry = (swp_entry_t){0}; 1014edf50470SMatthew Wilcox (Oracle) struct folio *prealloc = NULL; 10151da177e4SLinus Torvalds 10161da177e4SLinus Torvalds again: 101770e806e4SPeter Xu progress = 0; 1018d559db08SKAMEZAWA Hiroyuki init_rss_vec(rss); 1019d559db08SKAMEZAWA Hiroyuki 10203db82b93SHugh Dickins /* 10213db82b93SHugh Dickins * copy_pmd_range()'s prior pmd_none_or_clear_bad(src_pmd), and the 10223db82b93SHugh Dickins * error handling here, assume that exclusive mmap_lock on dst and src 10233db82b93SHugh Dickins * protects anon from unexpected THP transitions; with shmem and file 10243db82b93SHugh Dickins * protected by mmap_lock-less collapse skipping areas with anon_vma 10253db82b93SHugh Dickins * (whereas vma_needs_copy() skips areas without anon_vma). A rework 10263db82b93SHugh Dickins * can remove such assumptions later, but this is good enough for now. 10273db82b93SHugh Dickins */ 1028c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 102970e806e4SPeter Xu if (!dst_pte) { 103070e806e4SPeter Xu ret = -ENOMEM; 103170e806e4SPeter Xu goto out; 103270e806e4SPeter Xu } 10333db82b93SHugh Dickins src_pte = pte_offset_map_nolock(src_mm, src_pmd, addr, &src_ptl); 10343db82b93SHugh Dickins if (!src_pte) { 10353db82b93SHugh Dickins pte_unmap_unlock(dst_pte, dst_ptl); 10363db82b93SHugh Dickins /* ret == 0 */ 10373db82b93SHugh Dickins goto out; 10383db82b93SHugh Dickins } 1039f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 1040c36987e2SDaisuke Nishimura orig_src_pte = src_pte; 1041c36987e2SDaisuke Nishimura orig_dst_pte = dst_pte; 10426606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 10431da177e4SLinus Torvalds 10441da177e4SLinus Torvalds do { 10451da177e4SLinus Torvalds /* 10461da177e4SLinus Torvalds * We are holding two locks at this point - either of them 10471da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 10481da177e4SLinus Torvalds */ 1049e040f218SHugh Dickins if (progress >= 32) { 1050e040f218SHugh Dickins progress = 0; 1051e040f218SHugh Dickins if (need_resched() || 105295c354feSNick Piggin spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) 10531da177e4SLinus Torvalds break; 1054e040f218SHugh Dickins } 1055c33c7948SRyan Roberts ptent = ptep_get(src_pte); 1056c33c7948SRyan Roberts if (pte_none(ptent)) { 10571da177e4SLinus Torvalds progress++; 10581da177e4SLinus Torvalds continue; 10591da177e4SLinus Torvalds } 1060c33c7948SRyan Roberts if (unlikely(!pte_present(ptent))) { 10619a5cc85cSAlistair Popple ret = copy_nonpresent_pte(dst_mm, src_mm, 106279a1971cSLinus Torvalds dst_pte, src_pte, 10638f34f1eaSPeter Xu dst_vma, src_vma, 10648f34f1eaSPeter Xu addr, rss); 10659a5cc85cSAlistair Popple if (ret == -EIO) { 1066c33c7948SRyan Roberts entry = pte_to_swp_entry(ptep_get(src_pte)); 1067570a335bSHugh Dickins break; 1068b756a3b5SAlistair Popple } else if (ret == -EBUSY) { 1069b756a3b5SAlistair Popple break; 1070b756a3b5SAlistair Popple } else if (!ret) { 10711da177e4SLinus Torvalds progress += 8; 107279a1971cSLinus Torvalds continue; 107379a1971cSLinus Torvalds } 1074b756a3b5SAlistair Popple 1075b756a3b5SAlistair Popple /* 1076b756a3b5SAlistair Popple * Device exclusive entry restored, continue by copying 1077b756a3b5SAlistair Popple * the now present pte. 1078b756a3b5SAlistair Popple */ 1079b756a3b5SAlistair Popple WARN_ON_ONCE(ret != -ENOENT); 1080b756a3b5SAlistair Popple } 108170e806e4SPeter Xu /* copy_present_pte() will clear `*prealloc' if consumed */ 1082c78f4636SPeter Xu ret = copy_present_pte(dst_vma, src_vma, dst_pte, src_pte, 1083c78f4636SPeter Xu addr, rss, &prealloc); 108470e806e4SPeter Xu /* 108570e806e4SPeter Xu * If we need a pre-allocated page for this pte, drop the 108670e806e4SPeter Xu * locks, allocate, and try again. 108770e806e4SPeter Xu */ 108870e806e4SPeter Xu if (unlikely(ret == -EAGAIN)) 108970e806e4SPeter Xu break; 109070e806e4SPeter Xu if (unlikely(prealloc)) { 109170e806e4SPeter Xu /* 109270e806e4SPeter Xu * pre-alloc page cannot be reused by next time so as 109370e806e4SPeter Xu * to strictly follow mempolicy (e.g., alloc_page_vma() 109470e806e4SPeter Xu * will allocate page according to address). This 109570e806e4SPeter Xu * could only happen if one pinned pte changed. 109670e806e4SPeter Xu */ 1097edf50470SMatthew Wilcox (Oracle) folio_put(prealloc); 109870e806e4SPeter Xu prealloc = NULL; 109970e806e4SPeter Xu } 110079a1971cSLinus Torvalds progress += 8; 11011da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 11021da177e4SLinus Torvalds 11036606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 11043db82b93SHugh Dickins pte_unmap_unlock(orig_src_pte, src_ptl); 1105d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(dst_mm, rss); 1106c36987e2SDaisuke Nishimura pte_unmap_unlock(orig_dst_pte, dst_ptl); 1107c74df32cSHugh Dickins cond_resched(); 1108570a335bSHugh Dickins 11099a5cc85cSAlistair Popple if (ret == -EIO) { 11109a5cc85cSAlistair Popple VM_WARN_ON_ONCE(!entry.val); 111170e806e4SPeter Xu if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) { 111270e806e4SPeter Xu ret = -ENOMEM; 111370e806e4SPeter Xu goto out; 111470e806e4SPeter Xu } 111570e806e4SPeter Xu entry.val = 0; 1116b756a3b5SAlistair Popple } else if (ret == -EBUSY) { 1117b756a3b5SAlistair Popple goto out; 11189a5cc85cSAlistair Popple } else if (ret == -EAGAIN) { 1119c78f4636SPeter Xu prealloc = page_copy_prealloc(src_mm, src_vma, addr); 112070e806e4SPeter Xu if (!prealloc) 1121570a335bSHugh Dickins return -ENOMEM; 11229a5cc85cSAlistair Popple } else if (ret) { 11239a5cc85cSAlistair Popple VM_WARN_ON_ONCE(1); 11249a5cc85cSAlistair Popple } 11259a5cc85cSAlistair Popple 112670e806e4SPeter Xu /* We've captured and resolved the error. Reset, try again. */ 112770e806e4SPeter Xu ret = 0; 11289a5cc85cSAlistair Popple 11291da177e4SLinus Torvalds if (addr != end) 11301da177e4SLinus Torvalds goto again; 113170e806e4SPeter Xu out: 113270e806e4SPeter Xu if (unlikely(prealloc)) 1133edf50470SMatthew Wilcox (Oracle) folio_put(prealloc); 113470e806e4SPeter Xu return ret; 11351da177e4SLinus Torvalds } 11361da177e4SLinus Torvalds 1137c78f4636SPeter Xu static inline int 1138c78f4636SPeter Xu copy_pmd_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1139c78f4636SPeter Xu pud_t *dst_pud, pud_t *src_pud, unsigned long addr, 1140c78f4636SPeter Xu unsigned long end) 11411da177e4SLinus Torvalds { 1142c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1143c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 11441da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 11451da177e4SLinus Torvalds unsigned long next; 11461da177e4SLinus Torvalds 11471da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 11481da177e4SLinus Torvalds if (!dst_pmd) 11491da177e4SLinus Torvalds return -ENOMEM; 11501da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 11511da177e4SLinus Torvalds do { 11521da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 115384c3fc4eSZi Yan if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) 115484c3fc4eSZi Yan || pmd_devmap(*src_pmd)) { 115571e3aac0SAndrea Arcangeli int err; 1156c78f4636SPeter Xu VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, src_vma); 11578f34f1eaSPeter Xu err = copy_huge_pmd(dst_mm, src_mm, dst_pmd, src_pmd, 11588f34f1eaSPeter Xu addr, dst_vma, src_vma); 115971e3aac0SAndrea Arcangeli if (err == -ENOMEM) 116071e3aac0SAndrea Arcangeli return -ENOMEM; 116171e3aac0SAndrea Arcangeli if (!err) 116271e3aac0SAndrea Arcangeli continue; 116371e3aac0SAndrea Arcangeli /* fall through */ 116471e3aac0SAndrea Arcangeli } 11651da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 11661da177e4SLinus Torvalds continue; 1167c78f4636SPeter Xu if (copy_pte_range(dst_vma, src_vma, dst_pmd, src_pmd, 1168c78f4636SPeter Xu addr, next)) 11691da177e4SLinus Torvalds return -ENOMEM; 11701da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 11711da177e4SLinus Torvalds return 0; 11721da177e4SLinus Torvalds } 11731da177e4SLinus Torvalds 1174c78f4636SPeter Xu static inline int 1175c78f4636SPeter Xu copy_pud_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1176c78f4636SPeter Xu p4d_t *dst_p4d, p4d_t *src_p4d, unsigned long addr, 1177c78f4636SPeter Xu unsigned long end) 11781da177e4SLinus Torvalds { 1179c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1180c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 11811da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 11821da177e4SLinus Torvalds unsigned long next; 11831da177e4SLinus Torvalds 1184c2febafcSKirill A. Shutemov dst_pud = pud_alloc(dst_mm, dst_p4d, addr); 11851da177e4SLinus Torvalds if (!dst_pud) 11861da177e4SLinus Torvalds return -ENOMEM; 1187c2febafcSKirill A. Shutemov src_pud = pud_offset(src_p4d, addr); 11881da177e4SLinus Torvalds do { 11891da177e4SLinus Torvalds next = pud_addr_end(addr, end); 1190a00cc7d9SMatthew Wilcox if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { 1191a00cc7d9SMatthew Wilcox int err; 1192a00cc7d9SMatthew Wilcox 1193c78f4636SPeter Xu VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, src_vma); 1194a00cc7d9SMatthew Wilcox err = copy_huge_pud(dst_mm, src_mm, 1195c78f4636SPeter Xu dst_pud, src_pud, addr, src_vma); 1196a00cc7d9SMatthew Wilcox if (err == -ENOMEM) 1197a00cc7d9SMatthew Wilcox return -ENOMEM; 1198a00cc7d9SMatthew Wilcox if (!err) 1199a00cc7d9SMatthew Wilcox continue; 1200a00cc7d9SMatthew Wilcox /* fall through */ 1201a00cc7d9SMatthew Wilcox } 12021da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 12031da177e4SLinus Torvalds continue; 1204c78f4636SPeter Xu if (copy_pmd_range(dst_vma, src_vma, dst_pud, src_pud, 1205c78f4636SPeter Xu addr, next)) 12061da177e4SLinus Torvalds return -ENOMEM; 12071da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 12081da177e4SLinus Torvalds return 0; 12091da177e4SLinus Torvalds } 12101da177e4SLinus Torvalds 1211c78f4636SPeter Xu static inline int 1212c78f4636SPeter Xu copy_p4d_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1213c78f4636SPeter Xu pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long addr, 1214c78f4636SPeter Xu unsigned long end) 1215c2febafcSKirill A. Shutemov { 1216c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1217c2febafcSKirill A. Shutemov p4d_t *src_p4d, *dst_p4d; 1218c2febafcSKirill A. Shutemov unsigned long next; 1219c2febafcSKirill A. Shutemov 1220c2febafcSKirill A. Shutemov dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); 1221c2febafcSKirill A. Shutemov if (!dst_p4d) 1222c2febafcSKirill A. Shutemov return -ENOMEM; 1223c2febafcSKirill A. Shutemov src_p4d = p4d_offset(src_pgd, addr); 1224c2febafcSKirill A. Shutemov do { 1225c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 1226c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(src_p4d)) 1227c2febafcSKirill A. Shutemov continue; 1228c78f4636SPeter Xu if (copy_pud_range(dst_vma, src_vma, dst_p4d, src_p4d, 1229c78f4636SPeter Xu addr, next)) 1230c2febafcSKirill A. Shutemov return -ENOMEM; 1231c2febafcSKirill A. Shutemov } while (dst_p4d++, src_p4d++, addr = next, addr != end); 1232c2febafcSKirill A. Shutemov return 0; 1233c2febafcSKirill A. Shutemov } 1234c2febafcSKirill A. Shutemov 1235c56d1b62SPeter Xu /* 1236c56d1b62SPeter Xu * Return true if the vma needs to copy the pgtable during this fork(). Return 1237c56d1b62SPeter Xu * false when we can speed up fork() by allowing lazy page faults later until 1238c56d1b62SPeter Xu * when the child accesses the memory range. 1239c56d1b62SPeter Xu */ 1240bc70fbf2SPeter Xu static bool 1241c56d1b62SPeter Xu vma_needs_copy(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) 1242c56d1b62SPeter Xu { 1243c56d1b62SPeter Xu /* 1244c56d1b62SPeter Xu * Always copy pgtables when dst_vma has uffd-wp enabled even if it's 1245c56d1b62SPeter Xu * file-backed (e.g. shmem). Because when uffd-wp is enabled, pgtable 1246c56d1b62SPeter Xu * contains uffd-wp protection information, that's something we can't 1247c56d1b62SPeter Xu * retrieve from page cache, and skip copying will lose those info. 1248c56d1b62SPeter Xu */ 1249c56d1b62SPeter Xu if (userfaultfd_wp(dst_vma)) 1250c56d1b62SPeter Xu return true; 1251c56d1b62SPeter Xu 1252bcd51a3cSMike Kravetz if (src_vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 1253c56d1b62SPeter Xu return true; 1254c56d1b62SPeter Xu 1255c56d1b62SPeter Xu if (src_vma->anon_vma) 1256c56d1b62SPeter Xu return true; 1257c56d1b62SPeter Xu 1258c56d1b62SPeter Xu /* 1259c56d1b62SPeter Xu * Don't copy ptes where a page fault will fill them correctly. Fork 1260c56d1b62SPeter Xu * becomes much lighter when there are big shared or private readonly 1261c56d1b62SPeter Xu * mappings. The tradeoff is that copy_page_range is more efficient 1262c56d1b62SPeter Xu * than faulting. 1263c56d1b62SPeter Xu */ 1264c56d1b62SPeter Xu return false; 1265c56d1b62SPeter Xu } 1266c56d1b62SPeter Xu 1267c78f4636SPeter Xu int 1268c78f4636SPeter Xu copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) 12691da177e4SLinus Torvalds { 12701da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 12711da177e4SLinus Torvalds unsigned long next; 1272c78f4636SPeter Xu unsigned long addr = src_vma->vm_start; 1273c78f4636SPeter Xu unsigned long end = src_vma->vm_end; 1274c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1275c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 1276ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 12772ec74c3eSSagi Grimberg bool is_cow; 1278cddb8a5cSAndrea Arcangeli int ret; 12791da177e4SLinus Torvalds 1280c56d1b62SPeter Xu if (!vma_needs_copy(dst_vma, src_vma)) 1281d992895bSNick Piggin return 0; 1282d992895bSNick Piggin 1283c78f4636SPeter Xu if (is_vm_hugetlb_page(src_vma)) 1284bc70fbf2SPeter Xu return copy_hugetlb_page_range(dst_mm, src_mm, dst_vma, src_vma); 12851da177e4SLinus Torvalds 1286c78f4636SPeter Xu if (unlikely(src_vma->vm_flags & VM_PFNMAP)) { 12872ab64037Svenkatesh.pallipadi@intel.com /* 12882ab64037Svenkatesh.pallipadi@intel.com * We do not free on error cases below as remove_vma 12892ab64037Svenkatesh.pallipadi@intel.com * gets called on error from higher level routine 12902ab64037Svenkatesh.pallipadi@intel.com */ 1291c78f4636SPeter Xu ret = track_pfn_copy(src_vma); 12922ab64037Svenkatesh.pallipadi@intel.com if (ret) 12932ab64037Svenkatesh.pallipadi@intel.com return ret; 12942ab64037Svenkatesh.pallipadi@intel.com } 12952ab64037Svenkatesh.pallipadi@intel.com 1296cddb8a5cSAndrea Arcangeli /* 1297cddb8a5cSAndrea Arcangeli * We need to invalidate the secondary MMU mappings only when 1298cddb8a5cSAndrea Arcangeli * there could be a permission downgrade on the ptes of the 1299cddb8a5cSAndrea Arcangeli * parent mm. And a permission downgrade will only happen if 1300cddb8a5cSAndrea Arcangeli * is_cow_mapping() returns true. 1301cddb8a5cSAndrea Arcangeli */ 1302c78f4636SPeter Xu is_cow = is_cow_mapping(src_vma->vm_flags); 1303ac46d4f3SJérôme Glisse 1304ac46d4f3SJérôme Glisse if (is_cow) { 13057269f999SJérôme Glisse mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_PAGE, 13067d4a8be0SAlistair Popple 0, src_mm, addr, end); 1307ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 130857efa1feSJason Gunthorpe /* 130957efa1feSJason Gunthorpe * Disabling preemption is not needed for the write side, as 131057efa1feSJason Gunthorpe * the read side doesn't spin, but goes to the mmap_lock. 131157efa1feSJason Gunthorpe * 131257efa1feSJason Gunthorpe * Use the raw variant of the seqcount_t write API to avoid 131357efa1feSJason Gunthorpe * lockdep complaining about preemptibility. 131457efa1feSJason Gunthorpe */ 131557efa1feSJason Gunthorpe mmap_assert_write_locked(src_mm); 131657efa1feSJason Gunthorpe raw_write_seqcount_begin(&src_mm->write_protect_seq); 1317ac46d4f3SJérôme Glisse } 1318cddb8a5cSAndrea Arcangeli 1319cddb8a5cSAndrea Arcangeli ret = 0; 13201da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 13211da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 13221da177e4SLinus Torvalds do { 13231da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 13241da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 13251da177e4SLinus Torvalds continue; 1326c78f4636SPeter Xu if (unlikely(copy_p4d_range(dst_vma, src_vma, dst_pgd, src_pgd, 1327c78f4636SPeter Xu addr, next))) { 1328d155df53SMa Wupeng untrack_pfn_clear(dst_vma); 1329cddb8a5cSAndrea Arcangeli ret = -ENOMEM; 1330cddb8a5cSAndrea Arcangeli break; 1331cddb8a5cSAndrea Arcangeli } 13321da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 1333cddb8a5cSAndrea Arcangeli 133457efa1feSJason Gunthorpe if (is_cow) { 133557efa1feSJason Gunthorpe raw_write_seqcount_end(&src_mm->write_protect_seq); 1336ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 133757efa1feSJason Gunthorpe } 1338cddb8a5cSAndrea Arcangeli return ret; 13391da177e4SLinus Torvalds } 13401da177e4SLinus Torvalds 13415abfd71dSPeter Xu /* Whether we should zap all COWed (private) pages too */ 13425abfd71dSPeter Xu static inline bool should_zap_cows(struct zap_details *details) 13435abfd71dSPeter Xu { 13445abfd71dSPeter Xu /* By default, zap all pages */ 13455abfd71dSPeter Xu if (!details) 13465abfd71dSPeter Xu return true; 13475abfd71dSPeter Xu 13485abfd71dSPeter Xu /* Or, we zap COWed pages only if the caller wants to */ 13492e148f1eSPeter Xu return details->even_cows; 13505abfd71dSPeter Xu } 13515abfd71dSPeter Xu 13522e148f1eSPeter Xu /* Decides whether we should zap this page with the page pointer specified */ 1353254ab940SPeter Xu static inline bool should_zap_page(struct zap_details *details, struct page *page) 13543506659eSMatthew Wilcox (Oracle) { 13555abfd71dSPeter Xu /* If we can make a decision without *page.. */ 13565abfd71dSPeter Xu if (should_zap_cows(details)) 1357254ab940SPeter Xu return true; 13583506659eSMatthew Wilcox (Oracle) 13595abfd71dSPeter Xu /* E.g. the caller passes NULL for the case of a zero page */ 13605abfd71dSPeter Xu if (!page) 1361254ab940SPeter Xu return true; 13625abfd71dSPeter Xu 13632e148f1eSPeter Xu /* Otherwise we should only zap non-anon pages */ 13642e148f1eSPeter Xu return !PageAnon(page); 13653506659eSMatthew Wilcox (Oracle) } 13663506659eSMatthew Wilcox (Oracle) 1367999dad82SPeter Xu static inline bool zap_drop_file_uffd_wp(struct zap_details *details) 1368999dad82SPeter Xu { 1369999dad82SPeter Xu if (!details) 1370999dad82SPeter Xu return false; 1371999dad82SPeter Xu 1372999dad82SPeter Xu return details->zap_flags & ZAP_FLAG_DROP_MARKER; 1373999dad82SPeter Xu } 1374999dad82SPeter Xu 1375999dad82SPeter Xu /* 1376999dad82SPeter Xu * This function makes sure that we'll replace the none pte with an uffd-wp 1377999dad82SPeter Xu * swap special pte marker when necessary. Must be with the pgtable lock held. 1378999dad82SPeter Xu */ 1379999dad82SPeter Xu static inline void 1380999dad82SPeter Xu zap_install_uffd_wp_if_needed(struct vm_area_struct *vma, 1381999dad82SPeter Xu unsigned long addr, pte_t *pte, 1382999dad82SPeter Xu struct zap_details *details, pte_t pteval) 1383999dad82SPeter Xu { 13842bad466cSPeter Xu /* Zap on anonymous always means dropping everything */ 13852bad466cSPeter Xu if (vma_is_anonymous(vma)) 13862bad466cSPeter Xu return; 13872bad466cSPeter Xu 1388999dad82SPeter Xu if (zap_drop_file_uffd_wp(details)) 1389999dad82SPeter Xu return; 1390999dad82SPeter Xu 1391999dad82SPeter Xu pte_install_uffd_wp_if_needed(vma, addr, pte, pteval); 1392999dad82SPeter Xu } 1393999dad82SPeter Xu 139451c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 1395b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 13961da177e4SLinus Torvalds unsigned long addr, unsigned long end, 139797a89413SPeter Zijlstra struct zap_details *details) 13981da177e4SLinus Torvalds { 1399b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 1400d16dfc55SPeter Zijlstra int force_flush = 0; 1401d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 140297a89413SPeter Zijlstra spinlock_t *ptl; 14035f1a1907SSteven Rostedt pte_t *start_pte; 140497a89413SPeter Zijlstra pte_t *pte; 14058a5f14a2SKirill A. Shutemov swp_entry_t entry; 1406d559db08SKAMEZAWA Hiroyuki 1407ed6a7935SPeter Zijlstra tlb_change_page_size(tlb, PAGE_SIZE); 1408e303297eSPeter Zijlstra init_rss_vec(rss); 14093db82b93SHugh Dickins start_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 14103db82b93SHugh Dickins if (!pte) 14113db82b93SHugh Dickins return addr; 14123db82b93SHugh Dickins 14133ea27719SMel Gorman flush_tlb_batched_pending(mm); 14146606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 14151da177e4SLinus Torvalds do { 1416c33c7948SRyan Roberts pte_t ptent = ptep_get(pte); 14178018db85SPeter Xu struct page *page; 14188018db85SPeter Xu 1419166f61b9STobin C Harding if (pte_none(ptent)) 14201da177e4SLinus Torvalds continue; 142151c6f666SRobin Holt 14227b167b68SMinchan Kim if (need_resched()) 14237b167b68SMinchan Kim break; 14247b167b68SMinchan Kim 14256f5e6b9eSHugh Dickins if (pte_present(ptent)) { 14265df397deSLinus Torvalds unsigned int delay_rmap; 14275df397deSLinus Torvalds 142825b2995aSChristoph Hellwig page = vm_normal_page(vma, addr, ptent); 1429254ab940SPeter Xu if (unlikely(!should_zap_page(details, page))) 14301da177e4SLinus Torvalds continue; 1431b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 1432a600388dSZachary Amsden tlb->fullmm); 14331da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 1434999dad82SPeter Xu zap_install_uffd_wp_if_needed(vma, addr, pte, details, 1435999dad82SPeter Xu ptent); 14361da177e4SLinus Torvalds if (unlikely(!page)) 14371da177e4SLinus Torvalds continue; 1438eca56ff9SJerome Marchand 14395df397deSLinus Torvalds delay_rmap = 0; 1440eca56ff9SJerome Marchand if (!PageAnon(page)) { 14411cf35d47SLinus Torvalds if (pte_dirty(ptent)) { 14426237bcd9SHugh Dickins set_page_dirty(page); 14435df397deSLinus Torvalds if (tlb_delay_rmap(tlb)) { 14445df397deSLinus Torvalds delay_rmap = 1; 14455df397deSLinus Torvalds force_flush = 1; 14465df397deSLinus Torvalds } 14471cf35d47SLinus Torvalds } 14488788f678SYu Zhao if (pte_young(ptent) && likely(vma_has_recency(vma))) 1449bf3f3bc5SNick Piggin mark_page_accessed(page); 14506237bcd9SHugh Dickins } 1451eca56ff9SJerome Marchand rss[mm_counter(page)]--; 14525df397deSLinus Torvalds if (!delay_rmap) { 1453cea86fe2SHugh Dickins page_remove_rmap(page, vma, false); 14543dc14741SHugh Dickins if (unlikely(page_mapcount(page) < 0)) 14553dc14741SHugh Dickins print_bad_pte(vma, addr, ptent, page); 14565df397deSLinus Torvalds } 14575df397deSLinus Torvalds if (unlikely(__tlb_remove_page(tlb, page, delay_rmap))) { 14581cf35d47SLinus Torvalds force_flush = 1; 1459ce9ec37bSWill Deacon addr += PAGE_SIZE; 1460d16dfc55SPeter Zijlstra break; 14611cf35d47SLinus Torvalds } 14621da177e4SLinus Torvalds continue; 14631da177e4SLinus Torvalds } 14645042db43SJérôme Glisse 14655042db43SJérôme Glisse entry = pte_to_swp_entry(ptent); 1466b756a3b5SAlistair Popple if (is_device_private_entry(entry) || 1467b756a3b5SAlistair Popple is_device_exclusive_entry(entry)) { 14688018db85SPeter Xu page = pfn_swap_entry_to_page(entry); 1469254ab940SPeter Xu if (unlikely(!should_zap_page(details, page))) 14705042db43SJérôme Glisse continue; 1471999dad82SPeter Xu /* 1472999dad82SPeter Xu * Both device private/exclusive mappings should only 1473999dad82SPeter Xu * work with anonymous page so far, so we don't need to 1474999dad82SPeter Xu * consider uffd-wp bit when zap. For more information, 1475999dad82SPeter Xu * see zap_install_uffd_wp_if_needed(). 1476999dad82SPeter Xu */ 1477999dad82SPeter Xu WARN_ON_ONCE(!vma_is_anonymous(vma)); 14785042db43SJérôme Glisse rss[mm_counter(page)]--; 1479b756a3b5SAlistair Popple if (is_device_private_entry(entry)) 1480cea86fe2SHugh Dickins page_remove_rmap(page, vma, false); 14815042db43SJérôme Glisse put_page(page); 14828018db85SPeter Xu } else if (!non_swap_entry(entry)) { 14835abfd71dSPeter Xu /* Genuine swap entry, hence a private anon page */ 14845abfd71dSPeter Xu if (!should_zap_cows(details)) 14851da177e4SLinus Torvalds continue; 1486b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 14878018db85SPeter Xu if (unlikely(!free_swap_and_cache(entry))) 14888018db85SPeter Xu print_bad_pte(vma, addr, ptent, NULL); 14895abfd71dSPeter Xu } else if (is_migration_entry(entry)) { 1490af5cdaf8SAlistair Popple page = pfn_swap_entry_to_page(entry); 1491254ab940SPeter Xu if (!should_zap_page(details, page)) 14925abfd71dSPeter Xu continue; 1493eca56ff9SJerome Marchand rss[mm_counter(page)]--; 1494999dad82SPeter Xu } else if (pte_marker_entry_uffd_wp(entry)) { 14952bad466cSPeter Xu /* 14962bad466cSPeter Xu * For anon: always drop the marker; for file: only 14972bad466cSPeter Xu * drop the marker if explicitly requested. 14982bad466cSPeter Xu */ 14992bad466cSPeter Xu if (!vma_is_anonymous(vma) && 15002bad466cSPeter Xu !zap_drop_file_uffd_wp(details)) 1501999dad82SPeter Xu continue; 15029f186f9eSMiaohe Lin } else if (is_hwpoison_entry(entry) || 15039f186f9eSMiaohe Lin is_swapin_error_entry(entry)) { 15045abfd71dSPeter Xu if (!should_zap_cows(details)) 15055abfd71dSPeter Xu continue; 15065abfd71dSPeter Xu } else { 15075abfd71dSPeter Xu /* We should have covered all the swap entry types */ 15085abfd71dSPeter Xu WARN_ON_ONCE(1); 15099f9f1acdSKonstantin Khlebnikov } 15109888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 1511999dad82SPeter Xu zap_install_uffd_wp_if_needed(vma, addr, pte, details, ptent); 151297a89413SPeter Zijlstra } while (pte++, addr += PAGE_SIZE, addr != end); 1513ae859762SHugh Dickins 1514d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 15156606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 151651c6f666SRobin Holt 15171cf35d47SLinus Torvalds /* Do the actual TLB flush before dropping ptl */ 15185df397deSLinus Torvalds if (force_flush) { 15191cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 15205df397deSLinus Torvalds tlb_flush_rmaps(tlb, vma); 15215df397deSLinus Torvalds } 15221cf35d47SLinus Torvalds pte_unmap_unlock(start_pte, ptl); 15231cf35d47SLinus Torvalds 15241cf35d47SLinus Torvalds /* 15251cf35d47SLinus Torvalds * If we forced a TLB flush (either due to running out of 15261cf35d47SLinus Torvalds * batch buffers or because we needed to flush dirty TLB 15271cf35d47SLinus Torvalds * entries before releasing the ptl), free the batched 15283db82b93SHugh Dickins * memory too. Come back again if we didn't do everything. 15291cf35d47SLinus Torvalds */ 15303db82b93SHugh Dickins if (force_flush) 1531fa0aafb8SPeter Zijlstra tlb_flush_mmu(tlb); 1532d16dfc55SPeter Zijlstra 153351c6f666SRobin Holt return addr; 15341da177e4SLinus Torvalds } 15351da177e4SLinus Torvalds 153651c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1537b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 15381da177e4SLinus Torvalds unsigned long addr, unsigned long end, 153997a89413SPeter Zijlstra struct zap_details *details) 15401da177e4SLinus Torvalds { 15411da177e4SLinus Torvalds pmd_t *pmd; 15421da177e4SLinus Torvalds unsigned long next; 15431da177e4SLinus Torvalds 15441da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 15451da177e4SLinus Torvalds do { 15461da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 154784c3fc4eSZi Yan if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { 154853406ed1SHugh Dickins if (next - addr != HPAGE_PMD_SIZE) 1549fd60775aSDavid Rientjes __split_huge_pmd(vma, pmd, addr, false, NULL); 15503db82b93SHugh Dickins else if (zap_huge_pmd(tlb, vma, pmd, addr)) { 15513db82b93SHugh Dickins addr = next; 15523db82b93SHugh Dickins continue; 15533db82b93SHugh Dickins } 155471e3aac0SAndrea Arcangeli /* fall through */ 15553506659eSMatthew Wilcox (Oracle) } else if (details && details->single_folio && 15563506659eSMatthew Wilcox (Oracle) folio_test_pmd_mappable(details->single_folio) && 155722061a1fSHugh Dickins next - addr == HPAGE_PMD_SIZE && pmd_none(*pmd)) { 155822061a1fSHugh Dickins spinlock_t *ptl = pmd_lock(tlb->mm, pmd); 155922061a1fSHugh Dickins /* 156022061a1fSHugh Dickins * Take and drop THP pmd lock so that we cannot return 156122061a1fSHugh Dickins * prematurely, while zap_huge_pmd() has cleared *pmd, 156222061a1fSHugh Dickins * but not yet decremented compound_mapcount(). 156322061a1fSHugh Dickins */ 156422061a1fSHugh Dickins spin_unlock(ptl); 156571e3aac0SAndrea Arcangeli } 15663db82b93SHugh Dickins if (pmd_none(*pmd)) { 15673db82b93SHugh Dickins addr = next; 15683db82b93SHugh Dickins continue; 15693db82b93SHugh Dickins } 15703db82b93SHugh Dickins addr = zap_pte_range(tlb, vma, pmd, addr, next, details); 15713db82b93SHugh Dickins if (addr != next) 15723db82b93SHugh Dickins pmd--; 15733db82b93SHugh Dickins } while (pmd++, cond_resched(), addr != end); 157451c6f666SRobin Holt 157551c6f666SRobin Holt return addr; 15761da177e4SLinus Torvalds } 15771da177e4SLinus Torvalds 157851c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1579c2febafcSKirill A. Shutemov struct vm_area_struct *vma, p4d_t *p4d, 15801da177e4SLinus Torvalds unsigned long addr, unsigned long end, 158197a89413SPeter Zijlstra struct zap_details *details) 15821da177e4SLinus Torvalds { 15831da177e4SLinus Torvalds pud_t *pud; 15841da177e4SLinus Torvalds unsigned long next; 15851da177e4SLinus Torvalds 1586c2febafcSKirill A. Shutemov pud = pud_offset(p4d, addr); 15871da177e4SLinus Torvalds do { 15881da177e4SLinus Torvalds next = pud_addr_end(addr, end); 1589a00cc7d9SMatthew Wilcox if (pud_trans_huge(*pud) || pud_devmap(*pud)) { 1590a00cc7d9SMatthew Wilcox if (next - addr != HPAGE_PUD_SIZE) { 159142fc5414SMichel Lespinasse mmap_assert_locked(tlb->mm); 1592a00cc7d9SMatthew Wilcox split_huge_pud(vma, pud, addr); 1593a00cc7d9SMatthew Wilcox } else if (zap_huge_pud(tlb, vma, pud, addr)) 1594a00cc7d9SMatthew Wilcox goto next; 1595a00cc7d9SMatthew Wilcox /* fall through */ 1596a00cc7d9SMatthew Wilcox } 159797a89413SPeter Zijlstra if (pud_none_or_clear_bad(pud)) 15981da177e4SLinus Torvalds continue; 159997a89413SPeter Zijlstra next = zap_pmd_range(tlb, vma, pud, addr, next, details); 1600a00cc7d9SMatthew Wilcox next: 1601a00cc7d9SMatthew Wilcox cond_resched(); 160297a89413SPeter Zijlstra } while (pud++, addr = next, addr != end); 160351c6f666SRobin Holt 160451c6f666SRobin Holt return addr; 16051da177e4SLinus Torvalds } 16061da177e4SLinus Torvalds 1607c2febafcSKirill A. Shutemov static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, 1608c2febafcSKirill A. Shutemov struct vm_area_struct *vma, pgd_t *pgd, 1609c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 1610c2febafcSKirill A. Shutemov struct zap_details *details) 1611c2febafcSKirill A. Shutemov { 1612c2febafcSKirill A. Shutemov p4d_t *p4d; 1613c2febafcSKirill A. Shutemov unsigned long next; 1614c2febafcSKirill A. Shutemov 1615c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, addr); 1616c2febafcSKirill A. Shutemov do { 1617c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 1618c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(p4d)) 1619c2febafcSKirill A. Shutemov continue; 1620c2febafcSKirill A. Shutemov next = zap_pud_range(tlb, vma, p4d, addr, next, details); 1621c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 1622c2febafcSKirill A. Shutemov 1623c2febafcSKirill A. Shutemov return addr; 1624c2febafcSKirill A. Shutemov } 1625c2febafcSKirill A. Shutemov 1626aac45363SMichal Hocko void unmap_page_range(struct mmu_gather *tlb, 162751c6f666SRobin Holt struct vm_area_struct *vma, 16281da177e4SLinus Torvalds unsigned long addr, unsigned long end, 162997a89413SPeter Zijlstra struct zap_details *details) 16301da177e4SLinus Torvalds { 16311da177e4SLinus Torvalds pgd_t *pgd; 16321da177e4SLinus Torvalds unsigned long next; 16331da177e4SLinus Torvalds 16341da177e4SLinus Torvalds BUG_ON(addr >= end); 16351da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 16361da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 16371da177e4SLinus Torvalds do { 16381da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 163997a89413SPeter Zijlstra if (pgd_none_or_clear_bad(pgd)) 16401da177e4SLinus Torvalds continue; 1641c2febafcSKirill A. Shutemov next = zap_p4d_range(tlb, vma, pgd, addr, next, details); 164297a89413SPeter Zijlstra } while (pgd++, addr = next, addr != end); 16431da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 16441da177e4SLinus Torvalds } 16451da177e4SLinus Torvalds 1646f5cc4eefSAl Viro 1647f5cc4eefSAl Viro static void unmap_single_vma(struct mmu_gather *tlb, 16481da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 16494f74d2c8SLinus Torvalds unsigned long end_addr, 165068f48381SSuren Baghdasaryan struct zap_details *details, bool mm_wr_locked) 16511da177e4SLinus Torvalds { 1652f5cc4eefSAl Viro unsigned long start = max(vma->vm_start, start_addr); 16531da177e4SLinus Torvalds unsigned long end; 16541da177e4SLinus Torvalds 16551da177e4SLinus Torvalds if (start >= vma->vm_end) 1656f5cc4eefSAl Viro return; 16571da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 16581da177e4SLinus Torvalds if (end <= vma->vm_start) 1659f5cc4eefSAl Viro return; 16601da177e4SLinus Torvalds 1661cbc91f71SSrikar Dronamraju if (vma->vm_file) 1662cbc91f71SSrikar Dronamraju uprobe_munmap(vma, start, end); 1663cbc91f71SSrikar Dronamraju 1664b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) 166568f48381SSuren Baghdasaryan untrack_pfn(vma, 0, 0, mm_wr_locked); 16662ab64037Svenkatesh.pallipadi@intel.com 16678b2a1238SAl Viro if (start != end) { 166851c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1669a137e1ccSAndi Kleen /* 1670a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1671a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1672a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 16737aa6b4adSDavidlohr Bueso * cleanup path of mmap_region. When 1674a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 16757aa6b4adSDavidlohr Bueso * mmap_region() nullifies vma->vm_file 1676a137e1ccSAndi Kleen * before calling this function to clean up. 1677a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1678a137e1ccSAndi Kleen * safe to do nothing in this case. 1679a137e1ccSAndi Kleen */ 168024669e58SAneesh Kumar K.V if (vma->vm_file) { 168105e90bd0SPeter Xu zap_flags_t zap_flags = details ? 168205e90bd0SPeter Xu details->zap_flags : 0; 168305e90bd0SPeter Xu __unmap_hugepage_range_final(tlb, vma, start, end, 168405e90bd0SPeter Xu NULL, zap_flags); 168524669e58SAneesh Kumar K.V } 168651c6f666SRobin Holt } else 1687038c7aa1SAl Viro unmap_page_range(tlb, vma, start, end, details); 168897a89413SPeter Zijlstra } 168951c6f666SRobin Holt } 16901da177e4SLinus Torvalds 1691f5cc4eefSAl Viro /** 1692f5cc4eefSAl Viro * unmap_vmas - unmap a range of memory covered by a list of vma's 1693f5cc4eefSAl Viro * @tlb: address of the caller's struct mmu_gather 1694763ecb03SLiam R. Howlett * @mt: the maple tree 1695f5cc4eefSAl Viro * @vma: the starting vma 1696f5cc4eefSAl Viro * @start_addr: virtual address at which to start unmapping 1697f5cc4eefSAl Viro * @end_addr: virtual address at which to end unmapping 1698f5cc4eefSAl Viro * 1699f5cc4eefSAl Viro * Unmap all pages in the vma list. 1700f5cc4eefSAl Viro * 1701f5cc4eefSAl Viro * Only addresses between `start' and `end' will be unmapped. 1702f5cc4eefSAl Viro * 1703f5cc4eefSAl Viro * The VMA list must be sorted in ascending virtual address order. 1704f5cc4eefSAl Viro * 1705f5cc4eefSAl Viro * unmap_vmas() assumes that the caller will flush the whole unmapped address 1706f5cc4eefSAl Viro * range after unmap_vmas() returns. So the only responsibility here is to 1707f5cc4eefSAl Viro * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 1708f5cc4eefSAl Viro * drops the lock and schedules. 1709f5cc4eefSAl Viro */ 1710763ecb03SLiam R. Howlett void unmap_vmas(struct mmu_gather *tlb, struct maple_tree *mt, 1711f5cc4eefSAl Viro struct vm_area_struct *vma, unsigned long start_addr, 171268f48381SSuren Baghdasaryan unsigned long end_addr, bool mm_wr_locked) 1713f5cc4eefSAl Viro { 1714ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1715999dad82SPeter Xu struct zap_details details = { 171604ada095SMike Kravetz .zap_flags = ZAP_FLAG_DROP_MARKER | ZAP_FLAG_UNMAP, 1717999dad82SPeter Xu /* Careful - we need to zap private pages too! */ 1718999dad82SPeter Xu .even_cows = true, 1719999dad82SPeter Xu }; 1720763ecb03SLiam R. Howlett MA_STATE(mas, mt, vma->vm_end, vma->vm_end); 1721f5cc4eefSAl Viro 17227d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma->vm_mm, 17236f4f13e8SJérôme Glisse start_addr, end_addr); 1724ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 1725763ecb03SLiam R. Howlett do { 172668f48381SSuren Baghdasaryan unmap_single_vma(tlb, vma, start_addr, end_addr, &details, 172768f48381SSuren Baghdasaryan mm_wr_locked); 1728763ecb03SLiam R. Howlett } while ((vma = mas_find(&mas, end_addr - 1)) != NULL); 1729ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 17301da177e4SLinus Torvalds } 17311da177e4SLinus Torvalds 17321da177e4SLinus Torvalds /** 1733f5cc4eefSAl Viro * zap_page_range_single - remove user pages in a given range 1734f5cc4eefSAl Viro * @vma: vm_area_struct holding the applicable pages 1735f5cc4eefSAl Viro * @address: starting address of pages to zap 1736f5cc4eefSAl Viro * @size: number of bytes to zap 17378a5f14a2SKirill A. Shutemov * @details: details of shared cache invalidation 1738f5cc4eefSAl Viro * 1739f5cc4eefSAl Viro * The range must fit into one VMA. 1740f5cc4eefSAl Viro */ 174121b85b09SMike Kravetz void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 1742f5cc4eefSAl Viro unsigned long size, struct zap_details *details) 1743f5cc4eefSAl Viro { 174421b85b09SMike Kravetz const unsigned long end = address + size; 1745ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1746f5cc4eefSAl Viro struct mmu_gather tlb; 1747f5cc4eefSAl Viro 1748f5cc4eefSAl Viro lru_add_drain(); 17497d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma->vm_mm, 175021b85b09SMike Kravetz address, end); 175121b85b09SMike Kravetz if (is_vm_hugetlb_page(vma)) 175221b85b09SMike Kravetz adjust_range_if_pmd_sharing_possible(vma, &range.start, 175321b85b09SMike Kravetz &range.end); 1754a72afd87SWill Deacon tlb_gather_mmu(&tlb, vma->vm_mm); 1755ac46d4f3SJérôme Glisse update_hiwater_rss(vma->vm_mm); 1756ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 175721b85b09SMike Kravetz /* 175821b85b09SMike Kravetz * unmap 'address-end' not 'range.start-range.end' as range 175921b85b09SMike Kravetz * could have been expanded for hugetlb pmd sharing. 176021b85b09SMike Kravetz */ 176168f48381SSuren Baghdasaryan unmap_single_vma(&tlb, vma, address, end, details, false); 1762ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 1763ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 17641da177e4SLinus Torvalds } 17651da177e4SLinus Torvalds 1766c627f9ccSJack Steiner /** 1767c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1768c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1769c627f9ccSJack Steiner * @address: starting address of pages to zap 1770c627f9ccSJack Steiner * @size: number of bytes to zap 1771c627f9ccSJack Steiner * 1772c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1773c627f9ccSJack Steiner * 1774c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1775c627f9ccSJack Steiner * 1776c627f9ccSJack Steiner */ 177727d036e3SLeon Romanovsky void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1778c627f9ccSJack Steiner unsigned long size) 1779c627f9ccSJack Steiner { 178088a35912SMiaohe Lin if (!range_in_vma(vma, address, address + size) || 1781c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 178227d036e3SLeon Romanovsky return; 178327d036e3SLeon Romanovsky 1784f5cc4eefSAl Viro zap_page_range_single(vma, address, size, NULL); 1785c627f9ccSJack Steiner } 1786c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1787c627f9ccSJack Steiner 17888cd3984dSArjun Roy static pmd_t *walk_to_pmd(struct mm_struct *mm, unsigned long addr) 1789c9cfcddfSLinus Torvalds { 1790c2febafcSKirill A. Shutemov pgd_t *pgd; 1791c2febafcSKirill A. Shutemov p4d_t *p4d; 1792c2febafcSKirill A. Shutemov pud_t *pud; 1793c2febafcSKirill A. Shutemov pmd_t *pmd; 1794c2febafcSKirill A. Shutemov 1795c2febafcSKirill A. Shutemov pgd = pgd_offset(mm, addr); 1796c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 1797c2febafcSKirill A. Shutemov if (!p4d) 1798c2febafcSKirill A. Shutemov return NULL; 1799c2febafcSKirill A. Shutemov pud = pud_alloc(mm, p4d, addr); 1800c2febafcSKirill A. Shutemov if (!pud) 1801c2febafcSKirill A. Shutemov return NULL; 1802c2febafcSKirill A. Shutemov pmd = pmd_alloc(mm, pud, addr); 1803c2febafcSKirill A. Shutemov if (!pmd) 1804c2febafcSKirill A. Shutemov return NULL; 1805c2febafcSKirill A. Shutemov 1806f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 18078cd3984dSArjun Roy return pmd; 18088cd3984dSArjun Roy } 18098cd3984dSArjun Roy 18108cd3984dSArjun Roy pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 18118cd3984dSArjun Roy spinlock_t **ptl) 18128cd3984dSArjun Roy { 18138cd3984dSArjun Roy pmd_t *pmd = walk_to_pmd(mm, addr); 18148cd3984dSArjun Roy 18158cd3984dSArjun Roy if (!pmd) 18168cd3984dSArjun Roy return NULL; 1817c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 1818c9cfcddfSLinus Torvalds } 1819c9cfcddfSLinus Torvalds 18208efd6f5bSArjun Roy static int validate_page_before_insert(struct page *page) 18218efd6f5bSArjun Roy { 18228efd6f5bSArjun Roy if (PageAnon(page) || PageSlab(page) || page_has_type(page)) 18238efd6f5bSArjun Roy return -EINVAL; 18248efd6f5bSArjun Roy flush_dcache_page(page); 18258efd6f5bSArjun Roy return 0; 18268efd6f5bSArjun Roy } 18278efd6f5bSArjun Roy 1828cea86fe2SHugh Dickins static int insert_page_into_pte_locked(struct vm_area_struct *vma, pte_t *pte, 18298efd6f5bSArjun Roy unsigned long addr, struct page *page, pgprot_t prot) 18308efd6f5bSArjun Roy { 1831c33c7948SRyan Roberts if (!pte_none(ptep_get(pte))) 18328efd6f5bSArjun Roy return -EBUSY; 18338efd6f5bSArjun Roy /* Ok, finally just insert the thing.. */ 18348efd6f5bSArjun Roy get_page(page); 1835f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, mm_counter_file(page)); 1836cea86fe2SHugh Dickins page_add_file_rmap(page, vma, false); 1837cea86fe2SHugh Dickins set_pte_at(vma->vm_mm, addr, pte, mk_pte(page, prot)); 18388efd6f5bSArjun Roy return 0; 18398efd6f5bSArjun Roy } 18408efd6f5bSArjun Roy 18411da177e4SLinus Torvalds /* 1842238f58d8SLinus Torvalds * This is the old fallback for page remapping. 1843238f58d8SLinus Torvalds * 1844238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 1845238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 1846238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 1847238f58d8SLinus Torvalds */ 1848423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 1849423bad60SNick Piggin struct page *page, pgprot_t prot) 1850238f58d8SLinus Torvalds { 1851238f58d8SLinus Torvalds int retval; 1852238f58d8SLinus Torvalds pte_t *pte; 1853238f58d8SLinus Torvalds spinlock_t *ptl; 1854238f58d8SLinus Torvalds 18558efd6f5bSArjun Roy retval = validate_page_before_insert(page); 18568efd6f5bSArjun Roy if (retval) 18575b4e655eSKAMEZAWA Hiroyuki goto out; 1858238f58d8SLinus Torvalds retval = -ENOMEM; 1859cea86fe2SHugh Dickins pte = get_locked_pte(vma->vm_mm, addr, &ptl); 1860238f58d8SLinus Torvalds if (!pte) 18615b4e655eSKAMEZAWA Hiroyuki goto out; 1862cea86fe2SHugh Dickins retval = insert_page_into_pte_locked(vma, pte, addr, page, prot); 1863238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 1864238f58d8SLinus Torvalds out: 1865238f58d8SLinus Torvalds return retval; 1866238f58d8SLinus Torvalds } 1867238f58d8SLinus Torvalds 18688cd3984dSArjun Roy #ifdef pte_index 1869cea86fe2SHugh Dickins static int insert_page_in_batch_locked(struct vm_area_struct *vma, pte_t *pte, 18708cd3984dSArjun Roy unsigned long addr, struct page *page, pgprot_t prot) 18718cd3984dSArjun Roy { 18728cd3984dSArjun Roy int err; 18738cd3984dSArjun Roy 18748cd3984dSArjun Roy if (!page_count(page)) 18758cd3984dSArjun Roy return -EINVAL; 18768cd3984dSArjun Roy err = validate_page_before_insert(page); 18777f70c2a6SArjun Roy if (err) 18787f70c2a6SArjun Roy return err; 1879cea86fe2SHugh Dickins return insert_page_into_pte_locked(vma, pte, addr, page, prot); 18808cd3984dSArjun Roy } 18818cd3984dSArjun Roy 18828cd3984dSArjun Roy /* insert_pages() amortizes the cost of spinlock operations 18838cd3984dSArjun Roy * when inserting pages in a loop. Arch *must* define pte_index. 18848cd3984dSArjun Roy */ 18858cd3984dSArjun Roy static int insert_pages(struct vm_area_struct *vma, unsigned long addr, 18868cd3984dSArjun Roy struct page **pages, unsigned long *num, pgprot_t prot) 18878cd3984dSArjun Roy { 18888cd3984dSArjun Roy pmd_t *pmd = NULL; 18897f70c2a6SArjun Roy pte_t *start_pte, *pte; 18907f70c2a6SArjun Roy spinlock_t *pte_lock; 18918cd3984dSArjun Roy struct mm_struct *const mm = vma->vm_mm; 18928cd3984dSArjun Roy unsigned long curr_page_idx = 0; 18938cd3984dSArjun Roy unsigned long remaining_pages_total = *num; 18948cd3984dSArjun Roy unsigned long pages_to_write_in_pmd; 18958cd3984dSArjun Roy int ret; 18968cd3984dSArjun Roy more: 18978cd3984dSArjun Roy ret = -EFAULT; 18988cd3984dSArjun Roy pmd = walk_to_pmd(mm, addr); 18998cd3984dSArjun Roy if (!pmd) 19008cd3984dSArjun Roy goto out; 19018cd3984dSArjun Roy 19028cd3984dSArjun Roy pages_to_write_in_pmd = min_t(unsigned long, 19038cd3984dSArjun Roy remaining_pages_total, PTRS_PER_PTE - pte_index(addr)); 19048cd3984dSArjun Roy 19058cd3984dSArjun Roy /* Allocate the PTE if necessary; takes PMD lock once only. */ 19068cd3984dSArjun Roy ret = -ENOMEM; 19078cd3984dSArjun Roy if (pte_alloc(mm, pmd)) 19088cd3984dSArjun Roy goto out; 19098cd3984dSArjun Roy 19108cd3984dSArjun Roy while (pages_to_write_in_pmd) { 19118cd3984dSArjun Roy int pte_idx = 0; 19128cd3984dSArjun Roy const int batch_size = min_t(int, pages_to_write_in_pmd, 8); 19138cd3984dSArjun Roy 19147f70c2a6SArjun Roy start_pte = pte_offset_map_lock(mm, pmd, addr, &pte_lock); 19153db82b93SHugh Dickins if (!start_pte) { 19163db82b93SHugh Dickins ret = -EFAULT; 19173db82b93SHugh Dickins goto out; 19183db82b93SHugh Dickins } 19197f70c2a6SArjun Roy for (pte = start_pte; pte_idx < batch_size; ++pte, ++pte_idx) { 1920cea86fe2SHugh Dickins int err = insert_page_in_batch_locked(vma, pte, 19218cd3984dSArjun Roy addr, pages[curr_page_idx], prot); 19228cd3984dSArjun Roy if (unlikely(err)) { 19237f70c2a6SArjun Roy pte_unmap_unlock(start_pte, pte_lock); 19248cd3984dSArjun Roy ret = err; 19258cd3984dSArjun Roy remaining_pages_total -= pte_idx; 19268cd3984dSArjun Roy goto out; 19278cd3984dSArjun Roy } 19288cd3984dSArjun Roy addr += PAGE_SIZE; 19298cd3984dSArjun Roy ++curr_page_idx; 19308cd3984dSArjun Roy } 19317f70c2a6SArjun Roy pte_unmap_unlock(start_pte, pte_lock); 19328cd3984dSArjun Roy pages_to_write_in_pmd -= batch_size; 19338cd3984dSArjun Roy remaining_pages_total -= batch_size; 19348cd3984dSArjun Roy } 19358cd3984dSArjun Roy if (remaining_pages_total) 19368cd3984dSArjun Roy goto more; 19378cd3984dSArjun Roy ret = 0; 19388cd3984dSArjun Roy out: 19398cd3984dSArjun Roy *num = remaining_pages_total; 19408cd3984dSArjun Roy return ret; 19418cd3984dSArjun Roy } 19428cd3984dSArjun Roy #endif /* ifdef pte_index */ 19438cd3984dSArjun Roy 19448cd3984dSArjun Roy /** 19458cd3984dSArjun Roy * vm_insert_pages - insert multiple pages into user vma, batching the pmd lock. 19468cd3984dSArjun Roy * @vma: user vma to map to 19478cd3984dSArjun Roy * @addr: target start user address of these pages 19488cd3984dSArjun Roy * @pages: source kernel pages 19498cd3984dSArjun Roy * @num: in: number of pages to map. out: number of pages that were *not* 19508cd3984dSArjun Roy * mapped. (0 means all pages were successfully mapped). 19518cd3984dSArjun Roy * 19528cd3984dSArjun Roy * Preferred over vm_insert_page() when inserting multiple pages. 19538cd3984dSArjun Roy * 19548cd3984dSArjun Roy * In case of error, we may have mapped a subset of the provided 19558cd3984dSArjun Roy * pages. It is the caller's responsibility to account for this case. 19568cd3984dSArjun Roy * 19578cd3984dSArjun Roy * The same restrictions apply as in vm_insert_page(). 19588cd3984dSArjun Roy */ 19598cd3984dSArjun Roy int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr, 19608cd3984dSArjun Roy struct page **pages, unsigned long *num) 19618cd3984dSArjun Roy { 19628cd3984dSArjun Roy #ifdef pte_index 19638cd3984dSArjun Roy const unsigned long end_addr = addr + (*num * PAGE_SIZE) - 1; 19648cd3984dSArjun Roy 19658cd3984dSArjun Roy if (addr < vma->vm_start || end_addr >= vma->vm_end) 19668cd3984dSArjun Roy return -EFAULT; 19678cd3984dSArjun Roy if (!(vma->vm_flags & VM_MIXEDMAP)) { 1968d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(vma->vm_mm)); 19698cd3984dSArjun Roy BUG_ON(vma->vm_flags & VM_PFNMAP); 19701c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_MIXEDMAP); 19718cd3984dSArjun Roy } 19728cd3984dSArjun Roy /* Defer page refcount checking till we're about to map that page. */ 19738cd3984dSArjun Roy return insert_pages(vma, addr, pages, num, vma->vm_page_prot); 19748cd3984dSArjun Roy #else 19758cd3984dSArjun Roy unsigned long idx = 0, pgcount = *num; 197645779b03STom Rix int err = -EINVAL; 19778cd3984dSArjun Roy 19788cd3984dSArjun Roy for (; idx < pgcount; ++idx) { 19798cd3984dSArjun Roy err = vm_insert_page(vma, addr + (PAGE_SIZE * idx), pages[idx]); 19808cd3984dSArjun Roy if (err) 19818cd3984dSArjun Roy break; 19828cd3984dSArjun Roy } 19838cd3984dSArjun Roy *num = pgcount - idx; 19848cd3984dSArjun Roy return err; 19858cd3984dSArjun Roy #endif /* ifdef pte_index */ 19868cd3984dSArjun Roy } 19878cd3984dSArjun Roy EXPORT_SYMBOL(vm_insert_pages); 19888cd3984dSArjun Roy 1989bfa5bf6dSRolf Eike Beer /** 1990bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 1991bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1992bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 1993bfa5bf6dSRolf Eike Beer * @page: source kernel page 1994bfa5bf6dSRolf Eike Beer * 1995a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 1996a145dd41SLinus Torvalds * into a user vma. 1997a145dd41SLinus Torvalds * 1998a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 1999a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 2000a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 20018dfcc9baSNick Piggin * (see split_page()). 2002a145dd41SLinus Torvalds * 2003a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 2004a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 2005a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 2006a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 2007a145dd41SLinus Torvalds * ask for a shared writable mapping! 2008a145dd41SLinus Torvalds * 2009a145dd41SLinus Torvalds * The page does not need to be reserved. 20104b6e1e37SKonstantin Khlebnikov * 20114b6e1e37SKonstantin Khlebnikov * Usually this function is called from f_op->mmap() handler 2012c1e8d7c6SMichel Lespinasse * under mm->mmap_lock write-lock, so it can change vma->vm_flags. 20134b6e1e37SKonstantin Khlebnikov * Caller must set VM_MIXEDMAP on vma if it wants to call this 20144b6e1e37SKonstantin Khlebnikov * function from other places, for example from page-fault handler. 2015a862f68aSMike Rapoport * 2016a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 2017a145dd41SLinus Torvalds */ 2018423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 2019423bad60SNick Piggin struct page *page) 2020a145dd41SLinus Torvalds { 2021a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 2022a145dd41SLinus Torvalds return -EFAULT; 2023a145dd41SLinus Torvalds if (!page_count(page)) 2024a145dd41SLinus Torvalds return -EINVAL; 20254b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & VM_MIXEDMAP)) { 2026d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(vma->vm_mm)); 20274b6e1e37SKonstantin Khlebnikov BUG_ON(vma->vm_flags & VM_PFNMAP); 20281c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_MIXEDMAP); 20294b6e1e37SKonstantin Khlebnikov } 2030423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 2031a145dd41SLinus Torvalds } 2032e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 2033a145dd41SLinus Torvalds 2034a667d745SSouptick Joarder /* 2035a667d745SSouptick Joarder * __vm_map_pages - maps range of kernel pages into user vma 2036a667d745SSouptick Joarder * @vma: user vma to map to 2037a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2038a667d745SSouptick Joarder * @num: number of pages in page array 2039a667d745SSouptick Joarder * @offset: user's requested vm_pgoff 2040a667d745SSouptick Joarder * 2041a667d745SSouptick Joarder * This allows drivers to map range of kernel pages into a user vma. 2042a667d745SSouptick Joarder * 2043a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2044a667d745SSouptick Joarder */ 2045a667d745SSouptick Joarder static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages, 2046a667d745SSouptick Joarder unsigned long num, unsigned long offset) 2047a667d745SSouptick Joarder { 2048a667d745SSouptick Joarder unsigned long count = vma_pages(vma); 2049a667d745SSouptick Joarder unsigned long uaddr = vma->vm_start; 2050a667d745SSouptick Joarder int ret, i; 2051a667d745SSouptick Joarder 2052a667d745SSouptick Joarder /* Fail if the user requested offset is beyond the end of the object */ 205396756fcbSMiguel Ojeda if (offset >= num) 2054a667d745SSouptick Joarder return -ENXIO; 2055a667d745SSouptick Joarder 2056a667d745SSouptick Joarder /* Fail if the user requested size exceeds available object size */ 2057a667d745SSouptick Joarder if (count > num - offset) 2058a667d745SSouptick Joarder return -ENXIO; 2059a667d745SSouptick Joarder 2060a667d745SSouptick Joarder for (i = 0; i < count; i++) { 2061a667d745SSouptick Joarder ret = vm_insert_page(vma, uaddr, pages[offset + i]); 2062a667d745SSouptick Joarder if (ret < 0) 2063a667d745SSouptick Joarder return ret; 2064a667d745SSouptick Joarder uaddr += PAGE_SIZE; 2065a667d745SSouptick Joarder } 2066a667d745SSouptick Joarder 2067a667d745SSouptick Joarder return 0; 2068a667d745SSouptick Joarder } 2069a667d745SSouptick Joarder 2070a667d745SSouptick Joarder /** 2071a667d745SSouptick Joarder * vm_map_pages - maps range of kernel pages starts with non zero offset 2072a667d745SSouptick Joarder * @vma: user vma to map to 2073a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2074a667d745SSouptick Joarder * @num: number of pages in page array 2075a667d745SSouptick Joarder * 2076a667d745SSouptick Joarder * Maps an object consisting of @num pages, catering for the user's 2077a667d745SSouptick Joarder * requested vm_pgoff 2078a667d745SSouptick Joarder * 2079a667d745SSouptick Joarder * If we fail to insert any page into the vma, the function will return 2080a667d745SSouptick Joarder * immediately leaving any previously inserted pages present. Callers 2081a667d745SSouptick Joarder * from the mmap handler may immediately return the error as their caller 2082a667d745SSouptick Joarder * will destroy the vma, removing any successfully inserted pages. Other 2083a667d745SSouptick Joarder * callers should make their own arrangements for calling unmap_region(). 2084a667d745SSouptick Joarder * 2085a667d745SSouptick Joarder * Context: Process context. Called by mmap handlers. 2086a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2087a667d745SSouptick Joarder */ 2088a667d745SSouptick Joarder int vm_map_pages(struct vm_area_struct *vma, struct page **pages, 2089a667d745SSouptick Joarder unsigned long num) 2090a667d745SSouptick Joarder { 2091a667d745SSouptick Joarder return __vm_map_pages(vma, pages, num, vma->vm_pgoff); 2092a667d745SSouptick Joarder } 2093a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages); 2094a667d745SSouptick Joarder 2095a667d745SSouptick Joarder /** 2096a667d745SSouptick Joarder * vm_map_pages_zero - map range of kernel pages starts with zero offset 2097a667d745SSouptick Joarder * @vma: user vma to map to 2098a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2099a667d745SSouptick Joarder * @num: number of pages in page array 2100a667d745SSouptick Joarder * 2101a667d745SSouptick Joarder * Similar to vm_map_pages(), except that it explicitly sets the offset 2102a667d745SSouptick Joarder * to 0. This function is intended for the drivers that did not consider 2103a667d745SSouptick Joarder * vm_pgoff. 2104a667d745SSouptick Joarder * 2105a667d745SSouptick Joarder * Context: Process context. Called by mmap handlers. 2106a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2107a667d745SSouptick Joarder */ 2108a667d745SSouptick Joarder int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, 2109a667d745SSouptick Joarder unsigned long num) 2110a667d745SSouptick Joarder { 2111a667d745SSouptick Joarder return __vm_map_pages(vma, pages, num, 0); 2112a667d745SSouptick Joarder } 2113a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages_zero); 2114a667d745SSouptick Joarder 21159b5a8e00SMatthew Wilcox static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2116b2770da6SRoss Zwisler pfn_t pfn, pgprot_t prot, bool mkwrite) 2117423bad60SNick Piggin { 2118423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 2119423bad60SNick Piggin pte_t *pte, entry; 2120423bad60SNick Piggin spinlock_t *ptl; 2121423bad60SNick Piggin 2122423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 2123423bad60SNick Piggin if (!pte) 21249b5a8e00SMatthew Wilcox return VM_FAULT_OOM; 2125c33c7948SRyan Roberts entry = ptep_get(pte); 2126c33c7948SRyan Roberts if (!pte_none(entry)) { 2127b2770da6SRoss Zwisler if (mkwrite) { 2128b2770da6SRoss Zwisler /* 2129b2770da6SRoss Zwisler * For read faults on private mappings the PFN passed 2130b2770da6SRoss Zwisler * in may not match the PFN we have mapped if the 2131b2770da6SRoss Zwisler * mapped PFN is a writeable COW page. In the mkwrite 2132b2770da6SRoss Zwisler * case we are creating a writable PTE for a shared 2133f2c57d91SJan Kara * mapping and we expect the PFNs to match. If they 2134f2c57d91SJan Kara * don't match, we are likely racing with block 2135f2c57d91SJan Kara * allocation and mapping invalidation so just skip the 2136f2c57d91SJan Kara * update. 2137b2770da6SRoss Zwisler */ 2138c33c7948SRyan Roberts if (pte_pfn(entry) != pfn_t_to_pfn(pfn)) { 2139c33c7948SRyan Roberts WARN_ON_ONCE(!is_zero_pfn(pte_pfn(entry))); 2140423bad60SNick Piggin goto out_unlock; 2141f2c57d91SJan Kara } 2142c33c7948SRyan Roberts entry = pte_mkyoung(entry); 2143cae85cb8SJan Kara entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2144cae85cb8SJan Kara if (ptep_set_access_flags(vma, addr, pte, entry, 1)) 2145cae85cb8SJan Kara update_mmu_cache(vma, addr, pte); 2146cae85cb8SJan Kara } 2147b2770da6SRoss Zwisler goto out_unlock; 2148b2770da6SRoss Zwisler } 2149423bad60SNick Piggin 2150423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 215101c8f1c4SDan Williams if (pfn_t_devmap(pfn)) 215201c8f1c4SDan Williams entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); 215301c8f1c4SDan Williams else 215401c8f1c4SDan Williams entry = pte_mkspecial(pfn_t_pte(pfn, prot)); 2155b2770da6SRoss Zwisler 2156b2770da6SRoss Zwisler if (mkwrite) { 2157b2770da6SRoss Zwisler entry = pte_mkyoung(entry); 2158b2770da6SRoss Zwisler entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2159b2770da6SRoss Zwisler } 2160b2770da6SRoss Zwisler 2161423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 21624b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 2163423bad60SNick Piggin 2164423bad60SNick Piggin out_unlock: 2165423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 21669b5a8e00SMatthew Wilcox return VM_FAULT_NOPAGE; 2167423bad60SNick Piggin } 2168423bad60SNick Piggin 2169f5e6d1d5SMatthew Wilcox /** 2170f5e6d1d5SMatthew Wilcox * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot 2171f5e6d1d5SMatthew Wilcox * @vma: user vma to map to 2172f5e6d1d5SMatthew Wilcox * @addr: target user address of this page 2173f5e6d1d5SMatthew Wilcox * @pfn: source kernel pfn 2174f5e6d1d5SMatthew Wilcox * @pgprot: pgprot flags for the inserted page 2175f5e6d1d5SMatthew Wilcox * 2176a1a0aea5SRandy Dunlap * This is exactly like vmf_insert_pfn(), except that it allows drivers 2177f5e6d1d5SMatthew Wilcox * to override pgprot on a per-page basis. 2178f5e6d1d5SMatthew Wilcox * 2179f5e6d1d5SMatthew Wilcox * This only makes sense for IO mappings, and it makes no sense for 2180f5e6d1d5SMatthew Wilcox * COW mappings. In general, using multiple vmas is preferable; 2181ae2b01f3SMatthew Wilcox * vmf_insert_pfn_prot should only be used if using multiple VMAs is 2182f5e6d1d5SMatthew Wilcox * impractical. 2183f5e6d1d5SMatthew Wilcox * 218428d8b812SLorenzo Stoakes * pgprot typically only differs from @vma->vm_page_prot when drivers set 218528d8b812SLorenzo Stoakes * caching- and encryption bits different than those of @vma->vm_page_prot, 218628d8b812SLorenzo Stoakes * because the caching- or encryption mode may not be known at mmap() time. 218728d8b812SLorenzo Stoakes * 218828d8b812SLorenzo Stoakes * This is ok as long as @vma->vm_page_prot is not used by the core vm 218928d8b812SLorenzo Stoakes * to set caching and encryption bits for those vmas (except for COW pages). 219028d8b812SLorenzo Stoakes * This is ensured by core vm only modifying these page table entries using 219128d8b812SLorenzo Stoakes * functions that don't touch caching- or encryption bits, using pte_modify() 219228d8b812SLorenzo Stoakes * if needed. (See for example mprotect()). 219328d8b812SLorenzo Stoakes * 219428d8b812SLorenzo Stoakes * Also when new page-table entries are created, this is only done using the 219528d8b812SLorenzo Stoakes * fault() callback, and never using the value of vma->vm_page_prot, 219628d8b812SLorenzo Stoakes * except for page-table entries that point to anonymous pages as the result 219728d8b812SLorenzo Stoakes * of COW. 2198574c5b3dSThomas Hellstrom * 2199ae2b01f3SMatthew Wilcox * Context: Process context. May allocate using %GFP_KERNEL. 2200f5e6d1d5SMatthew Wilcox * Return: vm_fault_t value. 2201f5e6d1d5SMatthew Wilcox */ 2202f5e6d1d5SMatthew Wilcox vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, 2203f5e6d1d5SMatthew Wilcox unsigned long pfn, pgprot_t pgprot) 2204f5e6d1d5SMatthew Wilcox { 22056d958546SMatthew Wilcox /* 22066d958546SMatthew Wilcox * Technically, architectures with pte_special can avoid all these 22076d958546SMatthew Wilcox * restrictions (same for remap_pfn_range). However we would like 22086d958546SMatthew Wilcox * consistency in testing and feature parity among all, so we should 22096d958546SMatthew Wilcox * try to keep these invariants in place for everybody. 22106d958546SMatthew Wilcox */ 22116d958546SMatthew Wilcox BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 22126d958546SMatthew Wilcox BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 22136d958546SMatthew Wilcox (VM_PFNMAP|VM_MIXEDMAP)); 22146d958546SMatthew Wilcox BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 22156d958546SMatthew Wilcox BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 22166d958546SMatthew Wilcox 22176d958546SMatthew Wilcox if (addr < vma->vm_start || addr >= vma->vm_end) 22186d958546SMatthew Wilcox return VM_FAULT_SIGBUS; 22196d958546SMatthew Wilcox 22206d958546SMatthew Wilcox if (!pfn_modify_allowed(pfn, pgprot)) 22216d958546SMatthew Wilcox return VM_FAULT_SIGBUS; 22226d958546SMatthew Wilcox 22236d958546SMatthew Wilcox track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); 22246d958546SMatthew Wilcox 22259b5a8e00SMatthew Wilcox return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, 22266d958546SMatthew Wilcox false); 2227f5e6d1d5SMatthew Wilcox } 2228f5e6d1d5SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_pfn_prot); 2229e0dc0d8fSNick Piggin 2230ae2b01f3SMatthew Wilcox /** 2231ae2b01f3SMatthew Wilcox * vmf_insert_pfn - insert single pfn into user vma 2232ae2b01f3SMatthew Wilcox * @vma: user vma to map to 2233ae2b01f3SMatthew Wilcox * @addr: target user address of this page 2234ae2b01f3SMatthew Wilcox * @pfn: source kernel pfn 2235ae2b01f3SMatthew Wilcox * 2236ae2b01f3SMatthew Wilcox * Similar to vm_insert_page, this allows drivers to insert individual pages 2237ae2b01f3SMatthew Wilcox * they've allocated into a user vma. Same comments apply. 2238ae2b01f3SMatthew Wilcox * 2239ae2b01f3SMatthew Wilcox * This function should only be called from a vm_ops->fault handler, and 2240ae2b01f3SMatthew Wilcox * in that case the handler should return the result of this function. 2241ae2b01f3SMatthew Wilcox * 2242ae2b01f3SMatthew Wilcox * vma cannot be a COW mapping. 2243ae2b01f3SMatthew Wilcox * 2244ae2b01f3SMatthew Wilcox * As this is called only for pages that do not currently exist, we 2245ae2b01f3SMatthew Wilcox * do not need to flush old virtual caches or the TLB. 2246ae2b01f3SMatthew Wilcox * 2247ae2b01f3SMatthew Wilcox * Context: Process context. May allocate using %GFP_KERNEL. 2248ae2b01f3SMatthew Wilcox * Return: vm_fault_t value. 2249ae2b01f3SMatthew Wilcox */ 2250ae2b01f3SMatthew Wilcox vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2251ae2b01f3SMatthew Wilcox unsigned long pfn) 2252ae2b01f3SMatthew Wilcox { 2253ae2b01f3SMatthew Wilcox return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); 2254ae2b01f3SMatthew Wilcox } 2255ae2b01f3SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_pfn); 2256ae2b01f3SMatthew Wilcox 2257785a3fabSDan Williams static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) 2258785a3fabSDan Williams { 2259785a3fabSDan Williams /* these checks mirror the abort conditions in vm_normal_page */ 2260785a3fabSDan Williams if (vma->vm_flags & VM_MIXEDMAP) 2261785a3fabSDan Williams return true; 2262785a3fabSDan Williams if (pfn_t_devmap(pfn)) 2263785a3fabSDan Williams return true; 2264785a3fabSDan Williams if (pfn_t_special(pfn)) 2265785a3fabSDan Williams return true; 2266785a3fabSDan Williams if (is_zero_pfn(pfn_t_to_pfn(pfn))) 2267785a3fabSDan Williams return true; 2268785a3fabSDan Williams return false; 2269785a3fabSDan Williams } 2270785a3fabSDan Williams 227179f3aa5bSMatthew Wilcox static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma, 227228d8b812SLorenzo Stoakes unsigned long addr, pfn_t pfn, bool mkwrite) 2273423bad60SNick Piggin { 227428d8b812SLorenzo Stoakes pgprot_t pgprot = vma->vm_page_prot; 227579f3aa5bSMatthew Wilcox int err; 227687744ab3SDan Williams 2277785a3fabSDan Williams BUG_ON(!vm_mixed_ok(vma, pfn)); 2278423bad60SNick Piggin 2279423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 228079f3aa5bSMatthew Wilcox return VM_FAULT_SIGBUS; 2281308a047cSBorislav Petkov 2282308a047cSBorislav Petkov track_pfn_insert(vma, &pgprot, pfn); 2283423bad60SNick Piggin 228442e4089cSAndi Kleen if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) 228579f3aa5bSMatthew Wilcox return VM_FAULT_SIGBUS; 228642e4089cSAndi Kleen 2287423bad60SNick Piggin /* 2288423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 2289423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 2290423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 229162eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 229262eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 2293423bad60SNick Piggin */ 229400b3a331SLaurent Dufour if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && 229500b3a331SLaurent Dufour !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { 2296423bad60SNick Piggin struct page *page; 2297423bad60SNick Piggin 229803fc2da6SDan Williams /* 229903fc2da6SDan Williams * At this point we are committed to insert_page() 230003fc2da6SDan Williams * regardless of whether the caller specified flags that 230103fc2da6SDan Williams * result in pfn_t_has_page() == false. 230203fc2da6SDan Williams */ 230303fc2da6SDan Williams page = pfn_to_page(pfn_t_to_pfn(pfn)); 230479f3aa5bSMatthew Wilcox err = insert_page(vma, addr, page, pgprot); 230579f3aa5bSMatthew Wilcox } else { 23069b5a8e00SMatthew Wilcox return insert_pfn(vma, addr, pfn, pgprot, mkwrite); 2307423bad60SNick Piggin } 2308b2770da6SRoss Zwisler 23095d747637SMatthew Wilcox if (err == -ENOMEM) 23105d747637SMatthew Wilcox return VM_FAULT_OOM; 23115d747637SMatthew Wilcox if (err < 0 && err != -EBUSY) 23125d747637SMatthew Wilcox return VM_FAULT_SIGBUS; 23135d747637SMatthew Wilcox 23145d747637SMatthew Wilcox return VM_FAULT_NOPAGE; 2315423bad60SNick Piggin } 231679f3aa5bSMatthew Wilcox 231779f3aa5bSMatthew Wilcox vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 231879f3aa5bSMatthew Wilcox pfn_t pfn) 231979f3aa5bSMatthew Wilcox { 232028d8b812SLorenzo Stoakes return __vm_insert_mixed(vma, addr, pfn, false); 232179f3aa5bSMatthew Wilcox } 23225d747637SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_mixed); 2323423bad60SNick Piggin 2324ab77dab4SSouptick Joarder /* 2325ab77dab4SSouptick Joarder * If the insertion of PTE failed because someone else already added a 2326ab77dab4SSouptick Joarder * different entry in the mean time, we treat that as success as we assume 2327ab77dab4SSouptick Joarder * the same entry was actually inserted. 2328ab77dab4SSouptick Joarder */ 2329ab77dab4SSouptick Joarder vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, 2330ab77dab4SSouptick Joarder unsigned long addr, pfn_t pfn) 2331b2770da6SRoss Zwisler { 233228d8b812SLorenzo Stoakes return __vm_insert_mixed(vma, addr, pfn, true); 2333b2770da6SRoss Zwisler } 2334ab77dab4SSouptick Joarder EXPORT_SYMBOL(vmf_insert_mixed_mkwrite); 2335b2770da6SRoss Zwisler 2336a145dd41SLinus Torvalds /* 23371da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 23381da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 23391da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 23401da177e4SLinus Torvalds */ 23411da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 23421da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23431da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23441da177e4SLinus Torvalds { 234590a3e375SMiaohe Lin pte_t *pte, *mapped_pte; 2346c74df32cSHugh Dickins spinlock_t *ptl; 234742e4089cSAndi Kleen int err = 0; 23481da177e4SLinus Torvalds 234990a3e375SMiaohe Lin mapped_pte = pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 23501da177e4SLinus Torvalds if (!pte) 23511da177e4SLinus Torvalds return -ENOMEM; 23526606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 23531da177e4SLinus Torvalds do { 2354c33c7948SRyan Roberts BUG_ON(!pte_none(ptep_get(pte))); 235542e4089cSAndi Kleen if (!pfn_modify_allowed(pfn, prot)) { 235642e4089cSAndi Kleen err = -EACCES; 235742e4089cSAndi Kleen break; 235842e4089cSAndi Kleen } 23597e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 23601da177e4SLinus Torvalds pfn++; 23611da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 23626606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 236390a3e375SMiaohe Lin pte_unmap_unlock(mapped_pte, ptl); 236442e4089cSAndi Kleen return err; 23651da177e4SLinus Torvalds } 23661da177e4SLinus Torvalds 23671da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 23681da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23691da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23701da177e4SLinus Torvalds { 23711da177e4SLinus Torvalds pmd_t *pmd; 23721da177e4SLinus Torvalds unsigned long next; 237342e4089cSAndi Kleen int err; 23741da177e4SLinus Torvalds 23751da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 23761da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 23771da177e4SLinus Torvalds if (!pmd) 23781da177e4SLinus Torvalds return -ENOMEM; 2379f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 23801da177e4SLinus Torvalds do { 23811da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 238242e4089cSAndi Kleen err = remap_pte_range(mm, pmd, addr, next, 238342e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 238442e4089cSAndi Kleen if (err) 238542e4089cSAndi Kleen return err; 23861da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 23871da177e4SLinus Torvalds return 0; 23881da177e4SLinus Torvalds } 23891da177e4SLinus Torvalds 2390c2febafcSKirill A. Shutemov static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, 23911da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23921da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23931da177e4SLinus Torvalds { 23941da177e4SLinus Torvalds pud_t *pud; 23951da177e4SLinus Torvalds unsigned long next; 239642e4089cSAndi Kleen int err; 23971da177e4SLinus Torvalds 23981da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 2399c2febafcSKirill A. Shutemov pud = pud_alloc(mm, p4d, addr); 24001da177e4SLinus Torvalds if (!pud) 24011da177e4SLinus Torvalds return -ENOMEM; 24021da177e4SLinus Torvalds do { 24031da177e4SLinus Torvalds next = pud_addr_end(addr, end); 240442e4089cSAndi Kleen err = remap_pmd_range(mm, pud, addr, next, 240542e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 240642e4089cSAndi Kleen if (err) 240742e4089cSAndi Kleen return err; 24081da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 24091da177e4SLinus Torvalds return 0; 24101da177e4SLinus Torvalds } 24111da177e4SLinus Torvalds 2412c2febafcSKirill A. Shutemov static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, 2413c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 2414c2febafcSKirill A. Shutemov unsigned long pfn, pgprot_t prot) 2415c2febafcSKirill A. Shutemov { 2416c2febafcSKirill A. Shutemov p4d_t *p4d; 2417c2febafcSKirill A. Shutemov unsigned long next; 241842e4089cSAndi Kleen int err; 2419c2febafcSKirill A. Shutemov 2420c2febafcSKirill A. Shutemov pfn -= addr >> PAGE_SHIFT; 2421c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 2422c2febafcSKirill A. Shutemov if (!p4d) 2423c2febafcSKirill A. Shutemov return -ENOMEM; 2424c2febafcSKirill A. Shutemov do { 2425c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 242642e4089cSAndi Kleen err = remap_pud_range(mm, p4d, addr, next, 242742e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 242842e4089cSAndi Kleen if (err) 242942e4089cSAndi Kleen return err; 2430c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 2431c2febafcSKirill A. Shutemov return 0; 2432c2febafcSKirill A. Shutemov } 2433c2febafcSKirill A. Shutemov 243474ffa5a3SChristoph Hellwig /* 243574ffa5a3SChristoph Hellwig * Variant of remap_pfn_range that does not call track_pfn_remap. The caller 243674ffa5a3SChristoph Hellwig * must have pre-validated the caching bits of the pgprot_t. 2437bfa5bf6dSRolf Eike Beer */ 243874ffa5a3SChristoph Hellwig int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr, 24391da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 24401da177e4SLinus Torvalds { 24411da177e4SLinus Torvalds pgd_t *pgd; 24421da177e4SLinus Torvalds unsigned long next; 24432d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 24441da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 24451da177e4SLinus Torvalds int err; 24461da177e4SLinus Torvalds 24470c4123e3SAlex Zhang if (WARN_ON_ONCE(!PAGE_ALIGNED(addr))) 24480c4123e3SAlex Zhang return -EINVAL; 24490c4123e3SAlex Zhang 24501da177e4SLinus Torvalds /* 24511da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 24521da177e4SLinus Torvalds * rest of the world about it: 24531da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 24541da177e4SLinus Torvalds * (accesses can have side effects). 24556aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 24566aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 24576aab341eSLinus Torvalds * with them. 2458314e51b9SKonstantin Khlebnikov * VM_DONTEXPAND 2459314e51b9SKonstantin Khlebnikov * Disable vma merging and expanding with mremap(). 2460314e51b9SKonstantin Khlebnikov * VM_DONTDUMP 2461314e51b9SKonstantin Khlebnikov * Omit vma from core dump, even when VM_IO turned off. 2462fb155c16SLinus Torvalds * 2463fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 2464fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 2465fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 2466b3b9c293SKonstantin Khlebnikov * See vm_normal_page() for details. 24671da177e4SLinus Torvalds */ 2468b3b9c293SKonstantin Khlebnikov if (is_cow_mapping(vma->vm_flags)) { 2469b3b9c293SKonstantin Khlebnikov if (addr != vma->vm_start || end != vma->vm_end) 2470b3b9c293SKonstantin Khlebnikov return -EINVAL; 24716aab341eSLinus Torvalds vma->vm_pgoff = pfn; 2472b3b9c293SKonstantin Khlebnikov } 2473b3b9c293SKonstantin Khlebnikov 24741c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP); 24751da177e4SLinus Torvalds 24761da177e4SLinus Torvalds BUG_ON(addr >= end); 24771da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 24781da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 24791da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 24801da177e4SLinus Torvalds do { 24811da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 2482c2febafcSKirill A. Shutemov err = remap_p4d_range(mm, pgd, addr, next, 24831da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 24841da177e4SLinus Torvalds if (err) 248574ffa5a3SChristoph Hellwig return err; 24861da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 24872ab64037Svenkatesh.pallipadi@intel.com 248874ffa5a3SChristoph Hellwig return 0; 248974ffa5a3SChristoph Hellwig } 24902ab64037Svenkatesh.pallipadi@intel.com 249174ffa5a3SChristoph Hellwig /** 249274ffa5a3SChristoph Hellwig * remap_pfn_range - remap kernel memory to userspace 249374ffa5a3SChristoph Hellwig * @vma: user vma to map to 249474ffa5a3SChristoph Hellwig * @addr: target page aligned user address to start at 249574ffa5a3SChristoph Hellwig * @pfn: page frame number of kernel physical memory address 249674ffa5a3SChristoph Hellwig * @size: size of mapping area 249774ffa5a3SChristoph Hellwig * @prot: page protection flags for this mapping 249874ffa5a3SChristoph Hellwig * 249974ffa5a3SChristoph Hellwig * Note: this is only safe if the mm semaphore is held when called. 250074ffa5a3SChristoph Hellwig * 250174ffa5a3SChristoph Hellwig * Return: %0 on success, negative error code otherwise. 250274ffa5a3SChristoph Hellwig */ 250374ffa5a3SChristoph Hellwig int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 250474ffa5a3SChristoph Hellwig unsigned long pfn, unsigned long size, pgprot_t prot) 250574ffa5a3SChristoph Hellwig { 250674ffa5a3SChristoph Hellwig int err; 250774ffa5a3SChristoph Hellwig 250874ffa5a3SChristoph Hellwig err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); 250974ffa5a3SChristoph Hellwig if (err) 251074ffa5a3SChristoph Hellwig return -EINVAL; 251174ffa5a3SChristoph Hellwig 251274ffa5a3SChristoph Hellwig err = remap_pfn_range_notrack(vma, addr, pfn, size, prot); 251374ffa5a3SChristoph Hellwig if (err) 251468f48381SSuren Baghdasaryan untrack_pfn(vma, pfn, PAGE_ALIGN(size), true); 25151da177e4SLinus Torvalds return err; 25161da177e4SLinus Torvalds } 25171da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 25181da177e4SLinus Torvalds 2519b4cbb197SLinus Torvalds /** 2520b4cbb197SLinus Torvalds * vm_iomap_memory - remap memory to userspace 2521b4cbb197SLinus Torvalds * @vma: user vma to map to 2522abd69b9eSWang Wenhu * @start: start of the physical memory to be mapped 2523b4cbb197SLinus Torvalds * @len: size of area 2524b4cbb197SLinus Torvalds * 2525b4cbb197SLinus Torvalds * This is a simplified io_remap_pfn_range() for common driver use. The 2526b4cbb197SLinus Torvalds * driver just needs to give us the physical memory range to be mapped, 2527b4cbb197SLinus Torvalds * we'll figure out the rest from the vma information. 2528b4cbb197SLinus Torvalds * 2529b4cbb197SLinus Torvalds * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get 2530b4cbb197SLinus Torvalds * whatever write-combining details or similar. 2531a862f68aSMike Rapoport * 2532a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 2533b4cbb197SLinus Torvalds */ 2534b4cbb197SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 2535b4cbb197SLinus Torvalds { 2536b4cbb197SLinus Torvalds unsigned long vm_len, pfn, pages; 2537b4cbb197SLinus Torvalds 2538b4cbb197SLinus Torvalds /* Check that the physical memory area passed in looks valid */ 2539b4cbb197SLinus Torvalds if (start + len < start) 2540b4cbb197SLinus Torvalds return -EINVAL; 2541b4cbb197SLinus Torvalds /* 2542b4cbb197SLinus Torvalds * You *really* shouldn't map things that aren't page-aligned, 2543b4cbb197SLinus Torvalds * but we've historically allowed it because IO memory might 2544b4cbb197SLinus Torvalds * just have smaller alignment. 2545b4cbb197SLinus Torvalds */ 2546b4cbb197SLinus Torvalds len += start & ~PAGE_MASK; 2547b4cbb197SLinus Torvalds pfn = start >> PAGE_SHIFT; 2548b4cbb197SLinus Torvalds pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; 2549b4cbb197SLinus Torvalds if (pfn + pages < pfn) 2550b4cbb197SLinus Torvalds return -EINVAL; 2551b4cbb197SLinus Torvalds 2552b4cbb197SLinus Torvalds /* We start the mapping 'vm_pgoff' pages into the area */ 2553b4cbb197SLinus Torvalds if (vma->vm_pgoff > pages) 2554b4cbb197SLinus Torvalds return -EINVAL; 2555b4cbb197SLinus Torvalds pfn += vma->vm_pgoff; 2556b4cbb197SLinus Torvalds pages -= vma->vm_pgoff; 2557b4cbb197SLinus Torvalds 2558b4cbb197SLinus Torvalds /* Can we fit all of the mapping? */ 2559b4cbb197SLinus Torvalds vm_len = vma->vm_end - vma->vm_start; 2560b4cbb197SLinus Torvalds if (vm_len >> PAGE_SHIFT > pages) 2561b4cbb197SLinus Torvalds return -EINVAL; 2562b4cbb197SLinus Torvalds 2563b4cbb197SLinus Torvalds /* Ok, let it rip */ 2564b4cbb197SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 2565b4cbb197SLinus Torvalds } 2566b4cbb197SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 2567b4cbb197SLinus Torvalds 2568aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 2569aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2570e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2571e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2572aee16b3cSJeremy Fitzhardinge { 25738abb50c7SMiaohe Lin pte_t *pte, *mapped_pte; 2574be1db475SDaniel Axtens int err = 0; 25753f649ab7SKees Cook spinlock_t *ptl; 2576aee16b3cSJeremy Fitzhardinge 2577be1db475SDaniel Axtens if (create) { 25788abb50c7SMiaohe Lin mapped_pte = pte = (mm == &init_mm) ? 2579e80d3909SJoerg Roedel pte_alloc_kernel_track(pmd, addr, mask) : 2580aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 2581aee16b3cSJeremy Fitzhardinge if (!pte) 2582aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2583be1db475SDaniel Axtens } else { 25848abb50c7SMiaohe Lin mapped_pte = pte = (mm == &init_mm) ? 2585be1db475SDaniel Axtens pte_offset_kernel(pmd, addr) : 2586be1db475SDaniel Axtens pte_offset_map_lock(mm, pmd, addr, &ptl); 25873db82b93SHugh Dickins if (!pte) 25883db82b93SHugh Dickins return -EINVAL; 2589be1db475SDaniel Axtens } 2590aee16b3cSJeremy Fitzhardinge 259138e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 259238e0edb1SJeremy Fitzhardinge 2593eeb4a05fSChristoph Hellwig if (fn) { 2594aee16b3cSJeremy Fitzhardinge do { 2595c33c7948SRyan Roberts if (create || !pte_none(ptep_get(pte))) { 25968b1e0f81SAnshuman Khandual err = fn(pte++, addr, data); 2597aee16b3cSJeremy Fitzhardinge if (err) 2598aee16b3cSJeremy Fitzhardinge break; 2599be1db475SDaniel Axtens } 2600c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 2601eeb4a05fSChristoph Hellwig } 2602e80d3909SJoerg Roedel *mask |= PGTBL_PTE_MODIFIED; 2603aee16b3cSJeremy Fitzhardinge 260438e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 260538e0edb1SJeremy Fitzhardinge 2606aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 26078abb50c7SMiaohe Lin pte_unmap_unlock(mapped_pte, ptl); 2608aee16b3cSJeremy Fitzhardinge return err; 2609aee16b3cSJeremy Fitzhardinge } 2610aee16b3cSJeremy Fitzhardinge 2611aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 2612aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2613e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2614e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2615aee16b3cSJeremy Fitzhardinge { 2616aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 2617aee16b3cSJeremy Fitzhardinge unsigned long next; 2618be1db475SDaniel Axtens int err = 0; 2619aee16b3cSJeremy Fitzhardinge 2620ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 2621ceb86879SAndi Kleen 2622be1db475SDaniel Axtens if (create) { 2623e80d3909SJoerg Roedel pmd = pmd_alloc_track(mm, pud, addr, mask); 2624aee16b3cSJeremy Fitzhardinge if (!pmd) 2625aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2626be1db475SDaniel Axtens } else { 2627be1db475SDaniel Axtens pmd = pmd_offset(pud, addr); 2628be1db475SDaniel Axtens } 2629aee16b3cSJeremy Fitzhardinge do { 2630aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 26310c95cba4SNicholas Piggin if (pmd_none(*pmd) && !create) 26320c95cba4SNicholas Piggin continue; 26330c95cba4SNicholas Piggin if (WARN_ON_ONCE(pmd_leaf(*pmd))) 26340c95cba4SNicholas Piggin return -EINVAL; 26350c95cba4SNicholas Piggin if (!pmd_none(*pmd) && WARN_ON_ONCE(pmd_bad(*pmd))) { 26360c95cba4SNicholas Piggin if (!create) 26370c95cba4SNicholas Piggin continue; 26380c95cba4SNicholas Piggin pmd_clear_bad(pmd); 26390c95cba4SNicholas Piggin } 26400c95cba4SNicholas Piggin err = apply_to_pte_range(mm, pmd, addr, next, 26410c95cba4SNicholas Piggin fn, data, create, mask); 2642aee16b3cSJeremy Fitzhardinge if (err) 2643aee16b3cSJeremy Fitzhardinge break; 2644aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 26450c95cba4SNicholas Piggin 2646aee16b3cSJeremy Fitzhardinge return err; 2647aee16b3cSJeremy Fitzhardinge } 2648aee16b3cSJeremy Fitzhardinge 2649c2febafcSKirill A. Shutemov static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, 2650aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2651e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2652e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2653aee16b3cSJeremy Fitzhardinge { 2654aee16b3cSJeremy Fitzhardinge pud_t *pud; 2655aee16b3cSJeremy Fitzhardinge unsigned long next; 2656be1db475SDaniel Axtens int err = 0; 2657aee16b3cSJeremy Fitzhardinge 2658be1db475SDaniel Axtens if (create) { 2659e80d3909SJoerg Roedel pud = pud_alloc_track(mm, p4d, addr, mask); 2660aee16b3cSJeremy Fitzhardinge if (!pud) 2661aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2662be1db475SDaniel Axtens } else { 2663be1db475SDaniel Axtens pud = pud_offset(p4d, addr); 2664be1db475SDaniel Axtens } 2665aee16b3cSJeremy Fitzhardinge do { 2666aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 26670c95cba4SNicholas Piggin if (pud_none(*pud) && !create) 26680c95cba4SNicholas Piggin continue; 26690c95cba4SNicholas Piggin if (WARN_ON_ONCE(pud_leaf(*pud))) 26700c95cba4SNicholas Piggin return -EINVAL; 26710c95cba4SNicholas Piggin if (!pud_none(*pud) && WARN_ON_ONCE(pud_bad(*pud))) { 26720c95cba4SNicholas Piggin if (!create) 26730c95cba4SNicholas Piggin continue; 26740c95cba4SNicholas Piggin pud_clear_bad(pud); 26750c95cba4SNicholas Piggin } 26760c95cba4SNicholas Piggin err = apply_to_pmd_range(mm, pud, addr, next, 26770c95cba4SNicholas Piggin fn, data, create, mask); 2678aee16b3cSJeremy Fitzhardinge if (err) 2679aee16b3cSJeremy Fitzhardinge break; 2680aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 26810c95cba4SNicholas Piggin 2682aee16b3cSJeremy Fitzhardinge return err; 2683aee16b3cSJeremy Fitzhardinge } 2684aee16b3cSJeremy Fitzhardinge 2685c2febafcSKirill A. Shutemov static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, 2686c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 2687e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2688e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2689c2febafcSKirill A. Shutemov { 2690c2febafcSKirill A. Shutemov p4d_t *p4d; 2691c2febafcSKirill A. Shutemov unsigned long next; 2692be1db475SDaniel Axtens int err = 0; 2693c2febafcSKirill A. Shutemov 2694be1db475SDaniel Axtens if (create) { 2695e80d3909SJoerg Roedel p4d = p4d_alloc_track(mm, pgd, addr, mask); 2696c2febafcSKirill A. Shutemov if (!p4d) 2697c2febafcSKirill A. Shutemov return -ENOMEM; 2698be1db475SDaniel Axtens } else { 2699be1db475SDaniel Axtens p4d = p4d_offset(pgd, addr); 2700be1db475SDaniel Axtens } 2701c2febafcSKirill A. Shutemov do { 2702c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 27030c95cba4SNicholas Piggin if (p4d_none(*p4d) && !create) 27040c95cba4SNicholas Piggin continue; 27050c95cba4SNicholas Piggin if (WARN_ON_ONCE(p4d_leaf(*p4d))) 27060c95cba4SNicholas Piggin return -EINVAL; 27070c95cba4SNicholas Piggin if (!p4d_none(*p4d) && WARN_ON_ONCE(p4d_bad(*p4d))) { 27080c95cba4SNicholas Piggin if (!create) 27090c95cba4SNicholas Piggin continue; 27100c95cba4SNicholas Piggin p4d_clear_bad(p4d); 27110c95cba4SNicholas Piggin } 27120c95cba4SNicholas Piggin err = apply_to_pud_range(mm, p4d, addr, next, 27130c95cba4SNicholas Piggin fn, data, create, mask); 2714c2febafcSKirill A. Shutemov if (err) 2715c2febafcSKirill A. Shutemov break; 2716c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 27170c95cba4SNicholas Piggin 2718c2febafcSKirill A. Shutemov return err; 2719c2febafcSKirill A. Shutemov } 2720c2febafcSKirill A. Shutemov 2721be1db475SDaniel Axtens static int __apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2722be1db475SDaniel Axtens unsigned long size, pte_fn_t fn, 2723be1db475SDaniel Axtens void *data, bool create) 2724be1db475SDaniel Axtens { 2725be1db475SDaniel Axtens pgd_t *pgd; 2726e80d3909SJoerg Roedel unsigned long start = addr, next; 2727be1db475SDaniel Axtens unsigned long end = addr + size; 2728e80d3909SJoerg Roedel pgtbl_mod_mask mask = 0; 2729be1db475SDaniel Axtens int err = 0; 2730be1db475SDaniel Axtens 2731be1db475SDaniel Axtens if (WARN_ON(addr >= end)) 2732be1db475SDaniel Axtens return -EINVAL; 2733be1db475SDaniel Axtens 2734be1db475SDaniel Axtens pgd = pgd_offset(mm, addr); 2735be1db475SDaniel Axtens do { 2736be1db475SDaniel Axtens next = pgd_addr_end(addr, end); 27370c95cba4SNicholas Piggin if (pgd_none(*pgd) && !create) 2738be1db475SDaniel Axtens continue; 27390c95cba4SNicholas Piggin if (WARN_ON_ONCE(pgd_leaf(*pgd))) 27400c95cba4SNicholas Piggin return -EINVAL; 27410c95cba4SNicholas Piggin if (!pgd_none(*pgd) && WARN_ON_ONCE(pgd_bad(*pgd))) { 27420c95cba4SNicholas Piggin if (!create) 27430c95cba4SNicholas Piggin continue; 27440c95cba4SNicholas Piggin pgd_clear_bad(pgd); 27450c95cba4SNicholas Piggin } 27460c95cba4SNicholas Piggin err = apply_to_p4d_range(mm, pgd, addr, next, 27470c95cba4SNicholas Piggin fn, data, create, &mask); 2748be1db475SDaniel Axtens if (err) 2749be1db475SDaniel Axtens break; 2750be1db475SDaniel Axtens } while (pgd++, addr = next, addr != end); 2751be1db475SDaniel Axtens 2752e80d3909SJoerg Roedel if (mask & ARCH_PAGE_TABLE_SYNC_MASK) 2753e80d3909SJoerg Roedel arch_sync_kernel_mappings(start, start + size); 2754e80d3909SJoerg Roedel 2755be1db475SDaniel Axtens return err; 2756be1db475SDaniel Axtens } 2757be1db475SDaniel Axtens 2758aee16b3cSJeremy Fitzhardinge /* 2759aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 2760aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 2761aee16b3cSJeremy Fitzhardinge */ 2762aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2763aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 2764aee16b3cSJeremy Fitzhardinge { 2765be1db475SDaniel Axtens return __apply_to_page_range(mm, addr, size, fn, data, true); 2766aee16b3cSJeremy Fitzhardinge } 2767aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 2768aee16b3cSJeremy Fitzhardinge 27691da177e4SLinus Torvalds /* 2770be1db475SDaniel Axtens * Scan a region of virtual memory, calling a provided function on 2771be1db475SDaniel Axtens * each leaf page table where it exists. 2772be1db475SDaniel Axtens * 2773be1db475SDaniel Axtens * Unlike apply_to_page_range, this does _not_ fill in page tables 2774be1db475SDaniel Axtens * where they are absent. 2775be1db475SDaniel Axtens */ 2776be1db475SDaniel Axtens int apply_to_existing_page_range(struct mm_struct *mm, unsigned long addr, 2777be1db475SDaniel Axtens unsigned long size, pte_fn_t fn, void *data) 2778be1db475SDaniel Axtens { 2779be1db475SDaniel Axtens return __apply_to_page_range(mm, addr, size, fn, data, false); 2780be1db475SDaniel Axtens } 2781be1db475SDaniel Axtens EXPORT_SYMBOL_GPL(apply_to_existing_page_range); 2782be1db475SDaniel Axtens 2783be1db475SDaniel Axtens /* 27849b4bdd2fSKirill A. Shutemov * handle_pte_fault chooses page fault handler according to an entry which was 27859b4bdd2fSKirill A. Shutemov * read non-atomically. Before making any commitment, on those architectures 27869b4bdd2fSKirill A. Shutemov * or configurations (e.g. i386 with PAE) which might give a mix of unmatched 27879b4bdd2fSKirill A. Shutemov * parts, do_swap_page must check under lock before unmapping the pte and 27889b4bdd2fSKirill A. Shutemov * proceeding (but do_wp_page is only called after already making such a check; 2789a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 27908f4e2101SHugh Dickins */ 27912ca99358SPeter Xu static inline int pte_unmap_same(struct vm_fault *vmf) 27928f4e2101SHugh Dickins { 27938f4e2101SHugh Dickins int same = 1; 2794923717cbSThomas Gleixner #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPTION) 27958f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 2796c7ad0880SHugh Dickins spin_lock(vmf->ptl); 2797c33c7948SRyan Roberts same = pte_same(ptep_get(vmf->pte), vmf->orig_pte); 2798c7ad0880SHugh Dickins spin_unlock(vmf->ptl); 27998f4e2101SHugh Dickins } 28008f4e2101SHugh Dickins #endif 28012ca99358SPeter Xu pte_unmap(vmf->pte); 28022ca99358SPeter Xu vmf->pte = NULL; 28038f4e2101SHugh Dickins return same; 28048f4e2101SHugh Dickins } 28058f4e2101SHugh Dickins 2806a873dfe1STony Luck /* 2807a873dfe1STony Luck * Return: 2808a873dfe1STony Luck * 0: copied succeeded 2809a873dfe1STony Luck * -EHWPOISON: copy failed due to hwpoison in source page 2810a873dfe1STony Luck * -EAGAIN: copied failed (some other reason) 2811a873dfe1STony Luck */ 2812a873dfe1STony Luck static inline int __wp_page_copy_user(struct page *dst, struct page *src, 281383d116c5SJia He struct vm_fault *vmf) 28146aab341eSLinus Torvalds { 2815a873dfe1STony Luck int ret; 281683d116c5SJia He void *kaddr; 281783d116c5SJia He void __user *uaddr; 281883d116c5SJia He struct vm_area_struct *vma = vmf->vma; 281983d116c5SJia He struct mm_struct *mm = vma->vm_mm; 282083d116c5SJia He unsigned long addr = vmf->address; 282183d116c5SJia He 282283d116c5SJia He if (likely(src)) { 2823d302c239STony Luck if (copy_mc_user_highpage(dst, src, addr, vma)) { 2824d302c239STony Luck memory_failure_queue(page_to_pfn(src), 0); 2825a873dfe1STony Luck return -EHWPOISON; 2826d302c239STony Luck } 2827a873dfe1STony Luck return 0; 282883d116c5SJia He } 282983d116c5SJia He 28306aab341eSLinus Torvalds /* 28316aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 28326aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 28336aab341eSLinus Torvalds * just copying from the original user address. If that 28346aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 28356aab341eSLinus Torvalds */ 283683d116c5SJia He kaddr = kmap_atomic(dst); 283783d116c5SJia He uaddr = (void __user *)(addr & PAGE_MASK); 283883d116c5SJia He 283983d116c5SJia He /* 284083d116c5SJia He * On architectures with software "accessed" bits, we would 284183d116c5SJia He * take a double page fault, so mark it accessed here. 284283d116c5SJia He */ 28433db82b93SHugh Dickins vmf->pte = NULL; 2844e1fd09e3SYu Zhao if (!arch_has_hw_pte_young() && !pte_young(vmf->orig_pte)) { 284583d116c5SJia He pte_t entry; 284683d116c5SJia He 284783d116c5SJia He vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); 2848c33c7948SRyan Roberts if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 284983d116c5SJia He /* 285083d116c5SJia He * Other thread has already handled the fault 28517df67697SBibo Mao * and update local tlb only 285283d116c5SJia He */ 2853a92cbb82SHugh Dickins if (vmf->pte) 28547df67697SBibo Mao update_mmu_tlb(vma, addr, vmf->pte); 2855a873dfe1STony Luck ret = -EAGAIN; 285683d116c5SJia He goto pte_unlock; 285783d116c5SJia He } 285883d116c5SJia He 285983d116c5SJia He entry = pte_mkyoung(vmf->orig_pte); 286083d116c5SJia He if (ptep_set_access_flags(vma, addr, vmf->pte, entry, 0)) 286183d116c5SJia He update_mmu_cache(vma, addr, vmf->pte); 286283d116c5SJia He } 28635d2a2dbbSLinus Torvalds 28645d2a2dbbSLinus Torvalds /* 28655d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 28665d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 28675d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 28685d2a2dbbSLinus Torvalds * zeroes. 28695d2a2dbbSLinus Torvalds */ 287083d116c5SJia He if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { 28713db82b93SHugh Dickins if (vmf->pte) 2872c3e5ea6eSKirill A. Shutemov goto warn; 2873c3e5ea6eSKirill A. Shutemov 2874c3e5ea6eSKirill A. Shutemov /* Re-validate under PTL if the page is still mapped */ 2875c3e5ea6eSKirill A. Shutemov vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); 2876c33c7948SRyan Roberts if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 28777df67697SBibo Mao /* The PTE changed under us, update local tlb */ 2878a92cbb82SHugh Dickins if (vmf->pte) 28797df67697SBibo Mao update_mmu_tlb(vma, addr, vmf->pte); 2880a873dfe1STony Luck ret = -EAGAIN; 2881c3e5ea6eSKirill A. Shutemov goto pte_unlock; 2882c3e5ea6eSKirill A. Shutemov } 2883c3e5ea6eSKirill A. Shutemov 2884c3e5ea6eSKirill A. Shutemov /* 2885985ba004SEthon Paul * The same page can be mapped back since last copy attempt. 2886c3e5ea6eSKirill A. Shutemov * Try to copy again under PTL. 2887c3e5ea6eSKirill A. Shutemov */ 2888c3e5ea6eSKirill A. Shutemov if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { 288983d116c5SJia He /* 289083d116c5SJia He * Give a warn in case there can be some obscure 289183d116c5SJia He * use-case 289283d116c5SJia He */ 2893c3e5ea6eSKirill A. Shutemov warn: 289483d116c5SJia He WARN_ON_ONCE(1); 28953ecb01dfSJan Beulich clear_page(kaddr); 289683d116c5SJia He } 2897c3e5ea6eSKirill A. Shutemov } 289883d116c5SJia He 2899a873dfe1STony Luck ret = 0; 290083d116c5SJia He 290183d116c5SJia He pte_unlock: 29023db82b93SHugh Dickins if (vmf->pte) 290383d116c5SJia He pte_unmap_unlock(vmf->pte, vmf->ptl); 29049b04c5feSCong Wang kunmap_atomic(kaddr); 2905c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 290683d116c5SJia He 290783d116c5SJia He return ret; 29086aab341eSLinus Torvalds } 29096aab341eSLinus Torvalds 2910c20cd45eSMichal Hocko static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) 2911c20cd45eSMichal Hocko { 2912c20cd45eSMichal Hocko struct file *vm_file = vma->vm_file; 2913c20cd45eSMichal Hocko 2914c20cd45eSMichal Hocko if (vm_file) 2915c20cd45eSMichal Hocko return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; 2916c20cd45eSMichal Hocko 2917c20cd45eSMichal Hocko /* 2918c20cd45eSMichal Hocko * Special mappings (e.g. VDSO) do not have any file so fake 2919c20cd45eSMichal Hocko * a default GFP_KERNEL for them. 2920c20cd45eSMichal Hocko */ 2921c20cd45eSMichal Hocko return GFP_KERNEL; 2922c20cd45eSMichal Hocko } 2923c20cd45eSMichal Hocko 29241da177e4SLinus Torvalds /* 2925fb09a464SKirill A. Shutemov * Notify the address space that the page is about to become writable so that 2926fb09a464SKirill A. Shutemov * it can prohibit this or wait for the page to get into an appropriate state. 2927fb09a464SKirill A. Shutemov * 2928fb09a464SKirill A. Shutemov * We do this without the lock held, so that it can sleep if it needs to. 2929fb09a464SKirill A. Shutemov */ 29302b740303SSouptick Joarder static vm_fault_t do_page_mkwrite(struct vm_fault *vmf) 2931fb09a464SKirill A. Shutemov { 29322b740303SSouptick Joarder vm_fault_t ret; 293338b8cb7fSJan Kara struct page *page = vmf->page; 293438b8cb7fSJan Kara unsigned int old_flags = vmf->flags; 2935fb09a464SKirill A. Shutemov 293638b8cb7fSJan Kara vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 2937fb09a464SKirill A. Shutemov 2938dc617f29SDarrick J. Wong if (vmf->vma->vm_file && 2939dc617f29SDarrick J. Wong IS_SWAPFILE(vmf->vma->vm_file->f_mapping->host)) 2940dc617f29SDarrick J. Wong return VM_FAULT_SIGBUS; 2941dc617f29SDarrick J. Wong 294211bac800SDave Jiang ret = vmf->vma->vm_ops->page_mkwrite(vmf); 294338b8cb7fSJan Kara /* Restore original flags so that caller is not surprised */ 294438b8cb7fSJan Kara vmf->flags = old_flags; 2945fb09a464SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 2946fb09a464SKirill A. Shutemov return ret; 2947fb09a464SKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) { 2948fb09a464SKirill A. Shutemov lock_page(page); 2949fb09a464SKirill A. Shutemov if (!page->mapping) { 2950fb09a464SKirill A. Shutemov unlock_page(page); 2951fb09a464SKirill A. Shutemov return 0; /* retry */ 2952fb09a464SKirill A. Shutemov } 2953fb09a464SKirill A. Shutemov ret |= VM_FAULT_LOCKED; 2954fb09a464SKirill A. Shutemov } else 2955fb09a464SKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(page), page); 2956fb09a464SKirill A. Shutemov return ret; 2957fb09a464SKirill A. Shutemov } 2958fb09a464SKirill A. Shutemov 2959fb09a464SKirill A. Shutemov /* 296097ba0c2bSJan Kara * Handle dirtying of a page in shared file mapping on a write fault. 29614e047f89SShachar Raindel * 296297ba0c2bSJan Kara * The function expects the page to be locked and unlocks it. 29634e047f89SShachar Raindel */ 296489b15332SJohannes Weiner static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf) 29654e047f89SShachar Raindel { 296689b15332SJohannes Weiner struct vm_area_struct *vma = vmf->vma; 29674e047f89SShachar Raindel struct address_space *mapping; 296889b15332SJohannes Weiner struct page *page = vmf->page; 296997ba0c2bSJan Kara bool dirtied; 297097ba0c2bSJan Kara bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; 29714e047f89SShachar Raindel 29724e047f89SShachar Raindel dirtied = set_page_dirty(page); 29734e047f89SShachar Raindel VM_BUG_ON_PAGE(PageAnon(page), page); 297497ba0c2bSJan Kara /* 297597ba0c2bSJan Kara * Take a local copy of the address_space - page.mapping may be zeroed 297697ba0c2bSJan Kara * by truncate after unlock_page(). The address_space itself remains 297797ba0c2bSJan Kara * pinned by vma->vm_file's reference. We rely on unlock_page()'s 297897ba0c2bSJan Kara * release semantics to prevent the compiler from undoing this copying. 297997ba0c2bSJan Kara */ 298097ba0c2bSJan Kara mapping = page_rmapping(page); 29814e047f89SShachar Raindel unlock_page(page); 29824e047f89SShachar Raindel 29834e047f89SShachar Raindel if (!page_mkwrite) 29844e047f89SShachar Raindel file_update_time(vma->vm_file); 298589b15332SJohannes Weiner 298689b15332SJohannes Weiner /* 298789b15332SJohannes Weiner * Throttle page dirtying rate down to writeback speed. 298889b15332SJohannes Weiner * 298989b15332SJohannes Weiner * mapping may be NULL here because some device drivers do not 299089b15332SJohannes Weiner * set page.mapping but still dirty their pages 299189b15332SJohannes Weiner * 2992c1e8d7c6SMichel Lespinasse * Drop the mmap_lock before waiting on IO, if we can. The file 299389b15332SJohannes Weiner * is pinning the mapping, as per above. 299489b15332SJohannes Weiner */ 299589b15332SJohannes Weiner if ((dirtied || page_mkwrite) && mapping) { 299689b15332SJohannes Weiner struct file *fpin; 299789b15332SJohannes Weiner 299889b15332SJohannes Weiner fpin = maybe_unlock_mmap_for_io(vmf, NULL); 299989b15332SJohannes Weiner balance_dirty_pages_ratelimited(mapping); 300089b15332SJohannes Weiner if (fpin) { 300189b15332SJohannes Weiner fput(fpin); 3002d9272525SPeter Xu return VM_FAULT_COMPLETED; 300389b15332SJohannes Weiner } 300489b15332SJohannes Weiner } 300589b15332SJohannes Weiner 300689b15332SJohannes Weiner return 0; 30074e047f89SShachar Raindel } 30084e047f89SShachar Raindel 300997ba0c2bSJan Kara /* 30104e047f89SShachar Raindel * Handle write page faults for pages that can be reused in the current vma 30114e047f89SShachar Raindel * 30124e047f89SShachar Raindel * This can happen either due to the mapping being with the VM_SHARED flag, 30134e047f89SShachar Raindel * or due to us being the last reference standing to the page. In either 30144e047f89SShachar Raindel * case, all we need to do here is to mark the page as writable and update 30154e047f89SShachar Raindel * any related book-keeping. 30164e047f89SShachar Raindel */ 3017997dd98dSJan Kara static inline void wp_page_reuse(struct vm_fault *vmf) 301882b0f8c3SJan Kara __releases(vmf->ptl) 30194e047f89SShachar Raindel { 302082b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3021a41b70d6SJan Kara struct page *page = vmf->page; 30224e047f89SShachar Raindel pte_t entry; 30236c287605SDavid Hildenbrand 3024c89357e2SDavid Hildenbrand VM_BUG_ON(!(vmf->flags & FAULT_FLAG_WRITE)); 3025cdb281e6SQi Zheng VM_BUG_ON(page && PageAnon(page) && !PageAnonExclusive(page)); 30266c287605SDavid Hildenbrand 30274e047f89SShachar Raindel /* 30284e047f89SShachar Raindel * Clear the pages cpupid information as the existing 30294e047f89SShachar Raindel * information potentially belongs to a now completely 30304e047f89SShachar Raindel * unrelated process. 30314e047f89SShachar Raindel */ 30324e047f89SShachar Raindel if (page) 30334e047f89SShachar Raindel page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); 30344e047f89SShachar Raindel 30352994302bSJan Kara flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); 30362994302bSJan Kara entry = pte_mkyoung(vmf->orig_pte); 30374e047f89SShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 303882b0f8c3SJan Kara if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) 303982b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 304082b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3041798a6b87SPeter Xu count_vm_event(PGREUSE); 30424e047f89SShachar Raindel } 30434e047f89SShachar Raindel 30444e047f89SShachar Raindel /* 3045c89357e2SDavid Hildenbrand * Handle the case of a page which we actually need to copy to a new page, 3046c89357e2SDavid Hildenbrand * either due to COW or unsharing. 30472f38ab2cSShachar Raindel * 3048c1e8d7c6SMichel Lespinasse * Called with mmap_lock locked and the old page referenced, but 30492f38ab2cSShachar Raindel * without the ptl held. 30502f38ab2cSShachar Raindel * 30512f38ab2cSShachar Raindel * High level logic flow: 30522f38ab2cSShachar Raindel * 30532f38ab2cSShachar Raindel * - Allocate a page, copy the content of the old page to the new one. 30542f38ab2cSShachar Raindel * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. 30552f38ab2cSShachar Raindel * - Take the PTL. If the pte changed, bail out and release the allocated page 30562f38ab2cSShachar Raindel * - If the pte is still the way we remember it, update the page table and all 30572f38ab2cSShachar Raindel * relevant references. This includes dropping the reference the page-table 30582f38ab2cSShachar Raindel * held to the old page, as well as updating the rmap. 30592f38ab2cSShachar Raindel * - In any case, unlock the PTL and drop the reference we took to the old page. 30602f38ab2cSShachar Raindel */ 30612b740303SSouptick Joarder static vm_fault_t wp_page_copy(struct vm_fault *vmf) 30622f38ab2cSShachar Raindel { 3063c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 306482b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3065bae473a4SKirill A. Shutemov struct mm_struct *mm = vma->vm_mm; 306628d41a48SMatthew Wilcox (Oracle) struct folio *old_folio = NULL; 306728d41a48SMatthew Wilcox (Oracle) struct folio *new_folio = NULL; 30682f38ab2cSShachar Raindel pte_t entry; 30692f38ab2cSShachar Raindel int page_copied = 0; 3070ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 3071a873dfe1STony Luck int ret; 30722f38ab2cSShachar Raindel 3073662ce1dcSYang Yang delayacct_wpcopy_start(); 3074662ce1dcSYang Yang 307528d41a48SMatthew Wilcox (Oracle) if (vmf->page) 307628d41a48SMatthew Wilcox (Oracle) old_folio = page_folio(vmf->page); 30772f38ab2cSShachar Raindel if (unlikely(anon_vma_prepare(vma))) 30782f38ab2cSShachar Raindel goto oom; 30792f38ab2cSShachar Raindel 30802994302bSJan Kara if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { 30816bc56a4dSMatthew Wilcox (Oracle) new_folio = vma_alloc_zeroed_movable_folio(vma, vmf->address); 30826bc56a4dSMatthew Wilcox (Oracle) if (!new_folio) 30832f38ab2cSShachar Raindel goto oom; 30842f38ab2cSShachar Raindel } else { 308528d41a48SMatthew Wilcox (Oracle) new_folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma, 308628d41a48SMatthew Wilcox (Oracle) vmf->address, false); 308728d41a48SMatthew Wilcox (Oracle) if (!new_folio) 30882f38ab2cSShachar Raindel goto oom; 308983d116c5SJia He 309028d41a48SMatthew Wilcox (Oracle) ret = __wp_page_copy_user(&new_folio->page, vmf->page, vmf); 3091a873dfe1STony Luck if (ret) { 309283d116c5SJia He /* 309383d116c5SJia He * COW failed, if the fault was solved by other, 309483d116c5SJia He * it's fine. If not, userspace would re-fault on 309583d116c5SJia He * the same address and we will handle the fault 309683d116c5SJia He * from the second attempt. 3097a873dfe1STony Luck * The -EHWPOISON case will not be retried. 309883d116c5SJia He */ 309928d41a48SMatthew Wilcox (Oracle) folio_put(new_folio); 310028d41a48SMatthew Wilcox (Oracle) if (old_folio) 310128d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 3102662ce1dcSYang Yang 3103662ce1dcSYang Yang delayacct_wpcopy_end(); 3104a873dfe1STony Luck return ret == -EHWPOISON ? VM_FAULT_HWPOISON : 0; 310583d116c5SJia He } 310628d41a48SMatthew Wilcox (Oracle) kmsan_copy_page_meta(&new_folio->page, vmf->page); 31072f38ab2cSShachar Raindel } 31082f38ab2cSShachar Raindel 310928d41a48SMatthew Wilcox (Oracle) if (mem_cgroup_charge(new_folio, mm, GFP_KERNEL)) 31102f38ab2cSShachar Raindel goto oom_free_new; 31114d4f75bfSKefeng Wang folio_throttle_swaprate(new_folio, GFP_KERNEL); 31122f38ab2cSShachar Raindel 311328d41a48SMatthew Wilcox (Oracle) __folio_mark_uptodate(new_folio); 3114eb3c24f3SMel Gorman 31157d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, mm, 31166f4f13e8SJérôme Glisse vmf->address & PAGE_MASK, 3117ac46d4f3SJérôme Glisse (vmf->address & PAGE_MASK) + PAGE_SIZE); 3118ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 31192f38ab2cSShachar Raindel 31202f38ab2cSShachar Raindel /* 31212f38ab2cSShachar Raindel * Re-check the pte - we dropped the lock 31222f38ab2cSShachar Raindel */ 312382b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); 3124c33c7948SRyan Roberts if (likely(vmf->pte && pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 312528d41a48SMatthew Wilcox (Oracle) if (old_folio) { 312628d41a48SMatthew Wilcox (Oracle) if (!folio_test_anon(old_folio)) { 312728d41a48SMatthew Wilcox (Oracle) dec_mm_counter(mm, mm_counter_file(&old_folio->page)); 3128f1a79412SShakeel Butt inc_mm_counter(mm, MM_ANONPAGES); 31292f38ab2cSShachar Raindel } 31302f38ab2cSShachar Raindel } else { 3131f1a79412SShakeel Butt inc_mm_counter(mm, MM_ANONPAGES); 31322f38ab2cSShachar Raindel } 31332994302bSJan Kara flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); 313428d41a48SMatthew Wilcox (Oracle) entry = mk_pte(&new_folio->page, vma->vm_page_prot); 313550c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 3136c89357e2SDavid Hildenbrand if (unlikely(unshare)) { 3137c89357e2SDavid Hildenbrand if (pte_soft_dirty(vmf->orig_pte)) 3138c89357e2SDavid Hildenbrand entry = pte_mksoft_dirty(entry); 3139c89357e2SDavid Hildenbrand if (pte_uffd_wp(vmf->orig_pte)) 3140c89357e2SDavid Hildenbrand entry = pte_mkuffd_wp(entry); 3141c89357e2SDavid Hildenbrand } else { 31422f38ab2cSShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 3143c89357e2SDavid Hildenbrand } 3144111fe718SNicholas Piggin 31452f38ab2cSShachar Raindel /* 31462f38ab2cSShachar Raindel * Clear the pte entry and flush it first, before updating the 3147111fe718SNicholas Piggin * pte with the new entry, to keep TLBs on different CPUs in 3148111fe718SNicholas Piggin * sync. This code used to set the new PTE then flush TLBs, but 3149111fe718SNicholas Piggin * that left a window where the new PTE could be loaded into 3150111fe718SNicholas Piggin * some TLBs while the old PTE remains in others. 31512f38ab2cSShachar Raindel */ 315282b0f8c3SJan Kara ptep_clear_flush_notify(vma, vmf->address, vmf->pte); 315328d41a48SMatthew Wilcox (Oracle) folio_add_new_anon_rmap(new_folio, vma, vmf->address); 315428d41a48SMatthew Wilcox (Oracle) folio_add_lru_vma(new_folio, vma); 31552f38ab2cSShachar Raindel /* 31562f38ab2cSShachar Raindel * We call the notify macro here because, when using secondary 31572f38ab2cSShachar Raindel * mmu page tables (such as kvm shadow page tables), we want the 31582f38ab2cSShachar Raindel * new page to be mapped directly into the secondary page table. 31592f38ab2cSShachar Raindel */ 3160c89357e2SDavid Hildenbrand BUG_ON(unshare && pte_write(entry)); 316182b0f8c3SJan Kara set_pte_at_notify(mm, vmf->address, vmf->pte, entry); 316282b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 316328d41a48SMatthew Wilcox (Oracle) if (old_folio) { 31642f38ab2cSShachar Raindel /* 31652f38ab2cSShachar Raindel * Only after switching the pte to the new page may 31662f38ab2cSShachar Raindel * we remove the mapcount here. Otherwise another 31672f38ab2cSShachar Raindel * process may come and find the rmap count decremented 31682f38ab2cSShachar Raindel * before the pte is switched to the new page, and 31692f38ab2cSShachar Raindel * "reuse" the old page writing into it while our pte 31702f38ab2cSShachar Raindel * here still points into it and can be read by other 31712f38ab2cSShachar Raindel * threads. 31722f38ab2cSShachar Raindel * 31732f38ab2cSShachar Raindel * The critical issue is to order this 31742f38ab2cSShachar Raindel * page_remove_rmap with the ptp_clear_flush above. 31752f38ab2cSShachar Raindel * Those stores are ordered by (if nothing else,) 31762f38ab2cSShachar Raindel * the barrier present in the atomic_add_negative 31772f38ab2cSShachar Raindel * in page_remove_rmap. 31782f38ab2cSShachar Raindel * 31792f38ab2cSShachar Raindel * Then the TLB flush in ptep_clear_flush ensures that 31802f38ab2cSShachar Raindel * no process can access the old page before the 31812f38ab2cSShachar Raindel * decremented mapcount is visible. And the old page 31822f38ab2cSShachar Raindel * cannot be reused until after the decremented 31832f38ab2cSShachar Raindel * mapcount is visible. So transitively, TLBs to 31842f38ab2cSShachar Raindel * old page will be flushed before it can be reused. 31852f38ab2cSShachar Raindel */ 318628d41a48SMatthew Wilcox (Oracle) page_remove_rmap(vmf->page, vma, false); 31872f38ab2cSShachar Raindel } 31882f38ab2cSShachar Raindel 31892f38ab2cSShachar Raindel /* Free the old page.. */ 319028d41a48SMatthew Wilcox (Oracle) new_folio = old_folio; 31912f38ab2cSShachar Raindel page_copied = 1; 31923db82b93SHugh Dickins pte_unmap_unlock(vmf->pte, vmf->ptl); 31933db82b93SHugh Dickins } else if (vmf->pte) { 31947df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 31953db82b93SHugh Dickins pte_unmap_unlock(vmf->pte, vmf->ptl); 31962f38ab2cSShachar Raindel } 31972f38ab2cSShachar Raindel 31984645b9feSJérôme Glisse /* 31994645b9feSJérôme Glisse * No need to double call mmu_notifier->invalidate_range() callback as 32004645b9feSJérôme Glisse * the above ptep_clear_flush_notify() did already call it. 32014645b9feSJérôme Glisse */ 3202ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_only_end(&range); 32033db82b93SHugh Dickins 32043db82b93SHugh Dickins if (new_folio) 32053db82b93SHugh Dickins folio_put(new_folio); 320628d41a48SMatthew Wilcox (Oracle) if (old_folio) { 3207f4c4a3f4SHuang Ying if (page_copied) 320828d41a48SMatthew Wilcox (Oracle) free_swap_cache(&old_folio->page); 320928d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 32102f38ab2cSShachar Raindel } 3211662ce1dcSYang Yang 3212662ce1dcSYang Yang delayacct_wpcopy_end(); 3213cb8d8633SDavid Hildenbrand return 0; 32142f38ab2cSShachar Raindel oom_free_new: 321528d41a48SMatthew Wilcox (Oracle) folio_put(new_folio); 32162f38ab2cSShachar Raindel oom: 321728d41a48SMatthew Wilcox (Oracle) if (old_folio) 321828d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 3219662ce1dcSYang Yang 3220662ce1dcSYang Yang delayacct_wpcopy_end(); 32212f38ab2cSShachar Raindel return VM_FAULT_OOM; 32222f38ab2cSShachar Raindel } 32232f38ab2cSShachar Raindel 322466a6197cSJan Kara /** 322566a6197cSJan Kara * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE 322666a6197cSJan Kara * writeable once the page is prepared 322766a6197cSJan Kara * 322866a6197cSJan Kara * @vmf: structure describing the fault 322966a6197cSJan Kara * 323066a6197cSJan Kara * This function handles all that is needed to finish a write page fault in a 323166a6197cSJan Kara * shared mapping due to PTE being read-only once the mapped page is prepared. 3232a862f68aSMike Rapoport * It handles locking of PTE and modifying it. 323366a6197cSJan Kara * 323466a6197cSJan Kara * The function expects the page to be locked or other protection against 323566a6197cSJan Kara * concurrent faults / writeback (such as DAX radix tree locks). 3236a862f68aSMike Rapoport * 32372797e79fSLiu Xiang * Return: %0 on success, %VM_FAULT_NOPAGE when PTE got changed before 3238a862f68aSMike Rapoport * we acquired PTE lock. 323966a6197cSJan Kara */ 32402b740303SSouptick Joarder vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf) 324166a6197cSJan Kara { 324266a6197cSJan Kara WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); 324366a6197cSJan Kara vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, 324466a6197cSJan Kara &vmf->ptl); 32453db82b93SHugh Dickins if (!vmf->pte) 32463db82b93SHugh Dickins return VM_FAULT_NOPAGE; 324766a6197cSJan Kara /* 324866a6197cSJan Kara * We might have raced with another page fault while we released the 324966a6197cSJan Kara * pte_offset_map_lock. 325066a6197cSJan Kara */ 3251c33c7948SRyan Roberts if (!pte_same(ptep_get(vmf->pte), vmf->orig_pte)) { 32527df67697SBibo Mao update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 325366a6197cSJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3254a19e2553SJan Kara return VM_FAULT_NOPAGE; 325566a6197cSJan Kara } 325666a6197cSJan Kara wp_page_reuse(vmf); 3257a19e2553SJan Kara return 0; 325866a6197cSJan Kara } 325966a6197cSJan Kara 3260dd906184SBoaz Harrosh /* 3261dd906184SBoaz Harrosh * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED 3262dd906184SBoaz Harrosh * mapping 3263dd906184SBoaz Harrosh */ 32642b740303SSouptick Joarder static vm_fault_t wp_pfn_shared(struct vm_fault *vmf) 3265dd906184SBoaz Harrosh { 326682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3267bae473a4SKirill A. Shutemov 3268dd906184SBoaz Harrosh if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { 32692b740303SSouptick Joarder vm_fault_t ret; 3270dd906184SBoaz Harrosh 327182b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3272fe82221fSJan Kara vmf->flags |= FAULT_FLAG_MKWRITE; 327311bac800SDave Jiang ret = vma->vm_ops->pfn_mkwrite(vmf); 32742f89dc12SJan Kara if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) 3275dd906184SBoaz Harrosh return ret; 327666a6197cSJan Kara return finish_mkwrite_fault(vmf); 3277dd906184SBoaz Harrosh } 3278997dd98dSJan Kara wp_page_reuse(vmf); 3279cb8d8633SDavid Hildenbrand return 0; 3280dd906184SBoaz Harrosh } 3281dd906184SBoaz Harrosh 32822b740303SSouptick Joarder static vm_fault_t wp_page_shared(struct vm_fault *vmf) 328382b0f8c3SJan Kara __releases(vmf->ptl) 328493e478d4SShachar Raindel { 328582b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3286cb8d8633SDavid Hildenbrand vm_fault_t ret = 0; 328793e478d4SShachar Raindel 3288a41b70d6SJan Kara get_page(vmf->page); 328993e478d4SShachar Raindel 329093e478d4SShachar Raindel if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 32912b740303SSouptick Joarder vm_fault_t tmp; 329293e478d4SShachar Raindel 329382b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 329438b8cb7fSJan Kara tmp = do_page_mkwrite(vmf); 329593e478d4SShachar Raindel if (unlikely(!tmp || (tmp & 329693e478d4SShachar Raindel (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 3297a41b70d6SJan Kara put_page(vmf->page); 329893e478d4SShachar Raindel return tmp; 329993e478d4SShachar Raindel } 330066a6197cSJan Kara tmp = finish_mkwrite_fault(vmf); 3301a19e2553SJan Kara if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { 3302a41b70d6SJan Kara unlock_page(vmf->page); 3303a41b70d6SJan Kara put_page(vmf->page); 330466a6197cSJan Kara return tmp; 330593e478d4SShachar Raindel } 330666a6197cSJan Kara } else { 3307997dd98dSJan Kara wp_page_reuse(vmf); 3308997dd98dSJan Kara lock_page(vmf->page); 330993e478d4SShachar Raindel } 331089b15332SJohannes Weiner ret |= fault_dirty_shared_page(vmf); 3311997dd98dSJan Kara put_page(vmf->page); 331293e478d4SShachar Raindel 331389b15332SJohannes Weiner return ret; 331493e478d4SShachar Raindel } 331593e478d4SShachar Raindel 33162f38ab2cSShachar Raindel /* 3317c89357e2SDavid Hildenbrand * This routine handles present pages, when 3318c89357e2SDavid Hildenbrand * * users try to write to a shared page (FAULT_FLAG_WRITE) 3319c89357e2SDavid Hildenbrand * * GUP wants to take a R/O pin on a possibly shared anonymous page 3320c89357e2SDavid Hildenbrand * (FAULT_FLAG_UNSHARE) 3321c89357e2SDavid Hildenbrand * 3322c89357e2SDavid Hildenbrand * It is done by copying the page to a new address and decrementing the 3323c89357e2SDavid Hildenbrand * shared-page counter for the old page. 33241da177e4SLinus Torvalds * 33251da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 33261da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 3327c89357e2SDavid Hildenbrand * Thus, with FAULT_FLAG_WRITE, we can safely just mark it writable once we've 3328c89357e2SDavid Hildenbrand * done any necessary COW. 33291da177e4SLinus Torvalds * 3330c89357e2SDavid Hildenbrand * In case of FAULT_FLAG_WRITE, we also mark the page dirty at this point even 3331c89357e2SDavid Hildenbrand * though the page will change only once the write actually happens. This 3332c89357e2SDavid Hildenbrand * avoids a few races, and potentially makes it more efficient. 33331da177e4SLinus Torvalds * 3334c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 33358f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 3336c1e8d7c6SMichel Lespinasse * We return with mmap_lock still held, but pte unmapped and unlocked. 33371da177e4SLinus Torvalds */ 33382b740303SSouptick Joarder static vm_fault_t do_wp_page(struct vm_fault *vmf) 333982b0f8c3SJan Kara __releases(vmf->ptl) 33401da177e4SLinus Torvalds { 3341c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 334282b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3343b9086fdeSDavid Hildenbrand struct folio *folio = NULL; 33441da177e4SLinus Torvalds 3345c89357e2SDavid Hildenbrand if (likely(!unshare)) { 3346c33c7948SRyan Roberts if (userfaultfd_pte_wp(vma, ptep_get(vmf->pte))) { 3347529b930bSAndrea Arcangeli pte_unmap_unlock(vmf->pte, vmf->ptl); 3348529b930bSAndrea Arcangeli return handle_userfault(vmf, VM_UFFD_WP); 3349529b930bSAndrea Arcangeli } 3350529b930bSAndrea Arcangeli 33516ce64428SNadav Amit /* 33526ce64428SNadav Amit * Userfaultfd write-protect can defer flushes. Ensure the TLB 33536ce64428SNadav Amit * is flushed in this case before copying. 33546ce64428SNadav Amit */ 33556ce64428SNadav Amit if (unlikely(userfaultfd_wp(vmf->vma) && 33566ce64428SNadav Amit mm_tlb_flush_pending(vmf->vma->vm_mm))) 33576ce64428SNadav Amit flush_tlb_page(vmf->vma, vmf->address); 3358c89357e2SDavid Hildenbrand } 33596ce64428SNadav Amit 3360a41b70d6SJan Kara vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); 3361c89357e2SDavid Hildenbrand 3362251b97f5SPeter Zijlstra /* 3363b9086fdeSDavid Hildenbrand * Shared mapping: we are guaranteed to have VM_WRITE and 3364b9086fdeSDavid Hildenbrand * FAULT_FLAG_WRITE set at this point. 3365b9086fdeSDavid Hildenbrand */ 3366b9086fdeSDavid Hildenbrand if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 3367b9086fdeSDavid Hildenbrand /* 336864e45507SPeter Feiner * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a 336964e45507SPeter Feiner * VM_PFNMAP VMA. 3370251b97f5SPeter Zijlstra * 3371251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 3372dd906184SBoaz Harrosh * Just mark the pages writable and/or call ops->pfn_mkwrite. 3373251b97f5SPeter Zijlstra */ 3374b9086fdeSDavid Hildenbrand if (!vmf->page) 33752994302bSJan Kara return wp_pfn_shared(vmf); 3376b9086fdeSDavid Hildenbrand return wp_page_shared(vmf); 3377251b97f5SPeter Zijlstra } 33781da177e4SLinus Torvalds 3379b9086fdeSDavid Hildenbrand if (vmf->page) 3380e4a2ed94SMatthew Wilcox (Oracle) folio = page_folio(vmf->page); 3381b9086fdeSDavid Hildenbrand 3382b9086fdeSDavid Hildenbrand /* 3383b9086fdeSDavid Hildenbrand * Private mapping: create an exclusive anonymous page copy if reuse 3384b9086fdeSDavid Hildenbrand * is impossible. We might miss VM_WRITE for FOLL_FORCE handling. 3385b9086fdeSDavid Hildenbrand */ 3386b9086fdeSDavid Hildenbrand if (folio && folio_test_anon(folio)) { 338753a05ad9SDavid Hildenbrand /* 33886c287605SDavid Hildenbrand * If the page is exclusive to this process we must reuse the 33896c287605SDavid Hildenbrand * page without further checks. 33906c287605SDavid Hildenbrand */ 3391e4a2ed94SMatthew Wilcox (Oracle) if (PageAnonExclusive(vmf->page)) 33926c287605SDavid Hildenbrand goto reuse; 33936c287605SDavid Hildenbrand 33946c287605SDavid Hildenbrand /* 3395e4a2ed94SMatthew Wilcox (Oracle) * We have to verify under folio lock: these early checks are 3396e4a2ed94SMatthew Wilcox (Oracle) * just an optimization to avoid locking the folio and freeing 339753a05ad9SDavid Hildenbrand * the swapcache if there is little hope that we can reuse. 339853a05ad9SDavid Hildenbrand * 3399e4a2ed94SMatthew Wilcox (Oracle) * KSM doesn't necessarily raise the folio refcount. 340053a05ad9SDavid Hildenbrand */ 3401e4a2ed94SMatthew Wilcox (Oracle) if (folio_test_ksm(folio) || folio_ref_count(folio) > 3) 3402d4c47097SDavid Hildenbrand goto copy; 3403e4a2ed94SMatthew Wilcox (Oracle) if (!folio_test_lru(folio)) 3404d4c47097SDavid Hildenbrand /* 34051fec6890SMatthew Wilcox (Oracle) * We cannot easily detect+handle references from 34061fec6890SMatthew Wilcox (Oracle) * remote LRU caches or references to LRU folios. 3407d4c47097SDavid Hildenbrand */ 3408d4c47097SDavid Hildenbrand lru_add_drain(); 3409e4a2ed94SMatthew Wilcox (Oracle) if (folio_ref_count(folio) > 1 + folio_test_swapcache(folio)) 341052d1e606SKirill Tkhai goto copy; 3411e4a2ed94SMatthew Wilcox (Oracle) if (!folio_trylock(folio)) 341252d1e606SKirill Tkhai goto copy; 3413e4a2ed94SMatthew Wilcox (Oracle) if (folio_test_swapcache(folio)) 3414e4a2ed94SMatthew Wilcox (Oracle) folio_free_swap(folio); 3415e4a2ed94SMatthew Wilcox (Oracle) if (folio_test_ksm(folio) || folio_ref_count(folio) != 1) { 3416e4a2ed94SMatthew Wilcox (Oracle) folio_unlock(folio); 341709854ba9SLinus Torvalds goto copy; 341852d1e606SKirill Tkhai } 3419c44b6743SRik van Riel /* 3420e4a2ed94SMatthew Wilcox (Oracle) * Ok, we've got the only folio reference from our mapping 3421e4a2ed94SMatthew Wilcox (Oracle) * and the folio is locked, it's dark out, and we're wearing 342253a05ad9SDavid Hildenbrand * sunglasses. Hit it. 3423c44b6743SRik van Riel */ 3424e4a2ed94SMatthew Wilcox (Oracle) page_move_anon_rmap(vmf->page, vma); 3425e4a2ed94SMatthew Wilcox (Oracle) folio_unlock(folio); 34266c287605SDavid Hildenbrand reuse: 3427c89357e2SDavid Hildenbrand if (unlikely(unshare)) { 3428c89357e2SDavid Hildenbrand pte_unmap_unlock(vmf->pte, vmf->ptl); 3429c89357e2SDavid Hildenbrand return 0; 3430c89357e2SDavid Hildenbrand } 3431be068f29SLinus Torvalds wp_page_reuse(vmf); 3432cb8d8633SDavid Hildenbrand return 0; 34331da177e4SLinus Torvalds } 343452d1e606SKirill Tkhai copy: 34351da177e4SLinus Torvalds /* 34361da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 34371da177e4SLinus Torvalds */ 3438b9086fdeSDavid Hildenbrand if (folio) 3439b9086fdeSDavid Hildenbrand folio_get(folio); 344028766805SShachar Raindel 344182b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 344294bfe85bSYang Yang #ifdef CONFIG_KSM 3443b9086fdeSDavid Hildenbrand if (folio && folio_test_ksm(folio)) 344494bfe85bSYang Yang count_vm_event(COW_KSM); 344594bfe85bSYang Yang #endif 3446a41b70d6SJan Kara return wp_page_copy(vmf); 34471da177e4SLinus Torvalds } 34481da177e4SLinus Torvalds 344997a89413SPeter Zijlstra static void unmap_mapping_range_vma(struct vm_area_struct *vma, 34501da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 34511da177e4SLinus Torvalds struct zap_details *details) 34521da177e4SLinus Torvalds { 3453f5cc4eefSAl Viro zap_page_range_single(vma, start_addr, end_addr - start_addr, details); 34541da177e4SLinus Torvalds } 34551da177e4SLinus Torvalds 3456f808c13fSDavidlohr Bueso static inline void unmap_mapping_range_tree(struct rb_root_cached *root, 3457232a6a1cSPeter Xu pgoff_t first_index, 3458232a6a1cSPeter Xu pgoff_t last_index, 34591da177e4SLinus Torvalds struct zap_details *details) 34601da177e4SLinus Torvalds { 34611da177e4SLinus Torvalds struct vm_area_struct *vma; 34621da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 34631da177e4SLinus Torvalds 3464232a6a1cSPeter Xu vma_interval_tree_foreach(vma, root, first_index, last_index) { 34651da177e4SLinus Torvalds vba = vma->vm_pgoff; 3466d6e93217SLibin vea = vba + vma_pages(vma) - 1; 3467f9871da9SMiaohe Lin zba = max(first_index, vba); 3468f9871da9SMiaohe Lin zea = min(last_index, vea); 34691da177e4SLinus Torvalds 347097a89413SPeter Zijlstra unmap_mapping_range_vma(vma, 34711da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 34721da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 347397a89413SPeter Zijlstra details); 34741da177e4SLinus Torvalds } 34751da177e4SLinus Torvalds } 34761da177e4SLinus Torvalds 34771da177e4SLinus Torvalds /** 34783506659eSMatthew Wilcox (Oracle) * unmap_mapping_folio() - Unmap single folio from processes. 34793506659eSMatthew Wilcox (Oracle) * @folio: The locked folio to be unmapped. 348022061a1fSHugh Dickins * 34813506659eSMatthew Wilcox (Oracle) * Unmap this folio from any userspace process which still has it mmaped. 348222061a1fSHugh Dickins * Typically, for efficiency, the range of nearby pages has already been 348322061a1fSHugh Dickins * unmapped by unmap_mapping_pages() or unmap_mapping_range(). But once 34843506659eSMatthew Wilcox (Oracle) * truncation or invalidation holds the lock on a folio, it may find that 34853506659eSMatthew Wilcox (Oracle) * the page has been remapped again: and then uses unmap_mapping_folio() 348622061a1fSHugh Dickins * to unmap it finally. 348722061a1fSHugh Dickins */ 34883506659eSMatthew Wilcox (Oracle) void unmap_mapping_folio(struct folio *folio) 348922061a1fSHugh Dickins { 34903506659eSMatthew Wilcox (Oracle) struct address_space *mapping = folio->mapping; 349122061a1fSHugh Dickins struct zap_details details = { }; 3492232a6a1cSPeter Xu pgoff_t first_index; 3493232a6a1cSPeter Xu pgoff_t last_index; 349422061a1fSHugh Dickins 34953506659eSMatthew Wilcox (Oracle) VM_BUG_ON(!folio_test_locked(folio)); 349622061a1fSHugh Dickins 34973506659eSMatthew Wilcox (Oracle) first_index = folio->index; 34983506659eSMatthew Wilcox (Oracle) last_index = folio->index + folio_nr_pages(folio) - 1; 3499232a6a1cSPeter Xu 35002e148f1eSPeter Xu details.even_cows = false; 35013506659eSMatthew Wilcox (Oracle) details.single_folio = folio; 3502999dad82SPeter Xu details.zap_flags = ZAP_FLAG_DROP_MARKER; 350322061a1fSHugh Dickins 35042c865995SHugh Dickins i_mmap_lock_read(mapping); 350522061a1fSHugh Dickins if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) 3506232a6a1cSPeter Xu unmap_mapping_range_tree(&mapping->i_mmap, first_index, 3507232a6a1cSPeter Xu last_index, &details); 35082c865995SHugh Dickins i_mmap_unlock_read(mapping); 350922061a1fSHugh Dickins } 351022061a1fSHugh Dickins 351122061a1fSHugh Dickins /** 3512977fbdcdSMatthew Wilcox * unmap_mapping_pages() - Unmap pages from processes. 3513977fbdcdSMatthew Wilcox * @mapping: The address space containing pages to be unmapped. 3514977fbdcdSMatthew Wilcox * @start: Index of first page to be unmapped. 3515977fbdcdSMatthew Wilcox * @nr: Number of pages to be unmapped. 0 to unmap to end of file. 3516977fbdcdSMatthew Wilcox * @even_cows: Whether to unmap even private COWed pages. 3517977fbdcdSMatthew Wilcox * 3518977fbdcdSMatthew Wilcox * Unmap the pages in this address space from any userspace process which 3519977fbdcdSMatthew Wilcox * has them mmaped. Generally, you want to remove COWed pages as well when 3520977fbdcdSMatthew Wilcox * a file is being truncated, but not when invalidating pages from the page 3521977fbdcdSMatthew Wilcox * cache. 3522977fbdcdSMatthew Wilcox */ 3523977fbdcdSMatthew Wilcox void unmap_mapping_pages(struct address_space *mapping, pgoff_t start, 3524977fbdcdSMatthew Wilcox pgoff_t nr, bool even_cows) 3525977fbdcdSMatthew Wilcox { 3526977fbdcdSMatthew Wilcox struct zap_details details = { }; 3527232a6a1cSPeter Xu pgoff_t first_index = start; 3528232a6a1cSPeter Xu pgoff_t last_index = start + nr - 1; 3529977fbdcdSMatthew Wilcox 35302e148f1eSPeter Xu details.even_cows = even_cows; 3531232a6a1cSPeter Xu if (last_index < first_index) 3532232a6a1cSPeter Xu last_index = ULONG_MAX; 3533977fbdcdSMatthew Wilcox 35342c865995SHugh Dickins i_mmap_lock_read(mapping); 3535977fbdcdSMatthew Wilcox if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) 3536232a6a1cSPeter Xu unmap_mapping_range_tree(&mapping->i_mmap, first_index, 3537232a6a1cSPeter Xu last_index, &details); 35382c865995SHugh Dickins i_mmap_unlock_read(mapping); 3539977fbdcdSMatthew Wilcox } 35406e0e99d5SDavid Howells EXPORT_SYMBOL_GPL(unmap_mapping_pages); 3541977fbdcdSMatthew Wilcox 3542977fbdcdSMatthew Wilcox /** 35438a5f14a2SKirill A. Shutemov * unmap_mapping_range - unmap the portion of all mmaps in the specified 3544977fbdcdSMatthew Wilcox * address_space corresponding to the specified byte range in the underlying 35458a5f14a2SKirill A. Shutemov * file. 35468a5f14a2SKirill A. Shutemov * 35473d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 35481da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 35491da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 355025d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 35511da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 35521da177e4SLinus Torvalds * partial pages. 35531da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 35541da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 35551da177e4SLinus Torvalds * end of the file. 35561da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 35571da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 35581da177e4SLinus Torvalds */ 35591da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 35601da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 35611da177e4SLinus Torvalds { 35621da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 35631da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 35641da177e4SLinus Torvalds 35651da177e4SLinus Torvalds /* Check for overflow. */ 35661da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 35671da177e4SLinus Torvalds long long holeend = 35681da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 35691da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 35701da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 35711da177e4SLinus Torvalds } 35721da177e4SLinus Torvalds 3573977fbdcdSMatthew Wilcox unmap_mapping_pages(mapping, hba, hlen, even_cows); 35741da177e4SLinus Torvalds } 35751da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 35761da177e4SLinus Torvalds 35771da177e4SLinus Torvalds /* 3578b756a3b5SAlistair Popple * Restore a potential device exclusive pte to a working pte entry 3579b756a3b5SAlistair Popple */ 3580b756a3b5SAlistair Popple static vm_fault_t remove_device_exclusive_entry(struct vm_fault *vmf) 3581b756a3b5SAlistair Popple { 358219672a9eSMatthew Wilcox (Oracle) struct folio *folio = page_folio(vmf->page); 3583b756a3b5SAlistair Popple struct vm_area_struct *vma = vmf->vma; 3584b756a3b5SAlistair Popple struct mmu_notifier_range range; 3585b756a3b5SAlistair Popple 35867c7b9629SAlistair Popple /* 35877c7b9629SAlistair Popple * We need a reference to lock the folio because we don't hold 35887c7b9629SAlistair Popple * the PTL so a racing thread can remove the device-exclusive 35897c7b9629SAlistair Popple * entry and unmap it. If the folio is free the entry must 35907c7b9629SAlistair Popple * have been removed already. If it happens to have already 35917c7b9629SAlistair Popple * been re-allocated after being freed all we do is lock and 35927c7b9629SAlistair Popple * unlock it. 35937c7b9629SAlistair Popple */ 35947c7b9629SAlistair Popple if (!folio_try_get(folio)) 35957c7b9629SAlistair Popple return 0; 35967c7b9629SAlistair Popple 35977c7b9629SAlistair Popple if (!folio_lock_or_retry(folio, vma->vm_mm, vmf->flags)) { 35987c7b9629SAlistair Popple folio_put(folio); 3599b756a3b5SAlistair Popple return VM_FAULT_RETRY; 36007c7b9629SAlistair Popple } 36017d4a8be0SAlistair Popple mmu_notifier_range_init_owner(&range, MMU_NOTIFY_EXCLUSIVE, 0, 3602b756a3b5SAlistair Popple vma->vm_mm, vmf->address & PAGE_MASK, 3603b756a3b5SAlistair Popple (vmf->address & PAGE_MASK) + PAGE_SIZE, NULL); 3604b756a3b5SAlistair Popple mmu_notifier_invalidate_range_start(&range); 3605b756a3b5SAlistair Popple 3606b756a3b5SAlistair Popple vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 3607b756a3b5SAlistair Popple &vmf->ptl); 3608c33c7948SRyan Roberts if (likely(vmf->pte && pte_same(ptep_get(vmf->pte), vmf->orig_pte))) 360919672a9eSMatthew Wilcox (Oracle) restore_exclusive_pte(vma, vmf->page, vmf->address, vmf->pte); 3610b756a3b5SAlistair Popple 36113db82b93SHugh Dickins if (vmf->pte) 3612b756a3b5SAlistair Popple pte_unmap_unlock(vmf->pte, vmf->ptl); 361319672a9eSMatthew Wilcox (Oracle) folio_unlock(folio); 36147c7b9629SAlistair Popple folio_put(folio); 3615b756a3b5SAlistair Popple 3616b756a3b5SAlistair Popple mmu_notifier_invalidate_range_end(&range); 3617b756a3b5SAlistair Popple return 0; 3618b756a3b5SAlistair Popple } 3619b756a3b5SAlistair Popple 3620a160e537SMatthew Wilcox (Oracle) static inline bool should_try_to_free_swap(struct folio *folio, 3621c145e0b4SDavid Hildenbrand struct vm_area_struct *vma, 3622c145e0b4SDavid Hildenbrand unsigned int fault_flags) 3623c145e0b4SDavid Hildenbrand { 3624a160e537SMatthew Wilcox (Oracle) if (!folio_test_swapcache(folio)) 3625c145e0b4SDavid Hildenbrand return false; 36269202d527SMatthew Wilcox (Oracle) if (mem_cgroup_swap_full(folio) || (vma->vm_flags & VM_LOCKED) || 3627a160e537SMatthew Wilcox (Oracle) folio_test_mlocked(folio)) 3628c145e0b4SDavid Hildenbrand return true; 3629c145e0b4SDavid Hildenbrand /* 3630c145e0b4SDavid Hildenbrand * If we want to map a page that's in the swapcache writable, we 3631c145e0b4SDavid Hildenbrand * have to detect via the refcount if we're really the exclusive 3632c145e0b4SDavid Hildenbrand * user. Try freeing the swapcache to get rid of the swapcache 3633c145e0b4SDavid Hildenbrand * reference only in case it's likely that we'll be the exlusive user. 3634c145e0b4SDavid Hildenbrand */ 3635a160e537SMatthew Wilcox (Oracle) return (fault_flags & FAULT_FLAG_WRITE) && !folio_test_ksm(folio) && 3636a160e537SMatthew Wilcox (Oracle) folio_ref_count(folio) == 2; 3637c145e0b4SDavid Hildenbrand } 3638c145e0b4SDavid Hildenbrand 36399c28a205SPeter Xu static vm_fault_t pte_marker_clear(struct vm_fault *vmf) 36409c28a205SPeter Xu { 36419c28a205SPeter Xu vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, 36429c28a205SPeter Xu vmf->address, &vmf->ptl); 36433db82b93SHugh Dickins if (!vmf->pte) 36443db82b93SHugh Dickins return 0; 36459c28a205SPeter Xu /* 36469c28a205SPeter Xu * Be careful so that we will only recover a special uffd-wp pte into a 36479c28a205SPeter Xu * none pte. Otherwise it means the pte could have changed, so retry. 36487e3ce3f8SPeter Xu * 36497e3ce3f8SPeter Xu * This should also cover the case where e.g. the pte changed 36507e3ce3f8SPeter Xu * quickly from a PTE_MARKER_UFFD_WP into PTE_MARKER_SWAPIN_ERROR. 36517e3ce3f8SPeter Xu * So is_pte_marker() check is not enough to safely drop the pte. 36529c28a205SPeter Xu */ 3653c33c7948SRyan Roberts if (pte_same(vmf->orig_pte, ptep_get(vmf->pte))) 36549c28a205SPeter Xu pte_clear(vmf->vma->vm_mm, vmf->address, vmf->pte); 36559c28a205SPeter Xu pte_unmap_unlock(vmf->pte, vmf->ptl); 36569c28a205SPeter Xu return 0; 36579c28a205SPeter Xu } 36589c28a205SPeter Xu 36592bad466cSPeter Xu static vm_fault_t do_pte_missing(struct vm_fault *vmf) 36602bad466cSPeter Xu { 36612bad466cSPeter Xu if (vma_is_anonymous(vmf->vma)) 36622bad466cSPeter Xu return do_anonymous_page(vmf); 36632bad466cSPeter Xu else 36642bad466cSPeter Xu return do_fault(vmf); 36652bad466cSPeter Xu } 36662bad466cSPeter Xu 36679c28a205SPeter Xu /* 36689c28a205SPeter Xu * This is actually a page-missing access, but with uffd-wp special pte 36699c28a205SPeter Xu * installed. It means this pte was wr-protected before being unmapped. 36709c28a205SPeter Xu */ 36719c28a205SPeter Xu static vm_fault_t pte_marker_handle_uffd_wp(struct vm_fault *vmf) 36729c28a205SPeter Xu { 36739c28a205SPeter Xu /* 36749c28a205SPeter Xu * Just in case there're leftover special ptes even after the region 36757a079ba2SPeter Xu * got unregistered - we can simply clear them. 36769c28a205SPeter Xu */ 36772bad466cSPeter Xu if (unlikely(!userfaultfd_wp(vmf->vma))) 36789c28a205SPeter Xu return pte_marker_clear(vmf); 36799c28a205SPeter Xu 36802bad466cSPeter Xu return do_pte_missing(vmf); 36819c28a205SPeter Xu } 36829c28a205SPeter Xu 36835c041f5dSPeter Xu static vm_fault_t handle_pte_marker(struct vm_fault *vmf) 36845c041f5dSPeter Xu { 36855c041f5dSPeter Xu swp_entry_t entry = pte_to_swp_entry(vmf->orig_pte); 36865c041f5dSPeter Xu unsigned long marker = pte_marker_get(entry); 36875c041f5dSPeter Xu 36885c041f5dSPeter Xu /* 3689ca92ea3dSPeter Xu * PTE markers should never be empty. If anything weird happened, 3690ca92ea3dSPeter Xu * the best thing to do is to kill the process along with its mm. 36915c041f5dSPeter Xu */ 3692ca92ea3dSPeter Xu if (WARN_ON_ONCE(!marker)) 36935c041f5dSPeter Xu return VM_FAULT_SIGBUS; 36945c041f5dSPeter Xu 369515520a3fSPeter Xu /* Higher priority than uffd-wp when data corrupted */ 369615520a3fSPeter Xu if (marker & PTE_MARKER_SWAPIN_ERROR) 369715520a3fSPeter Xu return VM_FAULT_SIGBUS; 369815520a3fSPeter Xu 36999c28a205SPeter Xu if (pte_marker_entry_uffd_wp(entry)) 37009c28a205SPeter Xu return pte_marker_handle_uffd_wp(vmf); 37019c28a205SPeter Xu 37029c28a205SPeter Xu /* This is an unknown pte marker */ 37039c28a205SPeter Xu return VM_FAULT_SIGBUS; 37045c041f5dSPeter Xu } 37055c041f5dSPeter Xu 3706b756a3b5SAlistair Popple /* 3707c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 37088f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 37099a95f3cfSPaul Cassella * We return with pte unmapped and unlocked. 37109a95f3cfSPaul Cassella * 3711c1e8d7c6SMichel Lespinasse * We return with the mmap_lock locked or unlocked in the same cases 37129a95f3cfSPaul Cassella * as does filemap_fault(). 37131da177e4SLinus Torvalds */ 37142b740303SSouptick Joarder vm_fault_t do_swap_page(struct vm_fault *vmf) 37151da177e4SLinus Torvalds { 371682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3717d4f9565aSMatthew Wilcox (Oracle) struct folio *swapcache, *folio = NULL; 3718d4f9565aSMatthew Wilcox (Oracle) struct page *page; 37192799e775SMiaohe Lin struct swap_info_struct *si = NULL; 372014f9135dSDavid Hildenbrand rmap_t rmap_flags = RMAP_NONE; 37211493a191SDavid Hildenbrand bool exclusive = false; 372265500d23SHugh Dickins swp_entry_t entry; 37231da177e4SLinus Torvalds pte_t pte; 3724d065bd81SMichel Lespinasse int locked; 37252b740303SSouptick Joarder vm_fault_t ret = 0; 3726aae466b0SJoonsoo Kim void *shadow = NULL; 37271da177e4SLinus Torvalds 37282ca99358SPeter Xu if (!pte_unmap_same(vmf)) 37298f4e2101SHugh Dickins goto out; 373065500d23SHugh Dickins 373117c05f18SSuren Baghdasaryan if (vmf->flags & FAULT_FLAG_VMA_LOCK) { 373217c05f18SSuren Baghdasaryan ret = VM_FAULT_RETRY; 373317c05f18SSuren Baghdasaryan goto out; 373417c05f18SSuren Baghdasaryan } 373517c05f18SSuren Baghdasaryan 37362994302bSJan Kara entry = pte_to_swp_entry(vmf->orig_pte); 3737d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 37380697212aSChristoph Lameter if (is_migration_entry(entry)) { 373982b0f8c3SJan Kara migration_entry_wait(vma->vm_mm, vmf->pmd, 374082b0f8c3SJan Kara vmf->address); 3741b756a3b5SAlistair Popple } else if (is_device_exclusive_entry(entry)) { 3742b756a3b5SAlistair Popple vmf->page = pfn_swap_entry_to_page(entry); 3743b756a3b5SAlistair Popple ret = remove_device_exclusive_entry(vmf); 37445042db43SJérôme Glisse } else if (is_device_private_entry(entry)) { 3745af5cdaf8SAlistair Popple vmf->page = pfn_swap_entry_to_page(entry); 374616ce101dSAlistair Popple vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 374716ce101dSAlistair Popple vmf->address, &vmf->ptl); 37483db82b93SHugh Dickins if (unlikely(!vmf->pte || 3749c33c7948SRyan Roberts !pte_same(ptep_get(vmf->pte), 3750c33c7948SRyan Roberts vmf->orig_pte))) 37513b65f437SRyan Roberts goto unlock; 375216ce101dSAlistair Popple 375316ce101dSAlistair Popple /* 375416ce101dSAlistair Popple * Get a page reference while we know the page can't be 375516ce101dSAlistair Popple * freed. 375616ce101dSAlistair Popple */ 375716ce101dSAlistair Popple get_page(vmf->page); 375816ce101dSAlistair Popple pte_unmap_unlock(vmf->pte, vmf->ptl); 37594a955bedSAlistair Popple ret = vmf->page->pgmap->ops->migrate_to_ram(vmf); 376016ce101dSAlistair Popple put_page(vmf->page); 3761d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 3762d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 37635c041f5dSPeter Xu } else if (is_pte_marker_entry(entry)) { 37645c041f5dSPeter Xu ret = handle_pte_marker(vmf); 3765d1737fdbSAndi Kleen } else { 37662994302bSJan Kara print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); 3767d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 3768d1737fdbSAndi Kleen } 37690697212aSChristoph Lameter goto out; 37700697212aSChristoph Lameter } 37710bcac06fSMinchan Kim 37722799e775SMiaohe Lin /* Prevent swapoff from happening to us. */ 37732799e775SMiaohe Lin si = get_swap_device(entry); 37742799e775SMiaohe Lin if (unlikely(!si)) 37752799e775SMiaohe Lin goto out; 37760bcac06fSMinchan Kim 37775a423081SMatthew Wilcox (Oracle) folio = swap_cache_get_folio(entry, vma, vmf->address); 37785a423081SMatthew Wilcox (Oracle) if (folio) 37795a423081SMatthew Wilcox (Oracle) page = folio_file_page(folio, swp_offset(entry)); 3780d4f9565aSMatthew Wilcox (Oracle) swapcache = folio; 3781f8020772SMinchan Kim 3782d4f9565aSMatthew Wilcox (Oracle) if (!folio) { 3783a449bf58SQian Cai if (data_race(si->flags & SWP_SYNCHRONOUS_IO) && 3784eb085574SHuang Ying __swap_count(entry) == 1) { 37850bcac06fSMinchan Kim /* skip swapcache */ 378663ad4addSMatthew Wilcox (Oracle) folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, 378763ad4addSMatthew Wilcox (Oracle) vma, vmf->address, false); 378863ad4addSMatthew Wilcox (Oracle) page = &folio->page; 378963ad4addSMatthew Wilcox (Oracle) if (folio) { 379063ad4addSMatthew Wilcox (Oracle) __folio_set_locked(folio); 379163ad4addSMatthew Wilcox (Oracle) __folio_set_swapbacked(folio); 37924c6355b2SJohannes Weiner 379365995918SMatthew Wilcox (Oracle) if (mem_cgroup_swapin_charge_folio(folio, 379463ad4addSMatthew Wilcox (Oracle) vma->vm_mm, GFP_KERNEL, 379563ad4addSMatthew Wilcox (Oracle) entry)) { 3796545b1b07SMichal Hocko ret = VM_FAULT_OOM; 37974c6355b2SJohannes Weiner goto out_page; 3798545b1b07SMichal Hocko } 37990add0c77SShakeel Butt mem_cgroup_swapin_uncharge_swap(entry); 38004c6355b2SJohannes Weiner 3801aae466b0SJoonsoo Kim shadow = get_shadow_from_swap_cache(entry); 3802aae466b0SJoonsoo Kim if (shadow) 380363ad4addSMatthew Wilcox (Oracle) workingset_refault(folio, shadow); 38040076f029SJoonsoo Kim 380563ad4addSMatthew Wilcox (Oracle) folio_add_lru(folio); 38060add0c77SShakeel Butt 38070add0c77SShakeel Butt /* To provide entry to swap_readpage() */ 380863ad4addSMatthew Wilcox (Oracle) folio_set_swap_entry(folio, entry); 38095169b844SNeilBrown swap_readpage(page, true, NULL); 381063ad4addSMatthew Wilcox (Oracle) folio->private = NULL; 38110bcac06fSMinchan Kim } 3812aa8d22a1SMinchan Kim } else { 3813e9e9b7ecSMinchan Kim page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, 3814e9e9b7ecSMinchan Kim vmf); 381563ad4addSMatthew Wilcox (Oracle) if (page) 381663ad4addSMatthew Wilcox (Oracle) folio = page_folio(page); 3817d4f9565aSMatthew Wilcox (Oracle) swapcache = folio; 38180bcac06fSMinchan Kim } 38190bcac06fSMinchan Kim 3820d4f9565aSMatthew Wilcox (Oracle) if (!folio) { 38211da177e4SLinus Torvalds /* 38228f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 38238f4e2101SHugh Dickins * while we released the pte lock. 38241da177e4SLinus Torvalds */ 382582b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 382682b0f8c3SJan Kara vmf->address, &vmf->ptl); 3827c33c7948SRyan Roberts if (likely(vmf->pte && 3828c33c7948SRyan Roberts pte_same(ptep_get(vmf->pte), vmf->orig_pte))) 38291da177e4SLinus Torvalds ret = VM_FAULT_OOM; 383065500d23SHugh Dickins goto unlock; 38311da177e4SLinus Torvalds } 38321da177e4SLinus Torvalds 38331da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 38341da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 3835f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 38362262185cSRoman Gushchin count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); 3837d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 383871f72525SWu Fengguang /* 383971f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 384071f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 384171f72525SWu Fengguang */ 3842d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 38434779cb31SAndi Kleen goto out_release; 38441da177e4SLinus Torvalds } 38451da177e4SLinus Torvalds 384619672a9eSMatthew Wilcox (Oracle) locked = folio_lock_or_retry(folio, vma->vm_mm, vmf->flags); 3847e709ffd6SRik van Riel 3848d065bd81SMichel Lespinasse if (!locked) { 3849d065bd81SMichel Lespinasse ret |= VM_FAULT_RETRY; 3850d065bd81SMichel Lespinasse goto out_release; 3851d065bd81SMichel Lespinasse } 38521da177e4SLinus Torvalds 385384d60fddSDavid Hildenbrand if (swapcache) { 38544969c119SAndrea Arcangeli /* 38553b344157SMatthew Wilcox (Oracle) * Make sure folio_free_swap() or swapoff did not release the 385684d60fddSDavid Hildenbrand * swapcache from under us. The page pin, and pte_same test 385784d60fddSDavid Hildenbrand * below, are not enough to exclude that. Even if it is still 385884d60fddSDavid Hildenbrand * swapcache, we need to check that the page's swap has not 385984d60fddSDavid Hildenbrand * changed. 38604969c119SAndrea Arcangeli */ 386163ad4addSMatthew Wilcox (Oracle) if (unlikely(!folio_test_swapcache(folio) || 386284d60fddSDavid Hildenbrand page_private(page) != entry.val)) 38634969c119SAndrea Arcangeli goto out_page; 38644969c119SAndrea Arcangeli 386584d60fddSDavid Hildenbrand /* 386684d60fddSDavid Hildenbrand * KSM sometimes has to copy on read faults, for example, if 386784d60fddSDavid Hildenbrand * page->index of !PageKSM() pages would be nonlinear inside the 386884d60fddSDavid Hildenbrand * anon VMA -- PageKSM() is lost on actual swapout. 386984d60fddSDavid Hildenbrand */ 387082b0f8c3SJan Kara page = ksm_might_need_to_copy(page, vma, vmf->address); 38714969c119SAndrea Arcangeli if (unlikely(!page)) { 38725ad64688SHugh Dickins ret = VM_FAULT_OOM; 38734969c119SAndrea Arcangeli goto out_page; 38746b970599SKefeng Wang } else if (unlikely(PTR_ERR(page) == -EHWPOISON)) { 38756b970599SKefeng Wang ret = VM_FAULT_HWPOISON; 38766b970599SKefeng Wang goto out_page; 38774969c119SAndrea Arcangeli } 387863ad4addSMatthew Wilcox (Oracle) folio = page_folio(page); 3879c145e0b4SDavid Hildenbrand 3880c145e0b4SDavid Hildenbrand /* 3881c145e0b4SDavid Hildenbrand * If we want to map a page that's in the swapcache writable, we 3882c145e0b4SDavid Hildenbrand * have to detect via the refcount if we're really the exclusive 3883c145e0b4SDavid Hildenbrand * owner. Try removing the extra reference from the local LRU 38841fec6890SMatthew Wilcox (Oracle) * caches if required. 3885c145e0b4SDavid Hildenbrand */ 3886d4f9565aSMatthew Wilcox (Oracle) if ((vmf->flags & FAULT_FLAG_WRITE) && folio == swapcache && 388763ad4addSMatthew Wilcox (Oracle) !folio_test_ksm(folio) && !folio_test_lru(folio)) 3888c145e0b4SDavid Hildenbrand lru_add_drain(); 388984d60fddSDavid Hildenbrand } 38905ad64688SHugh Dickins 38914231f842SKefeng Wang folio_throttle_swaprate(folio, GFP_KERNEL); 3892073e587eSKAMEZAWA Hiroyuki 38931da177e4SLinus Torvalds /* 38948f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 38951da177e4SLinus Torvalds */ 389682b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 389782b0f8c3SJan Kara &vmf->ptl); 3898c33c7948SRyan Roberts if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) 3899b8107480SKirill Korotaev goto out_nomap; 3900b8107480SKirill Korotaev 390163ad4addSMatthew Wilcox (Oracle) if (unlikely(!folio_test_uptodate(folio))) { 3902b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 3903b8107480SKirill Korotaev goto out_nomap; 39041da177e4SLinus Torvalds } 39051da177e4SLinus Torvalds 39068c7c6e34SKAMEZAWA Hiroyuki /* 390778fbe906SDavid Hildenbrand * PG_anon_exclusive reuses PG_mappedtodisk for anon pages. A swap pte 390878fbe906SDavid Hildenbrand * must never point at an anonymous page in the swapcache that is 390978fbe906SDavid Hildenbrand * PG_anon_exclusive. Sanity check that this holds and especially, that 391078fbe906SDavid Hildenbrand * no filesystem set PG_mappedtodisk on a page in the swapcache. Sanity 391178fbe906SDavid Hildenbrand * check after taking the PT lock and making sure that nobody 391278fbe906SDavid Hildenbrand * concurrently faulted in this page and set PG_anon_exclusive. 391378fbe906SDavid Hildenbrand */ 391463ad4addSMatthew Wilcox (Oracle) BUG_ON(!folio_test_anon(folio) && folio_test_mappedtodisk(folio)); 391563ad4addSMatthew Wilcox (Oracle) BUG_ON(folio_test_anon(folio) && PageAnonExclusive(page)); 391678fbe906SDavid Hildenbrand 391778fbe906SDavid Hildenbrand /* 39181493a191SDavid Hildenbrand * Check under PT lock (to protect against concurrent fork() sharing 39191493a191SDavid Hildenbrand * the swap entry concurrently) for certainly exclusive pages. 39201493a191SDavid Hildenbrand */ 392163ad4addSMatthew Wilcox (Oracle) if (!folio_test_ksm(folio)) { 39221493a191SDavid Hildenbrand exclusive = pte_swp_exclusive(vmf->orig_pte); 3923d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache) { 39241493a191SDavid Hildenbrand /* 39251493a191SDavid Hildenbrand * We have a fresh page that is not exposed to the 39261493a191SDavid Hildenbrand * swapcache -> certainly exclusive. 39271493a191SDavid Hildenbrand */ 39281493a191SDavid Hildenbrand exclusive = true; 392963ad4addSMatthew Wilcox (Oracle) } else if (exclusive && folio_test_writeback(folio) && 3930eacde327SMiaohe Lin data_race(si->flags & SWP_STABLE_WRITES)) { 39311493a191SDavid Hildenbrand /* 39321493a191SDavid Hildenbrand * This is tricky: not all swap backends support 39331493a191SDavid Hildenbrand * concurrent page modifications while under writeback. 39341493a191SDavid Hildenbrand * 39351493a191SDavid Hildenbrand * So if we stumble over such a page in the swapcache 39361493a191SDavid Hildenbrand * we must not set the page exclusive, otherwise we can 39371493a191SDavid Hildenbrand * map it writable without further checks and modify it 39381493a191SDavid Hildenbrand * while still under writeback. 39391493a191SDavid Hildenbrand * 39401493a191SDavid Hildenbrand * For these problematic swap backends, simply drop the 39411493a191SDavid Hildenbrand * exclusive marker: this is perfectly fine as we start 39421493a191SDavid Hildenbrand * writeback only if we fully unmapped the page and 39431493a191SDavid Hildenbrand * there are no unexpected references on the page after 39441493a191SDavid Hildenbrand * unmapping succeeded. After fully unmapped, no 39451493a191SDavid Hildenbrand * further GUP references (FOLL_GET and FOLL_PIN) can 39461493a191SDavid Hildenbrand * appear, so dropping the exclusive marker and mapping 39471493a191SDavid Hildenbrand * it only R/O is fine. 39481493a191SDavid Hildenbrand */ 39491493a191SDavid Hildenbrand exclusive = false; 39501493a191SDavid Hildenbrand } 39511493a191SDavid Hildenbrand } 39521493a191SDavid Hildenbrand 39531493a191SDavid Hildenbrand /* 3954c145e0b4SDavid Hildenbrand * Remove the swap entry and conditionally try to free up the swapcache. 3955c145e0b4SDavid Hildenbrand * We're already holding a reference on the page but haven't mapped it 3956c145e0b4SDavid Hildenbrand * yet. 39578c7c6e34SKAMEZAWA Hiroyuki */ 3958c145e0b4SDavid Hildenbrand swap_free(entry); 3959a160e537SMatthew Wilcox (Oracle) if (should_try_to_free_swap(folio, vma, vmf->flags)) 3960a160e537SMatthew Wilcox (Oracle) folio_free_swap(folio); 39611da177e4SLinus Torvalds 3962f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, MM_ANONPAGES); 3963f1a79412SShakeel Butt dec_mm_counter(vma->vm_mm, MM_SWAPENTS); 39641da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 3965c145e0b4SDavid Hildenbrand 3966c145e0b4SDavid Hildenbrand /* 39671493a191SDavid Hildenbrand * Same logic as in do_wp_page(); however, optimize for pages that are 39681493a191SDavid Hildenbrand * certainly not shared either because we just allocated them without 39691493a191SDavid Hildenbrand * exposing them to the swapcache or because the swap entry indicates 39701493a191SDavid Hildenbrand * exclusivity. 3971c145e0b4SDavid Hildenbrand */ 397263ad4addSMatthew Wilcox (Oracle) if (!folio_test_ksm(folio) && 397363ad4addSMatthew Wilcox (Oracle) (exclusive || folio_ref_count(folio) == 1)) { 39746c287605SDavid Hildenbrand if (vmf->flags & FAULT_FLAG_WRITE) { 39751da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 397682b0f8c3SJan Kara vmf->flags &= ~FAULT_FLAG_WRITE; 39776c287605SDavid Hildenbrand } 397814f9135dSDavid Hildenbrand rmap_flags |= RMAP_EXCLUSIVE; 39791da177e4SLinus Torvalds } 39801da177e4SLinus Torvalds flush_icache_page(vma, page); 39812994302bSJan Kara if (pte_swp_soft_dirty(vmf->orig_pte)) 3982179ef71cSCyrill Gorcunov pte = pte_mksoft_dirty(pte); 3983f1eb1bacSPeter Xu if (pte_swp_uffd_wp(vmf->orig_pte)) 3984f45ec5ffSPeter Xu pte = pte_mkuffd_wp(pte); 39852994302bSJan Kara vmf->orig_pte = pte; 39860bcac06fSMinchan Kim 39870bcac06fSMinchan Kim /* ksm created a completely new copy */ 3988d4f9565aSMatthew Wilcox (Oracle) if (unlikely(folio != swapcache && swapcache)) { 398940f2bbf7SDavid Hildenbrand page_add_new_anon_rmap(page, vma, vmf->address); 399063ad4addSMatthew Wilcox (Oracle) folio_add_lru_vma(folio, vma); 39910bcac06fSMinchan Kim } else { 3992f1e2db12SDavid Hildenbrand page_add_anon_rmap(page, vma, vmf->address, rmap_flags); 399300501b53SJohannes Weiner } 39941da177e4SLinus Torvalds 399563ad4addSMatthew Wilcox (Oracle) VM_BUG_ON(!folio_test_anon(folio) || 399663ad4addSMatthew Wilcox (Oracle) (pte_write(pte) && !PageAnonExclusive(page))); 39971eba86c0SPasha Tatashin set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); 39981eba86c0SPasha Tatashin arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte); 39991eba86c0SPasha Tatashin 400063ad4addSMatthew Wilcox (Oracle) folio_unlock(folio); 4001d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache && swapcache) { 40024969c119SAndrea Arcangeli /* 40034969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 40044969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 40054969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 40064969c119SAndrea Arcangeli * further safety release the lock after the swap_free 40074969c119SAndrea Arcangeli * so that the swap count won't change under a 40084969c119SAndrea Arcangeli * parallel locked swapcache. 40094969c119SAndrea Arcangeli */ 4010d4f9565aSMatthew Wilcox (Oracle) folio_unlock(swapcache); 4011d4f9565aSMatthew Wilcox (Oracle) folio_put(swapcache); 40124969c119SAndrea Arcangeli } 4013c475a8abSHugh Dickins 401482b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) { 40152994302bSJan Kara ret |= do_wp_page(vmf); 401661469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 401761469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 40181da177e4SLinus Torvalds goto out; 40191da177e4SLinus Torvalds } 40201da177e4SLinus Torvalds 40211da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 402282b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 402365500d23SHugh Dickins unlock: 40243db82b93SHugh Dickins if (vmf->pte) 402582b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 40261da177e4SLinus Torvalds out: 40272799e775SMiaohe Lin if (si) 40282799e775SMiaohe Lin put_swap_device(si); 40291da177e4SLinus Torvalds return ret; 4030b8107480SKirill Korotaev out_nomap: 40313db82b93SHugh Dickins if (vmf->pte) 403282b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 4033bc43f75cSJohannes Weiner out_page: 403463ad4addSMatthew Wilcox (Oracle) folio_unlock(folio); 40354779cb31SAndi Kleen out_release: 403663ad4addSMatthew Wilcox (Oracle) folio_put(folio); 4037d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache && swapcache) { 4038d4f9565aSMatthew Wilcox (Oracle) folio_unlock(swapcache); 4039d4f9565aSMatthew Wilcox (Oracle) folio_put(swapcache); 40404969c119SAndrea Arcangeli } 40412799e775SMiaohe Lin if (si) 40422799e775SMiaohe Lin put_swap_device(si); 404365500d23SHugh Dickins return ret; 40441da177e4SLinus Torvalds } 40451da177e4SLinus Torvalds 40461da177e4SLinus Torvalds /* 4047c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 40488f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 4049c1e8d7c6SMichel Lespinasse * We return with mmap_lock still held, but pte unmapped and unlocked. 40501da177e4SLinus Torvalds */ 40512b740303SSouptick Joarder static vm_fault_t do_anonymous_page(struct vm_fault *vmf) 40521da177e4SLinus Torvalds { 40532bad466cSPeter Xu bool uffd_wp = vmf_orig_pte_uffd_wp(vmf); 405482b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 40556bc56a4dSMatthew Wilcox (Oracle) struct folio *folio; 40562b740303SSouptick Joarder vm_fault_t ret = 0; 40571da177e4SLinus Torvalds pte_t entry; 40581da177e4SLinus Torvalds 40596b7339f4SKirill A. Shutemov /* File mapping without ->vm_ops ? */ 40606b7339f4SKirill A. Shutemov if (vma->vm_flags & VM_SHARED) 40616b7339f4SKirill A. Shutemov return VM_FAULT_SIGBUS; 40626b7339f4SKirill A. Shutemov 40637267ec00SKirill A. Shutemov /* 40643db82b93SHugh Dickins * Use pte_alloc() instead of pte_alloc_map(), so that OOM can 40653db82b93SHugh Dickins * be distinguished from a transient failure of pte_offset_map(). 40667267ec00SKirill A. Shutemov */ 40674cf58924SJoel Fernandes (Google) if (pte_alloc(vma->vm_mm, vmf->pmd)) 40687267ec00SKirill A. Shutemov return VM_FAULT_OOM; 40697267ec00SKirill A. Shutemov 407011ac5524SLinus Torvalds /* Use the zero-page for reads */ 407182b0f8c3SJan Kara if (!(vmf->flags & FAULT_FLAG_WRITE) && 4072bae473a4SKirill A. Shutemov !mm_forbids_zeropage(vma->vm_mm)) { 407382b0f8c3SJan Kara entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), 407462eede62SHugh Dickins vma->vm_page_prot)); 407582b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 407682b0f8c3SJan Kara vmf->address, &vmf->ptl); 40773db82b93SHugh Dickins if (!vmf->pte) 40783db82b93SHugh Dickins goto unlock; 40792bad466cSPeter Xu if (vmf_pte_changed(vmf)) { 40807df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 4081a13ea5b7SHugh Dickins goto unlock; 40827df67697SBibo Mao } 40836b31d595SMichal Hocko ret = check_stable_address_space(vma->vm_mm); 40846b31d595SMichal Hocko if (ret) 40856b31d595SMichal Hocko goto unlock; 40866b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 40876b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 408882b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 408982b0f8c3SJan Kara return handle_userfault(vmf, VM_UFFD_MISSING); 40906b251fc9SAndrea Arcangeli } 4091a13ea5b7SHugh Dickins goto setpte; 4092a13ea5b7SHugh Dickins } 4093a13ea5b7SHugh Dickins 40941da177e4SLinus Torvalds /* Allocate our own private page. */ 40951da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 409665500d23SHugh Dickins goto oom; 40976bc56a4dSMatthew Wilcox (Oracle) folio = vma_alloc_zeroed_movable_folio(vma, vmf->address); 40986bc56a4dSMatthew Wilcox (Oracle) if (!folio) 409965500d23SHugh Dickins goto oom; 4100eb3c24f3SMel Gorman 41016bc56a4dSMatthew Wilcox (Oracle) if (mem_cgroup_charge(folio, vma->vm_mm, GFP_KERNEL)) 4102eb3c24f3SMel Gorman goto oom_free_page; 4103e2bf3e2cSKefeng Wang folio_throttle_swaprate(folio, GFP_KERNEL); 4104eb3c24f3SMel Gorman 410552f37629SMinchan Kim /* 4106cb3184deSMatthew Wilcox (Oracle) * The memory barrier inside __folio_mark_uptodate makes sure that 4107f4f5329dSWei Yang * preceding stores to the page contents become visible before 410852f37629SMinchan Kim * the set_pte_at() write. 410952f37629SMinchan Kim */ 4110cb3184deSMatthew Wilcox (Oracle) __folio_mark_uptodate(folio); 41111da177e4SLinus Torvalds 4112cb3184deSMatthew Wilcox (Oracle) entry = mk_pte(&folio->page, vma->vm_page_prot); 411350c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 41141ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 41151ac0cb5dSHugh Dickins entry = pte_mkwrite(pte_mkdirty(entry)); 41168f4e2101SHugh Dickins 411782b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 411882b0f8c3SJan Kara &vmf->ptl); 41193db82b93SHugh Dickins if (!vmf->pte) 41203db82b93SHugh Dickins goto release; 41212bad466cSPeter Xu if (vmf_pte_changed(vmf)) { 4122bce8cb3cSQi Zheng update_mmu_tlb(vma, vmf->address, vmf->pte); 41238f4e2101SHugh Dickins goto release; 41247df67697SBibo Mao } 41259ba69294SHugh Dickins 41266b31d595SMichal Hocko ret = check_stable_address_space(vma->vm_mm); 41276b31d595SMichal Hocko if (ret) 41286b31d595SMichal Hocko goto release; 41296b31d595SMichal Hocko 41306b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 41316b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 413282b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 4133cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 413482b0f8c3SJan Kara return handle_userfault(vmf, VM_UFFD_MISSING); 41356b251fc9SAndrea Arcangeli } 41366b251fc9SAndrea Arcangeli 4137f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, MM_ANONPAGES); 4138cb3184deSMatthew Wilcox (Oracle) folio_add_new_anon_rmap(folio, vma, vmf->address); 4139cb3184deSMatthew Wilcox (Oracle) folio_add_lru_vma(folio, vma); 4140a13ea5b7SHugh Dickins setpte: 41412bad466cSPeter Xu if (uffd_wp) 41422bad466cSPeter Xu entry = pte_mkuffd_wp(entry); 414382b0f8c3SJan Kara set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); 41441da177e4SLinus Torvalds 41451da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 414682b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 414765500d23SHugh Dickins unlock: 41483db82b93SHugh Dickins if (vmf->pte) 414982b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 41506b31d595SMichal Hocko return ret; 41518f4e2101SHugh Dickins release: 4152cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 41538f4e2101SHugh Dickins goto unlock; 41548a9f3ccdSBalbir Singh oom_free_page: 4155cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 415665500d23SHugh Dickins oom: 41571da177e4SLinus Torvalds return VM_FAULT_OOM; 41581da177e4SLinus Torvalds } 41591da177e4SLinus Torvalds 41609a95f3cfSPaul Cassella /* 4161c1e8d7c6SMichel Lespinasse * The mmap_lock must have been held on entry, and may have been 41629a95f3cfSPaul Cassella * released depending on flags and vma->vm_ops->fault() return value. 41639a95f3cfSPaul Cassella * See filemap_fault() and __lock_page_retry(). 41649a95f3cfSPaul Cassella */ 41652b740303SSouptick Joarder static vm_fault_t __do_fault(struct vm_fault *vmf) 41667eae74afSKirill A. Shutemov { 416782b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 41682b740303SSouptick Joarder vm_fault_t ret; 41697eae74afSKirill A. Shutemov 417063f3655fSMichal Hocko /* 417163f3655fSMichal Hocko * Preallocate pte before we take page_lock because this might lead to 417263f3655fSMichal Hocko * deadlocks for memcg reclaim which waits for pages under writeback: 417363f3655fSMichal Hocko * lock_page(A) 417463f3655fSMichal Hocko * SetPageWriteback(A) 417563f3655fSMichal Hocko * unlock_page(A) 417663f3655fSMichal Hocko * lock_page(B) 417763f3655fSMichal Hocko * lock_page(B) 4178d383807aSYanfei Xu * pte_alloc_one 417963f3655fSMichal Hocko * shrink_page_list 418063f3655fSMichal Hocko * wait_on_page_writeback(A) 418163f3655fSMichal Hocko * SetPageWriteback(B) 418263f3655fSMichal Hocko * unlock_page(B) 418363f3655fSMichal Hocko * # flush A, B to clear the writeback 418463f3655fSMichal Hocko */ 418563f3655fSMichal Hocko if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) { 4186a7069ee3SYanfei Xu vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); 418763f3655fSMichal Hocko if (!vmf->prealloc_pte) 418863f3655fSMichal Hocko return VM_FAULT_OOM; 418963f3655fSMichal Hocko } 419063f3655fSMichal Hocko 419111bac800SDave Jiang ret = vma->vm_ops->fault(vmf); 41923917048dSJan Kara if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | 4193b1aa812bSJan Kara VM_FAULT_DONE_COW))) 41947eae74afSKirill A. Shutemov return ret; 41957eae74afSKirill A. Shutemov 4196667240e0SJan Kara if (unlikely(PageHWPoison(vmf->page))) { 41973149c79fSRik van Riel struct page *page = vmf->page; 4198e53ac737SRik van Riel vm_fault_t poisonret = VM_FAULT_HWPOISON; 4199e53ac737SRik van Riel if (ret & VM_FAULT_LOCKED) { 42003149c79fSRik van Riel if (page_mapped(page)) 42013149c79fSRik van Riel unmap_mapping_pages(page_mapping(page), 42023149c79fSRik van Riel page->index, 1, false); 4203e53ac737SRik van Riel /* Retry if a clean page was removed from the cache. */ 42043149c79fSRik van Riel if (invalidate_inode_page(page)) 42053149c79fSRik van Riel poisonret = VM_FAULT_NOPAGE; 42063149c79fSRik van Riel unlock_page(page); 4207e53ac737SRik van Riel } 42083149c79fSRik van Riel put_page(page); 4209936ca80dSJan Kara vmf->page = NULL; 4210e53ac737SRik van Riel return poisonret; 42117eae74afSKirill A. Shutemov } 42127eae74afSKirill A. Shutemov 42137eae74afSKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) 4214667240e0SJan Kara lock_page(vmf->page); 42157eae74afSKirill A. Shutemov else 4216667240e0SJan Kara VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); 42177eae74afSKirill A. Shutemov 42187eae74afSKirill A. Shutemov return ret; 42197eae74afSKirill A. Shutemov } 42207eae74afSKirill A. Shutemov 4221396bcc52SMatthew Wilcox (Oracle) #ifdef CONFIG_TRANSPARENT_HUGEPAGE 422282b0f8c3SJan Kara static void deposit_prealloc_pte(struct vm_fault *vmf) 4223953c66c2SAneesh Kumar K.V { 422482b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4225953c66c2SAneesh Kumar K.V 422682b0f8c3SJan Kara pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); 4227953c66c2SAneesh Kumar K.V /* 4228953c66c2SAneesh Kumar K.V * We are going to consume the prealloc table, 4229953c66c2SAneesh Kumar K.V * count that as nr_ptes. 4230953c66c2SAneesh Kumar K.V */ 4231c4812909SKirill A. Shutemov mm_inc_nr_ptes(vma->vm_mm); 42327f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 4233953c66c2SAneesh Kumar K.V } 4234953c66c2SAneesh Kumar K.V 4235f9ce0be7SKirill A. Shutemov vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) 423610102459SKirill A. Shutemov { 423782b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 423882b0f8c3SJan Kara bool write = vmf->flags & FAULT_FLAG_WRITE; 423982b0f8c3SJan Kara unsigned long haddr = vmf->address & HPAGE_PMD_MASK; 424010102459SKirill A. Shutemov pmd_t entry; 42412b740303SSouptick Joarder int i; 4242d01ac3c3SMatthew Wilcox (Oracle) vm_fault_t ret = VM_FAULT_FALLBACK; 424310102459SKirill A. Shutemov 424410102459SKirill A. Shutemov if (!transhuge_vma_suitable(vma, haddr)) 4245d01ac3c3SMatthew Wilcox (Oracle) return ret; 424610102459SKirill A. Shutemov 424710102459SKirill A. Shutemov page = compound_head(page); 4248d01ac3c3SMatthew Wilcox (Oracle) if (compound_order(page) != HPAGE_PMD_ORDER) 4249d01ac3c3SMatthew Wilcox (Oracle) return ret; 425010102459SKirill A. Shutemov 4251953c66c2SAneesh Kumar K.V /* 4252eac96c3eSYang Shi * Just backoff if any subpage of a THP is corrupted otherwise 4253eac96c3eSYang Shi * the corrupted page may mapped by PMD silently to escape the 4254eac96c3eSYang Shi * check. This kind of THP just can be PTE mapped. Access to 4255eac96c3eSYang Shi * the corrupted subpage should trigger SIGBUS as expected. 4256eac96c3eSYang Shi */ 4257eac96c3eSYang Shi if (unlikely(PageHasHWPoisoned(page))) 4258eac96c3eSYang Shi return ret; 4259eac96c3eSYang Shi 4260eac96c3eSYang Shi /* 4261f0953a1bSIngo Molnar * Archs like ppc64 need additional space to store information 4262953c66c2SAneesh Kumar K.V * related to pte entry. Use the preallocated table for that. 4263953c66c2SAneesh Kumar K.V */ 426482b0f8c3SJan Kara if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { 42654cf58924SJoel Fernandes (Google) vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); 426682b0f8c3SJan Kara if (!vmf->prealloc_pte) 4267953c66c2SAneesh Kumar K.V return VM_FAULT_OOM; 4268953c66c2SAneesh Kumar K.V } 4269953c66c2SAneesh Kumar K.V 427082b0f8c3SJan Kara vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); 427182b0f8c3SJan Kara if (unlikely(!pmd_none(*vmf->pmd))) 427210102459SKirill A. Shutemov goto out; 427310102459SKirill A. Shutemov 427410102459SKirill A. Shutemov for (i = 0; i < HPAGE_PMD_NR; i++) 427510102459SKirill A. Shutemov flush_icache_page(vma, page + i); 427610102459SKirill A. Shutemov 427710102459SKirill A. Shutemov entry = mk_huge_pmd(page, vma->vm_page_prot); 427810102459SKirill A. Shutemov if (write) 4279f55e1014SLinus Torvalds entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); 428010102459SKirill A. Shutemov 4281fadae295SYang Shi add_mm_counter(vma->vm_mm, mm_counter_file(page), HPAGE_PMD_NR); 4282cea86fe2SHugh Dickins page_add_file_rmap(page, vma, true); 4283cea86fe2SHugh Dickins 4284953c66c2SAneesh Kumar K.V /* 4285953c66c2SAneesh Kumar K.V * deposit and withdraw with pmd lock held 4286953c66c2SAneesh Kumar K.V */ 4287953c66c2SAneesh Kumar K.V if (arch_needs_pgtable_deposit()) 428882b0f8c3SJan Kara deposit_prealloc_pte(vmf); 428910102459SKirill A. Shutemov 429082b0f8c3SJan Kara set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); 429110102459SKirill A. Shutemov 429282b0f8c3SJan Kara update_mmu_cache_pmd(vma, haddr, vmf->pmd); 429310102459SKirill A. Shutemov 429410102459SKirill A. Shutemov /* fault is handled */ 429510102459SKirill A. Shutemov ret = 0; 429695ecedcdSKirill A. Shutemov count_vm_event(THP_FILE_MAPPED); 429710102459SKirill A. Shutemov out: 429882b0f8c3SJan Kara spin_unlock(vmf->ptl); 429910102459SKirill A. Shutemov return ret; 430010102459SKirill A. Shutemov } 430110102459SKirill A. Shutemov #else 4302f9ce0be7SKirill A. Shutemov vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) 430310102459SKirill A. Shutemov { 4304f9ce0be7SKirill A. Shutemov return VM_FAULT_FALLBACK; 430510102459SKirill A. Shutemov } 430610102459SKirill A. Shutemov #endif 430710102459SKirill A. Shutemov 43089d3af4b4SWill Deacon void do_set_pte(struct vm_fault *vmf, struct page *page, unsigned long addr) 43093bb97794SKirill A. Shutemov { 431082b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 43112bad466cSPeter Xu bool uffd_wp = vmf_orig_pte_uffd_wp(vmf); 431282b0f8c3SJan Kara bool write = vmf->flags & FAULT_FLAG_WRITE; 43139d3af4b4SWill Deacon bool prefault = vmf->address != addr; 43143bb97794SKirill A. Shutemov pte_t entry; 43157267ec00SKirill A. Shutemov 43163bb97794SKirill A. Shutemov flush_icache_page(vma, page); 43173bb97794SKirill A. Shutemov entry = mk_pte(page, vma->vm_page_prot); 431846bdb427SWill Deacon 431946bdb427SWill Deacon if (prefault && arch_wants_old_prefaulted_pte()) 432046bdb427SWill Deacon entry = pte_mkold(entry); 432150c25ee9SThomas Bogendoerfer else 432250c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 432346bdb427SWill Deacon 43243bb97794SKirill A. Shutemov if (write) 43253bb97794SKirill A. Shutemov entry = maybe_mkwrite(pte_mkdirty(entry), vma); 43269c28a205SPeter Xu if (unlikely(uffd_wp)) 4327f1eb1bacSPeter Xu entry = pte_mkuffd_wp(entry); 4328bae473a4SKirill A. Shutemov /* copy-on-write page */ 4329bae473a4SKirill A. Shutemov if (write && !(vma->vm_flags & VM_SHARED)) { 4330f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, MM_ANONPAGES); 433140f2bbf7SDavid Hildenbrand page_add_new_anon_rmap(page, vma, addr); 4332b518154eSJoonsoo Kim lru_cache_add_inactive_or_unevictable(page, vma); 43333bb97794SKirill A. Shutemov } else { 4334f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, mm_counter_file(page)); 4335cea86fe2SHugh Dickins page_add_file_rmap(page, vma, false); 43363bb97794SKirill A. Shutemov } 43379d3af4b4SWill Deacon set_pte_at(vma->vm_mm, addr, vmf->pte, entry); 43383bb97794SKirill A. Shutemov } 43393bb97794SKirill A. Shutemov 4340f46f2adeSPeter Xu static bool vmf_pte_changed(struct vm_fault *vmf) 4341f46f2adeSPeter Xu { 4342f46f2adeSPeter Xu if (vmf->flags & FAULT_FLAG_ORIG_PTE_VALID) 4343c33c7948SRyan Roberts return !pte_same(ptep_get(vmf->pte), vmf->orig_pte); 4344f46f2adeSPeter Xu 4345c33c7948SRyan Roberts return !pte_none(ptep_get(vmf->pte)); 4346f46f2adeSPeter Xu } 4347f46f2adeSPeter Xu 43489118c0cbSJan Kara /** 43499118c0cbSJan Kara * finish_fault - finish page fault once we have prepared the page to fault 43509118c0cbSJan Kara * 43519118c0cbSJan Kara * @vmf: structure describing the fault 43529118c0cbSJan Kara * 43539118c0cbSJan Kara * This function handles all that is needed to finish a page fault once the 43549118c0cbSJan Kara * page to fault in is prepared. It handles locking of PTEs, inserts PTE for 43559118c0cbSJan Kara * given page, adds reverse page mapping, handles memcg charges and LRU 4356a862f68aSMike Rapoport * addition. 43579118c0cbSJan Kara * 43589118c0cbSJan Kara * The function expects the page to be locked and on success it consumes a 43599118c0cbSJan Kara * reference of a page being mapped (for the PTE which maps it). 4360a862f68aSMike Rapoport * 4361a862f68aSMike Rapoport * Return: %0 on success, %VM_FAULT_ code in case of error. 43629118c0cbSJan Kara */ 43632b740303SSouptick Joarder vm_fault_t finish_fault(struct vm_fault *vmf) 43649118c0cbSJan Kara { 4365f9ce0be7SKirill A. Shutemov struct vm_area_struct *vma = vmf->vma; 43669118c0cbSJan Kara struct page *page; 4367f9ce0be7SKirill A. Shutemov vm_fault_t ret; 43689118c0cbSJan Kara 43699118c0cbSJan Kara /* Did we COW the page? */ 4370f9ce0be7SKirill A. Shutemov if ((vmf->flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) 43719118c0cbSJan Kara page = vmf->cow_page; 43729118c0cbSJan Kara else 43739118c0cbSJan Kara page = vmf->page; 43746b31d595SMichal Hocko 43756b31d595SMichal Hocko /* 43766b31d595SMichal Hocko * check even for read faults because we might have lost our CoWed 43776b31d595SMichal Hocko * page 43786b31d595SMichal Hocko */ 4379f9ce0be7SKirill A. Shutemov if (!(vma->vm_flags & VM_SHARED)) { 4380f9ce0be7SKirill A. Shutemov ret = check_stable_address_space(vma->vm_mm); 4381f9ce0be7SKirill A. Shutemov if (ret) 4382f9ce0be7SKirill A. Shutemov return ret; 4383f9ce0be7SKirill A. Shutemov } 4384f9ce0be7SKirill A. Shutemov 4385f9ce0be7SKirill A. Shutemov if (pmd_none(*vmf->pmd)) { 4386f9ce0be7SKirill A. Shutemov if (PageTransCompound(page)) { 4387f9ce0be7SKirill A. Shutemov ret = do_set_pmd(vmf, page); 4388f9ce0be7SKirill A. Shutemov if (ret != VM_FAULT_FALLBACK) 4389f9ce0be7SKirill A. Shutemov return ret; 4390f9ce0be7SKirill A. Shutemov } 4391f9ce0be7SKirill A. Shutemov 439203c4f204SQi Zheng if (vmf->prealloc_pte) 439303c4f204SQi Zheng pmd_install(vma->vm_mm, vmf->pmd, &vmf->prealloc_pte); 439403c4f204SQi Zheng else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd))) 4395f9ce0be7SKirill A. Shutemov return VM_FAULT_OOM; 4396f9ce0be7SKirill A. Shutemov } 4397f9ce0be7SKirill A. Shutemov 4398f9ce0be7SKirill A. Shutemov vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 4399f9ce0be7SKirill A. Shutemov vmf->address, &vmf->ptl); 44003db82b93SHugh Dickins if (!vmf->pte) 44013db82b93SHugh Dickins return VM_FAULT_NOPAGE; 4402f9ce0be7SKirill A. Shutemov 440370427f6eSSergei Antonov /* Re-check under ptl */ 440470427f6eSSergei Antonov if (likely(!vmf_pte_changed(vmf))) { 440570427f6eSSergei Antonov do_set_pte(vmf, page, vmf->address); 440670427f6eSSergei Antonov 440770427f6eSSergei Antonov /* no need to invalidate: a not-present page won't be cached */ 440870427f6eSSergei Antonov update_mmu_cache(vma, vmf->address, vmf->pte); 440970427f6eSSergei Antonov 441070427f6eSSergei Antonov ret = 0; 441170427f6eSSergei Antonov } else { 4412f9ce0be7SKirill A. Shutemov update_mmu_tlb(vma, vmf->address, vmf->pte); 441370427f6eSSergei Antonov ret = VM_FAULT_NOPAGE; 441470427f6eSSergei Antonov } 441570427f6eSSergei Antonov 44169118c0cbSJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 44179118c0cbSJan Kara return ret; 44189118c0cbSJan Kara } 44199118c0cbSJan Kara 442053d36a56SLorenzo Stoakes static unsigned long fault_around_pages __read_mostly = 442153d36a56SLorenzo Stoakes 65536 >> PAGE_SHIFT; 4422a9b0f861SKirill A. Shutemov 44231592eef0SKirill A. Shutemov #ifdef CONFIG_DEBUG_FS 4424a9b0f861SKirill A. Shutemov static int fault_around_bytes_get(void *data, u64 *val) 44251592eef0SKirill A. Shutemov { 442653d36a56SLorenzo Stoakes *val = fault_around_pages << PAGE_SHIFT; 44271592eef0SKirill A. Shutemov return 0; 44281592eef0SKirill A. Shutemov } 44291592eef0SKirill A. Shutemov 4430b4903d6eSAndrey Ryabinin /* 4431da391d64SWilliam Kucharski * fault_around_bytes must be rounded down to the nearest page order as it's 4432da391d64SWilliam Kucharski * what do_fault_around() expects to see. 4433b4903d6eSAndrey Ryabinin */ 4434a9b0f861SKirill A. Shutemov static int fault_around_bytes_set(void *data, u64 val) 44351592eef0SKirill A. Shutemov { 4436a9b0f861SKirill A. Shutemov if (val / PAGE_SIZE > PTRS_PER_PTE) 44371592eef0SKirill A. Shutemov return -EINVAL; 443853d36a56SLorenzo Stoakes 443953d36a56SLorenzo Stoakes /* 444053d36a56SLorenzo Stoakes * The minimum value is 1 page, however this results in no fault-around 444153d36a56SLorenzo Stoakes * at all. See should_fault_around(). 444253d36a56SLorenzo Stoakes */ 444353d36a56SLorenzo Stoakes fault_around_pages = max(rounddown_pow_of_two(val) >> PAGE_SHIFT, 1UL); 444453d36a56SLorenzo Stoakes 44451592eef0SKirill A. Shutemov return 0; 44461592eef0SKirill A. Shutemov } 44470a1345f8SYevgen Pronenko DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, 4448a9b0f861SKirill A. Shutemov fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); 44491592eef0SKirill A. Shutemov 44501592eef0SKirill A. Shutemov static int __init fault_around_debugfs(void) 44511592eef0SKirill A. Shutemov { 4452d9f7979cSGreg Kroah-Hartman debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, 4453a9b0f861SKirill A. Shutemov &fault_around_bytes_fops); 44541592eef0SKirill A. Shutemov return 0; 44551592eef0SKirill A. Shutemov } 44561592eef0SKirill A. Shutemov late_initcall(fault_around_debugfs); 44571592eef0SKirill A. Shutemov #endif 44588c6e50b0SKirill A. Shutemov 44591fdb412bSKirill A. Shutemov /* 44601fdb412bSKirill A. Shutemov * do_fault_around() tries to map few pages around the fault address. The hope 44611fdb412bSKirill A. Shutemov * is that the pages will be needed soon and this will lower the number of 44621fdb412bSKirill A. Shutemov * faults to handle. 44631fdb412bSKirill A. Shutemov * 44641fdb412bSKirill A. Shutemov * It uses vm_ops->map_pages() to map the pages, which skips the page if it's 44651fdb412bSKirill A. Shutemov * not ready to be mapped: not up-to-date, locked, etc. 44661fdb412bSKirill A. Shutemov * 44679042599eSLorenzo Stoakes * This function doesn't cross VMA or page table boundaries, in order to call 44689042599eSLorenzo Stoakes * map_pages() and acquire a PTE lock only once. 44691fdb412bSKirill A. Shutemov * 447053d36a56SLorenzo Stoakes * fault_around_pages defines how many pages we'll try to map. 4471da391d64SWilliam Kucharski * do_fault_around() expects it to be set to a power of two less than or equal 4472da391d64SWilliam Kucharski * to PTRS_PER_PTE. 44731fdb412bSKirill A. Shutemov * 4474da391d64SWilliam Kucharski * The virtual address of the area that we map is naturally aligned to 447553d36a56SLorenzo Stoakes * fault_around_pages * PAGE_SIZE rounded down to the machine page size 4476da391d64SWilliam Kucharski * (and therefore to page order). This way it's easier to guarantee 4477da391d64SWilliam Kucharski * that we don't cross page table boundaries. 44781fdb412bSKirill A. Shutemov */ 44792b740303SSouptick Joarder static vm_fault_t do_fault_around(struct vm_fault *vmf) 44808c6e50b0SKirill A. Shutemov { 448153d36a56SLorenzo Stoakes pgoff_t nr_pages = READ_ONCE(fault_around_pages); 44829042599eSLorenzo Stoakes pgoff_t pte_off = pte_index(vmf->address); 44839042599eSLorenzo Stoakes /* The page offset of vmf->address within the VMA. */ 44849042599eSLorenzo Stoakes pgoff_t vma_off = vmf->pgoff - vmf->vma->vm_pgoff; 44859042599eSLorenzo Stoakes pgoff_t from_pte, to_pte; 448658ef47efSMatthew Wilcox (Oracle) vm_fault_t ret; 44878c6e50b0SKirill A. Shutemov 44889042599eSLorenzo Stoakes /* The PTE offset of the start address, clamped to the VMA. */ 44899042599eSLorenzo Stoakes from_pte = max(ALIGN_DOWN(pte_off, nr_pages), 44909042599eSLorenzo Stoakes pte_off - min(pte_off, vma_off)); 4491aecd6f44SKirill A. Shutemov 44929042599eSLorenzo Stoakes /* The PTE offset of the end address, clamped to the VMA and PTE. */ 44939042599eSLorenzo Stoakes to_pte = min3(from_pte + nr_pages, (pgoff_t)PTRS_PER_PTE, 44949042599eSLorenzo Stoakes pte_off + vma_pages(vmf->vma) - vma_off) - 1; 44958c6e50b0SKirill A. Shutemov 449682b0f8c3SJan Kara if (pmd_none(*vmf->pmd)) { 44974cf58924SJoel Fernandes (Google) vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm); 449882b0f8c3SJan Kara if (!vmf->prealloc_pte) 4499f9ce0be7SKirill A. Shutemov return VM_FAULT_OOM; 45008c6e50b0SKirill A. Shutemov } 45018c6e50b0SKirill A. Shutemov 450258ef47efSMatthew Wilcox (Oracle) rcu_read_lock(); 450358ef47efSMatthew Wilcox (Oracle) ret = vmf->vma->vm_ops->map_pages(vmf, 45049042599eSLorenzo Stoakes vmf->pgoff + from_pte - pte_off, 45059042599eSLorenzo Stoakes vmf->pgoff + to_pte - pte_off); 450658ef47efSMatthew Wilcox (Oracle) rcu_read_unlock(); 450758ef47efSMatthew Wilcox (Oracle) 450858ef47efSMatthew Wilcox (Oracle) return ret; 45097267ec00SKirill A. Shutemov } 45107267ec00SKirill A. Shutemov 45119c28a205SPeter Xu /* Return true if we should do read fault-around, false otherwise */ 45129c28a205SPeter Xu static inline bool should_fault_around(struct vm_fault *vmf) 45139c28a205SPeter Xu { 45149c28a205SPeter Xu /* No ->map_pages? No way to fault around... */ 45159c28a205SPeter Xu if (!vmf->vma->vm_ops->map_pages) 45169c28a205SPeter Xu return false; 45179c28a205SPeter Xu 45189c28a205SPeter Xu if (uffd_disable_fault_around(vmf->vma)) 45199c28a205SPeter Xu return false; 45209c28a205SPeter Xu 452153d36a56SLorenzo Stoakes /* A single page implies no faulting 'around' at all. */ 452253d36a56SLorenzo Stoakes return fault_around_pages > 1; 45239c28a205SPeter Xu } 45249c28a205SPeter Xu 45252b740303SSouptick Joarder static vm_fault_t do_read_fault(struct vm_fault *vmf) 4526e655fb29SKirill A. Shutemov { 45272b740303SSouptick Joarder vm_fault_t ret = 0; 45288c6e50b0SKirill A. Shutemov 45298c6e50b0SKirill A. Shutemov /* 45308c6e50b0SKirill A. Shutemov * Let's call ->map_pages() first and use ->fault() as fallback 45318c6e50b0SKirill A. Shutemov * if page by the offset is not ready to be mapped (cold cache or 45328c6e50b0SKirill A. Shutemov * something). 45338c6e50b0SKirill A. Shutemov */ 45349c28a205SPeter Xu if (should_fault_around(vmf)) { 45350721ec8bSJan Kara ret = do_fault_around(vmf); 45367267ec00SKirill A. Shutemov if (ret) 45377267ec00SKirill A. Shutemov return ret; 45388c6e50b0SKirill A. Shutemov } 4539e655fb29SKirill A. Shutemov 4540936ca80dSJan Kara ret = __do_fault(vmf); 4541e655fb29SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4542e655fb29SKirill A. Shutemov return ret; 4543e655fb29SKirill A. Shutemov 45449118c0cbSJan Kara ret |= finish_fault(vmf); 4545936ca80dSJan Kara unlock_page(vmf->page); 45467267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4547936ca80dSJan Kara put_page(vmf->page); 4548e655fb29SKirill A. Shutemov return ret; 4549e655fb29SKirill A. Shutemov } 4550e655fb29SKirill A. Shutemov 45512b740303SSouptick Joarder static vm_fault_t do_cow_fault(struct vm_fault *vmf) 4552ec47c3b9SKirill A. Shutemov { 455382b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 45542b740303SSouptick Joarder vm_fault_t ret; 4555ec47c3b9SKirill A. Shutemov 4556ec47c3b9SKirill A. Shutemov if (unlikely(anon_vma_prepare(vma))) 4557ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4558ec47c3b9SKirill A. Shutemov 4559936ca80dSJan Kara vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); 4560936ca80dSJan Kara if (!vmf->cow_page) 4561ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4562ec47c3b9SKirill A. Shutemov 45638f425e4eSMatthew Wilcox (Oracle) if (mem_cgroup_charge(page_folio(vmf->cow_page), vma->vm_mm, 45648f425e4eSMatthew Wilcox (Oracle) GFP_KERNEL)) { 4565936ca80dSJan Kara put_page(vmf->cow_page); 4566ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4567ec47c3b9SKirill A. Shutemov } 456868fa572bSKefeng Wang folio_throttle_swaprate(page_folio(vmf->cow_page), GFP_KERNEL); 4569ec47c3b9SKirill A. Shutemov 4570936ca80dSJan Kara ret = __do_fault(vmf); 4571ec47c3b9SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4572ec47c3b9SKirill A. Shutemov goto uncharge_out; 45733917048dSJan Kara if (ret & VM_FAULT_DONE_COW) 45743917048dSJan Kara return ret; 4575ec47c3b9SKirill A. Shutemov 4576936ca80dSJan Kara copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); 4577936ca80dSJan Kara __SetPageUptodate(vmf->cow_page); 4578ec47c3b9SKirill A. Shutemov 45799118c0cbSJan Kara ret |= finish_fault(vmf); 4580936ca80dSJan Kara unlock_page(vmf->page); 4581936ca80dSJan Kara put_page(vmf->page); 45827267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 45837267ec00SKirill A. Shutemov goto uncharge_out; 4584ec47c3b9SKirill A. Shutemov return ret; 4585ec47c3b9SKirill A. Shutemov uncharge_out: 4586936ca80dSJan Kara put_page(vmf->cow_page); 4587ec47c3b9SKirill A. Shutemov return ret; 4588ec47c3b9SKirill A. Shutemov } 4589ec47c3b9SKirill A. Shutemov 45902b740303SSouptick Joarder static vm_fault_t do_shared_fault(struct vm_fault *vmf) 45911da177e4SLinus Torvalds { 459282b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 45932b740303SSouptick Joarder vm_fault_t ret, tmp; 45941d65f86dSKAMEZAWA Hiroyuki 4595936ca80dSJan Kara ret = __do_fault(vmf); 45967eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4597f0c6d4d2SKirill A. Shutemov return ret; 45981da177e4SLinus Torvalds 45991da177e4SLinus Torvalds /* 4600f0c6d4d2SKirill A. Shutemov * Check if the backing address space wants to know that the page is 4601f0c6d4d2SKirill A. Shutemov * about to become writable 46021da177e4SLinus Torvalds */ 4603fb09a464SKirill A. Shutemov if (vma->vm_ops->page_mkwrite) { 4604936ca80dSJan Kara unlock_page(vmf->page); 460538b8cb7fSJan Kara tmp = do_page_mkwrite(vmf); 4606fb09a464SKirill A. Shutemov if (unlikely(!tmp || 4607fb09a464SKirill A. Shutemov (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 4608936ca80dSJan Kara put_page(vmf->page); 4609f0c6d4d2SKirill A. Shutemov return tmp; 461069676147SMark Fasheh } 4611d0217ac0SNick Piggin } 4612fb09a464SKirill A. Shutemov 46139118c0cbSJan Kara ret |= finish_fault(vmf); 46147267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | 46157267ec00SKirill A. Shutemov VM_FAULT_RETRY))) { 4616936ca80dSJan Kara unlock_page(vmf->page); 4617936ca80dSJan Kara put_page(vmf->page); 4618f0c6d4d2SKirill A. Shutemov return ret; 46199637a5efSDavid Howells } 4620d00806b1SNick Piggin 462189b15332SJohannes Weiner ret |= fault_dirty_shared_page(vmf); 4622b827e496SNick Piggin return ret; 462354cb8821SNick Piggin } 4624d00806b1SNick Piggin 46259a95f3cfSPaul Cassella /* 4626c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 46279a95f3cfSPaul Cassella * but allow concurrent faults). 4628c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 46299138e47eSMatthew Wilcox (Oracle) * return value. See filemap_fault() and __folio_lock_or_retry(). 4630c1e8d7c6SMichel Lespinasse * If mmap_lock is released, vma may become invalid (for example 4631fc8efd2dSJan Stancek * by other thread calling munmap()). 46329a95f3cfSPaul Cassella */ 46332b740303SSouptick Joarder static vm_fault_t do_fault(struct vm_fault *vmf) 463454cb8821SNick Piggin { 463582b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4636fc8efd2dSJan Stancek struct mm_struct *vm_mm = vma->vm_mm; 46372b740303SSouptick Joarder vm_fault_t ret; 463854cb8821SNick Piggin 4639ff09d7ecSAneesh Kumar K.V /* 4640ff09d7ecSAneesh Kumar K.V * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND 4641ff09d7ecSAneesh Kumar K.V */ 4642ff09d7ecSAneesh Kumar K.V if (!vma->vm_ops->fault) { 46433db82b93SHugh Dickins vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, 46443db82b93SHugh Dickins vmf->address, &vmf->ptl); 46453db82b93SHugh Dickins if (unlikely(!vmf->pte)) 4646b0b9b3dfSHugh Dickins ret = VM_FAULT_SIGBUS; 4647ff09d7ecSAneesh Kumar K.V else { 4648ff09d7ecSAneesh Kumar K.V /* 4649ff09d7ecSAneesh Kumar K.V * Make sure this is not a temporary clearing of pte 4650ff09d7ecSAneesh Kumar K.V * by holding ptl and checking again. A R/M/W update 4651ff09d7ecSAneesh Kumar K.V * of pte involves: take ptl, clearing the pte so that 4652ff09d7ecSAneesh Kumar K.V * we don't have concurrent modification by hardware 4653ff09d7ecSAneesh Kumar K.V * followed by an update. 4654ff09d7ecSAneesh Kumar K.V */ 4655c33c7948SRyan Roberts if (unlikely(pte_none(ptep_get(vmf->pte)))) 4656ff09d7ecSAneesh Kumar K.V ret = VM_FAULT_SIGBUS; 4657ff09d7ecSAneesh Kumar K.V else 4658ff09d7ecSAneesh Kumar K.V ret = VM_FAULT_NOPAGE; 4659ff09d7ecSAneesh Kumar K.V 4660ff09d7ecSAneesh Kumar K.V pte_unmap_unlock(vmf->pte, vmf->ptl); 4661ff09d7ecSAneesh Kumar K.V } 4662ff09d7ecSAneesh Kumar K.V } else if (!(vmf->flags & FAULT_FLAG_WRITE)) 4663b0b9b3dfSHugh Dickins ret = do_read_fault(vmf); 4664b0b9b3dfSHugh Dickins else if (!(vma->vm_flags & VM_SHARED)) 4665b0b9b3dfSHugh Dickins ret = do_cow_fault(vmf); 4666b0b9b3dfSHugh Dickins else 4667b0b9b3dfSHugh Dickins ret = do_shared_fault(vmf); 4668b0b9b3dfSHugh Dickins 4669b0b9b3dfSHugh Dickins /* preallocated pagetable is unused: free it */ 4670b0b9b3dfSHugh Dickins if (vmf->prealloc_pte) { 4671fc8efd2dSJan Stancek pte_free(vm_mm, vmf->prealloc_pte); 46727f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 4673b0b9b3dfSHugh Dickins } 4674b0b9b3dfSHugh Dickins return ret; 467554cb8821SNick Piggin } 467654cb8821SNick Piggin 4677f4c0d836SYang Shi int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, 4678f4c0d836SYang Shi unsigned long addr, int page_nid, int *flags) 46799532fec1SMel Gorman { 46809532fec1SMel Gorman get_page(page); 46819532fec1SMel Gorman 4682fc137c0dSRaghavendra K T /* Record the current PID acceesing VMA */ 4683fc137c0dSRaghavendra K T vma_set_access_pid_bit(vma); 4684fc137c0dSRaghavendra K T 46859532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS); 468604bb2f94SRik van Riel if (page_nid == numa_node_id()) { 46879532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); 468804bb2f94SRik van Riel *flags |= TNF_FAULT_LOCAL; 468904bb2f94SRik van Riel } 46909532fec1SMel Gorman 46919532fec1SMel Gorman return mpol_misplaced(page, vma, addr); 46929532fec1SMel Gorman } 46939532fec1SMel Gorman 46942b740303SSouptick Joarder static vm_fault_t do_numa_page(struct vm_fault *vmf) 4695d10e63f2SMel Gorman { 469682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 46974daae3b4SMel Gorman struct page *page = NULL; 469898fa15f3SAnshuman Khandual int page_nid = NUMA_NO_NODE; 46996a56ccbcSDavid Hildenbrand bool writable = false; 470090572890SPeter Zijlstra int last_cpupid; 4701cbee9f88SPeter Zijlstra int target_nid; 470204a86453SAneesh Kumar K.V pte_t pte, old_pte; 47036688cc05SPeter Zijlstra int flags = 0; 4704d10e63f2SMel Gorman 4705d10e63f2SMel Gorman /* 4706d10e63f2SMel Gorman * The "pte" at this point cannot be used safely without 4707d10e63f2SMel Gorman * validation through pte_unmap_same(). It's of NUMA type but 4708d10e63f2SMel Gorman * the pfn may be screwed if the read is non atomic. 4709d10e63f2SMel Gorman */ 471082b0f8c3SJan Kara spin_lock(vmf->ptl); 4711c33c7948SRyan Roberts if (unlikely(!pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 471282b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 47134daae3b4SMel Gorman goto out; 47144daae3b4SMel Gorman } 47154daae3b4SMel Gorman 4716b99a342dSHuang Ying /* Get the normal PTE */ 4717b99a342dSHuang Ying old_pte = ptep_get(vmf->pte); 471804a86453SAneesh Kumar K.V pte = pte_modify(old_pte, vma->vm_page_prot); 4719d10e63f2SMel Gorman 47206a56ccbcSDavid Hildenbrand /* 47216a56ccbcSDavid Hildenbrand * Detect now whether the PTE could be writable; this information 47226a56ccbcSDavid Hildenbrand * is only valid while holding the PT lock. 47236a56ccbcSDavid Hildenbrand */ 47246a56ccbcSDavid Hildenbrand writable = pte_write(pte); 47256a56ccbcSDavid Hildenbrand if (!writable && vma_wants_manual_pte_write_upgrade(vma) && 47266a56ccbcSDavid Hildenbrand can_change_pte_writable(vma, vmf->address, pte)) 47276a56ccbcSDavid Hildenbrand writable = true; 47286a56ccbcSDavid Hildenbrand 472982b0f8c3SJan Kara page = vm_normal_page(vma, vmf->address, pte); 47303218f871SAlex Sierra if (!page || is_zone_device_page(page)) 4731b99a342dSHuang Ying goto out_map; 4732d10e63f2SMel Gorman 4733e81c4802SKirill A. Shutemov /* TODO: handle PTE-mapped THP */ 4734b99a342dSHuang Ying if (PageCompound(page)) 4735b99a342dSHuang Ying goto out_map; 4736e81c4802SKirill A. Shutemov 47376688cc05SPeter Zijlstra /* 4738bea66fbdSMel Gorman * Avoid grouping on RO pages in general. RO pages shouldn't hurt as 4739bea66fbdSMel Gorman * much anyway since they can be in shared cache state. This misses 4740bea66fbdSMel Gorman * the case where a mapping is writable but the process never writes 4741bea66fbdSMel Gorman * to it but pte_write gets cleared during protection updates and 4742bea66fbdSMel Gorman * pte_dirty has unpredictable behaviour between PTE scan updates, 4743bea66fbdSMel Gorman * background writeback, dirty balancing and application behaviour. 47446688cc05SPeter Zijlstra */ 47456a56ccbcSDavid Hildenbrand if (!writable) 47466688cc05SPeter Zijlstra flags |= TNF_NO_GROUP; 47476688cc05SPeter Zijlstra 4748dabe1d99SRik van Riel /* 4749dabe1d99SRik van Riel * Flag if the page is shared between multiple address spaces. This 4750dabe1d99SRik van Riel * is later used when determining whether to group tasks together 4751dabe1d99SRik van Riel */ 4752dabe1d99SRik van Riel if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) 4753dabe1d99SRik van Riel flags |= TNF_SHARED; 4754dabe1d99SRik van Riel 47558191acbdSMel Gorman page_nid = page_to_nid(page); 475633024536SHuang Ying /* 475733024536SHuang Ying * For memory tiering mode, cpupid of slow memory page is used 475833024536SHuang Ying * to record page access time. So use default value. 475933024536SHuang Ying */ 476033024536SHuang Ying if ((sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) && 476133024536SHuang Ying !node_is_toptier(page_nid)) 476233024536SHuang Ying last_cpupid = (-1 & LAST_CPUPID_MASK); 476333024536SHuang Ying else 476433024536SHuang Ying last_cpupid = page_cpupid_last(page); 476582b0f8c3SJan Kara target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, 4766bae473a4SKirill A. Shutemov &flags); 476798fa15f3SAnshuman Khandual if (target_nid == NUMA_NO_NODE) { 47684daae3b4SMel Gorman put_page(page); 4769b99a342dSHuang Ying goto out_map; 47704daae3b4SMel Gorman } 4771b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 47726a56ccbcSDavid Hildenbrand writable = false; 47734daae3b4SMel Gorman 47744daae3b4SMel Gorman /* Migrate to the requested node */ 4775bf90ac19SWang Qing if (migrate_misplaced_page(page, vma, target_nid)) { 47768191acbdSMel Gorman page_nid = target_nid; 47776688cc05SPeter Zijlstra flags |= TNF_MIGRATED; 4778b99a342dSHuang Ying } else { 4779074c2381SMel Gorman flags |= TNF_MIGRATE_FAIL; 4780c7ad0880SHugh Dickins vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 4781c7ad0880SHugh Dickins vmf->address, &vmf->ptl); 4782c7ad0880SHugh Dickins if (unlikely(!vmf->pte)) 4783c7ad0880SHugh Dickins goto out; 4784c33c7948SRyan Roberts if (unlikely(!pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 4785b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 4786b99a342dSHuang Ying goto out; 4787b99a342dSHuang Ying } 4788b99a342dSHuang Ying goto out_map; 4789b99a342dSHuang Ying } 47904daae3b4SMel Gorman 47914daae3b4SMel Gorman out: 479298fa15f3SAnshuman Khandual if (page_nid != NUMA_NO_NODE) 47936688cc05SPeter Zijlstra task_numa_fault(last_cpupid, page_nid, 1, flags); 4794d10e63f2SMel Gorman return 0; 4795b99a342dSHuang Ying out_map: 4796b99a342dSHuang Ying /* 4797b99a342dSHuang Ying * Make it present again, depending on how arch implements 4798b99a342dSHuang Ying * non-accessible ptes, some can allow access by kernel mode. 4799b99a342dSHuang Ying */ 4800b99a342dSHuang Ying old_pte = ptep_modify_prot_start(vma, vmf->address, vmf->pte); 4801b99a342dSHuang Ying pte = pte_modify(old_pte, vma->vm_page_prot); 4802b99a342dSHuang Ying pte = pte_mkyoung(pte); 48036a56ccbcSDavid Hildenbrand if (writable) 4804b99a342dSHuang Ying pte = pte_mkwrite(pte); 4805b99a342dSHuang Ying ptep_modify_prot_commit(vma, vmf->address, vmf->pte, old_pte, pte); 4806b99a342dSHuang Ying update_mmu_cache(vma, vmf->address, vmf->pte); 4807b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 4808b99a342dSHuang Ying goto out; 4809d10e63f2SMel Gorman } 4810d10e63f2SMel Gorman 48112b740303SSouptick Joarder static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf) 4812b96375f7SMatthew Wilcox { 4813f4200391SDave Jiang if (vma_is_anonymous(vmf->vma)) 481482b0f8c3SJan Kara return do_huge_pmd_anonymous_page(vmf); 4815a2d58167SDave Jiang if (vmf->vma->vm_ops->huge_fault) 4816c791ace1SDave Jiang return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); 4817b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 4818b96375f7SMatthew Wilcox } 4819b96375f7SMatthew Wilcox 4820183f24aaSGeert Uytterhoeven /* `inline' is required to avoid gcc 4.1.2 build error */ 48215db4f15cSYang Shi static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf) 4822b96375f7SMatthew Wilcox { 4823c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 4824aea06577SDavid Hildenbrand vm_fault_t ret; 4825c89357e2SDavid Hildenbrand 4826529b930bSAndrea Arcangeli if (vma_is_anonymous(vmf->vma)) { 4827c89357e2SDavid Hildenbrand if (likely(!unshare) && 4828c89357e2SDavid Hildenbrand userfaultfd_huge_pmd_wp(vmf->vma, vmf->orig_pmd)) 4829529b930bSAndrea Arcangeli return handle_userfault(vmf, VM_UFFD_WP); 48305db4f15cSYang Shi return do_huge_pmd_wp_page(vmf); 4831529b930bSAndrea Arcangeli } 4832af9e4d5fSKirill A. Shutemov 4833aea06577SDavid Hildenbrand if (vmf->vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 4834aea06577SDavid Hildenbrand if (vmf->vma->vm_ops->huge_fault) { 4835aea06577SDavid Hildenbrand ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); 4836327e9fd4SThomas Hellstrom (VMware) if (!(ret & VM_FAULT_FALLBACK)) 4837327e9fd4SThomas Hellstrom (VMware) return ret; 4838327e9fd4SThomas Hellstrom (VMware) } 4839aea06577SDavid Hildenbrand } 4840327e9fd4SThomas Hellstrom (VMware) 4841327e9fd4SThomas Hellstrom (VMware) /* COW or write-notify handled on pte level: split pmd. */ 484282b0f8c3SJan Kara __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); 4843af9e4d5fSKirill A. Shutemov 4844b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 4845b96375f7SMatthew Wilcox } 4846b96375f7SMatthew Wilcox 48472b740303SSouptick Joarder static vm_fault_t create_huge_pud(struct vm_fault *vmf) 4848a00cc7d9SMatthew Wilcox { 4849327e9fd4SThomas Hellstrom (VMware) #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ 4850327e9fd4SThomas Hellstrom (VMware) defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) 4851a00cc7d9SMatthew Wilcox /* No support for anonymous transparent PUD pages yet */ 4852a00cc7d9SMatthew Wilcox if (vma_is_anonymous(vmf->vma)) 485314c99d65SGowans, James return VM_FAULT_FALLBACK; 485414c99d65SGowans, James if (vmf->vma->vm_ops->huge_fault) 485514c99d65SGowans, James return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); 485614c99d65SGowans, James #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 485714c99d65SGowans, James return VM_FAULT_FALLBACK; 485814c99d65SGowans, James } 485914c99d65SGowans, James 486014c99d65SGowans, James static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) 486114c99d65SGowans, James { 486214c99d65SGowans, James #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ 486314c99d65SGowans, James defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) 4864aea06577SDavid Hildenbrand vm_fault_t ret; 4865aea06577SDavid Hildenbrand 486614c99d65SGowans, James /* No support for anonymous transparent PUD pages yet */ 486714c99d65SGowans, James if (vma_is_anonymous(vmf->vma)) 4868327e9fd4SThomas Hellstrom (VMware) goto split; 4869aea06577SDavid Hildenbrand if (vmf->vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 4870327e9fd4SThomas Hellstrom (VMware) if (vmf->vma->vm_ops->huge_fault) { 4871aea06577SDavid Hildenbrand ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); 4872327e9fd4SThomas Hellstrom (VMware) if (!(ret & VM_FAULT_FALLBACK)) 4873327e9fd4SThomas Hellstrom (VMware) return ret; 4874327e9fd4SThomas Hellstrom (VMware) } 4875aea06577SDavid Hildenbrand } 4876327e9fd4SThomas Hellstrom (VMware) split: 4877327e9fd4SThomas Hellstrom (VMware) /* COW or write-notify not handled on PUD level: split pud.*/ 4878327e9fd4SThomas Hellstrom (VMware) __split_huge_pud(vmf->vma, vmf->pud, vmf->address); 487914c99d65SGowans, James #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 4880a00cc7d9SMatthew Wilcox return VM_FAULT_FALLBACK; 4881a00cc7d9SMatthew Wilcox } 4882a00cc7d9SMatthew Wilcox 48831da177e4SLinus Torvalds /* 48841da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 48851da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 48861da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 48871da177e4SLinus Torvalds * 48881da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 48891da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 48901da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 48911da177e4SLinus Torvalds * 4892c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, but allow 48937267ec00SKirill A. Shutemov * concurrent faults). 48949a95f3cfSPaul Cassella * 4895c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our return value. 48969138e47eSMatthew Wilcox (Oracle) * See filemap_fault() and __folio_lock_or_retry(). 48971da177e4SLinus Torvalds */ 48982b740303SSouptick Joarder static vm_fault_t handle_pte_fault(struct vm_fault *vmf) 48991da177e4SLinus Torvalds { 49001da177e4SLinus Torvalds pte_t entry; 49011da177e4SLinus Torvalds 490282b0f8c3SJan Kara if (unlikely(pmd_none(*vmf->pmd))) { 49037267ec00SKirill A. Shutemov /* 49047267ec00SKirill A. Shutemov * Leave __pte_alloc() until later: because vm_ops->fault may 49057267ec00SKirill A. Shutemov * want to allocate huge page, and if we expose page table 49067267ec00SKirill A. Shutemov * for an instant, it will be difficult to retract from 49077267ec00SKirill A. Shutemov * concurrent faults and from rmap lookups. 49087267ec00SKirill A. Shutemov */ 490982b0f8c3SJan Kara vmf->pte = NULL; 4910f46f2adeSPeter Xu vmf->flags &= ~FAULT_FLAG_ORIG_PTE_VALID; 49117267ec00SKirill A. Shutemov } else { 4912f9ce0be7SKirill A. Shutemov /* 49137267ec00SKirill A. Shutemov * A regular pmd is established and it can't morph into a huge 4914c7ad0880SHugh Dickins * pmd by anon khugepaged, since that takes mmap_lock in write 4915c7ad0880SHugh Dickins * mode; but shmem or file collapse to THP could still morph 4916c7ad0880SHugh Dickins * it into a huge pmd: just retry later if so. 49177267ec00SKirill A. Shutemov */ 4918c7ad0880SHugh Dickins vmf->pte = pte_offset_map_nolock(vmf->vma->vm_mm, vmf->pmd, 4919c7ad0880SHugh Dickins vmf->address, &vmf->ptl); 4920c7ad0880SHugh Dickins if (unlikely(!vmf->pte)) 4921c7ad0880SHugh Dickins return 0; 492226e1a0c3SHugh Dickins vmf->orig_pte = ptep_get_lockless(vmf->pte); 4923f46f2adeSPeter Xu vmf->flags |= FAULT_FLAG_ORIG_PTE_VALID; 49247267ec00SKirill A. Shutemov 49252994302bSJan Kara if (pte_none(vmf->orig_pte)) { 492682b0f8c3SJan Kara pte_unmap(vmf->pte); 492782b0f8c3SJan Kara vmf->pte = NULL; 49287267ec00SKirill A. Shutemov } 49297267ec00SKirill A. Shutemov } 49307267ec00SKirill A. Shutemov 49312bad466cSPeter Xu if (!vmf->pte) 49322bad466cSPeter Xu return do_pte_missing(vmf); 49337267ec00SKirill A. Shutemov 49342994302bSJan Kara if (!pte_present(vmf->orig_pte)) 49352994302bSJan Kara return do_swap_page(vmf); 49361da177e4SLinus Torvalds 49372994302bSJan Kara if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) 49382994302bSJan Kara return do_numa_page(vmf); 4939d10e63f2SMel Gorman 494082b0f8c3SJan Kara spin_lock(vmf->ptl); 49412994302bSJan Kara entry = vmf->orig_pte; 4942c33c7948SRyan Roberts if (unlikely(!pte_same(ptep_get(vmf->pte), entry))) { 49437df67697SBibo Mao update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 49448f4e2101SHugh Dickins goto unlock; 49457df67697SBibo Mao } 4946c89357e2SDavid Hildenbrand if (vmf->flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) { 4947f6f37321SLinus Torvalds if (!pte_write(entry)) 49482994302bSJan Kara return do_wp_page(vmf); 4949c89357e2SDavid Hildenbrand else if (likely(vmf->flags & FAULT_FLAG_WRITE)) 49501da177e4SLinus Torvalds entry = pte_mkdirty(entry); 49511da177e4SLinus Torvalds } 49521da177e4SLinus Torvalds entry = pte_mkyoung(entry); 495382b0f8c3SJan Kara if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, 495482b0f8c3SJan Kara vmf->flags & FAULT_FLAG_WRITE)) { 495582b0f8c3SJan Kara update_mmu_cache(vmf->vma, vmf->address, vmf->pte); 49561a44e149SAndrea Arcangeli } else { 4957b7333b58SYang Shi /* Skip spurious TLB flush for retried page fault */ 4958b7333b58SYang Shi if (vmf->flags & FAULT_FLAG_TRIED) 4959b7333b58SYang Shi goto unlock; 49601a44e149SAndrea Arcangeli /* 49611a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 49621a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 49631a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 49641a44e149SAndrea Arcangeli * with threads. 49651a44e149SAndrea Arcangeli */ 496682b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) 496799c29133SGerald Schaefer flush_tlb_fix_spurious_fault(vmf->vma, vmf->address, 496899c29133SGerald Schaefer vmf->pte); 49691a44e149SAndrea Arcangeli } 49708f4e2101SHugh Dickins unlock: 497182b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 497283c54070SNick Piggin return 0; 49731da177e4SLinus Torvalds } 49741da177e4SLinus Torvalds 49751da177e4SLinus Torvalds /* 49761da177e4SLinus Torvalds * By the time we get here, we already hold the mm semaphore 49779a95f3cfSPaul Cassella * 4978c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 49799138e47eSMatthew Wilcox (Oracle) * return value. See filemap_fault() and __folio_lock_or_retry(). 49801da177e4SLinus Torvalds */ 49812b740303SSouptick Joarder static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma, 49822b740303SSouptick Joarder unsigned long address, unsigned int flags) 49831da177e4SLinus Torvalds { 498482b0f8c3SJan Kara struct vm_fault vmf = { 4985bae473a4SKirill A. Shutemov .vma = vma, 49861a29d85eSJan Kara .address = address & PAGE_MASK, 4987824ddc60SNadav Amit .real_address = address, 4988bae473a4SKirill A. Shutemov .flags = flags, 49890721ec8bSJan Kara .pgoff = linear_page_index(vma, address), 4990667240e0SJan Kara .gfp_mask = __get_fault_gfp_mask(vma), 4991bae473a4SKirill A. Shutemov }; 4992dcddffd4SKirill A. Shutemov struct mm_struct *mm = vma->vm_mm; 49937da4e2cbSYang Shi unsigned long vm_flags = vma->vm_flags; 49941da177e4SLinus Torvalds pgd_t *pgd; 4995c2febafcSKirill A. Shutemov p4d_t *p4d; 49962b740303SSouptick Joarder vm_fault_t ret; 49971da177e4SLinus Torvalds 49981da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 4999c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, address); 5000c2febafcSKirill A. Shutemov if (!p4d) 5001c2febafcSKirill A. Shutemov return VM_FAULT_OOM; 5002a00cc7d9SMatthew Wilcox 5003c2febafcSKirill A. Shutemov vmf.pud = pud_alloc(mm, p4d, address); 5004a00cc7d9SMatthew Wilcox if (!vmf.pud) 5005c74df32cSHugh Dickins return VM_FAULT_OOM; 5006625110b5SThomas Hellstrom retry_pud: 50077da4e2cbSYang Shi if (pud_none(*vmf.pud) && 5008a7f4e6e4SZach O'Keefe hugepage_vma_check(vma, vm_flags, false, true, true)) { 5009a00cc7d9SMatthew Wilcox ret = create_huge_pud(&vmf); 5010a00cc7d9SMatthew Wilcox if (!(ret & VM_FAULT_FALLBACK)) 5011a00cc7d9SMatthew Wilcox return ret; 5012a00cc7d9SMatthew Wilcox } else { 5013a00cc7d9SMatthew Wilcox pud_t orig_pud = *vmf.pud; 5014a00cc7d9SMatthew Wilcox 5015a00cc7d9SMatthew Wilcox barrier(); 5016a00cc7d9SMatthew Wilcox if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { 5017a00cc7d9SMatthew Wilcox 5018c89357e2SDavid Hildenbrand /* 5019c89357e2SDavid Hildenbrand * TODO once we support anonymous PUDs: NUMA case and 5020c89357e2SDavid Hildenbrand * FAULT_FLAG_UNSHARE handling. 5021c89357e2SDavid Hildenbrand */ 5022c89357e2SDavid Hildenbrand if ((flags & FAULT_FLAG_WRITE) && !pud_write(orig_pud)) { 5023a00cc7d9SMatthew Wilcox ret = wp_huge_pud(&vmf, orig_pud); 5024a00cc7d9SMatthew Wilcox if (!(ret & VM_FAULT_FALLBACK)) 5025a00cc7d9SMatthew Wilcox return ret; 5026a00cc7d9SMatthew Wilcox } else { 5027a00cc7d9SMatthew Wilcox huge_pud_set_accessed(&vmf, orig_pud); 5028a00cc7d9SMatthew Wilcox return 0; 5029a00cc7d9SMatthew Wilcox } 5030a00cc7d9SMatthew Wilcox } 5031a00cc7d9SMatthew Wilcox } 5032a00cc7d9SMatthew Wilcox 5033a00cc7d9SMatthew Wilcox vmf.pmd = pmd_alloc(mm, vmf.pud, address); 503482b0f8c3SJan Kara if (!vmf.pmd) 5035c74df32cSHugh Dickins return VM_FAULT_OOM; 5036625110b5SThomas Hellstrom 5037625110b5SThomas Hellstrom /* Huge pud page fault raced with pmd_alloc? */ 5038625110b5SThomas Hellstrom if (pud_trans_unstable(vmf.pud)) 5039625110b5SThomas Hellstrom goto retry_pud; 5040625110b5SThomas Hellstrom 50417da4e2cbSYang Shi if (pmd_none(*vmf.pmd) && 5042a7f4e6e4SZach O'Keefe hugepage_vma_check(vma, vm_flags, false, true, true)) { 5043a2d58167SDave Jiang ret = create_huge_pmd(&vmf); 5044c0292554SKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 5045c0292554SKirill A. Shutemov return ret; 504671e3aac0SAndrea Arcangeli } else { 504726e1a0c3SHugh Dickins vmf.orig_pmd = pmdp_get_lockless(vmf.pmd); 50481f1d06c3SDavid Rientjes 50495db4f15cSYang Shi if (unlikely(is_swap_pmd(vmf.orig_pmd))) { 505084c3fc4eSZi Yan VM_BUG_ON(thp_migration_supported() && 50515db4f15cSYang Shi !is_pmd_migration_entry(vmf.orig_pmd)); 50525db4f15cSYang Shi if (is_pmd_migration_entry(vmf.orig_pmd)) 505384c3fc4eSZi Yan pmd_migration_entry_wait(mm, vmf.pmd); 505484c3fc4eSZi Yan return 0; 505584c3fc4eSZi Yan } 50565db4f15cSYang Shi if (pmd_trans_huge(vmf.orig_pmd) || pmd_devmap(vmf.orig_pmd)) { 50575db4f15cSYang Shi if (pmd_protnone(vmf.orig_pmd) && vma_is_accessible(vma)) 50585db4f15cSYang Shi return do_huge_pmd_numa_page(&vmf); 5059d10e63f2SMel Gorman 5060c89357e2SDavid Hildenbrand if ((flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) && 5061c89357e2SDavid Hildenbrand !pmd_write(vmf.orig_pmd)) { 50625db4f15cSYang Shi ret = wp_huge_pmd(&vmf); 50639845cbbdSKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 50641f1d06c3SDavid Rientjes return ret; 5065a1dd450bSWill Deacon } else { 50665db4f15cSYang Shi huge_pmd_set_accessed(&vmf); 506771e3aac0SAndrea Arcangeli return 0; 506871e3aac0SAndrea Arcangeli } 506971e3aac0SAndrea Arcangeli } 50709845cbbdSKirill A. Shutemov } 507171e3aac0SAndrea Arcangeli 507282b0f8c3SJan Kara return handle_pte_fault(&vmf); 50731da177e4SLinus Torvalds } 50741da177e4SLinus Torvalds 5075bce617edSPeter Xu /** 5076f0953a1bSIngo Molnar * mm_account_fault - Do page fault accounting 5077bce617edSPeter Xu * 5078bce617edSPeter Xu * @regs: the pt_regs struct pointer. When set to NULL, will skip accounting 5079bce617edSPeter Xu * of perf event counters, but we'll still do the per-task accounting to 5080bce617edSPeter Xu * the task who triggered this page fault. 5081bce617edSPeter Xu * @address: the faulted address. 5082bce617edSPeter Xu * @flags: the fault flags. 5083bce617edSPeter Xu * @ret: the fault retcode. 5084bce617edSPeter Xu * 5085f0953a1bSIngo Molnar * This will take care of most of the page fault accounting. Meanwhile, it 5086bce617edSPeter Xu * will also include the PERF_COUNT_SW_PAGE_FAULTS_[MAJ|MIN] perf counter 5087f0953a1bSIngo Molnar * updates. However, note that the handling of PERF_COUNT_SW_PAGE_FAULTS should 5088bce617edSPeter Xu * still be in per-arch page fault handlers at the entry of page fault. 5089bce617edSPeter Xu */ 509053156443SSuren Baghdasaryan static inline void mm_account_fault(struct mm_struct *mm, struct pt_regs *regs, 5091bce617edSPeter Xu unsigned long address, unsigned int flags, 5092bce617edSPeter Xu vm_fault_t ret) 5093bce617edSPeter Xu { 5094bce617edSPeter Xu bool major; 5095bce617edSPeter Xu 509653156443SSuren Baghdasaryan /* Incomplete faults will be accounted upon completion. */ 509753156443SSuren Baghdasaryan if (ret & VM_FAULT_RETRY) 509853156443SSuren Baghdasaryan return; 509953156443SSuren Baghdasaryan 5100bce617edSPeter Xu /* 510153156443SSuren Baghdasaryan * To preserve the behavior of older kernels, PGFAULT counters record 510253156443SSuren Baghdasaryan * both successful and failed faults, as opposed to perf counters, 510353156443SSuren Baghdasaryan * which ignore failed cases. 5104bce617edSPeter Xu */ 510553156443SSuren Baghdasaryan count_vm_event(PGFAULT); 510653156443SSuren Baghdasaryan count_memcg_event_mm(mm, PGFAULT); 510753156443SSuren Baghdasaryan 510853156443SSuren Baghdasaryan /* 510953156443SSuren Baghdasaryan * Do not account for unsuccessful faults (e.g. when the address wasn't 511053156443SSuren Baghdasaryan * valid). That includes arch_vma_access_permitted() failing before 511153156443SSuren Baghdasaryan * reaching here. So this is not a "this many hardware page faults" 511253156443SSuren Baghdasaryan * counter. We should use the hw profiling for that. 511353156443SSuren Baghdasaryan */ 511453156443SSuren Baghdasaryan if (ret & VM_FAULT_ERROR) 5115bce617edSPeter Xu return; 5116bce617edSPeter Xu 5117bce617edSPeter Xu /* 5118bce617edSPeter Xu * We define the fault as a major fault when the final successful fault 5119bce617edSPeter Xu * is VM_FAULT_MAJOR, or if it retried (which implies that we couldn't 5120bce617edSPeter Xu * handle it immediately previously). 5121bce617edSPeter Xu */ 5122bce617edSPeter Xu major = (ret & VM_FAULT_MAJOR) || (flags & FAULT_FLAG_TRIED); 5123bce617edSPeter Xu 5124a2beb5f1SPeter Xu if (major) 5125a2beb5f1SPeter Xu current->maj_flt++; 5126a2beb5f1SPeter Xu else 5127a2beb5f1SPeter Xu current->min_flt++; 5128a2beb5f1SPeter Xu 5129bce617edSPeter Xu /* 5130a2beb5f1SPeter Xu * If the fault is done for GUP, regs will be NULL. We only do the 5131a2beb5f1SPeter Xu * accounting for the per thread fault counters who triggered the 5132a2beb5f1SPeter Xu * fault, and we skip the perf event updates. 5133bce617edSPeter Xu */ 5134bce617edSPeter Xu if (!regs) 5135bce617edSPeter Xu return; 5136bce617edSPeter Xu 5137a2beb5f1SPeter Xu if (major) 5138bce617edSPeter Xu perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 5139a2beb5f1SPeter Xu else 5140bce617edSPeter Xu perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 5141bce617edSPeter Xu } 5142bce617edSPeter Xu 5143ec1c86b2SYu Zhao #ifdef CONFIG_LRU_GEN 5144ec1c86b2SYu Zhao static void lru_gen_enter_fault(struct vm_area_struct *vma) 5145ec1c86b2SYu Zhao { 51468788f678SYu Zhao /* the LRU algorithm only applies to accesses with recency */ 51478788f678SYu Zhao current->in_lru_fault = vma_has_recency(vma); 5148ec1c86b2SYu Zhao } 5149ec1c86b2SYu Zhao 5150ec1c86b2SYu Zhao static void lru_gen_exit_fault(void) 5151ec1c86b2SYu Zhao { 5152ec1c86b2SYu Zhao current->in_lru_fault = false; 5153ec1c86b2SYu Zhao } 5154ec1c86b2SYu Zhao #else 5155ec1c86b2SYu Zhao static void lru_gen_enter_fault(struct vm_area_struct *vma) 5156ec1c86b2SYu Zhao { 5157ec1c86b2SYu Zhao } 5158ec1c86b2SYu Zhao 5159ec1c86b2SYu Zhao static void lru_gen_exit_fault(void) 5160ec1c86b2SYu Zhao { 5161ec1c86b2SYu Zhao } 5162ec1c86b2SYu Zhao #endif /* CONFIG_LRU_GEN */ 5163ec1c86b2SYu Zhao 5164cdc5021cSDavid Hildenbrand static vm_fault_t sanitize_fault_flags(struct vm_area_struct *vma, 5165cdc5021cSDavid Hildenbrand unsigned int *flags) 5166cdc5021cSDavid Hildenbrand { 5167cdc5021cSDavid Hildenbrand if (unlikely(*flags & FAULT_FLAG_UNSHARE)) { 5168cdc5021cSDavid Hildenbrand if (WARN_ON_ONCE(*flags & FAULT_FLAG_WRITE)) 5169cdc5021cSDavid Hildenbrand return VM_FAULT_SIGSEGV; 5170cdc5021cSDavid Hildenbrand /* 5171cdc5021cSDavid Hildenbrand * FAULT_FLAG_UNSHARE only applies to COW mappings. Let's 5172cdc5021cSDavid Hildenbrand * just treat it like an ordinary read-fault otherwise. 5173cdc5021cSDavid Hildenbrand */ 5174cdc5021cSDavid Hildenbrand if (!is_cow_mapping(vma->vm_flags)) 5175cdc5021cSDavid Hildenbrand *flags &= ~FAULT_FLAG_UNSHARE; 517679881fedSDavid Hildenbrand } else if (*flags & FAULT_FLAG_WRITE) { 517779881fedSDavid Hildenbrand /* Write faults on read-only mappings are impossible ... */ 517879881fedSDavid Hildenbrand if (WARN_ON_ONCE(!(vma->vm_flags & VM_MAYWRITE))) 517979881fedSDavid Hildenbrand return VM_FAULT_SIGSEGV; 518079881fedSDavid Hildenbrand /* ... and FOLL_FORCE only applies to COW mappings. */ 518179881fedSDavid Hildenbrand if (WARN_ON_ONCE(!(vma->vm_flags & VM_WRITE) && 518279881fedSDavid Hildenbrand !is_cow_mapping(vma->vm_flags))) 518379881fedSDavid Hildenbrand return VM_FAULT_SIGSEGV; 5184cdc5021cSDavid Hildenbrand } 5185cdc5021cSDavid Hildenbrand return 0; 5186cdc5021cSDavid Hildenbrand } 5187cdc5021cSDavid Hildenbrand 51889a95f3cfSPaul Cassella /* 51899a95f3cfSPaul Cassella * By the time we get here, we already hold the mm semaphore 51909a95f3cfSPaul Cassella * 5191c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 51929138e47eSMatthew Wilcox (Oracle) * return value. See filemap_fault() and __folio_lock_or_retry(). 51939a95f3cfSPaul Cassella */ 51942b740303SSouptick Joarder vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address, 5195bce617edSPeter Xu unsigned int flags, struct pt_regs *regs) 5196519e5247SJohannes Weiner { 519753156443SSuren Baghdasaryan /* If the fault handler drops the mmap_lock, vma may be freed */ 519853156443SSuren Baghdasaryan struct mm_struct *mm = vma->vm_mm; 51992b740303SSouptick Joarder vm_fault_t ret; 5200519e5247SJohannes Weiner 5201519e5247SJohannes Weiner __set_current_state(TASK_RUNNING); 5202519e5247SJohannes Weiner 5203cdc5021cSDavid Hildenbrand ret = sanitize_fault_flags(vma, &flags); 5204cdc5021cSDavid Hildenbrand if (ret) 520553156443SSuren Baghdasaryan goto out; 5206cdc5021cSDavid Hildenbrand 5207de0c799bSLaurent Dufour if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, 5208de0c799bSLaurent Dufour flags & FAULT_FLAG_INSTRUCTION, 520953156443SSuren Baghdasaryan flags & FAULT_FLAG_REMOTE)) { 521053156443SSuren Baghdasaryan ret = VM_FAULT_SIGSEGV; 521153156443SSuren Baghdasaryan goto out; 521253156443SSuren Baghdasaryan } 5213de0c799bSLaurent Dufour 5214519e5247SJohannes Weiner /* 5215519e5247SJohannes Weiner * Enable the memcg OOM handling for faults triggered in user 5216519e5247SJohannes Weiner * space. Kernel faults are handled more gracefully. 5217519e5247SJohannes Weiner */ 5218519e5247SJohannes Weiner if (flags & FAULT_FLAG_USER) 521929ef680aSMichal Hocko mem_cgroup_enter_user_fault(); 5220519e5247SJohannes Weiner 5221ec1c86b2SYu Zhao lru_gen_enter_fault(vma); 5222ec1c86b2SYu Zhao 5223bae473a4SKirill A. Shutemov if (unlikely(is_vm_hugetlb_page(vma))) 5224bae473a4SKirill A. Shutemov ret = hugetlb_fault(vma->vm_mm, vma, address, flags); 5225bae473a4SKirill A. Shutemov else 5226dcddffd4SKirill A. Shutemov ret = __handle_mm_fault(vma, address, flags); 5227519e5247SJohannes Weiner 5228ec1c86b2SYu Zhao lru_gen_exit_fault(); 5229ec1c86b2SYu Zhao 523049426420SJohannes Weiner if (flags & FAULT_FLAG_USER) { 523129ef680aSMichal Hocko mem_cgroup_exit_user_fault(); 523249426420SJohannes Weiner /* 523349426420SJohannes Weiner * The task may have entered a memcg OOM situation but 523449426420SJohannes Weiner * if the allocation error was handled gracefully (no 523549426420SJohannes Weiner * VM_FAULT_OOM), there is no need to kill anything. 523649426420SJohannes Weiner * Just clean up the OOM state peacefully. 523749426420SJohannes Weiner */ 523849426420SJohannes Weiner if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) 523949426420SJohannes Weiner mem_cgroup_oom_synchronize(false); 524049426420SJohannes Weiner } 524153156443SSuren Baghdasaryan out: 524253156443SSuren Baghdasaryan mm_account_fault(mm, regs, address, flags, ret); 5243bce617edSPeter Xu 5244519e5247SJohannes Weiner return ret; 5245519e5247SJohannes Weiner } 5246e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(handle_mm_fault); 5247519e5247SJohannes Weiner 5248c2508ec5SLinus Torvalds #ifdef CONFIG_LOCK_MM_AND_FIND_VMA 5249c2508ec5SLinus Torvalds #include <linux/extable.h> 5250c2508ec5SLinus Torvalds 5251c2508ec5SLinus Torvalds static inline bool get_mmap_lock_carefully(struct mm_struct *mm, struct pt_regs *regs) 5252c2508ec5SLinus Torvalds { 5253c2508ec5SLinus Torvalds /* Even if this succeeds, make it clear we *might* have slept */ 5254c2508ec5SLinus Torvalds if (likely(mmap_read_trylock(mm))) { 5255c2508ec5SLinus Torvalds might_sleep(); 5256c2508ec5SLinus Torvalds return true; 5257c2508ec5SLinus Torvalds } 5258c2508ec5SLinus Torvalds 5259c2508ec5SLinus Torvalds if (regs && !user_mode(regs)) { 5260c2508ec5SLinus Torvalds unsigned long ip = instruction_pointer(regs); 5261c2508ec5SLinus Torvalds if (!search_exception_tables(ip)) 5262c2508ec5SLinus Torvalds return false; 5263c2508ec5SLinus Torvalds } 5264c2508ec5SLinus Torvalds 5265eda00472SLinus Torvalds return !mmap_read_lock_killable(mm); 5266c2508ec5SLinus Torvalds } 5267c2508ec5SLinus Torvalds 5268c2508ec5SLinus Torvalds static inline bool mmap_upgrade_trylock(struct mm_struct *mm) 5269c2508ec5SLinus Torvalds { 5270c2508ec5SLinus Torvalds /* 5271c2508ec5SLinus Torvalds * We don't have this operation yet. 5272c2508ec5SLinus Torvalds * 5273c2508ec5SLinus Torvalds * It should be easy enough to do: it's basically a 5274c2508ec5SLinus Torvalds * atomic_long_try_cmpxchg_acquire() 5275c2508ec5SLinus Torvalds * from RWSEM_READER_BIAS -> RWSEM_WRITER_LOCKED, but 5276c2508ec5SLinus Torvalds * it also needs the proper lockdep magic etc. 5277c2508ec5SLinus Torvalds */ 5278c2508ec5SLinus Torvalds return false; 5279c2508ec5SLinus Torvalds } 5280c2508ec5SLinus Torvalds 5281c2508ec5SLinus Torvalds static inline bool upgrade_mmap_lock_carefully(struct mm_struct *mm, struct pt_regs *regs) 5282c2508ec5SLinus Torvalds { 5283c2508ec5SLinus Torvalds mmap_read_unlock(mm); 5284c2508ec5SLinus Torvalds if (regs && !user_mode(regs)) { 5285c2508ec5SLinus Torvalds unsigned long ip = instruction_pointer(regs); 5286c2508ec5SLinus Torvalds if (!search_exception_tables(ip)) 5287c2508ec5SLinus Torvalds return false; 5288c2508ec5SLinus Torvalds } 5289eda00472SLinus Torvalds return !mmap_write_lock_killable(mm); 5290c2508ec5SLinus Torvalds } 5291c2508ec5SLinus Torvalds 5292c2508ec5SLinus Torvalds /* 5293c2508ec5SLinus Torvalds * Helper for page fault handling. 5294c2508ec5SLinus Torvalds * 5295c2508ec5SLinus Torvalds * This is kind of equivalend to "mmap_read_lock()" followed 5296c2508ec5SLinus Torvalds * by "find_extend_vma()", except it's a lot more careful about 5297c2508ec5SLinus Torvalds * the locking (and will drop the lock on failure). 5298c2508ec5SLinus Torvalds * 5299c2508ec5SLinus Torvalds * For example, if we have a kernel bug that causes a page 5300c2508ec5SLinus Torvalds * fault, we don't want to just use mmap_read_lock() to get 5301c2508ec5SLinus Torvalds * the mm lock, because that would deadlock if the bug were 5302c2508ec5SLinus Torvalds * to happen while we're holding the mm lock for writing. 5303c2508ec5SLinus Torvalds * 5304c2508ec5SLinus Torvalds * So this checks the exception tables on kernel faults in 5305c2508ec5SLinus Torvalds * order to only do this all for instructions that are actually 5306c2508ec5SLinus Torvalds * expected to fault. 5307c2508ec5SLinus Torvalds * 5308c2508ec5SLinus Torvalds * We can also actually take the mm lock for writing if we 5309c2508ec5SLinus Torvalds * need to extend the vma, which helps the VM layer a lot. 5310c2508ec5SLinus Torvalds */ 5311c2508ec5SLinus Torvalds struct vm_area_struct *lock_mm_and_find_vma(struct mm_struct *mm, 5312c2508ec5SLinus Torvalds unsigned long addr, struct pt_regs *regs) 5313c2508ec5SLinus Torvalds { 5314c2508ec5SLinus Torvalds struct vm_area_struct *vma; 5315c2508ec5SLinus Torvalds 5316c2508ec5SLinus Torvalds if (!get_mmap_lock_carefully(mm, regs)) 5317c2508ec5SLinus Torvalds return NULL; 5318c2508ec5SLinus Torvalds 5319c2508ec5SLinus Torvalds vma = find_vma(mm, addr); 5320c2508ec5SLinus Torvalds if (likely(vma && (vma->vm_start <= addr))) 5321c2508ec5SLinus Torvalds return vma; 5322c2508ec5SLinus Torvalds 5323c2508ec5SLinus Torvalds /* 5324c2508ec5SLinus Torvalds * Well, dang. We might still be successful, but only 5325c2508ec5SLinus Torvalds * if we can extend a vma to do so. 5326c2508ec5SLinus Torvalds */ 5327c2508ec5SLinus Torvalds if (!vma || !(vma->vm_flags & VM_GROWSDOWN)) { 5328c2508ec5SLinus Torvalds mmap_read_unlock(mm); 5329c2508ec5SLinus Torvalds return NULL; 5330c2508ec5SLinus Torvalds } 5331c2508ec5SLinus Torvalds 5332c2508ec5SLinus Torvalds /* 5333c2508ec5SLinus Torvalds * We can try to upgrade the mmap lock atomically, 5334c2508ec5SLinus Torvalds * in which case we can continue to use the vma 5335c2508ec5SLinus Torvalds * we already looked up. 5336c2508ec5SLinus Torvalds * 5337c2508ec5SLinus Torvalds * Otherwise we'll have to drop the mmap lock and 5338c2508ec5SLinus Torvalds * re-take it, and also look up the vma again, 5339c2508ec5SLinus Torvalds * re-checking it. 5340c2508ec5SLinus Torvalds */ 5341c2508ec5SLinus Torvalds if (!mmap_upgrade_trylock(mm)) { 5342c2508ec5SLinus Torvalds if (!upgrade_mmap_lock_carefully(mm, regs)) 5343c2508ec5SLinus Torvalds return NULL; 5344c2508ec5SLinus Torvalds 5345c2508ec5SLinus Torvalds vma = find_vma(mm, addr); 5346c2508ec5SLinus Torvalds if (!vma) 5347c2508ec5SLinus Torvalds goto fail; 5348c2508ec5SLinus Torvalds if (vma->vm_start <= addr) 5349c2508ec5SLinus Torvalds goto success; 5350c2508ec5SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 5351c2508ec5SLinus Torvalds goto fail; 5352c2508ec5SLinus Torvalds } 5353c2508ec5SLinus Torvalds 53548d7071afSLinus Torvalds if (expand_stack_locked(vma, addr)) 5355c2508ec5SLinus Torvalds goto fail; 5356c2508ec5SLinus Torvalds 5357c2508ec5SLinus Torvalds success: 5358c2508ec5SLinus Torvalds mmap_write_downgrade(mm); 5359c2508ec5SLinus Torvalds return vma; 5360c2508ec5SLinus Torvalds 5361c2508ec5SLinus Torvalds fail: 5362c2508ec5SLinus Torvalds mmap_write_unlock(mm); 5363c2508ec5SLinus Torvalds return NULL; 5364c2508ec5SLinus Torvalds } 5365c2508ec5SLinus Torvalds #endif 5366c2508ec5SLinus Torvalds 536750ee3253SSuren Baghdasaryan #ifdef CONFIG_PER_VMA_LOCK 536850ee3253SSuren Baghdasaryan /* 536950ee3253SSuren Baghdasaryan * Lookup and lock a VMA under RCU protection. Returned VMA is guaranteed to be 537050ee3253SSuren Baghdasaryan * stable and not isolated. If the VMA is not found or is being modified the 537150ee3253SSuren Baghdasaryan * function returns NULL. 537250ee3253SSuren Baghdasaryan */ 537350ee3253SSuren Baghdasaryan struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm, 537450ee3253SSuren Baghdasaryan unsigned long address) 537550ee3253SSuren Baghdasaryan { 537650ee3253SSuren Baghdasaryan MA_STATE(mas, &mm->mm_mt, address, address); 537750ee3253SSuren Baghdasaryan struct vm_area_struct *vma; 537850ee3253SSuren Baghdasaryan 537950ee3253SSuren Baghdasaryan rcu_read_lock(); 538050ee3253SSuren Baghdasaryan retry: 538150ee3253SSuren Baghdasaryan vma = mas_walk(&mas); 538250ee3253SSuren Baghdasaryan if (!vma) 538350ee3253SSuren Baghdasaryan goto inval; 538450ee3253SSuren Baghdasaryan 53857a7f0946SArjun Roy /* Only anonymous and tcp vmas are supported for now */ 53867a7f0946SArjun Roy if (!vma_is_anonymous(vma) && !vma_is_tcp(vma)) 538750ee3253SSuren Baghdasaryan goto inval; 538850ee3253SSuren Baghdasaryan 53892ac0af1bSSuren Baghdasaryan /* find_mergeable_anon_vma uses adjacent vmas which are not locked */ 53907a7f0946SArjun Roy if (!vma->anon_vma && !vma_is_tcp(vma)) 53912ac0af1bSSuren Baghdasaryan goto inval; 53922ac0af1bSSuren Baghdasaryan 539350ee3253SSuren Baghdasaryan if (!vma_start_read(vma)) 539450ee3253SSuren Baghdasaryan goto inval; 539550ee3253SSuren Baghdasaryan 5396444eeb17SSuren Baghdasaryan /* 5397444eeb17SSuren Baghdasaryan * Due to the possibility of userfault handler dropping mmap_lock, avoid 5398444eeb17SSuren Baghdasaryan * it for now and fall back to page fault handling under mmap_lock. 5399444eeb17SSuren Baghdasaryan */ 5400444eeb17SSuren Baghdasaryan if (userfaultfd_armed(vma)) { 5401444eeb17SSuren Baghdasaryan vma_end_read(vma); 5402444eeb17SSuren Baghdasaryan goto inval; 5403444eeb17SSuren Baghdasaryan } 5404444eeb17SSuren Baghdasaryan 540550ee3253SSuren Baghdasaryan /* Check since vm_start/vm_end might change before we lock the VMA */ 540650ee3253SSuren Baghdasaryan if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { 540750ee3253SSuren Baghdasaryan vma_end_read(vma); 540850ee3253SSuren Baghdasaryan goto inval; 540950ee3253SSuren Baghdasaryan } 541050ee3253SSuren Baghdasaryan 541150ee3253SSuren Baghdasaryan /* Check if the VMA got isolated after we found it */ 541250ee3253SSuren Baghdasaryan if (vma->detached) { 541350ee3253SSuren Baghdasaryan vma_end_read(vma); 541452f23865SSuren Baghdasaryan count_vm_vma_lock_event(VMA_LOCK_MISS); 541550ee3253SSuren Baghdasaryan /* The area was replaced with another one */ 541650ee3253SSuren Baghdasaryan goto retry; 541750ee3253SSuren Baghdasaryan } 541850ee3253SSuren Baghdasaryan 541950ee3253SSuren Baghdasaryan rcu_read_unlock(); 542050ee3253SSuren Baghdasaryan return vma; 542150ee3253SSuren Baghdasaryan inval: 542250ee3253SSuren Baghdasaryan rcu_read_unlock(); 542352f23865SSuren Baghdasaryan count_vm_vma_lock_event(VMA_LOCK_ABORT); 542450ee3253SSuren Baghdasaryan return NULL; 542550ee3253SSuren Baghdasaryan } 542650ee3253SSuren Baghdasaryan #endif /* CONFIG_PER_VMA_LOCK */ 542750ee3253SSuren Baghdasaryan 542890eceff1SKirill A. Shutemov #ifndef __PAGETABLE_P4D_FOLDED 542990eceff1SKirill A. Shutemov /* 543090eceff1SKirill A. Shutemov * Allocate p4d page table. 543190eceff1SKirill A. Shutemov * We've already handled the fast-path in-line. 543290eceff1SKirill A. Shutemov */ 543390eceff1SKirill A. Shutemov int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 543490eceff1SKirill A. Shutemov { 543590eceff1SKirill A. Shutemov p4d_t *new = p4d_alloc_one(mm, address); 543690eceff1SKirill A. Shutemov if (!new) 543790eceff1SKirill A. Shutemov return -ENOMEM; 543890eceff1SKirill A. Shutemov 543990eceff1SKirill A. Shutemov spin_lock(&mm->page_table_lock); 5440ed33b5a6SQi Zheng if (pgd_present(*pgd)) { /* Another has populated it */ 544190eceff1SKirill A. Shutemov p4d_free(mm, new); 5442ed33b5a6SQi Zheng } else { 5443ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 544490eceff1SKirill A. Shutemov pgd_populate(mm, pgd, new); 5445ed33b5a6SQi Zheng } 544690eceff1SKirill A. Shutemov spin_unlock(&mm->page_table_lock); 544790eceff1SKirill A. Shutemov return 0; 544890eceff1SKirill A. Shutemov } 544990eceff1SKirill A. Shutemov #endif /* __PAGETABLE_P4D_FOLDED */ 545090eceff1SKirill A. Shutemov 54511da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 54521da177e4SLinus Torvalds /* 54531da177e4SLinus Torvalds * Allocate page upper directory. 5454872fec16SHugh Dickins * We've already handled the fast-path in-line. 54551da177e4SLinus Torvalds */ 5456c2febafcSKirill A. Shutemov int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) 54571da177e4SLinus Torvalds { 5458c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 5459c74df32cSHugh Dickins if (!new) 54601bb3630eSHugh Dickins return -ENOMEM; 54611da177e4SLinus Torvalds 5462872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 5463b4e98d9aSKirill A. Shutemov if (!p4d_present(*p4d)) { 5464b4e98d9aSKirill A. Shutemov mm_inc_nr_puds(mm); 5465ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 5466c2febafcSKirill A. Shutemov p4d_populate(mm, p4d, new); 5467b4e98d9aSKirill A. Shutemov } else /* Another has populated it */ 5468c2febafcSKirill A. Shutemov pud_free(mm, new); 5469872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 54701bb3630eSHugh Dickins return 0; 54711da177e4SLinus Torvalds } 54721da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 54731da177e4SLinus Torvalds 54741da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 54751da177e4SLinus Torvalds /* 54761da177e4SLinus Torvalds * Allocate page middle directory. 5477872fec16SHugh Dickins * We've already handled the fast-path in-line. 54781da177e4SLinus Torvalds */ 54791bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 54801da177e4SLinus Torvalds { 5481a00cc7d9SMatthew Wilcox spinlock_t *ptl; 5482c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 5483c74df32cSHugh Dickins if (!new) 54841bb3630eSHugh Dickins return -ENOMEM; 54851da177e4SLinus Torvalds 5486a00cc7d9SMatthew Wilcox ptl = pud_lock(mm, pud); 5487dc6c9a35SKirill A. Shutemov if (!pud_present(*pud)) { 5488dc6c9a35SKirill A. Shutemov mm_inc_nr_pmds(mm); 5489ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 54901da177e4SLinus Torvalds pud_populate(mm, pud, new); 5491ed33b5a6SQi Zheng } else { /* Another has populated it */ 54925e541973SBenjamin Herrenschmidt pmd_free(mm, new); 5493ed33b5a6SQi Zheng } 5494a00cc7d9SMatthew Wilcox spin_unlock(ptl); 54951bb3630eSHugh Dickins return 0; 54961da177e4SLinus Torvalds } 54971da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 54981da177e4SLinus Torvalds 54993b6748e2SJohannes Weiner /** 55009fd6dad1SPaolo Bonzini * follow_pte - look up PTE at a user virtual address 55019fd6dad1SPaolo Bonzini * @mm: the mm_struct of the target address space 55029fd6dad1SPaolo Bonzini * @address: user virtual address 55039fd6dad1SPaolo Bonzini * @ptepp: location to store found PTE 55049fd6dad1SPaolo Bonzini * @ptlp: location to store the lock for the PTE 55059fd6dad1SPaolo Bonzini * 55069fd6dad1SPaolo Bonzini * On a successful return, the pointer to the PTE is stored in @ptepp; 55079fd6dad1SPaolo Bonzini * the corresponding lock is taken and its location is stored in @ptlp. 55089fd6dad1SPaolo Bonzini * The contents of the PTE are only stable until @ptlp is released; 55099fd6dad1SPaolo Bonzini * any further use, if any, must be protected against invalidation 55109fd6dad1SPaolo Bonzini * with MMU notifiers. 55119fd6dad1SPaolo Bonzini * 55129fd6dad1SPaolo Bonzini * Only IO mappings and raw PFN mappings are allowed. The mmap semaphore 55139fd6dad1SPaolo Bonzini * should be taken for read. 55149fd6dad1SPaolo Bonzini * 55159fd6dad1SPaolo Bonzini * KVM uses this function. While it is arguably less bad than ``follow_pfn``, 55169fd6dad1SPaolo Bonzini * it is not a good general-purpose API. 55179fd6dad1SPaolo Bonzini * 55189fd6dad1SPaolo Bonzini * Return: zero on success, -ve otherwise. 55199fd6dad1SPaolo Bonzini */ 55209fd6dad1SPaolo Bonzini int follow_pte(struct mm_struct *mm, unsigned long address, 55219fd6dad1SPaolo Bonzini pte_t **ptepp, spinlock_t **ptlp) 55229fd6dad1SPaolo Bonzini { 55230e5e64c0SMuchun Song pgd_t *pgd; 55240e5e64c0SMuchun Song p4d_t *p4d; 55250e5e64c0SMuchun Song pud_t *pud; 55260e5e64c0SMuchun Song pmd_t *pmd; 55270e5e64c0SMuchun Song pte_t *ptep; 55280e5e64c0SMuchun Song 55290e5e64c0SMuchun Song pgd = pgd_offset(mm, address); 55300e5e64c0SMuchun Song if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 55310e5e64c0SMuchun Song goto out; 55320e5e64c0SMuchun Song 55330e5e64c0SMuchun Song p4d = p4d_offset(pgd, address); 55340e5e64c0SMuchun Song if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) 55350e5e64c0SMuchun Song goto out; 55360e5e64c0SMuchun Song 55370e5e64c0SMuchun Song pud = pud_offset(p4d, address); 55380e5e64c0SMuchun Song if (pud_none(*pud) || unlikely(pud_bad(*pud))) 55390e5e64c0SMuchun Song goto out; 55400e5e64c0SMuchun Song 55410e5e64c0SMuchun Song pmd = pmd_offset(pud, address); 55420e5e64c0SMuchun Song VM_BUG_ON(pmd_trans_huge(*pmd)); 55430e5e64c0SMuchun Song 55440e5e64c0SMuchun Song ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 55453db82b93SHugh Dickins if (!ptep) 55463db82b93SHugh Dickins goto out; 5547c33c7948SRyan Roberts if (!pte_present(ptep_get(ptep))) 55480e5e64c0SMuchun Song goto unlock; 55490e5e64c0SMuchun Song *ptepp = ptep; 55500e5e64c0SMuchun Song return 0; 55510e5e64c0SMuchun Song unlock: 55520e5e64c0SMuchun Song pte_unmap_unlock(ptep, *ptlp); 55530e5e64c0SMuchun Song out: 55540e5e64c0SMuchun Song return -EINVAL; 55559fd6dad1SPaolo Bonzini } 55569fd6dad1SPaolo Bonzini EXPORT_SYMBOL_GPL(follow_pte); 55579fd6dad1SPaolo Bonzini 55589fd6dad1SPaolo Bonzini /** 55593b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 55603b6748e2SJohannes Weiner * @vma: memory mapping 55613b6748e2SJohannes Weiner * @address: user virtual address 55623b6748e2SJohannes Weiner * @pfn: location to store found PFN 55633b6748e2SJohannes Weiner * 55643b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 55653b6748e2SJohannes Weiner * 55669fd6dad1SPaolo Bonzini * This function does not allow the caller to read the permissions 55679fd6dad1SPaolo Bonzini * of the PTE. Do not use it. 55689fd6dad1SPaolo Bonzini * 5569a862f68aSMike Rapoport * Return: zero and the pfn at @pfn on success, -ve otherwise. 55703b6748e2SJohannes Weiner */ 55713b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 55723b6748e2SJohannes Weiner unsigned long *pfn) 55733b6748e2SJohannes Weiner { 55743b6748e2SJohannes Weiner int ret = -EINVAL; 55753b6748e2SJohannes Weiner spinlock_t *ptl; 55763b6748e2SJohannes Weiner pte_t *ptep; 55773b6748e2SJohannes Weiner 55783b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 55793b6748e2SJohannes Weiner return ret; 55803b6748e2SJohannes Weiner 55819fd6dad1SPaolo Bonzini ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 55823b6748e2SJohannes Weiner if (ret) 55833b6748e2SJohannes Weiner return ret; 5584c33c7948SRyan Roberts *pfn = pte_pfn(ptep_get(ptep)); 55853b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 55863b6748e2SJohannes Weiner return 0; 55873b6748e2SJohannes Weiner } 55883b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 55893b6748e2SJohannes Weiner 559028b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 5591d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 559228b2ee20SRik van Riel unsigned long address, unsigned int flags, 5593d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 559428b2ee20SRik van Riel { 559503668a4dSJohannes Weiner int ret = -EINVAL; 559628b2ee20SRik van Riel pte_t *ptep, pte; 559728b2ee20SRik van Riel spinlock_t *ptl; 559828b2ee20SRik van Riel 5599d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 5600d87fe660Svenkatesh.pallipadi@intel.com goto out; 560128b2ee20SRik van Riel 56029fd6dad1SPaolo Bonzini if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 5603d87fe660Svenkatesh.pallipadi@intel.com goto out; 5604c33c7948SRyan Roberts pte = ptep_get(ptep); 560503668a4dSJohannes Weiner 5606f6f37321SLinus Torvalds if ((flags & FOLL_WRITE) && !pte_write(pte)) 560728b2ee20SRik van Riel goto unlock; 560828b2ee20SRik van Riel 560928b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 561003668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 561128b2ee20SRik van Riel 561203668a4dSJohannes Weiner ret = 0; 561328b2ee20SRik van Riel unlock: 561428b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 561528b2ee20SRik van Riel out: 5616d87fe660Svenkatesh.pallipadi@intel.com return ret; 561728b2ee20SRik van Riel } 561828b2ee20SRik van Riel 561996667f8aSDaniel Vetter /** 562096667f8aSDaniel Vetter * generic_access_phys - generic implementation for iomem mmap access 562196667f8aSDaniel Vetter * @vma: the vma to access 5622f0953a1bSIngo Molnar * @addr: userspace address, not relative offset within @vma 562396667f8aSDaniel Vetter * @buf: buffer to read/write 562496667f8aSDaniel Vetter * @len: length of transfer 562596667f8aSDaniel Vetter * @write: set to FOLL_WRITE when writing, otherwise reading 562696667f8aSDaniel Vetter * 562796667f8aSDaniel Vetter * This is a generic implementation for &vm_operations_struct.access for an 562896667f8aSDaniel Vetter * iomem mapping. This callback is used by access_process_vm() when the @vma is 562996667f8aSDaniel Vetter * not page based. 563096667f8aSDaniel Vetter */ 563128b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 563228b2ee20SRik van Riel void *buf, int len, int write) 563328b2ee20SRik van Riel { 563428b2ee20SRik van Riel resource_size_t phys_addr; 563528b2ee20SRik van Riel unsigned long prot = 0; 56362bc7273bSKOSAKI Motohiro void __iomem *maddr; 563796667f8aSDaniel Vetter pte_t *ptep, pte; 563896667f8aSDaniel Vetter spinlock_t *ptl; 563996667f8aSDaniel Vetter int offset = offset_in_page(addr); 564096667f8aSDaniel Vetter int ret = -EINVAL; 564128b2ee20SRik van Riel 564296667f8aSDaniel Vetter if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 564396667f8aSDaniel Vetter return -EINVAL; 564496667f8aSDaniel Vetter 564596667f8aSDaniel Vetter retry: 5646e913a8cdSLinus Torvalds if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) 564796667f8aSDaniel Vetter return -EINVAL; 5648c33c7948SRyan Roberts pte = ptep_get(ptep); 564996667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 565096667f8aSDaniel Vetter 565196667f8aSDaniel Vetter prot = pgprot_val(pte_pgprot(pte)); 565296667f8aSDaniel Vetter phys_addr = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 565396667f8aSDaniel Vetter 565496667f8aSDaniel Vetter if ((write & FOLL_WRITE) && !pte_write(pte)) 565528b2ee20SRik van Riel return -EINVAL; 565628b2ee20SRik van Riel 56579cb12d7bSGrazvydas Ignotas maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); 565824eee1e4Sjie@chenjie6@huwei.com if (!maddr) 565924eee1e4Sjie@chenjie6@huwei.com return -ENOMEM; 566024eee1e4Sjie@chenjie6@huwei.com 5661e913a8cdSLinus Torvalds if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) 566296667f8aSDaniel Vetter goto out_unmap; 566396667f8aSDaniel Vetter 5664c33c7948SRyan Roberts if (!pte_same(pte, ptep_get(ptep))) { 566596667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 566696667f8aSDaniel Vetter iounmap(maddr); 566796667f8aSDaniel Vetter 566896667f8aSDaniel Vetter goto retry; 566996667f8aSDaniel Vetter } 567096667f8aSDaniel Vetter 567128b2ee20SRik van Riel if (write) 567228b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 567328b2ee20SRik van Riel else 567428b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 567596667f8aSDaniel Vetter ret = len; 567696667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 567796667f8aSDaniel Vetter out_unmap: 567828b2ee20SRik van Riel iounmap(maddr); 567928b2ee20SRik van Riel 568096667f8aSDaniel Vetter return ret; 568128b2ee20SRik van Riel } 56825a73633eSUwe Kleine-König EXPORT_SYMBOL_GPL(generic_access_phys); 568328b2ee20SRik van Riel #endif 568428b2ee20SRik van Riel 56850ec76a11SDavid Howells /* 5686d3f5ffcaSJohn Hubbard * Access another process' address space as given in mm. 56870ec76a11SDavid Howells */ 5688d3f5ffcaSJohn Hubbard int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf, 5689d3f5ffcaSJohn Hubbard int len, unsigned int gup_flags) 56900ec76a11SDavid Howells { 56910ec76a11SDavid Howells void *old_buf = buf; 5692442486ecSLorenzo Stoakes int write = gup_flags & FOLL_WRITE; 56930ec76a11SDavid Howells 5694d8ed45c5SMichel Lespinasse if (mmap_read_lock_killable(mm)) 56951e426fe2SKonstantin Khlebnikov return 0; 56961e426fe2SKonstantin Khlebnikov 5697*eee9c708SLinus Torvalds /* Avoid triggering the temporary warning in __get_user_pages */ 5698*eee9c708SLinus Torvalds if (!vma_lookup(mm, addr) && !expand_stack(mm, addr)) 5699*eee9c708SLinus Torvalds return 0; 5700*eee9c708SLinus Torvalds 5701183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 57020ec76a11SDavid Howells while (len) { 5703ca5e8632SLorenzo Stoakes int bytes, offset; 57040ec76a11SDavid Howells void *maddr; 5705ca5e8632SLorenzo Stoakes struct vm_area_struct *vma = NULL; 5706ca5e8632SLorenzo Stoakes struct page *page = get_user_page_vma_remote(mm, addr, 5707ca5e8632SLorenzo Stoakes gup_flags, &vma); 57080ec76a11SDavid Howells 5709ca5e8632SLorenzo Stoakes if (IS_ERR_OR_NULL(page)) { 57108d7071afSLinus Torvalds /* We might need to expand the stack to access it */ 57118d7071afSLinus Torvalds vma = vma_lookup(mm, addr); 57128d7071afSLinus Torvalds if (!vma) { 57138d7071afSLinus Torvalds vma = expand_stack(mm, addr); 57149471f1f2SLinus Torvalds 57159471f1f2SLinus Torvalds /* mmap_lock was dropped on failure */ 57168d7071afSLinus Torvalds if (!vma) 57179471f1f2SLinus Torvalds return buf - old_buf; 57189471f1f2SLinus Torvalds 57199471f1f2SLinus Torvalds /* Try again if stack expansion worked */ 57209471f1f2SLinus Torvalds continue; 57218d7071afSLinus Torvalds } 57228d7071afSLinus Torvalds 5723ca5e8632SLorenzo Stoakes 572428b2ee20SRik van Riel /* 572528b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 572628b2ee20SRik van Riel * we can access using slightly different code. 572728b2ee20SRik van Riel */ 57289471f1f2SLinus Torvalds bytes = 0; 57299471f1f2SLinus Torvalds #ifdef CONFIG_HAVE_IOREMAP_PROT 573028b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 57319471f1f2SLinus Torvalds bytes = vma->vm_ops->access(vma, addr, buf, 573228b2ee20SRik van Riel len, write); 5733dbffcd03SRik van Riel #endif 57349471f1f2SLinus Torvalds if (bytes <= 0) 57359471f1f2SLinus Torvalds break; 573628b2ee20SRik van Riel } else { 57370ec76a11SDavid Howells bytes = len; 57380ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 57390ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 57400ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 57410ec76a11SDavid Howells 57420ec76a11SDavid Howells maddr = kmap(page); 57430ec76a11SDavid Howells if (write) { 57440ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 57450ec76a11SDavid Howells maddr + offset, buf, bytes); 57460ec76a11SDavid Howells set_page_dirty_lock(page); 57470ec76a11SDavid Howells } else { 57480ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 57490ec76a11SDavid Howells buf, maddr + offset, bytes); 57500ec76a11SDavid Howells } 57510ec76a11SDavid Howells kunmap(page); 575209cbfeafSKirill A. Shutemov put_page(page); 575328b2ee20SRik van Riel } 57540ec76a11SDavid Howells len -= bytes; 57550ec76a11SDavid Howells buf += bytes; 57560ec76a11SDavid Howells addr += bytes; 57570ec76a11SDavid Howells } 5758d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 57590ec76a11SDavid Howells 57600ec76a11SDavid Howells return buf - old_buf; 57610ec76a11SDavid Howells } 576203252919SAndi Kleen 57635ddd36b9SStephen Wilson /** 5764ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 57655ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 57665ddd36b9SStephen Wilson * @addr: start address to access 57675ddd36b9SStephen Wilson * @buf: source or destination buffer 57685ddd36b9SStephen Wilson * @len: number of bytes to transfer 57696347e8d5SLorenzo Stoakes * @gup_flags: flags modifying lookup behaviour 57705ddd36b9SStephen Wilson * 57715ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 5772a862f68aSMike Rapoport * 5773a862f68aSMike Rapoport * Return: number of bytes copied from source to destination. 57745ddd36b9SStephen Wilson */ 57755ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 57766347e8d5SLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 57775ddd36b9SStephen Wilson { 5778d3f5ffcaSJohn Hubbard return __access_remote_vm(mm, addr, buf, len, gup_flags); 57795ddd36b9SStephen Wilson } 57805ddd36b9SStephen Wilson 578103252919SAndi Kleen /* 5782206cb636SStephen Wilson * Access another process' address space. 5783206cb636SStephen Wilson * Source/target buffer must be kernel space, 5784206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 5785206cb636SStephen Wilson */ 5786206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 5787f307ab6dSLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 5788206cb636SStephen Wilson { 5789206cb636SStephen Wilson struct mm_struct *mm; 5790206cb636SStephen Wilson int ret; 5791206cb636SStephen Wilson 5792206cb636SStephen Wilson mm = get_task_mm(tsk); 5793206cb636SStephen Wilson if (!mm) 5794206cb636SStephen Wilson return 0; 5795206cb636SStephen Wilson 5796d3f5ffcaSJohn Hubbard ret = __access_remote_vm(mm, addr, buf, len, gup_flags); 5797442486ecSLorenzo Stoakes 5798206cb636SStephen Wilson mmput(mm); 5799206cb636SStephen Wilson 5800206cb636SStephen Wilson return ret; 5801206cb636SStephen Wilson } 5802fcd35857SCatalin Marinas EXPORT_SYMBOL_GPL(access_process_vm); 5803206cb636SStephen Wilson 580403252919SAndi Kleen /* 580503252919SAndi Kleen * Print the name of a VMA. 580603252919SAndi Kleen */ 580703252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 580803252919SAndi Kleen { 580903252919SAndi Kleen struct mm_struct *mm = current->mm; 581003252919SAndi Kleen struct vm_area_struct *vma; 581103252919SAndi Kleen 5812e8bff74aSIngo Molnar /* 58130a7f682dSMichal Hocko * we might be running from an atomic context so we cannot sleep 5814e8bff74aSIngo Molnar */ 5815d8ed45c5SMichel Lespinasse if (!mmap_read_trylock(mm)) 5816e8bff74aSIngo Molnar return; 5817e8bff74aSIngo Molnar 581803252919SAndi Kleen vma = find_vma(mm, ip); 581903252919SAndi Kleen if (vma && vma->vm_file) { 582003252919SAndi Kleen struct file *f = vma->vm_file; 58210a7f682dSMichal Hocko char *buf = (char *)__get_free_page(GFP_NOWAIT); 582203252919SAndi Kleen if (buf) { 58232fbc57c5SAndy Shevchenko char *p; 582403252919SAndi Kleen 58259bf39ab2SMiklos Szeredi p = file_path(f, buf, PAGE_SIZE); 582603252919SAndi Kleen if (IS_ERR(p)) 582703252919SAndi Kleen p = "?"; 58282fbc57c5SAndy Shevchenko printk("%s%s[%lx+%lx]", prefix, kbasename(p), 582903252919SAndi Kleen vma->vm_start, 583003252919SAndi Kleen vma->vm_end - vma->vm_start); 583103252919SAndi Kleen free_page((unsigned long)buf); 583203252919SAndi Kleen } 583303252919SAndi Kleen } 5834d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 583503252919SAndi Kleen } 58363ee1afa3SNick Piggin 5837662bbcb2SMichael S. Tsirkin #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) 58389ec23531SDavid Hildenbrand void __might_fault(const char *file, int line) 58393ee1afa3SNick Piggin { 58409ec23531SDavid Hildenbrand if (pagefault_disabled()) 5841662bbcb2SMichael S. Tsirkin return; 584242a38756SThomas Gleixner __might_sleep(file, line); 58439ec23531SDavid Hildenbrand #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) 5844662bbcb2SMichael S. Tsirkin if (current->mm) 5845da1c55f1SMichel Lespinasse might_lock_read(¤t->mm->mmap_lock); 58469ec23531SDavid Hildenbrand #endif 58473ee1afa3SNick Piggin } 58489ec23531SDavid Hildenbrand EXPORT_SYMBOL(__might_fault); 58493ee1afa3SNick Piggin #endif 585047ad8475SAndrea Arcangeli 585147ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 5852c6ddfb6cSHuang Ying /* 5853c6ddfb6cSHuang Ying * Process all subpages of the specified huge page with the specified 5854c6ddfb6cSHuang Ying * operation. The target subpage will be processed last to keep its 5855c6ddfb6cSHuang Ying * cache lines hot. 5856c6ddfb6cSHuang Ying */ 58571cb9dc4bSLiu Shixin static inline int process_huge_page( 5858c6ddfb6cSHuang Ying unsigned long addr_hint, unsigned int pages_per_huge_page, 58591cb9dc4bSLiu Shixin int (*process_subpage)(unsigned long addr, int idx, void *arg), 5860c6ddfb6cSHuang Ying void *arg) 5861c6ddfb6cSHuang Ying { 58621cb9dc4bSLiu Shixin int i, n, base, l, ret; 5863c6ddfb6cSHuang Ying unsigned long addr = addr_hint & 5864c6ddfb6cSHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 5865c6ddfb6cSHuang Ying 5866c6ddfb6cSHuang Ying /* Process target subpage last to keep its cache lines hot */ 5867c6ddfb6cSHuang Ying might_sleep(); 5868c6ddfb6cSHuang Ying n = (addr_hint - addr) / PAGE_SIZE; 5869c6ddfb6cSHuang Ying if (2 * n <= pages_per_huge_page) { 5870c6ddfb6cSHuang Ying /* If target subpage in first half of huge page */ 5871c6ddfb6cSHuang Ying base = 0; 5872c6ddfb6cSHuang Ying l = n; 5873c6ddfb6cSHuang Ying /* Process subpages at the end of huge page */ 5874c6ddfb6cSHuang Ying for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { 5875c6ddfb6cSHuang Ying cond_resched(); 58761cb9dc4bSLiu Shixin ret = process_subpage(addr + i * PAGE_SIZE, i, arg); 58771cb9dc4bSLiu Shixin if (ret) 58781cb9dc4bSLiu Shixin return ret; 5879c6ddfb6cSHuang Ying } 5880c6ddfb6cSHuang Ying } else { 5881c6ddfb6cSHuang Ying /* If target subpage in second half of huge page */ 5882c6ddfb6cSHuang Ying base = pages_per_huge_page - 2 * (pages_per_huge_page - n); 5883c6ddfb6cSHuang Ying l = pages_per_huge_page - n; 5884c6ddfb6cSHuang Ying /* Process subpages at the begin of huge page */ 5885c6ddfb6cSHuang Ying for (i = 0; i < base; i++) { 5886c6ddfb6cSHuang Ying cond_resched(); 58871cb9dc4bSLiu Shixin ret = process_subpage(addr + i * PAGE_SIZE, i, arg); 58881cb9dc4bSLiu Shixin if (ret) 58891cb9dc4bSLiu Shixin return ret; 5890c6ddfb6cSHuang Ying } 5891c6ddfb6cSHuang Ying } 5892c6ddfb6cSHuang Ying /* 5893c6ddfb6cSHuang Ying * Process remaining subpages in left-right-left-right pattern 5894c6ddfb6cSHuang Ying * towards the target subpage 5895c6ddfb6cSHuang Ying */ 5896c6ddfb6cSHuang Ying for (i = 0; i < l; i++) { 5897c6ddfb6cSHuang Ying int left_idx = base + i; 5898c6ddfb6cSHuang Ying int right_idx = base + 2 * l - 1 - i; 5899c6ddfb6cSHuang Ying 5900c6ddfb6cSHuang Ying cond_resched(); 59011cb9dc4bSLiu Shixin ret = process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg); 59021cb9dc4bSLiu Shixin if (ret) 59031cb9dc4bSLiu Shixin return ret; 5904c6ddfb6cSHuang Ying cond_resched(); 59051cb9dc4bSLiu Shixin ret = process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg); 59061cb9dc4bSLiu Shixin if (ret) 59071cb9dc4bSLiu Shixin return ret; 5908c6ddfb6cSHuang Ying } 59091cb9dc4bSLiu Shixin return 0; 5910c6ddfb6cSHuang Ying } 5911c6ddfb6cSHuang Ying 591247ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 591347ad8475SAndrea Arcangeli unsigned long addr, 591447ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 591547ad8475SAndrea Arcangeli { 591647ad8475SAndrea Arcangeli int i; 591714455eabSCheng Li struct page *p; 591847ad8475SAndrea Arcangeli 591947ad8475SAndrea Arcangeli might_sleep(); 592014455eabSCheng Li for (i = 0; i < pages_per_huge_page; i++) { 592114455eabSCheng Li p = nth_page(page, i); 592247ad8475SAndrea Arcangeli cond_resched(); 592347ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 592447ad8475SAndrea Arcangeli } 592547ad8475SAndrea Arcangeli } 5926c6ddfb6cSHuang Ying 59271cb9dc4bSLiu Shixin static int clear_subpage(unsigned long addr, int idx, void *arg) 5928c6ddfb6cSHuang Ying { 5929c6ddfb6cSHuang Ying struct page *page = arg; 5930c6ddfb6cSHuang Ying 5931c6ddfb6cSHuang Ying clear_user_highpage(page + idx, addr); 59321cb9dc4bSLiu Shixin return 0; 5933c6ddfb6cSHuang Ying } 5934c6ddfb6cSHuang Ying 593547ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 5936c79b57e4SHuang Ying unsigned long addr_hint, unsigned int pages_per_huge_page) 593747ad8475SAndrea Arcangeli { 5938c79b57e4SHuang Ying unsigned long addr = addr_hint & 5939c79b57e4SHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 594047ad8475SAndrea Arcangeli 594147ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 594247ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 594347ad8475SAndrea Arcangeli return; 594447ad8475SAndrea Arcangeli } 594547ad8475SAndrea Arcangeli 5946c6ddfb6cSHuang Ying process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page); 594747ad8475SAndrea Arcangeli } 594847ad8475SAndrea Arcangeli 59491cb9dc4bSLiu Shixin static int copy_user_gigantic_page(struct folio *dst, struct folio *src, 595047ad8475SAndrea Arcangeli unsigned long addr, 595147ad8475SAndrea Arcangeli struct vm_area_struct *vma, 595247ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 595347ad8475SAndrea Arcangeli { 595447ad8475SAndrea Arcangeli int i; 5955c0e8150eSZhangPeng struct page *dst_page; 5956c0e8150eSZhangPeng struct page *src_page; 595747ad8475SAndrea Arcangeli 595814455eabSCheng Li for (i = 0; i < pages_per_huge_page; i++) { 5959c0e8150eSZhangPeng dst_page = folio_page(dst, i); 5960c0e8150eSZhangPeng src_page = folio_page(src, i); 596114455eabSCheng Li 596247ad8475SAndrea Arcangeli cond_resched(); 59631cb9dc4bSLiu Shixin if (copy_mc_user_highpage(dst_page, src_page, 59641cb9dc4bSLiu Shixin addr + i*PAGE_SIZE, vma)) { 59651cb9dc4bSLiu Shixin memory_failure_queue(page_to_pfn(src_page), 0); 59661cb9dc4bSLiu Shixin return -EHWPOISON; 596747ad8475SAndrea Arcangeli } 596847ad8475SAndrea Arcangeli } 59691cb9dc4bSLiu Shixin return 0; 59701cb9dc4bSLiu Shixin } 597147ad8475SAndrea Arcangeli 5972c9f4cd71SHuang Ying struct copy_subpage_arg { 5973c9f4cd71SHuang Ying struct page *dst; 5974c9f4cd71SHuang Ying struct page *src; 5975c9f4cd71SHuang Ying struct vm_area_struct *vma; 5976c9f4cd71SHuang Ying }; 5977c9f4cd71SHuang Ying 59781cb9dc4bSLiu Shixin static int copy_subpage(unsigned long addr, int idx, void *arg) 5979c9f4cd71SHuang Ying { 5980c9f4cd71SHuang Ying struct copy_subpage_arg *copy_arg = arg; 5981c9f4cd71SHuang Ying 59821cb9dc4bSLiu Shixin if (copy_mc_user_highpage(copy_arg->dst + idx, copy_arg->src + idx, 59831cb9dc4bSLiu Shixin addr, copy_arg->vma)) { 59841cb9dc4bSLiu Shixin memory_failure_queue(page_to_pfn(copy_arg->src + idx), 0); 59851cb9dc4bSLiu Shixin return -EHWPOISON; 59861cb9dc4bSLiu Shixin } 59871cb9dc4bSLiu Shixin return 0; 5988c9f4cd71SHuang Ying } 5989c9f4cd71SHuang Ying 59901cb9dc4bSLiu Shixin int copy_user_large_folio(struct folio *dst, struct folio *src, 5991c0e8150eSZhangPeng unsigned long addr_hint, struct vm_area_struct *vma) 599247ad8475SAndrea Arcangeli { 5993c0e8150eSZhangPeng unsigned int pages_per_huge_page = folio_nr_pages(dst); 5994c9f4cd71SHuang Ying unsigned long addr = addr_hint & 5995c9f4cd71SHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 5996c9f4cd71SHuang Ying struct copy_subpage_arg arg = { 5997c0e8150eSZhangPeng .dst = &dst->page, 5998c0e8150eSZhangPeng .src = &src->page, 5999c9f4cd71SHuang Ying .vma = vma, 6000c9f4cd71SHuang Ying }; 600147ad8475SAndrea Arcangeli 60021cb9dc4bSLiu Shixin if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) 60031cb9dc4bSLiu Shixin return copy_user_gigantic_page(dst, src, addr, vma, 600447ad8475SAndrea Arcangeli pages_per_huge_page); 600547ad8475SAndrea Arcangeli 60061cb9dc4bSLiu Shixin return process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg); 600747ad8475SAndrea Arcangeli } 6008fa4d75c1SMike Kravetz 6009e87340caSZhangPeng long copy_folio_from_user(struct folio *dst_folio, 6010fa4d75c1SMike Kravetz const void __user *usr_src, 6011810a56b9SMike Kravetz bool allow_pagefault) 6012fa4d75c1SMike Kravetz { 6013e87340caSZhangPeng void *kaddr; 6014fa4d75c1SMike Kravetz unsigned long i, rc = 0; 6015e87340caSZhangPeng unsigned int nr_pages = folio_nr_pages(dst_folio); 6016e87340caSZhangPeng unsigned long ret_val = nr_pages * PAGE_SIZE; 601714455eabSCheng Li struct page *subpage; 6018fa4d75c1SMike Kravetz 6019e87340caSZhangPeng for (i = 0; i < nr_pages; i++) { 6020e87340caSZhangPeng subpage = folio_page(dst_folio, i); 6021e87340caSZhangPeng kaddr = kmap_local_page(subpage); 60220d508c1fSZhangPeng if (!allow_pagefault) 60230d508c1fSZhangPeng pagefault_disable(); 6024e87340caSZhangPeng rc = copy_from_user(kaddr, usr_src + i * PAGE_SIZE, PAGE_SIZE); 60250d508c1fSZhangPeng if (!allow_pagefault) 60260d508c1fSZhangPeng pagefault_enable(); 6027e87340caSZhangPeng kunmap_local(kaddr); 6028fa4d75c1SMike Kravetz 6029fa4d75c1SMike Kravetz ret_val -= (PAGE_SIZE - rc); 6030fa4d75c1SMike Kravetz if (rc) 6031fa4d75c1SMike Kravetz break; 6032fa4d75c1SMike Kravetz 6033e763243cSMuchun Song flush_dcache_page(subpage); 6034e763243cSMuchun Song 6035fa4d75c1SMike Kravetz cond_resched(); 6036fa4d75c1SMike Kravetz } 6037fa4d75c1SMike Kravetz return ret_val; 6038fa4d75c1SMike Kravetz } 603947ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 604049076ec2SKirill A. Shutemov 604140b64acdSOlof Johansson #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS 6042b35f1819SKirill A. Shutemov 6043b35f1819SKirill A. Shutemov static struct kmem_cache *page_ptl_cachep; 6044b35f1819SKirill A. Shutemov 6045b35f1819SKirill A. Shutemov void __init ptlock_cache_init(void) 6046b35f1819SKirill A. Shutemov { 6047b35f1819SKirill A. Shutemov page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, 6048b35f1819SKirill A. Shutemov SLAB_PANIC, NULL); 6049b35f1819SKirill A. Shutemov } 6050b35f1819SKirill A. Shutemov 6051539edb58SPeter Zijlstra bool ptlock_alloc(struct page *page) 605249076ec2SKirill A. Shutemov { 605349076ec2SKirill A. Shutemov spinlock_t *ptl; 605449076ec2SKirill A. Shutemov 6055b35f1819SKirill A. Shutemov ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); 605649076ec2SKirill A. Shutemov if (!ptl) 605749076ec2SKirill A. Shutemov return false; 6058539edb58SPeter Zijlstra page->ptl = ptl; 605949076ec2SKirill A. Shutemov return true; 606049076ec2SKirill A. Shutemov } 606149076ec2SKirill A. Shutemov 6062539edb58SPeter Zijlstra void ptlock_free(struct page *page) 606349076ec2SKirill A. Shutemov { 6064b35f1819SKirill A. Shutemov kmem_cache_free(page_ptl_cachep, page->ptl); 606549076ec2SKirill A. Shutemov } 606649076ec2SKirill A. Shutemov #endif 6067