1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/mm/memory.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 61da177e4SLinus Torvalds */ 71da177e4SLinus Torvalds 81da177e4SLinus Torvalds /* 91da177e4SLinus Torvalds * demand-loading started 01.12.91 - seems it is high on the list of 101da177e4SLinus Torvalds * things wanted, and it should be easy to implement. - Linus 111da177e4SLinus Torvalds */ 121da177e4SLinus Torvalds 131da177e4SLinus Torvalds /* 141da177e4SLinus Torvalds * Ok, demand-loading was easy, shared pages a little bit tricker. Shared 151da177e4SLinus Torvalds * pages started 02.12.91, seems to work. - Linus. 161da177e4SLinus Torvalds * 171da177e4SLinus Torvalds * Tested sharing by executing about 30 /bin/sh: under the old kernel it 181da177e4SLinus Torvalds * would have taken more than the 6M I have free, but it worked well as 191da177e4SLinus Torvalds * far as I could see. 201da177e4SLinus Torvalds * 211da177e4SLinus Torvalds * Also corrected some "invalidate()"s - I wasn't doing enough of them. 221da177e4SLinus Torvalds */ 231da177e4SLinus Torvalds 241da177e4SLinus Torvalds /* 251da177e4SLinus Torvalds * Real VM (paging to/from disk) started 18.12.91. Much more work and 261da177e4SLinus Torvalds * thought has to go into this. Oh, well.. 271da177e4SLinus Torvalds * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. 281da177e4SLinus Torvalds * Found it. Everything seems to work now. 291da177e4SLinus Torvalds * 20.12.91 - Ok, making the swap-device changeable like the root. 301da177e4SLinus Torvalds */ 311da177e4SLinus Torvalds 321da177e4SLinus Torvalds /* 331da177e4SLinus Torvalds * 05.04.94 - Multi-page memory management added for v1.1. 341da177e4SLinus Torvalds * Idea by Alex Bligh (alex@cconcepts.co.uk) 351da177e4SLinus Torvalds * 361da177e4SLinus Torvalds * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG 371da177e4SLinus Torvalds * (Gerhard.Wichert@pdb.siemens.de) 381da177e4SLinus Torvalds * 391da177e4SLinus Torvalds * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) 401da177e4SLinus Torvalds */ 411da177e4SLinus Torvalds 421da177e4SLinus Torvalds #include <linux/kernel_stat.h> 431da177e4SLinus Torvalds #include <linux/mm.h> 4436090defSArnd Bergmann #include <linux/mm_inline.h> 456e84f315SIngo Molnar #include <linux/sched/mm.h> 46f7ccbae4SIngo Molnar #include <linux/sched/coredump.h> 476a3827d7SIngo Molnar #include <linux/sched/numa_balancing.h> 4829930025SIngo Molnar #include <linux/sched/task.h> 491da177e4SLinus Torvalds #include <linux/hugetlb.h> 501da177e4SLinus Torvalds #include <linux/mman.h> 511da177e4SLinus Torvalds #include <linux/swap.h> 521da177e4SLinus Torvalds #include <linux/highmem.h> 531da177e4SLinus Torvalds #include <linux/pagemap.h> 545042db43SJérôme Glisse #include <linux/memremap.h> 55b073d7f8SAlexander Potapenko #include <linux/kmsan.h> 569a840895SHugh Dickins #include <linux/ksm.h> 571da177e4SLinus Torvalds #include <linux/rmap.h> 58b95f1b31SPaul Gortmaker #include <linux/export.h> 590ff92245SShailabh Nagar #include <linux/delayacct.h> 601da177e4SLinus Torvalds #include <linux/init.h> 6101c8f1c4SDan Williams #include <linux/pfn_t.h> 62edc79b2aSPeter Zijlstra #include <linux/writeback.h> 638a9f3ccdSBalbir Singh #include <linux/memcontrol.h> 64cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 653dc14741SHugh Dickins #include <linux/swapops.h> 663dc14741SHugh Dickins #include <linux/elf.h> 675a0e3ad6STejun Heo #include <linux/gfp.h> 684daae3b4SMel Gorman #include <linux/migrate.h> 692fbc57c5SAndy Shevchenko #include <linux/string.h> 70467b171aSAneesh Kumar K.V #include <linux/memory-tiers.h> 711592eef0SKirill A. Shutemov #include <linux/debugfs.h> 726b251fc9SAndrea Arcangeli #include <linux/userfaultfd_k.h> 73bc2466e4SJan Kara #include <linux/dax.h> 746b31d595SMichal Hocko #include <linux/oom.h> 7598fa15f3SAnshuman Khandual #include <linux/numa.h> 76bce617edSPeter Xu #include <linux/perf_event.h> 77bce617edSPeter Xu #include <linux/ptrace.h> 78e80d3909SJoerg Roedel #include <linux/vmalloc.h> 7933024536SHuang Ying #include <linux/sched/sysctl.h> 801da177e4SLinus Torvalds 81b3d1411bSJoel Fernandes (Google) #include <trace/events/kmem.h> 82b3d1411bSJoel Fernandes (Google) 836952b61dSAlexey Dobriyan #include <asm/io.h> 8433a709b2SDave Hansen #include <asm/mmu_context.h> 851da177e4SLinus Torvalds #include <asm/pgalloc.h> 867c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 871da177e4SLinus Torvalds #include <asm/tlb.h> 881da177e4SLinus Torvalds #include <asm/tlbflush.h> 891da177e4SLinus Torvalds 90e80d3909SJoerg Roedel #include "pgalloc-track.h" 9142b77728SJan Beulich #include "internal.h" 92014bb1deSNeilBrown #include "swap.h" 9342b77728SJan Beulich 94af27d940SArnd Bergmann #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST) 9590572890SPeter Zijlstra #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. 9675980e97SPeter Zijlstra #endif 9775980e97SPeter Zijlstra 98a9ee6cf5SMike Rapoport #ifndef CONFIG_NUMA 991da177e4SLinus Torvalds unsigned long max_mapnr; 1001da177e4SLinus Torvalds EXPORT_SYMBOL(max_mapnr); 101166f61b9STobin C Harding 102166f61b9STobin C Harding struct page *mem_map; 1031da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 1041da177e4SLinus Torvalds #endif 1051da177e4SLinus Torvalds 1065c041f5dSPeter Xu static vm_fault_t do_fault(struct vm_fault *vmf); 1072bad466cSPeter Xu static vm_fault_t do_anonymous_page(struct vm_fault *vmf); 1082bad466cSPeter Xu static bool vmf_pte_changed(struct vm_fault *vmf); 1092bad466cSPeter Xu 1102bad466cSPeter Xu /* 1112bad466cSPeter Xu * Return true if the original pte was a uffd-wp pte marker (so the pte was 1122bad466cSPeter Xu * wr-protected). 1132bad466cSPeter Xu */ 1142bad466cSPeter Xu static bool vmf_orig_pte_uffd_wp(struct vm_fault *vmf) 1152bad466cSPeter Xu { 1162bad466cSPeter Xu if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID)) 1172bad466cSPeter Xu return false; 1182bad466cSPeter Xu 1192bad466cSPeter Xu return pte_marker_uffd_wp(vmf->orig_pte); 1202bad466cSPeter Xu } 1215c041f5dSPeter Xu 1221da177e4SLinus Torvalds /* 1231da177e4SLinus Torvalds * A number of key systems in x86 including ioremap() rely on the assumption 1241da177e4SLinus Torvalds * that high_memory defines the upper bound on direct map memory, then end 1251da177e4SLinus Torvalds * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and 1261da177e4SLinus Torvalds * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL 1271da177e4SLinus Torvalds * and ZONE_HIGHMEM. 1281da177e4SLinus Torvalds */ 1291da177e4SLinus Torvalds void *high_memory; 1301da177e4SLinus Torvalds EXPORT_SYMBOL(high_memory); 1311da177e4SLinus Torvalds 13232a93233SIngo Molnar /* 13332a93233SIngo Molnar * Randomize the address space (stacks, mmaps, brk, etc.). 13432a93233SIngo Molnar * 13532a93233SIngo Molnar * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, 13632a93233SIngo Molnar * as ancient (libc5 based) binaries can segfault. ) 13732a93233SIngo Molnar */ 13832a93233SIngo Molnar int randomize_va_space __read_mostly = 13932a93233SIngo Molnar #ifdef CONFIG_COMPAT_BRK 14032a93233SIngo Molnar 1; 14132a93233SIngo Molnar #else 14232a93233SIngo Molnar 2; 14332a93233SIngo Molnar #endif 144a62eaf15SAndi Kleen 14546bdb427SWill Deacon #ifndef arch_wants_old_prefaulted_pte 14646bdb427SWill Deacon static inline bool arch_wants_old_prefaulted_pte(void) 14746bdb427SWill Deacon { 14846bdb427SWill Deacon /* 14946bdb427SWill Deacon * Transitioning a PTE from 'old' to 'young' can be expensive on 15046bdb427SWill Deacon * some architectures, even if it's performed in hardware. By 15146bdb427SWill Deacon * default, "false" means prefaulted entries will be 'young'. 15246bdb427SWill Deacon */ 15346bdb427SWill Deacon return false; 15446bdb427SWill Deacon } 15546bdb427SWill Deacon #endif 15646bdb427SWill Deacon 157a62eaf15SAndi Kleen static int __init disable_randmaps(char *s) 158a62eaf15SAndi Kleen { 159a62eaf15SAndi Kleen randomize_va_space = 0; 1609b41046cSOGAWA Hirofumi return 1; 161a62eaf15SAndi Kleen } 162a62eaf15SAndi Kleen __setup("norandmaps", disable_randmaps); 163a62eaf15SAndi Kleen 16462eede62SHugh Dickins unsigned long zero_pfn __read_mostly; 1650b70068eSArd Biesheuvel EXPORT_SYMBOL(zero_pfn); 1660b70068eSArd Biesheuvel 167166f61b9STobin C Harding unsigned long highest_memmap_pfn __read_mostly; 168166f61b9STobin C Harding 169a13ea5b7SHugh Dickins /* 170a13ea5b7SHugh Dickins * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() 171a13ea5b7SHugh Dickins */ 172a13ea5b7SHugh Dickins static int __init init_zero_pfn(void) 173a13ea5b7SHugh Dickins { 174a13ea5b7SHugh Dickins zero_pfn = page_to_pfn(ZERO_PAGE(0)); 175a13ea5b7SHugh Dickins return 0; 176a13ea5b7SHugh Dickins } 177e720e7d0SIlya Lipnitskiy early_initcall(init_zero_pfn); 178a62eaf15SAndi Kleen 179f1a79412SShakeel Butt void mm_trace_rss_stat(struct mm_struct *mm, int member) 180b3d1411bSJoel Fernandes (Google) { 181f1a79412SShakeel Butt trace_rss_stat(mm, member); 182b3d1411bSJoel Fernandes (Google) } 183d559db08SKAMEZAWA Hiroyuki 1841da177e4SLinus Torvalds /* 1851da177e4SLinus Torvalds * Note: this doesn't free the actual pages themselves. That 1861da177e4SLinus Torvalds * has been handled earlier when unmapping all the memory regions. 1871da177e4SLinus Torvalds */ 1889e1b32caSBenjamin Herrenschmidt static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, 1899e1b32caSBenjamin Herrenschmidt unsigned long addr) 1901da177e4SLinus Torvalds { 1912f569afdSMartin Schwidefsky pgtable_t token = pmd_pgtable(*pmd); 1921da177e4SLinus Torvalds pmd_clear(pmd); 1939e1b32caSBenjamin Herrenschmidt pte_free_tlb(tlb, token, addr); 194c4812909SKirill A. Shutemov mm_dec_nr_ptes(tlb->mm); 1951da177e4SLinus Torvalds } 1961da177e4SLinus Torvalds 197e0da382cSHugh Dickins static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, 198e0da382cSHugh Dickins unsigned long addr, unsigned long end, 199e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 2001da177e4SLinus Torvalds { 2011da177e4SLinus Torvalds pmd_t *pmd; 2021da177e4SLinus Torvalds unsigned long next; 203e0da382cSHugh Dickins unsigned long start; 2041da177e4SLinus Torvalds 205e0da382cSHugh Dickins start = addr; 2061da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 2071da177e4SLinus Torvalds do { 2081da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 2091da177e4SLinus Torvalds if (pmd_none_or_clear_bad(pmd)) 2101da177e4SLinus Torvalds continue; 2119e1b32caSBenjamin Herrenschmidt free_pte_range(tlb, pmd, addr); 2121da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 2131da177e4SLinus Torvalds 214e0da382cSHugh Dickins start &= PUD_MASK; 215e0da382cSHugh Dickins if (start < floor) 216e0da382cSHugh Dickins return; 217e0da382cSHugh Dickins if (ceiling) { 218e0da382cSHugh Dickins ceiling &= PUD_MASK; 219e0da382cSHugh Dickins if (!ceiling) 220e0da382cSHugh Dickins return; 2211da177e4SLinus Torvalds } 222e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 223e0da382cSHugh Dickins return; 224e0da382cSHugh Dickins 225e0da382cSHugh Dickins pmd = pmd_offset(pud, start); 226e0da382cSHugh Dickins pud_clear(pud); 2279e1b32caSBenjamin Herrenschmidt pmd_free_tlb(tlb, pmd, start); 228dc6c9a35SKirill A. Shutemov mm_dec_nr_pmds(tlb->mm); 2291da177e4SLinus Torvalds } 2301da177e4SLinus Torvalds 231c2febafcSKirill A. Shutemov static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, 232e0da382cSHugh Dickins unsigned long addr, unsigned long end, 233e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 2341da177e4SLinus Torvalds { 2351da177e4SLinus Torvalds pud_t *pud; 2361da177e4SLinus Torvalds unsigned long next; 237e0da382cSHugh Dickins unsigned long start; 2381da177e4SLinus Torvalds 239e0da382cSHugh Dickins start = addr; 240c2febafcSKirill A. Shutemov pud = pud_offset(p4d, addr); 2411da177e4SLinus Torvalds do { 2421da177e4SLinus Torvalds next = pud_addr_end(addr, end); 2431da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 2441da177e4SLinus Torvalds continue; 245e0da382cSHugh Dickins free_pmd_range(tlb, pud, addr, next, floor, ceiling); 2461da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 2471da177e4SLinus Torvalds 248c2febafcSKirill A. Shutemov start &= P4D_MASK; 249c2febafcSKirill A. Shutemov if (start < floor) 250c2febafcSKirill A. Shutemov return; 251c2febafcSKirill A. Shutemov if (ceiling) { 252c2febafcSKirill A. Shutemov ceiling &= P4D_MASK; 253c2febafcSKirill A. Shutemov if (!ceiling) 254c2febafcSKirill A. Shutemov return; 255c2febafcSKirill A. Shutemov } 256c2febafcSKirill A. Shutemov if (end - 1 > ceiling - 1) 257c2febafcSKirill A. Shutemov return; 258c2febafcSKirill A. Shutemov 259c2febafcSKirill A. Shutemov pud = pud_offset(p4d, start); 260c2febafcSKirill A. Shutemov p4d_clear(p4d); 261c2febafcSKirill A. Shutemov pud_free_tlb(tlb, pud, start); 262b4e98d9aSKirill A. Shutemov mm_dec_nr_puds(tlb->mm); 263c2febafcSKirill A. Shutemov } 264c2febafcSKirill A. Shutemov 265c2febafcSKirill A. Shutemov static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, 266c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 267c2febafcSKirill A. Shutemov unsigned long floor, unsigned long ceiling) 268c2febafcSKirill A. Shutemov { 269c2febafcSKirill A. Shutemov p4d_t *p4d; 270c2febafcSKirill A. Shutemov unsigned long next; 271c2febafcSKirill A. Shutemov unsigned long start; 272c2febafcSKirill A. Shutemov 273c2febafcSKirill A. Shutemov start = addr; 274c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, addr); 275c2febafcSKirill A. Shutemov do { 276c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 277c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(p4d)) 278c2febafcSKirill A. Shutemov continue; 279c2febafcSKirill A. Shutemov free_pud_range(tlb, p4d, addr, next, floor, ceiling); 280c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 281c2febafcSKirill A. Shutemov 282e0da382cSHugh Dickins start &= PGDIR_MASK; 283e0da382cSHugh Dickins if (start < floor) 284e0da382cSHugh Dickins return; 285e0da382cSHugh Dickins if (ceiling) { 286e0da382cSHugh Dickins ceiling &= PGDIR_MASK; 287e0da382cSHugh Dickins if (!ceiling) 288e0da382cSHugh Dickins return; 2891da177e4SLinus Torvalds } 290e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 291e0da382cSHugh Dickins return; 292e0da382cSHugh Dickins 293c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, start); 294e0da382cSHugh Dickins pgd_clear(pgd); 295c2febafcSKirill A. Shutemov p4d_free_tlb(tlb, p4d, start); 2961da177e4SLinus Torvalds } 2971da177e4SLinus Torvalds 2981da177e4SLinus Torvalds /* 299e0da382cSHugh Dickins * This function frees user-level page tables of a process. 3001da177e4SLinus Torvalds */ 30142b77728SJan Beulich void free_pgd_range(struct mmu_gather *tlb, 302e0da382cSHugh Dickins unsigned long addr, unsigned long end, 303e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 3041da177e4SLinus Torvalds { 3051da177e4SLinus Torvalds pgd_t *pgd; 3061da177e4SLinus Torvalds unsigned long next; 3071da177e4SLinus Torvalds 308e0da382cSHugh Dickins /* 309e0da382cSHugh Dickins * The next few lines have given us lots of grief... 310e0da382cSHugh Dickins * 311e0da382cSHugh Dickins * Why are we testing PMD* at this top level? Because often 312e0da382cSHugh Dickins * there will be no work to do at all, and we'd prefer not to 313e0da382cSHugh Dickins * go all the way down to the bottom just to discover that. 314e0da382cSHugh Dickins * 315e0da382cSHugh Dickins * Why all these "- 1"s? Because 0 represents both the bottom 316e0da382cSHugh Dickins * of the address space and the top of it (using -1 for the 317e0da382cSHugh Dickins * top wouldn't help much: the masks would do the wrong thing). 318e0da382cSHugh Dickins * The rule is that addr 0 and floor 0 refer to the bottom of 319e0da382cSHugh Dickins * the address space, but end 0 and ceiling 0 refer to the top 320e0da382cSHugh Dickins * Comparisons need to use "end - 1" and "ceiling - 1" (though 321e0da382cSHugh Dickins * that end 0 case should be mythical). 322e0da382cSHugh Dickins * 323e0da382cSHugh Dickins * Wherever addr is brought up or ceiling brought down, we must 324e0da382cSHugh Dickins * be careful to reject "the opposite 0" before it confuses the 325e0da382cSHugh Dickins * subsequent tests. But what about where end is brought down 326e0da382cSHugh Dickins * by PMD_SIZE below? no, end can't go down to 0 there. 327e0da382cSHugh Dickins * 328e0da382cSHugh Dickins * Whereas we round start (addr) and ceiling down, by different 329e0da382cSHugh Dickins * masks at different levels, in order to test whether a table 330e0da382cSHugh Dickins * now has no other vmas using it, so can be freed, we don't 331e0da382cSHugh Dickins * bother to round floor or end up - the tests don't need that. 332e0da382cSHugh Dickins */ 333e0da382cSHugh Dickins 334e0da382cSHugh Dickins addr &= PMD_MASK; 335e0da382cSHugh Dickins if (addr < floor) { 336e0da382cSHugh Dickins addr += PMD_SIZE; 337e0da382cSHugh Dickins if (!addr) 338e0da382cSHugh Dickins return; 339e0da382cSHugh Dickins } 340e0da382cSHugh Dickins if (ceiling) { 341e0da382cSHugh Dickins ceiling &= PMD_MASK; 342e0da382cSHugh Dickins if (!ceiling) 343e0da382cSHugh Dickins return; 344e0da382cSHugh Dickins } 345e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 346e0da382cSHugh Dickins end -= PMD_SIZE; 347e0da382cSHugh Dickins if (addr > end - 1) 348e0da382cSHugh Dickins return; 34907e32661SAneesh Kumar K.V /* 35007e32661SAneesh Kumar K.V * We add page table cache pages with PAGE_SIZE, 35107e32661SAneesh Kumar K.V * (see pte_free_tlb()), flush the tlb if we need 35207e32661SAneesh Kumar K.V */ 353ed6a7935SPeter Zijlstra tlb_change_page_size(tlb, PAGE_SIZE); 35442b77728SJan Beulich pgd = pgd_offset(tlb->mm, addr); 3551da177e4SLinus Torvalds do { 3561da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 3571da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 3581da177e4SLinus Torvalds continue; 359c2febafcSKirill A. Shutemov free_p4d_range(tlb, pgd, addr, next, floor, ceiling); 3601da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 361e0da382cSHugh Dickins } 362e0da382cSHugh Dickins 363fd892593SLiam R. Howlett void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas, 364763ecb03SLiam R. Howlett struct vm_area_struct *vma, unsigned long floor, 36598e51a22SSuren Baghdasaryan unsigned long ceiling, bool mm_wr_locked) 366e0da382cSHugh Dickins { 367763ecb03SLiam R. Howlett do { 368e0da382cSHugh Dickins unsigned long addr = vma->vm_start; 369763ecb03SLiam R. Howlett struct vm_area_struct *next; 370763ecb03SLiam R. Howlett 371763ecb03SLiam R. Howlett /* 372763ecb03SLiam R. Howlett * Note: USER_PGTABLES_CEILING may be passed as ceiling and may 373763ecb03SLiam R. Howlett * be 0. This will underflow and is okay. 374763ecb03SLiam R. Howlett */ 375fd892593SLiam R. Howlett next = mas_find(mas, ceiling - 1); 376e0da382cSHugh Dickins 3778f4f8c16SHugh Dickins /* 37825d9e2d1Snpiggin@suse.de * Hide vma from rmap and truncate_pagecache before freeing 37925d9e2d1Snpiggin@suse.de * pgtables 3808f4f8c16SHugh Dickins */ 38198e51a22SSuren Baghdasaryan if (mm_wr_locked) 38298e51a22SSuren Baghdasaryan vma_start_write(vma); 3835beb4930SRik van Riel unlink_anon_vmas(vma); 3848f4f8c16SHugh Dickins unlink_file_vma(vma); 3858f4f8c16SHugh Dickins 3869da61aefSDavid Gibson if (is_vm_hugetlb_page(vma)) { 3873bf5ee95SHugh Dickins hugetlb_free_pgd_range(tlb, addr, vma->vm_end, 3883bf5ee95SHugh Dickins floor, next ? next->vm_start : ceiling); 3893bf5ee95SHugh Dickins } else { 3903bf5ee95SHugh Dickins /* 3913bf5ee95SHugh Dickins * Optimization: gather nearby vmas into one call down 3923bf5ee95SHugh Dickins */ 3933bf5ee95SHugh Dickins while (next && next->vm_start <= vma->vm_end + PMD_SIZE 3944866920bSDavid Gibson && !is_vm_hugetlb_page(next)) { 395e0da382cSHugh Dickins vma = next; 396fd892593SLiam R. Howlett next = mas_find(mas, ceiling - 1); 39798e51a22SSuren Baghdasaryan if (mm_wr_locked) 39898e51a22SSuren Baghdasaryan vma_start_write(vma); 3995beb4930SRik van Riel unlink_anon_vmas(vma); 4008f4f8c16SHugh Dickins unlink_file_vma(vma); 401e0da382cSHugh Dickins } 4023bf5ee95SHugh Dickins free_pgd_range(tlb, addr, vma->vm_end, 403e0da382cSHugh Dickins floor, next ? next->vm_start : ceiling); 4043bf5ee95SHugh Dickins } 405e0da382cSHugh Dickins vma = next; 406763ecb03SLiam R. Howlett } while (vma); 4071da177e4SLinus Torvalds } 4081da177e4SLinus Torvalds 40903c4f204SQi Zheng void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte) 4101da177e4SLinus Torvalds { 41103c4f204SQi Zheng spinlock_t *ptl = pmd_lock(mm, pmd); 4121bb3630eSHugh Dickins 41303c4f204SQi Zheng if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 41403c4f204SQi Zheng mm_inc_nr_ptes(mm); 415362a61adSNick Piggin /* 416362a61adSNick Piggin * Ensure all pte setup (eg. pte page lock and page clearing) are 417362a61adSNick Piggin * visible before the pte is made visible to other CPUs by being 418362a61adSNick Piggin * put into page tables. 419362a61adSNick Piggin * 420362a61adSNick Piggin * The other side of the story is the pointer chasing in the page 421362a61adSNick Piggin * table walking code (when walking the page table without locking; 422362a61adSNick Piggin * ie. most of the time). Fortunately, these data accesses consist 423362a61adSNick Piggin * of a chain of data-dependent loads, meaning most CPUs (alpha 424362a61adSNick Piggin * being the notable exception) will already guarantee loads are 425362a61adSNick Piggin * seen in-order. See the alpha page table accessors for the 426bb7cdd38SWill Deacon * smp_rmb() barriers in page table walking code. 427362a61adSNick Piggin */ 428362a61adSNick Piggin smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ 429ed33b5a6SQi Zheng pmd_populate(mm, pmd, *pte); 430ed33b5a6SQi Zheng *pte = NULL; 4314b471e88SKirill A. Shutemov } 432c4088ebdSKirill A. Shutemov spin_unlock(ptl); 433ed33b5a6SQi Zheng } 434ed33b5a6SQi Zheng 435ed33b5a6SQi Zheng int __pte_alloc(struct mm_struct *mm, pmd_t *pmd) 436ed33b5a6SQi Zheng { 437ed33b5a6SQi Zheng pgtable_t new = pte_alloc_one(mm); 438ed33b5a6SQi Zheng if (!new) 439ed33b5a6SQi Zheng return -ENOMEM; 4408ac1f832SAndrea Arcangeli 44103c4f204SQi Zheng pmd_install(mm, pmd, &new); 4422f569afdSMartin Schwidefsky if (new) 4432f569afdSMartin Schwidefsky pte_free(mm, new); 4441bb3630eSHugh Dickins return 0; 4451da177e4SLinus Torvalds } 4461da177e4SLinus Torvalds 4474cf58924SJoel Fernandes (Google) int __pte_alloc_kernel(pmd_t *pmd) 4481da177e4SLinus Torvalds { 4494cf58924SJoel Fernandes (Google) pte_t *new = pte_alloc_one_kernel(&init_mm); 4501da177e4SLinus Torvalds if (!new) 4511bb3630eSHugh Dickins return -ENOMEM; 4521da177e4SLinus Torvalds 453872fec16SHugh Dickins spin_lock(&init_mm.page_table_lock); 4548ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 455ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 456872fec16SHugh Dickins pmd_populate_kernel(&init_mm, pmd, new); 4572f569afdSMartin Schwidefsky new = NULL; 4584b471e88SKirill A. Shutemov } 459872fec16SHugh Dickins spin_unlock(&init_mm.page_table_lock); 4602f569afdSMartin Schwidefsky if (new) 4612f569afdSMartin Schwidefsky pte_free_kernel(&init_mm, new); 4621bb3630eSHugh Dickins return 0; 4631da177e4SLinus Torvalds } 4641da177e4SLinus Torvalds 465d559db08SKAMEZAWA Hiroyuki static inline void init_rss_vec(int *rss) 466ae859762SHugh Dickins { 467d559db08SKAMEZAWA Hiroyuki memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); 468d559db08SKAMEZAWA Hiroyuki } 469d559db08SKAMEZAWA Hiroyuki 470d559db08SKAMEZAWA Hiroyuki static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) 471d559db08SKAMEZAWA Hiroyuki { 472d559db08SKAMEZAWA Hiroyuki int i; 473d559db08SKAMEZAWA Hiroyuki 47434e55232SKAMEZAWA Hiroyuki if (current->mm == mm) 47505af2e10SDavid Rientjes sync_mm_rss(mm); 476d559db08SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) 477d559db08SKAMEZAWA Hiroyuki if (rss[i]) 478d559db08SKAMEZAWA Hiroyuki add_mm_counter(mm, i, rss[i]); 479ae859762SHugh Dickins } 480ae859762SHugh Dickins 4811da177e4SLinus Torvalds /* 4826aab341eSLinus Torvalds * This function is called to print an error when a bad pte 4836aab341eSLinus Torvalds * is found. For example, we might have a PFN-mapped pte in 4846aab341eSLinus Torvalds * a region that doesn't allow it. 485b5810039SNick Piggin * 486b5810039SNick Piggin * The calling function must still handle the error. 487b5810039SNick Piggin */ 4883dc14741SHugh Dickins static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, 4893dc14741SHugh Dickins pte_t pte, struct page *page) 490b5810039SNick Piggin { 4913dc14741SHugh Dickins pgd_t *pgd = pgd_offset(vma->vm_mm, addr); 492c2febafcSKirill A. Shutemov p4d_t *p4d = p4d_offset(pgd, addr); 493c2febafcSKirill A. Shutemov pud_t *pud = pud_offset(p4d, addr); 4943dc14741SHugh Dickins pmd_t *pmd = pmd_offset(pud, addr); 4953dc14741SHugh Dickins struct address_space *mapping; 4963dc14741SHugh Dickins pgoff_t index; 497d936cf9bSHugh Dickins static unsigned long resume; 498d936cf9bSHugh Dickins static unsigned long nr_shown; 499d936cf9bSHugh Dickins static unsigned long nr_unshown; 500d936cf9bSHugh Dickins 501d936cf9bSHugh Dickins /* 502d936cf9bSHugh Dickins * Allow a burst of 60 reports, then keep quiet for that minute; 503d936cf9bSHugh Dickins * or allow a steady drip of one report per second. 504d936cf9bSHugh Dickins */ 505d936cf9bSHugh Dickins if (nr_shown == 60) { 506d936cf9bSHugh Dickins if (time_before(jiffies, resume)) { 507d936cf9bSHugh Dickins nr_unshown++; 508d936cf9bSHugh Dickins return; 509d936cf9bSHugh Dickins } 510d936cf9bSHugh Dickins if (nr_unshown) { 5111170532bSJoe Perches pr_alert("BUG: Bad page map: %lu messages suppressed\n", 512d936cf9bSHugh Dickins nr_unshown); 513d936cf9bSHugh Dickins nr_unshown = 0; 514d936cf9bSHugh Dickins } 515d936cf9bSHugh Dickins nr_shown = 0; 516d936cf9bSHugh Dickins } 517d936cf9bSHugh Dickins if (nr_shown++ == 0) 518d936cf9bSHugh Dickins resume = jiffies + 60 * HZ; 5193dc14741SHugh Dickins 5203dc14741SHugh Dickins mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; 5213dc14741SHugh Dickins index = linear_page_index(vma, addr); 5223dc14741SHugh Dickins 5231170532bSJoe Perches pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", 5243dc14741SHugh Dickins current->comm, 5253dc14741SHugh Dickins (long long)pte_val(pte), (long long)pmd_val(*pmd)); 526718a3821SWu Fengguang if (page) 527f0b791a3SDave Hansen dump_page(page, "bad pte"); 5286aa9b8b2SKefeng Wang pr_alert("addr:%px vm_flags:%08lx anon_vma:%px mapping:%px index:%lx\n", 5293dc14741SHugh Dickins (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); 5307e0a1265SMatthew Wilcox (Oracle) pr_alert("file:%pD fault:%ps mmap:%ps read_folio:%ps\n", 5312682582aSKonstantin Khlebnikov vma->vm_file, 5322682582aSKonstantin Khlebnikov vma->vm_ops ? vma->vm_ops->fault : NULL, 5332682582aSKonstantin Khlebnikov vma->vm_file ? vma->vm_file->f_op->mmap : NULL, 5347e0a1265SMatthew Wilcox (Oracle) mapping ? mapping->a_ops->read_folio : NULL); 535b5810039SNick Piggin dump_stack(); 536373d4d09SRusty Russell add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); 537b5810039SNick Piggin } 538b5810039SNick Piggin 539b5810039SNick Piggin /* 5407e675137SNick Piggin * vm_normal_page -- This function gets the "struct page" associated with a pte. 5416aab341eSLinus Torvalds * 5427e675137SNick Piggin * "Special" mappings do not wish to be associated with a "struct page" (either 5437e675137SNick Piggin * it doesn't exist, or it exists but they don't want to touch it). In this 5447e675137SNick Piggin * case, NULL is returned here. "Normal" mappings do have a struct page. 545b379d790SJared Hulbert * 5467e675137SNick Piggin * There are 2 broad cases. Firstly, an architecture may define a pte_special() 5477e675137SNick Piggin * pte bit, in which case this function is trivial. Secondly, an architecture 5487e675137SNick Piggin * may not have a spare pte bit, which requires a more complicated scheme, 5497e675137SNick Piggin * described below. 5507e675137SNick Piggin * 5517e675137SNick Piggin * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a 5527e675137SNick Piggin * special mapping (even if there are underlying and valid "struct pages"). 5537e675137SNick Piggin * COWed pages of a VM_PFNMAP are always normal. 5546aab341eSLinus Torvalds * 555b379d790SJared Hulbert * The way we recognize COWed pages within VM_PFNMAP mappings is through the 556b379d790SJared Hulbert * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit 5577e675137SNick Piggin * set, and the vm_pgoff will point to the first PFN mapped: thus every special 5587e675137SNick Piggin * mapping will always honor the rule 5596aab341eSLinus Torvalds * 5606aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 5616aab341eSLinus Torvalds * 5627e675137SNick Piggin * And for normal mappings this is false. 563b379d790SJared Hulbert * 5647e675137SNick Piggin * This restricts such mappings to be a linear translation from virtual address 5657e675137SNick Piggin * to pfn. To get around this restriction, we allow arbitrary mappings so long 5667e675137SNick Piggin * as the vma is not a COW mapping; in that case, we know that all ptes are 5677e675137SNick Piggin * special (because none can have been COWed). 568b379d790SJared Hulbert * 569b379d790SJared Hulbert * 5707e675137SNick Piggin * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. 5717e675137SNick Piggin * 572b379d790SJared Hulbert * VM_MIXEDMAP mappings can likewise contain memory with or without "struct 573b379d790SJared Hulbert * page" backing, however the difference is that _all_ pages with a struct 574b379d790SJared Hulbert * page (that is, those where pfn_valid is true) are refcounted and considered 575b379d790SJared Hulbert * normal pages by the VM. The disadvantage is that pages are refcounted 576b379d790SJared Hulbert * (which can be slower and simply not an option for some PFNMAP users). The 577b379d790SJared Hulbert * advantage is that we don't have to follow the strict linearity rule of 578b379d790SJared Hulbert * PFNMAP mappings in order to support COWable mappings. 579b379d790SJared Hulbert * 580ee498ed7SHugh Dickins */ 58125b2995aSChristoph Hellwig struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 58225b2995aSChristoph Hellwig pte_t pte) 583ee498ed7SHugh Dickins { 58422b31eecSHugh Dickins unsigned long pfn = pte_pfn(pte); 5857e675137SNick Piggin 58600b3a331SLaurent Dufour if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) { 587b38af472SHugh Dickins if (likely(!pte_special(pte))) 58822b31eecSHugh Dickins goto check_pfn; 589667a0a06SDavid Vrabel if (vma->vm_ops && vma->vm_ops->find_special_page) 590667a0a06SDavid Vrabel return vma->vm_ops->find_special_page(vma, addr); 591a13ea5b7SHugh Dickins if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 592a13ea5b7SHugh Dickins return NULL; 593df6ad698SJérôme Glisse if (is_zero_pfn(pfn)) 594df6ad698SJérôme Glisse return NULL; 595e1fb4a08SDave Jiang if (pte_devmap(pte)) 5963218f871SAlex Sierra /* 5973218f871SAlex Sierra * NOTE: New users of ZONE_DEVICE will not set pte_devmap() 5983218f871SAlex Sierra * and will have refcounts incremented on their struct pages 5993218f871SAlex Sierra * when they are inserted into PTEs, thus they are safe to 6003218f871SAlex Sierra * return here. Legacy ZONE_DEVICE pages that set pte_devmap() 6013218f871SAlex Sierra * do not have refcounts. Example of legacy ZONE_DEVICE is 6023218f871SAlex Sierra * MEMORY_DEVICE_FS_DAX type in pmem or virtio_fs drivers. 6033218f871SAlex Sierra */ 604e1fb4a08SDave Jiang return NULL; 605e1fb4a08SDave Jiang 60622b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 6077e675137SNick Piggin return NULL; 6087e675137SNick Piggin } 6097e675137SNick Piggin 61000b3a331SLaurent Dufour /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */ 6117e675137SNick Piggin 612b379d790SJared Hulbert if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 613b379d790SJared Hulbert if (vma->vm_flags & VM_MIXEDMAP) { 614b379d790SJared Hulbert if (!pfn_valid(pfn)) 615b379d790SJared Hulbert return NULL; 616b379d790SJared Hulbert goto out; 617b379d790SJared Hulbert } else { 6187e675137SNick Piggin unsigned long off; 6197e675137SNick Piggin off = (addr - vma->vm_start) >> PAGE_SHIFT; 6206aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 6216aab341eSLinus Torvalds return NULL; 62267121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 623fb155c16SLinus Torvalds return NULL; 6246aab341eSLinus Torvalds } 625b379d790SJared Hulbert } 6266aab341eSLinus Torvalds 627b38af472SHugh Dickins if (is_zero_pfn(pfn)) 628b38af472SHugh Dickins return NULL; 62900b3a331SLaurent Dufour 63022b31eecSHugh Dickins check_pfn: 63122b31eecSHugh Dickins if (unlikely(pfn > highest_memmap_pfn)) { 63222b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 63322b31eecSHugh Dickins return NULL; 63422b31eecSHugh Dickins } 6356aab341eSLinus Torvalds 6366aab341eSLinus Torvalds /* 6377e675137SNick Piggin * NOTE! We still have PageReserved() pages in the page tables. 6387e675137SNick Piggin * eg. VDSO mappings can cause them to exist. 6396aab341eSLinus Torvalds */ 640b379d790SJared Hulbert out: 6416aab341eSLinus Torvalds return pfn_to_page(pfn); 642ee498ed7SHugh Dickins } 643ee498ed7SHugh Dickins 644318e9342SVishal Moola (Oracle) struct folio *vm_normal_folio(struct vm_area_struct *vma, unsigned long addr, 645318e9342SVishal Moola (Oracle) pte_t pte) 646318e9342SVishal Moola (Oracle) { 647318e9342SVishal Moola (Oracle) struct page *page = vm_normal_page(vma, addr, pte); 648318e9342SVishal Moola (Oracle) 649318e9342SVishal Moola (Oracle) if (page) 650318e9342SVishal Moola (Oracle) return page_folio(page); 651318e9342SVishal Moola (Oracle) return NULL; 652318e9342SVishal Moola (Oracle) } 653318e9342SVishal Moola (Oracle) 65428093f9fSGerald Schaefer #ifdef CONFIG_TRANSPARENT_HUGEPAGE 65528093f9fSGerald Schaefer struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, 65628093f9fSGerald Schaefer pmd_t pmd) 65728093f9fSGerald Schaefer { 65828093f9fSGerald Schaefer unsigned long pfn = pmd_pfn(pmd); 65928093f9fSGerald Schaefer 66028093f9fSGerald Schaefer /* 66128093f9fSGerald Schaefer * There is no pmd_special() but there may be special pmds, e.g. 66228093f9fSGerald Schaefer * in a direct-access (dax) mapping, so let's just replicate the 66300b3a331SLaurent Dufour * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here. 66428093f9fSGerald Schaefer */ 66528093f9fSGerald Schaefer if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 66628093f9fSGerald Schaefer if (vma->vm_flags & VM_MIXEDMAP) { 66728093f9fSGerald Schaefer if (!pfn_valid(pfn)) 66828093f9fSGerald Schaefer return NULL; 66928093f9fSGerald Schaefer goto out; 67028093f9fSGerald Schaefer } else { 67128093f9fSGerald Schaefer unsigned long off; 67228093f9fSGerald Schaefer off = (addr - vma->vm_start) >> PAGE_SHIFT; 67328093f9fSGerald Schaefer if (pfn == vma->vm_pgoff + off) 67428093f9fSGerald Schaefer return NULL; 67528093f9fSGerald Schaefer if (!is_cow_mapping(vma->vm_flags)) 67628093f9fSGerald Schaefer return NULL; 67728093f9fSGerald Schaefer } 67828093f9fSGerald Schaefer } 67928093f9fSGerald Schaefer 680e1fb4a08SDave Jiang if (pmd_devmap(pmd)) 681e1fb4a08SDave Jiang return NULL; 6823cde287bSYu Zhao if (is_huge_zero_pmd(pmd)) 68328093f9fSGerald Schaefer return NULL; 68428093f9fSGerald Schaefer if (unlikely(pfn > highest_memmap_pfn)) 68528093f9fSGerald Schaefer return NULL; 68628093f9fSGerald Schaefer 68728093f9fSGerald Schaefer /* 68828093f9fSGerald Schaefer * NOTE! We still have PageReserved() pages in the page tables. 68928093f9fSGerald Schaefer * eg. VDSO mappings can cause them to exist. 69028093f9fSGerald Schaefer */ 69128093f9fSGerald Schaefer out: 69228093f9fSGerald Schaefer return pfn_to_page(pfn); 69328093f9fSGerald Schaefer } 69428093f9fSGerald Schaefer #endif 69528093f9fSGerald Schaefer 696b756a3b5SAlistair Popple static void restore_exclusive_pte(struct vm_area_struct *vma, 697b756a3b5SAlistair Popple struct page *page, unsigned long address, 698b756a3b5SAlistair Popple pte_t *ptep) 699b756a3b5SAlistair Popple { 700c33c7948SRyan Roberts pte_t orig_pte; 701b756a3b5SAlistair Popple pte_t pte; 702b756a3b5SAlistair Popple swp_entry_t entry; 703b756a3b5SAlistair Popple 704c33c7948SRyan Roberts orig_pte = ptep_get(ptep); 705b756a3b5SAlistair Popple pte = pte_mkold(mk_pte(page, READ_ONCE(vma->vm_page_prot))); 706c33c7948SRyan Roberts if (pte_swp_soft_dirty(orig_pte)) 707b756a3b5SAlistair Popple pte = pte_mksoft_dirty(pte); 708b756a3b5SAlistair Popple 709c33c7948SRyan Roberts entry = pte_to_swp_entry(orig_pte); 710c33c7948SRyan Roberts if (pte_swp_uffd_wp(orig_pte)) 711b756a3b5SAlistair Popple pte = pte_mkuffd_wp(pte); 712b756a3b5SAlistair Popple else if (is_writable_device_exclusive_entry(entry)) 713b756a3b5SAlistair Popple pte = maybe_mkwrite(pte_mkdirty(pte), vma); 714b756a3b5SAlistair Popple 7156c287605SDavid Hildenbrand VM_BUG_ON(pte_write(pte) && !(PageAnon(page) && PageAnonExclusive(page))); 7166c287605SDavid Hildenbrand 717b756a3b5SAlistair Popple /* 718b756a3b5SAlistair Popple * No need to take a page reference as one was already 719b756a3b5SAlistair Popple * created when the swap entry was made. 720b756a3b5SAlistair Popple */ 721b756a3b5SAlistair Popple if (PageAnon(page)) 722f1e2db12SDavid Hildenbrand page_add_anon_rmap(page, vma, address, RMAP_NONE); 723b756a3b5SAlistair Popple else 724b756a3b5SAlistair Popple /* 725b756a3b5SAlistair Popple * Currently device exclusive access only supports anonymous 726b756a3b5SAlistair Popple * memory so the entry shouldn't point to a filebacked page. 727b756a3b5SAlistair Popple */ 7284d8ff640SMiaohe Lin WARN_ON_ONCE(1); 729b756a3b5SAlistair Popple 7301eba86c0SPasha Tatashin set_pte_at(vma->vm_mm, address, ptep, pte); 7311eba86c0SPasha Tatashin 732b756a3b5SAlistair Popple /* 733b756a3b5SAlistair Popple * No need to invalidate - it was non-present before. However 734b756a3b5SAlistair Popple * secondary CPUs may have mappings that need invalidating. 735b756a3b5SAlistair Popple */ 736b756a3b5SAlistair Popple update_mmu_cache(vma, address, ptep); 737b756a3b5SAlistair Popple } 738b756a3b5SAlistair Popple 739b756a3b5SAlistair Popple /* 740b756a3b5SAlistair Popple * Tries to restore an exclusive pte if the page lock can be acquired without 741b756a3b5SAlistair Popple * sleeping. 742b756a3b5SAlistair Popple */ 743b756a3b5SAlistair Popple static int 744b756a3b5SAlistair Popple try_restore_exclusive_pte(pte_t *src_pte, struct vm_area_struct *vma, 745b756a3b5SAlistair Popple unsigned long addr) 746b756a3b5SAlistair Popple { 747c33c7948SRyan Roberts swp_entry_t entry = pte_to_swp_entry(ptep_get(src_pte)); 748b756a3b5SAlistair Popple struct page *page = pfn_swap_entry_to_page(entry); 749b756a3b5SAlistair Popple 750b756a3b5SAlistair Popple if (trylock_page(page)) { 751b756a3b5SAlistair Popple restore_exclusive_pte(vma, page, addr, src_pte); 752b756a3b5SAlistair Popple unlock_page(page); 753b756a3b5SAlistair Popple return 0; 754b756a3b5SAlistair Popple } 755b756a3b5SAlistair Popple 756b756a3b5SAlistair Popple return -EBUSY; 757b756a3b5SAlistair Popple } 758b756a3b5SAlistair Popple 759ee498ed7SHugh Dickins /* 7601da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 7611da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 7621da177e4SLinus Torvalds * covered by this vma. 7631da177e4SLinus Torvalds */ 7641da177e4SLinus Torvalds 765df3a57d1SLinus Torvalds static unsigned long 766df3a57d1SLinus Torvalds copy_nonpresent_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 7678f34f1eaSPeter Xu pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *dst_vma, 7688f34f1eaSPeter Xu struct vm_area_struct *src_vma, unsigned long addr, int *rss) 7691da177e4SLinus Torvalds { 7708f34f1eaSPeter Xu unsigned long vm_flags = dst_vma->vm_flags; 771c33c7948SRyan Roberts pte_t orig_pte = ptep_get(src_pte); 772c33c7948SRyan Roberts pte_t pte = orig_pte; 7731da177e4SLinus Torvalds struct page *page; 774c33c7948SRyan Roberts swp_entry_t entry = pte_to_swp_entry(orig_pte); 7750697212aSChristoph Lameter 7762022b4d1SHugh Dickins if (likely(!non_swap_entry(entry))) { 777570a335bSHugh Dickins if (swap_duplicate(entry) < 0) 7789a5cc85cSAlistair Popple return -EIO; 779570a335bSHugh Dickins 7801da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 7811da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 7821da177e4SLinus Torvalds spin_lock(&mmlist_lock); 783f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 784f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 785f412ac08SHugh Dickins &src_mm->mmlist); 7861da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 7871da177e4SLinus Torvalds } 7881493a191SDavid Hildenbrand /* Mark the swap entry as shared. */ 789c33c7948SRyan Roberts if (pte_swp_exclusive(orig_pte)) { 790c33c7948SRyan Roberts pte = pte_swp_clear_exclusive(orig_pte); 7911493a191SDavid Hildenbrand set_pte_at(src_mm, addr, src_pte, pte); 7921493a191SDavid Hildenbrand } 793b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]++; 7942022b4d1SHugh Dickins } else if (is_migration_entry(entry)) { 795af5cdaf8SAlistair Popple page = pfn_swap_entry_to_page(entry); 7969f9f1acdSKonstantin Khlebnikov 797eca56ff9SJerome Marchand rss[mm_counter(page)]++; 7989f9f1acdSKonstantin Khlebnikov 7996c287605SDavid Hildenbrand if (!is_readable_migration_entry(entry) && 8000697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 8010697212aSChristoph Lameter /* 8026c287605SDavid Hildenbrand * COW mappings require pages in both parent and child 8036c287605SDavid Hildenbrand * to be set to read. A previously exclusive entry is 8046c287605SDavid Hildenbrand * now shared. 8050697212aSChristoph Lameter */ 8064dd845b5SAlistair Popple entry = make_readable_migration_entry( 8074dd845b5SAlistair Popple swp_offset(entry)); 8080697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 809c33c7948SRyan Roberts if (pte_swp_soft_dirty(orig_pte)) 810c3d16e16SCyrill Gorcunov pte = pte_swp_mksoft_dirty(pte); 811c33c7948SRyan Roberts if (pte_swp_uffd_wp(orig_pte)) 812f45ec5ffSPeter Xu pte = pte_swp_mkuffd_wp(pte); 8130697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 8140697212aSChristoph Lameter } 8155042db43SJérôme Glisse } else if (is_device_private_entry(entry)) { 816af5cdaf8SAlistair Popple page = pfn_swap_entry_to_page(entry); 8175042db43SJérôme Glisse 8185042db43SJérôme Glisse /* 8195042db43SJérôme Glisse * Update rss count even for unaddressable pages, as 8205042db43SJérôme Glisse * they should treated just like normal pages in this 8215042db43SJérôme Glisse * respect. 8225042db43SJérôme Glisse * 8235042db43SJérôme Glisse * We will likely want to have some new rss counters 8245042db43SJérôme Glisse * for unaddressable pages, at some point. But for now 8255042db43SJérôme Glisse * keep things as they are. 8265042db43SJérôme Glisse */ 8275042db43SJérôme Glisse get_page(page); 8285042db43SJérôme Glisse rss[mm_counter(page)]++; 829fb3d824dSDavid Hildenbrand /* Cannot fail as these pages cannot get pinned. */ 830fb3d824dSDavid Hildenbrand BUG_ON(page_try_dup_anon_rmap(page, false, src_vma)); 8315042db43SJérôme Glisse 8325042db43SJérôme Glisse /* 8335042db43SJérôme Glisse * We do not preserve soft-dirty information, because so 8345042db43SJérôme Glisse * far, checkpoint/restore is the only feature that 8355042db43SJérôme Glisse * requires that. And checkpoint/restore does not work 8365042db43SJérôme Glisse * when a device driver is involved (you cannot easily 8375042db43SJérôme Glisse * save and restore device driver state). 8385042db43SJérôme Glisse */ 8394dd845b5SAlistair Popple if (is_writable_device_private_entry(entry) && 8405042db43SJérôme Glisse is_cow_mapping(vm_flags)) { 8414dd845b5SAlistair Popple entry = make_readable_device_private_entry( 8424dd845b5SAlistair Popple swp_offset(entry)); 8435042db43SJérôme Glisse pte = swp_entry_to_pte(entry); 844c33c7948SRyan Roberts if (pte_swp_uffd_wp(orig_pte)) 845f45ec5ffSPeter Xu pte = pte_swp_mkuffd_wp(pte); 8465042db43SJérôme Glisse set_pte_at(src_mm, addr, src_pte, pte); 8475042db43SJérôme Glisse } 848b756a3b5SAlistair Popple } else if (is_device_exclusive_entry(entry)) { 849b756a3b5SAlistair Popple /* 850b756a3b5SAlistair Popple * Make device exclusive entries present by restoring the 851b756a3b5SAlistair Popple * original entry then copying as for a present pte. Device 852b756a3b5SAlistair Popple * exclusive entries currently only support private writable 853b756a3b5SAlistair Popple * (ie. COW) mappings. 854b756a3b5SAlistair Popple */ 855b756a3b5SAlistair Popple VM_BUG_ON(!is_cow_mapping(src_vma->vm_flags)); 856b756a3b5SAlistair Popple if (try_restore_exclusive_pte(src_pte, src_vma, addr)) 857b756a3b5SAlistair Popple return -EBUSY; 858b756a3b5SAlistair Popple return -ENOENT; 859c56d1b62SPeter Xu } else if (is_pte_marker_entry(entry)) { 860af19487fSAxel Rasmussen pte_marker marker = copy_pte_marker(entry, dst_vma); 861af19487fSAxel Rasmussen 862af19487fSAxel Rasmussen if (marker) 863af19487fSAxel Rasmussen set_pte_at(dst_mm, addr, dst_pte, 864af19487fSAxel Rasmussen make_pte_marker(marker)); 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); 1436e2942062Sxu xin if (unlikely(!page)) { 14376080d19fSxu xin ksm_might_unmap_zero_page(mm, ptent); 14381da177e4SLinus Torvalds continue; 1439e2942062Sxu xin } 1440eca56ff9SJerome Marchand 14415df397deSLinus Torvalds delay_rmap = 0; 1442eca56ff9SJerome Marchand if (!PageAnon(page)) { 14431cf35d47SLinus Torvalds if (pte_dirty(ptent)) { 14446237bcd9SHugh Dickins set_page_dirty(page); 14455df397deSLinus Torvalds if (tlb_delay_rmap(tlb)) { 14465df397deSLinus Torvalds delay_rmap = 1; 14475df397deSLinus Torvalds force_flush = 1; 14485df397deSLinus Torvalds } 14491cf35d47SLinus Torvalds } 14508788f678SYu Zhao if (pte_young(ptent) && likely(vma_has_recency(vma))) 1451bf3f3bc5SNick Piggin mark_page_accessed(page); 14526237bcd9SHugh Dickins } 1453eca56ff9SJerome Marchand rss[mm_counter(page)]--; 14545df397deSLinus Torvalds if (!delay_rmap) { 1455cea86fe2SHugh Dickins page_remove_rmap(page, vma, false); 14563dc14741SHugh Dickins if (unlikely(page_mapcount(page) < 0)) 14573dc14741SHugh Dickins print_bad_pte(vma, addr, ptent, page); 14585df397deSLinus Torvalds } 14595df397deSLinus Torvalds if (unlikely(__tlb_remove_page(tlb, page, delay_rmap))) { 14601cf35d47SLinus Torvalds force_flush = 1; 1461ce9ec37bSWill Deacon addr += PAGE_SIZE; 1462d16dfc55SPeter Zijlstra break; 14631cf35d47SLinus Torvalds } 14641da177e4SLinus Torvalds continue; 14651da177e4SLinus Torvalds } 14665042db43SJérôme Glisse 14675042db43SJérôme Glisse entry = pte_to_swp_entry(ptent); 1468b756a3b5SAlistair Popple if (is_device_private_entry(entry) || 1469b756a3b5SAlistair Popple is_device_exclusive_entry(entry)) { 14708018db85SPeter Xu page = pfn_swap_entry_to_page(entry); 1471254ab940SPeter Xu if (unlikely(!should_zap_page(details, page))) 14725042db43SJérôme Glisse continue; 1473999dad82SPeter Xu /* 1474999dad82SPeter Xu * Both device private/exclusive mappings should only 1475999dad82SPeter Xu * work with anonymous page so far, so we don't need to 1476999dad82SPeter Xu * consider uffd-wp bit when zap. For more information, 1477999dad82SPeter Xu * see zap_install_uffd_wp_if_needed(). 1478999dad82SPeter Xu */ 1479999dad82SPeter Xu WARN_ON_ONCE(!vma_is_anonymous(vma)); 14805042db43SJérôme Glisse rss[mm_counter(page)]--; 1481b756a3b5SAlistair Popple if (is_device_private_entry(entry)) 1482cea86fe2SHugh Dickins page_remove_rmap(page, vma, false); 14835042db43SJérôme Glisse put_page(page); 14848018db85SPeter Xu } else if (!non_swap_entry(entry)) { 14855abfd71dSPeter Xu /* Genuine swap entry, hence a private anon page */ 14865abfd71dSPeter Xu if (!should_zap_cows(details)) 14871da177e4SLinus Torvalds continue; 1488b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 14898018db85SPeter Xu if (unlikely(!free_swap_and_cache(entry))) 14908018db85SPeter Xu print_bad_pte(vma, addr, ptent, NULL); 14915abfd71dSPeter Xu } else if (is_migration_entry(entry)) { 1492af5cdaf8SAlistair Popple page = pfn_swap_entry_to_page(entry); 1493254ab940SPeter Xu if (!should_zap_page(details, page)) 14945abfd71dSPeter Xu continue; 1495eca56ff9SJerome Marchand rss[mm_counter(page)]--; 1496999dad82SPeter Xu } else if (pte_marker_entry_uffd_wp(entry)) { 14972bad466cSPeter Xu /* 14982bad466cSPeter Xu * For anon: always drop the marker; for file: only 14992bad466cSPeter Xu * drop the marker if explicitly requested. 15002bad466cSPeter Xu */ 15012bad466cSPeter Xu if (!vma_is_anonymous(vma) && 15022bad466cSPeter Xu !zap_drop_file_uffd_wp(details)) 1503999dad82SPeter Xu continue; 15049f186f9eSMiaohe Lin } else if (is_hwpoison_entry(entry) || 1505af19487fSAxel Rasmussen is_poisoned_swp_entry(entry)) { 15065abfd71dSPeter Xu if (!should_zap_cows(details)) 15075abfd71dSPeter Xu continue; 15085abfd71dSPeter Xu } else { 15095abfd71dSPeter Xu /* We should have covered all the swap entry types */ 15105abfd71dSPeter Xu WARN_ON_ONCE(1); 15119f9f1acdSKonstantin Khlebnikov } 15129888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 1513999dad82SPeter Xu zap_install_uffd_wp_if_needed(vma, addr, pte, details, ptent); 151497a89413SPeter Zijlstra } while (pte++, addr += PAGE_SIZE, addr != end); 1515ae859762SHugh Dickins 1516d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 15176606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 151851c6f666SRobin Holt 15191cf35d47SLinus Torvalds /* Do the actual TLB flush before dropping ptl */ 15205df397deSLinus Torvalds if (force_flush) { 15211cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 15225df397deSLinus Torvalds tlb_flush_rmaps(tlb, vma); 15235df397deSLinus Torvalds } 15241cf35d47SLinus Torvalds pte_unmap_unlock(start_pte, ptl); 15251cf35d47SLinus Torvalds 15261cf35d47SLinus Torvalds /* 15271cf35d47SLinus Torvalds * If we forced a TLB flush (either due to running out of 15281cf35d47SLinus Torvalds * batch buffers or because we needed to flush dirty TLB 15291cf35d47SLinus Torvalds * entries before releasing the ptl), free the batched 15303db82b93SHugh Dickins * memory too. Come back again if we didn't do everything. 15311cf35d47SLinus Torvalds */ 15323db82b93SHugh Dickins if (force_flush) 1533fa0aafb8SPeter Zijlstra tlb_flush_mmu(tlb); 1534d16dfc55SPeter Zijlstra 153551c6f666SRobin Holt return addr; 15361da177e4SLinus Torvalds } 15371da177e4SLinus Torvalds 153851c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1539b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 15401da177e4SLinus Torvalds unsigned long addr, unsigned long end, 154197a89413SPeter Zijlstra struct zap_details *details) 15421da177e4SLinus Torvalds { 15431da177e4SLinus Torvalds pmd_t *pmd; 15441da177e4SLinus Torvalds unsigned long next; 15451da177e4SLinus Torvalds 15461da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 15471da177e4SLinus Torvalds do { 15481da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 154984c3fc4eSZi Yan if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { 155053406ed1SHugh Dickins if (next - addr != HPAGE_PMD_SIZE) 1551fd60775aSDavid Rientjes __split_huge_pmd(vma, pmd, addr, false, NULL); 15523db82b93SHugh Dickins else if (zap_huge_pmd(tlb, vma, pmd, addr)) { 15533db82b93SHugh Dickins addr = next; 15543db82b93SHugh Dickins continue; 15553db82b93SHugh Dickins } 155671e3aac0SAndrea Arcangeli /* fall through */ 15573506659eSMatthew Wilcox (Oracle) } else if (details && details->single_folio && 15583506659eSMatthew Wilcox (Oracle) folio_test_pmd_mappable(details->single_folio) && 155922061a1fSHugh Dickins next - addr == HPAGE_PMD_SIZE && pmd_none(*pmd)) { 156022061a1fSHugh Dickins spinlock_t *ptl = pmd_lock(tlb->mm, pmd); 156122061a1fSHugh Dickins /* 156222061a1fSHugh Dickins * Take and drop THP pmd lock so that we cannot return 156322061a1fSHugh Dickins * prematurely, while zap_huge_pmd() has cleared *pmd, 156422061a1fSHugh Dickins * but not yet decremented compound_mapcount(). 156522061a1fSHugh Dickins */ 156622061a1fSHugh Dickins spin_unlock(ptl); 156771e3aac0SAndrea Arcangeli } 15683db82b93SHugh Dickins if (pmd_none(*pmd)) { 15693db82b93SHugh Dickins addr = next; 15703db82b93SHugh Dickins continue; 15713db82b93SHugh Dickins } 15723db82b93SHugh Dickins addr = zap_pte_range(tlb, vma, pmd, addr, next, details); 15733db82b93SHugh Dickins if (addr != next) 15743db82b93SHugh Dickins pmd--; 15753db82b93SHugh Dickins } while (pmd++, cond_resched(), addr != end); 157651c6f666SRobin Holt 157751c6f666SRobin Holt return addr; 15781da177e4SLinus Torvalds } 15791da177e4SLinus Torvalds 158051c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1581c2febafcSKirill A. Shutemov struct vm_area_struct *vma, p4d_t *p4d, 15821da177e4SLinus Torvalds unsigned long addr, unsigned long end, 158397a89413SPeter Zijlstra struct zap_details *details) 15841da177e4SLinus Torvalds { 15851da177e4SLinus Torvalds pud_t *pud; 15861da177e4SLinus Torvalds unsigned long next; 15871da177e4SLinus Torvalds 1588c2febafcSKirill A. Shutemov pud = pud_offset(p4d, addr); 15891da177e4SLinus Torvalds do { 15901da177e4SLinus Torvalds next = pud_addr_end(addr, end); 1591a00cc7d9SMatthew Wilcox if (pud_trans_huge(*pud) || pud_devmap(*pud)) { 1592a00cc7d9SMatthew Wilcox if (next - addr != HPAGE_PUD_SIZE) { 159342fc5414SMichel Lespinasse mmap_assert_locked(tlb->mm); 1594a00cc7d9SMatthew Wilcox split_huge_pud(vma, pud, addr); 1595a00cc7d9SMatthew Wilcox } else if (zap_huge_pud(tlb, vma, pud, addr)) 1596a00cc7d9SMatthew Wilcox goto next; 1597a00cc7d9SMatthew Wilcox /* fall through */ 1598a00cc7d9SMatthew Wilcox } 159997a89413SPeter Zijlstra if (pud_none_or_clear_bad(pud)) 16001da177e4SLinus Torvalds continue; 160197a89413SPeter Zijlstra next = zap_pmd_range(tlb, vma, pud, addr, next, details); 1602a00cc7d9SMatthew Wilcox next: 1603a00cc7d9SMatthew Wilcox cond_resched(); 160497a89413SPeter Zijlstra } while (pud++, addr = next, addr != end); 160551c6f666SRobin Holt 160651c6f666SRobin Holt return addr; 16071da177e4SLinus Torvalds } 16081da177e4SLinus Torvalds 1609c2febafcSKirill A. Shutemov static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, 1610c2febafcSKirill A. Shutemov struct vm_area_struct *vma, pgd_t *pgd, 1611c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 1612c2febafcSKirill A. Shutemov struct zap_details *details) 1613c2febafcSKirill A. Shutemov { 1614c2febafcSKirill A. Shutemov p4d_t *p4d; 1615c2febafcSKirill A. Shutemov unsigned long next; 1616c2febafcSKirill A. Shutemov 1617c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, addr); 1618c2febafcSKirill A. Shutemov do { 1619c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 1620c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(p4d)) 1621c2febafcSKirill A. Shutemov continue; 1622c2febafcSKirill A. Shutemov next = zap_pud_range(tlb, vma, p4d, addr, next, details); 1623c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 1624c2febafcSKirill A. Shutemov 1625c2febafcSKirill A. Shutemov return addr; 1626c2febafcSKirill A. Shutemov } 1627c2febafcSKirill A. Shutemov 1628aac45363SMichal Hocko void unmap_page_range(struct mmu_gather *tlb, 162951c6f666SRobin Holt struct vm_area_struct *vma, 16301da177e4SLinus Torvalds unsigned long addr, unsigned long end, 163197a89413SPeter Zijlstra struct zap_details *details) 16321da177e4SLinus Torvalds { 16331da177e4SLinus Torvalds pgd_t *pgd; 16341da177e4SLinus Torvalds unsigned long next; 16351da177e4SLinus Torvalds 16361da177e4SLinus Torvalds BUG_ON(addr >= end); 16371da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 16381da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 16391da177e4SLinus Torvalds do { 16401da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 164197a89413SPeter Zijlstra if (pgd_none_or_clear_bad(pgd)) 16421da177e4SLinus Torvalds continue; 1643c2febafcSKirill A. Shutemov next = zap_p4d_range(tlb, vma, pgd, addr, next, details); 164497a89413SPeter Zijlstra } while (pgd++, addr = next, addr != end); 16451da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 16461da177e4SLinus Torvalds } 16471da177e4SLinus Torvalds 1648f5cc4eefSAl Viro 1649f5cc4eefSAl Viro static void unmap_single_vma(struct mmu_gather *tlb, 16501da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 16514f74d2c8SLinus Torvalds unsigned long end_addr, 165268f48381SSuren Baghdasaryan struct zap_details *details, bool mm_wr_locked) 16531da177e4SLinus Torvalds { 1654f5cc4eefSAl Viro unsigned long start = max(vma->vm_start, start_addr); 16551da177e4SLinus Torvalds unsigned long end; 16561da177e4SLinus Torvalds 16571da177e4SLinus Torvalds if (start >= vma->vm_end) 1658f5cc4eefSAl Viro return; 16591da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 16601da177e4SLinus Torvalds if (end <= vma->vm_start) 1661f5cc4eefSAl Viro return; 16621da177e4SLinus Torvalds 1663cbc91f71SSrikar Dronamraju if (vma->vm_file) 1664cbc91f71SSrikar Dronamraju uprobe_munmap(vma, start, end); 1665cbc91f71SSrikar Dronamraju 1666b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) 166768f48381SSuren Baghdasaryan untrack_pfn(vma, 0, 0, mm_wr_locked); 16682ab64037Svenkatesh.pallipadi@intel.com 16698b2a1238SAl Viro if (start != end) { 167051c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1671a137e1ccSAndi Kleen /* 1672a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1673a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1674a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 16757aa6b4adSDavidlohr Bueso * cleanup path of mmap_region. When 1676a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 16777aa6b4adSDavidlohr Bueso * mmap_region() nullifies vma->vm_file 1678a137e1ccSAndi Kleen * before calling this function to clean up. 1679a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1680a137e1ccSAndi Kleen * safe to do nothing in this case. 1681a137e1ccSAndi Kleen */ 168224669e58SAneesh Kumar K.V if (vma->vm_file) { 168305e90bd0SPeter Xu zap_flags_t zap_flags = details ? 168405e90bd0SPeter Xu details->zap_flags : 0; 168505e90bd0SPeter Xu __unmap_hugepage_range_final(tlb, vma, start, end, 168605e90bd0SPeter Xu NULL, zap_flags); 168724669e58SAneesh Kumar K.V } 168851c6f666SRobin Holt } else 1689038c7aa1SAl Viro unmap_page_range(tlb, vma, start, end, details); 169097a89413SPeter Zijlstra } 169151c6f666SRobin Holt } 16921da177e4SLinus Torvalds 1693f5cc4eefSAl Viro /** 1694f5cc4eefSAl Viro * unmap_vmas - unmap a range of memory covered by a list of vma's 1695f5cc4eefSAl Viro * @tlb: address of the caller's struct mmu_gather 1696763ecb03SLiam R. Howlett * @mt: the maple tree 1697f5cc4eefSAl Viro * @vma: the starting vma 1698f5cc4eefSAl Viro * @start_addr: virtual address at which to start unmapping 1699f5cc4eefSAl Viro * @end_addr: virtual address at which to end unmapping 1700809ef83cSYang Li * @mm_wr_locked: lock flag 1701f5cc4eefSAl Viro * 1702f5cc4eefSAl Viro * Unmap all pages in the vma list. 1703f5cc4eefSAl Viro * 1704f5cc4eefSAl Viro * Only addresses between `start' and `end' will be unmapped. 1705f5cc4eefSAl Viro * 1706f5cc4eefSAl Viro * The VMA list must be sorted in ascending virtual address order. 1707f5cc4eefSAl Viro * 1708f5cc4eefSAl Viro * unmap_vmas() assumes that the caller will flush the whole unmapped address 1709f5cc4eefSAl Viro * range after unmap_vmas() returns. So the only responsibility here is to 1710f5cc4eefSAl Viro * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 1711f5cc4eefSAl Viro * drops the lock and schedules. 1712f5cc4eefSAl Viro */ 1713fd892593SLiam R. Howlett void unmap_vmas(struct mmu_gather *tlb, struct ma_state *mas, 1714f5cc4eefSAl Viro struct vm_area_struct *vma, unsigned long start_addr, 1715fd892593SLiam R. Howlett unsigned long end_addr, unsigned long tree_end, 1716fd892593SLiam R. Howlett bool mm_wr_locked) 1717f5cc4eefSAl Viro { 1718ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1719999dad82SPeter Xu struct zap_details details = { 172004ada095SMike Kravetz .zap_flags = ZAP_FLAG_DROP_MARKER | ZAP_FLAG_UNMAP, 1721999dad82SPeter Xu /* Careful - we need to zap private pages too! */ 1722999dad82SPeter Xu .even_cows = true, 1723999dad82SPeter Xu }; 1724f5cc4eefSAl Viro 17257d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma->vm_mm, 17266f4f13e8SJérôme Glisse start_addr, end_addr); 1727ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 1728763ecb03SLiam R. Howlett do { 172968f48381SSuren Baghdasaryan unmap_single_vma(tlb, vma, start_addr, end_addr, &details, 173068f48381SSuren Baghdasaryan mm_wr_locked); 1731fd892593SLiam R. Howlett } while ((vma = mas_find(mas, tree_end - 1)) != NULL); 1732ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 17331da177e4SLinus Torvalds } 17341da177e4SLinus Torvalds 17351da177e4SLinus Torvalds /** 1736f5cc4eefSAl Viro * zap_page_range_single - remove user pages in a given range 1737f5cc4eefSAl Viro * @vma: vm_area_struct holding the applicable pages 1738f5cc4eefSAl Viro * @address: starting address of pages to zap 1739f5cc4eefSAl Viro * @size: number of bytes to zap 17408a5f14a2SKirill A. Shutemov * @details: details of shared cache invalidation 1741f5cc4eefSAl Viro * 1742f5cc4eefSAl Viro * The range must fit into one VMA. 1743f5cc4eefSAl Viro */ 174421b85b09SMike Kravetz void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 1745f5cc4eefSAl Viro unsigned long size, struct zap_details *details) 1746f5cc4eefSAl Viro { 174721b85b09SMike Kravetz const unsigned long end = address + size; 1748ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1749f5cc4eefSAl Viro struct mmu_gather tlb; 1750f5cc4eefSAl Viro 1751f5cc4eefSAl Viro lru_add_drain(); 17527d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma->vm_mm, 175321b85b09SMike Kravetz address, end); 175421b85b09SMike Kravetz if (is_vm_hugetlb_page(vma)) 175521b85b09SMike Kravetz adjust_range_if_pmd_sharing_possible(vma, &range.start, 175621b85b09SMike Kravetz &range.end); 1757a72afd87SWill Deacon tlb_gather_mmu(&tlb, vma->vm_mm); 1758ac46d4f3SJérôme Glisse update_hiwater_rss(vma->vm_mm); 1759ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 176021b85b09SMike Kravetz /* 176121b85b09SMike Kravetz * unmap 'address-end' not 'range.start-range.end' as range 176221b85b09SMike Kravetz * could have been expanded for hugetlb pmd sharing. 176321b85b09SMike Kravetz */ 176468f48381SSuren Baghdasaryan unmap_single_vma(&tlb, vma, address, end, details, false); 1765ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 1766ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 17671da177e4SLinus Torvalds } 17681da177e4SLinus Torvalds 1769c627f9ccSJack Steiner /** 1770c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1771c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1772c627f9ccSJack Steiner * @address: starting address of pages to zap 1773c627f9ccSJack Steiner * @size: number of bytes to zap 1774c627f9ccSJack Steiner * 1775c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1776c627f9ccSJack Steiner * 1777c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1778c627f9ccSJack Steiner * 1779c627f9ccSJack Steiner */ 178027d036e3SLeon Romanovsky void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1781c627f9ccSJack Steiner unsigned long size) 1782c627f9ccSJack Steiner { 178388a35912SMiaohe Lin if (!range_in_vma(vma, address, address + size) || 1784c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 178527d036e3SLeon Romanovsky return; 178627d036e3SLeon Romanovsky 1787f5cc4eefSAl Viro zap_page_range_single(vma, address, size, NULL); 1788c627f9ccSJack Steiner } 1789c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1790c627f9ccSJack Steiner 17918cd3984dSArjun Roy static pmd_t *walk_to_pmd(struct mm_struct *mm, unsigned long addr) 1792c9cfcddfSLinus Torvalds { 1793c2febafcSKirill A. Shutemov pgd_t *pgd; 1794c2febafcSKirill A. Shutemov p4d_t *p4d; 1795c2febafcSKirill A. Shutemov pud_t *pud; 1796c2febafcSKirill A. Shutemov pmd_t *pmd; 1797c2febafcSKirill A. Shutemov 1798c2febafcSKirill A. Shutemov pgd = pgd_offset(mm, addr); 1799c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 1800c2febafcSKirill A. Shutemov if (!p4d) 1801c2febafcSKirill A. Shutemov return NULL; 1802c2febafcSKirill A. Shutemov pud = pud_alloc(mm, p4d, addr); 1803c2febafcSKirill A. Shutemov if (!pud) 1804c2febafcSKirill A. Shutemov return NULL; 1805c2febafcSKirill A. Shutemov pmd = pmd_alloc(mm, pud, addr); 1806c2febafcSKirill A. Shutemov if (!pmd) 1807c2febafcSKirill A. Shutemov return NULL; 1808c2febafcSKirill A. Shutemov 1809f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 18108cd3984dSArjun Roy return pmd; 18118cd3984dSArjun Roy } 18128cd3984dSArjun Roy 18138cd3984dSArjun Roy pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 18148cd3984dSArjun Roy spinlock_t **ptl) 18158cd3984dSArjun Roy { 18168cd3984dSArjun Roy pmd_t *pmd = walk_to_pmd(mm, addr); 18178cd3984dSArjun Roy 18188cd3984dSArjun Roy if (!pmd) 18198cd3984dSArjun Roy return NULL; 1820c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 1821c9cfcddfSLinus Torvalds } 1822c9cfcddfSLinus Torvalds 18238efd6f5bSArjun Roy static int validate_page_before_insert(struct page *page) 18248efd6f5bSArjun Roy { 18258efd6f5bSArjun Roy if (PageAnon(page) || PageSlab(page) || page_has_type(page)) 18268efd6f5bSArjun Roy return -EINVAL; 18278efd6f5bSArjun Roy flush_dcache_page(page); 18288efd6f5bSArjun Roy return 0; 18298efd6f5bSArjun Roy } 18308efd6f5bSArjun Roy 1831cea86fe2SHugh Dickins static int insert_page_into_pte_locked(struct vm_area_struct *vma, pte_t *pte, 18328efd6f5bSArjun Roy unsigned long addr, struct page *page, pgprot_t prot) 18338efd6f5bSArjun Roy { 1834c33c7948SRyan Roberts if (!pte_none(ptep_get(pte))) 18358efd6f5bSArjun Roy return -EBUSY; 18368efd6f5bSArjun Roy /* Ok, finally just insert the thing.. */ 18378efd6f5bSArjun Roy get_page(page); 1838f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, mm_counter_file(page)); 1839cea86fe2SHugh Dickins page_add_file_rmap(page, vma, false); 1840cea86fe2SHugh Dickins set_pte_at(vma->vm_mm, addr, pte, mk_pte(page, prot)); 18418efd6f5bSArjun Roy return 0; 18428efd6f5bSArjun Roy } 18438efd6f5bSArjun Roy 18441da177e4SLinus Torvalds /* 1845238f58d8SLinus Torvalds * This is the old fallback for page remapping. 1846238f58d8SLinus Torvalds * 1847238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 1848238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 1849238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 1850238f58d8SLinus Torvalds */ 1851423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 1852423bad60SNick Piggin struct page *page, pgprot_t prot) 1853238f58d8SLinus Torvalds { 1854238f58d8SLinus Torvalds int retval; 1855238f58d8SLinus Torvalds pte_t *pte; 1856238f58d8SLinus Torvalds spinlock_t *ptl; 1857238f58d8SLinus Torvalds 18588efd6f5bSArjun Roy retval = validate_page_before_insert(page); 18598efd6f5bSArjun Roy if (retval) 18605b4e655eSKAMEZAWA Hiroyuki goto out; 1861238f58d8SLinus Torvalds retval = -ENOMEM; 1862cea86fe2SHugh Dickins pte = get_locked_pte(vma->vm_mm, addr, &ptl); 1863238f58d8SLinus Torvalds if (!pte) 18645b4e655eSKAMEZAWA Hiroyuki goto out; 1865cea86fe2SHugh Dickins retval = insert_page_into_pte_locked(vma, pte, addr, page, prot); 1866238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 1867238f58d8SLinus Torvalds out: 1868238f58d8SLinus Torvalds return retval; 1869238f58d8SLinus Torvalds } 1870238f58d8SLinus Torvalds 18718cd3984dSArjun Roy #ifdef pte_index 1872cea86fe2SHugh Dickins static int insert_page_in_batch_locked(struct vm_area_struct *vma, pte_t *pte, 18738cd3984dSArjun Roy unsigned long addr, struct page *page, pgprot_t prot) 18748cd3984dSArjun Roy { 18758cd3984dSArjun Roy int err; 18768cd3984dSArjun Roy 18778cd3984dSArjun Roy if (!page_count(page)) 18788cd3984dSArjun Roy return -EINVAL; 18798cd3984dSArjun Roy err = validate_page_before_insert(page); 18807f70c2a6SArjun Roy if (err) 18817f70c2a6SArjun Roy return err; 1882cea86fe2SHugh Dickins return insert_page_into_pte_locked(vma, pte, addr, page, prot); 18838cd3984dSArjun Roy } 18848cd3984dSArjun Roy 18858cd3984dSArjun Roy /* insert_pages() amortizes the cost of spinlock operations 18868cd3984dSArjun Roy * when inserting pages in a loop. Arch *must* define pte_index. 18878cd3984dSArjun Roy */ 18888cd3984dSArjun Roy static int insert_pages(struct vm_area_struct *vma, unsigned long addr, 18898cd3984dSArjun Roy struct page **pages, unsigned long *num, pgprot_t prot) 18908cd3984dSArjun Roy { 18918cd3984dSArjun Roy pmd_t *pmd = NULL; 18927f70c2a6SArjun Roy pte_t *start_pte, *pte; 18937f70c2a6SArjun Roy spinlock_t *pte_lock; 18948cd3984dSArjun Roy struct mm_struct *const mm = vma->vm_mm; 18958cd3984dSArjun Roy unsigned long curr_page_idx = 0; 18968cd3984dSArjun Roy unsigned long remaining_pages_total = *num; 18978cd3984dSArjun Roy unsigned long pages_to_write_in_pmd; 18988cd3984dSArjun Roy int ret; 18998cd3984dSArjun Roy more: 19008cd3984dSArjun Roy ret = -EFAULT; 19018cd3984dSArjun Roy pmd = walk_to_pmd(mm, addr); 19028cd3984dSArjun Roy if (!pmd) 19038cd3984dSArjun Roy goto out; 19048cd3984dSArjun Roy 19058cd3984dSArjun Roy pages_to_write_in_pmd = min_t(unsigned long, 19068cd3984dSArjun Roy remaining_pages_total, PTRS_PER_PTE - pte_index(addr)); 19078cd3984dSArjun Roy 19088cd3984dSArjun Roy /* Allocate the PTE if necessary; takes PMD lock once only. */ 19098cd3984dSArjun Roy ret = -ENOMEM; 19108cd3984dSArjun Roy if (pte_alloc(mm, pmd)) 19118cd3984dSArjun Roy goto out; 19128cd3984dSArjun Roy 19138cd3984dSArjun Roy while (pages_to_write_in_pmd) { 19148cd3984dSArjun Roy int pte_idx = 0; 19158cd3984dSArjun Roy const int batch_size = min_t(int, pages_to_write_in_pmd, 8); 19168cd3984dSArjun Roy 19177f70c2a6SArjun Roy start_pte = pte_offset_map_lock(mm, pmd, addr, &pte_lock); 19183db82b93SHugh Dickins if (!start_pte) { 19193db82b93SHugh Dickins ret = -EFAULT; 19203db82b93SHugh Dickins goto out; 19213db82b93SHugh Dickins } 19227f70c2a6SArjun Roy for (pte = start_pte; pte_idx < batch_size; ++pte, ++pte_idx) { 1923cea86fe2SHugh Dickins int err = insert_page_in_batch_locked(vma, pte, 19248cd3984dSArjun Roy addr, pages[curr_page_idx], prot); 19258cd3984dSArjun Roy if (unlikely(err)) { 19267f70c2a6SArjun Roy pte_unmap_unlock(start_pte, pte_lock); 19278cd3984dSArjun Roy ret = err; 19288cd3984dSArjun Roy remaining_pages_total -= pte_idx; 19298cd3984dSArjun Roy goto out; 19308cd3984dSArjun Roy } 19318cd3984dSArjun Roy addr += PAGE_SIZE; 19328cd3984dSArjun Roy ++curr_page_idx; 19338cd3984dSArjun Roy } 19347f70c2a6SArjun Roy pte_unmap_unlock(start_pte, pte_lock); 19358cd3984dSArjun Roy pages_to_write_in_pmd -= batch_size; 19368cd3984dSArjun Roy remaining_pages_total -= batch_size; 19378cd3984dSArjun Roy } 19388cd3984dSArjun Roy if (remaining_pages_total) 19398cd3984dSArjun Roy goto more; 19408cd3984dSArjun Roy ret = 0; 19418cd3984dSArjun Roy out: 19428cd3984dSArjun Roy *num = remaining_pages_total; 19438cd3984dSArjun Roy return ret; 19448cd3984dSArjun Roy } 19458cd3984dSArjun Roy #endif /* ifdef pte_index */ 19468cd3984dSArjun Roy 19478cd3984dSArjun Roy /** 19488cd3984dSArjun Roy * vm_insert_pages - insert multiple pages into user vma, batching the pmd lock. 19498cd3984dSArjun Roy * @vma: user vma to map to 19508cd3984dSArjun Roy * @addr: target start user address of these pages 19518cd3984dSArjun Roy * @pages: source kernel pages 19528cd3984dSArjun Roy * @num: in: number of pages to map. out: number of pages that were *not* 19538cd3984dSArjun Roy * mapped. (0 means all pages were successfully mapped). 19548cd3984dSArjun Roy * 19558cd3984dSArjun Roy * Preferred over vm_insert_page() when inserting multiple pages. 19568cd3984dSArjun Roy * 19578cd3984dSArjun Roy * In case of error, we may have mapped a subset of the provided 19588cd3984dSArjun Roy * pages. It is the caller's responsibility to account for this case. 19598cd3984dSArjun Roy * 19608cd3984dSArjun Roy * The same restrictions apply as in vm_insert_page(). 19618cd3984dSArjun Roy */ 19628cd3984dSArjun Roy int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr, 19638cd3984dSArjun Roy struct page **pages, unsigned long *num) 19648cd3984dSArjun Roy { 19658cd3984dSArjun Roy #ifdef pte_index 19668cd3984dSArjun Roy const unsigned long end_addr = addr + (*num * PAGE_SIZE) - 1; 19678cd3984dSArjun Roy 19688cd3984dSArjun Roy if (addr < vma->vm_start || end_addr >= vma->vm_end) 19698cd3984dSArjun Roy return -EFAULT; 19708cd3984dSArjun Roy if (!(vma->vm_flags & VM_MIXEDMAP)) { 1971d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(vma->vm_mm)); 19728cd3984dSArjun Roy BUG_ON(vma->vm_flags & VM_PFNMAP); 19731c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_MIXEDMAP); 19748cd3984dSArjun Roy } 19758cd3984dSArjun Roy /* Defer page refcount checking till we're about to map that page. */ 19768cd3984dSArjun Roy return insert_pages(vma, addr, pages, num, vma->vm_page_prot); 19778cd3984dSArjun Roy #else 19788cd3984dSArjun Roy unsigned long idx = 0, pgcount = *num; 197945779b03STom Rix int err = -EINVAL; 19808cd3984dSArjun Roy 19818cd3984dSArjun Roy for (; idx < pgcount; ++idx) { 19828cd3984dSArjun Roy err = vm_insert_page(vma, addr + (PAGE_SIZE * idx), pages[idx]); 19838cd3984dSArjun Roy if (err) 19848cd3984dSArjun Roy break; 19858cd3984dSArjun Roy } 19868cd3984dSArjun Roy *num = pgcount - idx; 19878cd3984dSArjun Roy return err; 19888cd3984dSArjun Roy #endif /* ifdef pte_index */ 19898cd3984dSArjun Roy } 19908cd3984dSArjun Roy EXPORT_SYMBOL(vm_insert_pages); 19918cd3984dSArjun Roy 1992bfa5bf6dSRolf Eike Beer /** 1993bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 1994bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1995bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 1996bfa5bf6dSRolf Eike Beer * @page: source kernel page 1997bfa5bf6dSRolf Eike Beer * 1998a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 1999a145dd41SLinus Torvalds * into a user vma. 2000a145dd41SLinus Torvalds * 2001a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 2002a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 2003a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 20048dfcc9baSNick Piggin * (see split_page()). 2005a145dd41SLinus Torvalds * 2006a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 2007a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 2008a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 2009a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 2010a145dd41SLinus Torvalds * ask for a shared writable mapping! 2011a145dd41SLinus Torvalds * 2012a145dd41SLinus Torvalds * The page does not need to be reserved. 20134b6e1e37SKonstantin Khlebnikov * 20144b6e1e37SKonstantin Khlebnikov * Usually this function is called from f_op->mmap() handler 2015c1e8d7c6SMichel Lespinasse * under mm->mmap_lock write-lock, so it can change vma->vm_flags. 20164b6e1e37SKonstantin Khlebnikov * Caller must set VM_MIXEDMAP on vma if it wants to call this 20174b6e1e37SKonstantin Khlebnikov * function from other places, for example from page-fault handler. 2018a862f68aSMike Rapoport * 2019a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 2020a145dd41SLinus Torvalds */ 2021423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 2022423bad60SNick Piggin struct page *page) 2023a145dd41SLinus Torvalds { 2024a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 2025a145dd41SLinus Torvalds return -EFAULT; 2026a145dd41SLinus Torvalds if (!page_count(page)) 2027a145dd41SLinus Torvalds return -EINVAL; 20284b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & VM_MIXEDMAP)) { 2029d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(vma->vm_mm)); 20304b6e1e37SKonstantin Khlebnikov BUG_ON(vma->vm_flags & VM_PFNMAP); 20311c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_MIXEDMAP); 20324b6e1e37SKonstantin Khlebnikov } 2033423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 2034a145dd41SLinus Torvalds } 2035e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 2036a145dd41SLinus Torvalds 2037a667d745SSouptick Joarder /* 2038a667d745SSouptick Joarder * __vm_map_pages - maps range of kernel pages into user vma 2039a667d745SSouptick Joarder * @vma: user vma to map to 2040a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2041a667d745SSouptick Joarder * @num: number of pages in page array 2042a667d745SSouptick Joarder * @offset: user's requested vm_pgoff 2043a667d745SSouptick Joarder * 2044a667d745SSouptick Joarder * This allows drivers to map range of kernel pages into a user vma. 2045a667d745SSouptick Joarder * 2046a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2047a667d745SSouptick Joarder */ 2048a667d745SSouptick Joarder static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages, 2049a667d745SSouptick Joarder unsigned long num, unsigned long offset) 2050a667d745SSouptick Joarder { 2051a667d745SSouptick Joarder unsigned long count = vma_pages(vma); 2052a667d745SSouptick Joarder unsigned long uaddr = vma->vm_start; 2053a667d745SSouptick Joarder int ret, i; 2054a667d745SSouptick Joarder 2055a667d745SSouptick Joarder /* Fail if the user requested offset is beyond the end of the object */ 205696756fcbSMiguel Ojeda if (offset >= num) 2057a667d745SSouptick Joarder return -ENXIO; 2058a667d745SSouptick Joarder 2059a667d745SSouptick Joarder /* Fail if the user requested size exceeds available object size */ 2060a667d745SSouptick Joarder if (count > num - offset) 2061a667d745SSouptick Joarder return -ENXIO; 2062a667d745SSouptick Joarder 2063a667d745SSouptick Joarder for (i = 0; i < count; i++) { 2064a667d745SSouptick Joarder ret = vm_insert_page(vma, uaddr, pages[offset + i]); 2065a667d745SSouptick Joarder if (ret < 0) 2066a667d745SSouptick Joarder return ret; 2067a667d745SSouptick Joarder uaddr += PAGE_SIZE; 2068a667d745SSouptick Joarder } 2069a667d745SSouptick Joarder 2070a667d745SSouptick Joarder return 0; 2071a667d745SSouptick Joarder } 2072a667d745SSouptick Joarder 2073a667d745SSouptick Joarder /** 2074a667d745SSouptick Joarder * vm_map_pages - maps range of kernel pages starts with non zero offset 2075a667d745SSouptick Joarder * @vma: user vma to map to 2076a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2077a667d745SSouptick Joarder * @num: number of pages in page array 2078a667d745SSouptick Joarder * 2079a667d745SSouptick Joarder * Maps an object consisting of @num pages, catering for the user's 2080a667d745SSouptick Joarder * requested vm_pgoff 2081a667d745SSouptick Joarder * 2082a667d745SSouptick Joarder * If we fail to insert any page into the vma, the function will return 2083a667d745SSouptick Joarder * immediately leaving any previously inserted pages present. Callers 2084a667d745SSouptick Joarder * from the mmap handler may immediately return the error as their caller 2085a667d745SSouptick Joarder * will destroy the vma, removing any successfully inserted pages. Other 2086a667d745SSouptick Joarder * callers should make their own arrangements for calling unmap_region(). 2087a667d745SSouptick Joarder * 2088a667d745SSouptick Joarder * Context: Process context. Called by mmap handlers. 2089a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2090a667d745SSouptick Joarder */ 2091a667d745SSouptick Joarder int vm_map_pages(struct vm_area_struct *vma, struct page **pages, 2092a667d745SSouptick Joarder unsigned long num) 2093a667d745SSouptick Joarder { 2094a667d745SSouptick Joarder return __vm_map_pages(vma, pages, num, vma->vm_pgoff); 2095a667d745SSouptick Joarder } 2096a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages); 2097a667d745SSouptick Joarder 2098a667d745SSouptick Joarder /** 2099a667d745SSouptick Joarder * vm_map_pages_zero - map range of kernel pages starts with zero offset 2100a667d745SSouptick Joarder * @vma: user vma to map to 2101a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 2102a667d745SSouptick Joarder * @num: number of pages in page array 2103a667d745SSouptick Joarder * 2104a667d745SSouptick Joarder * Similar to vm_map_pages(), except that it explicitly sets the offset 2105a667d745SSouptick Joarder * to 0. This function is intended for the drivers that did not consider 2106a667d745SSouptick Joarder * vm_pgoff. 2107a667d745SSouptick Joarder * 2108a667d745SSouptick Joarder * Context: Process context. Called by mmap handlers. 2109a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 2110a667d745SSouptick Joarder */ 2111a667d745SSouptick Joarder int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, 2112a667d745SSouptick Joarder unsigned long num) 2113a667d745SSouptick Joarder { 2114a667d745SSouptick Joarder return __vm_map_pages(vma, pages, num, 0); 2115a667d745SSouptick Joarder } 2116a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages_zero); 2117a667d745SSouptick Joarder 21189b5a8e00SMatthew Wilcox static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2119b2770da6SRoss Zwisler pfn_t pfn, pgprot_t prot, bool mkwrite) 2120423bad60SNick Piggin { 2121423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 2122423bad60SNick Piggin pte_t *pte, entry; 2123423bad60SNick Piggin spinlock_t *ptl; 2124423bad60SNick Piggin 2125423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 2126423bad60SNick Piggin if (!pte) 21279b5a8e00SMatthew Wilcox return VM_FAULT_OOM; 2128c33c7948SRyan Roberts entry = ptep_get(pte); 2129c33c7948SRyan Roberts if (!pte_none(entry)) { 2130b2770da6SRoss Zwisler if (mkwrite) { 2131b2770da6SRoss Zwisler /* 2132b2770da6SRoss Zwisler * For read faults on private mappings the PFN passed 2133b2770da6SRoss Zwisler * in may not match the PFN we have mapped if the 2134b2770da6SRoss Zwisler * mapped PFN is a writeable COW page. In the mkwrite 2135b2770da6SRoss Zwisler * case we are creating a writable PTE for a shared 2136f2c57d91SJan Kara * mapping and we expect the PFNs to match. If they 2137f2c57d91SJan Kara * don't match, we are likely racing with block 2138f2c57d91SJan Kara * allocation and mapping invalidation so just skip the 2139f2c57d91SJan Kara * update. 2140b2770da6SRoss Zwisler */ 2141c33c7948SRyan Roberts if (pte_pfn(entry) != pfn_t_to_pfn(pfn)) { 2142c33c7948SRyan Roberts WARN_ON_ONCE(!is_zero_pfn(pte_pfn(entry))); 2143423bad60SNick Piggin goto out_unlock; 2144f2c57d91SJan Kara } 2145c33c7948SRyan Roberts entry = pte_mkyoung(entry); 2146cae85cb8SJan Kara entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2147cae85cb8SJan Kara if (ptep_set_access_flags(vma, addr, pte, entry, 1)) 2148cae85cb8SJan Kara update_mmu_cache(vma, addr, pte); 2149cae85cb8SJan Kara } 2150b2770da6SRoss Zwisler goto out_unlock; 2151b2770da6SRoss Zwisler } 2152423bad60SNick Piggin 2153423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 215401c8f1c4SDan Williams if (pfn_t_devmap(pfn)) 215501c8f1c4SDan Williams entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); 215601c8f1c4SDan Williams else 215701c8f1c4SDan Williams entry = pte_mkspecial(pfn_t_pte(pfn, prot)); 2158b2770da6SRoss Zwisler 2159b2770da6SRoss Zwisler if (mkwrite) { 2160b2770da6SRoss Zwisler entry = pte_mkyoung(entry); 2161b2770da6SRoss Zwisler entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2162b2770da6SRoss Zwisler } 2163b2770da6SRoss Zwisler 2164423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 21654b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 2166423bad60SNick Piggin 2167423bad60SNick Piggin out_unlock: 2168423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 21699b5a8e00SMatthew Wilcox return VM_FAULT_NOPAGE; 2170423bad60SNick Piggin } 2171423bad60SNick Piggin 2172f5e6d1d5SMatthew Wilcox /** 2173f5e6d1d5SMatthew Wilcox * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot 2174f5e6d1d5SMatthew Wilcox * @vma: user vma to map to 2175f5e6d1d5SMatthew Wilcox * @addr: target user address of this page 2176f5e6d1d5SMatthew Wilcox * @pfn: source kernel pfn 2177f5e6d1d5SMatthew Wilcox * @pgprot: pgprot flags for the inserted page 2178f5e6d1d5SMatthew Wilcox * 2179a1a0aea5SRandy Dunlap * This is exactly like vmf_insert_pfn(), except that it allows drivers 2180f5e6d1d5SMatthew Wilcox * to override pgprot on a per-page basis. 2181f5e6d1d5SMatthew Wilcox * 2182f5e6d1d5SMatthew Wilcox * This only makes sense for IO mappings, and it makes no sense for 2183f5e6d1d5SMatthew Wilcox * COW mappings. In general, using multiple vmas is preferable; 2184ae2b01f3SMatthew Wilcox * vmf_insert_pfn_prot should only be used if using multiple VMAs is 2185f5e6d1d5SMatthew Wilcox * impractical. 2186f5e6d1d5SMatthew Wilcox * 218728d8b812SLorenzo Stoakes * pgprot typically only differs from @vma->vm_page_prot when drivers set 218828d8b812SLorenzo Stoakes * caching- and encryption bits different than those of @vma->vm_page_prot, 218928d8b812SLorenzo Stoakes * because the caching- or encryption mode may not be known at mmap() time. 219028d8b812SLorenzo Stoakes * 219128d8b812SLorenzo Stoakes * This is ok as long as @vma->vm_page_prot is not used by the core vm 219228d8b812SLorenzo Stoakes * to set caching and encryption bits for those vmas (except for COW pages). 219328d8b812SLorenzo Stoakes * This is ensured by core vm only modifying these page table entries using 219428d8b812SLorenzo Stoakes * functions that don't touch caching- or encryption bits, using pte_modify() 219528d8b812SLorenzo Stoakes * if needed. (See for example mprotect()). 219628d8b812SLorenzo Stoakes * 219728d8b812SLorenzo Stoakes * Also when new page-table entries are created, this is only done using the 219828d8b812SLorenzo Stoakes * fault() callback, and never using the value of vma->vm_page_prot, 219928d8b812SLorenzo Stoakes * except for page-table entries that point to anonymous pages as the result 220028d8b812SLorenzo Stoakes * of COW. 2201574c5b3dSThomas Hellstrom * 2202ae2b01f3SMatthew Wilcox * Context: Process context. May allocate using %GFP_KERNEL. 2203f5e6d1d5SMatthew Wilcox * Return: vm_fault_t value. 2204f5e6d1d5SMatthew Wilcox */ 2205f5e6d1d5SMatthew Wilcox vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, 2206f5e6d1d5SMatthew Wilcox unsigned long pfn, pgprot_t pgprot) 2207f5e6d1d5SMatthew Wilcox { 22086d958546SMatthew Wilcox /* 22096d958546SMatthew Wilcox * Technically, architectures with pte_special can avoid all these 22106d958546SMatthew Wilcox * restrictions (same for remap_pfn_range). However we would like 22116d958546SMatthew Wilcox * consistency in testing and feature parity among all, so we should 22126d958546SMatthew Wilcox * try to keep these invariants in place for everybody. 22136d958546SMatthew Wilcox */ 22146d958546SMatthew Wilcox BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 22156d958546SMatthew Wilcox BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 22166d958546SMatthew Wilcox (VM_PFNMAP|VM_MIXEDMAP)); 22176d958546SMatthew Wilcox BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 22186d958546SMatthew Wilcox BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 22196d958546SMatthew Wilcox 22206d958546SMatthew Wilcox if (addr < vma->vm_start || addr >= vma->vm_end) 22216d958546SMatthew Wilcox return VM_FAULT_SIGBUS; 22226d958546SMatthew Wilcox 22236d958546SMatthew Wilcox if (!pfn_modify_allowed(pfn, pgprot)) 22246d958546SMatthew Wilcox return VM_FAULT_SIGBUS; 22256d958546SMatthew Wilcox 22266d958546SMatthew Wilcox track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); 22276d958546SMatthew Wilcox 22289b5a8e00SMatthew Wilcox return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, 22296d958546SMatthew Wilcox false); 2230f5e6d1d5SMatthew Wilcox } 2231f5e6d1d5SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_pfn_prot); 2232e0dc0d8fSNick Piggin 2233ae2b01f3SMatthew Wilcox /** 2234ae2b01f3SMatthew Wilcox * vmf_insert_pfn - insert single pfn into user vma 2235ae2b01f3SMatthew Wilcox * @vma: user vma to map to 2236ae2b01f3SMatthew Wilcox * @addr: target user address of this page 2237ae2b01f3SMatthew Wilcox * @pfn: source kernel pfn 2238ae2b01f3SMatthew Wilcox * 2239ae2b01f3SMatthew Wilcox * Similar to vm_insert_page, this allows drivers to insert individual pages 2240ae2b01f3SMatthew Wilcox * they've allocated into a user vma. Same comments apply. 2241ae2b01f3SMatthew Wilcox * 2242ae2b01f3SMatthew Wilcox * This function should only be called from a vm_ops->fault handler, and 2243ae2b01f3SMatthew Wilcox * in that case the handler should return the result of this function. 2244ae2b01f3SMatthew Wilcox * 2245ae2b01f3SMatthew Wilcox * vma cannot be a COW mapping. 2246ae2b01f3SMatthew Wilcox * 2247ae2b01f3SMatthew Wilcox * As this is called only for pages that do not currently exist, we 2248ae2b01f3SMatthew Wilcox * do not need to flush old virtual caches or the TLB. 2249ae2b01f3SMatthew Wilcox * 2250ae2b01f3SMatthew Wilcox * Context: Process context. May allocate using %GFP_KERNEL. 2251ae2b01f3SMatthew Wilcox * Return: vm_fault_t value. 2252ae2b01f3SMatthew Wilcox */ 2253ae2b01f3SMatthew Wilcox vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2254ae2b01f3SMatthew Wilcox unsigned long pfn) 2255ae2b01f3SMatthew Wilcox { 2256ae2b01f3SMatthew Wilcox return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); 2257ae2b01f3SMatthew Wilcox } 2258ae2b01f3SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_pfn); 2259ae2b01f3SMatthew Wilcox 2260785a3fabSDan Williams static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) 2261785a3fabSDan Williams { 2262785a3fabSDan Williams /* these checks mirror the abort conditions in vm_normal_page */ 2263785a3fabSDan Williams if (vma->vm_flags & VM_MIXEDMAP) 2264785a3fabSDan Williams return true; 2265785a3fabSDan Williams if (pfn_t_devmap(pfn)) 2266785a3fabSDan Williams return true; 2267785a3fabSDan Williams if (pfn_t_special(pfn)) 2268785a3fabSDan Williams return true; 2269785a3fabSDan Williams if (is_zero_pfn(pfn_t_to_pfn(pfn))) 2270785a3fabSDan Williams return true; 2271785a3fabSDan Williams return false; 2272785a3fabSDan Williams } 2273785a3fabSDan Williams 227479f3aa5bSMatthew Wilcox static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma, 227528d8b812SLorenzo Stoakes unsigned long addr, pfn_t pfn, bool mkwrite) 2276423bad60SNick Piggin { 227728d8b812SLorenzo Stoakes pgprot_t pgprot = vma->vm_page_prot; 227879f3aa5bSMatthew Wilcox int err; 227987744ab3SDan Williams 2280785a3fabSDan Williams BUG_ON(!vm_mixed_ok(vma, pfn)); 2281423bad60SNick Piggin 2282423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 228379f3aa5bSMatthew Wilcox return VM_FAULT_SIGBUS; 2284308a047cSBorislav Petkov 2285308a047cSBorislav Petkov track_pfn_insert(vma, &pgprot, pfn); 2286423bad60SNick Piggin 228742e4089cSAndi Kleen if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) 228879f3aa5bSMatthew Wilcox return VM_FAULT_SIGBUS; 228942e4089cSAndi Kleen 2290423bad60SNick Piggin /* 2291423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 2292423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 2293423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 229462eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 229562eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 2296423bad60SNick Piggin */ 229700b3a331SLaurent Dufour if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && 229800b3a331SLaurent Dufour !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { 2299423bad60SNick Piggin struct page *page; 2300423bad60SNick Piggin 230103fc2da6SDan Williams /* 230203fc2da6SDan Williams * At this point we are committed to insert_page() 230303fc2da6SDan Williams * regardless of whether the caller specified flags that 230403fc2da6SDan Williams * result in pfn_t_has_page() == false. 230503fc2da6SDan Williams */ 230603fc2da6SDan Williams page = pfn_to_page(pfn_t_to_pfn(pfn)); 230779f3aa5bSMatthew Wilcox err = insert_page(vma, addr, page, pgprot); 230879f3aa5bSMatthew Wilcox } else { 23099b5a8e00SMatthew Wilcox return insert_pfn(vma, addr, pfn, pgprot, mkwrite); 2310423bad60SNick Piggin } 2311b2770da6SRoss Zwisler 23125d747637SMatthew Wilcox if (err == -ENOMEM) 23135d747637SMatthew Wilcox return VM_FAULT_OOM; 23145d747637SMatthew Wilcox if (err < 0 && err != -EBUSY) 23155d747637SMatthew Wilcox return VM_FAULT_SIGBUS; 23165d747637SMatthew Wilcox 23175d747637SMatthew Wilcox return VM_FAULT_NOPAGE; 2318423bad60SNick Piggin } 231979f3aa5bSMatthew Wilcox 232079f3aa5bSMatthew Wilcox vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 232179f3aa5bSMatthew Wilcox pfn_t pfn) 232279f3aa5bSMatthew Wilcox { 232328d8b812SLorenzo Stoakes return __vm_insert_mixed(vma, addr, pfn, false); 232479f3aa5bSMatthew Wilcox } 23255d747637SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_mixed); 2326423bad60SNick Piggin 2327ab77dab4SSouptick Joarder /* 2328ab77dab4SSouptick Joarder * If the insertion of PTE failed because someone else already added a 2329ab77dab4SSouptick Joarder * different entry in the mean time, we treat that as success as we assume 2330ab77dab4SSouptick Joarder * the same entry was actually inserted. 2331ab77dab4SSouptick Joarder */ 2332ab77dab4SSouptick Joarder vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, 2333ab77dab4SSouptick Joarder unsigned long addr, pfn_t pfn) 2334b2770da6SRoss Zwisler { 233528d8b812SLorenzo Stoakes return __vm_insert_mixed(vma, addr, pfn, true); 2336b2770da6SRoss Zwisler } 2337ab77dab4SSouptick Joarder EXPORT_SYMBOL(vmf_insert_mixed_mkwrite); 2338b2770da6SRoss Zwisler 2339a145dd41SLinus Torvalds /* 23401da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 23411da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 23421da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 23431da177e4SLinus Torvalds */ 23441da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 23451da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23461da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23471da177e4SLinus Torvalds { 234890a3e375SMiaohe Lin pte_t *pte, *mapped_pte; 2349c74df32cSHugh Dickins spinlock_t *ptl; 235042e4089cSAndi Kleen int err = 0; 23511da177e4SLinus Torvalds 235290a3e375SMiaohe Lin mapped_pte = pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 23531da177e4SLinus Torvalds if (!pte) 23541da177e4SLinus Torvalds return -ENOMEM; 23556606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 23561da177e4SLinus Torvalds do { 2357c33c7948SRyan Roberts BUG_ON(!pte_none(ptep_get(pte))); 235842e4089cSAndi Kleen if (!pfn_modify_allowed(pfn, prot)) { 235942e4089cSAndi Kleen err = -EACCES; 236042e4089cSAndi Kleen break; 236142e4089cSAndi Kleen } 23627e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 23631da177e4SLinus Torvalds pfn++; 23641da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 23656606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 236690a3e375SMiaohe Lin pte_unmap_unlock(mapped_pte, ptl); 236742e4089cSAndi Kleen return err; 23681da177e4SLinus Torvalds } 23691da177e4SLinus Torvalds 23701da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 23711da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23721da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23731da177e4SLinus Torvalds { 23741da177e4SLinus Torvalds pmd_t *pmd; 23751da177e4SLinus Torvalds unsigned long next; 237642e4089cSAndi Kleen int err; 23771da177e4SLinus Torvalds 23781da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 23791da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 23801da177e4SLinus Torvalds if (!pmd) 23811da177e4SLinus Torvalds return -ENOMEM; 2382f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 23831da177e4SLinus Torvalds do { 23841da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 238542e4089cSAndi Kleen err = remap_pte_range(mm, pmd, addr, next, 238642e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 238742e4089cSAndi Kleen if (err) 238842e4089cSAndi Kleen return err; 23891da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 23901da177e4SLinus Torvalds return 0; 23911da177e4SLinus Torvalds } 23921da177e4SLinus Torvalds 2393c2febafcSKirill A. Shutemov static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, 23941da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23951da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23961da177e4SLinus Torvalds { 23971da177e4SLinus Torvalds pud_t *pud; 23981da177e4SLinus Torvalds unsigned long next; 239942e4089cSAndi Kleen int err; 24001da177e4SLinus Torvalds 24011da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 2402c2febafcSKirill A. Shutemov pud = pud_alloc(mm, p4d, addr); 24031da177e4SLinus Torvalds if (!pud) 24041da177e4SLinus Torvalds return -ENOMEM; 24051da177e4SLinus Torvalds do { 24061da177e4SLinus Torvalds next = pud_addr_end(addr, end); 240742e4089cSAndi Kleen err = remap_pmd_range(mm, pud, addr, next, 240842e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 240942e4089cSAndi Kleen if (err) 241042e4089cSAndi Kleen return err; 24111da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 24121da177e4SLinus Torvalds return 0; 24131da177e4SLinus Torvalds } 24141da177e4SLinus Torvalds 2415c2febafcSKirill A. Shutemov static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, 2416c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 2417c2febafcSKirill A. Shutemov unsigned long pfn, pgprot_t prot) 2418c2febafcSKirill A. Shutemov { 2419c2febafcSKirill A. Shutemov p4d_t *p4d; 2420c2febafcSKirill A. Shutemov unsigned long next; 242142e4089cSAndi Kleen int err; 2422c2febafcSKirill A. Shutemov 2423c2febafcSKirill A. Shutemov pfn -= addr >> PAGE_SHIFT; 2424c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 2425c2febafcSKirill A. Shutemov if (!p4d) 2426c2febafcSKirill A. Shutemov return -ENOMEM; 2427c2febafcSKirill A. Shutemov do { 2428c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 242942e4089cSAndi Kleen err = remap_pud_range(mm, p4d, addr, next, 243042e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 243142e4089cSAndi Kleen if (err) 243242e4089cSAndi Kleen return err; 2433c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 2434c2febafcSKirill A. Shutemov return 0; 2435c2febafcSKirill A. Shutemov } 2436c2febafcSKirill A. Shutemov 243774ffa5a3SChristoph Hellwig /* 243874ffa5a3SChristoph Hellwig * Variant of remap_pfn_range that does not call track_pfn_remap. The caller 243974ffa5a3SChristoph Hellwig * must have pre-validated the caching bits of the pgprot_t. 2440bfa5bf6dSRolf Eike Beer */ 244174ffa5a3SChristoph Hellwig int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr, 24421da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 24431da177e4SLinus Torvalds { 24441da177e4SLinus Torvalds pgd_t *pgd; 24451da177e4SLinus Torvalds unsigned long next; 24462d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 24471da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 24481da177e4SLinus Torvalds int err; 24491da177e4SLinus Torvalds 24500c4123e3SAlex Zhang if (WARN_ON_ONCE(!PAGE_ALIGNED(addr))) 24510c4123e3SAlex Zhang return -EINVAL; 24520c4123e3SAlex Zhang 24531da177e4SLinus Torvalds /* 24541da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 24551da177e4SLinus Torvalds * rest of the world about it: 24561da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 24571da177e4SLinus Torvalds * (accesses can have side effects). 24586aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 24596aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 24606aab341eSLinus Torvalds * with them. 2461314e51b9SKonstantin Khlebnikov * VM_DONTEXPAND 2462314e51b9SKonstantin Khlebnikov * Disable vma merging and expanding with mremap(). 2463314e51b9SKonstantin Khlebnikov * VM_DONTDUMP 2464314e51b9SKonstantin Khlebnikov * Omit vma from core dump, even when VM_IO turned off. 2465fb155c16SLinus Torvalds * 2466fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 2467fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 2468fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 2469b3b9c293SKonstantin Khlebnikov * See vm_normal_page() for details. 24701da177e4SLinus Torvalds */ 2471b3b9c293SKonstantin Khlebnikov if (is_cow_mapping(vma->vm_flags)) { 2472b3b9c293SKonstantin Khlebnikov if (addr != vma->vm_start || end != vma->vm_end) 2473b3b9c293SKonstantin Khlebnikov return -EINVAL; 24746aab341eSLinus Torvalds vma->vm_pgoff = pfn; 2475b3b9c293SKonstantin Khlebnikov } 2476b3b9c293SKonstantin Khlebnikov 24771c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP); 24781da177e4SLinus Torvalds 24791da177e4SLinus Torvalds BUG_ON(addr >= end); 24801da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 24811da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 24821da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 24831da177e4SLinus Torvalds do { 24841da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 2485c2febafcSKirill A. Shutemov err = remap_p4d_range(mm, pgd, addr, next, 24861da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 24871da177e4SLinus Torvalds if (err) 248874ffa5a3SChristoph Hellwig return err; 24891da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 24902ab64037Svenkatesh.pallipadi@intel.com 249174ffa5a3SChristoph Hellwig return 0; 249274ffa5a3SChristoph Hellwig } 24932ab64037Svenkatesh.pallipadi@intel.com 249474ffa5a3SChristoph Hellwig /** 249574ffa5a3SChristoph Hellwig * remap_pfn_range - remap kernel memory to userspace 249674ffa5a3SChristoph Hellwig * @vma: user vma to map to 249774ffa5a3SChristoph Hellwig * @addr: target page aligned user address to start at 249874ffa5a3SChristoph Hellwig * @pfn: page frame number of kernel physical memory address 249974ffa5a3SChristoph Hellwig * @size: size of mapping area 250074ffa5a3SChristoph Hellwig * @prot: page protection flags for this mapping 250174ffa5a3SChristoph Hellwig * 250274ffa5a3SChristoph Hellwig * Note: this is only safe if the mm semaphore is held when called. 250374ffa5a3SChristoph Hellwig * 250474ffa5a3SChristoph Hellwig * Return: %0 on success, negative error code otherwise. 250574ffa5a3SChristoph Hellwig */ 250674ffa5a3SChristoph Hellwig int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 250774ffa5a3SChristoph Hellwig unsigned long pfn, unsigned long size, pgprot_t prot) 250874ffa5a3SChristoph Hellwig { 250974ffa5a3SChristoph Hellwig int err; 251074ffa5a3SChristoph Hellwig 251174ffa5a3SChristoph Hellwig err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); 251274ffa5a3SChristoph Hellwig if (err) 251374ffa5a3SChristoph Hellwig return -EINVAL; 251474ffa5a3SChristoph Hellwig 251574ffa5a3SChristoph Hellwig err = remap_pfn_range_notrack(vma, addr, pfn, size, prot); 251674ffa5a3SChristoph Hellwig if (err) 251768f48381SSuren Baghdasaryan untrack_pfn(vma, pfn, PAGE_ALIGN(size), true); 25181da177e4SLinus Torvalds return err; 25191da177e4SLinus Torvalds } 25201da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 25211da177e4SLinus Torvalds 2522b4cbb197SLinus Torvalds /** 2523b4cbb197SLinus Torvalds * vm_iomap_memory - remap memory to userspace 2524b4cbb197SLinus Torvalds * @vma: user vma to map to 2525abd69b9eSWang Wenhu * @start: start of the physical memory to be mapped 2526b4cbb197SLinus Torvalds * @len: size of area 2527b4cbb197SLinus Torvalds * 2528b4cbb197SLinus Torvalds * This is a simplified io_remap_pfn_range() for common driver use. The 2529b4cbb197SLinus Torvalds * driver just needs to give us the physical memory range to be mapped, 2530b4cbb197SLinus Torvalds * we'll figure out the rest from the vma information. 2531b4cbb197SLinus Torvalds * 2532b4cbb197SLinus Torvalds * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get 2533b4cbb197SLinus Torvalds * whatever write-combining details or similar. 2534a862f68aSMike Rapoport * 2535a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 2536b4cbb197SLinus Torvalds */ 2537b4cbb197SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 2538b4cbb197SLinus Torvalds { 2539b4cbb197SLinus Torvalds unsigned long vm_len, pfn, pages; 2540b4cbb197SLinus Torvalds 2541b4cbb197SLinus Torvalds /* Check that the physical memory area passed in looks valid */ 2542b4cbb197SLinus Torvalds if (start + len < start) 2543b4cbb197SLinus Torvalds return -EINVAL; 2544b4cbb197SLinus Torvalds /* 2545b4cbb197SLinus Torvalds * You *really* shouldn't map things that aren't page-aligned, 2546b4cbb197SLinus Torvalds * but we've historically allowed it because IO memory might 2547b4cbb197SLinus Torvalds * just have smaller alignment. 2548b4cbb197SLinus Torvalds */ 2549b4cbb197SLinus Torvalds len += start & ~PAGE_MASK; 2550b4cbb197SLinus Torvalds pfn = start >> PAGE_SHIFT; 2551b4cbb197SLinus Torvalds pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; 2552b4cbb197SLinus Torvalds if (pfn + pages < pfn) 2553b4cbb197SLinus Torvalds return -EINVAL; 2554b4cbb197SLinus Torvalds 2555b4cbb197SLinus Torvalds /* We start the mapping 'vm_pgoff' pages into the area */ 2556b4cbb197SLinus Torvalds if (vma->vm_pgoff > pages) 2557b4cbb197SLinus Torvalds return -EINVAL; 2558b4cbb197SLinus Torvalds pfn += vma->vm_pgoff; 2559b4cbb197SLinus Torvalds pages -= vma->vm_pgoff; 2560b4cbb197SLinus Torvalds 2561b4cbb197SLinus Torvalds /* Can we fit all of the mapping? */ 2562b4cbb197SLinus Torvalds vm_len = vma->vm_end - vma->vm_start; 2563b4cbb197SLinus Torvalds if (vm_len >> PAGE_SHIFT > pages) 2564b4cbb197SLinus Torvalds return -EINVAL; 2565b4cbb197SLinus Torvalds 2566b4cbb197SLinus Torvalds /* Ok, let it rip */ 2567b4cbb197SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 2568b4cbb197SLinus Torvalds } 2569b4cbb197SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 2570b4cbb197SLinus Torvalds 2571aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 2572aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2573e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2574e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2575aee16b3cSJeremy Fitzhardinge { 25768abb50c7SMiaohe Lin pte_t *pte, *mapped_pte; 2577be1db475SDaniel Axtens int err = 0; 25783f649ab7SKees Cook spinlock_t *ptl; 2579aee16b3cSJeremy Fitzhardinge 2580be1db475SDaniel Axtens if (create) { 25818abb50c7SMiaohe Lin mapped_pte = pte = (mm == &init_mm) ? 2582e80d3909SJoerg Roedel pte_alloc_kernel_track(pmd, addr, mask) : 2583aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 2584aee16b3cSJeremy Fitzhardinge if (!pte) 2585aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2586be1db475SDaniel Axtens } else { 25878abb50c7SMiaohe Lin mapped_pte = pte = (mm == &init_mm) ? 2588be1db475SDaniel Axtens pte_offset_kernel(pmd, addr) : 2589be1db475SDaniel Axtens pte_offset_map_lock(mm, pmd, addr, &ptl); 25903db82b93SHugh Dickins if (!pte) 25913db82b93SHugh Dickins return -EINVAL; 2592be1db475SDaniel Axtens } 2593aee16b3cSJeremy Fitzhardinge 259438e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 259538e0edb1SJeremy Fitzhardinge 2596eeb4a05fSChristoph Hellwig if (fn) { 2597aee16b3cSJeremy Fitzhardinge do { 2598c33c7948SRyan Roberts if (create || !pte_none(ptep_get(pte))) { 25998b1e0f81SAnshuman Khandual err = fn(pte++, addr, data); 2600aee16b3cSJeremy Fitzhardinge if (err) 2601aee16b3cSJeremy Fitzhardinge break; 2602be1db475SDaniel Axtens } 2603c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 2604eeb4a05fSChristoph Hellwig } 2605e80d3909SJoerg Roedel *mask |= PGTBL_PTE_MODIFIED; 2606aee16b3cSJeremy Fitzhardinge 260738e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 260838e0edb1SJeremy Fitzhardinge 2609aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 26108abb50c7SMiaohe Lin pte_unmap_unlock(mapped_pte, ptl); 2611aee16b3cSJeremy Fitzhardinge return err; 2612aee16b3cSJeremy Fitzhardinge } 2613aee16b3cSJeremy Fitzhardinge 2614aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 2615aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2616e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2617e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2618aee16b3cSJeremy Fitzhardinge { 2619aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 2620aee16b3cSJeremy Fitzhardinge unsigned long next; 2621be1db475SDaniel Axtens int err = 0; 2622aee16b3cSJeremy Fitzhardinge 2623ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 2624ceb86879SAndi Kleen 2625be1db475SDaniel Axtens if (create) { 2626e80d3909SJoerg Roedel pmd = pmd_alloc_track(mm, pud, addr, mask); 2627aee16b3cSJeremy Fitzhardinge if (!pmd) 2628aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2629be1db475SDaniel Axtens } else { 2630be1db475SDaniel Axtens pmd = pmd_offset(pud, addr); 2631be1db475SDaniel Axtens } 2632aee16b3cSJeremy Fitzhardinge do { 2633aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 26340c95cba4SNicholas Piggin if (pmd_none(*pmd) && !create) 26350c95cba4SNicholas Piggin continue; 26360c95cba4SNicholas Piggin if (WARN_ON_ONCE(pmd_leaf(*pmd))) 26370c95cba4SNicholas Piggin return -EINVAL; 26380c95cba4SNicholas Piggin if (!pmd_none(*pmd) && WARN_ON_ONCE(pmd_bad(*pmd))) { 26390c95cba4SNicholas Piggin if (!create) 26400c95cba4SNicholas Piggin continue; 26410c95cba4SNicholas Piggin pmd_clear_bad(pmd); 26420c95cba4SNicholas Piggin } 26430c95cba4SNicholas Piggin err = apply_to_pte_range(mm, pmd, addr, next, 26440c95cba4SNicholas Piggin fn, data, create, mask); 2645aee16b3cSJeremy Fitzhardinge if (err) 2646aee16b3cSJeremy Fitzhardinge break; 2647aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 26480c95cba4SNicholas Piggin 2649aee16b3cSJeremy Fitzhardinge return err; 2650aee16b3cSJeremy Fitzhardinge } 2651aee16b3cSJeremy Fitzhardinge 2652c2febafcSKirill A. Shutemov static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, 2653aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2654e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2655e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2656aee16b3cSJeremy Fitzhardinge { 2657aee16b3cSJeremy Fitzhardinge pud_t *pud; 2658aee16b3cSJeremy Fitzhardinge unsigned long next; 2659be1db475SDaniel Axtens int err = 0; 2660aee16b3cSJeremy Fitzhardinge 2661be1db475SDaniel Axtens if (create) { 2662e80d3909SJoerg Roedel pud = pud_alloc_track(mm, p4d, addr, mask); 2663aee16b3cSJeremy Fitzhardinge if (!pud) 2664aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2665be1db475SDaniel Axtens } else { 2666be1db475SDaniel Axtens pud = pud_offset(p4d, addr); 2667be1db475SDaniel Axtens } 2668aee16b3cSJeremy Fitzhardinge do { 2669aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 26700c95cba4SNicholas Piggin if (pud_none(*pud) && !create) 26710c95cba4SNicholas Piggin continue; 26720c95cba4SNicholas Piggin if (WARN_ON_ONCE(pud_leaf(*pud))) 26730c95cba4SNicholas Piggin return -EINVAL; 26740c95cba4SNicholas Piggin if (!pud_none(*pud) && WARN_ON_ONCE(pud_bad(*pud))) { 26750c95cba4SNicholas Piggin if (!create) 26760c95cba4SNicholas Piggin continue; 26770c95cba4SNicholas Piggin pud_clear_bad(pud); 26780c95cba4SNicholas Piggin } 26790c95cba4SNicholas Piggin err = apply_to_pmd_range(mm, pud, addr, next, 26800c95cba4SNicholas Piggin fn, data, create, mask); 2681aee16b3cSJeremy Fitzhardinge if (err) 2682aee16b3cSJeremy Fitzhardinge break; 2683aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 26840c95cba4SNicholas Piggin 2685aee16b3cSJeremy Fitzhardinge return err; 2686aee16b3cSJeremy Fitzhardinge } 2687aee16b3cSJeremy Fitzhardinge 2688c2febafcSKirill A. Shutemov static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, 2689c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 2690e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2691e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2692c2febafcSKirill A. Shutemov { 2693c2febafcSKirill A. Shutemov p4d_t *p4d; 2694c2febafcSKirill A. Shutemov unsigned long next; 2695be1db475SDaniel Axtens int err = 0; 2696c2febafcSKirill A. Shutemov 2697be1db475SDaniel Axtens if (create) { 2698e80d3909SJoerg Roedel p4d = p4d_alloc_track(mm, pgd, addr, mask); 2699c2febafcSKirill A. Shutemov if (!p4d) 2700c2febafcSKirill A. Shutemov return -ENOMEM; 2701be1db475SDaniel Axtens } else { 2702be1db475SDaniel Axtens p4d = p4d_offset(pgd, addr); 2703be1db475SDaniel Axtens } 2704c2febafcSKirill A. Shutemov do { 2705c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 27060c95cba4SNicholas Piggin if (p4d_none(*p4d) && !create) 27070c95cba4SNicholas Piggin continue; 27080c95cba4SNicholas Piggin if (WARN_ON_ONCE(p4d_leaf(*p4d))) 27090c95cba4SNicholas Piggin return -EINVAL; 27100c95cba4SNicholas Piggin if (!p4d_none(*p4d) && WARN_ON_ONCE(p4d_bad(*p4d))) { 27110c95cba4SNicholas Piggin if (!create) 27120c95cba4SNicholas Piggin continue; 27130c95cba4SNicholas Piggin p4d_clear_bad(p4d); 27140c95cba4SNicholas Piggin } 27150c95cba4SNicholas Piggin err = apply_to_pud_range(mm, p4d, addr, next, 27160c95cba4SNicholas Piggin fn, data, create, mask); 2717c2febafcSKirill A. Shutemov if (err) 2718c2febafcSKirill A. Shutemov break; 2719c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 27200c95cba4SNicholas Piggin 2721c2febafcSKirill A. Shutemov return err; 2722c2febafcSKirill A. Shutemov } 2723c2febafcSKirill A. Shutemov 2724be1db475SDaniel Axtens static int __apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2725be1db475SDaniel Axtens unsigned long size, pte_fn_t fn, 2726be1db475SDaniel Axtens void *data, bool create) 2727be1db475SDaniel Axtens { 2728be1db475SDaniel Axtens pgd_t *pgd; 2729e80d3909SJoerg Roedel unsigned long start = addr, next; 2730be1db475SDaniel Axtens unsigned long end = addr + size; 2731e80d3909SJoerg Roedel pgtbl_mod_mask mask = 0; 2732be1db475SDaniel Axtens int err = 0; 2733be1db475SDaniel Axtens 2734be1db475SDaniel Axtens if (WARN_ON(addr >= end)) 2735be1db475SDaniel Axtens return -EINVAL; 2736be1db475SDaniel Axtens 2737be1db475SDaniel Axtens pgd = pgd_offset(mm, addr); 2738be1db475SDaniel Axtens do { 2739be1db475SDaniel Axtens next = pgd_addr_end(addr, end); 27400c95cba4SNicholas Piggin if (pgd_none(*pgd) && !create) 2741be1db475SDaniel Axtens continue; 27420c95cba4SNicholas Piggin if (WARN_ON_ONCE(pgd_leaf(*pgd))) 27430c95cba4SNicholas Piggin return -EINVAL; 27440c95cba4SNicholas Piggin if (!pgd_none(*pgd) && WARN_ON_ONCE(pgd_bad(*pgd))) { 27450c95cba4SNicholas Piggin if (!create) 27460c95cba4SNicholas Piggin continue; 27470c95cba4SNicholas Piggin pgd_clear_bad(pgd); 27480c95cba4SNicholas Piggin } 27490c95cba4SNicholas Piggin err = apply_to_p4d_range(mm, pgd, addr, next, 27500c95cba4SNicholas Piggin fn, data, create, &mask); 2751be1db475SDaniel Axtens if (err) 2752be1db475SDaniel Axtens break; 2753be1db475SDaniel Axtens } while (pgd++, addr = next, addr != end); 2754be1db475SDaniel Axtens 2755e80d3909SJoerg Roedel if (mask & ARCH_PAGE_TABLE_SYNC_MASK) 2756e80d3909SJoerg Roedel arch_sync_kernel_mappings(start, start + size); 2757e80d3909SJoerg Roedel 2758be1db475SDaniel Axtens return err; 2759be1db475SDaniel Axtens } 2760be1db475SDaniel Axtens 2761aee16b3cSJeremy Fitzhardinge /* 2762aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 2763aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 2764aee16b3cSJeremy Fitzhardinge */ 2765aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2766aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 2767aee16b3cSJeremy Fitzhardinge { 2768be1db475SDaniel Axtens return __apply_to_page_range(mm, addr, size, fn, data, true); 2769aee16b3cSJeremy Fitzhardinge } 2770aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 2771aee16b3cSJeremy Fitzhardinge 27721da177e4SLinus Torvalds /* 2773be1db475SDaniel Axtens * Scan a region of virtual memory, calling a provided function on 2774be1db475SDaniel Axtens * each leaf page table where it exists. 2775be1db475SDaniel Axtens * 2776be1db475SDaniel Axtens * Unlike apply_to_page_range, this does _not_ fill in page tables 2777be1db475SDaniel Axtens * where they are absent. 2778be1db475SDaniel Axtens */ 2779be1db475SDaniel Axtens int apply_to_existing_page_range(struct mm_struct *mm, unsigned long addr, 2780be1db475SDaniel Axtens unsigned long size, pte_fn_t fn, void *data) 2781be1db475SDaniel Axtens { 2782be1db475SDaniel Axtens return __apply_to_page_range(mm, addr, size, fn, data, false); 2783be1db475SDaniel Axtens } 2784be1db475SDaniel Axtens EXPORT_SYMBOL_GPL(apply_to_existing_page_range); 2785be1db475SDaniel Axtens 2786be1db475SDaniel Axtens /* 27879b4bdd2fSKirill A. Shutemov * handle_pte_fault chooses page fault handler according to an entry which was 27889b4bdd2fSKirill A. Shutemov * read non-atomically. Before making any commitment, on those architectures 27899b4bdd2fSKirill A. Shutemov * or configurations (e.g. i386 with PAE) which might give a mix of unmatched 27909b4bdd2fSKirill A. Shutemov * parts, do_swap_page must check under lock before unmapping the pte and 27919b4bdd2fSKirill A. Shutemov * proceeding (but do_wp_page is only called after already making such a check; 2792a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 27938f4e2101SHugh Dickins */ 27942ca99358SPeter Xu static inline int pte_unmap_same(struct vm_fault *vmf) 27958f4e2101SHugh Dickins { 27968f4e2101SHugh Dickins int same = 1; 2797923717cbSThomas Gleixner #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPTION) 27988f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 2799c7ad0880SHugh Dickins spin_lock(vmf->ptl); 2800c33c7948SRyan Roberts same = pte_same(ptep_get(vmf->pte), vmf->orig_pte); 2801c7ad0880SHugh Dickins spin_unlock(vmf->ptl); 28028f4e2101SHugh Dickins } 28038f4e2101SHugh Dickins #endif 28042ca99358SPeter Xu pte_unmap(vmf->pte); 28052ca99358SPeter Xu vmf->pte = NULL; 28068f4e2101SHugh Dickins return same; 28078f4e2101SHugh Dickins } 28088f4e2101SHugh Dickins 2809a873dfe1STony Luck /* 2810a873dfe1STony Luck * Return: 2811a873dfe1STony Luck * 0: copied succeeded 2812a873dfe1STony Luck * -EHWPOISON: copy failed due to hwpoison in source page 2813a873dfe1STony Luck * -EAGAIN: copied failed (some other reason) 2814a873dfe1STony Luck */ 2815a873dfe1STony Luck static inline int __wp_page_copy_user(struct page *dst, struct page *src, 281683d116c5SJia He struct vm_fault *vmf) 28176aab341eSLinus Torvalds { 2818a873dfe1STony Luck int ret; 281983d116c5SJia He void *kaddr; 282083d116c5SJia He void __user *uaddr; 282183d116c5SJia He struct vm_area_struct *vma = vmf->vma; 282283d116c5SJia He struct mm_struct *mm = vma->vm_mm; 282383d116c5SJia He unsigned long addr = vmf->address; 282483d116c5SJia He 282583d116c5SJia He if (likely(src)) { 2826d302c239STony Luck if (copy_mc_user_highpage(dst, src, addr, vma)) { 2827d302c239STony Luck memory_failure_queue(page_to_pfn(src), 0); 2828a873dfe1STony Luck return -EHWPOISON; 2829d302c239STony Luck } 2830a873dfe1STony Luck return 0; 283183d116c5SJia He } 283283d116c5SJia He 28336aab341eSLinus Torvalds /* 28346aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 28356aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 28366aab341eSLinus Torvalds * just copying from the original user address. If that 28376aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 28386aab341eSLinus Torvalds */ 283983d116c5SJia He kaddr = kmap_atomic(dst); 284083d116c5SJia He uaddr = (void __user *)(addr & PAGE_MASK); 284183d116c5SJia He 284283d116c5SJia He /* 284383d116c5SJia He * On architectures with software "accessed" bits, we would 284483d116c5SJia He * take a double page fault, so mark it accessed here. 284583d116c5SJia He */ 28463db82b93SHugh Dickins vmf->pte = NULL; 2847e1fd09e3SYu Zhao if (!arch_has_hw_pte_young() && !pte_young(vmf->orig_pte)) { 284883d116c5SJia He pte_t entry; 284983d116c5SJia He 285083d116c5SJia He vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); 2851c33c7948SRyan Roberts if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 285283d116c5SJia He /* 285383d116c5SJia He * Other thread has already handled the fault 28547df67697SBibo Mao * and update local tlb only 285583d116c5SJia He */ 2856a92cbb82SHugh Dickins if (vmf->pte) 28577df67697SBibo Mao update_mmu_tlb(vma, addr, vmf->pte); 2858a873dfe1STony Luck ret = -EAGAIN; 285983d116c5SJia He goto pte_unlock; 286083d116c5SJia He } 286183d116c5SJia He 286283d116c5SJia He entry = pte_mkyoung(vmf->orig_pte); 286383d116c5SJia He if (ptep_set_access_flags(vma, addr, vmf->pte, entry, 0)) 286483d116c5SJia He update_mmu_cache(vma, addr, vmf->pte); 286583d116c5SJia He } 28665d2a2dbbSLinus Torvalds 28675d2a2dbbSLinus Torvalds /* 28685d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 28695d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 28705d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 28715d2a2dbbSLinus Torvalds * zeroes. 28725d2a2dbbSLinus Torvalds */ 287383d116c5SJia He if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { 28743db82b93SHugh Dickins if (vmf->pte) 2875c3e5ea6eSKirill A. Shutemov goto warn; 2876c3e5ea6eSKirill A. Shutemov 2877c3e5ea6eSKirill A. Shutemov /* Re-validate under PTL if the page is still mapped */ 2878c3e5ea6eSKirill A. Shutemov vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); 2879c33c7948SRyan Roberts if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 28807df67697SBibo Mao /* The PTE changed under us, update local tlb */ 2881a92cbb82SHugh Dickins if (vmf->pte) 28827df67697SBibo Mao update_mmu_tlb(vma, addr, vmf->pte); 2883a873dfe1STony Luck ret = -EAGAIN; 2884c3e5ea6eSKirill A. Shutemov goto pte_unlock; 2885c3e5ea6eSKirill A. Shutemov } 2886c3e5ea6eSKirill A. Shutemov 2887c3e5ea6eSKirill A. Shutemov /* 2888985ba004SEthon Paul * The same page can be mapped back since last copy attempt. 2889c3e5ea6eSKirill A. Shutemov * Try to copy again under PTL. 2890c3e5ea6eSKirill A. Shutemov */ 2891c3e5ea6eSKirill A. Shutemov if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { 289283d116c5SJia He /* 289383d116c5SJia He * Give a warn in case there can be some obscure 289483d116c5SJia He * use-case 289583d116c5SJia He */ 2896c3e5ea6eSKirill A. Shutemov warn: 289783d116c5SJia He WARN_ON_ONCE(1); 28983ecb01dfSJan Beulich clear_page(kaddr); 289983d116c5SJia He } 2900c3e5ea6eSKirill A. Shutemov } 290183d116c5SJia He 2902a873dfe1STony Luck ret = 0; 290383d116c5SJia He 290483d116c5SJia He pte_unlock: 29053db82b93SHugh Dickins if (vmf->pte) 290683d116c5SJia He pte_unmap_unlock(vmf->pte, vmf->ptl); 29079b04c5feSCong Wang kunmap_atomic(kaddr); 2908c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 290983d116c5SJia He 291083d116c5SJia He return ret; 29116aab341eSLinus Torvalds } 29126aab341eSLinus Torvalds 2913c20cd45eSMichal Hocko static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) 2914c20cd45eSMichal Hocko { 2915c20cd45eSMichal Hocko struct file *vm_file = vma->vm_file; 2916c20cd45eSMichal Hocko 2917c20cd45eSMichal Hocko if (vm_file) 2918c20cd45eSMichal Hocko return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; 2919c20cd45eSMichal Hocko 2920c20cd45eSMichal Hocko /* 2921c20cd45eSMichal Hocko * Special mappings (e.g. VDSO) do not have any file so fake 2922c20cd45eSMichal Hocko * a default GFP_KERNEL for them. 2923c20cd45eSMichal Hocko */ 2924c20cd45eSMichal Hocko return GFP_KERNEL; 2925c20cd45eSMichal Hocko } 2926c20cd45eSMichal Hocko 29271da177e4SLinus Torvalds /* 2928fb09a464SKirill A. Shutemov * Notify the address space that the page is about to become writable so that 2929fb09a464SKirill A. Shutemov * it can prohibit this or wait for the page to get into an appropriate state. 2930fb09a464SKirill A. Shutemov * 2931fb09a464SKirill A. Shutemov * We do this without the lock held, so that it can sleep if it needs to. 2932fb09a464SKirill A. Shutemov */ 293386aa6998SSidhartha Kumar static vm_fault_t do_page_mkwrite(struct vm_fault *vmf, struct folio *folio) 2934fb09a464SKirill A. Shutemov { 29352b740303SSouptick Joarder vm_fault_t ret; 293638b8cb7fSJan Kara unsigned int old_flags = vmf->flags; 2937fb09a464SKirill A. Shutemov 293838b8cb7fSJan Kara vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 2939fb09a464SKirill A. Shutemov 2940dc617f29SDarrick J. Wong if (vmf->vma->vm_file && 2941dc617f29SDarrick J. Wong IS_SWAPFILE(vmf->vma->vm_file->f_mapping->host)) 2942dc617f29SDarrick J. Wong return VM_FAULT_SIGBUS; 2943dc617f29SDarrick J. Wong 294411bac800SDave Jiang ret = vmf->vma->vm_ops->page_mkwrite(vmf); 294538b8cb7fSJan Kara /* Restore original flags so that caller is not surprised */ 294638b8cb7fSJan Kara vmf->flags = old_flags; 2947fb09a464SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 2948fb09a464SKirill A. Shutemov return ret; 2949fb09a464SKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) { 29503d243659SSidhartha Kumar folio_lock(folio); 29513d243659SSidhartha Kumar if (!folio->mapping) { 29523d243659SSidhartha Kumar folio_unlock(folio); 2953fb09a464SKirill A. Shutemov return 0; /* retry */ 2954fb09a464SKirill A. Shutemov } 2955fb09a464SKirill A. Shutemov ret |= VM_FAULT_LOCKED; 2956fb09a464SKirill A. Shutemov } else 29573d243659SSidhartha Kumar VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); 2958fb09a464SKirill A. Shutemov return ret; 2959fb09a464SKirill A. Shutemov } 2960fb09a464SKirill A. Shutemov 2961fb09a464SKirill A. Shutemov /* 296297ba0c2bSJan Kara * Handle dirtying of a page in shared file mapping on a write fault. 29634e047f89SShachar Raindel * 296497ba0c2bSJan Kara * The function expects the page to be locked and unlocks it. 29654e047f89SShachar Raindel */ 296689b15332SJohannes Weiner static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf) 29674e047f89SShachar Raindel { 296889b15332SJohannes Weiner struct vm_area_struct *vma = vmf->vma; 29694e047f89SShachar Raindel struct address_space *mapping; 297015b4919aSZhangPeng struct folio *folio = page_folio(vmf->page); 297197ba0c2bSJan Kara bool dirtied; 297297ba0c2bSJan Kara bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; 29734e047f89SShachar Raindel 297415b4919aSZhangPeng dirtied = folio_mark_dirty(folio); 297515b4919aSZhangPeng VM_BUG_ON_FOLIO(folio_test_anon(folio), folio); 297697ba0c2bSJan Kara /* 297715b4919aSZhangPeng * Take a local copy of the address_space - folio.mapping may be zeroed 297815b4919aSZhangPeng * by truncate after folio_unlock(). The address_space itself remains 297915b4919aSZhangPeng * pinned by vma->vm_file's reference. We rely on folio_unlock()'s 298097ba0c2bSJan Kara * release semantics to prevent the compiler from undoing this copying. 298197ba0c2bSJan Kara */ 298215b4919aSZhangPeng mapping = folio_raw_mapping(folio); 298315b4919aSZhangPeng folio_unlock(folio); 29844e047f89SShachar Raindel 29854e047f89SShachar Raindel if (!page_mkwrite) 29864e047f89SShachar Raindel file_update_time(vma->vm_file); 298789b15332SJohannes Weiner 298889b15332SJohannes Weiner /* 298989b15332SJohannes Weiner * Throttle page dirtying rate down to writeback speed. 299089b15332SJohannes Weiner * 299189b15332SJohannes Weiner * mapping may be NULL here because some device drivers do not 299289b15332SJohannes Weiner * set page.mapping but still dirty their pages 299389b15332SJohannes Weiner * 2994c1e8d7c6SMichel Lespinasse * Drop the mmap_lock before waiting on IO, if we can. The file 299589b15332SJohannes Weiner * is pinning the mapping, as per above. 299689b15332SJohannes Weiner */ 299789b15332SJohannes Weiner if ((dirtied || page_mkwrite) && mapping) { 299889b15332SJohannes Weiner struct file *fpin; 299989b15332SJohannes Weiner 300089b15332SJohannes Weiner fpin = maybe_unlock_mmap_for_io(vmf, NULL); 300189b15332SJohannes Weiner balance_dirty_pages_ratelimited(mapping); 300289b15332SJohannes Weiner if (fpin) { 300389b15332SJohannes Weiner fput(fpin); 3004d9272525SPeter Xu return VM_FAULT_COMPLETED; 300589b15332SJohannes Weiner } 300689b15332SJohannes Weiner } 300789b15332SJohannes Weiner 300889b15332SJohannes Weiner return 0; 30094e047f89SShachar Raindel } 30104e047f89SShachar Raindel 301197ba0c2bSJan Kara /* 30124e047f89SShachar Raindel * Handle write page faults for pages that can be reused in the current vma 30134e047f89SShachar Raindel * 30144e047f89SShachar Raindel * This can happen either due to the mapping being with the VM_SHARED flag, 30154e047f89SShachar Raindel * or due to us being the last reference standing to the page. In either 30164e047f89SShachar Raindel * case, all we need to do here is to mark the page as writable and update 30174e047f89SShachar Raindel * any related book-keeping. 30184e047f89SShachar Raindel */ 3019997dd98dSJan Kara static inline void wp_page_reuse(struct vm_fault *vmf) 302082b0f8c3SJan Kara __releases(vmf->ptl) 30214e047f89SShachar Raindel { 302282b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3023a41b70d6SJan Kara struct page *page = vmf->page; 30244e047f89SShachar Raindel pte_t entry; 30256c287605SDavid Hildenbrand 3026c89357e2SDavid Hildenbrand VM_BUG_ON(!(vmf->flags & FAULT_FLAG_WRITE)); 3027cdb281e6SQi Zheng VM_BUG_ON(page && PageAnon(page) && !PageAnonExclusive(page)); 30286c287605SDavid Hildenbrand 30294e047f89SShachar Raindel /* 30304e047f89SShachar Raindel * Clear the pages cpupid information as the existing 30314e047f89SShachar Raindel * information potentially belongs to a now completely 30324e047f89SShachar Raindel * unrelated process. 30334e047f89SShachar Raindel */ 30344e047f89SShachar Raindel if (page) 30354e047f89SShachar Raindel page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); 30364e047f89SShachar Raindel 30372994302bSJan Kara flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); 30382994302bSJan Kara entry = pte_mkyoung(vmf->orig_pte); 30394e047f89SShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 304082b0f8c3SJan Kara if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) 304182b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 304282b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3043798a6b87SPeter Xu count_vm_event(PGREUSE); 30444e047f89SShachar Raindel } 30454e047f89SShachar Raindel 30464e047f89SShachar Raindel /* 3047c89357e2SDavid Hildenbrand * Handle the case of a page which we actually need to copy to a new page, 3048c89357e2SDavid Hildenbrand * either due to COW or unsharing. 30492f38ab2cSShachar Raindel * 3050c1e8d7c6SMichel Lespinasse * Called with mmap_lock locked and the old page referenced, but 30512f38ab2cSShachar Raindel * without the ptl held. 30522f38ab2cSShachar Raindel * 30532f38ab2cSShachar Raindel * High level logic flow: 30542f38ab2cSShachar Raindel * 30552f38ab2cSShachar Raindel * - Allocate a page, copy the content of the old page to the new one. 30562f38ab2cSShachar Raindel * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. 30572f38ab2cSShachar Raindel * - Take the PTL. If the pte changed, bail out and release the allocated page 30582f38ab2cSShachar Raindel * - If the pte is still the way we remember it, update the page table and all 30592f38ab2cSShachar Raindel * relevant references. This includes dropping the reference the page-table 30602f38ab2cSShachar Raindel * held to the old page, as well as updating the rmap. 30612f38ab2cSShachar Raindel * - In any case, unlock the PTL and drop the reference we took to the old page. 30622f38ab2cSShachar Raindel */ 30632b740303SSouptick Joarder static vm_fault_t wp_page_copy(struct vm_fault *vmf) 30642f38ab2cSShachar Raindel { 3065c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 306682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3067bae473a4SKirill A. Shutemov struct mm_struct *mm = vma->vm_mm; 306828d41a48SMatthew Wilcox (Oracle) struct folio *old_folio = NULL; 306928d41a48SMatthew Wilcox (Oracle) struct folio *new_folio = NULL; 30702f38ab2cSShachar Raindel pte_t entry; 30712f38ab2cSShachar Raindel int page_copied = 0; 3072ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 3073a873dfe1STony Luck int ret; 30742f38ab2cSShachar Raindel 3075662ce1dcSYang Yang delayacct_wpcopy_start(); 3076662ce1dcSYang Yang 307728d41a48SMatthew Wilcox (Oracle) if (vmf->page) 307828d41a48SMatthew Wilcox (Oracle) old_folio = page_folio(vmf->page); 30792f38ab2cSShachar Raindel if (unlikely(anon_vma_prepare(vma))) 30802f38ab2cSShachar Raindel goto oom; 30812f38ab2cSShachar Raindel 30822994302bSJan Kara if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { 30836bc56a4dSMatthew Wilcox (Oracle) new_folio = vma_alloc_zeroed_movable_folio(vma, vmf->address); 30846bc56a4dSMatthew Wilcox (Oracle) if (!new_folio) 30852f38ab2cSShachar Raindel goto oom; 30862f38ab2cSShachar Raindel } else { 308728d41a48SMatthew Wilcox (Oracle) new_folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma, 308828d41a48SMatthew Wilcox (Oracle) vmf->address, false); 308928d41a48SMatthew Wilcox (Oracle) if (!new_folio) 30902f38ab2cSShachar Raindel goto oom; 309183d116c5SJia He 309228d41a48SMatthew Wilcox (Oracle) ret = __wp_page_copy_user(&new_folio->page, vmf->page, vmf); 3093a873dfe1STony Luck if (ret) { 309483d116c5SJia He /* 309583d116c5SJia He * COW failed, if the fault was solved by other, 309683d116c5SJia He * it's fine. If not, userspace would re-fault on 309783d116c5SJia He * the same address and we will handle the fault 309883d116c5SJia He * from the second attempt. 3099a873dfe1STony Luck * The -EHWPOISON case will not be retried. 310083d116c5SJia He */ 310128d41a48SMatthew Wilcox (Oracle) folio_put(new_folio); 310228d41a48SMatthew Wilcox (Oracle) if (old_folio) 310328d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 3104662ce1dcSYang Yang 3105662ce1dcSYang Yang delayacct_wpcopy_end(); 3106a873dfe1STony Luck return ret == -EHWPOISON ? VM_FAULT_HWPOISON : 0; 310783d116c5SJia He } 310828d41a48SMatthew Wilcox (Oracle) kmsan_copy_page_meta(&new_folio->page, vmf->page); 31092f38ab2cSShachar Raindel } 31102f38ab2cSShachar Raindel 311128d41a48SMatthew Wilcox (Oracle) if (mem_cgroup_charge(new_folio, mm, GFP_KERNEL)) 31122f38ab2cSShachar Raindel goto oom_free_new; 31134d4f75bfSKefeng Wang folio_throttle_swaprate(new_folio, GFP_KERNEL); 31142f38ab2cSShachar Raindel 311528d41a48SMatthew Wilcox (Oracle) __folio_mark_uptodate(new_folio); 3116eb3c24f3SMel Gorman 31177d4a8be0SAlistair Popple mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, mm, 31186f4f13e8SJérôme Glisse vmf->address & PAGE_MASK, 3119ac46d4f3SJérôme Glisse (vmf->address & PAGE_MASK) + PAGE_SIZE); 3120ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 31212f38ab2cSShachar Raindel 31222f38ab2cSShachar Raindel /* 31232f38ab2cSShachar Raindel * Re-check the pte - we dropped the lock 31242f38ab2cSShachar Raindel */ 312582b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); 3126c33c7948SRyan Roberts if (likely(vmf->pte && pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 312728d41a48SMatthew Wilcox (Oracle) if (old_folio) { 312828d41a48SMatthew Wilcox (Oracle) if (!folio_test_anon(old_folio)) { 312928d41a48SMatthew Wilcox (Oracle) dec_mm_counter(mm, mm_counter_file(&old_folio->page)); 3130f1a79412SShakeel Butt inc_mm_counter(mm, MM_ANONPAGES); 31312f38ab2cSShachar Raindel } 31322f38ab2cSShachar Raindel } else { 31336080d19fSxu xin ksm_might_unmap_zero_page(mm, vmf->orig_pte); 3134f1a79412SShakeel Butt inc_mm_counter(mm, MM_ANONPAGES); 31352f38ab2cSShachar Raindel } 31362994302bSJan Kara flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); 313728d41a48SMatthew Wilcox (Oracle) entry = mk_pte(&new_folio->page, vma->vm_page_prot); 313850c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 3139c89357e2SDavid Hildenbrand if (unlikely(unshare)) { 3140c89357e2SDavid Hildenbrand if (pte_soft_dirty(vmf->orig_pte)) 3141c89357e2SDavid Hildenbrand entry = pte_mksoft_dirty(entry); 3142c89357e2SDavid Hildenbrand if (pte_uffd_wp(vmf->orig_pte)) 3143c89357e2SDavid Hildenbrand entry = pte_mkuffd_wp(entry); 3144c89357e2SDavid Hildenbrand } else { 31452f38ab2cSShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 3146c89357e2SDavid Hildenbrand } 3147111fe718SNicholas Piggin 31482f38ab2cSShachar Raindel /* 31492f38ab2cSShachar Raindel * Clear the pte entry and flush it first, before updating the 3150111fe718SNicholas Piggin * pte with the new entry, to keep TLBs on different CPUs in 3151111fe718SNicholas Piggin * sync. This code used to set the new PTE then flush TLBs, but 3152111fe718SNicholas Piggin * that left a window where the new PTE could be loaded into 3153111fe718SNicholas Piggin * some TLBs while the old PTE remains in others. 31542f38ab2cSShachar Raindel */ 3155ec8832d0SAlistair Popple ptep_clear_flush(vma, vmf->address, vmf->pte); 315628d41a48SMatthew Wilcox (Oracle) folio_add_new_anon_rmap(new_folio, vma, vmf->address); 315728d41a48SMatthew Wilcox (Oracle) folio_add_lru_vma(new_folio, vma); 31582f38ab2cSShachar Raindel /* 31592f38ab2cSShachar Raindel * We call the notify macro here because, when using secondary 31602f38ab2cSShachar Raindel * mmu page tables (such as kvm shadow page tables), we want the 31612f38ab2cSShachar Raindel * new page to be mapped directly into the secondary page table. 31622f38ab2cSShachar Raindel */ 3163c89357e2SDavid Hildenbrand BUG_ON(unshare && pte_write(entry)); 316482b0f8c3SJan Kara set_pte_at_notify(mm, vmf->address, vmf->pte, entry); 316582b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 316628d41a48SMatthew Wilcox (Oracle) if (old_folio) { 31672f38ab2cSShachar Raindel /* 31682f38ab2cSShachar Raindel * Only after switching the pte to the new page may 31692f38ab2cSShachar Raindel * we remove the mapcount here. Otherwise another 31702f38ab2cSShachar Raindel * process may come and find the rmap count decremented 31712f38ab2cSShachar Raindel * before the pte is switched to the new page, and 31722f38ab2cSShachar Raindel * "reuse" the old page writing into it while our pte 31732f38ab2cSShachar Raindel * here still points into it and can be read by other 31742f38ab2cSShachar Raindel * threads. 31752f38ab2cSShachar Raindel * 31762f38ab2cSShachar Raindel * The critical issue is to order this 31772f38ab2cSShachar Raindel * page_remove_rmap with the ptp_clear_flush above. 31782f38ab2cSShachar Raindel * Those stores are ordered by (if nothing else,) 31792f38ab2cSShachar Raindel * the barrier present in the atomic_add_negative 31802f38ab2cSShachar Raindel * in page_remove_rmap. 31812f38ab2cSShachar Raindel * 31822f38ab2cSShachar Raindel * Then the TLB flush in ptep_clear_flush ensures that 31832f38ab2cSShachar Raindel * no process can access the old page before the 31842f38ab2cSShachar Raindel * decremented mapcount is visible. And the old page 31852f38ab2cSShachar Raindel * cannot be reused until after the decremented 31862f38ab2cSShachar Raindel * mapcount is visible. So transitively, TLBs to 31872f38ab2cSShachar Raindel * old page will be flushed before it can be reused. 31882f38ab2cSShachar Raindel */ 318928d41a48SMatthew Wilcox (Oracle) page_remove_rmap(vmf->page, vma, false); 31902f38ab2cSShachar Raindel } 31912f38ab2cSShachar Raindel 31922f38ab2cSShachar Raindel /* Free the old page.. */ 319328d41a48SMatthew Wilcox (Oracle) new_folio = old_folio; 31942f38ab2cSShachar Raindel page_copied = 1; 31953db82b93SHugh Dickins pte_unmap_unlock(vmf->pte, vmf->ptl); 31963db82b93SHugh Dickins } else if (vmf->pte) { 31977df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 31983db82b93SHugh Dickins pte_unmap_unlock(vmf->pte, vmf->ptl); 31992f38ab2cSShachar Raindel } 32002f38ab2cSShachar Raindel 3201ec8832d0SAlistair Popple mmu_notifier_invalidate_range_end(&range); 32023db82b93SHugh Dickins 32033db82b93SHugh Dickins if (new_folio) 32043db82b93SHugh Dickins folio_put(new_folio); 320528d41a48SMatthew Wilcox (Oracle) if (old_folio) { 3206f4c4a3f4SHuang Ying if (page_copied) 320728d41a48SMatthew Wilcox (Oracle) free_swap_cache(&old_folio->page); 320828d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 32092f38ab2cSShachar Raindel } 3210662ce1dcSYang Yang 3211662ce1dcSYang Yang delayacct_wpcopy_end(); 3212cb8d8633SDavid Hildenbrand return 0; 32132f38ab2cSShachar Raindel oom_free_new: 321428d41a48SMatthew Wilcox (Oracle) folio_put(new_folio); 32152f38ab2cSShachar Raindel oom: 321628d41a48SMatthew Wilcox (Oracle) if (old_folio) 321728d41a48SMatthew Wilcox (Oracle) folio_put(old_folio); 3218662ce1dcSYang Yang 3219662ce1dcSYang Yang delayacct_wpcopy_end(); 32202f38ab2cSShachar Raindel return VM_FAULT_OOM; 32212f38ab2cSShachar Raindel } 32222f38ab2cSShachar Raindel 322366a6197cSJan Kara /** 322466a6197cSJan Kara * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE 322566a6197cSJan Kara * writeable once the page is prepared 322666a6197cSJan Kara * 322766a6197cSJan Kara * @vmf: structure describing the fault 322866a6197cSJan Kara * 322966a6197cSJan Kara * This function handles all that is needed to finish a write page fault in a 323066a6197cSJan Kara * shared mapping due to PTE being read-only once the mapped page is prepared. 3231a862f68aSMike Rapoport * It handles locking of PTE and modifying it. 323266a6197cSJan Kara * 323366a6197cSJan Kara * The function expects the page to be locked or other protection against 323466a6197cSJan Kara * concurrent faults / writeback (such as DAX radix tree locks). 3235a862f68aSMike Rapoport * 32362797e79fSLiu Xiang * Return: %0 on success, %VM_FAULT_NOPAGE when PTE got changed before 3237a862f68aSMike Rapoport * we acquired PTE lock. 323866a6197cSJan Kara */ 32392b740303SSouptick Joarder vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf) 324066a6197cSJan Kara { 324166a6197cSJan Kara WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); 324266a6197cSJan Kara vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, 324366a6197cSJan Kara &vmf->ptl); 32443db82b93SHugh Dickins if (!vmf->pte) 32453db82b93SHugh Dickins return VM_FAULT_NOPAGE; 324666a6197cSJan Kara /* 324766a6197cSJan Kara * We might have raced with another page fault while we released the 324866a6197cSJan Kara * pte_offset_map_lock. 324966a6197cSJan Kara */ 3250c33c7948SRyan Roberts if (!pte_same(ptep_get(vmf->pte), vmf->orig_pte)) { 32517df67697SBibo Mao update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 325266a6197cSJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3253a19e2553SJan Kara return VM_FAULT_NOPAGE; 325466a6197cSJan Kara } 325566a6197cSJan Kara wp_page_reuse(vmf); 3256a19e2553SJan Kara return 0; 325766a6197cSJan Kara } 325866a6197cSJan Kara 3259dd906184SBoaz Harrosh /* 3260dd906184SBoaz Harrosh * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED 3261dd906184SBoaz Harrosh * mapping 3262dd906184SBoaz Harrosh */ 32632b740303SSouptick Joarder static vm_fault_t wp_pfn_shared(struct vm_fault *vmf) 3264dd906184SBoaz Harrosh { 326582b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3266bae473a4SKirill A. Shutemov 3267dd906184SBoaz Harrosh if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { 32682b740303SSouptick Joarder vm_fault_t ret; 3269dd906184SBoaz Harrosh 327082b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3271fe82221fSJan Kara vmf->flags |= FAULT_FLAG_MKWRITE; 327211bac800SDave Jiang ret = vma->vm_ops->pfn_mkwrite(vmf); 32732f89dc12SJan Kara if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) 3274dd906184SBoaz Harrosh return ret; 327566a6197cSJan Kara return finish_mkwrite_fault(vmf); 3276dd906184SBoaz Harrosh } 3277997dd98dSJan Kara wp_page_reuse(vmf); 3278cb8d8633SDavid Hildenbrand return 0; 3279dd906184SBoaz Harrosh } 3280dd906184SBoaz Harrosh 32815a97858bSSidhartha Kumar static vm_fault_t wp_page_shared(struct vm_fault *vmf, struct folio *folio) 328282b0f8c3SJan Kara __releases(vmf->ptl) 328393e478d4SShachar Raindel { 328482b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3285cb8d8633SDavid Hildenbrand vm_fault_t ret = 0; 328693e478d4SShachar Raindel 32875a97858bSSidhartha Kumar folio_get(folio); 328893e478d4SShachar Raindel 328993e478d4SShachar Raindel if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 32902b740303SSouptick Joarder vm_fault_t tmp; 329193e478d4SShachar Raindel 329282b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 329386aa6998SSidhartha Kumar tmp = do_page_mkwrite(vmf, folio); 329493e478d4SShachar Raindel if (unlikely(!tmp || (tmp & 329593e478d4SShachar Raindel (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 32965a97858bSSidhartha Kumar folio_put(folio); 329793e478d4SShachar Raindel return tmp; 329893e478d4SShachar Raindel } 329966a6197cSJan Kara tmp = finish_mkwrite_fault(vmf); 3300a19e2553SJan Kara if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { 33015a97858bSSidhartha Kumar folio_unlock(folio); 33025a97858bSSidhartha Kumar folio_put(folio); 330366a6197cSJan Kara return tmp; 330493e478d4SShachar Raindel } 330566a6197cSJan Kara } else { 3306997dd98dSJan Kara wp_page_reuse(vmf); 33075a97858bSSidhartha Kumar folio_lock(folio); 330893e478d4SShachar Raindel } 330989b15332SJohannes Weiner ret |= fault_dirty_shared_page(vmf); 33105a97858bSSidhartha Kumar folio_put(folio); 331193e478d4SShachar Raindel 331289b15332SJohannes Weiner return ret; 331393e478d4SShachar Raindel } 331493e478d4SShachar Raindel 33152f38ab2cSShachar Raindel /* 3316c89357e2SDavid Hildenbrand * This routine handles present pages, when 3317c89357e2SDavid Hildenbrand * * users try to write to a shared page (FAULT_FLAG_WRITE) 3318c89357e2SDavid Hildenbrand * * GUP wants to take a R/O pin on a possibly shared anonymous page 3319c89357e2SDavid Hildenbrand * (FAULT_FLAG_UNSHARE) 3320c89357e2SDavid Hildenbrand * 3321c89357e2SDavid Hildenbrand * It is done by copying the page to a new address and decrementing the 3322c89357e2SDavid Hildenbrand * shared-page counter for the old page. 33231da177e4SLinus Torvalds * 33241da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 33251da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 3326c89357e2SDavid Hildenbrand * Thus, with FAULT_FLAG_WRITE, we can safely just mark it writable once we've 3327c89357e2SDavid Hildenbrand * done any necessary COW. 33281da177e4SLinus Torvalds * 3329c89357e2SDavid Hildenbrand * In case of FAULT_FLAG_WRITE, we also mark the page dirty at this point even 3330c89357e2SDavid Hildenbrand * though the page will change only once the write actually happens. This 3331c89357e2SDavid Hildenbrand * avoids a few races, and potentially makes it more efficient. 33321da177e4SLinus Torvalds * 3333c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 33348f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 3335c1e8d7c6SMichel Lespinasse * We return with mmap_lock still held, but pte unmapped and unlocked. 33361da177e4SLinus Torvalds */ 33372b740303SSouptick Joarder static vm_fault_t do_wp_page(struct vm_fault *vmf) 333882b0f8c3SJan Kara __releases(vmf->ptl) 33391da177e4SLinus Torvalds { 3340c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 334182b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3342b9086fdeSDavid Hildenbrand struct folio *folio = NULL; 33431da177e4SLinus Torvalds 3344c89357e2SDavid Hildenbrand if (likely(!unshare)) { 3345c33c7948SRyan Roberts if (userfaultfd_pte_wp(vma, ptep_get(vmf->pte))) { 3346529b930bSAndrea Arcangeli pte_unmap_unlock(vmf->pte, vmf->ptl); 3347529b930bSAndrea Arcangeli return handle_userfault(vmf, VM_UFFD_WP); 3348529b930bSAndrea Arcangeli } 3349529b930bSAndrea Arcangeli 33506ce64428SNadav Amit /* 33516ce64428SNadav Amit * Userfaultfd write-protect can defer flushes. Ensure the TLB 33526ce64428SNadav Amit * is flushed in this case before copying. 33536ce64428SNadav Amit */ 33546ce64428SNadav Amit if (unlikely(userfaultfd_wp(vmf->vma) && 33556ce64428SNadav Amit mm_tlb_flush_pending(vmf->vma->vm_mm))) 33566ce64428SNadav Amit flush_tlb_page(vmf->vma, vmf->address); 3357c89357e2SDavid Hildenbrand } 33586ce64428SNadav Amit 3359a41b70d6SJan Kara vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); 3360c89357e2SDavid Hildenbrand 33615a97858bSSidhartha Kumar if (vmf->page) 33625a97858bSSidhartha Kumar folio = page_folio(vmf->page); 33635a97858bSSidhartha Kumar 3364251b97f5SPeter Zijlstra /* 3365b9086fdeSDavid Hildenbrand * Shared mapping: we are guaranteed to have VM_WRITE and 3366b9086fdeSDavid Hildenbrand * FAULT_FLAG_WRITE set at this point. 3367b9086fdeSDavid Hildenbrand */ 3368b9086fdeSDavid Hildenbrand if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 3369b9086fdeSDavid Hildenbrand /* 337064e45507SPeter Feiner * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a 337164e45507SPeter Feiner * VM_PFNMAP VMA. 3372251b97f5SPeter Zijlstra * 3373251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 3374dd906184SBoaz Harrosh * Just mark the pages writable and/or call ops->pfn_mkwrite. 3375251b97f5SPeter Zijlstra */ 3376b9086fdeSDavid Hildenbrand if (!vmf->page) 33772994302bSJan Kara return wp_pfn_shared(vmf); 33785a97858bSSidhartha Kumar return wp_page_shared(vmf, folio); 3379251b97f5SPeter Zijlstra } 33801da177e4SLinus Torvalds 3381b9086fdeSDavid Hildenbrand /* 3382b9086fdeSDavid Hildenbrand * Private mapping: create an exclusive anonymous page copy if reuse 3383b9086fdeSDavid Hildenbrand * is impossible. We might miss VM_WRITE for FOLL_FORCE handling. 3384b9086fdeSDavid Hildenbrand */ 3385b9086fdeSDavid Hildenbrand if (folio && folio_test_anon(folio)) { 338653a05ad9SDavid Hildenbrand /* 33876c287605SDavid Hildenbrand * If the page is exclusive to this process we must reuse the 33886c287605SDavid Hildenbrand * page without further checks. 33896c287605SDavid Hildenbrand */ 3390e4a2ed94SMatthew Wilcox (Oracle) if (PageAnonExclusive(vmf->page)) 33916c287605SDavid Hildenbrand goto reuse; 33926c287605SDavid Hildenbrand 33936c287605SDavid Hildenbrand /* 3394e4a2ed94SMatthew Wilcox (Oracle) * We have to verify under folio lock: these early checks are 3395e4a2ed94SMatthew Wilcox (Oracle) * just an optimization to avoid locking the folio and freeing 339653a05ad9SDavid Hildenbrand * the swapcache if there is little hope that we can reuse. 339753a05ad9SDavid Hildenbrand * 3398e4a2ed94SMatthew Wilcox (Oracle) * KSM doesn't necessarily raise the folio refcount. 339953a05ad9SDavid Hildenbrand */ 3400e4a2ed94SMatthew Wilcox (Oracle) if (folio_test_ksm(folio) || folio_ref_count(folio) > 3) 3401d4c47097SDavid Hildenbrand goto copy; 3402e4a2ed94SMatthew Wilcox (Oracle) if (!folio_test_lru(folio)) 3403d4c47097SDavid Hildenbrand /* 34041fec6890SMatthew Wilcox (Oracle) * We cannot easily detect+handle references from 34051fec6890SMatthew Wilcox (Oracle) * remote LRU caches or references to LRU folios. 3406d4c47097SDavid Hildenbrand */ 3407d4c47097SDavid Hildenbrand lru_add_drain(); 3408e4a2ed94SMatthew Wilcox (Oracle) if (folio_ref_count(folio) > 1 + folio_test_swapcache(folio)) 340952d1e606SKirill Tkhai goto copy; 3410e4a2ed94SMatthew Wilcox (Oracle) if (!folio_trylock(folio)) 341152d1e606SKirill Tkhai goto copy; 3412e4a2ed94SMatthew Wilcox (Oracle) if (folio_test_swapcache(folio)) 3413e4a2ed94SMatthew Wilcox (Oracle) folio_free_swap(folio); 3414e4a2ed94SMatthew Wilcox (Oracle) if (folio_test_ksm(folio) || folio_ref_count(folio) != 1) { 3415e4a2ed94SMatthew Wilcox (Oracle) folio_unlock(folio); 341609854ba9SLinus Torvalds goto copy; 341752d1e606SKirill Tkhai } 3418c44b6743SRik van Riel /* 3419e4a2ed94SMatthew Wilcox (Oracle) * Ok, we've got the only folio reference from our mapping 3420e4a2ed94SMatthew Wilcox (Oracle) * and the folio is locked, it's dark out, and we're wearing 342153a05ad9SDavid Hildenbrand * sunglasses. Hit it. 3422c44b6743SRik van Riel */ 3423e4a2ed94SMatthew Wilcox (Oracle) page_move_anon_rmap(vmf->page, vma); 3424e4a2ed94SMatthew Wilcox (Oracle) folio_unlock(folio); 34256c287605SDavid Hildenbrand reuse: 3426c89357e2SDavid Hildenbrand if (unlikely(unshare)) { 3427c89357e2SDavid Hildenbrand pte_unmap_unlock(vmf->pte, vmf->ptl); 3428c89357e2SDavid Hildenbrand return 0; 3429c89357e2SDavid Hildenbrand } 3430be068f29SLinus Torvalds wp_page_reuse(vmf); 3431cb8d8633SDavid Hildenbrand return 0; 34321da177e4SLinus Torvalds } 343352d1e606SKirill Tkhai copy: 34341da177e4SLinus Torvalds /* 34351da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 34361da177e4SLinus Torvalds */ 3437b9086fdeSDavid Hildenbrand if (folio) 3438b9086fdeSDavid Hildenbrand folio_get(folio); 343928766805SShachar Raindel 344082b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 344194bfe85bSYang Yang #ifdef CONFIG_KSM 3442b9086fdeSDavid Hildenbrand if (folio && folio_test_ksm(folio)) 344394bfe85bSYang Yang count_vm_event(COW_KSM); 344494bfe85bSYang Yang #endif 3445a41b70d6SJan Kara return wp_page_copy(vmf); 34461da177e4SLinus Torvalds } 34471da177e4SLinus Torvalds 344897a89413SPeter Zijlstra static void unmap_mapping_range_vma(struct vm_area_struct *vma, 34491da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 34501da177e4SLinus Torvalds struct zap_details *details) 34511da177e4SLinus Torvalds { 3452f5cc4eefSAl Viro zap_page_range_single(vma, start_addr, end_addr - start_addr, details); 34531da177e4SLinus Torvalds } 34541da177e4SLinus Torvalds 3455f808c13fSDavidlohr Bueso static inline void unmap_mapping_range_tree(struct rb_root_cached *root, 3456232a6a1cSPeter Xu pgoff_t first_index, 3457232a6a1cSPeter Xu pgoff_t last_index, 34581da177e4SLinus Torvalds struct zap_details *details) 34591da177e4SLinus Torvalds { 34601da177e4SLinus Torvalds struct vm_area_struct *vma; 34611da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 34621da177e4SLinus Torvalds 3463232a6a1cSPeter Xu vma_interval_tree_foreach(vma, root, first_index, last_index) { 34641da177e4SLinus Torvalds vba = vma->vm_pgoff; 3465d6e93217SLibin vea = vba + vma_pages(vma) - 1; 3466f9871da9SMiaohe Lin zba = max(first_index, vba); 3467f9871da9SMiaohe Lin zea = min(last_index, vea); 34681da177e4SLinus Torvalds 346997a89413SPeter Zijlstra unmap_mapping_range_vma(vma, 34701da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 34711da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 347297a89413SPeter Zijlstra details); 34731da177e4SLinus Torvalds } 34741da177e4SLinus Torvalds } 34751da177e4SLinus Torvalds 34761da177e4SLinus Torvalds /** 34773506659eSMatthew Wilcox (Oracle) * unmap_mapping_folio() - Unmap single folio from processes. 34783506659eSMatthew Wilcox (Oracle) * @folio: The locked folio to be unmapped. 347922061a1fSHugh Dickins * 34803506659eSMatthew Wilcox (Oracle) * Unmap this folio from any userspace process which still has it mmaped. 348122061a1fSHugh Dickins * Typically, for efficiency, the range of nearby pages has already been 348222061a1fSHugh Dickins * unmapped by unmap_mapping_pages() or unmap_mapping_range(). But once 34833506659eSMatthew Wilcox (Oracle) * truncation or invalidation holds the lock on a folio, it may find that 34843506659eSMatthew Wilcox (Oracle) * the page has been remapped again: and then uses unmap_mapping_folio() 348522061a1fSHugh Dickins * to unmap it finally. 348622061a1fSHugh Dickins */ 34873506659eSMatthew Wilcox (Oracle) void unmap_mapping_folio(struct folio *folio) 348822061a1fSHugh Dickins { 34893506659eSMatthew Wilcox (Oracle) struct address_space *mapping = folio->mapping; 349022061a1fSHugh Dickins struct zap_details details = { }; 3491232a6a1cSPeter Xu pgoff_t first_index; 3492232a6a1cSPeter Xu pgoff_t last_index; 349322061a1fSHugh Dickins 34943506659eSMatthew Wilcox (Oracle) VM_BUG_ON(!folio_test_locked(folio)); 349522061a1fSHugh Dickins 34963506659eSMatthew Wilcox (Oracle) first_index = folio->index; 349787b11f86SSidhartha Kumar last_index = folio_next_index(folio) - 1; 3498232a6a1cSPeter Xu 34992e148f1eSPeter Xu details.even_cows = false; 35003506659eSMatthew Wilcox (Oracle) details.single_folio = folio; 3501999dad82SPeter Xu details.zap_flags = ZAP_FLAG_DROP_MARKER; 350222061a1fSHugh Dickins 35032c865995SHugh Dickins i_mmap_lock_read(mapping); 350422061a1fSHugh Dickins if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) 3505232a6a1cSPeter Xu unmap_mapping_range_tree(&mapping->i_mmap, first_index, 3506232a6a1cSPeter Xu last_index, &details); 35072c865995SHugh Dickins i_mmap_unlock_read(mapping); 350822061a1fSHugh Dickins } 350922061a1fSHugh Dickins 351022061a1fSHugh Dickins /** 3511977fbdcdSMatthew Wilcox * unmap_mapping_pages() - Unmap pages from processes. 3512977fbdcdSMatthew Wilcox * @mapping: The address space containing pages to be unmapped. 3513977fbdcdSMatthew Wilcox * @start: Index of first page to be unmapped. 3514977fbdcdSMatthew Wilcox * @nr: Number of pages to be unmapped. 0 to unmap to end of file. 3515977fbdcdSMatthew Wilcox * @even_cows: Whether to unmap even private COWed pages. 3516977fbdcdSMatthew Wilcox * 3517977fbdcdSMatthew Wilcox * Unmap the pages in this address space from any userspace process which 3518977fbdcdSMatthew Wilcox * has them mmaped. Generally, you want to remove COWed pages as well when 3519977fbdcdSMatthew Wilcox * a file is being truncated, but not when invalidating pages from the page 3520977fbdcdSMatthew Wilcox * cache. 3521977fbdcdSMatthew Wilcox */ 3522977fbdcdSMatthew Wilcox void unmap_mapping_pages(struct address_space *mapping, pgoff_t start, 3523977fbdcdSMatthew Wilcox pgoff_t nr, bool even_cows) 3524977fbdcdSMatthew Wilcox { 3525977fbdcdSMatthew Wilcox struct zap_details details = { }; 3526232a6a1cSPeter Xu pgoff_t first_index = start; 3527232a6a1cSPeter Xu pgoff_t last_index = start + nr - 1; 3528977fbdcdSMatthew Wilcox 35292e148f1eSPeter Xu details.even_cows = even_cows; 3530232a6a1cSPeter Xu if (last_index < first_index) 3531232a6a1cSPeter Xu last_index = ULONG_MAX; 3532977fbdcdSMatthew Wilcox 35332c865995SHugh Dickins i_mmap_lock_read(mapping); 3534977fbdcdSMatthew Wilcox if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) 3535232a6a1cSPeter Xu unmap_mapping_range_tree(&mapping->i_mmap, first_index, 3536232a6a1cSPeter Xu last_index, &details); 35372c865995SHugh Dickins i_mmap_unlock_read(mapping); 3538977fbdcdSMatthew Wilcox } 35396e0e99d5SDavid Howells EXPORT_SYMBOL_GPL(unmap_mapping_pages); 3540977fbdcdSMatthew Wilcox 3541977fbdcdSMatthew Wilcox /** 35428a5f14a2SKirill A. Shutemov * unmap_mapping_range - unmap the portion of all mmaps in the specified 3543977fbdcdSMatthew Wilcox * address_space corresponding to the specified byte range in the underlying 35448a5f14a2SKirill A. Shutemov * file. 35458a5f14a2SKirill A. Shutemov * 35463d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 35471da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 35481da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 354925d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 35501da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 35511da177e4SLinus Torvalds * partial pages. 35521da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 35531da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 35541da177e4SLinus Torvalds * end of the file. 35551da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 35561da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 35571da177e4SLinus Torvalds */ 35581da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 35591da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 35601da177e4SLinus Torvalds { 35611da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 35621da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 35631da177e4SLinus Torvalds 35641da177e4SLinus Torvalds /* Check for overflow. */ 35651da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 35661da177e4SLinus Torvalds long long holeend = 35671da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 35681da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 35691da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 35701da177e4SLinus Torvalds } 35711da177e4SLinus Torvalds 3572977fbdcdSMatthew Wilcox unmap_mapping_pages(mapping, hba, hlen, even_cows); 35731da177e4SLinus Torvalds } 35741da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 35751da177e4SLinus Torvalds 35761da177e4SLinus Torvalds /* 3577b756a3b5SAlistair Popple * Restore a potential device exclusive pte to a working pte entry 3578b756a3b5SAlistair Popple */ 3579b756a3b5SAlistair Popple static vm_fault_t remove_device_exclusive_entry(struct vm_fault *vmf) 3580b756a3b5SAlistair Popple { 358119672a9eSMatthew Wilcox (Oracle) struct folio *folio = page_folio(vmf->page); 3582b756a3b5SAlistair Popple struct vm_area_struct *vma = vmf->vma; 3583b756a3b5SAlistair Popple struct mmu_notifier_range range; 3584b756a3b5SAlistair Popple 35857c7b9629SAlistair Popple /* 35867c7b9629SAlistair Popple * We need a reference to lock the folio because we don't hold 35877c7b9629SAlistair Popple * the PTL so a racing thread can remove the device-exclusive 35887c7b9629SAlistair Popple * entry and unmap it. If the folio is free the entry must 35897c7b9629SAlistair Popple * have been removed already. If it happens to have already 35907c7b9629SAlistair Popple * been re-allocated after being freed all we do is lock and 35917c7b9629SAlistair Popple * unlock it. 35927c7b9629SAlistair Popple */ 35937c7b9629SAlistair Popple if (!folio_try_get(folio)) 35947c7b9629SAlistair Popple return 0; 35957c7b9629SAlistair Popple 35967c7b9629SAlistair Popple if (!folio_lock_or_retry(folio, vma->vm_mm, vmf->flags)) { 35977c7b9629SAlistair Popple folio_put(folio); 3598b756a3b5SAlistair Popple return VM_FAULT_RETRY; 35997c7b9629SAlistair Popple } 36007d4a8be0SAlistair Popple mmu_notifier_range_init_owner(&range, MMU_NOTIFY_EXCLUSIVE, 0, 3601b756a3b5SAlistair Popple vma->vm_mm, vmf->address & PAGE_MASK, 3602b756a3b5SAlistair Popple (vmf->address & PAGE_MASK) + PAGE_SIZE, NULL); 3603b756a3b5SAlistair Popple mmu_notifier_invalidate_range_start(&range); 3604b756a3b5SAlistair Popple 3605b756a3b5SAlistair Popple vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 3606b756a3b5SAlistair Popple &vmf->ptl); 3607c33c7948SRyan Roberts if (likely(vmf->pte && pte_same(ptep_get(vmf->pte), vmf->orig_pte))) 360819672a9eSMatthew Wilcox (Oracle) restore_exclusive_pte(vma, vmf->page, vmf->address, vmf->pte); 3609b756a3b5SAlistair Popple 36103db82b93SHugh Dickins if (vmf->pte) 3611b756a3b5SAlistair Popple pte_unmap_unlock(vmf->pte, vmf->ptl); 361219672a9eSMatthew Wilcox (Oracle) folio_unlock(folio); 36137c7b9629SAlistair Popple folio_put(folio); 3614b756a3b5SAlistair Popple 3615b756a3b5SAlistair Popple mmu_notifier_invalidate_range_end(&range); 3616b756a3b5SAlistair Popple return 0; 3617b756a3b5SAlistair Popple } 3618b756a3b5SAlistair Popple 3619a160e537SMatthew Wilcox (Oracle) static inline bool should_try_to_free_swap(struct folio *folio, 3620c145e0b4SDavid Hildenbrand struct vm_area_struct *vma, 3621c145e0b4SDavid Hildenbrand unsigned int fault_flags) 3622c145e0b4SDavid Hildenbrand { 3623a160e537SMatthew Wilcox (Oracle) if (!folio_test_swapcache(folio)) 3624c145e0b4SDavid Hildenbrand return false; 36259202d527SMatthew Wilcox (Oracle) if (mem_cgroup_swap_full(folio) || (vma->vm_flags & VM_LOCKED) || 3626a160e537SMatthew Wilcox (Oracle) folio_test_mlocked(folio)) 3627c145e0b4SDavid Hildenbrand return true; 3628c145e0b4SDavid Hildenbrand /* 3629c145e0b4SDavid Hildenbrand * If we want to map a page that's in the swapcache writable, we 3630c145e0b4SDavid Hildenbrand * have to detect via the refcount if we're really the exclusive 3631c145e0b4SDavid Hildenbrand * user. Try freeing the swapcache to get rid of the swapcache 3632c145e0b4SDavid Hildenbrand * reference only in case it's likely that we'll be the exlusive user. 3633c145e0b4SDavid Hildenbrand */ 3634a160e537SMatthew Wilcox (Oracle) return (fault_flags & FAULT_FLAG_WRITE) && !folio_test_ksm(folio) && 3635a160e537SMatthew Wilcox (Oracle) folio_ref_count(folio) == 2; 3636c145e0b4SDavid Hildenbrand } 3637c145e0b4SDavid Hildenbrand 36389c28a205SPeter Xu static vm_fault_t pte_marker_clear(struct vm_fault *vmf) 36399c28a205SPeter Xu { 36409c28a205SPeter Xu vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, 36419c28a205SPeter Xu vmf->address, &vmf->ptl); 36423db82b93SHugh Dickins if (!vmf->pte) 36433db82b93SHugh Dickins return 0; 36449c28a205SPeter Xu /* 36459c28a205SPeter Xu * Be careful so that we will only recover a special uffd-wp pte into a 36469c28a205SPeter Xu * none pte. Otherwise it means the pte could have changed, so retry. 36477e3ce3f8SPeter Xu * 36487e3ce3f8SPeter Xu * This should also cover the case where e.g. the pte changed 3649af19487fSAxel Rasmussen * quickly from a PTE_MARKER_UFFD_WP into PTE_MARKER_POISONED. 36507e3ce3f8SPeter Xu * So is_pte_marker() check is not enough to safely drop the pte. 36519c28a205SPeter Xu */ 3652c33c7948SRyan Roberts if (pte_same(vmf->orig_pte, ptep_get(vmf->pte))) 36539c28a205SPeter Xu pte_clear(vmf->vma->vm_mm, vmf->address, vmf->pte); 36549c28a205SPeter Xu pte_unmap_unlock(vmf->pte, vmf->ptl); 36559c28a205SPeter Xu return 0; 36569c28a205SPeter Xu } 36579c28a205SPeter Xu 36582bad466cSPeter Xu static vm_fault_t do_pte_missing(struct vm_fault *vmf) 36592bad466cSPeter Xu { 36602bad466cSPeter Xu if (vma_is_anonymous(vmf->vma)) 36612bad466cSPeter Xu return do_anonymous_page(vmf); 36622bad466cSPeter Xu else 36632bad466cSPeter Xu return do_fault(vmf); 36642bad466cSPeter Xu } 36652bad466cSPeter Xu 36669c28a205SPeter Xu /* 36679c28a205SPeter Xu * This is actually a page-missing access, but with uffd-wp special pte 36689c28a205SPeter Xu * installed. It means this pte was wr-protected before being unmapped. 36699c28a205SPeter Xu */ 36709c28a205SPeter Xu static vm_fault_t pte_marker_handle_uffd_wp(struct vm_fault *vmf) 36719c28a205SPeter Xu { 36729c28a205SPeter Xu /* 36739c28a205SPeter Xu * Just in case there're leftover special ptes even after the region 36747a079ba2SPeter Xu * got unregistered - we can simply clear them. 36759c28a205SPeter Xu */ 36762bad466cSPeter Xu if (unlikely(!userfaultfd_wp(vmf->vma))) 36779c28a205SPeter Xu return pte_marker_clear(vmf); 36789c28a205SPeter Xu 36792bad466cSPeter Xu return do_pte_missing(vmf); 36809c28a205SPeter Xu } 36819c28a205SPeter Xu 36825c041f5dSPeter Xu static vm_fault_t handle_pte_marker(struct vm_fault *vmf) 36835c041f5dSPeter Xu { 36845c041f5dSPeter Xu swp_entry_t entry = pte_to_swp_entry(vmf->orig_pte); 36855c041f5dSPeter Xu unsigned long marker = pte_marker_get(entry); 36865c041f5dSPeter Xu 36875c041f5dSPeter Xu /* 3688ca92ea3dSPeter Xu * PTE markers should never be empty. If anything weird happened, 3689ca92ea3dSPeter Xu * the best thing to do is to kill the process along with its mm. 36905c041f5dSPeter Xu */ 3691ca92ea3dSPeter Xu if (WARN_ON_ONCE(!marker)) 36925c041f5dSPeter Xu return VM_FAULT_SIGBUS; 36935c041f5dSPeter Xu 369415520a3fSPeter Xu /* Higher priority than uffd-wp when data corrupted */ 3695af19487fSAxel Rasmussen if (marker & PTE_MARKER_POISONED) 3696af19487fSAxel Rasmussen return VM_FAULT_HWPOISON; 369715520a3fSPeter Xu 36989c28a205SPeter Xu if (pte_marker_entry_uffd_wp(entry)) 36999c28a205SPeter Xu return pte_marker_handle_uffd_wp(vmf); 37009c28a205SPeter Xu 37019c28a205SPeter Xu /* This is an unknown pte marker */ 37029c28a205SPeter Xu return VM_FAULT_SIGBUS; 37035c041f5dSPeter Xu } 37045c041f5dSPeter Xu 3705b756a3b5SAlistair Popple /* 3706c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 37078f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 37089a95f3cfSPaul Cassella * We return with pte unmapped and unlocked. 37099a95f3cfSPaul Cassella * 3710c1e8d7c6SMichel Lespinasse * We return with the mmap_lock locked or unlocked in the same cases 37119a95f3cfSPaul Cassella * as does filemap_fault(). 37121da177e4SLinus Torvalds */ 37132b740303SSouptick Joarder vm_fault_t do_swap_page(struct vm_fault *vmf) 37141da177e4SLinus Torvalds { 371582b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3716d4f9565aSMatthew Wilcox (Oracle) struct folio *swapcache, *folio = NULL; 3717d4f9565aSMatthew Wilcox (Oracle) struct page *page; 37182799e775SMiaohe Lin struct swap_info_struct *si = NULL; 371914f9135dSDavid Hildenbrand rmap_t rmap_flags = RMAP_NONE; 37201493a191SDavid Hildenbrand bool exclusive = false; 372165500d23SHugh Dickins swp_entry_t entry; 37221da177e4SLinus Torvalds pte_t pte; 3723d065bd81SMichel Lespinasse int locked; 37242b740303SSouptick Joarder vm_fault_t ret = 0; 3725aae466b0SJoonsoo Kim void *shadow = NULL; 37261da177e4SLinus Torvalds 37272ca99358SPeter Xu if (!pte_unmap_same(vmf)) 37288f4e2101SHugh Dickins goto out; 372965500d23SHugh Dickins 373017c05f18SSuren Baghdasaryan if (vmf->flags & FAULT_FLAG_VMA_LOCK) { 373117c05f18SSuren Baghdasaryan ret = VM_FAULT_RETRY; 373217c05f18SSuren Baghdasaryan goto out; 373317c05f18SSuren Baghdasaryan } 373417c05f18SSuren Baghdasaryan 37352994302bSJan Kara entry = pte_to_swp_entry(vmf->orig_pte); 3736d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 37370697212aSChristoph Lameter if (is_migration_entry(entry)) { 373882b0f8c3SJan Kara migration_entry_wait(vma->vm_mm, vmf->pmd, 373982b0f8c3SJan Kara vmf->address); 3740b756a3b5SAlistair Popple } else if (is_device_exclusive_entry(entry)) { 3741b756a3b5SAlistair Popple vmf->page = pfn_swap_entry_to_page(entry); 3742b756a3b5SAlistair Popple ret = remove_device_exclusive_entry(vmf); 37435042db43SJérôme Glisse } else if (is_device_private_entry(entry)) { 3744af5cdaf8SAlistair Popple vmf->page = pfn_swap_entry_to_page(entry); 374516ce101dSAlistair Popple vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 374616ce101dSAlistair Popple vmf->address, &vmf->ptl); 37473db82b93SHugh Dickins if (unlikely(!vmf->pte || 3748c33c7948SRyan Roberts !pte_same(ptep_get(vmf->pte), 3749c33c7948SRyan Roberts vmf->orig_pte))) 37503b65f437SRyan Roberts goto unlock; 375116ce101dSAlistair Popple 375216ce101dSAlistair Popple /* 375316ce101dSAlistair Popple * Get a page reference while we know the page can't be 375416ce101dSAlistair Popple * freed. 375516ce101dSAlistair Popple */ 375616ce101dSAlistair Popple get_page(vmf->page); 375716ce101dSAlistair Popple pte_unmap_unlock(vmf->pte, vmf->ptl); 37584a955bedSAlistair Popple ret = vmf->page->pgmap->ops->migrate_to_ram(vmf); 375916ce101dSAlistair Popple put_page(vmf->page); 3760d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 3761d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 37625c041f5dSPeter Xu } else if (is_pte_marker_entry(entry)) { 37635c041f5dSPeter Xu ret = handle_pte_marker(vmf); 3764d1737fdbSAndi Kleen } else { 37652994302bSJan Kara print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); 3766d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 3767d1737fdbSAndi Kleen } 37680697212aSChristoph Lameter goto out; 37690697212aSChristoph Lameter } 37700bcac06fSMinchan Kim 37712799e775SMiaohe Lin /* Prevent swapoff from happening to us. */ 37722799e775SMiaohe Lin si = get_swap_device(entry); 37732799e775SMiaohe Lin if (unlikely(!si)) 37742799e775SMiaohe Lin goto out; 37750bcac06fSMinchan Kim 37765a423081SMatthew Wilcox (Oracle) folio = swap_cache_get_folio(entry, vma, vmf->address); 37775a423081SMatthew Wilcox (Oracle) if (folio) 37785a423081SMatthew Wilcox (Oracle) page = folio_file_page(folio, swp_offset(entry)); 3779d4f9565aSMatthew Wilcox (Oracle) swapcache = folio; 3780f8020772SMinchan Kim 3781d4f9565aSMatthew Wilcox (Oracle) if (!folio) { 3782a449bf58SQian Cai if (data_race(si->flags & SWP_SYNCHRONOUS_IO) && 3783eb085574SHuang Ying __swap_count(entry) == 1) { 37840bcac06fSMinchan Kim /* skip swapcache */ 378563ad4addSMatthew Wilcox (Oracle) folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, 378663ad4addSMatthew Wilcox (Oracle) vma, vmf->address, false); 378763ad4addSMatthew Wilcox (Oracle) page = &folio->page; 378863ad4addSMatthew Wilcox (Oracle) if (folio) { 378963ad4addSMatthew Wilcox (Oracle) __folio_set_locked(folio); 379063ad4addSMatthew Wilcox (Oracle) __folio_set_swapbacked(folio); 37914c6355b2SJohannes Weiner 379265995918SMatthew Wilcox (Oracle) if (mem_cgroup_swapin_charge_folio(folio, 379363ad4addSMatthew Wilcox (Oracle) vma->vm_mm, GFP_KERNEL, 379463ad4addSMatthew Wilcox (Oracle) entry)) { 3795545b1b07SMichal Hocko ret = VM_FAULT_OOM; 37964c6355b2SJohannes Weiner goto out_page; 3797545b1b07SMichal Hocko } 37980add0c77SShakeel Butt mem_cgroup_swapin_uncharge_swap(entry); 37994c6355b2SJohannes Weiner 3800aae466b0SJoonsoo Kim shadow = get_shadow_from_swap_cache(entry); 3801aae466b0SJoonsoo Kim if (shadow) 380263ad4addSMatthew Wilcox (Oracle) workingset_refault(folio, shadow); 38030076f029SJoonsoo Kim 380463ad4addSMatthew Wilcox (Oracle) folio_add_lru(folio); 38050add0c77SShakeel Butt 38060add0c77SShakeel Butt /* To provide entry to swap_readpage() */ 380763ad4addSMatthew Wilcox (Oracle) folio_set_swap_entry(folio, entry); 38085169b844SNeilBrown swap_readpage(page, true, NULL); 380963ad4addSMatthew Wilcox (Oracle) folio->private = NULL; 38100bcac06fSMinchan Kim } 3811aa8d22a1SMinchan Kim } else { 3812e9e9b7ecSMinchan Kim page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, 3813e9e9b7ecSMinchan Kim vmf); 381463ad4addSMatthew Wilcox (Oracle) if (page) 381563ad4addSMatthew Wilcox (Oracle) folio = page_folio(page); 3816d4f9565aSMatthew Wilcox (Oracle) swapcache = folio; 38170bcac06fSMinchan Kim } 38180bcac06fSMinchan Kim 3819d4f9565aSMatthew Wilcox (Oracle) if (!folio) { 38201da177e4SLinus Torvalds /* 38218f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 38228f4e2101SHugh Dickins * while we released the pte lock. 38231da177e4SLinus Torvalds */ 382482b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 382582b0f8c3SJan Kara vmf->address, &vmf->ptl); 3826c33c7948SRyan Roberts if (likely(vmf->pte && 3827c33c7948SRyan Roberts pte_same(ptep_get(vmf->pte), vmf->orig_pte))) 38281da177e4SLinus Torvalds ret = VM_FAULT_OOM; 382965500d23SHugh Dickins goto unlock; 38301da177e4SLinus Torvalds } 38311da177e4SLinus Torvalds 38321da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 38331da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 3834f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 38352262185cSRoman Gushchin count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); 3836d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 383771f72525SWu Fengguang /* 383871f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 383971f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 384071f72525SWu Fengguang */ 3841d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 38424779cb31SAndi Kleen goto out_release; 38431da177e4SLinus Torvalds } 38441da177e4SLinus Torvalds 384519672a9eSMatthew Wilcox (Oracle) locked = folio_lock_or_retry(folio, vma->vm_mm, vmf->flags); 3846e709ffd6SRik van Riel 3847d065bd81SMichel Lespinasse if (!locked) { 3848d065bd81SMichel Lespinasse ret |= VM_FAULT_RETRY; 3849d065bd81SMichel Lespinasse goto out_release; 3850d065bd81SMichel Lespinasse } 38511da177e4SLinus Torvalds 385284d60fddSDavid Hildenbrand if (swapcache) { 38534969c119SAndrea Arcangeli /* 38543b344157SMatthew Wilcox (Oracle) * Make sure folio_free_swap() or swapoff did not release the 385584d60fddSDavid Hildenbrand * swapcache from under us. The page pin, and pte_same test 385684d60fddSDavid Hildenbrand * below, are not enough to exclude that. Even if it is still 385784d60fddSDavid Hildenbrand * swapcache, we need to check that the page's swap has not 385884d60fddSDavid Hildenbrand * changed. 38594969c119SAndrea Arcangeli */ 386063ad4addSMatthew Wilcox (Oracle) if (unlikely(!folio_test_swapcache(folio) || 386184d60fddSDavid Hildenbrand page_private(page) != entry.val)) 38624969c119SAndrea Arcangeli goto out_page; 38634969c119SAndrea Arcangeli 386484d60fddSDavid Hildenbrand /* 386584d60fddSDavid Hildenbrand * KSM sometimes has to copy on read faults, for example, if 386684d60fddSDavid Hildenbrand * page->index of !PageKSM() pages would be nonlinear inside the 386784d60fddSDavid Hildenbrand * anon VMA -- PageKSM() is lost on actual swapout. 386884d60fddSDavid Hildenbrand */ 386982b0f8c3SJan Kara page = ksm_might_need_to_copy(page, vma, vmf->address); 38704969c119SAndrea Arcangeli if (unlikely(!page)) { 38715ad64688SHugh Dickins ret = VM_FAULT_OOM; 38724969c119SAndrea Arcangeli goto out_page; 38736b970599SKefeng Wang } else if (unlikely(PTR_ERR(page) == -EHWPOISON)) { 38746b970599SKefeng Wang ret = VM_FAULT_HWPOISON; 38756b970599SKefeng Wang goto out_page; 38764969c119SAndrea Arcangeli } 387763ad4addSMatthew Wilcox (Oracle) folio = page_folio(page); 3878c145e0b4SDavid Hildenbrand 3879c145e0b4SDavid Hildenbrand /* 3880c145e0b4SDavid Hildenbrand * If we want to map a page that's in the swapcache writable, we 3881c145e0b4SDavid Hildenbrand * have to detect via the refcount if we're really the exclusive 3882c145e0b4SDavid Hildenbrand * owner. Try removing the extra reference from the local LRU 38831fec6890SMatthew Wilcox (Oracle) * caches if required. 3884c145e0b4SDavid Hildenbrand */ 3885d4f9565aSMatthew Wilcox (Oracle) if ((vmf->flags & FAULT_FLAG_WRITE) && folio == swapcache && 388663ad4addSMatthew Wilcox (Oracle) !folio_test_ksm(folio) && !folio_test_lru(folio)) 3887c145e0b4SDavid Hildenbrand lru_add_drain(); 388884d60fddSDavid Hildenbrand } 38895ad64688SHugh Dickins 38904231f842SKefeng Wang folio_throttle_swaprate(folio, GFP_KERNEL); 3891073e587eSKAMEZAWA Hiroyuki 38921da177e4SLinus Torvalds /* 38938f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 38941da177e4SLinus Torvalds */ 389582b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 389682b0f8c3SJan Kara &vmf->ptl); 3897c33c7948SRyan Roberts if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) 3898b8107480SKirill Korotaev goto out_nomap; 3899b8107480SKirill Korotaev 390063ad4addSMatthew Wilcox (Oracle) if (unlikely(!folio_test_uptodate(folio))) { 3901b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 3902b8107480SKirill Korotaev goto out_nomap; 39031da177e4SLinus Torvalds } 39041da177e4SLinus Torvalds 39058c7c6e34SKAMEZAWA Hiroyuki /* 390678fbe906SDavid Hildenbrand * PG_anon_exclusive reuses PG_mappedtodisk for anon pages. A swap pte 390778fbe906SDavid Hildenbrand * must never point at an anonymous page in the swapcache that is 390878fbe906SDavid Hildenbrand * PG_anon_exclusive. Sanity check that this holds and especially, that 390978fbe906SDavid Hildenbrand * no filesystem set PG_mappedtodisk on a page in the swapcache. Sanity 391078fbe906SDavid Hildenbrand * check after taking the PT lock and making sure that nobody 391178fbe906SDavid Hildenbrand * concurrently faulted in this page and set PG_anon_exclusive. 391278fbe906SDavid Hildenbrand */ 391363ad4addSMatthew Wilcox (Oracle) BUG_ON(!folio_test_anon(folio) && folio_test_mappedtodisk(folio)); 391463ad4addSMatthew Wilcox (Oracle) BUG_ON(folio_test_anon(folio) && PageAnonExclusive(page)); 391578fbe906SDavid Hildenbrand 391678fbe906SDavid Hildenbrand /* 39171493a191SDavid Hildenbrand * Check under PT lock (to protect against concurrent fork() sharing 39181493a191SDavid Hildenbrand * the swap entry concurrently) for certainly exclusive pages. 39191493a191SDavid Hildenbrand */ 392063ad4addSMatthew Wilcox (Oracle) if (!folio_test_ksm(folio)) { 39211493a191SDavid Hildenbrand exclusive = pte_swp_exclusive(vmf->orig_pte); 3922d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache) { 39231493a191SDavid Hildenbrand /* 39241493a191SDavid Hildenbrand * We have a fresh page that is not exposed to the 39251493a191SDavid Hildenbrand * swapcache -> certainly exclusive. 39261493a191SDavid Hildenbrand */ 39271493a191SDavid Hildenbrand exclusive = true; 392863ad4addSMatthew Wilcox (Oracle) } else if (exclusive && folio_test_writeback(folio) && 3929eacde327SMiaohe Lin data_race(si->flags & SWP_STABLE_WRITES)) { 39301493a191SDavid Hildenbrand /* 39311493a191SDavid Hildenbrand * This is tricky: not all swap backends support 39321493a191SDavid Hildenbrand * concurrent page modifications while under writeback. 39331493a191SDavid Hildenbrand * 39341493a191SDavid Hildenbrand * So if we stumble over such a page in the swapcache 39351493a191SDavid Hildenbrand * we must not set the page exclusive, otherwise we can 39361493a191SDavid Hildenbrand * map it writable without further checks and modify it 39371493a191SDavid Hildenbrand * while still under writeback. 39381493a191SDavid Hildenbrand * 39391493a191SDavid Hildenbrand * For these problematic swap backends, simply drop the 39401493a191SDavid Hildenbrand * exclusive marker: this is perfectly fine as we start 39411493a191SDavid Hildenbrand * writeback only if we fully unmapped the page and 39421493a191SDavid Hildenbrand * there are no unexpected references on the page after 39431493a191SDavid Hildenbrand * unmapping succeeded. After fully unmapped, no 39441493a191SDavid Hildenbrand * further GUP references (FOLL_GET and FOLL_PIN) can 39451493a191SDavid Hildenbrand * appear, so dropping the exclusive marker and mapping 39461493a191SDavid Hildenbrand * it only R/O is fine. 39471493a191SDavid Hildenbrand */ 39481493a191SDavid Hildenbrand exclusive = false; 39491493a191SDavid Hildenbrand } 39501493a191SDavid Hildenbrand } 39511493a191SDavid Hildenbrand 39521493a191SDavid Hildenbrand /* 39536dca4ac6SPeter Collingbourne * Some architectures may have to restore extra metadata to the page 39546dca4ac6SPeter Collingbourne * when reading from swap. This metadata may be indexed by swap entry 39556dca4ac6SPeter Collingbourne * so this must be called before swap_free(). 39566dca4ac6SPeter Collingbourne */ 39576dca4ac6SPeter Collingbourne arch_swap_restore(entry, folio); 39586dca4ac6SPeter Collingbourne 39596dca4ac6SPeter Collingbourne /* 3960c145e0b4SDavid Hildenbrand * Remove the swap entry and conditionally try to free up the swapcache. 3961c145e0b4SDavid Hildenbrand * We're already holding a reference on the page but haven't mapped it 3962c145e0b4SDavid Hildenbrand * yet. 39638c7c6e34SKAMEZAWA Hiroyuki */ 3964c145e0b4SDavid Hildenbrand swap_free(entry); 3965a160e537SMatthew Wilcox (Oracle) if (should_try_to_free_swap(folio, vma, vmf->flags)) 3966a160e537SMatthew Wilcox (Oracle) folio_free_swap(folio); 39671da177e4SLinus Torvalds 3968f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, MM_ANONPAGES); 3969f1a79412SShakeel Butt dec_mm_counter(vma->vm_mm, MM_SWAPENTS); 39701da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 3971c145e0b4SDavid Hildenbrand 3972c145e0b4SDavid Hildenbrand /* 39731493a191SDavid Hildenbrand * Same logic as in do_wp_page(); however, optimize for pages that are 39741493a191SDavid Hildenbrand * certainly not shared either because we just allocated them without 39751493a191SDavid Hildenbrand * exposing them to the swapcache or because the swap entry indicates 39761493a191SDavid Hildenbrand * exclusivity. 3977c145e0b4SDavid Hildenbrand */ 397863ad4addSMatthew Wilcox (Oracle) if (!folio_test_ksm(folio) && 397963ad4addSMatthew Wilcox (Oracle) (exclusive || folio_ref_count(folio) == 1)) { 39806c287605SDavid Hildenbrand if (vmf->flags & FAULT_FLAG_WRITE) { 39811da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 398282b0f8c3SJan Kara vmf->flags &= ~FAULT_FLAG_WRITE; 39836c287605SDavid Hildenbrand } 398414f9135dSDavid Hildenbrand rmap_flags |= RMAP_EXCLUSIVE; 39851da177e4SLinus Torvalds } 39861da177e4SLinus Torvalds flush_icache_page(vma, page); 39872994302bSJan Kara if (pte_swp_soft_dirty(vmf->orig_pte)) 3988179ef71cSCyrill Gorcunov pte = pte_mksoft_dirty(pte); 3989f1eb1bacSPeter Xu if (pte_swp_uffd_wp(vmf->orig_pte)) 3990f45ec5ffSPeter Xu pte = pte_mkuffd_wp(pte); 39912994302bSJan Kara vmf->orig_pte = pte; 39920bcac06fSMinchan Kim 39930bcac06fSMinchan Kim /* ksm created a completely new copy */ 3994d4f9565aSMatthew Wilcox (Oracle) if (unlikely(folio != swapcache && swapcache)) { 399540f2bbf7SDavid Hildenbrand page_add_new_anon_rmap(page, vma, vmf->address); 399663ad4addSMatthew Wilcox (Oracle) folio_add_lru_vma(folio, vma); 39970bcac06fSMinchan Kim } else { 3998f1e2db12SDavid Hildenbrand page_add_anon_rmap(page, vma, vmf->address, rmap_flags); 399900501b53SJohannes Weiner } 40001da177e4SLinus Torvalds 400163ad4addSMatthew Wilcox (Oracle) VM_BUG_ON(!folio_test_anon(folio) || 400263ad4addSMatthew Wilcox (Oracle) (pte_write(pte) && !PageAnonExclusive(page))); 40031eba86c0SPasha Tatashin set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); 40041eba86c0SPasha Tatashin arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte); 40051eba86c0SPasha Tatashin 400663ad4addSMatthew Wilcox (Oracle) folio_unlock(folio); 4007d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache && swapcache) { 40084969c119SAndrea Arcangeli /* 40094969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 40104969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 40114969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 40124969c119SAndrea Arcangeli * further safety release the lock after the swap_free 40134969c119SAndrea Arcangeli * so that the swap count won't change under a 40144969c119SAndrea Arcangeli * parallel locked swapcache. 40154969c119SAndrea Arcangeli */ 4016d4f9565aSMatthew Wilcox (Oracle) folio_unlock(swapcache); 4017d4f9565aSMatthew Wilcox (Oracle) folio_put(swapcache); 40184969c119SAndrea Arcangeli } 4019c475a8abSHugh Dickins 402082b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) { 40212994302bSJan Kara ret |= do_wp_page(vmf); 402261469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 402361469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 40241da177e4SLinus Torvalds goto out; 40251da177e4SLinus Torvalds } 40261da177e4SLinus Torvalds 40271da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 402882b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 402965500d23SHugh Dickins unlock: 40303db82b93SHugh Dickins if (vmf->pte) 403182b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 40321da177e4SLinus Torvalds out: 40332799e775SMiaohe Lin if (si) 40342799e775SMiaohe Lin put_swap_device(si); 40351da177e4SLinus Torvalds return ret; 4036b8107480SKirill Korotaev out_nomap: 40373db82b93SHugh Dickins if (vmf->pte) 403882b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 4039bc43f75cSJohannes Weiner out_page: 404063ad4addSMatthew Wilcox (Oracle) folio_unlock(folio); 40414779cb31SAndi Kleen out_release: 404263ad4addSMatthew Wilcox (Oracle) folio_put(folio); 4043d4f9565aSMatthew Wilcox (Oracle) if (folio != swapcache && swapcache) { 4044d4f9565aSMatthew Wilcox (Oracle) folio_unlock(swapcache); 4045d4f9565aSMatthew Wilcox (Oracle) folio_put(swapcache); 40464969c119SAndrea Arcangeli } 40472799e775SMiaohe Lin if (si) 40482799e775SMiaohe Lin put_swap_device(si); 404965500d23SHugh Dickins return ret; 40501da177e4SLinus Torvalds } 40511da177e4SLinus Torvalds 40521da177e4SLinus Torvalds /* 4053c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 40548f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 4055c1e8d7c6SMichel Lespinasse * We return with mmap_lock still held, but pte unmapped and unlocked. 40561da177e4SLinus Torvalds */ 40572b740303SSouptick Joarder static vm_fault_t do_anonymous_page(struct vm_fault *vmf) 40581da177e4SLinus Torvalds { 40592bad466cSPeter Xu bool uffd_wp = vmf_orig_pte_uffd_wp(vmf); 406082b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 40616bc56a4dSMatthew Wilcox (Oracle) struct folio *folio; 40622b740303SSouptick Joarder vm_fault_t ret = 0; 40631da177e4SLinus Torvalds pte_t entry; 40641da177e4SLinus Torvalds 40656b7339f4SKirill A. Shutemov /* File mapping without ->vm_ops ? */ 40666b7339f4SKirill A. Shutemov if (vma->vm_flags & VM_SHARED) 40676b7339f4SKirill A. Shutemov return VM_FAULT_SIGBUS; 40686b7339f4SKirill A. Shutemov 40697267ec00SKirill A. Shutemov /* 40703db82b93SHugh Dickins * Use pte_alloc() instead of pte_alloc_map(), so that OOM can 40713db82b93SHugh Dickins * be distinguished from a transient failure of pte_offset_map(). 40727267ec00SKirill A. Shutemov */ 40734cf58924SJoel Fernandes (Google) if (pte_alloc(vma->vm_mm, vmf->pmd)) 40747267ec00SKirill A. Shutemov return VM_FAULT_OOM; 40757267ec00SKirill A. Shutemov 407611ac5524SLinus Torvalds /* Use the zero-page for reads */ 407782b0f8c3SJan Kara if (!(vmf->flags & FAULT_FLAG_WRITE) && 4078bae473a4SKirill A. Shutemov !mm_forbids_zeropage(vma->vm_mm)) { 407982b0f8c3SJan Kara entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), 408062eede62SHugh Dickins vma->vm_page_prot)); 408182b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 408282b0f8c3SJan Kara vmf->address, &vmf->ptl); 40833db82b93SHugh Dickins if (!vmf->pte) 40843db82b93SHugh Dickins goto unlock; 40852bad466cSPeter Xu if (vmf_pte_changed(vmf)) { 40867df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 4087a13ea5b7SHugh Dickins goto unlock; 40887df67697SBibo Mao } 40896b31d595SMichal Hocko ret = check_stable_address_space(vma->vm_mm); 40906b31d595SMichal Hocko if (ret) 40916b31d595SMichal Hocko goto unlock; 40926b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 40936b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 409482b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 409582b0f8c3SJan Kara return handle_userfault(vmf, VM_UFFD_MISSING); 40966b251fc9SAndrea Arcangeli } 4097a13ea5b7SHugh Dickins goto setpte; 4098a13ea5b7SHugh Dickins } 4099a13ea5b7SHugh Dickins 41001da177e4SLinus Torvalds /* Allocate our own private page. */ 41011da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 410265500d23SHugh Dickins goto oom; 41036bc56a4dSMatthew Wilcox (Oracle) folio = vma_alloc_zeroed_movable_folio(vma, vmf->address); 41046bc56a4dSMatthew Wilcox (Oracle) if (!folio) 410565500d23SHugh Dickins goto oom; 4106eb3c24f3SMel Gorman 41076bc56a4dSMatthew Wilcox (Oracle) if (mem_cgroup_charge(folio, vma->vm_mm, GFP_KERNEL)) 4108eb3c24f3SMel Gorman goto oom_free_page; 4109e2bf3e2cSKefeng Wang folio_throttle_swaprate(folio, GFP_KERNEL); 4110eb3c24f3SMel Gorman 411152f37629SMinchan Kim /* 4112cb3184deSMatthew Wilcox (Oracle) * The memory barrier inside __folio_mark_uptodate makes sure that 4113f4f5329dSWei Yang * preceding stores to the page contents become visible before 411452f37629SMinchan Kim * the set_pte_at() write. 411552f37629SMinchan Kim */ 4116cb3184deSMatthew Wilcox (Oracle) __folio_mark_uptodate(folio); 41171da177e4SLinus Torvalds 4118cb3184deSMatthew Wilcox (Oracle) entry = mk_pte(&folio->page, vma->vm_page_prot); 411950c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 41201ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 41211ac0cb5dSHugh Dickins entry = pte_mkwrite(pte_mkdirty(entry)); 41228f4e2101SHugh Dickins 412382b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 412482b0f8c3SJan Kara &vmf->ptl); 41253db82b93SHugh Dickins if (!vmf->pte) 41263db82b93SHugh Dickins goto release; 41272bad466cSPeter Xu if (vmf_pte_changed(vmf)) { 4128bce8cb3cSQi Zheng update_mmu_tlb(vma, vmf->address, vmf->pte); 41298f4e2101SHugh Dickins goto release; 41307df67697SBibo Mao } 41319ba69294SHugh Dickins 41326b31d595SMichal Hocko ret = check_stable_address_space(vma->vm_mm); 41336b31d595SMichal Hocko if (ret) 41346b31d595SMichal Hocko goto release; 41356b31d595SMichal Hocko 41366b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 41376b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 413882b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 4139cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 414082b0f8c3SJan Kara return handle_userfault(vmf, VM_UFFD_MISSING); 41416b251fc9SAndrea Arcangeli } 41426b251fc9SAndrea Arcangeli 4143f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, MM_ANONPAGES); 4144cb3184deSMatthew Wilcox (Oracle) folio_add_new_anon_rmap(folio, vma, vmf->address); 4145cb3184deSMatthew Wilcox (Oracle) folio_add_lru_vma(folio, vma); 4146a13ea5b7SHugh Dickins setpte: 41472bad466cSPeter Xu if (uffd_wp) 41482bad466cSPeter Xu entry = pte_mkuffd_wp(entry); 414982b0f8c3SJan Kara set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); 41501da177e4SLinus Torvalds 41511da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 415282b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 415365500d23SHugh Dickins unlock: 41543db82b93SHugh Dickins if (vmf->pte) 415582b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 41566b31d595SMichal Hocko return ret; 41578f4e2101SHugh Dickins release: 4158cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 41598f4e2101SHugh Dickins goto unlock; 41608a9f3ccdSBalbir Singh oom_free_page: 4161cb3184deSMatthew Wilcox (Oracle) folio_put(folio); 416265500d23SHugh Dickins oom: 41631da177e4SLinus Torvalds return VM_FAULT_OOM; 41641da177e4SLinus Torvalds } 41651da177e4SLinus Torvalds 41669a95f3cfSPaul Cassella /* 4167c1e8d7c6SMichel Lespinasse * The mmap_lock must have been held on entry, and may have been 41689a95f3cfSPaul Cassella * released depending on flags and vma->vm_ops->fault() return value. 41699a95f3cfSPaul Cassella * See filemap_fault() and __lock_page_retry(). 41709a95f3cfSPaul Cassella */ 41712b740303SSouptick Joarder static vm_fault_t __do_fault(struct vm_fault *vmf) 41727eae74afSKirill A. Shutemov { 417382b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 41742b740303SSouptick Joarder vm_fault_t ret; 41757eae74afSKirill A. Shutemov 417663f3655fSMichal Hocko /* 417763f3655fSMichal Hocko * Preallocate pte before we take page_lock because this might lead to 417863f3655fSMichal Hocko * deadlocks for memcg reclaim which waits for pages under writeback: 417963f3655fSMichal Hocko * lock_page(A) 418063f3655fSMichal Hocko * SetPageWriteback(A) 418163f3655fSMichal Hocko * unlock_page(A) 418263f3655fSMichal Hocko * lock_page(B) 418363f3655fSMichal Hocko * lock_page(B) 4184d383807aSYanfei Xu * pte_alloc_one 418563f3655fSMichal Hocko * shrink_page_list 418663f3655fSMichal Hocko * wait_on_page_writeback(A) 418763f3655fSMichal Hocko * SetPageWriteback(B) 418863f3655fSMichal Hocko * unlock_page(B) 418963f3655fSMichal Hocko * # flush A, B to clear the writeback 419063f3655fSMichal Hocko */ 419163f3655fSMichal Hocko if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) { 4192a7069ee3SYanfei Xu vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); 419363f3655fSMichal Hocko if (!vmf->prealloc_pte) 419463f3655fSMichal Hocko return VM_FAULT_OOM; 419563f3655fSMichal Hocko } 419663f3655fSMichal Hocko 419711bac800SDave Jiang ret = vma->vm_ops->fault(vmf); 41983917048dSJan Kara if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | 4199b1aa812bSJan Kara VM_FAULT_DONE_COW))) 42007eae74afSKirill A. Shutemov return ret; 42017eae74afSKirill A. Shutemov 4202667240e0SJan Kara if (unlikely(PageHWPoison(vmf->page))) { 42033149c79fSRik van Riel struct page *page = vmf->page; 4204e53ac737SRik van Riel vm_fault_t poisonret = VM_FAULT_HWPOISON; 4205e53ac737SRik van Riel if (ret & VM_FAULT_LOCKED) { 42063149c79fSRik van Riel if (page_mapped(page)) 42073149c79fSRik van Riel unmap_mapping_pages(page_mapping(page), 42083149c79fSRik van Riel page->index, 1, false); 4209e53ac737SRik van Riel /* Retry if a clean page was removed from the cache. */ 42103149c79fSRik van Riel if (invalidate_inode_page(page)) 42113149c79fSRik van Riel poisonret = VM_FAULT_NOPAGE; 42123149c79fSRik van Riel unlock_page(page); 4213e53ac737SRik van Riel } 42143149c79fSRik van Riel put_page(page); 4215936ca80dSJan Kara vmf->page = NULL; 4216e53ac737SRik van Riel return poisonret; 42177eae74afSKirill A. Shutemov } 42187eae74afSKirill A. Shutemov 42197eae74afSKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) 4220667240e0SJan Kara lock_page(vmf->page); 42217eae74afSKirill A. Shutemov else 4222667240e0SJan Kara VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); 42237eae74afSKirill A. Shutemov 42247eae74afSKirill A. Shutemov return ret; 42257eae74afSKirill A. Shutemov } 42267eae74afSKirill A. Shutemov 4227396bcc52SMatthew Wilcox (Oracle) #ifdef CONFIG_TRANSPARENT_HUGEPAGE 422882b0f8c3SJan Kara static void deposit_prealloc_pte(struct vm_fault *vmf) 4229953c66c2SAneesh Kumar K.V { 423082b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4231953c66c2SAneesh Kumar K.V 423282b0f8c3SJan Kara pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); 4233953c66c2SAneesh Kumar K.V /* 4234953c66c2SAneesh Kumar K.V * We are going to consume the prealloc table, 4235953c66c2SAneesh Kumar K.V * count that as nr_ptes. 4236953c66c2SAneesh Kumar K.V */ 4237c4812909SKirill A. Shutemov mm_inc_nr_ptes(vma->vm_mm); 42387f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 4239953c66c2SAneesh Kumar K.V } 4240953c66c2SAneesh Kumar K.V 4241f9ce0be7SKirill A. Shutemov vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) 424210102459SKirill A. Shutemov { 424382b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 424482b0f8c3SJan Kara bool write = vmf->flags & FAULT_FLAG_WRITE; 424582b0f8c3SJan Kara unsigned long haddr = vmf->address & HPAGE_PMD_MASK; 424610102459SKirill A. Shutemov pmd_t entry; 42472b740303SSouptick Joarder int i; 4248d01ac3c3SMatthew Wilcox (Oracle) vm_fault_t ret = VM_FAULT_FALLBACK; 424910102459SKirill A. Shutemov 425010102459SKirill A. Shutemov if (!transhuge_vma_suitable(vma, haddr)) 4251d01ac3c3SMatthew Wilcox (Oracle) return ret; 425210102459SKirill A. Shutemov 425310102459SKirill A. Shutemov page = compound_head(page); 4254d01ac3c3SMatthew Wilcox (Oracle) if (compound_order(page) != HPAGE_PMD_ORDER) 4255d01ac3c3SMatthew Wilcox (Oracle) return ret; 425610102459SKirill A. Shutemov 4257953c66c2SAneesh Kumar K.V /* 4258eac96c3eSYang Shi * Just backoff if any subpage of a THP is corrupted otherwise 4259eac96c3eSYang Shi * the corrupted page may mapped by PMD silently to escape the 4260eac96c3eSYang Shi * check. This kind of THP just can be PTE mapped. Access to 4261eac96c3eSYang Shi * the corrupted subpage should trigger SIGBUS as expected. 4262eac96c3eSYang Shi */ 4263eac96c3eSYang Shi if (unlikely(PageHasHWPoisoned(page))) 4264eac96c3eSYang Shi return ret; 4265eac96c3eSYang Shi 4266eac96c3eSYang Shi /* 4267f0953a1bSIngo Molnar * Archs like ppc64 need additional space to store information 4268953c66c2SAneesh Kumar K.V * related to pte entry. Use the preallocated table for that. 4269953c66c2SAneesh Kumar K.V */ 427082b0f8c3SJan Kara if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { 42714cf58924SJoel Fernandes (Google) vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); 427282b0f8c3SJan Kara if (!vmf->prealloc_pte) 4273953c66c2SAneesh Kumar K.V return VM_FAULT_OOM; 4274953c66c2SAneesh Kumar K.V } 4275953c66c2SAneesh Kumar K.V 427682b0f8c3SJan Kara vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); 427782b0f8c3SJan Kara if (unlikely(!pmd_none(*vmf->pmd))) 427810102459SKirill A. Shutemov goto out; 427910102459SKirill A. Shutemov 428010102459SKirill A. Shutemov for (i = 0; i < HPAGE_PMD_NR; i++) 428110102459SKirill A. Shutemov flush_icache_page(vma, page + i); 428210102459SKirill A. Shutemov 428310102459SKirill A. Shutemov entry = mk_huge_pmd(page, vma->vm_page_prot); 428410102459SKirill A. Shutemov if (write) 4285f55e1014SLinus Torvalds entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); 428610102459SKirill A. Shutemov 4287fadae295SYang Shi add_mm_counter(vma->vm_mm, mm_counter_file(page), HPAGE_PMD_NR); 4288cea86fe2SHugh Dickins page_add_file_rmap(page, vma, true); 4289cea86fe2SHugh Dickins 4290953c66c2SAneesh Kumar K.V /* 4291953c66c2SAneesh Kumar K.V * deposit and withdraw with pmd lock held 4292953c66c2SAneesh Kumar K.V */ 4293953c66c2SAneesh Kumar K.V if (arch_needs_pgtable_deposit()) 429482b0f8c3SJan Kara deposit_prealloc_pte(vmf); 429510102459SKirill A. Shutemov 429682b0f8c3SJan Kara set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); 429710102459SKirill A. Shutemov 429882b0f8c3SJan Kara update_mmu_cache_pmd(vma, haddr, vmf->pmd); 429910102459SKirill A. Shutemov 430010102459SKirill A. Shutemov /* fault is handled */ 430110102459SKirill A. Shutemov ret = 0; 430295ecedcdSKirill A. Shutemov count_vm_event(THP_FILE_MAPPED); 430310102459SKirill A. Shutemov out: 430482b0f8c3SJan Kara spin_unlock(vmf->ptl); 430510102459SKirill A. Shutemov return ret; 430610102459SKirill A. Shutemov } 430710102459SKirill A. Shutemov #else 4308f9ce0be7SKirill A. Shutemov vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) 430910102459SKirill A. Shutemov { 4310f9ce0be7SKirill A. Shutemov return VM_FAULT_FALLBACK; 431110102459SKirill A. Shutemov } 431210102459SKirill A. Shutemov #endif 431310102459SKirill A. Shutemov 43149d3af4b4SWill Deacon void do_set_pte(struct vm_fault *vmf, struct page *page, unsigned long addr) 43153bb97794SKirill A. Shutemov { 431682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 43172bad466cSPeter Xu bool uffd_wp = vmf_orig_pte_uffd_wp(vmf); 431882b0f8c3SJan Kara bool write = vmf->flags & FAULT_FLAG_WRITE; 43199d3af4b4SWill Deacon bool prefault = vmf->address != addr; 43203bb97794SKirill A. Shutemov pte_t entry; 43217267ec00SKirill A. Shutemov 43223bb97794SKirill A. Shutemov flush_icache_page(vma, page); 43233bb97794SKirill A. Shutemov entry = mk_pte(page, vma->vm_page_prot); 432446bdb427SWill Deacon 432546bdb427SWill Deacon if (prefault && arch_wants_old_prefaulted_pte()) 432646bdb427SWill Deacon entry = pte_mkold(entry); 432750c25ee9SThomas Bogendoerfer else 432850c25ee9SThomas Bogendoerfer entry = pte_sw_mkyoung(entry); 432946bdb427SWill Deacon 43303bb97794SKirill A. Shutemov if (write) 43313bb97794SKirill A. Shutemov entry = maybe_mkwrite(pte_mkdirty(entry), vma); 43329c28a205SPeter Xu if (unlikely(uffd_wp)) 4333f1eb1bacSPeter Xu entry = pte_mkuffd_wp(entry); 4334bae473a4SKirill A. Shutemov /* copy-on-write page */ 4335bae473a4SKirill A. Shutemov if (write && !(vma->vm_flags & VM_SHARED)) { 4336f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, MM_ANONPAGES); 433740f2bbf7SDavid Hildenbrand page_add_new_anon_rmap(page, vma, addr); 4338b518154eSJoonsoo Kim lru_cache_add_inactive_or_unevictable(page, vma); 43393bb97794SKirill A. Shutemov } else { 4340f1a79412SShakeel Butt inc_mm_counter(vma->vm_mm, mm_counter_file(page)); 4341cea86fe2SHugh Dickins page_add_file_rmap(page, vma, false); 43423bb97794SKirill A. Shutemov } 43439d3af4b4SWill Deacon set_pte_at(vma->vm_mm, addr, vmf->pte, entry); 43443bb97794SKirill A. Shutemov } 43453bb97794SKirill A. Shutemov 4346f46f2adeSPeter Xu static bool vmf_pte_changed(struct vm_fault *vmf) 4347f46f2adeSPeter Xu { 4348f46f2adeSPeter Xu if (vmf->flags & FAULT_FLAG_ORIG_PTE_VALID) 4349c33c7948SRyan Roberts return !pte_same(ptep_get(vmf->pte), vmf->orig_pte); 4350f46f2adeSPeter Xu 4351c33c7948SRyan Roberts return !pte_none(ptep_get(vmf->pte)); 4352f46f2adeSPeter Xu } 4353f46f2adeSPeter Xu 43549118c0cbSJan Kara /** 43559118c0cbSJan Kara * finish_fault - finish page fault once we have prepared the page to fault 43569118c0cbSJan Kara * 43579118c0cbSJan Kara * @vmf: structure describing the fault 43589118c0cbSJan Kara * 43599118c0cbSJan Kara * This function handles all that is needed to finish a page fault once the 43609118c0cbSJan Kara * page to fault in is prepared. It handles locking of PTEs, inserts PTE for 43619118c0cbSJan Kara * given page, adds reverse page mapping, handles memcg charges and LRU 4362a862f68aSMike Rapoport * addition. 43639118c0cbSJan Kara * 43649118c0cbSJan Kara * The function expects the page to be locked and on success it consumes a 43659118c0cbSJan Kara * reference of a page being mapped (for the PTE which maps it). 4366a862f68aSMike Rapoport * 4367a862f68aSMike Rapoport * Return: %0 on success, %VM_FAULT_ code in case of error. 43689118c0cbSJan Kara */ 43692b740303SSouptick Joarder vm_fault_t finish_fault(struct vm_fault *vmf) 43709118c0cbSJan Kara { 4371f9ce0be7SKirill A. Shutemov struct vm_area_struct *vma = vmf->vma; 43729118c0cbSJan Kara struct page *page; 4373f9ce0be7SKirill A. Shutemov vm_fault_t ret; 43749118c0cbSJan Kara 43759118c0cbSJan Kara /* Did we COW the page? */ 4376f9ce0be7SKirill A. Shutemov if ((vmf->flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) 43779118c0cbSJan Kara page = vmf->cow_page; 43789118c0cbSJan Kara else 43799118c0cbSJan Kara page = vmf->page; 43806b31d595SMichal Hocko 43816b31d595SMichal Hocko /* 43826b31d595SMichal Hocko * check even for read faults because we might have lost our CoWed 43836b31d595SMichal Hocko * page 43846b31d595SMichal Hocko */ 4385f9ce0be7SKirill A. Shutemov if (!(vma->vm_flags & VM_SHARED)) { 4386f9ce0be7SKirill A. Shutemov ret = check_stable_address_space(vma->vm_mm); 4387f9ce0be7SKirill A. Shutemov if (ret) 4388f9ce0be7SKirill A. Shutemov return ret; 4389f9ce0be7SKirill A. Shutemov } 4390f9ce0be7SKirill A. Shutemov 4391f9ce0be7SKirill A. Shutemov if (pmd_none(*vmf->pmd)) { 4392f9ce0be7SKirill A. Shutemov if (PageTransCompound(page)) { 4393f9ce0be7SKirill A. Shutemov ret = do_set_pmd(vmf, page); 4394f9ce0be7SKirill A. Shutemov if (ret != VM_FAULT_FALLBACK) 4395f9ce0be7SKirill A. Shutemov return ret; 4396f9ce0be7SKirill A. Shutemov } 4397f9ce0be7SKirill A. Shutemov 439803c4f204SQi Zheng if (vmf->prealloc_pte) 439903c4f204SQi Zheng pmd_install(vma->vm_mm, vmf->pmd, &vmf->prealloc_pte); 440003c4f204SQi Zheng else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd))) 4401f9ce0be7SKirill A. Shutemov return VM_FAULT_OOM; 4402f9ce0be7SKirill A. Shutemov } 4403f9ce0be7SKirill A. Shutemov 4404f9ce0be7SKirill A. Shutemov vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 4405f9ce0be7SKirill A. Shutemov vmf->address, &vmf->ptl); 44063db82b93SHugh Dickins if (!vmf->pte) 44073db82b93SHugh Dickins return VM_FAULT_NOPAGE; 4408f9ce0be7SKirill A. Shutemov 440970427f6eSSergei Antonov /* Re-check under ptl */ 441070427f6eSSergei Antonov if (likely(!vmf_pte_changed(vmf))) { 441170427f6eSSergei Antonov do_set_pte(vmf, page, vmf->address); 441270427f6eSSergei Antonov 441370427f6eSSergei Antonov /* no need to invalidate: a not-present page won't be cached */ 441470427f6eSSergei Antonov update_mmu_cache(vma, vmf->address, vmf->pte); 441570427f6eSSergei Antonov 441670427f6eSSergei Antonov ret = 0; 441770427f6eSSergei Antonov } else { 4418f9ce0be7SKirill A. Shutemov update_mmu_tlb(vma, vmf->address, vmf->pte); 441970427f6eSSergei Antonov ret = VM_FAULT_NOPAGE; 442070427f6eSSergei Antonov } 442170427f6eSSergei Antonov 44229118c0cbSJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 44239118c0cbSJan Kara return ret; 44249118c0cbSJan Kara } 44259118c0cbSJan Kara 442653d36a56SLorenzo Stoakes static unsigned long fault_around_pages __read_mostly = 442753d36a56SLorenzo Stoakes 65536 >> PAGE_SHIFT; 4428a9b0f861SKirill A. Shutemov 44291592eef0SKirill A. Shutemov #ifdef CONFIG_DEBUG_FS 4430a9b0f861SKirill A. Shutemov static int fault_around_bytes_get(void *data, u64 *val) 44311592eef0SKirill A. Shutemov { 443253d36a56SLorenzo Stoakes *val = fault_around_pages << PAGE_SHIFT; 44331592eef0SKirill A. Shutemov return 0; 44341592eef0SKirill A. Shutemov } 44351592eef0SKirill A. Shutemov 4436b4903d6eSAndrey Ryabinin /* 4437da391d64SWilliam Kucharski * fault_around_bytes must be rounded down to the nearest page order as it's 4438da391d64SWilliam Kucharski * what do_fault_around() expects to see. 4439b4903d6eSAndrey Ryabinin */ 4440a9b0f861SKirill A. Shutemov static int fault_around_bytes_set(void *data, u64 val) 44411592eef0SKirill A. Shutemov { 4442a9b0f861SKirill A. Shutemov if (val / PAGE_SIZE > PTRS_PER_PTE) 44431592eef0SKirill A. Shutemov return -EINVAL; 444453d36a56SLorenzo Stoakes 444553d36a56SLorenzo Stoakes /* 444653d36a56SLorenzo Stoakes * The minimum value is 1 page, however this results in no fault-around 444753d36a56SLorenzo Stoakes * at all. See should_fault_around(). 444853d36a56SLorenzo Stoakes */ 444953d36a56SLorenzo Stoakes fault_around_pages = max(rounddown_pow_of_two(val) >> PAGE_SHIFT, 1UL); 445053d36a56SLorenzo Stoakes 44511592eef0SKirill A. Shutemov return 0; 44521592eef0SKirill A. Shutemov } 44530a1345f8SYevgen Pronenko DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, 4454a9b0f861SKirill A. Shutemov fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); 44551592eef0SKirill A. Shutemov 44561592eef0SKirill A. Shutemov static int __init fault_around_debugfs(void) 44571592eef0SKirill A. Shutemov { 4458d9f7979cSGreg Kroah-Hartman debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, 4459a9b0f861SKirill A. Shutemov &fault_around_bytes_fops); 44601592eef0SKirill A. Shutemov return 0; 44611592eef0SKirill A. Shutemov } 44621592eef0SKirill A. Shutemov late_initcall(fault_around_debugfs); 44631592eef0SKirill A. Shutemov #endif 44648c6e50b0SKirill A. Shutemov 44651fdb412bSKirill A. Shutemov /* 44661fdb412bSKirill A. Shutemov * do_fault_around() tries to map few pages around the fault address. The hope 44671fdb412bSKirill A. Shutemov * is that the pages will be needed soon and this will lower the number of 44681fdb412bSKirill A. Shutemov * faults to handle. 44691fdb412bSKirill A. Shutemov * 44701fdb412bSKirill A. Shutemov * It uses vm_ops->map_pages() to map the pages, which skips the page if it's 44711fdb412bSKirill A. Shutemov * not ready to be mapped: not up-to-date, locked, etc. 44721fdb412bSKirill A. Shutemov * 44739042599eSLorenzo Stoakes * This function doesn't cross VMA or page table boundaries, in order to call 44749042599eSLorenzo Stoakes * map_pages() and acquire a PTE lock only once. 44751fdb412bSKirill A. Shutemov * 447653d36a56SLorenzo Stoakes * fault_around_pages defines how many pages we'll try to map. 4477da391d64SWilliam Kucharski * do_fault_around() expects it to be set to a power of two less than or equal 4478da391d64SWilliam Kucharski * to PTRS_PER_PTE. 44791fdb412bSKirill A. Shutemov * 4480da391d64SWilliam Kucharski * The virtual address of the area that we map is naturally aligned to 448153d36a56SLorenzo Stoakes * fault_around_pages * PAGE_SIZE rounded down to the machine page size 4482da391d64SWilliam Kucharski * (and therefore to page order). This way it's easier to guarantee 4483da391d64SWilliam Kucharski * that we don't cross page table boundaries. 44841fdb412bSKirill A. Shutemov */ 44852b740303SSouptick Joarder static vm_fault_t do_fault_around(struct vm_fault *vmf) 44868c6e50b0SKirill A. Shutemov { 448753d36a56SLorenzo Stoakes pgoff_t nr_pages = READ_ONCE(fault_around_pages); 44889042599eSLorenzo Stoakes pgoff_t pte_off = pte_index(vmf->address); 44899042599eSLorenzo Stoakes /* The page offset of vmf->address within the VMA. */ 44909042599eSLorenzo Stoakes pgoff_t vma_off = vmf->pgoff - vmf->vma->vm_pgoff; 44919042599eSLorenzo Stoakes pgoff_t from_pte, to_pte; 449258ef47efSMatthew Wilcox (Oracle) vm_fault_t ret; 44938c6e50b0SKirill A. Shutemov 44949042599eSLorenzo Stoakes /* The PTE offset of the start address, clamped to the VMA. */ 44959042599eSLorenzo Stoakes from_pte = max(ALIGN_DOWN(pte_off, nr_pages), 44969042599eSLorenzo Stoakes pte_off - min(pte_off, vma_off)); 4497aecd6f44SKirill A. Shutemov 44989042599eSLorenzo Stoakes /* The PTE offset of the end address, clamped to the VMA and PTE. */ 44999042599eSLorenzo Stoakes to_pte = min3(from_pte + nr_pages, (pgoff_t)PTRS_PER_PTE, 45009042599eSLorenzo Stoakes pte_off + vma_pages(vmf->vma) - vma_off) - 1; 45018c6e50b0SKirill A. Shutemov 450282b0f8c3SJan Kara if (pmd_none(*vmf->pmd)) { 45034cf58924SJoel Fernandes (Google) vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm); 450482b0f8c3SJan Kara if (!vmf->prealloc_pte) 4505f9ce0be7SKirill A. Shutemov return VM_FAULT_OOM; 45068c6e50b0SKirill A. Shutemov } 45078c6e50b0SKirill A. Shutemov 450858ef47efSMatthew Wilcox (Oracle) rcu_read_lock(); 450958ef47efSMatthew Wilcox (Oracle) ret = vmf->vma->vm_ops->map_pages(vmf, 45109042599eSLorenzo Stoakes vmf->pgoff + from_pte - pte_off, 45119042599eSLorenzo Stoakes vmf->pgoff + to_pte - pte_off); 451258ef47efSMatthew Wilcox (Oracle) rcu_read_unlock(); 451358ef47efSMatthew Wilcox (Oracle) 451458ef47efSMatthew Wilcox (Oracle) return ret; 45157267ec00SKirill A. Shutemov } 45167267ec00SKirill A. Shutemov 45179c28a205SPeter Xu /* Return true if we should do read fault-around, false otherwise */ 45189c28a205SPeter Xu static inline bool should_fault_around(struct vm_fault *vmf) 45199c28a205SPeter Xu { 45209c28a205SPeter Xu /* No ->map_pages? No way to fault around... */ 45219c28a205SPeter Xu if (!vmf->vma->vm_ops->map_pages) 45229c28a205SPeter Xu return false; 45239c28a205SPeter Xu 45249c28a205SPeter Xu if (uffd_disable_fault_around(vmf->vma)) 45259c28a205SPeter Xu return false; 45269c28a205SPeter Xu 452753d36a56SLorenzo Stoakes /* A single page implies no faulting 'around' at all. */ 452853d36a56SLorenzo Stoakes return fault_around_pages > 1; 45299c28a205SPeter Xu } 45309c28a205SPeter Xu 45312b740303SSouptick Joarder static vm_fault_t do_read_fault(struct vm_fault *vmf) 4532e655fb29SKirill A. Shutemov { 45332b740303SSouptick Joarder vm_fault_t ret = 0; 453422d1e68fSSidhartha Kumar struct folio *folio; 45358c6e50b0SKirill A. Shutemov 45368c6e50b0SKirill A. Shutemov /* 45378c6e50b0SKirill A. Shutemov * Let's call ->map_pages() first and use ->fault() as fallback 45388c6e50b0SKirill A. Shutemov * if page by the offset is not ready to be mapped (cold cache or 45398c6e50b0SKirill A. Shutemov * something). 45408c6e50b0SKirill A. Shutemov */ 45419c28a205SPeter Xu if (should_fault_around(vmf)) { 45420721ec8bSJan Kara ret = do_fault_around(vmf); 45437267ec00SKirill A. Shutemov if (ret) 45447267ec00SKirill A. Shutemov return ret; 45458c6e50b0SKirill A. Shutemov } 4546e655fb29SKirill A. Shutemov 4547936ca80dSJan Kara ret = __do_fault(vmf); 4548e655fb29SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4549e655fb29SKirill A. Shutemov return ret; 4550e655fb29SKirill A. Shutemov 45519118c0cbSJan Kara ret |= finish_fault(vmf); 455222d1e68fSSidhartha Kumar folio = page_folio(vmf->page); 455322d1e68fSSidhartha Kumar folio_unlock(folio); 45547267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 455522d1e68fSSidhartha Kumar folio_put(folio); 4556e655fb29SKirill A. Shutemov return ret; 4557e655fb29SKirill A. Shutemov } 4558e655fb29SKirill A. Shutemov 45592b740303SSouptick Joarder static vm_fault_t do_cow_fault(struct vm_fault *vmf) 4560ec47c3b9SKirill A. Shutemov { 456182b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 45622b740303SSouptick Joarder vm_fault_t ret; 4563ec47c3b9SKirill A. Shutemov 4564ec47c3b9SKirill A. Shutemov if (unlikely(anon_vma_prepare(vma))) 4565ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4566ec47c3b9SKirill A. Shutemov 4567936ca80dSJan Kara vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); 4568936ca80dSJan Kara if (!vmf->cow_page) 4569ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4570ec47c3b9SKirill A. Shutemov 45718f425e4eSMatthew Wilcox (Oracle) if (mem_cgroup_charge(page_folio(vmf->cow_page), vma->vm_mm, 45728f425e4eSMatthew Wilcox (Oracle) GFP_KERNEL)) { 4573936ca80dSJan Kara put_page(vmf->cow_page); 4574ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4575ec47c3b9SKirill A. Shutemov } 457668fa572bSKefeng Wang folio_throttle_swaprate(page_folio(vmf->cow_page), GFP_KERNEL); 4577ec47c3b9SKirill A. Shutemov 4578936ca80dSJan Kara ret = __do_fault(vmf); 4579ec47c3b9SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4580ec47c3b9SKirill A. Shutemov goto uncharge_out; 45813917048dSJan Kara if (ret & VM_FAULT_DONE_COW) 45823917048dSJan Kara return ret; 4583ec47c3b9SKirill A. Shutemov 4584936ca80dSJan Kara copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); 4585936ca80dSJan Kara __SetPageUptodate(vmf->cow_page); 4586ec47c3b9SKirill A. Shutemov 45879118c0cbSJan Kara ret |= finish_fault(vmf); 4588936ca80dSJan Kara unlock_page(vmf->page); 4589936ca80dSJan Kara put_page(vmf->page); 45907267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 45917267ec00SKirill A. Shutemov goto uncharge_out; 4592ec47c3b9SKirill A. Shutemov return ret; 4593ec47c3b9SKirill A. Shutemov uncharge_out: 4594936ca80dSJan Kara put_page(vmf->cow_page); 4595ec47c3b9SKirill A. Shutemov return ret; 4596ec47c3b9SKirill A. Shutemov } 4597ec47c3b9SKirill A. Shutemov 45982b740303SSouptick Joarder static vm_fault_t do_shared_fault(struct vm_fault *vmf) 45991da177e4SLinus Torvalds { 460082b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 46012b740303SSouptick Joarder vm_fault_t ret, tmp; 46026f609b7eSSidhartha Kumar struct folio *folio; 46031d65f86dSKAMEZAWA Hiroyuki 4604936ca80dSJan Kara ret = __do_fault(vmf); 46057eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4606f0c6d4d2SKirill A. Shutemov return ret; 46071da177e4SLinus Torvalds 46086f609b7eSSidhartha Kumar folio = page_folio(vmf->page); 46096f609b7eSSidhartha Kumar 46101da177e4SLinus Torvalds /* 4611f0c6d4d2SKirill A. Shutemov * Check if the backing address space wants to know that the page is 4612f0c6d4d2SKirill A. Shutemov * about to become writable 46131da177e4SLinus Torvalds */ 4614fb09a464SKirill A. Shutemov if (vma->vm_ops->page_mkwrite) { 46156f609b7eSSidhartha Kumar folio_unlock(folio); 461686aa6998SSidhartha Kumar tmp = do_page_mkwrite(vmf, folio); 4617fb09a464SKirill A. Shutemov if (unlikely(!tmp || 4618fb09a464SKirill A. Shutemov (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 46196f609b7eSSidhartha Kumar folio_put(folio); 4620f0c6d4d2SKirill A. Shutemov return tmp; 462169676147SMark Fasheh } 4622d0217ac0SNick Piggin } 4623fb09a464SKirill A. Shutemov 46249118c0cbSJan Kara ret |= finish_fault(vmf); 46257267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | 46267267ec00SKirill A. Shutemov VM_FAULT_RETRY))) { 46276f609b7eSSidhartha Kumar folio_unlock(folio); 46286f609b7eSSidhartha Kumar folio_put(folio); 4629f0c6d4d2SKirill A. Shutemov return ret; 46309637a5efSDavid Howells } 4631d00806b1SNick Piggin 463289b15332SJohannes Weiner ret |= fault_dirty_shared_page(vmf); 4633b827e496SNick Piggin return ret; 463454cb8821SNick Piggin } 4635d00806b1SNick Piggin 46369a95f3cfSPaul Cassella /* 4637c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 46389a95f3cfSPaul Cassella * but allow concurrent faults). 4639c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 46409138e47eSMatthew Wilcox (Oracle) * return value. See filemap_fault() and __folio_lock_or_retry(). 4641c1e8d7c6SMichel Lespinasse * If mmap_lock is released, vma may become invalid (for example 4642fc8efd2dSJan Stancek * by other thread calling munmap()). 46439a95f3cfSPaul Cassella */ 46442b740303SSouptick Joarder static vm_fault_t do_fault(struct vm_fault *vmf) 464554cb8821SNick Piggin { 464682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4647fc8efd2dSJan Stancek struct mm_struct *vm_mm = vma->vm_mm; 46482b740303SSouptick Joarder vm_fault_t ret; 464954cb8821SNick Piggin 4650ff09d7ecSAneesh Kumar K.V /* 4651ff09d7ecSAneesh Kumar K.V * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND 4652ff09d7ecSAneesh Kumar K.V */ 4653ff09d7ecSAneesh Kumar K.V if (!vma->vm_ops->fault) { 46543db82b93SHugh Dickins vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, 46553db82b93SHugh Dickins vmf->address, &vmf->ptl); 46563db82b93SHugh Dickins if (unlikely(!vmf->pte)) 4657b0b9b3dfSHugh Dickins ret = VM_FAULT_SIGBUS; 4658ff09d7ecSAneesh Kumar K.V else { 4659ff09d7ecSAneesh Kumar K.V /* 4660ff09d7ecSAneesh Kumar K.V * Make sure this is not a temporary clearing of pte 4661ff09d7ecSAneesh Kumar K.V * by holding ptl and checking again. A R/M/W update 4662ff09d7ecSAneesh Kumar K.V * of pte involves: take ptl, clearing the pte so that 4663ff09d7ecSAneesh Kumar K.V * we don't have concurrent modification by hardware 4664ff09d7ecSAneesh Kumar K.V * followed by an update. 4665ff09d7ecSAneesh Kumar K.V */ 4666c33c7948SRyan Roberts if (unlikely(pte_none(ptep_get(vmf->pte)))) 4667ff09d7ecSAneesh Kumar K.V ret = VM_FAULT_SIGBUS; 4668ff09d7ecSAneesh Kumar K.V else 4669ff09d7ecSAneesh Kumar K.V ret = VM_FAULT_NOPAGE; 4670ff09d7ecSAneesh Kumar K.V 4671ff09d7ecSAneesh Kumar K.V pte_unmap_unlock(vmf->pte, vmf->ptl); 4672ff09d7ecSAneesh Kumar K.V } 4673ff09d7ecSAneesh Kumar K.V } else if (!(vmf->flags & FAULT_FLAG_WRITE)) 4674b0b9b3dfSHugh Dickins ret = do_read_fault(vmf); 4675b0b9b3dfSHugh Dickins else if (!(vma->vm_flags & VM_SHARED)) 4676b0b9b3dfSHugh Dickins ret = do_cow_fault(vmf); 4677b0b9b3dfSHugh Dickins else 4678b0b9b3dfSHugh Dickins ret = do_shared_fault(vmf); 4679b0b9b3dfSHugh Dickins 4680b0b9b3dfSHugh Dickins /* preallocated pagetable is unused: free it */ 4681b0b9b3dfSHugh Dickins if (vmf->prealloc_pte) { 4682fc8efd2dSJan Stancek pte_free(vm_mm, vmf->prealloc_pte); 46837f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 4684b0b9b3dfSHugh Dickins } 4685b0b9b3dfSHugh Dickins return ret; 468654cb8821SNick Piggin } 468754cb8821SNick Piggin 4688f4c0d836SYang Shi int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, 4689f4c0d836SYang Shi unsigned long addr, int page_nid, int *flags) 46909532fec1SMel Gorman { 46919532fec1SMel Gorman get_page(page); 46929532fec1SMel Gorman 4693fc137c0dSRaghavendra K T /* Record the current PID acceesing VMA */ 4694fc137c0dSRaghavendra K T vma_set_access_pid_bit(vma); 4695fc137c0dSRaghavendra K T 46969532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS); 469704bb2f94SRik van Riel if (page_nid == numa_node_id()) { 46989532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); 469904bb2f94SRik van Riel *flags |= TNF_FAULT_LOCAL; 470004bb2f94SRik van Riel } 47019532fec1SMel Gorman 47029532fec1SMel Gorman return mpol_misplaced(page, vma, addr); 47039532fec1SMel Gorman } 47049532fec1SMel Gorman 47052b740303SSouptick Joarder static vm_fault_t do_numa_page(struct vm_fault *vmf) 4706d10e63f2SMel Gorman { 470782b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 47084daae3b4SMel Gorman struct page *page = NULL; 470998fa15f3SAnshuman Khandual int page_nid = NUMA_NO_NODE; 47106a56ccbcSDavid Hildenbrand bool writable = false; 471190572890SPeter Zijlstra int last_cpupid; 4712cbee9f88SPeter Zijlstra int target_nid; 471304a86453SAneesh Kumar K.V pte_t pte, old_pte; 47146688cc05SPeter Zijlstra int flags = 0; 4715d10e63f2SMel Gorman 4716d10e63f2SMel Gorman /* 4717d10e63f2SMel Gorman * The "pte" at this point cannot be used safely without 4718d10e63f2SMel Gorman * validation through pte_unmap_same(). It's of NUMA type but 4719d10e63f2SMel Gorman * the pfn may be screwed if the read is non atomic. 4720d10e63f2SMel Gorman */ 472182b0f8c3SJan Kara spin_lock(vmf->ptl); 4722c33c7948SRyan Roberts if (unlikely(!pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 472382b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 47244daae3b4SMel Gorman goto out; 47254daae3b4SMel Gorman } 47264daae3b4SMel Gorman 4727b99a342dSHuang Ying /* Get the normal PTE */ 4728b99a342dSHuang Ying old_pte = ptep_get(vmf->pte); 472904a86453SAneesh Kumar K.V pte = pte_modify(old_pte, vma->vm_page_prot); 4730d10e63f2SMel Gorman 47316a56ccbcSDavid Hildenbrand /* 47326a56ccbcSDavid Hildenbrand * Detect now whether the PTE could be writable; this information 47336a56ccbcSDavid Hildenbrand * is only valid while holding the PT lock. 47346a56ccbcSDavid Hildenbrand */ 47356a56ccbcSDavid Hildenbrand writable = pte_write(pte); 47366a56ccbcSDavid Hildenbrand if (!writable && vma_wants_manual_pte_write_upgrade(vma) && 47376a56ccbcSDavid Hildenbrand can_change_pte_writable(vma, vmf->address, pte)) 47386a56ccbcSDavid Hildenbrand writable = true; 47396a56ccbcSDavid Hildenbrand 474082b0f8c3SJan Kara page = vm_normal_page(vma, vmf->address, pte); 47413218f871SAlex Sierra if (!page || is_zone_device_page(page)) 4742b99a342dSHuang Ying goto out_map; 4743d10e63f2SMel Gorman 4744e81c4802SKirill A. Shutemov /* TODO: handle PTE-mapped THP */ 4745b99a342dSHuang Ying if (PageCompound(page)) 4746b99a342dSHuang Ying goto out_map; 4747e81c4802SKirill A. Shutemov 47486688cc05SPeter Zijlstra /* 4749bea66fbdSMel Gorman * Avoid grouping on RO pages in general. RO pages shouldn't hurt as 4750bea66fbdSMel Gorman * much anyway since they can be in shared cache state. This misses 4751bea66fbdSMel Gorman * the case where a mapping is writable but the process never writes 4752bea66fbdSMel Gorman * to it but pte_write gets cleared during protection updates and 4753bea66fbdSMel Gorman * pte_dirty has unpredictable behaviour between PTE scan updates, 4754bea66fbdSMel Gorman * background writeback, dirty balancing and application behaviour. 47556688cc05SPeter Zijlstra */ 47566a56ccbcSDavid Hildenbrand if (!writable) 47576688cc05SPeter Zijlstra flags |= TNF_NO_GROUP; 47586688cc05SPeter Zijlstra 4759dabe1d99SRik van Riel /* 4760dabe1d99SRik van Riel * Flag if the page is shared between multiple address spaces. This 4761dabe1d99SRik van Riel * is later used when determining whether to group tasks together 4762dabe1d99SRik van Riel */ 4763dabe1d99SRik van Riel if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) 4764dabe1d99SRik van Riel flags |= TNF_SHARED; 4765dabe1d99SRik van Riel 47668191acbdSMel Gorman page_nid = page_to_nid(page); 476733024536SHuang Ying /* 476833024536SHuang Ying * For memory tiering mode, cpupid of slow memory page is used 476933024536SHuang Ying * to record page access time. So use default value. 477033024536SHuang Ying */ 477133024536SHuang Ying if ((sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) && 477233024536SHuang Ying !node_is_toptier(page_nid)) 477333024536SHuang Ying last_cpupid = (-1 & LAST_CPUPID_MASK); 477433024536SHuang Ying else 477533024536SHuang Ying last_cpupid = page_cpupid_last(page); 477682b0f8c3SJan Kara target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, 4777bae473a4SKirill A. Shutemov &flags); 477898fa15f3SAnshuman Khandual if (target_nid == NUMA_NO_NODE) { 47794daae3b4SMel Gorman put_page(page); 4780b99a342dSHuang Ying goto out_map; 47814daae3b4SMel Gorman } 4782b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 47836a56ccbcSDavid Hildenbrand writable = false; 47844daae3b4SMel Gorman 47854daae3b4SMel Gorman /* Migrate to the requested node */ 4786bf90ac19SWang Qing if (migrate_misplaced_page(page, vma, target_nid)) { 47878191acbdSMel Gorman page_nid = target_nid; 47886688cc05SPeter Zijlstra flags |= TNF_MIGRATED; 4789b99a342dSHuang Ying } else { 4790074c2381SMel Gorman flags |= TNF_MIGRATE_FAIL; 4791c7ad0880SHugh Dickins vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 4792c7ad0880SHugh Dickins vmf->address, &vmf->ptl); 4793c7ad0880SHugh Dickins if (unlikely(!vmf->pte)) 4794c7ad0880SHugh Dickins goto out; 4795c33c7948SRyan Roberts if (unlikely(!pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { 4796b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 4797b99a342dSHuang Ying goto out; 4798b99a342dSHuang Ying } 4799b99a342dSHuang Ying goto out_map; 4800b99a342dSHuang Ying } 48014daae3b4SMel Gorman 48024daae3b4SMel Gorman out: 480398fa15f3SAnshuman Khandual if (page_nid != NUMA_NO_NODE) 48046688cc05SPeter Zijlstra task_numa_fault(last_cpupid, page_nid, 1, flags); 4805d10e63f2SMel Gorman return 0; 4806b99a342dSHuang Ying out_map: 4807b99a342dSHuang Ying /* 4808b99a342dSHuang Ying * Make it present again, depending on how arch implements 4809b99a342dSHuang Ying * non-accessible ptes, some can allow access by kernel mode. 4810b99a342dSHuang Ying */ 4811b99a342dSHuang Ying old_pte = ptep_modify_prot_start(vma, vmf->address, vmf->pte); 4812b99a342dSHuang Ying pte = pte_modify(old_pte, vma->vm_page_prot); 4813b99a342dSHuang Ying pte = pte_mkyoung(pte); 48146a56ccbcSDavid Hildenbrand if (writable) 4815b99a342dSHuang Ying pte = pte_mkwrite(pte); 4816b99a342dSHuang Ying ptep_modify_prot_commit(vma, vmf->address, vmf->pte, old_pte, pte); 4817b99a342dSHuang Ying update_mmu_cache(vma, vmf->address, vmf->pte); 4818b99a342dSHuang Ying pte_unmap_unlock(vmf->pte, vmf->ptl); 4819b99a342dSHuang Ying goto out; 4820d10e63f2SMel Gorman } 4821d10e63f2SMel Gorman 48222b740303SSouptick Joarder static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf) 4823b96375f7SMatthew Wilcox { 4824f4200391SDave Jiang if (vma_is_anonymous(vmf->vma)) 482582b0f8c3SJan Kara return do_huge_pmd_anonymous_page(vmf); 4826a2d58167SDave Jiang if (vmf->vma->vm_ops->huge_fault) 4827c791ace1SDave Jiang return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); 4828b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 4829b96375f7SMatthew Wilcox } 4830b96375f7SMatthew Wilcox 4831183f24aaSGeert Uytterhoeven /* `inline' is required to avoid gcc 4.1.2 build error */ 48325db4f15cSYang Shi static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf) 4833b96375f7SMatthew Wilcox { 4834c89357e2SDavid Hildenbrand const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; 4835aea06577SDavid Hildenbrand vm_fault_t ret; 4836c89357e2SDavid Hildenbrand 4837529b930bSAndrea Arcangeli if (vma_is_anonymous(vmf->vma)) { 4838c89357e2SDavid Hildenbrand if (likely(!unshare) && 4839c89357e2SDavid Hildenbrand userfaultfd_huge_pmd_wp(vmf->vma, vmf->orig_pmd)) 4840529b930bSAndrea Arcangeli return handle_userfault(vmf, VM_UFFD_WP); 48415db4f15cSYang Shi return do_huge_pmd_wp_page(vmf); 4842529b930bSAndrea Arcangeli } 4843af9e4d5fSKirill A. Shutemov 4844aea06577SDavid Hildenbrand if (vmf->vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 4845aea06577SDavid Hildenbrand if (vmf->vma->vm_ops->huge_fault) { 4846aea06577SDavid Hildenbrand ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); 4847327e9fd4SThomas Hellstrom (VMware) if (!(ret & VM_FAULT_FALLBACK)) 4848327e9fd4SThomas Hellstrom (VMware) return ret; 4849327e9fd4SThomas Hellstrom (VMware) } 4850aea06577SDavid Hildenbrand } 4851327e9fd4SThomas Hellstrom (VMware) 4852327e9fd4SThomas Hellstrom (VMware) /* COW or write-notify handled on pte level: split pmd. */ 485382b0f8c3SJan Kara __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); 4854af9e4d5fSKirill A. Shutemov 4855b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 4856b96375f7SMatthew Wilcox } 4857b96375f7SMatthew Wilcox 48582b740303SSouptick Joarder static vm_fault_t create_huge_pud(struct vm_fault *vmf) 4859a00cc7d9SMatthew Wilcox { 4860327e9fd4SThomas Hellstrom (VMware) #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ 4861327e9fd4SThomas Hellstrom (VMware) defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) 4862a00cc7d9SMatthew Wilcox /* No support for anonymous transparent PUD pages yet */ 4863a00cc7d9SMatthew Wilcox if (vma_is_anonymous(vmf->vma)) 486414c99d65SGowans, James return VM_FAULT_FALLBACK; 486514c99d65SGowans, James if (vmf->vma->vm_ops->huge_fault) 486614c99d65SGowans, James return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); 486714c99d65SGowans, James #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 486814c99d65SGowans, James return VM_FAULT_FALLBACK; 486914c99d65SGowans, James } 487014c99d65SGowans, James 487114c99d65SGowans, James static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) 487214c99d65SGowans, James { 487314c99d65SGowans, James #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ 487414c99d65SGowans, James defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) 4875aea06577SDavid Hildenbrand vm_fault_t ret; 4876aea06577SDavid Hildenbrand 487714c99d65SGowans, James /* No support for anonymous transparent PUD pages yet */ 487814c99d65SGowans, James if (vma_is_anonymous(vmf->vma)) 4879327e9fd4SThomas Hellstrom (VMware) goto split; 4880aea06577SDavid Hildenbrand if (vmf->vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { 4881327e9fd4SThomas Hellstrom (VMware) if (vmf->vma->vm_ops->huge_fault) { 4882aea06577SDavid Hildenbrand ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); 4883327e9fd4SThomas Hellstrom (VMware) if (!(ret & VM_FAULT_FALLBACK)) 4884327e9fd4SThomas Hellstrom (VMware) return ret; 4885327e9fd4SThomas Hellstrom (VMware) } 4886aea06577SDavid Hildenbrand } 4887327e9fd4SThomas Hellstrom (VMware) split: 4888327e9fd4SThomas Hellstrom (VMware) /* COW or write-notify not handled on PUD level: split pud.*/ 4889327e9fd4SThomas Hellstrom (VMware) __split_huge_pud(vmf->vma, vmf->pud, vmf->address); 489014c99d65SGowans, James #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ 4891a00cc7d9SMatthew Wilcox return VM_FAULT_FALLBACK; 4892a00cc7d9SMatthew Wilcox } 4893a00cc7d9SMatthew Wilcox 48941da177e4SLinus Torvalds /* 48951da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 48961da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 48971da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 48981da177e4SLinus Torvalds * 48991da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 49001da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 49011da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 49021da177e4SLinus Torvalds * 4903c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, but allow 49047267ec00SKirill A. Shutemov * concurrent faults). 49059a95f3cfSPaul Cassella * 4906c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our return value. 49079138e47eSMatthew Wilcox (Oracle) * See filemap_fault() and __folio_lock_or_retry(). 49081da177e4SLinus Torvalds */ 49092b740303SSouptick Joarder static vm_fault_t handle_pte_fault(struct vm_fault *vmf) 49101da177e4SLinus Torvalds { 49111da177e4SLinus Torvalds pte_t entry; 49121da177e4SLinus Torvalds 491382b0f8c3SJan Kara if (unlikely(pmd_none(*vmf->pmd))) { 49147267ec00SKirill A. Shutemov /* 49157267ec00SKirill A. Shutemov * Leave __pte_alloc() until later: because vm_ops->fault may 49167267ec00SKirill A. Shutemov * want to allocate huge page, and if we expose page table 49177267ec00SKirill A. Shutemov * for an instant, it will be difficult to retract from 49187267ec00SKirill A. Shutemov * concurrent faults and from rmap lookups. 49197267ec00SKirill A. Shutemov */ 492082b0f8c3SJan Kara vmf->pte = NULL; 4921f46f2adeSPeter Xu vmf->flags &= ~FAULT_FLAG_ORIG_PTE_VALID; 49227267ec00SKirill A. Shutemov } else { 4923f9ce0be7SKirill A. Shutemov /* 49247267ec00SKirill A. Shutemov * A regular pmd is established and it can't morph into a huge 4925c7ad0880SHugh Dickins * pmd by anon khugepaged, since that takes mmap_lock in write 4926c7ad0880SHugh Dickins * mode; but shmem or file collapse to THP could still morph 4927c7ad0880SHugh Dickins * it into a huge pmd: just retry later if so. 49287267ec00SKirill A. Shutemov */ 4929c7ad0880SHugh Dickins vmf->pte = pte_offset_map_nolock(vmf->vma->vm_mm, vmf->pmd, 4930c7ad0880SHugh Dickins vmf->address, &vmf->ptl); 4931c7ad0880SHugh Dickins if (unlikely(!vmf->pte)) 4932c7ad0880SHugh Dickins return 0; 493326e1a0c3SHugh Dickins vmf->orig_pte = ptep_get_lockless(vmf->pte); 4934f46f2adeSPeter Xu vmf->flags |= FAULT_FLAG_ORIG_PTE_VALID; 49357267ec00SKirill A. Shutemov 49362994302bSJan Kara if (pte_none(vmf->orig_pte)) { 493782b0f8c3SJan Kara pte_unmap(vmf->pte); 493882b0f8c3SJan Kara vmf->pte = NULL; 49397267ec00SKirill A. Shutemov } 49407267ec00SKirill A. Shutemov } 49417267ec00SKirill A. Shutemov 49422bad466cSPeter Xu if (!vmf->pte) 49432bad466cSPeter Xu return do_pte_missing(vmf); 49447267ec00SKirill A. Shutemov 49452994302bSJan Kara if (!pte_present(vmf->orig_pte)) 49462994302bSJan Kara return do_swap_page(vmf); 49471da177e4SLinus Torvalds 49482994302bSJan Kara if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) 49492994302bSJan Kara return do_numa_page(vmf); 4950d10e63f2SMel Gorman 495182b0f8c3SJan Kara spin_lock(vmf->ptl); 49522994302bSJan Kara entry = vmf->orig_pte; 4953c33c7948SRyan Roberts if (unlikely(!pte_same(ptep_get(vmf->pte), entry))) { 49547df67697SBibo Mao update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 49558f4e2101SHugh Dickins goto unlock; 49567df67697SBibo Mao } 4957c89357e2SDavid Hildenbrand if (vmf->flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) { 4958f6f37321SLinus Torvalds if (!pte_write(entry)) 49592994302bSJan Kara return do_wp_page(vmf); 4960c89357e2SDavid Hildenbrand else if (likely(vmf->flags & FAULT_FLAG_WRITE)) 49611da177e4SLinus Torvalds entry = pte_mkdirty(entry); 49621da177e4SLinus Torvalds } 49631da177e4SLinus Torvalds entry = pte_mkyoung(entry); 496482b0f8c3SJan Kara if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, 496582b0f8c3SJan Kara vmf->flags & FAULT_FLAG_WRITE)) { 496682b0f8c3SJan Kara update_mmu_cache(vmf->vma, vmf->address, vmf->pte); 49671a44e149SAndrea Arcangeli } else { 4968b7333b58SYang Shi /* Skip spurious TLB flush for retried page fault */ 4969b7333b58SYang Shi if (vmf->flags & FAULT_FLAG_TRIED) 4970b7333b58SYang Shi goto unlock; 49711a44e149SAndrea Arcangeli /* 49721a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 49731a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 49741a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 49751a44e149SAndrea Arcangeli * with threads. 49761a44e149SAndrea Arcangeli */ 497782b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) 497899c29133SGerald Schaefer flush_tlb_fix_spurious_fault(vmf->vma, vmf->address, 497999c29133SGerald Schaefer vmf->pte); 49801a44e149SAndrea Arcangeli } 49818f4e2101SHugh Dickins unlock: 498282b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 498383c54070SNick Piggin return 0; 49841da177e4SLinus Torvalds } 49851da177e4SLinus Torvalds 49861da177e4SLinus Torvalds /* 4987*4ec31152SMatthew Wilcox (Oracle) * On entry, we hold either the VMA lock or the mmap_lock 4988*4ec31152SMatthew Wilcox (Oracle) * (FAULT_FLAG_VMA_LOCK tells you which). If VM_FAULT_RETRY is set in 4989*4ec31152SMatthew Wilcox (Oracle) * the result, the mmap_lock is not held on exit. See filemap_fault() 4990*4ec31152SMatthew Wilcox (Oracle) * and __folio_lock_or_retry(). 49911da177e4SLinus Torvalds */ 49922b740303SSouptick Joarder static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma, 49932b740303SSouptick Joarder unsigned long address, unsigned int flags) 49941da177e4SLinus Torvalds { 499582b0f8c3SJan Kara struct vm_fault vmf = { 4996bae473a4SKirill A. Shutemov .vma = vma, 49971a29d85eSJan Kara .address = address & PAGE_MASK, 4998824ddc60SNadav Amit .real_address = address, 4999bae473a4SKirill A. Shutemov .flags = flags, 50000721ec8bSJan Kara .pgoff = linear_page_index(vma, address), 5001667240e0SJan Kara .gfp_mask = __get_fault_gfp_mask(vma), 5002bae473a4SKirill A. Shutemov }; 5003dcddffd4SKirill A. Shutemov struct mm_struct *mm = vma->vm_mm; 50047da4e2cbSYang Shi unsigned long vm_flags = vma->vm_flags; 50051da177e4SLinus Torvalds pgd_t *pgd; 5006c2febafcSKirill A. Shutemov p4d_t *p4d; 50072b740303SSouptick Joarder vm_fault_t ret; 50081da177e4SLinus Torvalds 5009*4ec31152SMatthew Wilcox (Oracle) if ((flags & FAULT_FLAG_VMA_LOCK) && !vma_is_anonymous(vma)) { 5010*4ec31152SMatthew Wilcox (Oracle) vma_end_read(vma); 5011*4ec31152SMatthew Wilcox (Oracle) return VM_FAULT_RETRY; 5012*4ec31152SMatthew Wilcox (Oracle) } 5013*4ec31152SMatthew Wilcox (Oracle) 50141da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 5015c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, address); 5016c2febafcSKirill A. Shutemov if (!p4d) 5017c2febafcSKirill A. Shutemov return VM_FAULT_OOM; 5018a00cc7d9SMatthew Wilcox 5019c2febafcSKirill A. Shutemov vmf.pud = pud_alloc(mm, p4d, address); 5020a00cc7d9SMatthew Wilcox if (!vmf.pud) 5021c74df32cSHugh Dickins return VM_FAULT_OOM; 5022625110b5SThomas Hellstrom retry_pud: 50237da4e2cbSYang Shi if (pud_none(*vmf.pud) && 5024a7f4e6e4SZach O'Keefe hugepage_vma_check(vma, vm_flags, false, true, true)) { 5025a00cc7d9SMatthew Wilcox ret = create_huge_pud(&vmf); 5026a00cc7d9SMatthew Wilcox if (!(ret & VM_FAULT_FALLBACK)) 5027a00cc7d9SMatthew Wilcox return ret; 5028a00cc7d9SMatthew Wilcox } else { 5029a00cc7d9SMatthew Wilcox pud_t orig_pud = *vmf.pud; 5030a00cc7d9SMatthew Wilcox 5031a00cc7d9SMatthew Wilcox barrier(); 5032a00cc7d9SMatthew Wilcox if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { 5033a00cc7d9SMatthew Wilcox 5034c89357e2SDavid Hildenbrand /* 5035c89357e2SDavid Hildenbrand * TODO once we support anonymous PUDs: NUMA case and 5036c89357e2SDavid Hildenbrand * FAULT_FLAG_UNSHARE handling. 5037c89357e2SDavid Hildenbrand */ 5038c89357e2SDavid Hildenbrand if ((flags & FAULT_FLAG_WRITE) && !pud_write(orig_pud)) { 5039a00cc7d9SMatthew Wilcox ret = wp_huge_pud(&vmf, orig_pud); 5040a00cc7d9SMatthew Wilcox if (!(ret & VM_FAULT_FALLBACK)) 5041a00cc7d9SMatthew Wilcox return ret; 5042a00cc7d9SMatthew Wilcox } else { 5043a00cc7d9SMatthew Wilcox huge_pud_set_accessed(&vmf, orig_pud); 5044a00cc7d9SMatthew Wilcox return 0; 5045a00cc7d9SMatthew Wilcox } 5046a00cc7d9SMatthew Wilcox } 5047a00cc7d9SMatthew Wilcox } 5048a00cc7d9SMatthew Wilcox 5049a00cc7d9SMatthew Wilcox vmf.pmd = pmd_alloc(mm, vmf.pud, address); 505082b0f8c3SJan Kara if (!vmf.pmd) 5051c74df32cSHugh Dickins return VM_FAULT_OOM; 5052625110b5SThomas Hellstrom 5053625110b5SThomas Hellstrom /* Huge pud page fault raced with pmd_alloc? */ 5054625110b5SThomas Hellstrom if (pud_trans_unstable(vmf.pud)) 5055625110b5SThomas Hellstrom goto retry_pud; 5056625110b5SThomas Hellstrom 50577da4e2cbSYang Shi if (pmd_none(*vmf.pmd) && 5058a7f4e6e4SZach O'Keefe hugepage_vma_check(vma, vm_flags, false, true, true)) { 5059a2d58167SDave Jiang ret = create_huge_pmd(&vmf); 5060c0292554SKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 5061c0292554SKirill A. Shutemov return ret; 506271e3aac0SAndrea Arcangeli } else { 506326e1a0c3SHugh Dickins vmf.orig_pmd = pmdp_get_lockless(vmf.pmd); 50641f1d06c3SDavid Rientjes 50655db4f15cSYang Shi if (unlikely(is_swap_pmd(vmf.orig_pmd))) { 506684c3fc4eSZi Yan VM_BUG_ON(thp_migration_supported() && 50675db4f15cSYang Shi !is_pmd_migration_entry(vmf.orig_pmd)); 50685db4f15cSYang Shi if (is_pmd_migration_entry(vmf.orig_pmd)) 506984c3fc4eSZi Yan pmd_migration_entry_wait(mm, vmf.pmd); 507084c3fc4eSZi Yan return 0; 507184c3fc4eSZi Yan } 50725db4f15cSYang Shi if (pmd_trans_huge(vmf.orig_pmd) || pmd_devmap(vmf.orig_pmd)) { 50735db4f15cSYang Shi if (pmd_protnone(vmf.orig_pmd) && vma_is_accessible(vma)) 50745db4f15cSYang Shi return do_huge_pmd_numa_page(&vmf); 5075d10e63f2SMel Gorman 5076c89357e2SDavid Hildenbrand if ((flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) && 5077c89357e2SDavid Hildenbrand !pmd_write(vmf.orig_pmd)) { 50785db4f15cSYang Shi ret = wp_huge_pmd(&vmf); 50799845cbbdSKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 50801f1d06c3SDavid Rientjes return ret; 5081a1dd450bSWill Deacon } else { 50825db4f15cSYang Shi huge_pmd_set_accessed(&vmf); 508371e3aac0SAndrea Arcangeli return 0; 508471e3aac0SAndrea Arcangeli } 508571e3aac0SAndrea Arcangeli } 50869845cbbdSKirill A. Shutemov } 508771e3aac0SAndrea Arcangeli 508882b0f8c3SJan Kara return handle_pte_fault(&vmf); 50891da177e4SLinus Torvalds } 50901da177e4SLinus Torvalds 5091bce617edSPeter Xu /** 5092f0953a1bSIngo Molnar * mm_account_fault - Do page fault accounting 5093809ef83cSYang Li * @mm: mm from which memcg should be extracted. It can be NULL. 5094bce617edSPeter Xu * @regs: the pt_regs struct pointer. When set to NULL, will skip accounting 5095bce617edSPeter Xu * of perf event counters, but we'll still do the per-task accounting to 5096bce617edSPeter Xu * the task who triggered this page fault. 5097bce617edSPeter Xu * @address: the faulted address. 5098bce617edSPeter Xu * @flags: the fault flags. 5099bce617edSPeter Xu * @ret: the fault retcode. 5100bce617edSPeter Xu * 5101f0953a1bSIngo Molnar * This will take care of most of the page fault accounting. Meanwhile, it 5102bce617edSPeter Xu * will also include the PERF_COUNT_SW_PAGE_FAULTS_[MAJ|MIN] perf counter 5103f0953a1bSIngo Molnar * updates. However, note that the handling of PERF_COUNT_SW_PAGE_FAULTS should 5104bce617edSPeter Xu * still be in per-arch page fault handlers at the entry of page fault. 5105bce617edSPeter Xu */ 510653156443SSuren Baghdasaryan static inline void mm_account_fault(struct mm_struct *mm, struct pt_regs *regs, 5107bce617edSPeter Xu unsigned long address, unsigned int flags, 5108bce617edSPeter Xu vm_fault_t ret) 5109bce617edSPeter Xu { 5110bce617edSPeter Xu bool major; 5111bce617edSPeter Xu 511253156443SSuren Baghdasaryan /* Incomplete faults will be accounted upon completion. */ 511353156443SSuren Baghdasaryan if (ret & VM_FAULT_RETRY) 511453156443SSuren Baghdasaryan return; 511553156443SSuren Baghdasaryan 5116bce617edSPeter Xu /* 511753156443SSuren Baghdasaryan * To preserve the behavior of older kernels, PGFAULT counters record 511853156443SSuren Baghdasaryan * both successful and failed faults, as opposed to perf counters, 511953156443SSuren Baghdasaryan * which ignore failed cases. 5120bce617edSPeter Xu */ 512153156443SSuren Baghdasaryan count_vm_event(PGFAULT); 512253156443SSuren Baghdasaryan count_memcg_event_mm(mm, PGFAULT); 512353156443SSuren Baghdasaryan 512453156443SSuren Baghdasaryan /* 512553156443SSuren Baghdasaryan * Do not account for unsuccessful faults (e.g. when the address wasn't 512653156443SSuren Baghdasaryan * valid). That includes arch_vma_access_permitted() failing before 512753156443SSuren Baghdasaryan * reaching here. So this is not a "this many hardware page faults" 512853156443SSuren Baghdasaryan * counter. We should use the hw profiling for that. 512953156443SSuren Baghdasaryan */ 513053156443SSuren Baghdasaryan if (ret & VM_FAULT_ERROR) 5131bce617edSPeter Xu return; 5132bce617edSPeter Xu 5133bce617edSPeter Xu /* 5134bce617edSPeter Xu * We define the fault as a major fault when the final successful fault 5135bce617edSPeter Xu * is VM_FAULT_MAJOR, or if it retried (which implies that we couldn't 5136bce617edSPeter Xu * handle it immediately previously). 5137bce617edSPeter Xu */ 5138bce617edSPeter Xu major = (ret & VM_FAULT_MAJOR) || (flags & FAULT_FLAG_TRIED); 5139bce617edSPeter Xu 5140a2beb5f1SPeter Xu if (major) 5141a2beb5f1SPeter Xu current->maj_flt++; 5142a2beb5f1SPeter Xu else 5143a2beb5f1SPeter Xu current->min_flt++; 5144a2beb5f1SPeter Xu 5145bce617edSPeter Xu /* 5146a2beb5f1SPeter Xu * If the fault is done for GUP, regs will be NULL. We only do the 5147a2beb5f1SPeter Xu * accounting for the per thread fault counters who triggered the 5148a2beb5f1SPeter Xu * fault, and we skip the perf event updates. 5149bce617edSPeter Xu */ 5150bce617edSPeter Xu if (!regs) 5151bce617edSPeter Xu return; 5152bce617edSPeter Xu 5153a2beb5f1SPeter Xu if (major) 5154bce617edSPeter Xu perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 5155a2beb5f1SPeter Xu else 5156bce617edSPeter Xu perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 5157bce617edSPeter Xu } 5158bce617edSPeter Xu 5159ec1c86b2SYu Zhao #ifdef CONFIG_LRU_GEN 5160ec1c86b2SYu Zhao static void lru_gen_enter_fault(struct vm_area_struct *vma) 5161ec1c86b2SYu Zhao { 51628788f678SYu Zhao /* the LRU algorithm only applies to accesses with recency */ 51638788f678SYu Zhao current->in_lru_fault = vma_has_recency(vma); 5164ec1c86b2SYu Zhao } 5165ec1c86b2SYu Zhao 5166ec1c86b2SYu Zhao static void lru_gen_exit_fault(void) 5167ec1c86b2SYu Zhao { 5168ec1c86b2SYu Zhao current->in_lru_fault = false; 5169ec1c86b2SYu Zhao } 5170ec1c86b2SYu Zhao #else 5171ec1c86b2SYu Zhao static void lru_gen_enter_fault(struct vm_area_struct *vma) 5172ec1c86b2SYu Zhao { 5173ec1c86b2SYu Zhao } 5174ec1c86b2SYu Zhao 5175ec1c86b2SYu Zhao static void lru_gen_exit_fault(void) 5176ec1c86b2SYu Zhao { 5177ec1c86b2SYu Zhao } 5178ec1c86b2SYu Zhao #endif /* CONFIG_LRU_GEN */ 5179ec1c86b2SYu Zhao 5180cdc5021cSDavid Hildenbrand static vm_fault_t sanitize_fault_flags(struct vm_area_struct *vma, 5181cdc5021cSDavid Hildenbrand unsigned int *flags) 5182cdc5021cSDavid Hildenbrand { 5183cdc5021cSDavid Hildenbrand if (unlikely(*flags & FAULT_FLAG_UNSHARE)) { 5184cdc5021cSDavid Hildenbrand if (WARN_ON_ONCE(*flags & FAULT_FLAG_WRITE)) 5185cdc5021cSDavid Hildenbrand return VM_FAULT_SIGSEGV; 5186cdc5021cSDavid Hildenbrand /* 5187cdc5021cSDavid Hildenbrand * FAULT_FLAG_UNSHARE only applies to COW mappings. Let's 5188cdc5021cSDavid Hildenbrand * just treat it like an ordinary read-fault otherwise. 5189cdc5021cSDavid Hildenbrand */ 5190cdc5021cSDavid Hildenbrand if (!is_cow_mapping(vma->vm_flags)) 5191cdc5021cSDavid Hildenbrand *flags &= ~FAULT_FLAG_UNSHARE; 519279881fedSDavid Hildenbrand } else if (*flags & FAULT_FLAG_WRITE) { 519379881fedSDavid Hildenbrand /* Write faults on read-only mappings are impossible ... */ 519479881fedSDavid Hildenbrand if (WARN_ON_ONCE(!(vma->vm_flags & VM_MAYWRITE))) 519579881fedSDavid Hildenbrand return VM_FAULT_SIGSEGV; 519679881fedSDavid Hildenbrand /* ... and FOLL_FORCE only applies to COW mappings. */ 519779881fedSDavid Hildenbrand if (WARN_ON_ONCE(!(vma->vm_flags & VM_WRITE) && 519879881fedSDavid Hildenbrand !is_cow_mapping(vma->vm_flags))) 519979881fedSDavid Hildenbrand return VM_FAULT_SIGSEGV; 5200cdc5021cSDavid Hildenbrand } 5201cdc5021cSDavid Hildenbrand return 0; 5202cdc5021cSDavid Hildenbrand } 5203cdc5021cSDavid Hildenbrand 52049a95f3cfSPaul Cassella /* 52059a95f3cfSPaul Cassella * By the time we get here, we already hold the mm semaphore 52069a95f3cfSPaul Cassella * 5207c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 52089138e47eSMatthew Wilcox (Oracle) * return value. See filemap_fault() and __folio_lock_or_retry(). 52099a95f3cfSPaul Cassella */ 52102b740303SSouptick Joarder vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address, 5211bce617edSPeter Xu unsigned int flags, struct pt_regs *regs) 5212519e5247SJohannes Weiner { 521353156443SSuren Baghdasaryan /* If the fault handler drops the mmap_lock, vma may be freed */ 521453156443SSuren Baghdasaryan struct mm_struct *mm = vma->vm_mm; 52152b740303SSouptick Joarder vm_fault_t ret; 5216519e5247SJohannes Weiner 5217519e5247SJohannes Weiner __set_current_state(TASK_RUNNING); 5218519e5247SJohannes Weiner 5219cdc5021cSDavid Hildenbrand ret = sanitize_fault_flags(vma, &flags); 5220cdc5021cSDavid Hildenbrand if (ret) 522153156443SSuren Baghdasaryan goto out; 5222cdc5021cSDavid Hildenbrand 5223de0c799bSLaurent Dufour if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, 5224de0c799bSLaurent Dufour flags & FAULT_FLAG_INSTRUCTION, 522553156443SSuren Baghdasaryan flags & FAULT_FLAG_REMOTE)) { 522653156443SSuren Baghdasaryan ret = VM_FAULT_SIGSEGV; 522753156443SSuren Baghdasaryan goto out; 522853156443SSuren Baghdasaryan } 5229de0c799bSLaurent Dufour 5230519e5247SJohannes Weiner /* 5231519e5247SJohannes Weiner * Enable the memcg OOM handling for faults triggered in user 5232519e5247SJohannes Weiner * space. Kernel faults are handled more gracefully. 5233519e5247SJohannes Weiner */ 5234519e5247SJohannes Weiner if (flags & FAULT_FLAG_USER) 523529ef680aSMichal Hocko mem_cgroup_enter_user_fault(); 5236519e5247SJohannes Weiner 5237ec1c86b2SYu Zhao lru_gen_enter_fault(vma); 5238ec1c86b2SYu Zhao 5239bae473a4SKirill A. Shutemov if (unlikely(is_vm_hugetlb_page(vma))) 5240bae473a4SKirill A. Shutemov ret = hugetlb_fault(vma->vm_mm, vma, address, flags); 5241bae473a4SKirill A. Shutemov else 5242dcddffd4SKirill A. Shutemov ret = __handle_mm_fault(vma, address, flags); 5243519e5247SJohannes Weiner 5244ec1c86b2SYu Zhao lru_gen_exit_fault(); 5245ec1c86b2SYu Zhao 524649426420SJohannes Weiner if (flags & FAULT_FLAG_USER) { 524729ef680aSMichal Hocko mem_cgroup_exit_user_fault(); 524849426420SJohannes Weiner /* 524949426420SJohannes Weiner * The task may have entered a memcg OOM situation but 525049426420SJohannes Weiner * if the allocation error was handled gracefully (no 525149426420SJohannes Weiner * VM_FAULT_OOM), there is no need to kill anything. 525249426420SJohannes Weiner * Just clean up the OOM state peacefully. 525349426420SJohannes Weiner */ 525449426420SJohannes Weiner if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) 525549426420SJohannes Weiner mem_cgroup_oom_synchronize(false); 525649426420SJohannes Weiner } 525753156443SSuren Baghdasaryan out: 525853156443SSuren Baghdasaryan mm_account_fault(mm, regs, address, flags, ret); 5259bce617edSPeter Xu 5260519e5247SJohannes Weiner return ret; 5261519e5247SJohannes Weiner } 5262e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(handle_mm_fault); 5263519e5247SJohannes Weiner 5264c2508ec5SLinus Torvalds #ifdef CONFIG_LOCK_MM_AND_FIND_VMA 5265c2508ec5SLinus Torvalds #include <linux/extable.h> 5266c2508ec5SLinus Torvalds 5267c2508ec5SLinus Torvalds static inline bool get_mmap_lock_carefully(struct mm_struct *mm, struct pt_regs *regs) 5268c2508ec5SLinus Torvalds { 5269c2508ec5SLinus Torvalds /* Even if this succeeds, make it clear we *might* have slept */ 5270c2508ec5SLinus Torvalds if (likely(mmap_read_trylock(mm))) { 5271c2508ec5SLinus Torvalds might_sleep(); 5272c2508ec5SLinus Torvalds return true; 5273c2508ec5SLinus Torvalds } 5274c2508ec5SLinus Torvalds 5275c2508ec5SLinus Torvalds if (regs && !user_mode(regs)) { 5276c2508ec5SLinus Torvalds unsigned long ip = instruction_pointer(regs); 5277c2508ec5SLinus Torvalds if (!search_exception_tables(ip)) 5278c2508ec5SLinus Torvalds return false; 5279c2508ec5SLinus Torvalds } 5280c2508ec5SLinus Torvalds 5281eda00472SLinus Torvalds return !mmap_read_lock_killable(mm); 5282c2508ec5SLinus Torvalds } 5283c2508ec5SLinus Torvalds 5284c2508ec5SLinus Torvalds static inline bool mmap_upgrade_trylock(struct mm_struct *mm) 5285c2508ec5SLinus Torvalds { 5286c2508ec5SLinus Torvalds /* 5287c2508ec5SLinus Torvalds * We don't have this operation yet. 5288c2508ec5SLinus Torvalds * 5289c2508ec5SLinus Torvalds * It should be easy enough to do: it's basically a 5290c2508ec5SLinus Torvalds * atomic_long_try_cmpxchg_acquire() 5291c2508ec5SLinus Torvalds * from RWSEM_READER_BIAS -> RWSEM_WRITER_LOCKED, but 5292c2508ec5SLinus Torvalds * it also needs the proper lockdep magic etc. 5293c2508ec5SLinus Torvalds */ 5294c2508ec5SLinus Torvalds return false; 5295c2508ec5SLinus Torvalds } 5296c2508ec5SLinus Torvalds 5297c2508ec5SLinus Torvalds static inline bool upgrade_mmap_lock_carefully(struct mm_struct *mm, struct pt_regs *regs) 5298c2508ec5SLinus Torvalds { 5299c2508ec5SLinus Torvalds mmap_read_unlock(mm); 5300c2508ec5SLinus Torvalds if (regs && !user_mode(regs)) { 5301c2508ec5SLinus Torvalds unsigned long ip = instruction_pointer(regs); 5302c2508ec5SLinus Torvalds if (!search_exception_tables(ip)) 5303c2508ec5SLinus Torvalds return false; 5304c2508ec5SLinus Torvalds } 5305eda00472SLinus Torvalds return !mmap_write_lock_killable(mm); 5306c2508ec5SLinus Torvalds } 5307c2508ec5SLinus Torvalds 5308c2508ec5SLinus Torvalds /* 5309c2508ec5SLinus Torvalds * Helper for page fault handling. 5310c2508ec5SLinus Torvalds * 5311c2508ec5SLinus Torvalds * This is kind of equivalend to "mmap_read_lock()" followed 5312c2508ec5SLinus Torvalds * by "find_extend_vma()", except it's a lot more careful about 5313c2508ec5SLinus Torvalds * the locking (and will drop the lock on failure). 5314c2508ec5SLinus Torvalds * 5315c2508ec5SLinus Torvalds * For example, if we have a kernel bug that causes a page 5316c2508ec5SLinus Torvalds * fault, we don't want to just use mmap_read_lock() to get 5317c2508ec5SLinus Torvalds * the mm lock, because that would deadlock if the bug were 5318c2508ec5SLinus Torvalds * to happen while we're holding the mm lock for writing. 5319c2508ec5SLinus Torvalds * 5320c2508ec5SLinus Torvalds * So this checks the exception tables on kernel faults in 5321c2508ec5SLinus Torvalds * order to only do this all for instructions that are actually 5322c2508ec5SLinus Torvalds * expected to fault. 5323c2508ec5SLinus Torvalds * 5324c2508ec5SLinus Torvalds * We can also actually take the mm lock for writing if we 5325c2508ec5SLinus Torvalds * need to extend the vma, which helps the VM layer a lot. 5326c2508ec5SLinus Torvalds */ 5327c2508ec5SLinus Torvalds struct vm_area_struct *lock_mm_and_find_vma(struct mm_struct *mm, 5328c2508ec5SLinus Torvalds unsigned long addr, struct pt_regs *regs) 5329c2508ec5SLinus Torvalds { 5330c2508ec5SLinus Torvalds struct vm_area_struct *vma; 5331c2508ec5SLinus Torvalds 5332c2508ec5SLinus Torvalds if (!get_mmap_lock_carefully(mm, regs)) 5333c2508ec5SLinus Torvalds return NULL; 5334c2508ec5SLinus Torvalds 5335c2508ec5SLinus Torvalds vma = find_vma(mm, addr); 5336c2508ec5SLinus Torvalds if (likely(vma && (vma->vm_start <= addr))) 5337c2508ec5SLinus Torvalds return vma; 5338c2508ec5SLinus Torvalds 5339c2508ec5SLinus Torvalds /* 5340c2508ec5SLinus Torvalds * Well, dang. We might still be successful, but only 5341c2508ec5SLinus Torvalds * if we can extend a vma to do so. 5342c2508ec5SLinus Torvalds */ 5343c2508ec5SLinus Torvalds if (!vma || !(vma->vm_flags & VM_GROWSDOWN)) { 5344c2508ec5SLinus Torvalds mmap_read_unlock(mm); 5345c2508ec5SLinus Torvalds return NULL; 5346c2508ec5SLinus Torvalds } 5347c2508ec5SLinus Torvalds 5348c2508ec5SLinus Torvalds /* 5349c2508ec5SLinus Torvalds * We can try to upgrade the mmap lock atomically, 5350c2508ec5SLinus Torvalds * in which case we can continue to use the vma 5351c2508ec5SLinus Torvalds * we already looked up. 5352c2508ec5SLinus Torvalds * 5353c2508ec5SLinus Torvalds * Otherwise we'll have to drop the mmap lock and 5354c2508ec5SLinus Torvalds * re-take it, and also look up the vma again, 5355c2508ec5SLinus Torvalds * re-checking it. 5356c2508ec5SLinus Torvalds */ 5357c2508ec5SLinus Torvalds if (!mmap_upgrade_trylock(mm)) { 5358c2508ec5SLinus Torvalds if (!upgrade_mmap_lock_carefully(mm, regs)) 5359c2508ec5SLinus Torvalds return NULL; 5360c2508ec5SLinus Torvalds 5361c2508ec5SLinus Torvalds vma = find_vma(mm, addr); 5362c2508ec5SLinus Torvalds if (!vma) 5363c2508ec5SLinus Torvalds goto fail; 5364c2508ec5SLinus Torvalds if (vma->vm_start <= addr) 5365c2508ec5SLinus Torvalds goto success; 5366c2508ec5SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 5367c2508ec5SLinus Torvalds goto fail; 5368c2508ec5SLinus Torvalds } 5369c2508ec5SLinus Torvalds 53708d7071afSLinus Torvalds if (expand_stack_locked(vma, addr)) 5371c2508ec5SLinus Torvalds goto fail; 5372c2508ec5SLinus Torvalds 5373c2508ec5SLinus Torvalds success: 5374c2508ec5SLinus Torvalds mmap_write_downgrade(mm); 5375c2508ec5SLinus Torvalds return vma; 5376c2508ec5SLinus Torvalds 5377c2508ec5SLinus Torvalds fail: 5378c2508ec5SLinus Torvalds mmap_write_unlock(mm); 5379c2508ec5SLinus Torvalds return NULL; 5380c2508ec5SLinus Torvalds } 5381c2508ec5SLinus Torvalds #endif 5382c2508ec5SLinus Torvalds 538350ee3253SSuren Baghdasaryan #ifdef CONFIG_PER_VMA_LOCK 538450ee3253SSuren Baghdasaryan /* 538550ee3253SSuren Baghdasaryan * Lookup and lock a VMA under RCU protection. Returned VMA is guaranteed to be 538650ee3253SSuren Baghdasaryan * stable and not isolated. If the VMA is not found or is being modified the 538750ee3253SSuren Baghdasaryan * function returns NULL. 538850ee3253SSuren Baghdasaryan */ 538950ee3253SSuren Baghdasaryan struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm, 539050ee3253SSuren Baghdasaryan unsigned long address) 539150ee3253SSuren Baghdasaryan { 539250ee3253SSuren Baghdasaryan MA_STATE(mas, &mm->mm_mt, address, address); 539350ee3253SSuren Baghdasaryan struct vm_area_struct *vma; 539450ee3253SSuren Baghdasaryan 539550ee3253SSuren Baghdasaryan rcu_read_lock(); 539650ee3253SSuren Baghdasaryan retry: 539750ee3253SSuren Baghdasaryan vma = mas_walk(&mas); 539850ee3253SSuren Baghdasaryan if (!vma) 539950ee3253SSuren Baghdasaryan goto inval; 540050ee3253SSuren Baghdasaryan 540150ee3253SSuren Baghdasaryan if (!vma_start_read(vma)) 540250ee3253SSuren Baghdasaryan goto inval; 540350ee3253SSuren Baghdasaryan 5404444eeb17SSuren Baghdasaryan /* 5405657b5146SJann Horn * find_mergeable_anon_vma uses adjacent vmas which are not locked. 5406657b5146SJann Horn * This check must happen after vma_start_read(); otherwise, a 5407657b5146SJann Horn * concurrent mremap() with MREMAP_DONTUNMAP could dissociate the VMA 5408657b5146SJann Horn * from its anon_vma. 5409657b5146SJann Horn */ 5410657b5146SJann Horn if (unlikely(!vma->anon_vma && !vma_is_tcp(vma))) 5411657b5146SJann Horn goto inval_end_read; 5412657b5146SJann Horn 5413657b5146SJann Horn /* 5414444eeb17SSuren Baghdasaryan * Due to the possibility of userfault handler dropping mmap_lock, avoid 5415444eeb17SSuren Baghdasaryan * it for now and fall back to page fault handling under mmap_lock. 5416444eeb17SSuren Baghdasaryan */ 5417657b5146SJann Horn if (userfaultfd_armed(vma)) 5418657b5146SJann Horn goto inval_end_read; 5419444eeb17SSuren Baghdasaryan 542050ee3253SSuren Baghdasaryan /* Check since vm_start/vm_end might change before we lock the VMA */ 5421657b5146SJann Horn if (unlikely(address < vma->vm_start || address >= vma->vm_end)) 5422657b5146SJann Horn goto inval_end_read; 542350ee3253SSuren Baghdasaryan 542450ee3253SSuren Baghdasaryan /* Check if the VMA got isolated after we found it */ 542550ee3253SSuren Baghdasaryan if (vma->detached) { 542650ee3253SSuren Baghdasaryan vma_end_read(vma); 542752f23865SSuren Baghdasaryan count_vm_vma_lock_event(VMA_LOCK_MISS); 542850ee3253SSuren Baghdasaryan /* The area was replaced with another one */ 542950ee3253SSuren Baghdasaryan goto retry; 543050ee3253SSuren Baghdasaryan } 543150ee3253SSuren Baghdasaryan 543250ee3253SSuren Baghdasaryan rcu_read_unlock(); 543350ee3253SSuren Baghdasaryan return vma; 5434657b5146SJann Horn 5435657b5146SJann Horn inval_end_read: 5436657b5146SJann Horn vma_end_read(vma); 543750ee3253SSuren Baghdasaryan inval: 543850ee3253SSuren Baghdasaryan rcu_read_unlock(); 543952f23865SSuren Baghdasaryan count_vm_vma_lock_event(VMA_LOCK_ABORT); 544050ee3253SSuren Baghdasaryan return NULL; 544150ee3253SSuren Baghdasaryan } 544250ee3253SSuren Baghdasaryan #endif /* CONFIG_PER_VMA_LOCK */ 544350ee3253SSuren Baghdasaryan 544490eceff1SKirill A. Shutemov #ifndef __PAGETABLE_P4D_FOLDED 544590eceff1SKirill A. Shutemov /* 544690eceff1SKirill A. Shutemov * Allocate p4d page table. 544790eceff1SKirill A. Shutemov * We've already handled the fast-path in-line. 544890eceff1SKirill A. Shutemov */ 544990eceff1SKirill A. Shutemov int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 545090eceff1SKirill A. Shutemov { 545190eceff1SKirill A. Shutemov p4d_t *new = p4d_alloc_one(mm, address); 545290eceff1SKirill A. Shutemov if (!new) 545390eceff1SKirill A. Shutemov return -ENOMEM; 545490eceff1SKirill A. Shutemov 545590eceff1SKirill A. Shutemov spin_lock(&mm->page_table_lock); 5456ed33b5a6SQi Zheng if (pgd_present(*pgd)) { /* Another has populated it */ 545790eceff1SKirill A. Shutemov p4d_free(mm, new); 5458ed33b5a6SQi Zheng } else { 5459ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 546090eceff1SKirill A. Shutemov pgd_populate(mm, pgd, new); 5461ed33b5a6SQi Zheng } 546290eceff1SKirill A. Shutemov spin_unlock(&mm->page_table_lock); 546390eceff1SKirill A. Shutemov return 0; 546490eceff1SKirill A. Shutemov } 546590eceff1SKirill A. Shutemov #endif /* __PAGETABLE_P4D_FOLDED */ 546690eceff1SKirill A. Shutemov 54671da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 54681da177e4SLinus Torvalds /* 54691da177e4SLinus Torvalds * Allocate page upper directory. 5470872fec16SHugh Dickins * We've already handled the fast-path in-line. 54711da177e4SLinus Torvalds */ 5472c2febafcSKirill A. Shutemov int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) 54731da177e4SLinus Torvalds { 5474c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 5475c74df32cSHugh Dickins if (!new) 54761bb3630eSHugh Dickins return -ENOMEM; 54771da177e4SLinus Torvalds 5478872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 5479b4e98d9aSKirill A. Shutemov if (!p4d_present(*p4d)) { 5480b4e98d9aSKirill A. Shutemov mm_inc_nr_puds(mm); 5481ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 5482c2febafcSKirill A. Shutemov p4d_populate(mm, p4d, new); 5483b4e98d9aSKirill A. Shutemov } else /* Another has populated it */ 5484c2febafcSKirill A. Shutemov pud_free(mm, new); 5485872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 54861bb3630eSHugh Dickins return 0; 54871da177e4SLinus Torvalds } 54881da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 54891da177e4SLinus Torvalds 54901da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 54911da177e4SLinus Torvalds /* 54921da177e4SLinus Torvalds * Allocate page middle directory. 5493872fec16SHugh Dickins * We've already handled the fast-path in-line. 54941da177e4SLinus Torvalds */ 54951bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 54961da177e4SLinus Torvalds { 5497a00cc7d9SMatthew Wilcox spinlock_t *ptl; 5498c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 5499c74df32cSHugh Dickins if (!new) 55001bb3630eSHugh Dickins return -ENOMEM; 55011da177e4SLinus Torvalds 5502a00cc7d9SMatthew Wilcox ptl = pud_lock(mm, pud); 5503dc6c9a35SKirill A. Shutemov if (!pud_present(*pud)) { 5504dc6c9a35SKirill A. Shutemov mm_inc_nr_pmds(mm); 5505ed33b5a6SQi Zheng smp_wmb(); /* See comment in pmd_install() */ 55061da177e4SLinus Torvalds pud_populate(mm, pud, new); 5507ed33b5a6SQi Zheng } else { /* Another has populated it */ 55085e541973SBenjamin Herrenschmidt pmd_free(mm, new); 5509ed33b5a6SQi Zheng } 5510a00cc7d9SMatthew Wilcox spin_unlock(ptl); 55111bb3630eSHugh Dickins return 0; 55121da177e4SLinus Torvalds } 55131da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 55141da177e4SLinus Torvalds 55153b6748e2SJohannes Weiner /** 55169fd6dad1SPaolo Bonzini * follow_pte - look up PTE at a user virtual address 55179fd6dad1SPaolo Bonzini * @mm: the mm_struct of the target address space 55189fd6dad1SPaolo Bonzini * @address: user virtual address 55199fd6dad1SPaolo Bonzini * @ptepp: location to store found PTE 55209fd6dad1SPaolo Bonzini * @ptlp: location to store the lock for the PTE 55219fd6dad1SPaolo Bonzini * 55229fd6dad1SPaolo Bonzini * On a successful return, the pointer to the PTE is stored in @ptepp; 55239fd6dad1SPaolo Bonzini * the corresponding lock is taken and its location is stored in @ptlp. 55249fd6dad1SPaolo Bonzini * The contents of the PTE are only stable until @ptlp is released; 55259fd6dad1SPaolo Bonzini * any further use, if any, must be protected against invalidation 55269fd6dad1SPaolo Bonzini * with MMU notifiers. 55279fd6dad1SPaolo Bonzini * 55289fd6dad1SPaolo Bonzini * Only IO mappings and raw PFN mappings are allowed. The mmap semaphore 55299fd6dad1SPaolo Bonzini * should be taken for read. 55309fd6dad1SPaolo Bonzini * 55319fd6dad1SPaolo Bonzini * KVM uses this function. While it is arguably less bad than ``follow_pfn``, 55329fd6dad1SPaolo Bonzini * it is not a good general-purpose API. 55339fd6dad1SPaolo Bonzini * 55349fd6dad1SPaolo Bonzini * Return: zero on success, -ve otherwise. 55359fd6dad1SPaolo Bonzini */ 55369fd6dad1SPaolo Bonzini int follow_pte(struct mm_struct *mm, unsigned long address, 55379fd6dad1SPaolo Bonzini pte_t **ptepp, spinlock_t **ptlp) 55389fd6dad1SPaolo Bonzini { 55390e5e64c0SMuchun Song pgd_t *pgd; 55400e5e64c0SMuchun Song p4d_t *p4d; 55410e5e64c0SMuchun Song pud_t *pud; 55420e5e64c0SMuchun Song pmd_t *pmd; 55430e5e64c0SMuchun Song pte_t *ptep; 55440e5e64c0SMuchun Song 55450e5e64c0SMuchun Song pgd = pgd_offset(mm, address); 55460e5e64c0SMuchun Song if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 55470e5e64c0SMuchun Song goto out; 55480e5e64c0SMuchun Song 55490e5e64c0SMuchun Song p4d = p4d_offset(pgd, address); 55500e5e64c0SMuchun Song if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) 55510e5e64c0SMuchun Song goto out; 55520e5e64c0SMuchun Song 55530e5e64c0SMuchun Song pud = pud_offset(p4d, address); 55540e5e64c0SMuchun Song if (pud_none(*pud) || unlikely(pud_bad(*pud))) 55550e5e64c0SMuchun Song goto out; 55560e5e64c0SMuchun Song 55570e5e64c0SMuchun Song pmd = pmd_offset(pud, address); 55580e5e64c0SMuchun Song VM_BUG_ON(pmd_trans_huge(*pmd)); 55590e5e64c0SMuchun Song 55600e5e64c0SMuchun Song ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 55613db82b93SHugh Dickins if (!ptep) 55623db82b93SHugh Dickins goto out; 5563c33c7948SRyan Roberts if (!pte_present(ptep_get(ptep))) 55640e5e64c0SMuchun Song goto unlock; 55650e5e64c0SMuchun Song *ptepp = ptep; 55660e5e64c0SMuchun Song return 0; 55670e5e64c0SMuchun Song unlock: 55680e5e64c0SMuchun Song pte_unmap_unlock(ptep, *ptlp); 55690e5e64c0SMuchun Song out: 55700e5e64c0SMuchun Song return -EINVAL; 55719fd6dad1SPaolo Bonzini } 55729fd6dad1SPaolo Bonzini EXPORT_SYMBOL_GPL(follow_pte); 55739fd6dad1SPaolo Bonzini 55749fd6dad1SPaolo Bonzini /** 55753b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 55763b6748e2SJohannes Weiner * @vma: memory mapping 55773b6748e2SJohannes Weiner * @address: user virtual address 55783b6748e2SJohannes Weiner * @pfn: location to store found PFN 55793b6748e2SJohannes Weiner * 55803b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 55813b6748e2SJohannes Weiner * 55829fd6dad1SPaolo Bonzini * This function does not allow the caller to read the permissions 55839fd6dad1SPaolo Bonzini * of the PTE. Do not use it. 55849fd6dad1SPaolo Bonzini * 5585a862f68aSMike Rapoport * Return: zero and the pfn at @pfn on success, -ve otherwise. 55863b6748e2SJohannes Weiner */ 55873b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 55883b6748e2SJohannes Weiner unsigned long *pfn) 55893b6748e2SJohannes Weiner { 55903b6748e2SJohannes Weiner int ret = -EINVAL; 55913b6748e2SJohannes Weiner spinlock_t *ptl; 55923b6748e2SJohannes Weiner pte_t *ptep; 55933b6748e2SJohannes Weiner 55943b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 55953b6748e2SJohannes Weiner return ret; 55963b6748e2SJohannes Weiner 55979fd6dad1SPaolo Bonzini ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 55983b6748e2SJohannes Weiner if (ret) 55993b6748e2SJohannes Weiner return ret; 5600c33c7948SRyan Roberts *pfn = pte_pfn(ptep_get(ptep)); 56013b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 56023b6748e2SJohannes Weiner return 0; 56033b6748e2SJohannes Weiner } 56043b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 56053b6748e2SJohannes Weiner 560628b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 5607d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 560828b2ee20SRik van Riel unsigned long address, unsigned int flags, 5609d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 561028b2ee20SRik van Riel { 561103668a4dSJohannes Weiner int ret = -EINVAL; 561228b2ee20SRik van Riel pte_t *ptep, pte; 561328b2ee20SRik van Riel spinlock_t *ptl; 561428b2ee20SRik van Riel 5615d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 5616d87fe660Svenkatesh.pallipadi@intel.com goto out; 561728b2ee20SRik van Riel 56189fd6dad1SPaolo Bonzini if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 5619d87fe660Svenkatesh.pallipadi@intel.com goto out; 5620c33c7948SRyan Roberts pte = ptep_get(ptep); 562103668a4dSJohannes Weiner 5622f6f37321SLinus Torvalds if ((flags & FOLL_WRITE) && !pte_write(pte)) 562328b2ee20SRik van Riel goto unlock; 562428b2ee20SRik van Riel 562528b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 562603668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 562728b2ee20SRik van Riel 562803668a4dSJohannes Weiner ret = 0; 562928b2ee20SRik van Riel unlock: 563028b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 563128b2ee20SRik van Riel out: 5632d87fe660Svenkatesh.pallipadi@intel.com return ret; 563328b2ee20SRik van Riel } 563428b2ee20SRik van Riel 563596667f8aSDaniel Vetter /** 563696667f8aSDaniel Vetter * generic_access_phys - generic implementation for iomem mmap access 563796667f8aSDaniel Vetter * @vma: the vma to access 5638f0953a1bSIngo Molnar * @addr: userspace address, not relative offset within @vma 563996667f8aSDaniel Vetter * @buf: buffer to read/write 564096667f8aSDaniel Vetter * @len: length of transfer 564196667f8aSDaniel Vetter * @write: set to FOLL_WRITE when writing, otherwise reading 564296667f8aSDaniel Vetter * 564396667f8aSDaniel Vetter * This is a generic implementation for &vm_operations_struct.access for an 564496667f8aSDaniel Vetter * iomem mapping. This callback is used by access_process_vm() when the @vma is 564596667f8aSDaniel Vetter * not page based. 564696667f8aSDaniel Vetter */ 564728b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 564828b2ee20SRik van Riel void *buf, int len, int write) 564928b2ee20SRik van Riel { 565028b2ee20SRik van Riel resource_size_t phys_addr; 565128b2ee20SRik van Riel unsigned long prot = 0; 56522bc7273bSKOSAKI Motohiro void __iomem *maddr; 565396667f8aSDaniel Vetter pte_t *ptep, pte; 565496667f8aSDaniel Vetter spinlock_t *ptl; 565596667f8aSDaniel Vetter int offset = offset_in_page(addr); 565696667f8aSDaniel Vetter int ret = -EINVAL; 565728b2ee20SRik van Riel 565896667f8aSDaniel Vetter if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 565996667f8aSDaniel Vetter return -EINVAL; 566096667f8aSDaniel Vetter 566196667f8aSDaniel Vetter retry: 5662e913a8cdSLinus Torvalds if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) 566396667f8aSDaniel Vetter return -EINVAL; 5664c33c7948SRyan Roberts pte = ptep_get(ptep); 566596667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 566696667f8aSDaniel Vetter 566796667f8aSDaniel Vetter prot = pgprot_val(pte_pgprot(pte)); 566896667f8aSDaniel Vetter phys_addr = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 566996667f8aSDaniel Vetter 567096667f8aSDaniel Vetter if ((write & FOLL_WRITE) && !pte_write(pte)) 567128b2ee20SRik van Riel return -EINVAL; 567228b2ee20SRik van Riel 56739cb12d7bSGrazvydas Ignotas maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); 567424eee1e4Sjie@chenjie6@huwei.com if (!maddr) 567524eee1e4Sjie@chenjie6@huwei.com return -ENOMEM; 567624eee1e4Sjie@chenjie6@huwei.com 5677e913a8cdSLinus Torvalds if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) 567896667f8aSDaniel Vetter goto out_unmap; 567996667f8aSDaniel Vetter 5680c33c7948SRyan Roberts if (!pte_same(pte, ptep_get(ptep))) { 568196667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 568296667f8aSDaniel Vetter iounmap(maddr); 568396667f8aSDaniel Vetter 568496667f8aSDaniel Vetter goto retry; 568596667f8aSDaniel Vetter } 568696667f8aSDaniel Vetter 568728b2ee20SRik van Riel if (write) 568828b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 568928b2ee20SRik van Riel else 569028b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 569196667f8aSDaniel Vetter ret = len; 569296667f8aSDaniel Vetter pte_unmap_unlock(ptep, ptl); 569396667f8aSDaniel Vetter out_unmap: 569428b2ee20SRik van Riel iounmap(maddr); 569528b2ee20SRik van Riel 569696667f8aSDaniel Vetter return ret; 569728b2ee20SRik van Riel } 56985a73633eSUwe Kleine-König EXPORT_SYMBOL_GPL(generic_access_phys); 569928b2ee20SRik van Riel #endif 570028b2ee20SRik van Riel 57010ec76a11SDavid Howells /* 5702d3f5ffcaSJohn Hubbard * Access another process' address space as given in mm. 57030ec76a11SDavid Howells */ 5704d3f5ffcaSJohn Hubbard int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf, 5705d3f5ffcaSJohn Hubbard int len, unsigned int gup_flags) 57060ec76a11SDavid Howells { 57070ec76a11SDavid Howells void *old_buf = buf; 5708442486ecSLorenzo Stoakes int write = gup_flags & FOLL_WRITE; 57090ec76a11SDavid Howells 5710d8ed45c5SMichel Lespinasse if (mmap_read_lock_killable(mm)) 57111e426fe2SKonstantin Khlebnikov return 0; 57121e426fe2SKonstantin Khlebnikov 5713eee9c708SLinus Torvalds /* Avoid triggering the temporary warning in __get_user_pages */ 5714eee9c708SLinus Torvalds if (!vma_lookup(mm, addr) && !expand_stack(mm, addr)) 5715eee9c708SLinus Torvalds return 0; 5716eee9c708SLinus Torvalds 5717183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 57180ec76a11SDavid Howells while (len) { 5719ca5e8632SLorenzo Stoakes int bytes, offset; 57200ec76a11SDavid Howells void *maddr; 5721ca5e8632SLorenzo Stoakes struct vm_area_struct *vma = NULL; 5722ca5e8632SLorenzo Stoakes struct page *page = get_user_page_vma_remote(mm, addr, 5723ca5e8632SLorenzo Stoakes gup_flags, &vma); 57240ec76a11SDavid Howells 5725ca5e8632SLorenzo Stoakes if (IS_ERR_OR_NULL(page)) { 57268d7071afSLinus Torvalds /* We might need to expand the stack to access it */ 57278d7071afSLinus Torvalds vma = vma_lookup(mm, addr); 57288d7071afSLinus Torvalds if (!vma) { 57298d7071afSLinus Torvalds vma = expand_stack(mm, addr); 57309471f1f2SLinus Torvalds 57319471f1f2SLinus Torvalds /* mmap_lock was dropped on failure */ 57328d7071afSLinus Torvalds if (!vma) 57339471f1f2SLinus Torvalds return buf - old_buf; 57349471f1f2SLinus Torvalds 57359471f1f2SLinus Torvalds /* Try again if stack expansion worked */ 57369471f1f2SLinus Torvalds continue; 57378d7071afSLinus Torvalds } 57388d7071afSLinus Torvalds 5739ca5e8632SLorenzo Stoakes 574028b2ee20SRik van Riel /* 574128b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 574228b2ee20SRik van Riel * we can access using slightly different code. 574328b2ee20SRik van Riel */ 57449471f1f2SLinus Torvalds bytes = 0; 57459471f1f2SLinus Torvalds #ifdef CONFIG_HAVE_IOREMAP_PROT 574628b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 57479471f1f2SLinus Torvalds bytes = vma->vm_ops->access(vma, addr, buf, 574828b2ee20SRik van Riel len, write); 5749dbffcd03SRik van Riel #endif 57509471f1f2SLinus Torvalds if (bytes <= 0) 57519471f1f2SLinus Torvalds break; 575228b2ee20SRik van Riel } else { 57530ec76a11SDavid Howells bytes = len; 57540ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 57550ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 57560ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 57570ec76a11SDavid Howells 57580ec76a11SDavid Howells maddr = kmap(page); 57590ec76a11SDavid Howells if (write) { 57600ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 57610ec76a11SDavid Howells maddr + offset, buf, bytes); 57620ec76a11SDavid Howells set_page_dirty_lock(page); 57630ec76a11SDavid Howells } else { 57640ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 57650ec76a11SDavid Howells buf, maddr + offset, bytes); 57660ec76a11SDavid Howells } 57670ec76a11SDavid Howells kunmap(page); 576809cbfeafSKirill A. Shutemov put_page(page); 576928b2ee20SRik van Riel } 57700ec76a11SDavid Howells len -= bytes; 57710ec76a11SDavid Howells buf += bytes; 57720ec76a11SDavid Howells addr += bytes; 57730ec76a11SDavid Howells } 5774d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 57750ec76a11SDavid Howells 57760ec76a11SDavid Howells return buf - old_buf; 57770ec76a11SDavid Howells } 577803252919SAndi Kleen 57795ddd36b9SStephen Wilson /** 5780ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 57815ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 57825ddd36b9SStephen Wilson * @addr: start address to access 57835ddd36b9SStephen Wilson * @buf: source or destination buffer 57845ddd36b9SStephen Wilson * @len: number of bytes to transfer 57856347e8d5SLorenzo Stoakes * @gup_flags: flags modifying lookup behaviour 57865ddd36b9SStephen Wilson * 57875ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 5788a862f68aSMike Rapoport * 5789a862f68aSMike Rapoport * Return: number of bytes copied from source to destination. 57905ddd36b9SStephen Wilson */ 57915ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 57926347e8d5SLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 57935ddd36b9SStephen Wilson { 5794d3f5ffcaSJohn Hubbard return __access_remote_vm(mm, addr, buf, len, gup_flags); 57955ddd36b9SStephen Wilson } 57965ddd36b9SStephen Wilson 579703252919SAndi Kleen /* 5798206cb636SStephen Wilson * Access another process' address space. 5799206cb636SStephen Wilson * Source/target buffer must be kernel space, 5800206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 5801206cb636SStephen Wilson */ 5802206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 5803f307ab6dSLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 5804206cb636SStephen Wilson { 5805206cb636SStephen Wilson struct mm_struct *mm; 5806206cb636SStephen Wilson int ret; 5807206cb636SStephen Wilson 5808206cb636SStephen Wilson mm = get_task_mm(tsk); 5809206cb636SStephen Wilson if (!mm) 5810206cb636SStephen Wilson return 0; 5811206cb636SStephen Wilson 5812d3f5ffcaSJohn Hubbard ret = __access_remote_vm(mm, addr, buf, len, gup_flags); 5813442486ecSLorenzo Stoakes 5814206cb636SStephen Wilson mmput(mm); 5815206cb636SStephen Wilson 5816206cb636SStephen Wilson return ret; 5817206cb636SStephen Wilson } 5818fcd35857SCatalin Marinas EXPORT_SYMBOL_GPL(access_process_vm); 5819206cb636SStephen Wilson 582003252919SAndi Kleen /* 582103252919SAndi Kleen * Print the name of a VMA. 582203252919SAndi Kleen */ 582303252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 582403252919SAndi Kleen { 582503252919SAndi Kleen struct mm_struct *mm = current->mm; 582603252919SAndi Kleen struct vm_area_struct *vma; 582703252919SAndi Kleen 5828e8bff74aSIngo Molnar /* 58290a7f682dSMichal Hocko * we might be running from an atomic context so we cannot sleep 5830e8bff74aSIngo Molnar */ 5831d8ed45c5SMichel Lespinasse if (!mmap_read_trylock(mm)) 5832e8bff74aSIngo Molnar return; 5833e8bff74aSIngo Molnar 583403252919SAndi Kleen vma = find_vma(mm, ip); 583503252919SAndi Kleen if (vma && vma->vm_file) { 583603252919SAndi Kleen struct file *f = vma->vm_file; 58370a7f682dSMichal Hocko char *buf = (char *)__get_free_page(GFP_NOWAIT); 583803252919SAndi Kleen if (buf) { 58392fbc57c5SAndy Shevchenko char *p; 584003252919SAndi Kleen 58419bf39ab2SMiklos Szeredi p = file_path(f, buf, PAGE_SIZE); 584203252919SAndi Kleen if (IS_ERR(p)) 584303252919SAndi Kleen p = "?"; 58442fbc57c5SAndy Shevchenko printk("%s%s[%lx+%lx]", prefix, kbasename(p), 584503252919SAndi Kleen vma->vm_start, 584603252919SAndi Kleen vma->vm_end - vma->vm_start); 584703252919SAndi Kleen free_page((unsigned long)buf); 584803252919SAndi Kleen } 584903252919SAndi Kleen } 5850d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 585103252919SAndi Kleen } 58523ee1afa3SNick Piggin 5853662bbcb2SMichael S. Tsirkin #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) 58549ec23531SDavid Hildenbrand void __might_fault(const char *file, int line) 58553ee1afa3SNick Piggin { 58569ec23531SDavid Hildenbrand if (pagefault_disabled()) 5857662bbcb2SMichael S. Tsirkin return; 585842a38756SThomas Gleixner __might_sleep(file, line); 58599ec23531SDavid Hildenbrand #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) 5860662bbcb2SMichael S. Tsirkin if (current->mm) 5861da1c55f1SMichel Lespinasse might_lock_read(¤t->mm->mmap_lock); 58629ec23531SDavid Hildenbrand #endif 58633ee1afa3SNick Piggin } 58649ec23531SDavid Hildenbrand EXPORT_SYMBOL(__might_fault); 58653ee1afa3SNick Piggin #endif 586647ad8475SAndrea Arcangeli 586747ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 5868c6ddfb6cSHuang Ying /* 5869c6ddfb6cSHuang Ying * Process all subpages of the specified huge page with the specified 5870c6ddfb6cSHuang Ying * operation. The target subpage will be processed last to keep its 5871c6ddfb6cSHuang Ying * cache lines hot. 5872c6ddfb6cSHuang Ying */ 58731cb9dc4bSLiu Shixin static inline int process_huge_page( 5874c6ddfb6cSHuang Ying unsigned long addr_hint, unsigned int pages_per_huge_page, 58751cb9dc4bSLiu Shixin int (*process_subpage)(unsigned long addr, int idx, void *arg), 5876c6ddfb6cSHuang Ying void *arg) 5877c6ddfb6cSHuang Ying { 58781cb9dc4bSLiu Shixin int i, n, base, l, ret; 5879c6ddfb6cSHuang Ying unsigned long addr = addr_hint & 5880c6ddfb6cSHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 5881c6ddfb6cSHuang Ying 5882c6ddfb6cSHuang Ying /* Process target subpage last to keep its cache lines hot */ 5883c6ddfb6cSHuang Ying might_sleep(); 5884c6ddfb6cSHuang Ying n = (addr_hint - addr) / PAGE_SIZE; 5885c6ddfb6cSHuang Ying if (2 * n <= pages_per_huge_page) { 5886c6ddfb6cSHuang Ying /* If target subpage in first half of huge page */ 5887c6ddfb6cSHuang Ying base = 0; 5888c6ddfb6cSHuang Ying l = n; 5889c6ddfb6cSHuang Ying /* Process subpages at the end of huge page */ 5890c6ddfb6cSHuang Ying for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { 5891c6ddfb6cSHuang Ying cond_resched(); 58921cb9dc4bSLiu Shixin ret = process_subpage(addr + i * PAGE_SIZE, i, arg); 58931cb9dc4bSLiu Shixin if (ret) 58941cb9dc4bSLiu Shixin return ret; 5895c6ddfb6cSHuang Ying } 5896c6ddfb6cSHuang Ying } else { 5897c6ddfb6cSHuang Ying /* If target subpage in second half of huge page */ 5898c6ddfb6cSHuang Ying base = pages_per_huge_page - 2 * (pages_per_huge_page - n); 5899c6ddfb6cSHuang Ying l = pages_per_huge_page - n; 5900c6ddfb6cSHuang Ying /* Process subpages at the begin of huge page */ 5901c6ddfb6cSHuang Ying for (i = 0; i < base; i++) { 5902c6ddfb6cSHuang Ying cond_resched(); 59031cb9dc4bSLiu Shixin ret = process_subpage(addr + i * PAGE_SIZE, i, arg); 59041cb9dc4bSLiu Shixin if (ret) 59051cb9dc4bSLiu Shixin return ret; 5906c6ddfb6cSHuang Ying } 5907c6ddfb6cSHuang Ying } 5908c6ddfb6cSHuang Ying /* 5909c6ddfb6cSHuang Ying * Process remaining subpages in left-right-left-right pattern 5910c6ddfb6cSHuang Ying * towards the target subpage 5911c6ddfb6cSHuang Ying */ 5912c6ddfb6cSHuang Ying for (i = 0; i < l; i++) { 5913c6ddfb6cSHuang Ying int left_idx = base + i; 5914c6ddfb6cSHuang Ying int right_idx = base + 2 * l - 1 - i; 5915c6ddfb6cSHuang Ying 5916c6ddfb6cSHuang Ying cond_resched(); 59171cb9dc4bSLiu Shixin ret = process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg); 59181cb9dc4bSLiu Shixin if (ret) 59191cb9dc4bSLiu Shixin return ret; 5920c6ddfb6cSHuang Ying cond_resched(); 59211cb9dc4bSLiu Shixin ret = process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg); 59221cb9dc4bSLiu Shixin if (ret) 59231cb9dc4bSLiu Shixin return ret; 5924c6ddfb6cSHuang Ying } 59251cb9dc4bSLiu Shixin return 0; 5926c6ddfb6cSHuang Ying } 5927c6ddfb6cSHuang Ying 592847ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 592947ad8475SAndrea Arcangeli unsigned long addr, 593047ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 593147ad8475SAndrea Arcangeli { 593247ad8475SAndrea Arcangeli int i; 593314455eabSCheng Li struct page *p; 593447ad8475SAndrea Arcangeli 593547ad8475SAndrea Arcangeli might_sleep(); 593614455eabSCheng Li for (i = 0; i < pages_per_huge_page; i++) { 593714455eabSCheng Li p = nth_page(page, i); 593847ad8475SAndrea Arcangeli cond_resched(); 593947ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 594047ad8475SAndrea Arcangeli } 594147ad8475SAndrea Arcangeli } 5942c6ddfb6cSHuang Ying 59431cb9dc4bSLiu Shixin static int clear_subpage(unsigned long addr, int idx, void *arg) 5944c6ddfb6cSHuang Ying { 5945c6ddfb6cSHuang Ying struct page *page = arg; 5946c6ddfb6cSHuang Ying 5947c6ddfb6cSHuang Ying clear_user_highpage(page + idx, addr); 59481cb9dc4bSLiu Shixin return 0; 5949c6ddfb6cSHuang Ying } 5950c6ddfb6cSHuang Ying 595147ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 5952c79b57e4SHuang Ying unsigned long addr_hint, unsigned int pages_per_huge_page) 595347ad8475SAndrea Arcangeli { 5954c79b57e4SHuang Ying unsigned long addr = addr_hint & 5955c79b57e4SHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 595647ad8475SAndrea Arcangeli 595747ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 595847ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 595947ad8475SAndrea Arcangeli return; 596047ad8475SAndrea Arcangeli } 596147ad8475SAndrea Arcangeli 5962c6ddfb6cSHuang Ying process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page); 596347ad8475SAndrea Arcangeli } 596447ad8475SAndrea Arcangeli 59651cb9dc4bSLiu Shixin static int copy_user_gigantic_page(struct folio *dst, struct folio *src, 596647ad8475SAndrea Arcangeli unsigned long addr, 596747ad8475SAndrea Arcangeli struct vm_area_struct *vma, 596847ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 596947ad8475SAndrea Arcangeli { 597047ad8475SAndrea Arcangeli int i; 5971c0e8150eSZhangPeng struct page *dst_page; 5972c0e8150eSZhangPeng struct page *src_page; 597347ad8475SAndrea Arcangeli 597414455eabSCheng Li for (i = 0; i < pages_per_huge_page; i++) { 5975c0e8150eSZhangPeng dst_page = folio_page(dst, i); 5976c0e8150eSZhangPeng src_page = folio_page(src, i); 597714455eabSCheng Li 597847ad8475SAndrea Arcangeli cond_resched(); 59791cb9dc4bSLiu Shixin if (copy_mc_user_highpage(dst_page, src_page, 59801cb9dc4bSLiu Shixin addr + i*PAGE_SIZE, vma)) { 59811cb9dc4bSLiu Shixin memory_failure_queue(page_to_pfn(src_page), 0); 59821cb9dc4bSLiu Shixin return -EHWPOISON; 598347ad8475SAndrea Arcangeli } 598447ad8475SAndrea Arcangeli } 59851cb9dc4bSLiu Shixin return 0; 59861cb9dc4bSLiu Shixin } 598747ad8475SAndrea Arcangeli 5988c9f4cd71SHuang Ying struct copy_subpage_arg { 5989c9f4cd71SHuang Ying struct page *dst; 5990c9f4cd71SHuang Ying struct page *src; 5991c9f4cd71SHuang Ying struct vm_area_struct *vma; 5992c9f4cd71SHuang Ying }; 5993c9f4cd71SHuang Ying 59941cb9dc4bSLiu Shixin static int copy_subpage(unsigned long addr, int idx, void *arg) 5995c9f4cd71SHuang Ying { 5996c9f4cd71SHuang Ying struct copy_subpage_arg *copy_arg = arg; 5997c9f4cd71SHuang Ying 59981cb9dc4bSLiu Shixin if (copy_mc_user_highpage(copy_arg->dst + idx, copy_arg->src + idx, 59991cb9dc4bSLiu Shixin addr, copy_arg->vma)) { 60001cb9dc4bSLiu Shixin memory_failure_queue(page_to_pfn(copy_arg->src + idx), 0); 60011cb9dc4bSLiu Shixin return -EHWPOISON; 60021cb9dc4bSLiu Shixin } 60031cb9dc4bSLiu Shixin return 0; 6004c9f4cd71SHuang Ying } 6005c9f4cd71SHuang Ying 60061cb9dc4bSLiu Shixin int copy_user_large_folio(struct folio *dst, struct folio *src, 6007c0e8150eSZhangPeng unsigned long addr_hint, struct vm_area_struct *vma) 600847ad8475SAndrea Arcangeli { 6009c0e8150eSZhangPeng unsigned int pages_per_huge_page = folio_nr_pages(dst); 6010c9f4cd71SHuang Ying unsigned long addr = addr_hint & 6011c9f4cd71SHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 6012c9f4cd71SHuang Ying struct copy_subpage_arg arg = { 6013c0e8150eSZhangPeng .dst = &dst->page, 6014c0e8150eSZhangPeng .src = &src->page, 6015c9f4cd71SHuang Ying .vma = vma, 6016c9f4cd71SHuang Ying }; 601747ad8475SAndrea Arcangeli 60181cb9dc4bSLiu Shixin if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) 60191cb9dc4bSLiu Shixin return copy_user_gigantic_page(dst, src, addr, vma, 602047ad8475SAndrea Arcangeli pages_per_huge_page); 602147ad8475SAndrea Arcangeli 60221cb9dc4bSLiu Shixin return process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg); 602347ad8475SAndrea Arcangeli } 6024fa4d75c1SMike Kravetz 6025e87340caSZhangPeng long copy_folio_from_user(struct folio *dst_folio, 6026fa4d75c1SMike Kravetz const void __user *usr_src, 6027810a56b9SMike Kravetz bool allow_pagefault) 6028fa4d75c1SMike Kravetz { 6029e87340caSZhangPeng void *kaddr; 6030fa4d75c1SMike Kravetz unsigned long i, rc = 0; 6031e87340caSZhangPeng unsigned int nr_pages = folio_nr_pages(dst_folio); 6032e87340caSZhangPeng unsigned long ret_val = nr_pages * PAGE_SIZE; 603314455eabSCheng Li struct page *subpage; 6034fa4d75c1SMike Kravetz 6035e87340caSZhangPeng for (i = 0; i < nr_pages; i++) { 6036e87340caSZhangPeng subpage = folio_page(dst_folio, i); 6037e87340caSZhangPeng kaddr = kmap_local_page(subpage); 60380d508c1fSZhangPeng if (!allow_pagefault) 60390d508c1fSZhangPeng pagefault_disable(); 6040e87340caSZhangPeng rc = copy_from_user(kaddr, usr_src + i * PAGE_SIZE, PAGE_SIZE); 60410d508c1fSZhangPeng if (!allow_pagefault) 60420d508c1fSZhangPeng pagefault_enable(); 6043e87340caSZhangPeng kunmap_local(kaddr); 6044fa4d75c1SMike Kravetz 6045fa4d75c1SMike Kravetz ret_val -= (PAGE_SIZE - rc); 6046fa4d75c1SMike Kravetz if (rc) 6047fa4d75c1SMike Kravetz break; 6048fa4d75c1SMike Kravetz 6049e763243cSMuchun Song flush_dcache_page(subpage); 6050e763243cSMuchun Song 6051fa4d75c1SMike Kravetz cond_resched(); 6052fa4d75c1SMike Kravetz } 6053fa4d75c1SMike Kravetz return ret_val; 6054fa4d75c1SMike Kravetz } 605547ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 605649076ec2SKirill A. Shutemov 605740b64acdSOlof Johansson #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS 6058b35f1819SKirill A. Shutemov 6059b35f1819SKirill A. Shutemov static struct kmem_cache *page_ptl_cachep; 6060b35f1819SKirill A. Shutemov 6061b35f1819SKirill A. Shutemov void __init ptlock_cache_init(void) 6062b35f1819SKirill A. Shutemov { 6063b35f1819SKirill A. Shutemov page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, 6064b35f1819SKirill A. Shutemov SLAB_PANIC, NULL); 6065b35f1819SKirill A. Shutemov } 6066b35f1819SKirill A. Shutemov 6067539edb58SPeter Zijlstra bool ptlock_alloc(struct page *page) 606849076ec2SKirill A. Shutemov { 606949076ec2SKirill A. Shutemov spinlock_t *ptl; 607049076ec2SKirill A. Shutemov 6071b35f1819SKirill A. Shutemov ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); 607249076ec2SKirill A. Shutemov if (!ptl) 607349076ec2SKirill A. Shutemov return false; 6074539edb58SPeter Zijlstra page->ptl = ptl; 607549076ec2SKirill A. Shutemov return true; 607649076ec2SKirill A. Shutemov } 607749076ec2SKirill A. Shutemov 6078539edb58SPeter Zijlstra void ptlock_free(struct page *page) 607949076ec2SKirill A. Shutemov { 6080b35f1819SKirill A. Shutemov kmem_cache_free(page_ptl_cachep, page->ptl); 608149076ec2SKirill A. Shutemov } 608249076ec2SKirill A. Shutemov #endif 6083