11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/mm/memory.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * demand-loading started 01.12.91 - seems it is high on the list of 91da177e4SLinus Torvalds * things wanted, and it should be easy to implement. - Linus 101da177e4SLinus Torvalds */ 111da177e4SLinus Torvalds 121da177e4SLinus Torvalds /* 131da177e4SLinus Torvalds * Ok, demand-loading was easy, shared pages a little bit tricker. Shared 141da177e4SLinus Torvalds * pages started 02.12.91, seems to work. - Linus. 151da177e4SLinus Torvalds * 161da177e4SLinus Torvalds * Tested sharing by executing about 30 /bin/sh: under the old kernel it 171da177e4SLinus Torvalds * would have taken more than the 6M I have free, but it worked well as 181da177e4SLinus Torvalds * far as I could see. 191da177e4SLinus Torvalds * 201da177e4SLinus Torvalds * Also corrected some "invalidate()"s - I wasn't doing enough of them. 211da177e4SLinus Torvalds */ 221da177e4SLinus Torvalds 231da177e4SLinus Torvalds /* 241da177e4SLinus Torvalds * Real VM (paging to/from disk) started 18.12.91. Much more work and 251da177e4SLinus Torvalds * thought has to go into this. Oh, well.. 261da177e4SLinus Torvalds * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. 271da177e4SLinus Torvalds * Found it. Everything seems to work now. 281da177e4SLinus Torvalds * 20.12.91 - Ok, making the swap-device changeable like the root. 291da177e4SLinus Torvalds */ 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds /* 321da177e4SLinus Torvalds * 05.04.94 - Multi-page memory management added for v1.1. 331da177e4SLinus Torvalds * Idea by Alex Bligh (alex@cconcepts.co.uk) 341da177e4SLinus Torvalds * 351da177e4SLinus Torvalds * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG 361da177e4SLinus Torvalds * (Gerhard.Wichert@pdb.siemens.de) 371da177e4SLinus Torvalds * 381da177e4SLinus Torvalds * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) 391da177e4SLinus Torvalds */ 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds #include <linux/kernel_stat.h> 421da177e4SLinus Torvalds #include <linux/mm.h> 431da177e4SLinus Torvalds #include <linux/hugetlb.h> 441da177e4SLinus Torvalds #include <linux/mman.h> 451da177e4SLinus Torvalds #include <linux/swap.h> 461da177e4SLinus Torvalds #include <linux/highmem.h> 471da177e4SLinus Torvalds #include <linux/pagemap.h> 489a840895SHugh Dickins #include <linux/ksm.h> 491da177e4SLinus Torvalds #include <linux/rmap.h> 501da177e4SLinus Torvalds #include <linux/module.h> 510ff92245SShailabh Nagar #include <linux/delayacct.h> 521da177e4SLinus Torvalds #include <linux/init.h> 53edc79b2aSPeter Zijlstra #include <linux/writeback.h> 548a9f3ccdSBalbir Singh #include <linux/memcontrol.h> 55cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 563dc14741SHugh Dickins #include <linux/kallsyms.h> 573dc14741SHugh Dickins #include <linux/swapops.h> 583dc14741SHugh Dickins #include <linux/elf.h> 595a0e3ad6STejun Heo #include <linux/gfp.h> 601da177e4SLinus Torvalds 616952b61dSAlexey Dobriyan #include <asm/io.h> 621da177e4SLinus Torvalds #include <asm/pgalloc.h> 631da177e4SLinus Torvalds #include <asm/uaccess.h> 641da177e4SLinus Torvalds #include <asm/tlb.h> 651da177e4SLinus Torvalds #include <asm/tlbflush.h> 661da177e4SLinus Torvalds #include <asm/pgtable.h> 671da177e4SLinus Torvalds 6842b77728SJan Beulich #include "internal.h" 6942b77728SJan Beulich 70d41dee36SAndy Whitcroft #ifndef CONFIG_NEED_MULTIPLE_NODES 711da177e4SLinus Torvalds /* use the per-pgdat data instead for discontigmem - mbligh */ 721da177e4SLinus Torvalds unsigned long max_mapnr; 731da177e4SLinus Torvalds struct page *mem_map; 741da177e4SLinus Torvalds 751da177e4SLinus Torvalds EXPORT_SYMBOL(max_mapnr); 761da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 771da177e4SLinus Torvalds #endif 781da177e4SLinus Torvalds 791da177e4SLinus Torvalds unsigned long num_physpages; 801da177e4SLinus Torvalds /* 811da177e4SLinus Torvalds * A number of key systems in x86 including ioremap() rely on the assumption 821da177e4SLinus Torvalds * that high_memory defines the upper bound on direct map memory, then end 831da177e4SLinus Torvalds * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and 841da177e4SLinus Torvalds * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL 851da177e4SLinus Torvalds * and ZONE_HIGHMEM. 861da177e4SLinus Torvalds */ 871da177e4SLinus Torvalds void * high_memory; 881da177e4SLinus Torvalds 891da177e4SLinus Torvalds EXPORT_SYMBOL(num_physpages); 901da177e4SLinus Torvalds EXPORT_SYMBOL(high_memory); 911da177e4SLinus Torvalds 9232a93233SIngo Molnar /* 9332a93233SIngo Molnar * Randomize the address space (stacks, mmaps, brk, etc.). 9432a93233SIngo Molnar * 9532a93233SIngo Molnar * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, 9632a93233SIngo Molnar * as ancient (libc5 based) binaries can segfault. ) 9732a93233SIngo Molnar */ 9832a93233SIngo Molnar int randomize_va_space __read_mostly = 9932a93233SIngo Molnar #ifdef CONFIG_COMPAT_BRK 10032a93233SIngo Molnar 1; 10132a93233SIngo Molnar #else 10232a93233SIngo Molnar 2; 10332a93233SIngo Molnar #endif 104a62eaf15SAndi Kleen 105a62eaf15SAndi Kleen static int __init disable_randmaps(char *s) 106a62eaf15SAndi Kleen { 107a62eaf15SAndi Kleen randomize_va_space = 0; 1089b41046cSOGAWA Hirofumi return 1; 109a62eaf15SAndi Kleen } 110a62eaf15SAndi Kleen __setup("norandmaps", disable_randmaps); 111a62eaf15SAndi Kleen 11262eede62SHugh Dickins unsigned long zero_pfn __read_mostly; 11303f6462aSHugh Dickins unsigned long highest_memmap_pfn __read_mostly; 114a13ea5b7SHugh Dickins 115a13ea5b7SHugh Dickins /* 116a13ea5b7SHugh Dickins * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() 117a13ea5b7SHugh Dickins */ 118a13ea5b7SHugh Dickins static int __init init_zero_pfn(void) 119a13ea5b7SHugh Dickins { 120a13ea5b7SHugh Dickins zero_pfn = page_to_pfn(ZERO_PAGE(0)); 121a13ea5b7SHugh Dickins return 0; 122a13ea5b7SHugh Dickins } 123a13ea5b7SHugh Dickins core_initcall(init_zero_pfn); 124a62eaf15SAndi Kleen 125d559db08SKAMEZAWA Hiroyuki 12634e55232SKAMEZAWA Hiroyuki #if defined(SPLIT_RSS_COUNTING) 12734e55232SKAMEZAWA Hiroyuki 128a3a2e76cSKAMEZAWA Hiroyuki static void __sync_task_rss_stat(struct task_struct *task, struct mm_struct *mm) 12934e55232SKAMEZAWA Hiroyuki { 13034e55232SKAMEZAWA Hiroyuki int i; 13134e55232SKAMEZAWA Hiroyuki 13234e55232SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) { 13334e55232SKAMEZAWA Hiroyuki if (task->rss_stat.count[i]) { 13434e55232SKAMEZAWA Hiroyuki add_mm_counter(mm, i, task->rss_stat.count[i]); 13534e55232SKAMEZAWA Hiroyuki task->rss_stat.count[i] = 0; 13634e55232SKAMEZAWA Hiroyuki } 13734e55232SKAMEZAWA Hiroyuki } 13834e55232SKAMEZAWA Hiroyuki task->rss_stat.events = 0; 13934e55232SKAMEZAWA Hiroyuki } 14034e55232SKAMEZAWA Hiroyuki 14134e55232SKAMEZAWA Hiroyuki static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) 14234e55232SKAMEZAWA Hiroyuki { 14334e55232SKAMEZAWA Hiroyuki struct task_struct *task = current; 14434e55232SKAMEZAWA Hiroyuki 14534e55232SKAMEZAWA Hiroyuki if (likely(task->mm == mm)) 14634e55232SKAMEZAWA Hiroyuki task->rss_stat.count[member] += val; 14734e55232SKAMEZAWA Hiroyuki else 14834e55232SKAMEZAWA Hiroyuki add_mm_counter(mm, member, val); 14934e55232SKAMEZAWA Hiroyuki } 15034e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) 15134e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) 15234e55232SKAMEZAWA Hiroyuki 15334e55232SKAMEZAWA Hiroyuki /* sync counter once per 64 page faults */ 15434e55232SKAMEZAWA Hiroyuki #define TASK_RSS_EVENTS_THRESH (64) 15534e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 15634e55232SKAMEZAWA Hiroyuki { 15734e55232SKAMEZAWA Hiroyuki if (unlikely(task != current)) 15834e55232SKAMEZAWA Hiroyuki return; 15934e55232SKAMEZAWA Hiroyuki if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) 16034e55232SKAMEZAWA Hiroyuki __sync_task_rss_stat(task, task->mm); 16134e55232SKAMEZAWA Hiroyuki } 16234e55232SKAMEZAWA Hiroyuki 16334e55232SKAMEZAWA Hiroyuki unsigned long get_mm_counter(struct mm_struct *mm, int member) 16434e55232SKAMEZAWA Hiroyuki { 16534e55232SKAMEZAWA Hiroyuki long val = 0; 16634e55232SKAMEZAWA Hiroyuki 16734e55232SKAMEZAWA Hiroyuki /* 16834e55232SKAMEZAWA Hiroyuki * Don't use task->mm here...for avoiding to use task_get_mm().. 16934e55232SKAMEZAWA Hiroyuki * The caller must guarantee task->mm is not invalid. 17034e55232SKAMEZAWA Hiroyuki */ 17134e55232SKAMEZAWA Hiroyuki val = atomic_long_read(&mm->rss_stat.count[member]); 17234e55232SKAMEZAWA Hiroyuki /* 17334e55232SKAMEZAWA Hiroyuki * counter is updated in asynchronous manner and may go to minus. 17434e55232SKAMEZAWA Hiroyuki * But it's never be expected number for users. 17534e55232SKAMEZAWA Hiroyuki */ 17634e55232SKAMEZAWA Hiroyuki if (val < 0) 17734e55232SKAMEZAWA Hiroyuki return 0; 17834e55232SKAMEZAWA Hiroyuki return (unsigned long)val; 17934e55232SKAMEZAWA Hiroyuki } 18034e55232SKAMEZAWA Hiroyuki 18134e55232SKAMEZAWA Hiroyuki void sync_mm_rss(struct task_struct *task, struct mm_struct *mm) 18234e55232SKAMEZAWA Hiroyuki { 18334e55232SKAMEZAWA Hiroyuki __sync_task_rss_stat(task, mm); 18434e55232SKAMEZAWA Hiroyuki } 18534e55232SKAMEZAWA Hiroyuki #else 18634e55232SKAMEZAWA Hiroyuki 18734e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) 18834e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) 18934e55232SKAMEZAWA Hiroyuki 19034e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 19134e55232SKAMEZAWA Hiroyuki { 19234e55232SKAMEZAWA Hiroyuki } 19334e55232SKAMEZAWA Hiroyuki 19434e55232SKAMEZAWA Hiroyuki #endif 19534e55232SKAMEZAWA Hiroyuki 19626723911SPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 19726723911SPeter Zijlstra 19826723911SPeter Zijlstra /* 19926723911SPeter Zijlstra * See the comment near struct mmu_table_batch. 20026723911SPeter Zijlstra */ 20126723911SPeter Zijlstra 20226723911SPeter Zijlstra static void tlb_remove_table_smp_sync(void *arg) 20326723911SPeter Zijlstra { 20426723911SPeter Zijlstra /* Simply deliver the interrupt */ 20526723911SPeter Zijlstra } 20626723911SPeter Zijlstra 20726723911SPeter Zijlstra static void tlb_remove_table_one(void *table) 20826723911SPeter Zijlstra { 20926723911SPeter Zijlstra /* 21026723911SPeter Zijlstra * This isn't an RCU grace period and hence the page-tables cannot be 21126723911SPeter Zijlstra * assumed to be actually RCU-freed. 21226723911SPeter Zijlstra * 21326723911SPeter Zijlstra * It is however sufficient for software page-table walkers that rely on 21426723911SPeter Zijlstra * IRQ disabling. See the comment near struct mmu_table_batch. 21526723911SPeter Zijlstra */ 21626723911SPeter Zijlstra smp_call_function(tlb_remove_table_smp_sync, NULL, 1); 21726723911SPeter Zijlstra __tlb_remove_table(table); 21826723911SPeter Zijlstra } 21926723911SPeter Zijlstra 22026723911SPeter Zijlstra static void tlb_remove_table_rcu(struct rcu_head *head) 22126723911SPeter Zijlstra { 22226723911SPeter Zijlstra struct mmu_table_batch *batch; 22326723911SPeter Zijlstra int i; 22426723911SPeter Zijlstra 22526723911SPeter Zijlstra batch = container_of(head, struct mmu_table_batch, rcu); 22626723911SPeter Zijlstra 22726723911SPeter Zijlstra for (i = 0; i < batch->nr; i++) 22826723911SPeter Zijlstra __tlb_remove_table(batch->tables[i]); 22926723911SPeter Zijlstra 23026723911SPeter Zijlstra free_page((unsigned long)batch); 23126723911SPeter Zijlstra } 23226723911SPeter Zijlstra 23326723911SPeter Zijlstra void tlb_table_flush(struct mmu_gather *tlb) 23426723911SPeter Zijlstra { 23526723911SPeter Zijlstra struct mmu_table_batch **batch = &tlb->batch; 23626723911SPeter Zijlstra 23726723911SPeter Zijlstra if (*batch) { 23826723911SPeter Zijlstra call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); 23926723911SPeter Zijlstra *batch = NULL; 24026723911SPeter Zijlstra } 24126723911SPeter Zijlstra } 24226723911SPeter Zijlstra 24326723911SPeter Zijlstra void tlb_remove_table(struct mmu_gather *tlb, void *table) 24426723911SPeter Zijlstra { 24526723911SPeter Zijlstra struct mmu_table_batch **batch = &tlb->batch; 24626723911SPeter Zijlstra 24726723911SPeter Zijlstra tlb->need_flush = 1; 24826723911SPeter Zijlstra 24926723911SPeter Zijlstra /* 25026723911SPeter Zijlstra * When there's less then two users of this mm there cannot be a 25126723911SPeter Zijlstra * concurrent page-table walk. 25226723911SPeter Zijlstra */ 25326723911SPeter Zijlstra if (atomic_read(&tlb->mm->mm_users) < 2) { 25426723911SPeter Zijlstra __tlb_remove_table(table); 25526723911SPeter Zijlstra return; 25626723911SPeter Zijlstra } 25726723911SPeter Zijlstra 25826723911SPeter Zijlstra if (*batch == NULL) { 25926723911SPeter Zijlstra *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); 26026723911SPeter Zijlstra if (*batch == NULL) { 26126723911SPeter Zijlstra tlb_remove_table_one(table); 26226723911SPeter Zijlstra return; 26326723911SPeter Zijlstra } 26426723911SPeter Zijlstra (*batch)->nr = 0; 26526723911SPeter Zijlstra } 26626723911SPeter Zijlstra (*batch)->tables[(*batch)->nr++] = table; 26726723911SPeter Zijlstra if ((*batch)->nr == MAX_TABLE_BATCH) 26826723911SPeter Zijlstra tlb_table_flush(tlb); 26926723911SPeter Zijlstra } 27026723911SPeter Zijlstra 27126723911SPeter Zijlstra #endif 27226723911SPeter Zijlstra 2731da177e4SLinus Torvalds /* 2741da177e4SLinus Torvalds * If a p?d_bad entry is found while walking page tables, report 2751da177e4SLinus Torvalds * the error, before resetting entry to p?d_none. Usually (but 2761da177e4SLinus Torvalds * very seldom) called out from the p?d_none_or_clear_bad macros. 2771da177e4SLinus Torvalds */ 2781da177e4SLinus Torvalds 2791da177e4SLinus Torvalds void pgd_clear_bad(pgd_t *pgd) 2801da177e4SLinus Torvalds { 2811da177e4SLinus Torvalds pgd_ERROR(*pgd); 2821da177e4SLinus Torvalds pgd_clear(pgd); 2831da177e4SLinus Torvalds } 2841da177e4SLinus Torvalds 2851da177e4SLinus Torvalds void pud_clear_bad(pud_t *pud) 2861da177e4SLinus Torvalds { 2871da177e4SLinus Torvalds pud_ERROR(*pud); 2881da177e4SLinus Torvalds pud_clear(pud); 2891da177e4SLinus Torvalds } 2901da177e4SLinus Torvalds 2911da177e4SLinus Torvalds void pmd_clear_bad(pmd_t *pmd) 2921da177e4SLinus Torvalds { 2931da177e4SLinus Torvalds pmd_ERROR(*pmd); 2941da177e4SLinus Torvalds pmd_clear(pmd); 2951da177e4SLinus Torvalds } 2961da177e4SLinus Torvalds 2971da177e4SLinus Torvalds /* 2981da177e4SLinus Torvalds * Note: this doesn't free the actual pages themselves. That 2991da177e4SLinus Torvalds * has been handled earlier when unmapping all the memory regions. 3001da177e4SLinus Torvalds */ 3019e1b32caSBenjamin Herrenschmidt static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, 3029e1b32caSBenjamin Herrenschmidt unsigned long addr) 3031da177e4SLinus Torvalds { 3042f569afdSMartin Schwidefsky pgtable_t token = pmd_pgtable(*pmd); 3051da177e4SLinus Torvalds pmd_clear(pmd); 3069e1b32caSBenjamin Herrenschmidt pte_free_tlb(tlb, token, addr); 3071da177e4SLinus Torvalds tlb->mm->nr_ptes--; 3081da177e4SLinus Torvalds } 3091da177e4SLinus Torvalds 310e0da382cSHugh Dickins static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, 311e0da382cSHugh Dickins unsigned long addr, unsigned long end, 312e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 3131da177e4SLinus Torvalds { 3141da177e4SLinus Torvalds pmd_t *pmd; 3151da177e4SLinus Torvalds unsigned long next; 316e0da382cSHugh Dickins unsigned long start; 3171da177e4SLinus Torvalds 318e0da382cSHugh Dickins start = addr; 3191da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 3201da177e4SLinus Torvalds do { 3211da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 3221da177e4SLinus Torvalds if (pmd_none_or_clear_bad(pmd)) 3231da177e4SLinus Torvalds continue; 3249e1b32caSBenjamin Herrenschmidt free_pte_range(tlb, pmd, addr); 3251da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 3261da177e4SLinus Torvalds 327e0da382cSHugh Dickins start &= PUD_MASK; 328e0da382cSHugh Dickins if (start < floor) 329e0da382cSHugh Dickins return; 330e0da382cSHugh Dickins if (ceiling) { 331e0da382cSHugh Dickins ceiling &= PUD_MASK; 332e0da382cSHugh Dickins if (!ceiling) 333e0da382cSHugh Dickins return; 3341da177e4SLinus Torvalds } 335e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 336e0da382cSHugh Dickins return; 337e0da382cSHugh Dickins 338e0da382cSHugh Dickins pmd = pmd_offset(pud, start); 339e0da382cSHugh Dickins pud_clear(pud); 3409e1b32caSBenjamin Herrenschmidt pmd_free_tlb(tlb, pmd, start); 3411da177e4SLinus Torvalds } 3421da177e4SLinus Torvalds 343e0da382cSHugh Dickins static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, 344e0da382cSHugh Dickins unsigned long addr, unsigned long end, 345e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 3461da177e4SLinus Torvalds { 3471da177e4SLinus Torvalds pud_t *pud; 3481da177e4SLinus Torvalds unsigned long next; 349e0da382cSHugh Dickins unsigned long start; 3501da177e4SLinus Torvalds 351e0da382cSHugh Dickins start = addr; 3521da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 3531da177e4SLinus Torvalds do { 3541da177e4SLinus Torvalds next = pud_addr_end(addr, end); 3551da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 3561da177e4SLinus Torvalds continue; 357e0da382cSHugh Dickins free_pmd_range(tlb, pud, addr, next, floor, ceiling); 3581da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 3591da177e4SLinus Torvalds 360e0da382cSHugh Dickins start &= PGDIR_MASK; 361e0da382cSHugh Dickins if (start < floor) 362e0da382cSHugh Dickins return; 363e0da382cSHugh Dickins if (ceiling) { 364e0da382cSHugh Dickins ceiling &= PGDIR_MASK; 365e0da382cSHugh Dickins if (!ceiling) 366e0da382cSHugh Dickins return; 3671da177e4SLinus Torvalds } 368e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 369e0da382cSHugh Dickins return; 370e0da382cSHugh Dickins 371e0da382cSHugh Dickins pud = pud_offset(pgd, start); 372e0da382cSHugh Dickins pgd_clear(pgd); 3739e1b32caSBenjamin Herrenschmidt pud_free_tlb(tlb, pud, start); 3741da177e4SLinus Torvalds } 3751da177e4SLinus Torvalds 3761da177e4SLinus Torvalds /* 377e0da382cSHugh Dickins * This function frees user-level page tables of a process. 378e0da382cSHugh Dickins * 3791da177e4SLinus Torvalds * Must be called with pagetable lock held. 3801da177e4SLinus Torvalds */ 38142b77728SJan Beulich void free_pgd_range(struct mmu_gather *tlb, 382e0da382cSHugh Dickins unsigned long addr, unsigned long end, 383e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 3841da177e4SLinus Torvalds { 3851da177e4SLinus Torvalds pgd_t *pgd; 3861da177e4SLinus Torvalds unsigned long next; 3871da177e4SLinus Torvalds 388e0da382cSHugh Dickins /* 389e0da382cSHugh Dickins * The next few lines have given us lots of grief... 390e0da382cSHugh Dickins * 391e0da382cSHugh Dickins * Why are we testing PMD* at this top level? Because often 392e0da382cSHugh Dickins * there will be no work to do at all, and we'd prefer not to 393e0da382cSHugh Dickins * go all the way down to the bottom just to discover that. 394e0da382cSHugh Dickins * 395e0da382cSHugh Dickins * Why all these "- 1"s? Because 0 represents both the bottom 396e0da382cSHugh Dickins * of the address space and the top of it (using -1 for the 397e0da382cSHugh Dickins * top wouldn't help much: the masks would do the wrong thing). 398e0da382cSHugh Dickins * The rule is that addr 0 and floor 0 refer to the bottom of 399e0da382cSHugh Dickins * the address space, but end 0 and ceiling 0 refer to the top 400e0da382cSHugh Dickins * Comparisons need to use "end - 1" and "ceiling - 1" (though 401e0da382cSHugh Dickins * that end 0 case should be mythical). 402e0da382cSHugh Dickins * 403e0da382cSHugh Dickins * Wherever addr is brought up or ceiling brought down, we must 404e0da382cSHugh Dickins * be careful to reject "the opposite 0" before it confuses the 405e0da382cSHugh Dickins * subsequent tests. But what about where end is brought down 406e0da382cSHugh Dickins * by PMD_SIZE below? no, end can't go down to 0 there. 407e0da382cSHugh Dickins * 408e0da382cSHugh Dickins * Whereas we round start (addr) and ceiling down, by different 409e0da382cSHugh Dickins * masks at different levels, in order to test whether a table 410e0da382cSHugh Dickins * now has no other vmas using it, so can be freed, we don't 411e0da382cSHugh Dickins * bother to round floor or end up - the tests don't need that. 412e0da382cSHugh Dickins */ 413e0da382cSHugh Dickins 414e0da382cSHugh Dickins addr &= PMD_MASK; 415e0da382cSHugh Dickins if (addr < floor) { 416e0da382cSHugh Dickins addr += PMD_SIZE; 417e0da382cSHugh Dickins if (!addr) 418e0da382cSHugh Dickins return; 419e0da382cSHugh Dickins } 420e0da382cSHugh Dickins if (ceiling) { 421e0da382cSHugh Dickins ceiling &= PMD_MASK; 422e0da382cSHugh Dickins if (!ceiling) 423e0da382cSHugh Dickins return; 424e0da382cSHugh Dickins } 425e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 426e0da382cSHugh Dickins end -= PMD_SIZE; 427e0da382cSHugh Dickins if (addr > end - 1) 428e0da382cSHugh Dickins return; 429e0da382cSHugh Dickins 43042b77728SJan Beulich pgd = pgd_offset(tlb->mm, addr); 4311da177e4SLinus Torvalds do { 4321da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 4331da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 4341da177e4SLinus Torvalds continue; 43542b77728SJan Beulich free_pud_range(tlb, pgd, addr, next, floor, ceiling); 4361da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 437e0da382cSHugh Dickins } 438e0da382cSHugh Dickins 43942b77728SJan Beulich void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, 440e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 441e0da382cSHugh Dickins { 442e0da382cSHugh Dickins while (vma) { 443e0da382cSHugh Dickins struct vm_area_struct *next = vma->vm_next; 444e0da382cSHugh Dickins unsigned long addr = vma->vm_start; 445e0da382cSHugh Dickins 4468f4f8c16SHugh Dickins /* 44725d9e2d1Snpiggin@suse.de * Hide vma from rmap and truncate_pagecache before freeing 44825d9e2d1Snpiggin@suse.de * pgtables 4498f4f8c16SHugh Dickins */ 4505beb4930SRik van Riel unlink_anon_vmas(vma); 4518f4f8c16SHugh Dickins unlink_file_vma(vma); 4528f4f8c16SHugh Dickins 4539da61aefSDavid Gibson if (is_vm_hugetlb_page(vma)) { 4543bf5ee95SHugh Dickins hugetlb_free_pgd_range(tlb, addr, vma->vm_end, 4553bf5ee95SHugh Dickins floor, next? next->vm_start: ceiling); 4563bf5ee95SHugh Dickins } else { 4573bf5ee95SHugh Dickins /* 4583bf5ee95SHugh Dickins * Optimization: gather nearby vmas into one call down 4593bf5ee95SHugh Dickins */ 4603bf5ee95SHugh Dickins while (next && next->vm_start <= vma->vm_end + PMD_SIZE 4614866920bSDavid Gibson && !is_vm_hugetlb_page(next)) { 462e0da382cSHugh Dickins vma = next; 463e0da382cSHugh Dickins next = vma->vm_next; 4645beb4930SRik van Riel unlink_anon_vmas(vma); 4658f4f8c16SHugh Dickins unlink_file_vma(vma); 466e0da382cSHugh Dickins } 4673bf5ee95SHugh Dickins free_pgd_range(tlb, addr, vma->vm_end, 468e0da382cSHugh Dickins floor, next? next->vm_start: ceiling); 4693bf5ee95SHugh Dickins } 470e0da382cSHugh Dickins vma = next; 471e0da382cSHugh Dickins } 4721da177e4SLinus Torvalds } 4731da177e4SLinus Torvalds 4748ac1f832SAndrea Arcangeli int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma, 4758ac1f832SAndrea Arcangeli pmd_t *pmd, unsigned long address) 4761da177e4SLinus Torvalds { 4772f569afdSMartin Schwidefsky pgtable_t new = pte_alloc_one(mm, address); 4788ac1f832SAndrea Arcangeli int wait_split_huge_page; 4791da177e4SLinus Torvalds if (!new) 4801bb3630eSHugh Dickins return -ENOMEM; 4811bb3630eSHugh Dickins 482362a61adSNick Piggin /* 483362a61adSNick Piggin * Ensure all pte setup (eg. pte page lock and page clearing) are 484362a61adSNick Piggin * visible before the pte is made visible to other CPUs by being 485362a61adSNick Piggin * put into page tables. 486362a61adSNick Piggin * 487362a61adSNick Piggin * The other side of the story is the pointer chasing in the page 488362a61adSNick Piggin * table walking code (when walking the page table without locking; 489362a61adSNick Piggin * ie. most of the time). Fortunately, these data accesses consist 490362a61adSNick Piggin * of a chain of data-dependent loads, meaning most CPUs (alpha 491362a61adSNick Piggin * being the notable exception) will already guarantee loads are 492362a61adSNick Piggin * seen in-order. See the alpha page table accessors for the 493362a61adSNick Piggin * smp_read_barrier_depends() barriers in page table walking code. 494362a61adSNick Piggin */ 495362a61adSNick Piggin smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ 496362a61adSNick Piggin 497c74df32cSHugh Dickins spin_lock(&mm->page_table_lock); 4988ac1f832SAndrea Arcangeli wait_split_huge_page = 0; 4998ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 5001da177e4SLinus Torvalds mm->nr_ptes++; 5011da177e4SLinus Torvalds pmd_populate(mm, pmd, new); 5022f569afdSMartin Schwidefsky new = NULL; 5038ac1f832SAndrea Arcangeli } else if (unlikely(pmd_trans_splitting(*pmd))) 5048ac1f832SAndrea Arcangeli wait_split_huge_page = 1; 505c74df32cSHugh Dickins spin_unlock(&mm->page_table_lock); 5062f569afdSMartin Schwidefsky if (new) 5072f569afdSMartin Schwidefsky pte_free(mm, new); 5088ac1f832SAndrea Arcangeli if (wait_split_huge_page) 5098ac1f832SAndrea Arcangeli wait_split_huge_page(vma->anon_vma, pmd); 5101bb3630eSHugh Dickins return 0; 5111da177e4SLinus Torvalds } 5121da177e4SLinus Torvalds 5131bb3630eSHugh Dickins int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) 5141da177e4SLinus Torvalds { 5151bb3630eSHugh Dickins pte_t *new = pte_alloc_one_kernel(&init_mm, address); 5161da177e4SLinus Torvalds if (!new) 5171bb3630eSHugh Dickins return -ENOMEM; 5181da177e4SLinus Torvalds 519362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 520362a61adSNick Piggin 521872fec16SHugh Dickins spin_lock(&init_mm.page_table_lock); 5228ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 523872fec16SHugh Dickins pmd_populate_kernel(&init_mm, pmd, new); 5242f569afdSMartin Schwidefsky new = NULL; 5258ac1f832SAndrea Arcangeli } else 5268ac1f832SAndrea Arcangeli VM_BUG_ON(pmd_trans_splitting(*pmd)); 527872fec16SHugh Dickins spin_unlock(&init_mm.page_table_lock); 5282f569afdSMartin Schwidefsky if (new) 5292f569afdSMartin Schwidefsky pte_free_kernel(&init_mm, new); 5301bb3630eSHugh Dickins return 0; 5311da177e4SLinus Torvalds } 5321da177e4SLinus Torvalds 533d559db08SKAMEZAWA Hiroyuki static inline void init_rss_vec(int *rss) 534ae859762SHugh Dickins { 535d559db08SKAMEZAWA Hiroyuki memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); 536d559db08SKAMEZAWA Hiroyuki } 537d559db08SKAMEZAWA Hiroyuki 538d559db08SKAMEZAWA Hiroyuki static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) 539d559db08SKAMEZAWA Hiroyuki { 540d559db08SKAMEZAWA Hiroyuki int i; 541d559db08SKAMEZAWA Hiroyuki 54234e55232SKAMEZAWA Hiroyuki if (current->mm == mm) 54334e55232SKAMEZAWA Hiroyuki sync_mm_rss(current, mm); 544d559db08SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) 545d559db08SKAMEZAWA Hiroyuki if (rss[i]) 546d559db08SKAMEZAWA Hiroyuki add_mm_counter(mm, i, rss[i]); 547ae859762SHugh Dickins } 548ae859762SHugh Dickins 5491da177e4SLinus Torvalds /* 5506aab341eSLinus Torvalds * This function is called to print an error when a bad pte 5516aab341eSLinus Torvalds * is found. For example, we might have a PFN-mapped pte in 5526aab341eSLinus Torvalds * a region that doesn't allow it. 553b5810039SNick Piggin * 554b5810039SNick Piggin * The calling function must still handle the error. 555b5810039SNick Piggin */ 5563dc14741SHugh Dickins static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, 5573dc14741SHugh Dickins pte_t pte, struct page *page) 558b5810039SNick Piggin { 5593dc14741SHugh Dickins pgd_t *pgd = pgd_offset(vma->vm_mm, addr); 5603dc14741SHugh Dickins pud_t *pud = pud_offset(pgd, addr); 5613dc14741SHugh Dickins pmd_t *pmd = pmd_offset(pud, addr); 5623dc14741SHugh Dickins struct address_space *mapping; 5633dc14741SHugh Dickins pgoff_t index; 564d936cf9bSHugh Dickins static unsigned long resume; 565d936cf9bSHugh Dickins static unsigned long nr_shown; 566d936cf9bSHugh Dickins static unsigned long nr_unshown; 567d936cf9bSHugh Dickins 568d936cf9bSHugh Dickins /* 569d936cf9bSHugh Dickins * Allow a burst of 60 reports, then keep quiet for that minute; 570d936cf9bSHugh Dickins * or allow a steady drip of one report per second. 571d936cf9bSHugh Dickins */ 572d936cf9bSHugh Dickins if (nr_shown == 60) { 573d936cf9bSHugh Dickins if (time_before(jiffies, resume)) { 574d936cf9bSHugh Dickins nr_unshown++; 575d936cf9bSHugh Dickins return; 576d936cf9bSHugh Dickins } 577d936cf9bSHugh Dickins if (nr_unshown) { 5781e9e6365SHugh Dickins printk(KERN_ALERT 5791e9e6365SHugh Dickins "BUG: Bad page map: %lu messages suppressed\n", 580d936cf9bSHugh Dickins nr_unshown); 581d936cf9bSHugh Dickins nr_unshown = 0; 582d936cf9bSHugh Dickins } 583d936cf9bSHugh Dickins nr_shown = 0; 584d936cf9bSHugh Dickins } 585d936cf9bSHugh Dickins if (nr_shown++ == 0) 586d936cf9bSHugh Dickins resume = jiffies + 60 * HZ; 5873dc14741SHugh Dickins 5883dc14741SHugh Dickins mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; 5893dc14741SHugh Dickins index = linear_page_index(vma, addr); 5903dc14741SHugh Dickins 5911e9e6365SHugh Dickins printk(KERN_ALERT 5921e9e6365SHugh Dickins "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", 5933dc14741SHugh Dickins current->comm, 5943dc14741SHugh Dickins (long long)pte_val(pte), (long long)pmd_val(*pmd)); 595718a3821SWu Fengguang if (page) 596718a3821SWu Fengguang dump_page(page); 5971e9e6365SHugh Dickins printk(KERN_ALERT 5983dc14741SHugh Dickins "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", 5993dc14741SHugh Dickins (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); 6003dc14741SHugh Dickins /* 6013dc14741SHugh Dickins * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y 6023dc14741SHugh Dickins */ 6033dc14741SHugh Dickins if (vma->vm_ops) 6041e9e6365SHugh Dickins print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n", 6053dc14741SHugh Dickins (unsigned long)vma->vm_ops->fault); 6063dc14741SHugh Dickins if (vma->vm_file && vma->vm_file->f_op) 6071e9e6365SHugh Dickins print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n", 6083dc14741SHugh Dickins (unsigned long)vma->vm_file->f_op->mmap); 609b5810039SNick Piggin dump_stack(); 6103dc14741SHugh Dickins add_taint(TAINT_BAD_PAGE); 611b5810039SNick Piggin } 612b5810039SNick Piggin 61367121172SLinus Torvalds static inline int is_cow_mapping(unsigned int flags) 61467121172SLinus Torvalds { 61567121172SLinus Torvalds return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; 61667121172SLinus Torvalds } 61767121172SLinus Torvalds 61862eede62SHugh Dickins #ifndef is_zero_pfn 61962eede62SHugh Dickins static inline int is_zero_pfn(unsigned long pfn) 62062eede62SHugh Dickins { 62162eede62SHugh Dickins return pfn == zero_pfn; 62262eede62SHugh Dickins } 62362eede62SHugh Dickins #endif 62462eede62SHugh Dickins 62562eede62SHugh Dickins #ifndef my_zero_pfn 62662eede62SHugh Dickins static inline unsigned long my_zero_pfn(unsigned long addr) 62762eede62SHugh Dickins { 62862eede62SHugh Dickins return zero_pfn; 62962eede62SHugh Dickins } 63062eede62SHugh Dickins #endif 63162eede62SHugh Dickins 632b5810039SNick Piggin /* 6337e675137SNick Piggin * vm_normal_page -- This function gets the "struct page" associated with a pte. 6346aab341eSLinus Torvalds * 6357e675137SNick Piggin * "Special" mappings do not wish to be associated with a "struct page" (either 6367e675137SNick Piggin * it doesn't exist, or it exists but they don't want to touch it). In this 6377e675137SNick Piggin * case, NULL is returned here. "Normal" mappings do have a struct page. 638b379d790SJared Hulbert * 6397e675137SNick Piggin * There are 2 broad cases. Firstly, an architecture may define a pte_special() 6407e675137SNick Piggin * pte bit, in which case this function is trivial. Secondly, an architecture 6417e675137SNick Piggin * may not have a spare pte bit, which requires a more complicated scheme, 6427e675137SNick Piggin * described below. 6437e675137SNick Piggin * 6447e675137SNick Piggin * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a 6457e675137SNick Piggin * special mapping (even if there are underlying and valid "struct pages"). 6467e675137SNick Piggin * COWed pages of a VM_PFNMAP are always normal. 6476aab341eSLinus Torvalds * 648b379d790SJared Hulbert * The way we recognize COWed pages within VM_PFNMAP mappings is through the 649b379d790SJared Hulbert * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit 6507e675137SNick Piggin * set, and the vm_pgoff will point to the first PFN mapped: thus every special 6517e675137SNick Piggin * mapping will always honor the rule 6526aab341eSLinus Torvalds * 6536aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 6546aab341eSLinus Torvalds * 6557e675137SNick Piggin * And for normal mappings this is false. 656b379d790SJared Hulbert * 6577e675137SNick Piggin * This restricts such mappings to be a linear translation from virtual address 6587e675137SNick Piggin * to pfn. To get around this restriction, we allow arbitrary mappings so long 6597e675137SNick Piggin * as the vma is not a COW mapping; in that case, we know that all ptes are 6607e675137SNick Piggin * special (because none can have been COWed). 661b379d790SJared Hulbert * 662b379d790SJared Hulbert * 6637e675137SNick Piggin * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. 6647e675137SNick Piggin * 665b379d790SJared Hulbert * VM_MIXEDMAP mappings can likewise contain memory with or without "struct 666b379d790SJared Hulbert * page" backing, however the difference is that _all_ pages with a struct 667b379d790SJared Hulbert * page (that is, those where pfn_valid is true) are refcounted and considered 668b379d790SJared Hulbert * normal pages by the VM. The disadvantage is that pages are refcounted 669b379d790SJared Hulbert * (which can be slower and simply not an option for some PFNMAP users). The 670b379d790SJared Hulbert * advantage is that we don't have to follow the strict linearity rule of 671b379d790SJared Hulbert * PFNMAP mappings in order to support COWable mappings. 672b379d790SJared Hulbert * 673ee498ed7SHugh Dickins */ 6747e675137SNick Piggin #ifdef __HAVE_ARCH_PTE_SPECIAL 6757e675137SNick Piggin # define HAVE_PTE_SPECIAL 1 6767e675137SNick Piggin #else 6777e675137SNick Piggin # define HAVE_PTE_SPECIAL 0 6787e675137SNick Piggin #endif 6797e675137SNick Piggin struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 6807e675137SNick Piggin pte_t pte) 681ee498ed7SHugh Dickins { 68222b31eecSHugh Dickins unsigned long pfn = pte_pfn(pte); 6837e675137SNick Piggin 6847e675137SNick Piggin if (HAVE_PTE_SPECIAL) { 68522b31eecSHugh Dickins if (likely(!pte_special(pte))) 68622b31eecSHugh Dickins goto check_pfn; 687a13ea5b7SHugh Dickins if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 688a13ea5b7SHugh Dickins return NULL; 68962eede62SHugh Dickins if (!is_zero_pfn(pfn)) 69022b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 6917e675137SNick Piggin return NULL; 6927e675137SNick Piggin } 6937e675137SNick Piggin 6947e675137SNick Piggin /* !HAVE_PTE_SPECIAL case follows: */ 6957e675137SNick Piggin 696b379d790SJared Hulbert if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 697b379d790SJared Hulbert if (vma->vm_flags & VM_MIXEDMAP) { 698b379d790SJared Hulbert if (!pfn_valid(pfn)) 699b379d790SJared Hulbert return NULL; 700b379d790SJared Hulbert goto out; 701b379d790SJared Hulbert } else { 7027e675137SNick Piggin unsigned long off; 7037e675137SNick Piggin off = (addr - vma->vm_start) >> PAGE_SHIFT; 7046aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 7056aab341eSLinus Torvalds return NULL; 70667121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 707fb155c16SLinus Torvalds return NULL; 7086aab341eSLinus Torvalds } 709b379d790SJared Hulbert } 7106aab341eSLinus Torvalds 71162eede62SHugh Dickins if (is_zero_pfn(pfn)) 71262eede62SHugh Dickins return NULL; 71322b31eecSHugh Dickins check_pfn: 71422b31eecSHugh Dickins if (unlikely(pfn > highest_memmap_pfn)) { 71522b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 71622b31eecSHugh Dickins return NULL; 71722b31eecSHugh Dickins } 7186aab341eSLinus Torvalds 7196aab341eSLinus Torvalds /* 7207e675137SNick Piggin * NOTE! We still have PageReserved() pages in the page tables. 7217e675137SNick Piggin * eg. VDSO mappings can cause them to exist. 7226aab341eSLinus Torvalds */ 723b379d790SJared Hulbert out: 7246aab341eSLinus Torvalds return pfn_to_page(pfn); 725ee498ed7SHugh Dickins } 726ee498ed7SHugh Dickins 727ee498ed7SHugh Dickins /* 7281da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 7291da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 7301da177e4SLinus Torvalds * covered by this vma. 7311da177e4SLinus Torvalds */ 7321da177e4SLinus Torvalds 733570a335bSHugh Dickins static inline unsigned long 7341da177e4SLinus Torvalds copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 735b5810039SNick Piggin pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, 7368c103762SHugh Dickins unsigned long addr, int *rss) 7371da177e4SLinus Torvalds { 738b5810039SNick Piggin unsigned long vm_flags = vma->vm_flags; 7391da177e4SLinus Torvalds pte_t pte = *src_pte; 7401da177e4SLinus Torvalds struct page *page; 7411da177e4SLinus Torvalds 7421da177e4SLinus Torvalds /* pte contains position in swap or file, so copy. */ 7431da177e4SLinus Torvalds if (unlikely(!pte_present(pte))) { 7441da177e4SLinus Torvalds if (!pte_file(pte)) { 7450697212aSChristoph Lameter swp_entry_t entry = pte_to_swp_entry(pte); 7460697212aSChristoph Lameter 747570a335bSHugh Dickins if (swap_duplicate(entry) < 0) 748570a335bSHugh Dickins return entry.val; 749570a335bSHugh Dickins 7501da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 7511da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 7521da177e4SLinus Torvalds spin_lock(&mmlist_lock); 753f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 754f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 755f412ac08SHugh Dickins &src_mm->mmlist); 7561da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 7571da177e4SLinus Torvalds } 758b084d435SKAMEZAWA Hiroyuki if (likely(!non_swap_entry(entry))) 759b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]++; 760b084d435SKAMEZAWA Hiroyuki else if (is_write_migration_entry(entry) && 7610697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 7620697212aSChristoph Lameter /* 7630697212aSChristoph Lameter * COW mappings require pages in both parent 7640697212aSChristoph Lameter * and child to be set to read. 7650697212aSChristoph Lameter */ 7660697212aSChristoph Lameter make_migration_entry_read(&entry); 7670697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 7680697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 7690697212aSChristoph Lameter } 7701da177e4SLinus Torvalds } 771ae859762SHugh Dickins goto out_set_pte; 7721da177e4SLinus Torvalds } 7731da177e4SLinus Torvalds 7741da177e4SLinus Torvalds /* 7751da177e4SLinus Torvalds * If it's a COW mapping, write protect it both 7761da177e4SLinus Torvalds * in the parent and the child 7771da177e4SLinus Torvalds */ 77867121172SLinus Torvalds if (is_cow_mapping(vm_flags)) { 7791da177e4SLinus Torvalds ptep_set_wrprotect(src_mm, addr, src_pte); 7803dc90795SZachary Amsden pte = pte_wrprotect(pte); 7811da177e4SLinus Torvalds } 7821da177e4SLinus Torvalds 7831da177e4SLinus Torvalds /* 7841da177e4SLinus Torvalds * If it's a shared mapping, mark it clean in 7851da177e4SLinus Torvalds * the child 7861da177e4SLinus Torvalds */ 7871da177e4SLinus Torvalds if (vm_flags & VM_SHARED) 7881da177e4SLinus Torvalds pte = pte_mkclean(pte); 7891da177e4SLinus Torvalds pte = pte_mkold(pte); 7906aab341eSLinus Torvalds 7916aab341eSLinus Torvalds page = vm_normal_page(vma, addr, pte); 7926aab341eSLinus Torvalds if (page) { 7931da177e4SLinus Torvalds get_page(page); 79421333b2bSHugh Dickins page_dup_rmap(page); 795d559db08SKAMEZAWA Hiroyuki if (PageAnon(page)) 796d559db08SKAMEZAWA Hiroyuki rss[MM_ANONPAGES]++; 797d559db08SKAMEZAWA Hiroyuki else 798d559db08SKAMEZAWA Hiroyuki rss[MM_FILEPAGES]++; 7996aab341eSLinus Torvalds } 800ae859762SHugh Dickins 801ae859762SHugh Dickins out_set_pte: 802ae859762SHugh Dickins set_pte_at(dst_mm, addr, dst_pte, pte); 803570a335bSHugh Dickins return 0; 8041da177e4SLinus Torvalds } 8051da177e4SLinus Torvalds 80671e3aac0SAndrea Arcangeli int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 8071da177e4SLinus Torvalds pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, 8081da177e4SLinus Torvalds unsigned long addr, unsigned long end) 8091da177e4SLinus Torvalds { 810c36987e2SDaisuke Nishimura pte_t *orig_src_pte, *orig_dst_pte; 8111da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 812c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 813e040f218SHugh Dickins int progress = 0; 814d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 815570a335bSHugh Dickins swp_entry_t entry = (swp_entry_t){0}; 8161da177e4SLinus Torvalds 8171da177e4SLinus Torvalds again: 818d559db08SKAMEZAWA Hiroyuki init_rss_vec(rss); 819d559db08SKAMEZAWA Hiroyuki 820c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 8211da177e4SLinus Torvalds if (!dst_pte) 8221da177e4SLinus Torvalds return -ENOMEM; 823ece0e2b6SPeter Zijlstra src_pte = pte_offset_map(src_pmd, addr); 8244c21e2f2SHugh Dickins src_ptl = pte_lockptr(src_mm, src_pmd); 825f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 826c36987e2SDaisuke Nishimura orig_src_pte = src_pte; 827c36987e2SDaisuke Nishimura orig_dst_pte = dst_pte; 8286606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 8291da177e4SLinus Torvalds 8301da177e4SLinus Torvalds do { 8311da177e4SLinus Torvalds /* 8321da177e4SLinus Torvalds * We are holding two locks at this point - either of them 8331da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 8341da177e4SLinus Torvalds */ 835e040f218SHugh Dickins if (progress >= 32) { 836e040f218SHugh Dickins progress = 0; 837e040f218SHugh Dickins if (need_resched() || 83895c354feSNick Piggin spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) 8391da177e4SLinus Torvalds break; 840e040f218SHugh Dickins } 8411da177e4SLinus Torvalds if (pte_none(*src_pte)) { 8421da177e4SLinus Torvalds progress++; 8431da177e4SLinus Torvalds continue; 8441da177e4SLinus Torvalds } 845570a335bSHugh Dickins entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, 846570a335bSHugh Dickins vma, addr, rss); 847570a335bSHugh Dickins if (entry.val) 848570a335bSHugh Dickins break; 8491da177e4SLinus Torvalds progress += 8; 8501da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 8511da177e4SLinus Torvalds 8526606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 853c74df32cSHugh Dickins spin_unlock(src_ptl); 854ece0e2b6SPeter Zijlstra pte_unmap(orig_src_pte); 855d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(dst_mm, rss); 856c36987e2SDaisuke Nishimura pte_unmap_unlock(orig_dst_pte, dst_ptl); 857c74df32cSHugh Dickins cond_resched(); 858570a335bSHugh Dickins 859570a335bSHugh Dickins if (entry.val) { 860570a335bSHugh Dickins if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) 861570a335bSHugh Dickins return -ENOMEM; 862570a335bSHugh Dickins progress = 0; 863570a335bSHugh Dickins } 8641da177e4SLinus Torvalds if (addr != end) 8651da177e4SLinus Torvalds goto again; 8661da177e4SLinus Torvalds return 0; 8671da177e4SLinus Torvalds } 8681da177e4SLinus Torvalds 8691da177e4SLinus Torvalds static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 8701da177e4SLinus Torvalds pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, 8711da177e4SLinus Torvalds unsigned long addr, unsigned long end) 8721da177e4SLinus Torvalds { 8731da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 8741da177e4SLinus Torvalds unsigned long next; 8751da177e4SLinus Torvalds 8761da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 8771da177e4SLinus Torvalds if (!dst_pmd) 8781da177e4SLinus Torvalds return -ENOMEM; 8791da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 8801da177e4SLinus Torvalds do { 8811da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 88271e3aac0SAndrea Arcangeli if (pmd_trans_huge(*src_pmd)) { 88371e3aac0SAndrea Arcangeli int err; 88414d1a55cSAndrea Arcangeli VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); 88571e3aac0SAndrea Arcangeli err = copy_huge_pmd(dst_mm, src_mm, 88671e3aac0SAndrea Arcangeli dst_pmd, src_pmd, addr, vma); 88771e3aac0SAndrea Arcangeli if (err == -ENOMEM) 88871e3aac0SAndrea Arcangeli return -ENOMEM; 88971e3aac0SAndrea Arcangeli if (!err) 89071e3aac0SAndrea Arcangeli continue; 89171e3aac0SAndrea Arcangeli /* fall through */ 89271e3aac0SAndrea Arcangeli } 8931da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 8941da177e4SLinus Torvalds continue; 8951da177e4SLinus Torvalds if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, 8961da177e4SLinus Torvalds vma, addr, next)) 8971da177e4SLinus Torvalds return -ENOMEM; 8981da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 8991da177e4SLinus Torvalds return 0; 9001da177e4SLinus Torvalds } 9011da177e4SLinus Torvalds 9021da177e4SLinus Torvalds static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9031da177e4SLinus Torvalds pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, 9041da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9051da177e4SLinus Torvalds { 9061da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 9071da177e4SLinus Torvalds unsigned long next; 9081da177e4SLinus Torvalds 9091da177e4SLinus Torvalds dst_pud = pud_alloc(dst_mm, dst_pgd, addr); 9101da177e4SLinus Torvalds if (!dst_pud) 9111da177e4SLinus Torvalds return -ENOMEM; 9121da177e4SLinus Torvalds src_pud = pud_offset(src_pgd, addr); 9131da177e4SLinus Torvalds do { 9141da177e4SLinus Torvalds next = pud_addr_end(addr, end); 9151da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 9161da177e4SLinus Torvalds continue; 9171da177e4SLinus Torvalds if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, 9181da177e4SLinus Torvalds vma, addr, next)) 9191da177e4SLinus Torvalds return -ENOMEM; 9201da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 9211da177e4SLinus Torvalds return 0; 9221da177e4SLinus Torvalds } 9231da177e4SLinus Torvalds 9241da177e4SLinus Torvalds int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9251da177e4SLinus Torvalds struct vm_area_struct *vma) 9261da177e4SLinus Torvalds { 9271da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 9281da177e4SLinus Torvalds unsigned long next; 9291da177e4SLinus Torvalds unsigned long addr = vma->vm_start; 9301da177e4SLinus Torvalds unsigned long end = vma->vm_end; 931cddb8a5cSAndrea Arcangeli int ret; 9321da177e4SLinus Torvalds 933d992895bSNick Piggin /* 934d992895bSNick Piggin * Don't copy ptes where a page fault will fill them correctly. 935d992895bSNick Piggin * Fork becomes much lighter when there are big shared or private 936d992895bSNick Piggin * readonly mappings. The tradeoff is that copy_page_range is more 937d992895bSNick Piggin * efficient than faulting. 938d992895bSNick Piggin */ 9394d7672b4SLinus Torvalds if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) { 940d992895bSNick Piggin if (!vma->anon_vma) 941d992895bSNick Piggin return 0; 942d992895bSNick Piggin } 943d992895bSNick Piggin 9441da177e4SLinus Torvalds if (is_vm_hugetlb_page(vma)) 9451da177e4SLinus Torvalds return copy_hugetlb_page_range(dst_mm, src_mm, vma); 9461da177e4SLinus Torvalds 94734801ba9Svenkatesh.pallipadi@intel.com if (unlikely(is_pfn_mapping(vma))) { 9482ab64037Svenkatesh.pallipadi@intel.com /* 9492ab64037Svenkatesh.pallipadi@intel.com * We do not free on error cases below as remove_vma 9502ab64037Svenkatesh.pallipadi@intel.com * gets called on error from higher level routine 9512ab64037Svenkatesh.pallipadi@intel.com */ 9522ab64037Svenkatesh.pallipadi@intel.com ret = track_pfn_vma_copy(vma); 9532ab64037Svenkatesh.pallipadi@intel.com if (ret) 9542ab64037Svenkatesh.pallipadi@intel.com return ret; 9552ab64037Svenkatesh.pallipadi@intel.com } 9562ab64037Svenkatesh.pallipadi@intel.com 957cddb8a5cSAndrea Arcangeli /* 958cddb8a5cSAndrea Arcangeli * We need to invalidate the secondary MMU mappings only when 959cddb8a5cSAndrea Arcangeli * there could be a permission downgrade on the ptes of the 960cddb8a5cSAndrea Arcangeli * parent mm. And a permission downgrade will only happen if 961cddb8a5cSAndrea Arcangeli * is_cow_mapping() returns true. 962cddb8a5cSAndrea Arcangeli */ 963cddb8a5cSAndrea Arcangeli if (is_cow_mapping(vma->vm_flags)) 964cddb8a5cSAndrea Arcangeli mmu_notifier_invalidate_range_start(src_mm, addr, end); 965cddb8a5cSAndrea Arcangeli 966cddb8a5cSAndrea Arcangeli ret = 0; 9671da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 9681da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 9691da177e4SLinus Torvalds do { 9701da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 9711da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 9721da177e4SLinus Torvalds continue; 973cddb8a5cSAndrea Arcangeli if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, 974cddb8a5cSAndrea Arcangeli vma, addr, next))) { 975cddb8a5cSAndrea Arcangeli ret = -ENOMEM; 976cddb8a5cSAndrea Arcangeli break; 977cddb8a5cSAndrea Arcangeli } 9781da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 979cddb8a5cSAndrea Arcangeli 980cddb8a5cSAndrea Arcangeli if (is_cow_mapping(vma->vm_flags)) 981cddb8a5cSAndrea Arcangeli mmu_notifier_invalidate_range_end(src_mm, 982cddb8a5cSAndrea Arcangeli vma->vm_start, end); 983cddb8a5cSAndrea Arcangeli return ret; 9841da177e4SLinus Torvalds } 9851da177e4SLinus Torvalds 98651c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 987b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 9881da177e4SLinus Torvalds unsigned long addr, unsigned long end, 98951c6f666SRobin Holt long *zap_work, struct zap_details *details) 9901da177e4SLinus Torvalds { 991b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 992d16dfc55SPeter Zijlstra int force_flush = 0; 9931da177e4SLinus Torvalds pte_t *pte; 994508034a3SHugh Dickins spinlock_t *ptl; 995d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 996d559db08SKAMEZAWA Hiroyuki 997d16dfc55SPeter Zijlstra again: 998*e303297eSPeter Zijlstra init_rss_vec(rss); 999508034a3SHugh Dickins pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 10006606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 10011da177e4SLinus Torvalds do { 10021da177e4SLinus Torvalds pte_t ptent = *pte; 100351c6f666SRobin Holt if (pte_none(ptent)) { 100451c6f666SRobin Holt (*zap_work)--; 10051da177e4SLinus Torvalds continue; 100651c6f666SRobin Holt } 100751c6f666SRobin Holt 100851c6f666SRobin Holt (*zap_work) -= PAGE_SIZE; 100951c6f666SRobin Holt 10106f5e6b9eSHugh Dickins if (pte_present(ptent)) { 10116f5e6b9eSHugh Dickins struct page *page; 10126f5e6b9eSHugh Dickins 10136aab341eSLinus Torvalds page = vm_normal_page(vma, addr, ptent); 10141da177e4SLinus Torvalds if (unlikely(details) && page) { 10151da177e4SLinus Torvalds /* 10161da177e4SLinus Torvalds * unmap_shared_mapping_pages() wants to 10171da177e4SLinus Torvalds * invalidate cache without truncating: 10181da177e4SLinus Torvalds * unmap shared but keep private pages. 10191da177e4SLinus Torvalds */ 10201da177e4SLinus Torvalds if (details->check_mapping && 10211da177e4SLinus Torvalds details->check_mapping != page->mapping) 10221da177e4SLinus Torvalds continue; 10231da177e4SLinus Torvalds /* 10241da177e4SLinus Torvalds * Each page->index must be checked when 10251da177e4SLinus Torvalds * invalidating or truncating nonlinear. 10261da177e4SLinus Torvalds */ 10271da177e4SLinus Torvalds if (details->nonlinear_vma && 10281da177e4SLinus Torvalds (page->index < details->first_index || 10291da177e4SLinus Torvalds page->index > details->last_index)) 10301da177e4SLinus Torvalds continue; 10311da177e4SLinus Torvalds } 1032b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 1033a600388dSZachary Amsden tlb->fullmm); 10341da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 10351da177e4SLinus Torvalds if (unlikely(!page)) 10361da177e4SLinus Torvalds continue; 10371da177e4SLinus Torvalds if (unlikely(details) && details->nonlinear_vma 10381da177e4SLinus Torvalds && linear_page_index(details->nonlinear_vma, 10391da177e4SLinus Torvalds addr) != page->index) 1040b5810039SNick Piggin set_pte_at(mm, addr, pte, 10411da177e4SLinus Torvalds pgoff_to_pte(page->index)); 10421da177e4SLinus Torvalds if (PageAnon(page)) 1043d559db08SKAMEZAWA Hiroyuki rss[MM_ANONPAGES]--; 10446237bcd9SHugh Dickins else { 10456237bcd9SHugh Dickins if (pte_dirty(ptent)) 10466237bcd9SHugh Dickins set_page_dirty(page); 10474917e5d0SJohannes Weiner if (pte_young(ptent) && 10484917e5d0SJohannes Weiner likely(!VM_SequentialReadHint(vma))) 1049bf3f3bc5SNick Piggin mark_page_accessed(page); 1050d559db08SKAMEZAWA Hiroyuki rss[MM_FILEPAGES]--; 10516237bcd9SHugh Dickins } 1052edc315fdSHugh Dickins page_remove_rmap(page); 10533dc14741SHugh Dickins if (unlikely(page_mapcount(page) < 0)) 10543dc14741SHugh Dickins print_bad_pte(vma, addr, ptent, page); 1055d16dfc55SPeter Zijlstra force_flush = !__tlb_remove_page(tlb, page); 1056d16dfc55SPeter Zijlstra if (force_flush) 1057d16dfc55SPeter Zijlstra break; 10581da177e4SLinus Torvalds continue; 10591da177e4SLinus Torvalds } 10601da177e4SLinus Torvalds /* 10611da177e4SLinus Torvalds * If details->check_mapping, we leave swap entries; 10621da177e4SLinus Torvalds * if details->nonlinear_vma, we leave file entries. 10631da177e4SLinus Torvalds */ 10641da177e4SLinus Torvalds if (unlikely(details)) 10651da177e4SLinus Torvalds continue; 10662509ef26SHugh Dickins if (pte_file(ptent)) { 10672509ef26SHugh Dickins if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) 10682509ef26SHugh Dickins print_bad_pte(vma, addr, ptent, NULL); 1069b084d435SKAMEZAWA Hiroyuki } else { 1070b084d435SKAMEZAWA Hiroyuki swp_entry_t entry = pte_to_swp_entry(ptent); 1071b084d435SKAMEZAWA Hiroyuki 1072b084d435SKAMEZAWA Hiroyuki if (!non_swap_entry(entry)) 1073b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 1074b084d435SKAMEZAWA Hiroyuki if (unlikely(!free_swap_and_cache(entry))) 10752509ef26SHugh Dickins print_bad_pte(vma, addr, ptent, NULL); 1076b084d435SKAMEZAWA Hiroyuki } 10779888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 107851c6f666SRobin Holt } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0)); 1079ae859762SHugh Dickins 1080d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 10816606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 1082508034a3SHugh Dickins pte_unmap_unlock(pte - 1, ptl); 108351c6f666SRobin Holt 1084d16dfc55SPeter Zijlstra /* 1085d16dfc55SPeter Zijlstra * mmu_gather ran out of room to batch pages, we break out of 1086d16dfc55SPeter Zijlstra * the PTE lock to avoid doing the potential expensive TLB invalidate 1087d16dfc55SPeter Zijlstra * and page-free while holding it. 1088d16dfc55SPeter Zijlstra */ 1089d16dfc55SPeter Zijlstra if (force_flush) { 1090d16dfc55SPeter Zijlstra force_flush = 0; 1091d16dfc55SPeter Zijlstra tlb_flush_mmu(tlb); 1092d16dfc55SPeter Zijlstra if (addr != end) 1093d16dfc55SPeter Zijlstra goto again; 1094d16dfc55SPeter Zijlstra } 1095d16dfc55SPeter Zijlstra 109651c6f666SRobin Holt return addr; 10971da177e4SLinus Torvalds } 10981da177e4SLinus Torvalds 109951c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1100b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 11011da177e4SLinus Torvalds unsigned long addr, unsigned long end, 110251c6f666SRobin Holt long *zap_work, struct zap_details *details) 11031da177e4SLinus Torvalds { 11041da177e4SLinus Torvalds pmd_t *pmd; 11051da177e4SLinus Torvalds unsigned long next; 11061da177e4SLinus Torvalds 11071da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 11081da177e4SLinus Torvalds do { 11091da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 111071e3aac0SAndrea Arcangeli if (pmd_trans_huge(*pmd)) { 111114d1a55cSAndrea Arcangeli if (next-addr != HPAGE_PMD_SIZE) { 111214d1a55cSAndrea Arcangeli VM_BUG_ON(!rwsem_is_locked(&tlb->mm->mmap_sem)); 111371e3aac0SAndrea Arcangeli split_huge_page_pmd(vma->vm_mm, pmd); 111414d1a55cSAndrea Arcangeli } else if (zap_huge_pmd(tlb, vma, pmd)) { 111571e3aac0SAndrea Arcangeli (*zap_work)--; 111671e3aac0SAndrea Arcangeli continue; 111771e3aac0SAndrea Arcangeli } 111871e3aac0SAndrea Arcangeli /* fall through */ 111971e3aac0SAndrea Arcangeli } 112051c6f666SRobin Holt if (pmd_none_or_clear_bad(pmd)) { 112151c6f666SRobin Holt (*zap_work)--; 11221da177e4SLinus Torvalds continue; 112351c6f666SRobin Holt } 112451c6f666SRobin Holt next = zap_pte_range(tlb, vma, pmd, addr, next, 112551c6f666SRobin Holt zap_work, details); 112651c6f666SRobin Holt } while (pmd++, addr = next, (addr != end && *zap_work > 0)); 112751c6f666SRobin Holt 112851c6f666SRobin Holt return addr; 11291da177e4SLinus Torvalds } 11301da177e4SLinus Torvalds 113151c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1132b5810039SNick Piggin struct vm_area_struct *vma, pgd_t *pgd, 11331da177e4SLinus Torvalds unsigned long addr, unsigned long end, 113451c6f666SRobin Holt long *zap_work, struct zap_details *details) 11351da177e4SLinus Torvalds { 11361da177e4SLinus Torvalds pud_t *pud; 11371da177e4SLinus Torvalds unsigned long next; 11381da177e4SLinus Torvalds 11391da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 11401da177e4SLinus Torvalds do { 11411da177e4SLinus Torvalds next = pud_addr_end(addr, end); 114251c6f666SRobin Holt if (pud_none_or_clear_bad(pud)) { 114351c6f666SRobin Holt (*zap_work)--; 11441da177e4SLinus Torvalds continue; 114551c6f666SRobin Holt } 114651c6f666SRobin Holt next = zap_pmd_range(tlb, vma, pud, addr, next, 114751c6f666SRobin Holt zap_work, details); 114851c6f666SRobin Holt } while (pud++, addr = next, (addr != end && *zap_work > 0)); 114951c6f666SRobin Holt 115051c6f666SRobin Holt return addr; 11511da177e4SLinus Torvalds } 11521da177e4SLinus Torvalds 115351c6f666SRobin Holt static unsigned long unmap_page_range(struct mmu_gather *tlb, 115451c6f666SRobin Holt struct vm_area_struct *vma, 11551da177e4SLinus Torvalds unsigned long addr, unsigned long end, 115651c6f666SRobin Holt long *zap_work, struct zap_details *details) 11571da177e4SLinus Torvalds { 11581da177e4SLinus Torvalds pgd_t *pgd; 11591da177e4SLinus Torvalds unsigned long next; 11601da177e4SLinus Torvalds 11611da177e4SLinus Torvalds if (details && !details->check_mapping && !details->nonlinear_vma) 11621da177e4SLinus Torvalds details = NULL; 11631da177e4SLinus Torvalds 11641da177e4SLinus Torvalds BUG_ON(addr >= end); 1165569b846dSKAMEZAWA Hiroyuki mem_cgroup_uncharge_start(); 11661da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 11671da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 11681da177e4SLinus Torvalds do { 11691da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 117051c6f666SRobin Holt if (pgd_none_or_clear_bad(pgd)) { 117151c6f666SRobin Holt (*zap_work)--; 11721da177e4SLinus Torvalds continue; 117351c6f666SRobin Holt } 117451c6f666SRobin Holt next = zap_pud_range(tlb, vma, pgd, addr, next, 117551c6f666SRobin Holt zap_work, details); 117651c6f666SRobin Holt } while (pgd++, addr = next, (addr != end && *zap_work > 0)); 11771da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 1178569b846dSKAMEZAWA Hiroyuki mem_cgroup_uncharge_end(); 117951c6f666SRobin Holt 118051c6f666SRobin Holt return addr; 11811da177e4SLinus Torvalds } 11821da177e4SLinus Torvalds 11831da177e4SLinus Torvalds #ifdef CONFIG_PREEMPT 11841da177e4SLinus Torvalds # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE) 11851da177e4SLinus Torvalds #else 11861da177e4SLinus Torvalds /* No preempt: go for improved straight-line efficiency */ 11871da177e4SLinus Torvalds # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE) 11881da177e4SLinus Torvalds #endif 11891da177e4SLinus Torvalds 11901da177e4SLinus Torvalds /** 11911da177e4SLinus Torvalds * unmap_vmas - unmap a range of memory covered by a list of vma's 11921da177e4SLinus Torvalds * @tlbp: address of the caller's struct mmu_gather 11931da177e4SLinus Torvalds * @vma: the starting vma 11941da177e4SLinus Torvalds * @start_addr: virtual address at which to start unmapping 11951da177e4SLinus Torvalds * @end_addr: virtual address at which to end unmapping 11961da177e4SLinus Torvalds * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here 11971da177e4SLinus Torvalds * @details: details of nonlinear truncation or shared cache invalidation 11981da177e4SLinus Torvalds * 1199ee39b37bSHugh Dickins * Returns the end address of the unmapping (restart addr if interrupted). 12001da177e4SLinus Torvalds * 1201508034a3SHugh Dickins * Unmap all pages in the vma list. 12021da177e4SLinus Torvalds * 1203508034a3SHugh Dickins * We aim to not hold locks for too long (for scheduling latency reasons). 1204508034a3SHugh Dickins * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to 12051da177e4SLinus Torvalds * return the ending mmu_gather to the caller. 12061da177e4SLinus Torvalds * 12071da177e4SLinus Torvalds * Only addresses between `start' and `end' will be unmapped. 12081da177e4SLinus Torvalds * 12091da177e4SLinus Torvalds * The VMA list must be sorted in ascending virtual address order. 12101da177e4SLinus Torvalds * 12111da177e4SLinus Torvalds * unmap_vmas() assumes that the caller will flush the whole unmapped address 12121da177e4SLinus Torvalds * range after unmap_vmas() returns. So the only responsibility here is to 12131da177e4SLinus Torvalds * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 12141da177e4SLinus Torvalds * drops the lock and schedules. 12151da177e4SLinus Torvalds */ 1216d16dfc55SPeter Zijlstra unsigned long unmap_vmas(struct mmu_gather *tlb, 12171da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 12181da177e4SLinus Torvalds unsigned long end_addr, unsigned long *nr_accounted, 12191da177e4SLinus Torvalds struct zap_details *details) 12201da177e4SLinus Torvalds { 122151c6f666SRobin Holt long zap_work = ZAP_BLOCK_SIZE; 1222ee39b37bSHugh Dickins unsigned long start = start_addr; 12231da177e4SLinus Torvalds spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL; 1224cddb8a5cSAndrea Arcangeli struct mm_struct *mm = vma->vm_mm; 12251da177e4SLinus Torvalds 1226cddb8a5cSAndrea Arcangeli mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); 12271da177e4SLinus Torvalds for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) { 12281da177e4SLinus Torvalds unsigned long end; 12291da177e4SLinus Torvalds 12301da177e4SLinus Torvalds start = max(vma->vm_start, start_addr); 12311da177e4SLinus Torvalds if (start >= vma->vm_end) 12321da177e4SLinus Torvalds continue; 12331da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 12341da177e4SLinus Torvalds if (end <= vma->vm_start) 12351da177e4SLinus Torvalds continue; 12361da177e4SLinus Torvalds 12371da177e4SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 12381da177e4SLinus Torvalds *nr_accounted += (end - start) >> PAGE_SHIFT; 12391da177e4SLinus Torvalds 124034801ba9Svenkatesh.pallipadi@intel.com if (unlikely(is_pfn_mapping(vma))) 12412ab64037Svenkatesh.pallipadi@intel.com untrack_pfn_vma(vma, 0, 0); 12422ab64037Svenkatesh.pallipadi@intel.com 12431da177e4SLinus Torvalds while (start != end) { 124451c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1245a137e1ccSAndi Kleen /* 1246a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1247a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1248a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 1249a137e1ccSAndi Kleen * cleanup path of do_mmap_pgoff. When 1250a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 1251a137e1ccSAndi Kleen * do_mmap_pgoff() nullifies vma->vm_file 1252a137e1ccSAndi Kleen * before calling this function to clean up. 1253a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1254a137e1ccSAndi Kleen * safe to do nothing in this case. 1255a137e1ccSAndi Kleen */ 1256a137e1ccSAndi Kleen if (vma->vm_file) { 125704f2cbe3SMel Gorman unmap_hugepage_range(vma, start, end, NULL); 125851c6f666SRobin Holt zap_work -= (end - start) / 1259a5516438SAndi Kleen pages_per_huge_page(hstate_vma(vma)); 1260a137e1ccSAndi Kleen } 1261a137e1ccSAndi Kleen 126251c6f666SRobin Holt start = end; 126351c6f666SRobin Holt } else 1264d16dfc55SPeter Zijlstra start = unmap_page_range(tlb, vma, 126551c6f666SRobin Holt start, end, &zap_work, details); 12661da177e4SLinus Torvalds 126751c6f666SRobin Holt if (zap_work > 0) { 126851c6f666SRobin Holt BUG_ON(start != end); 126951c6f666SRobin Holt break; 127051c6f666SRobin Holt } 12711da177e4SLinus Torvalds 12721da177e4SLinus Torvalds if (need_resched() || 127395c354feSNick Piggin (i_mmap_lock && spin_needbreak(i_mmap_lock))) { 1274d16dfc55SPeter Zijlstra if (i_mmap_lock) 12751da177e4SLinus Torvalds goto out; 12761da177e4SLinus Torvalds cond_resched(); 12771da177e4SLinus Torvalds } 12781da177e4SLinus Torvalds 127951c6f666SRobin Holt zap_work = ZAP_BLOCK_SIZE; 12801da177e4SLinus Torvalds } 12811da177e4SLinus Torvalds } 12821da177e4SLinus Torvalds out: 1283cddb8a5cSAndrea Arcangeli mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); 1284ee39b37bSHugh Dickins return start; /* which is now the end (or restart) address */ 12851da177e4SLinus Torvalds } 12861da177e4SLinus Torvalds 12871da177e4SLinus Torvalds /** 12881da177e4SLinus Torvalds * zap_page_range - remove user pages in a given range 12891da177e4SLinus Torvalds * @vma: vm_area_struct holding the applicable pages 12901da177e4SLinus Torvalds * @address: starting address of pages to zap 12911da177e4SLinus Torvalds * @size: number of bytes to zap 12921da177e4SLinus Torvalds * @details: details of nonlinear truncation or shared cache invalidation 12931da177e4SLinus Torvalds */ 1294ee39b37bSHugh Dickins unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address, 12951da177e4SLinus Torvalds unsigned long size, struct zap_details *details) 12961da177e4SLinus Torvalds { 12971da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 1298d16dfc55SPeter Zijlstra struct mmu_gather tlb; 12991da177e4SLinus Torvalds unsigned long end = address + size; 13001da177e4SLinus Torvalds unsigned long nr_accounted = 0; 13011da177e4SLinus Torvalds 13021da177e4SLinus Torvalds lru_add_drain(); 1303d16dfc55SPeter Zijlstra tlb_gather_mmu(&tlb, mm, 0); 1304365e9c87SHugh Dickins update_hiwater_rss(mm); 1305508034a3SHugh Dickins end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details); 1306d16dfc55SPeter Zijlstra tlb_finish_mmu(&tlb, address, end); 1307ee39b37bSHugh Dickins return end; 13081da177e4SLinus Torvalds } 13091da177e4SLinus Torvalds 1310c627f9ccSJack Steiner /** 1311c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1312c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1313c627f9ccSJack Steiner * @address: starting address of pages to zap 1314c627f9ccSJack Steiner * @size: number of bytes to zap 1315c627f9ccSJack Steiner * 1316c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1317c627f9ccSJack Steiner * 1318c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1319c627f9ccSJack Steiner * 1320c627f9ccSJack Steiner * Returns 0 if successful. 1321c627f9ccSJack Steiner */ 1322c627f9ccSJack Steiner int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1323c627f9ccSJack Steiner unsigned long size) 1324c627f9ccSJack Steiner { 1325c627f9ccSJack Steiner if (address < vma->vm_start || address + size > vma->vm_end || 1326c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 1327c627f9ccSJack Steiner return -1; 1328c627f9ccSJack Steiner zap_page_range(vma, address, size, NULL); 1329c627f9ccSJack Steiner return 0; 1330c627f9ccSJack Steiner } 1331c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1332c627f9ccSJack Steiner 1333142762bdSJohannes Weiner /** 1334142762bdSJohannes Weiner * follow_page - look up a page descriptor from a user-virtual address 1335142762bdSJohannes Weiner * @vma: vm_area_struct mapping @address 1336142762bdSJohannes Weiner * @address: virtual address to look up 1337142762bdSJohannes Weiner * @flags: flags modifying lookup behaviour 1338142762bdSJohannes Weiner * 1339142762bdSJohannes Weiner * @flags can have FOLL_ flags set, defined in <linux/mm.h> 1340142762bdSJohannes Weiner * 1341142762bdSJohannes Weiner * Returns the mapped (struct page *), %NULL if no mapping exists, or 1342142762bdSJohannes Weiner * an error pointer if there is a mapping to something not represented 1343142762bdSJohannes Weiner * by a page descriptor (see also vm_normal_page()). 13441da177e4SLinus Torvalds */ 13456aab341eSLinus Torvalds struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 1346deceb6cdSHugh Dickins unsigned int flags) 13471da177e4SLinus Torvalds { 13481da177e4SLinus Torvalds pgd_t *pgd; 13491da177e4SLinus Torvalds pud_t *pud; 13501da177e4SLinus Torvalds pmd_t *pmd; 13511da177e4SLinus Torvalds pte_t *ptep, pte; 1352deceb6cdSHugh Dickins spinlock_t *ptl; 13531da177e4SLinus Torvalds struct page *page; 13546aab341eSLinus Torvalds struct mm_struct *mm = vma->vm_mm; 13551da177e4SLinus Torvalds 1356deceb6cdSHugh Dickins page = follow_huge_addr(mm, address, flags & FOLL_WRITE); 1357deceb6cdSHugh Dickins if (!IS_ERR(page)) { 1358deceb6cdSHugh Dickins BUG_ON(flags & FOLL_GET); 1359deceb6cdSHugh Dickins goto out; 1360deceb6cdSHugh Dickins } 13611da177e4SLinus Torvalds 1362deceb6cdSHugh Dickins page = NULL; 13631da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 13641da177e4SLinus Torvalds if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 1365deceb6cdSHugh Dickins goto no_page_table; 13661da177e4SLinus Torvalds 13671da177e4SLinus Torvalds pud = pud_offset(pgd, address); 1368ceb86879SAndi Kleen if (pud_none(*pud)) 1369ceb86879SAndi Kleen goto no_page_table; 13708a07651eSHugh Dickins if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) { 1371ceb86879SAndi Kleen BUG_ON(flags & FOLL_GET); 1372ceb86879SAndi Kleen page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE); 1373ceb86879SAndi Kleen goto out; 1374ceb86879SAndi Kleen } 1375ceb86879SAndi Kleen if (unlikely(pud_bad(*pud))) 1376deceb6cdSHugh Dickins goto no_page_table; 13771da177e4SLinus Torvalds 13781da177e4SLinus Torvalds pmd = pmd_offset(pud, address); 1379aeed5fceSHugh Dickins if (pmd_none(*pmd)) 1380deceb6cdSHugh Dickins goto no_page_table; 138171e3aac0SAndrea Arcangeli if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) { 1382deceb6cdSHugh Dickins BUG_ON(flags & FOLL_GET); 1383deceb6cdSHugh Dickins page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); 1384deceb6cdSHugh Dickins goto out; 1385deceb6cdSHugh Dickins } 138671e3aac0SAndrea Arcangeli if (pmd_trans_huge(*pmd)) { 1387500d65d4SAndrea Arcangeli if (flags & FOLL_SPLIT) { 1388500d65d4SAndrea Arcangeli split_huge_page_pmd(mm, pmd); 1389500d65d4SAndrea Arcangeli goto split_fallthrough; 1390500d65d4SAndrea Arcangeli } 139171e3aac0SAndrea Arcangeli spin_lock(&mm->page_table_lock); 139271e3aac0SAndrea Arcangeli if (likely(pmd_trans_huge(*pmd))) { 139371e3aac0SAndrea Arcangeli if (unlikely(pmd_trans_splitting(*pmd))) { 139471e3aac0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 139571e3aac0SAndrea Arcangeli wait_split_huge_page(vma->anon_vma, pmd); 139671e3aac0SAndrea Arcangeli } else { 139771e3aac0SAndrea Arcangeli page = follow_trans_huge_pmd(mm, address, 139871e3aac0SAndrea Arcangeli pmd, flags); 139971e3aac0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 140071e3aac0SAndrea Arcangeli goto out; 140171e3aac0SAndrea Arcangeli } 140271e3aac0SAndrea Arcangeli } else 140371e3aac0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 140471e3aac0SAndrea Arcangeli /* fall through */ 140571e3aac0SAndrea Arcangeli } 1406500d65d4SAndrea Arcangeli split_fallthrough: 1407aeed5fceSHugh Dickins if (unlikely(pmd_bad(*pmd))) 1408aeed5fceSHugh Dickins goto no_page_table; 1409aeed5fceSHugh Dickins 1410deceb6cdSHugh Dickins ptep = pte_offset_map_lock(mm, pmd, address, &ptl); 14111da177e4SLinus Torvalds 14121da177e4SLinus Torvalds pte = *ptep; 1413deceb6cdSHugh Dickins if (!pte_present(pte)) 141489f5b7daSLinus Torvalds goto no_page; 1415deceb6cdSHugh Dickins if ((flags & FOLL_WRITE) && !pte_write(pte)) 1416deceb6cdSHugh Dickins goto unlock; 1417a13ea5b7SHugh Dickins 14186aab341eSLinus Torvalds page = vm_normal_page(vma, address, pte); 1419a13ea5b7SHugh Dickins if (unlikely(!page)) { 1420a13ea5b7SHugh Dickins if ((flags & FOLL_DUMP) || 142162eede62SHugh Dickins !is_zero_pfn(pte_pfn(pte))) 142289f5b7daSLinus Torvalds goto bad_page; 1423a13ea5b7SHugh Dickins page = pte_page(pte); 1424a13ea5b7SHugh Dickins } 1425deceb6cdSHugh Dickins 1426deceb6cdSHugh Dickins if (flags & FOLL_GET) 1427deceb6cdSHugh Dickins get_page(page); 1428deceb6cdSHugh Dickins if (flags & FOLL_TOUCH) { 1429deceb6cdSHugh Dickins if ((flags & FOLL_WRITE) && 1430deceb6cdSHugh Dickins !pte_dirty(pte) && !PageDirty(page)) 1431f33ea7f4SNick Piggin set_page_dirty(page); 1432bd775c42SKOSAKI Motohiro /* 1433bd775c42SKOSAKI Motohiro * pte_mkyoung() would be more correct here, but atomic care 1434bd775c42SKOSAKI Motohiro * is needed to avoid losing the dirty bit: it is easier to use 1435bd775c42SKOSAKI Motohiro * mark_page_accessed(). 1436bd775c42SKOSAKI Motohiro */ 14371da177e4SLinus Torvalds mark_page_accessed(page); 14381da177e4SLinus Torvalds } 1439a1fde08cSLinus Torvalds if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) { 1440110d74a9SMichel Lespinasse /* 1441110d74a9SMichel Lespinasse * The preliminary mapping check is mainly to avoid the 1442110d74a9SMichel Lespinasse * pointless overhead of lock_page on the ZERO_PAGE 1443110d74a9SMichel Lespinasse * which might bounce very badly if there is contention. 1444110d74a9SMichel Lespinasse * 1445110d74a9SMichel Lespinasse * If the page is already locked, we don't need to 1446110d74a9SMichel Lespinasse * handle it now - vmscan will handle it later if and 1447110d74a9SMichel Lespinasse * when it attempts to reclaim the page. 1448110d74a9SMichel Lespinasse */ 1449110d74a9SMichel Lespinasse if (page->mapping && trylock_page(page)) { 1450110d74a9SMichel Lespinasse lru_add_drain(); /* push cached pages to LRU */ 1451110d74a9SMichel Lespinasse /* 1452110d74a9SMichel Lespinasse * Because we lock page here and migration is 1453110d74a9SMichel Lespinasse * blocked by the pte's page reference, we need 1454110d74a9SMichel Lespinasse * only check for file-cache page truncation. 1455110d74a9SMichel Lespinasse */ 1456110d74a9SMichel Lespinasse if (page->mapping) 1457110d74a9SMichel Lespinasse mlock_vma_page(page); 1458110d74a9SMichel Lespinasse unlock_page(page); 1459110d74a9SMichel Lespinasse } 1460110d74a9SMichel Lespinasse } 1461deceb6cdSHugh Dickins unlock: 1462deceb6cdSHugh Dickins pte_unmap_unlock(ptep, ptl); 14631da177e4SLinus Torvalds out: 1464deceb6cdSHugh Dickins return page; 1465deceb6cdSHugh Dickins 146689f5b7daSLinus Torvalds bad_page: 146789f5b7daSLinus Torvalds pte_unmap_unlock(ptep, ptl); 146889f5b7daSLinus Torvalds return ERR_PTR(-EFAULT); 146989f5b7daSLinus Torvalds 147089f5b7daSLinus Torvalds no_page: 147189f5b7daSLinus Torvalds pte_unmap_unlock(ptep, ptl); 147289f5b7daSLinus Torvalds if (!pte_none(pte)) 147389f5b7daSLinus Torvalds return page; 14748e4b9a60SHugh Dickins 1475deceb6cdSHugh Dickins no_page_table: 1476deceb6cdSHugh Dickins /* 1477deceb6cdSHugh Dickins * When core dumping an enormous anonymous area that nobody 14788e4b9a60SHugh Dickins * has touched so far, we don't want to allocate unnecessary pages or 14798e4b9a60SHugh Dickins * page tables. Return error instead of NULL to skip handle_mm_fault, 14808e4b9a60SHugh Dickins * then get_dump_page() will return NULL to leave a hole in the dump. 14818e4b9a60SHugh Dickins * But we can only make this optimization where a hole would surely 14828e4b9a60SHugh Dickins * be zero-filled if handle_mm_fault() actually did handle it. 1483deceb6cdSHugh Dickins */ 14848e4b9a60SHugh Dickins if ((flags & FOLL_DUMP) && 14858e4b9a60SHugh Dickins (!vma->vm_ops || !vma->vm_ops->fault)) 14868e4b9a60SHugh Dickins return ERR_PTR(-EFAULT); 1487deceb6cdSHugh Dickins return page; 14881da177e4SLinus Torvalds } 14891da177e4SLinus Torvalds 149095042f9eSLinus Torvalds static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr) 149195042f9eSLinus Torvalds { 1492a09a79f6SMikulas Patocka return stack_guard_page_start(vma, addr) || 1493a09a79f6SMikulas Patocka stack_guard_page_end(vma, addr+PAGE_SIZE); 149495042f9eSLinus Torvalds } 149595042f9eSLinus Torvalds 14960014bd99SHuang Ying /** 14970014bd99SHuang Ying * __get_user_pages() - pin user pages in memory 14980014bd99SHuang Ying * @tsk: task_struct of target task 14990014bd99SHuang Ying * @mm: mm_struct of target mm 15000014bd99SHuang Ying * @start: starting user address 15010014bd99SHuang Ying * @nr_pages: number of pages from start to pin 15020014bd99SHuang Ying * @gup_flags: flags modifying pin behaviour 15030014bd99SHuang Ying * @pages: array that receives pointers to the pages pinned. 15040014bd99SHuang Ying * Should be at least nr_pages long. Or NULL, if caller 15050014bd99SHuang Ying * only intends to ensure the pages are faulted in. 15060014bd99SHuang Ying * @vmas: array of pointers to vmas corresponding to each page. 15070014bd99SHuang Ying * Or NULL if the caller does not require them. 15080014bd99SHuang Ying * @nonblocking: whether waiting for disk IO or mmap_sem contention 15090014bd99SHuang Ying * 15100014bd99SHuang Ying * Returns number of pages pinned. This may be fewer than the number 15110014bd99SHuang Ying * requested. If nr_pages is 0 or negative, returns 0. If no pages 15120014bd99SHuang Ying * were pinned, returns -errno. Each page returned must be released 15130014bd99SHuang Ying * with a put_page() call when it is finished with. vmas will only 15140014bd99SHuang Ying * remain valid while mmap_sem is held. 15150014bd99SHuang Ying * 15160014bd99SHuang Ying * Must be called with mmap_sem held for read or write. 15170014bd99SHuang Ying * 15180014bd99SHuang Ying * __get_user_pages walks a process's page tables and takes a reference to 15190014bd99SHuang Ying * each struct page that each user address corresponds to at a given 15200014bd99SHuang Ying * instant. That is, it takes the page that would be accessed if a user 15210014bd99SHuang Ying * thread accesses the given user virtual address at that instant. 15220014bd99SHuang Ying * 15230014bd99SHuang Ying * This does not guarantee that the page exists in the user mappings when 15240014bd99SHuang Ying * __get_user_pages returns, and there may even be a completely different 15250014bd99SHuang Ying * page there in some cases (eg. if mmapped pagecache has been invalidated 15260014bd99SHuang Ying * and subsequently re faulted). However it does guarantee that the page 15270014bd99SHuang Ying * won't be freed completely. And mostly callers simply care that the page 15280014bd99SHuang Ying * contains data that was valid *at some point in time*. Typically, an IO 15290014bd99SHuang Ying * or similar operation cannot guarantee anything stronger anyway because 15300014bd99SHuang Ying * locks can't be held over the syscall boundary. 15310014bd99SHuang Ying * 15320014bd99SHuang Ying * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If 15330014bd99SHuang Ying * the page is written to, set_page_dirty (or set_page_dirty_lock, as 15340014bd99SHuang Ying * appropriate) must be called after the page is finished with, and 15350014bd99SHuang Ying * before put_page is called. 15360014bd99SHuang Ying * 15370014bd99SHuang Ying * If @nonblocking != NULL, __get_user_pages will not wait for disk IO 15380014bd99SHuang Ying * or mmap_sem contention, and if waiting is needed to pin all pages, 15390014bd99SHuang Ying * *@nonblocking will be set to 0. 15400014bd99SHuang Ying * 15410014bd99SHuang Ying * In most cases, get_user_pages or get_user_pages_fast should be used 15420014bd99SHuang Ying * instead of __get_user_pages. __get_user_pages should be used only if 15430014bd99SHuang Ying * you need some special @gup_flags. 15440014bd99SHuang Ying */ 1545b291f000SNick Piggin int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 154658fa879eSHugh Dickins unsigned long start, int nr_pages, unsigned int gup_flags, 154753a7706dSMichel Lespinasse struct page **pages, struct vm_area_struct **vmas, 154853a7706dSMichel Lespinasse int *nonblocking) 15491da177e4SLinus Torvalds { 15501da177e4SLinus Torvalds int i; 155158fa879eSHugh Dickins unsigned long vm_flags; 15521da177e4SLinus Torvalds 15539d73777eSPeter Zijlstra if (nr_pages <= 0) 1554900cf086SJonathan Corbet return 0; 155558fa879eSHugh Dickins 155658fa879eSHugh Dickins VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET)); 155758fa879eSHugh Dickins 15581da177e4SLinus Torvalds /* 15591da177e4SLinus Torvalds * Require read or write permissions. 156058fa879eSHugh Dickins * If FOLL_FORCE is set, we only require the "MAY" flags. 15611da177e4SLinus Torvalds */ 156258fa879eSHugh Dickins vm_flags = (gup_flags & FOLL_WRITE) ? 156358fa879eSHugh Dickins (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 156458fa879eSHugh Dickins vm_flags &= (gup_flags & FOLL_FORCE) ? 156558fa879eSHugh Dickins (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 15661da177e4SLinus Torvalds i = 0; 15671da177e4SLinus Torvalds 15681da177e4SLinus Torvalds do { 15691da177e4SLinus Torvalds struct vm_area_struct *vma; 15701da177e4SLinus Torvalds 15711da177e4SLinus Torvalds vma = find_extend_vma(mm, start); 1572e7f22e20SStephen Wilson if (!vma && in_gate_area(mm, start)) { 15731da177e4SLinus Torvalds unsigned long pg = start & PAGE_MASK; 15741da177e4SLinus Torvalds pgd_t *pgd; 15751da177e4SLinus Torvalds pud_t *pud; 15761da177e4SLinus Torvalds pmd_t *pmd; 15771da177e4SLinus Torvalds pte_t *pte; 1578b291f000SNick Piggin 1579b291f000SNick Piggin /* user gate pages are read-only */ 158058fa879eSHugh Dickins if (gup_flags & FOLL_WRITE) 15811da177e4SLinus Torvalds return i ? : -EFAULT; 15821da177e4SLinus Torvalds if (pg > TASK_SIZE) 15831da177e4SLinus Torvalds pgd = pgd_offset_k(pg); 15841da177e4SLinus Torvalds else 15851da177e4SLinus Torvalds pgd = pgd_offset_gate(mm, pg); 15861da177e4SLinus Torvalds BUG_ON(pgd_none(*pgd)); 15871da177e4SLinus Torvalds pud = pud_offset(pgd, pg); 15881da177e4SLinus Torvalds BUG_ON(pud_none(*pud)); 15891da177e4SLinus Torvalds pmd = pmd_offset(pud, pg); 1590690dbe1cSHugh Dickins if (pmd_none(*pmd)) 1591690dbe1cSHugh Dickins return i ? : -EFAULT; 1592f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 15931da177e4SLinus Torvalds pte = pte_offset_map(pmd, pg); 1594690dbe1cSHugh Dickins if (pte_none(*pte)) { 1595690dbe1cSHugh Dickins pte_unmap(pte); 1596690dbe1cSHugh Dickins return i ? : -EFAULT; 1597690dbe1cSHugh Dickins } 159895042f9eSLinus Torvalds vma = get_gate_vma(mm); 15991da177e4SLinus Torvalds if (pages) { 1600de51257aSHugh Dickins struct page *page; 1601de51257aSHugh Dickins 160295042f9eSLinus Torvalds page = vm_normal_page(vma, start, *pte); 1603de51257aSHugh Dickins if (!page) { 1604de51257aSHugh Dickins if (!(gup_flags & FOLL_DUMP) && 1605de51257aSHugh Dickins is_zero_pfn(pte_pfn(*pte))) 1606de51257aSHugh Dickins page = pte_page(*pte); 1607de51257aSHugh Dickins else { 1608de51257aSHugh Dickins pte_unmap(pte); 1609de51257aSHugh Dickins return i ? : -EFAULT; 1610de51257aSHugh Dickins } 1611de51257aSHugh Dickins } 16126aab341eSLinus Torvalds pages[i] = page; 16136aab341eSLinus Torvalds get_page(page); 16141da177e4SLinus Torvalds } 16151da177e4SLinus Torvalds pte_unmap(pte); 161695042f9eSLinus Torvalds goto next_page; 16171da177e4SLinus Torvalds } 16181da177e4SLinus Torvalds 1619b291f000SNick Piggin if (!vma || 1620b291f000SNick Piggin (vma->vm_flags & (VM_IO | VM_PFNMAP)) || 16211c3aff1cSHugh Dickins !(vm_flags & vma->vm_flags)) 16221da177e4SLinus Torvalds return i ? : -EFAULT; 16231da177e4SLinus Torvalds 16242a15efc9SHugh Dickins if (is_vm_hugetlb_page(vma)) { 16252a15efc9SHugh Dickins i = follow_hugetlb_page(mm, vma, pages, vmas, 162658fa879eSHugh Dickins &start, &nr_pages, i, gup_flags); 16272a15efc9SHugh Dickins continue; 16282a15efc9SHugh Dickins } 1629deceb6cdSHugh Dickins 16301da177e4SLinus Torvalds do { 163108ef4729SHugh Dickins struct page *page; 163258fa879eSHugh Dickins unsigned int foll_flags = gup_flags; 16331da177e4SLinus Torvalds 1634462e00ccSEthan Solomita /* 16354779280dSYing Han * If we have a pending SIGKILL, don't keep faulting 16361c3aff1cSHugh Dickins * pages and potentially allocating memory. 1637462e00ccSEthan Solomita */ 16381c3aff1cSHugh Dickins if (unlikely(fatal_signal_pending(current))) 16394779280dSYing Han return i ? i : -ERESTARTSYS; 1640462e00ccSEthan Solomita 1641deceb6cdSHugh Dickins cond_resched(); 16426aab341eSLinus Torvalds while (!(page = follow_page(vma, start, foll_flags))) { 1643a68d2ebcSLinus Torvalds int ret; 164453a7706dSMichel Lespinasse unsigned int fault_flags = 0; 164553a7706dSMichel Lespinasse 1646a09a79f6SMikulas Patocka /* For mlock, just skip the stack guard page. */ 1647a09a79f6SMikulas Patocka if (foll_flags & FOLL_MLOCK) { 1648a09a79f6SMikulas Patocka if (stack_guard_page(vma, start)) 1649a09a79f6SMikulas Patocka goto next_page; 1650a09a79f6SMikulas Patocka } 165153a7706dSMichel Lespinasse if (foll_flags & FOLL_WRITE) 165253a7706dSMichel Lespinasse fault_flags |= FAULT_FLAG_WRITE; 165353a7706dSMichel Lespinasse if (nonblocking) 165453a7706dSMichel Lespinasse fault_flags |= FAULT_FLAG_ALLOW_RETRY; 1655318b275fSGleb Natapov if (foll_flags & FOLL_NOWAIT) 1656318b275fSGleb Natapov fault_flags |= (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT); 1657d06063ccSLinus Torvalds 1658d26ed650SHugh Dickins ret = handle_mm_fault(mm, vma, start, 165953a7706dSMichel Lespinasse fault_flags); 1660d26ed650SHugh Dickins 166183c54070SNick Piggin if (ret & VM_FAULT_ERROR) { 166283c54070SNick Piggin if (ret & VM_FAULT_OOM) 166383c54070SNick Piggin return i ? i : -ENOMEM; 166469ebb83eSHuang Ying if (ret & (VM_FAULT_HWPOISON | 166569ebb83eSHuang Ying VM_FAULT_HWPOISON_LARGE)) { 166669ebb83eSHuang Ying if (i) 166769ebb83eSHuang Ying return i; 166869ebb83eSHuang Ying else if (gup_flags & FOLL_HWPOISON) 166969ebb83eSHuang Ying return -EHWPOISON; 167069ebb83eSHuang Ying else 167169ebb83eSHuang Ying return -EFAULT; 167269ebb83eSHuang Ying } 167369ebb83eSHuang Ying if (ret & VM_FAULT_SIGBUS) 167483c54070SNick Piggin return i ? i : -EFAULT; 167583c54070SNick Piggin BUG(); 167683c54070SNick Piggin } 1677e7f22e20SStephen Wilson 1678e7f22e20SStephen Wilson if (tsk) { 167983c54070SNick Piggin if (ret & VM_FAULT_MAJOR) 168083c54070SNick Piggin tsk->maj_flt++; 168183c54070SNick Piggin else 168283c54070SNick Piggin tsk->min_flt++; 1683e7f22e20SStephen Wilson } 168483c54070SNick Piggin 168553a7706dSMichel Lespinasse if (ret & VM_FAULT_RETRY) { 1686318b275fSGleb Natapov if (nonblocking) 168753a7706dSMichel Lespinasse *nonblocking = 0; 168853a7706dSMichel Lespinasse return i; 168953a7706dSMichel Lespinasse } 169053a7706dSMichel Lespinasse 1691f33ea7f4SNick Piggin /* 169283c54070SNick Piggin * The VM_FAULT_WRITE bit tells us that 169383c54070SNick Piggin * do_wp_page has broken COW when necessary, 169483c54070SNick Piggin * even if maybe_mkwrite decided not to set 169583c54070SNick Piggin * pte_write. We can thus safely do subsequent 1696878b63acSHugh Dickins * page lookups as if they were reads. But only 1697878b63acSHugh Dickins * do so when looping for pte_write is futile: 1698878b63acSHugh Dickins * in some cases userspace may also be wanting 1699878b63acSHugh Dickins * to write to the gotten user page, which a 1700878b63acSHugh Dickins * read fault here might prevent (a readonly 1701878b63acSHugh Dickins * page might get reCOWed by userspace write). 1702f33ea7f4SNick Piggin */ 1703878b63acSHugh Dickins if ((ret & VM_FAULT_WRITE) && 1704878b63acSHugh Dickins !(vma->vm_flags & VM_WRITE)) 1705deceb6cdSHugh Dickins foll_flags &= ~FOLL_WRITE; 1706a68d2ebcSLinus Torvalds 17077f7bbbe5SBenjamin Herrenschmidt cond_resched(); 17081da177e4SLinus Torvalds } 170989f5b7daSLinus Torvalds if (IS_ERR(page)) 171089f5b7daSLinus Torvalds return i ? i : PTR_ERR(page); 17111da177e4SLinus Torvalds if (pages) { 171208ef4729SHugh Dickins pages[i] = page; 171303beb076SJames Bottomley 1714a6f36be3SRussell King flush_anon_page(vma, page, start); 171508ef4729SHugh Dickins flush_dcache_page(page); 17161da177e4SLinus Torvalds } 171795042f9eSLinus Torvalds next_page: 17181da177e4SLinus Torvalds if (vmas) 17191da177e4SLinus Torvalds vmas[i] = vma; 17201da177e4SLinus Torvalds i++; 17211da177e4SLinus Torvalds start += PAGE_SIZE; 17229d73777eSPeter Zijlstra nr_pages--; 17239d73777eSPeter Zijlstra } while (nr_pages && start < vma->vm_end); 17249d73777eSPeter Zijlstra } while (nr_pages); 17251da177e4SLinus Torvalds return i; 17261da177e4SLinus Torvalds } 17270014bd99SHuang Ying EXPORT_SYMBOL(__get_user_pages); 1728b291f000SNick Piggin 1729d2bf6be8SNick Piggin /** 1730d2bf6be8SNick Piggin * get_user_pages() - pin user pages in memory 1731e7f22e20SStephen Wilson * @tsk: the task_struct to use for page fault accounting, or 1732e7f22e20SStephen Wilson * NULL if faults are not to be recorded. 1733d2bf6be8SNick Piggin * @mm: mm_struct of target mm 1734d2bf6be8SNick Piggin * @start: starting user address 17359d73777eSPeter Zijlstra * @nr_pages: number of pages from start to pin 1736d2bf6be8SNick Piggin * @write: whether pages will be written to by the caller 1737d2bf6be8SNick Piggin * @force: whether to force write access even if user mapping is 1738d2bf6be8SNick Piggin * readonly. This will result in the page being COWed even 1739d2bf6be8SNick Piggin * in MAP_SHARED mappings. You do not want this. 1740d2bf6be8SNick Piggin * @pages: array that receives pointers to the pages pinned. 1741d2bf6be8SNick Piggin * Should be at least nr_pages long. Or NULL, if caller 1742d2bf6be8SNick Piggin * only intends to ensure the pages are faulted in. 1743d2bf6be8SNick Piggin * @vmas: array of pointers to vmas corresponding to each page. 1744d2bf6be8SNick Piggin * Or NULL if the caller does not require them. 1745d2bf6be8SNick Piggin * 1746d2bf6be8SNick Piggin * Returns number of pages pinned. This may be fewer than the number 17479d73777eSPeter Zijlstra * requested. If nr_pages is 0 or negative, returns 0. If no pages 1748d2bf6be8SNick Piggin * were pinned, returns -errno. Each page returned must be released 1749d2bf6be8SNick Piggin * with a put_page() call when it is finished with. vmas will only 1750d2bf6be8SNick Piggin * remain valid while mmap_sem is held. 1751d2bf6be8SNick Piggin * 1752d2bf6be8SNick Piggin * Must be called with mmap_sem held for read or write. 1753d2bf6be8SNick Piggin * 1754d2bf6be8SNick Piggin * get_user_pages walks a process's page tables and takes a reference to 1755d2bf6be8SNick Piggin * each struct page that each user address corresponds to at a given 1756d2bf6be8SNick Piggin * instant. That is, it takes the page that would be accessed if a user 1757d2bf6be8SNick Piggin * thread accesses the given user virtual address at that instant. 1758d2bf6be8SNick Piggin * 1759d2bf6be8SNick Piggin * This does not guarantee that the page exists in the user mappings when 1760d2bf6be8SNick Piggin * get_user_pages returns, and there may even be a completely different 1761d2bf6be8SNick Piggin * page there in some cases (eg. if mmapped pagecache has been invalidated 1762d2bf6be8SNick Piggin * and subsequently re faulted). However it does guarantee that the page 1763d2bf6be8SNick Piggin * won't be freed completely. And mostly callers simply care that the page 1764d2bf6be8SNick Piggin * contains data that was valid *at some point in time*. Typically, an IO 1765d2bf6be8SNick Piggin * or similar operation cannot guarantee anything stronger anyway because 1766d2bf6be8SNick Piggin * locks can't be held over the syscall boundary. 1767d2bf6be8SNick Piggin * 1768d2bf6be8SNick Piggin * If write=0, the page must not be written to. If the page is written to, 1769d2bf6be8SNick Piggin * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called 1770d2bf6be8SNick Piggin * after the page is finished with, and before put_page is called. 1771d2bf6be8SNick Piggin * 1772d2bf6be8SNick Piggin * get_user_pages is typically used for fewer-copy IO operations, to get a 1773d2bf6be8SNick Piggin * handle on the memory by some means other than accesses via the user virtual 1774d2bf6be8SNick Piggin * addresses. The pages may be submitted for DMA to devices or accessed via 1775d2bf6be8SNick Piggin * their kernel linear mapping (via the kmap APIs). Care should be taken to 1776d2bf6be8SNick Piggin * use the correct cache flushing APIs. 1777d2bf6be8SNick Piggin * 1778d2bf6be8SNick Piggin * See also get_user_pages_fast, for performance critical applications. 1779d2bf6be8SNick Piggin */ 1780b291f000SNick Piggin int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 17819d73777eSPeter Zijlstra unsigned long start, int nr_pages, int write, int force, 1782b291f000SNick Piggin struct page **pages, struct vm_area_struct **vmas) 1783b291f000SNick Piggin { 178458fa879eSHugh Dickins int flags = FOLL_TOUCH; 1785b291f000SNick Piggin 178658fa879eSHugh Dickins if (pages) 178758fa879eSHugh Dickins flags |= FOLL_GET; 1788b291f000SNick Piggin if (write) 178958fa879eSHugh Dickins flags |= FOLL_WRITE; 1790b291f000SNick Piggin if (force) 179158fa879eSHugh Dickins flags |= FOLL_FORCE; 1792b291f000SNick Piggin 179353a7706dSMichel Lespinasse return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, 179453a7706dSMichel Lespinasse NULL); 1795b291f000SNick Piggin } 17961da177e4SLinus Torvalds EXPORT_SYMBOL(get_user_pages); 17971da177e4SLinus Torvalds 1798f3e8fccdSHugh Dickins /** 1799f3e8fccdSHugh Dickins * get_dump_page() - pin user page in memory while writing it to core dump 1800f3e8fccdSHugh Dickins * @addr: user address 1801f3e8fccdSHugh Dickins * 1802f3e8fccdSHugh Dickins * Returns struct page pointer of user page pinned for dump, 1803f3e8fccdSHugh Dickins * to be freed afterwards by page_cache_release() or put_page(). 1804f3e8fccdSHugh Dickins * 1805f3e8fccdSHugh Dickins * Returns NULL on any kind of failure - a hole must then be inserted into 1806f3e8fccdSHugh Dickins * the corefile, to preserve alignment with its headers; and also returns 1807f3e8fccdSHugh Dickins * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - 1808f3e8fccdSHugh Dickins * allowing a hole to be left in the corefile to save diskspace. 1809f3e8fccdSHugh Dickins * 1810f3e8fccdSHugh Dickins * Called without mmap_sem, but after all other threads have been killed. 1811f3e8fccdSHugh Dickins */ 1812f3e8fccdSHugh Dickins #ifdef CONFIG_ELF_CORE 1813f3e8fccdSHugh Dickins struct page *get_dump_page(unsigned long addr) 1814f3e8fccdSHugh Dickins { 1815f3e8fccdSHugh Dickins struct vm_area_struct *vma; 1816f3e8fccdSHugh Dickins struct page *page; 1817f3e8fccdSHugh Dickins 1818f3e8fccdSHugh Dickins if (__get_user_pages(current, current->mm, addr, 1, 181953a7706dSMichel Lespinasse FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma, 182053a7706dSMichel Lespinasse NULL) < 1) 1821f3e8fccdSHugh Dickins return NULL; 1822f3e8fccdSHugh Dickins flush_cache_page(vma, addr, page_to_pfn(page)); 1823f3e8fccdSHugh Dickins return page; 1824f3e8fccdSHugh Dickins } 1825f3e8fccdSHugh Dickins #endif /* CONFIG_ELF_CORE */ 1826f3e8fccdSHugh Dickins 182725ca1d6cSNamhyung Kim pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 1828920c7a5dSHarvey Harrison spinlock_t **ptl) 1829c9cfcddfSLinus Torvalds { 1830c9cfcddfSLinus Torvalds pgd_t * pgd = pgd_offset(mm, addr); 1831c9cfcddfSLinus Torvalds pud_t * pud = pud_alloc(mm, pgd, addr); 1832c9cfcddfSLinus Torvalds if (pud) { 183349c91fb0STrond Myklebust pmd_t * pmd = pmd_alloc(mm, pud, addr); 1834f66055abSAndrea Arcangeli if (pmd) { 1835f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 1836c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 1837c9cfcddfSLinus Torvalds } 1838f66055abSAndrea Arcangeli } 1839c9cfcddfSLinus Torvalds return NULL; 1840c9cfcddfSLinus Torvalds } 1841c9cfcddfSLinus Torvalds 18421da177e4SLinus Torvalds /* 1843238f58d8SLinus Torvalds * This is the old fallback for page remapping. 1844238f58d8SLinus Torvalds * 1845238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 1846238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 1847238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 1848238f58d8SLinus Torvalds */ 1849423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 1850423bad60SNick Piggin struct page *page, pgprot_t prot) 1851238f58d8SLinus Torvalds { 1852423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 1853238f58d8SLinus Torvalds int retval; 1854238f58d8SLinus Torvalds pte_t *pte; 1855238f58d8SLinus Torvalds spinlock_t *ptl; 1856238f58d8SLinus Torvalds 1857238f58d8SLinus Torvalds retval = -EINVAL; 1858a145dd41SLinus Torvalds if (PageAnon(page)) 18595b4e655eSKAMEZAWA Hiroyuki goto out; 1860238f58d8SLinus Torvalds retval = -ENOMEM; 1861238f58d8SLinus Torvalds flush_dcache_page(page); 1862c9cfcddfSLinus Torvalds pte = get_locked_pte(mm, addr, &ptl); 1863238f58d8SLinus Torvalds if (!pte) 18645b4e655eSKAMEZAWA Hiroyuki goto out; 1865238f58d8SLinus Torvalds retval = -EBUSY; 1866238f58d8SLinus Torvalds if (!pte_none(*pte)) 1867238f58d8SLinus Torvalds goto out_unlock; 1868238f58d8SLinus Torvalds 1869238f58d8SLinus Torvalds /* Ok, finally just insert the thing.. */ 1870238f58d8SLinus Torvalds get_page(page); 187134e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_FILEPAGES); 1872238f58d8SLinus Torvalds page_add_file_rmap(page); 1873238f58d8SLinus Torvalds set_pte_at(mm, addr, pte, mk_pte(page, prot)); 1874238f58d8SLinus Torvalds 1875238f58d8SLinus Torvalds retval = 0; 18768a9f3ccdSBalbir Singh pte_unmap_unlock(pte, ptl); 18778a9f3ccdSBalbir Singh return retval; 1878238f58d8SLinus Torvalds out_unlock: 1879238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 1880238f58d8SLinus Torvalds out: 1881238f58d8SLinus Torvalds return retval; 1882238f58d8SLinus Torvalds } 1883238f58d8SLinus Torvalds 1884bfa5bf6dSRolf Eike Beer /** 1885bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 1886bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1887bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 1888bfa5bf6dSRolf Eike Beer * @page: source kernel page 1889bfa5bf6dSRolf Eike Beer * 1890a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 1891a145dd41SLinus Torvalds * into a user vma. 1892a145dd41SLinus Torvalds * 1893a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 1894a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 1895a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 18968dfcc9baSNick Piggin * (see split_page()). 1897a145dd41SLinus Torvalds * 1898a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 1899a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 1900a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 1901a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 1902a145dd41SLinus Torvalds * ask for a shared writable mapping! 1903a145dd41SLinus Torvalds * 1904a145dd41SLinus Torvalds * The page does not need to be reserved. 1905a145dd41SLinus Torvalds */ 1906423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 1907423bad60SNick Piggin struct page *page) 1908a145dd41SLinus Torvalds { 1909a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 1910a145dd41SLinus Torvalds return -EFAULT; 1911a145dd41SLinus Torvalds if (!page_count(page)) 1912a145dd41SLinus Torvalds return -EINVAL; 19134d7672b4SLinus Torvalds vma->vm_flags |= VM_INSERTPAGE; 1914423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 1915a145dd41SLinus Torvalds } 1916e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 1917a145dd41SLinus Torvalds 1918423bad60SNick Piggin static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, 1919423bad60SNick Piggin unsigned long pfn, pgprot_t prot) 1920423bad60SNick Piggin { 1921423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 1922423bad60SNick Piggin int retval; 1923423bad60SNick Piggin pte_t *pte, entry; 1924423bad60SNick Piggin spinlock_t *ptl; 1925423bad60SNick Piggin 1926423bad60SNick Piggin retval = -ENOMEM; 1927423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 1928423bad60SNick Piggin if (!pte) 1929423bad60SNick Piggin goto out; 1930423bad60SNick Piggin retval = -EBUSY; 1931423bad60SNick Piggin if (!pte_none(*pte)) 1932423bad60SNick Piggin goto out_unlock; 1933423bad60SNick Piggin 1934423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 1935423bad60SNick Piggin entry = pte_mkspecial(pfn_pte(pfn, prot)); 1936423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 19374b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 1938423bad60SNick Piggin 1939423bad60SNick Piggin retval = 0; 1940423bad60SNick Piggin out_unlock: 1941423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 1942423bad60SNick Piggin out: 1943423bad60SNick Piggin return retval; 1944423bad60SNick Piggin } 1945423bad60SNick Piggin 1946e0dc0d8fSNick Piggin /** 1947e0dc0d8fSNick Piggin * vm_insert_pfn - insert single pfn into user vma 1948e0dc0d8fSNick Piggin * @vma: user vma to map to 1949e0dc0d8fSNick Piggin * @addr: target user address of this page 1950e0dc0d8fSNick Piggin * @pfn: source kernel pfn 1951e0dc0d8fSNick Piggin * 1952e0dc0d8fSNick Piggin * Similar to vm_inert_page, this allows drivers to insert individual pages 1953e0dc0d8fSNick Piggin * they've allocated into a user vma. Same comments apply. 1954e0dc0d8fSNick Piggin * 1955e0dc0d8fSNick Piggin * This function should only be called from a vm_ops->fault handler, and 1956e0dc0d8fSNick Piggin * in that case the handler should return NULL. 19570d71d10aSNick Piggin * 19580d71d10aSNick Piggin * vma cannot be a COW mapping. 19590d71d10aSNick Piggin * 19600d71d10aSNick Piggin * As this is called only for pages that do not currently exist, we 19610d71d10aSNick Piggin * do not need to flush old virtual caches or the TLB. 1962e0dc0d8fSNick Piggin */ 1963e0dc0d8fSNick Piggin int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 1964e0dc0d8fSNick Piggin unsigned long pfn) 1965e0dc0d8fSNick Piggin { 19662ab64037Svenkatesh.pallipadi@intel.com int ret; 1967e4b866edSvenkatesh.pallipadi@intel.com pgprot_t pgprot = vma->vm_page_prot; 19687e675137SNick Piggin /* 19697e675137SNick Piggin * Technically, architectures with pte_special can avoid all these 19707e675137SNick Piggin * restrictions (same for remap_pfn_range). However we would like 19717e675137SNick Piggin * consistency in testing and feature parity among all, so we should 19727e675137SNick Piggin * try to keep these invariants in place for everybody. 19737e675137SNick Piggin */ 1974b379d790SJared Hulbert BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 1975b379d790SJared Hulbert BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 1976b379d790SJared Hulbert (VM_PFNMAP|VM_MIXEDMAP)); 1977b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 1978b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 1979e0dc0d8fSNick Piggin 1980423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 1981423bad60SNick Piggin return -EFAULT; 1982e4b866edSvenkatesh.pallipadi@intel.com if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE)) 19832ab64037Svenkatesh.pallipadi@intel.com return -EINVAL; 19842ab64037Svenkatesh.pallipadi@intel.com 1985e4b866edSvenkatesh.pallipadi@intel.com ret = insert_pfn(vma, addr, pfn, pgprot); 19862ab64037Svenkatesh.pallipadi@intel.com 19872ab64037Svenkatesh.pallipadi@intel.com if (ret) 19882ab64037Svenkatesh.pallipadi@intel.com untrack_pfn_vma(vma, pfn, PAGE_SIZE); 19892ab64037Svenkatesh.pallipadi@intel.com 19902ab64037Svenkatesh.pallipadi@intel.com return ret; 1991e0dc0d8fSNick Piggin } 1992e0dc0d8fSNick Piggin EXPORT_SYMBOL(vm_insert_pfn); 1993e0dc0d8fSNick Piggin 1994423bad60SNick Piggin int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 1995423bad60SNick Piggin unsigned long pfn) 1996423bad60SNick Piggin { 1997423bad60SNick Piggin BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); 1998423bad60SNick Piggin 1999423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 2000423bad60SNick Piggin return -EFAULT; 2001423bad60SNick Piggin 2002423bad60SNick Piggin /* 2003423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 2004423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 2005423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 200662eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 200762eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 2008423bad60SNick Piggin */ 2009423bad60SNick Piggin if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { 2010423bad60SNick Piggin struct page *page; 2011423bad60SNick Piggin 2012423bad60SNick Piggin page = pfn_to_page(pfn); 2013423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 2014423bad60SNick Piggin } 2015423bad60SNick Piggin return insert_pfn(vma, addr, pfn, vma->vm_page_prot); 2016423bad60SNick Piggin } 2017423bad60SNick Piggin EXPORT_SYMBOL(vm_insert_mixed); 2018423bad60SNick Piggin 2019a145dd41SLinus Torvalds /* 20201da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 20211da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 20221da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 20231da177e4SLinus Torvalds */ 20241da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 20251da177e4SLinus Torvalds unsigned long addr, unsigned long end, 20261da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 20271da177e4SLinus Torvalds { 20281da177e4SLinus Torvalds pte_t *pte; 2029c74df32cSHugh Dickins spinlock_t *ptl; 20301da177e4SLinus Torvalds 2031c74df32cSHugh Dickins pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 20321da177e4SLinus Torvalds if (!pte) 20331da177e4SLinus Torvalds return -ENOMEM; 20346606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 20351da177e4SLinus Torvalds do { 20361da177e4SLinus Torvalds BUG_ON(!pte_none(*pte)); 20377e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 20381da177e4SLinus Torvalds pfn++; 20391da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 20406606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 2041c74df32cSHugh Dickins pte_unmap_unlock(pte - 1, ptl); 20421da177e4SLinus Torvalds return 0; 20431da177e4SLinus Torvalds } 20441da177e4SLinus Torvalds 20451da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 20461da177e4SLinus Torvalds unsigned long addr, unsigned long end, 20471da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 20481da177e4SLinus Torvalds { 20491da177e4SLinus Torvalds pmd_t *pmd; 20501da177e4SLinus Torvalds unsigned long next; 20511da177e4SLinus Torvalds 20521da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 20531da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 20541da177e4SLinus Torvalds if (!pmd) 20551da177e4SLinus Torvalds return -ENOMEM; 2056f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 20571da177e4SLinus Torvalds do { 20581da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 20591da177e4SLinus Torvalds if (remap_pte_range(mm, pmd, addr, next, 20601da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 20611da177e4SLinus Torvalds return -ENOMEM; 20621da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 20631da177e4SLinus Torvalds return 0; 20641da177e4SLinus Torvalds } 20651da177e4SLinus Torvalds 20661da177e4SLinus Torvalds static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, 20671da177e4SLinus Torvalds unsigned long addr, unsigned long end, 20681da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 20691da177e4SLinus Torvalds { 20701da177e4SLinus Torvalds pud_t *pud; 20711da177e4SLinus Torvalds unsigned long next; 20721da177e4SLinus Torvalds 20731da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 20741da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, addr); 20751da177e4SLinus Torvalds if (!pud) 20761da177e4SLinus Torvalds return -ENOMEM; 20771da177e4SLinus Torvalds do { 20781da177e4SLinus Torvalds next = pud_addr_end(addr, end); 20791da177e4SLinus Torvalds if (remap_pmd_range(mm, pud, addr, next, 20801da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 20811da177e4SLinus Torvalds return -ENOMEM; 20821da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 20831da177e4SLinus Torvalds return 0; 20841da177e4SLinus Torvalds } 20851da177e4SLinus Torvalds 2086bfa5bf6dSRolf Eike Beer /** 2087bfa5bf6dSRolf Eike Beer * remap_pfn_range - remap kernel memory to userspace 2088bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 2089bfa5bf6dSRolf Eike Beer * @addr: target user address to start at 2090bfa5bf6dSRolf Eike Beer * @pfn: physical address of kernel memory 2091bfa5bf6dSRolf Eike Beer * @size: size of map area 2092bfa5bf6dSRolf Eike Beer * @prot: page protection flags for this mapping 2093bfa5bf6dSRolf Eike Beer * 2094bfa5bf6dSRolf Eike Beer * Note: this is only safe if the mm semaphore is held when called. 2095bfa5bf6dSRolf Eike Beer */ 20961da177e4SLinus Torvalds int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 20971da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 20981da177e4SLinus Torvalds { 20991da177e4SLinus Torvalds pgd_t *pgd; 21001da177e4SLinus Torvalds unsigned long next; 21012d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 21021da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 21031da177e4SLinus Torvalds int err; 21041da177e4SLinus Torvalds 21051da177e4SLinus Torvalds /* 21061da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 21071da177e4SLinus Torvalds * rest of the world about it: 21081da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 21091da177e4SLinus Torvalds * (accesses can have side effects). 21100b14c179SHugh Dickins * VM_RESERVED is specified all over the place, because 21110b14c179SHugh Dickins * in 2.4 it kept swapout's vma scan off this vma; but 21120b14c179SHugh Dickins * in 2.6 the LRU scan won't even find its pages, so this 21130b14c179SHugh Dickins * flag means no more than count its pages in reserved_vm, 21140b14c179SHugh Dickins * and omit it from core dump, even when VM_IO turned off. 21156aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 21166aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 21176aab341eSLinus Torvalds * with them. 2118fb155c16SLinus Torvalds * 2119fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 2120fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 2121fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 21221da177e4SLinus Torvalds */ 21234bb9c5c0SPallipadi, Venkatesh if (addr == vma->vm_start && end == vma->vm_end) { 21246aab341eSLinus Torvalds vma->vm_pgoff = pfn; 2125895791daSPallipadi, Venkatesh vma->vm_flags |= VM_PFN_AT_MMAP; 21264bb9c5c0SPallipadi, Venkatesh } else if (is_cow_mapping(vma->vm_flags)) 21273c8bb73aSvenkatesh.pallipadi@intel.com return -EINVAL; 2128fb155c16SLinus Torvalds 2129fb155c16SLinus Torvalds vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; 21301da177e4SLinus Torvalds 2131e4b866edSvenkatesh.pallipadi@intel.com err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size)); 2132a3670613Svenkatesh.pallipadi@intel.com if (err) { 2133a3670613Svenkatesh.pallipadi@intel.com /* 2134a3670613Svenkatesh.pallipadi@intel.com * To indicate that track_pfn related cleanup is not 2135a3670613Svenkatesh.pallipadi@intel.com * needed from higher level routine calling unmap_vmas 2136a3670613Svenkatesh.pallipadi@intel.com */ 2137a3670613Svenkatesh.pallipadi@intel.com vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP); 2138895791daSPallipadi, Venkatesh vma->vm_flags &= ~VM_PFN_AT_MMAP; 21392ab64037Svenkatesh.pallipadi@intel.com return -EINVAL; 2140a3670613Svenkatesh.pallipadi@intel.com } 21412ab64037Svenkatesh.pallipadi@intel.com 21421da177e4SLinus Torvalds BUG_ON(addr >= end); 21431da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 21441da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 21451da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 21461da177e4SLinus Torvalds do { 21471da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 21481da177e4SLinus Torvalds err = remap_pud_range(mm, pgd, addr, next, 21491da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 21501da177e4SLinus Torvalds if (err) 21511da177e4SLinus Torvalds break; 21521da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 21532ab64037Svenkatesh.pallipadi@intel.com 21542ab64037Svenkatesh.pallipadi@intel.com if (err) 21552ab64037Svenkatesh.pallipadi@intel.com untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size)); 21562ab64037Svenkatesh.pallipadi@intel.com 21571da177e4SLinus Torvalds return err; 21581da177e4SLinus Torvalds } 21591da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 21601da177e4SLinus Torvalds 2161aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 2162aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2163aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 2164aee16b3cSJeremy Fitzhardinge { 2165aee16b3cSJeremy Fitzhardinge pte_t *pte; 2166aee16b3cSJeremy Fitzhardinge int err; 21672f569afdSMartin Schwidefsky pgtable_t token; 216894909914SBorislav Petkov spinlock_t *uninitialized_var(ptl); 2169aee16b3cSJeremy Fitzhardinge 2170aee16b3cSJeremy Fitzhardinge pte = (mm == &init_mm) ? 2171aee16b3cSJeremy Fitzhardinge pte_alloc_kernel(pmd, addr) : 2172aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 2173aee16b3cSJeremy Fitzhardinge if (!pte) 2174aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2175aee16b3cSJeremy Fitzhardinge 2176aee16b3cSJeremy Fitzhardinge BUG_ON(pmd_huge(*pmd)); 2177aee16b3cSJeremy Fitzhardinge 217838e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 217938e0edb1SJeremy Fitzhardinge 21802f569afdSMartin Schwidefsky token = pmd_pgtable(*pmd); 2181aee16b3cSJeremy Fitzhardinge 2182aee16b3cSJeremy Fitzhardinge do { 2183c36987e2SDaisuke Nishimura err = fn(pte++, token, addr, data); 2184aee16b3cSJeremy Fitzhardinge if (err) 2185aee16b3cSJeremy Fitzhardinge break; 2186c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 2187aee16b3cSJeremy Fitzhardinge 218838e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 218938e0edb1SJeremy Fitzhardinge 2190aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 2191aee16b3cSJeremy Fitzhardinge pte_unmap_unlock(pte-1, ptl); 2192aee16b3cSJeremy Fitzhardinge return err; 2193aee16b3cSJeremy Fitzhardinge } 2194aee16b3cSJeremy Fitzhardinge 2195aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 2196aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2197aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 2198aee16b3cSJeremy Fitzhardinge { 2199aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 2200aee16b3cSJeremy Fitzhardinge unsigned long next; 2201aee16b3cSJeremy Fitzhardinge int err; 2202aee16b3cSJeremy Fitzhardinge 2203ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 2204ceb86879SAndi Kleen 2205aee16b3cSJeremy Fitzhardinge pmd = pmd_alloc(mm, pud, addr); 2206aee16b3cSJeremy Fitzhardinge if (!pmd) 2207aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2208aee16b3cSJeremy Fitzhardinge do { 2209aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 2210aee16b3cSJeremy Fitzhardinge err = apply_to_pte_range(mm, pmd, addr, next, fn, data); 2211aee16b3cSJeremy Fitzhardinge if (err) 2212aee16b3cSJeremy Fitzhardinge break; 2213aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 2214aee16b3cSJeremy Fitzhardinge return err; 2215aee16b3cSJeremy Fitzhardinge } 2216aee16b3cSJeremy Fitzhardinge 2217aee16b3cSJeremy Fitzhardinge static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, 2218aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2219aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 2220aee16b3cSJeremy Fitzhardinge { 2221aee16b3cSJeremy Fitzhardinge pud_t *pud; 2222aee16b3cSJeremy Fitzhardinge unsigned long next; 2223aee16b3cSJeremy Fitzhardinge int err; 2224aee16b3cSJeremy Fitzhardinge 2225aee16b3cSJeremy Fitzhardinge pud = pud_alloc(mm, pgd, addr); 2226aee16b3cSJeremy Fitzhardinge if (!pud) 2227aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2228aee16b3cSJeremy Fitzhardinge do { 2229aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 2230aee16b3cSJeremy Fitzhardinge err = apply_to_pmd_range(mm, pud, addr, next, fn, data); 2231aee16b3cSJeremy Fitzhardinge if (err) 2232aee16b3cSJeremy Fitzhardinge break; 2233aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 2234aee16b3cSJeremy Fitzhardinge return err; 2235aee16b3cSJeremy Fitzhardinge } 2236aee16b3cSJeremy Fitzhardinge 2237aee16b3cSJeremy Fitzhardinge /* 2238aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 2239aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 2240aee16b3cSJeremy Fitzhardinge */ 2241aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2242aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 2243aee16b3cSJeremy Fitzhardinge { 2244aee16b3cSJeremy Fitzhardinge pgd_t *pgd; 2245aee16b3cSJeremy Fitzhardinge unsigned long next; 224657250a5bSJeremy Fitzhardinge unsigned long end = addr + size; 2247aee16b3cSJeremy Fitzhardinge int err; 2248aee16b3cSJeremy Fitzhardinge 2249aee16b3cSJeremy Fitzhardinge BUG_ON(addr >= end); 2250aee16b3cSJeremy Fitzhardinge pgd = pgd_offset(mm, addr); 2251aee16b3cSJeremy Fitzhardinge do { 2252aee16b3cSJeremy Fitzhardinge next = pgd_addr_end(addr, end); 2253aee16b3cSJeremy Fitzhardinge err = apply_to_pud_range(mm, pgd, addr, next, fn, data); 2254aee16b3cSJeremy Fitzhardinge if (err) 2255aee16b3cSJeremy Fitzhardinge break; 2256aee16b3cSJeremy Fitzhardinge } while (pgd++, addr = next, addr != end); 225757250a5bSJeremy Fitzhardinge 2258aee16b3cSJeremy Fitzhardinge return err; 2259aee16b3cSJeremy Fitzhardinge } 2260aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 2261aee16b3cSJeremy Fitzhardinge 22621da177e4SLinus Torvalds /* 22638f4e2101SHugh Dickins * handle_pte_fault chooses page fault handler according to an entry 22648f4e2101SHugh Dickins * which was read non-atomically. Before making any commitment, on 22658f4e2101SHugh Dickins * those architectures or configurations (e.g. i386 with PAE) which 2266a335b2e1SRyota Ozaki * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault 22678f4e2101SHugh Dickins * must check under lock before unmapping the pte and proceeding 22688f4e2101SHugh Dickins * (but do_wp_page is only called after already making such a check; 2269a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 22708f4e2101SHugh Dickins */ 22714c21e2f2SHugh Dickins static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, 22728f4e2101SHugh Dickins pte_t *page_table, pte_t orig_pte) 22738f4e2101SHugh Dickins { 22748f4e2101SHugh Dickins int same = 1; 22758f4e2101SHugh Dickins #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) 22768f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 22774c21e2f2SHugh Dickins spinlock_t *ptl = pte_lockptr(mm, pmd); 22784c21e2f2SHugh Dickins spin_lock(ptl); 22798f4e2101SHugh Dickins same = pte_same(*page_table, orig_pte); 22804c21e2f2SHugh Dickins spin_unlock(ptl); 22818f4e2101SHugh Dickins } 22828f4e2101SHugh Dickins #endif 22838f4e2101SHugh Dickins pte_unmap(page_table); 22848f4e2101SHugh Dickins return same; 22858f4e2101SHugh Dickins } 22868f4e2101SHugh Dickins 22879de455b2SAtsushi Nemoto static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) 22886aab341eSLinus Torvalds { 22896aab341eSLinus Torvalds /* 22906aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 22916aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 22926aab341eSLinus Torvalds * just copying from the original user address. If that 22936aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 22946aab341eSLinus Torvalds */ 22956aab341eSLinus Torvalds if (unlikely(!src)) { 22966aab341eSLinus Torvalds void *kaddr = kmap_atomic(dst, KM_USER0); 22975d2a2dbbSLinus Torvalds void __user *uaddr = (void __user *)(va & PAGE_MASK); 22985d2a2dbbSLinus Torvalds 22995d2a2dbbSLinus Torvalds /* 23005d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 23015d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 23025d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 23035d2a2dbbSLinus Torvalds * zeroes. 23045d2a2dbbSLinus Torvalds */ 23055d2a2dbbSLinus Torvalds if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) 23063ecb01dfSJan Beulich clear_page(kaddr); 23076aab341eSLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 2308c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 23090ed361deSNick Piggin } else 23109de455b2SAtsushi Nemoto copy_user_highpage(dst, src, va, vma); 23116aab341eSLinus Torvalds } 23126aab341eSLinus Torvalds 23131da177e4SLinus Torvalds /* 23141da177e4SLinus Torvalds * This routine handles present pages, when users try to write 23151da177e4SLinus Torvalds * to a shared page. It is done by copying the page to a new address 23161da177e4SLinus Torvalds * and decrementing the shared-page counter for the old page. 23171da177e4SLinus Torvalds * 23181da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 23191da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 23201da177e4SLinus Torvalds * Thus we can safely just mark it writable once we've done any necessary 23211da177e4SLinus Torvalds * COW. 23221da177e4SLinus Torvalds * 23231da177e4SLinus Torvalds * We also mark the page dirty at this point even though the page will 23241da177e4SLinus Torvalds * change only once the write actually happens. This avoids a few races, 23251da177e4SLinus Torvalds * and potentially makes it more efficient. 23261da177e4SLinus Torvalds * 23278f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 23288f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 23298f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 23301da177e4SLinus Torvalds */ 23311da177e4SLinus Torvalds static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, 233265500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 23338f4e2101SHugh Dickins spinlock_t *ptl, pte_t orig_pte) 2334e6219ec8SNamhyung Kim __releases(ptl) 23351da177e4SLinus Torvalds { 2336e5bbe4dfSHugh Dickins struct page *old_page, *new_page; 23371da177e4SLinus Torvalds pte_t entry; 2338b009c024SMichel Lespinasse int ret = 0; 2339a200ee18SPeter Zijlstra int page_mkwrite = 0; 2340d08b3851SPeter Zijlstra struct page *dirty_page = NULL; 23411da177e4SLinus Torvalds 23426aab341eSLinus Torvalds old_page = vm_normal_page(vma, address, orig_pte); 2343251b97f5SPeter Zijlstra if (!old_page) { 2344251b97f5SPeter Zijlstra /* 2345251b97f5SPeter Zijlstra * VM_MIXEDMAP !pfn_valid() case 2346251b97f5SPeter Zijlstra * 2347251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 2348251b97f5SPeter Zijlstra * Just mark the pages writable as we can't do any dirty 2349251b97f5SPeter Zijlstra * accounting on raw pfn maps. 2350251b97f5SPeter Zijlstra */ 2351251b97f5SPeter Zijlstra if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2352251b97f5SPeter Zijlstra (VM_WRITE|VM_SHARED)) 2353251b97f5SPeter Zijlstra goto reuse; 2354920fc356SHugh Dickins goto gotten; 2355251b97f5SPeter Zijlstra } 23561da177e4SLinus Torvalds 2357d08b3851SPeter Zijlstra /* 2358ee6a6457SPeter Zijlstra * Take out anonymous pages first, anonymous shared vmas are 2359ee6a6457SPeter Zijlstra * not dirty accountable. 2360d08b3851SPeter Zijlstra */ 23619a840895SHugh Dickins if (PageAnon(old_page) && !PageKsm(old_page)) { 2362ab967d86SHugh Dickins if (!trylock_page(old_page)) { 2363ab967d86SHugh Dickins page_cache_get(old_page); 2364ab967d86SHugh Dickins pte_unmap_unlock(page_table, ptl); 2365ab967d86SHugh Dickins lock_page(old_page); 2366ab967d86SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, 2367ab967d86SHugh Dickins &ptl); 2368ab967d86SHugh Dickins if (!pte_same(*page_table, orig_pte)) { 2369ab967d86SHugh Dickins unlock_page(old_page); 2370ab967d86SHugh Dickins goto unlock; 2371ab967d86SHugh Dickins } 2372ab967d86SHugh Dickins page_cache_release(old_page); 2373ab967d86SHugh Dickins } 2374b009c024SMichel Lespinasse if (reuse_swap_page(old_page)) { 2375c44b6743SRik van Riel /* 2376c44b6743SRik van Riel * The page is all ours. Move it to our anon_vma so 2377c44b6743SRik van Riel * the rmap code will not search our parent or siblings. 2378c44b6743SRik van Riel * Protected against the rmap code by the page lock. 2379c44b6743SRik van Riel */ 2380c44b6743SRik van Riel page_move_anon_rmap(old_page, vma, address); 2381ee6a6457SPeter Zijlstra unlock_page(old_page); 2382b009c024SMichel Lespinasse goto reuse; 2383b009c024SMichel Lespinasse } 2384b009c024SMichel Lespinasse unlock_page(old_page); 2385ee6a6457SPeter Zijlstra } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2386d08b3851SPeter Zijlstra (VM_WRITE|VM_SHARED))) { 2387ee6a6457SPeter Zijlstra /* 2388ee6a6457SPeter Zijlstra * Only catch write-faults on shared writable pages, 2389ee6a6457SPeter Zijlstra * read-only shared pages can get COWed by 2390ee6a6457SPeter Zijlstra * get_user_pages(.write=1, .force=1). 2391ee6a6457SPeter Zijlstra */ 23929637a5efSDavid Howells if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 2393c2ec175cSNick Piggin struct vm_fault vmf; 2394c2ec175cSNick Piggin int tmp; 2395c2ec175cSNick Piggin 2396c2ec175cSNick Piggin vmf.virtual_address = (void __user *)(address & 2397c2ec175cSNick Piggin PAGE_MASK); 2398c2ec175cSNick Piggin vmf.pgoff = old_page->index; 2399c2ec175cSNick Piggin vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 2400c2ec175cSNick Piggin vmf.page = old_page; 2401c2ec175cSNick Piggin 24029637a5efSDavid Howells /* 24039637a5efSDavid Howells * Notify the address space that the page is about to 24049637a5efSDavid Howells * become writable so that it can prohibit this or wait 24059637a5efSDavid Howells * for the page to get into an appropriate state. 24069637a5efSDavid Howells * 24079637a5efSDavid Howells * We do this without the lock held, so that it can 24089637a5efSDavid Howells * sleep if it needs to. 24099637a5efSDavid Howells */ 24109637a5efSDavid Howells page_cache_get(old_page); 24119637a5efSDavid Howells pte_unmap_unlock(page_table, ptl); 24129637a5efSDavid Howells 2413c2ec175cSNick Piggin tmp = vma->vm_ops->page_mkwrite(vma, &vmf); 2414c2ec175cSNick Piggin if (unlikely(tmp & 2415c2ec175cSNick Piggin (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { 2416c2ec175cSNick Piggin ret = tmp; 24179637a5efSDavid Howells goto unwritable_page; 2418c2ec175cSNick Piggin } 2419b827e496SNick Piggin if (unlikely(!(tmp & VM_FAULT_LOCKED))) { 2420b827e496SNick Piggin lock_page(old_page); 2421b827e496SNick Piggin if (!old_page->mapping) { 2422b827e496SNick Piggin ret = 0; /* retry the fault */ 2423b827e496SNick Piggin unlock_page(old_page); 2424b827e496SNick Piggin goto unwritable_page; 2425b827e496SNick Piggin } 2426b827e496SNick Piggin } else 2427b827e496SNick Piggin VM_BUG_ON(!PageLocked(old_page)); 24289637a5efSDavid Howells 24299637a5efSDavid Howells /* 24309637a5efSDavid Howells * Since we dropped the lock we need to revalidate 24319637a5efSDavid Howells * the PTE as someone else may have changed it. If 24329637a5efSDavid Howells * they did, we just return, as we can count on the 24339637a5efSDavid Howells * MMU to tell us if they didn't also make it writable. 24349637a5efSDavid Howells */ 24359637a5efSDavid Howells page_table = pte_offset_map_lock(mm, pmd, address, 24369637a5efSDavid Howells &ptl); 2437b827e496SNick Piggin if (!pte_same(*page_table, orig_pte)) { 2438b827e496SNick Piggin unlock_page(old_page); 24399637a5efSDavid Howells goto unlock; 2440b827e496SNick Piggin } 2441a200ee18SPeter Zijlstra 2442a200ee18SPeter Zijlstra page_mkwrite = 1; 24439637a5efSDavid Howells } 2444d08b3851SPeter Zijlstra dirty_page = old_page; 2445d08b3851SPeter Zijlstra get_page(dirty_page); 24469637a5efSDavid Howells 2447251b97f5SPeter Zijlstra reuse: 2448eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 244965500d23SHugh Dickins entry = pte_mkyoung(orig_pte); 245065500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2451954ffcb3SKAMEZAWA Hiroyuki if (ptep_set_access_flags(vma, address, page_table, entry,1)) 24524b3073e1SRussell King update_mmu_cache(vma, address, page_table); 245372ddc8f7SMichel Lespinasse pte_unmap_unlock(page_table, ptl); 245465500d23SHugh Dickins ret |= VM_FAULT_WRITE; 245572ddc8f7SMichel Lespinasse 245672ddc8f7SMichel Lespinasse if (!dirty_page) 245772ddc8f7SMichel Lespinasse return ret; 245872ddc8f7SMichel Lespinasse 245972ddc8f7SMichel Lespinasse /* 246072ddc8f7SMichel Lespinasse * Yes, Virginia, this is actually required to prevent a race 246172ddc8f7SMichel Lespinasse * with clear_page_dirty_for_io() from clearing the page dirty 246272ddc8f7SMichel Lespinasse * bit after it clear all dirty ptes, but before a racing 246372ddc8f7SMichel Lespinasse * do_wp_page installs a dirty pte. 246472ddc8f7SMichel Lespinasse * 2465a335b2e1SRyota Ozaki * __do_fault is protected similarly. 246672ddc8f7SMichel Lespinasse */ 246772ddc8f7SMichel Lespinasse if (!page_mkwrite) { 246872ddc8f7SMichel Lespinasse wait_on_page_locked(dirty_page); 246972ddc8f7SMichel Lespinasse set_page_dirty_balance(dirty_page, page_mkwrite); 247072ddc8f7SMichel Lespinasse } 247172ddc8f7SMichel Lespinasse put_page(dirty_page); 247272ddc8f7SMichel Lespinasse if (page_mkwrite) { 247372ddc8f7SMichel Lespinasse struct address_space *mapping = dirty_page->mapping; 247472ddc8f7SMichel Lespinasse 247572ddc8f7SMichel Lespinasse set_page_dirty(dirty_page); 247672ddc8f7SMichel Lespinasse unlock_page(dirty_page); 247772ddc8f7SMichel Lespinasse page_cache_release(dirty_page); 247872ddc8f7SMichel Lespinasse if (mapping) { 247972ddc8f7SMichel Lespinasse /* 248072ddc8f7SMichel Lespinasse * Some device drivers do not set page.mapping 248172ddc8f7SMichel Lespinasse * but still dirty their pages 248272ddc8f7SMichel Lespinasse */ 248372ddc8f7SMichel Lespinasse balance_dirty_pages_ratelimited(mapping); 248472ddc8f7SMichel Lespinasse } 248572ddc8f7SMichel Lespinasse } 248672ddc8f7SMichel Lespinasse 248772ddc8f7SMichel Lespinasse /* file_update_time outside page_lock */ 248872ddc8f7SMichel Lespinasse if (vma->vm_file) 248972ddc8f7SMichel Lespinasse file_update_time(vma->vm_file); 249072ddc8f7SMichel Lespinasse 249172ddc8f7SMichel Lespinasse return ret; 24921da177e4SLinus Torvalds } 24931da177e4SLinus Torvalds 24941da177e4SLinus Torvalds /* 24951da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 24961da177e4SLinus Torvalds */ 24971da177e4SLinus Torvalds page_cache_get(old_page); 2498920fc356SHugh Dickins gotten: 24998f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 25001da177e4SLinus Torvalds 25011da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 250265500d23SHugh Dickins goto oom; 2503a13ea5b7SHugh Dickins 250462eede62SHugh Dickins if (is_zero_pfn(pte_pfn(orig_pte))) { 2505a13ea5b7SHugh Dickins new_page = alloc_zeroed_user_highpage_movable(vma, address); 2506a13ea5b7SHugh Dickins if (!new_page) 2507a13ea5b7SHugh Dickins goto oom; 2508a13ea5b7SHugh Dickins } else { 2509769848c0SMel Gorman new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 25101da177e4SLinus Torvalds if (!new_page) 251165500d23SHugh Dickins goto oom; 2512a13ea5b7SHugh Dickins cow_user_page(new_page, old_page, address, vma); 2513a13ea5b7SHugh Dickins } 2514a13ea5b7SHugh Dickins __SetPageUptodate(new_page); 2515a13ea5b7SHugh Dickins 25162c26fdd7SKAMEZAWA Hiroyuki if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL)) 25178a9f3ccdSBalbir Singh goto oom_free_new; 25188a9f3ccdSBalbir Singh 25191da177e4SLinus Torvalds /* 25201da177e4SLinus Torvalds * Re-check the pte - we dropped the lock 25211da177e4SLinus Torvalds */ 25228f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 252365500d23SHugh Dickins if (likely(pte_same(*page_table, orig_pte))) { 2524920fc356SHugh Dickins if (old_page) { 25254294621fSHugh Dickins if (!PageAnon(old_page)) { 252634e55232SKAMEZAWA Hiroyuki dec_mm_counter_fast(mm, MM_FILEPAGES); 252734e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 25284294621fSHugh Dickins } 2529920fc356SHugh Dickins } else 253034e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 2531eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 253265500d23SHugh Dickins entry = mk_pte(new_page, vma->vm_page_prot); 253365500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 25344ce072f1SSiddha, Suresh B /* 25354ce072f1SSiddha, Suresh B * Clear the pte entry and flush it first, before updating the 25364ce072f1SSiddha, Suresh B * pte with the new entry. This will avoid a race condition 25374ce072f1SSiddha, Suresh B * seen in the presence of one thread doing SMC and another 25384ce072f1SSiddha, Suresh B * thread doing COW. 25394ce072f1SSiddha, Suresh B */ 2540828502d3SIzik Eidus ptep_clear_flush(vma, address, page_table); 25419617d95eSNick Piggin page_add_new_anon_rmap(new_page, vma, address); 2542828502d3SIzik Eidus /* 2543828502d3SIzik Eidus * We call the notify macro here because, when using secondary 2544828502d3SIzik Eidus * mmu page tables (such as kvm shadow page tables), we want the 2545828502d3SIzik Eidus * new page to be mapped directly into the secondary page table. 2546828502d3SIzik Eidus */ 2547828502d3SIzik Eidus set_pte_at_notify(mm, address, page_table, entry); 25484b3073e1SRussell King update_mmu_cache(vma, address, page_table); 2549945754a1SNick Piggin if (old_page) { 2550945754a1SNick Piggin /* 2551945754a1SNick Piggin * Only after switching the pte to the new page may 2552945754a1SNick Piggin * we remove the mapcount here. Otherwise another 2553945754a1SNick Piggin * process may come and find the rmap count decremented 2554945754a1SNick Piggin * before the pte is switched to the new page, and 2555945754a1SNick Piggin * "reuse" the old page writing into it while our pte 2556945754a1SNick Piggin * here still points into it and can be read by other 2557945754a1SNick Piggin * threads. 2558945754a1SNick Piggin * 2559945754a1SNick Piggin * The critical issue is to order this 2560945754a1SNick Piggin * page_remove_rmap with the ptp_clear_flush above. 2561945754a1SNick Piggin * Those stores are ordered by (if nothing else,) 2562945754a1SNick Piggin * the barrier present in the atomic_add_negative 2563945754a1SNick Piggin * in page_remove_rmap. 2564945754a1SNick Piggin * 2565945754a1SNick Piggin * Then the TLB flush in ptep_clear_flush ensures that 2566945754a1SNick Piggin * no process can access the old page before the 2567945754a1SNick Piggin * decremented mapcount is visible. And the old page 2568945754a1SNick Piggin * cannot be reused until after the decremented 2569945754a1SNick Piggin * mapcount is visible. So transitively, TLBs to 2570945754a1SNick Piggin * old page will be flushed before it can be reused. 2571945754a1SNick Piggin */ 2572edc315fdSHugh Dickins page_remove_rmap(old_page); 2573945754a1SNick Piggin } 2574945754a1SNick Piggin 25751da177e4SLinus Torvalds /* Free the old page.. */ 25761da177e4SLinus Torvalds new_page = old_page; 2577f33ea7f4SNick Piggin ret |= VM_FAULT_WRITE; 25788a9f3ccdSBalbir Singh } else 25798a9f3ccdSBalbir Singh mem_cgroup_uncharge_page(new_page); 25808a9f3ccdSBalbir Singh 2581920fc356SHugh Dickins if (new_page) 25821da177e4SLinus Torvalds page_cache_release(new_page); 258365500d23SHugh Dickins unlock: 25848f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 2585e15f8c01SMichel Lespinasse if (old_page) { 2586e15f8c01SMichel Lespinasse /* 2587e15f8c01SMichel Lespinasse * Don't let another task, with possibly unlocked vma, 2588e15f8c01SMichel Lespinasse * keep the mlocked page. 2589e15f8c01SMichel Lespinasse */ 2590e15f8c01SMichel Lespinasse if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) { 2591e15f8c01SMichel Lespinasse lock_page(old_page); /* LRU manipulation */ 2592e15f8c01SMichel Lespinasse munlock_vma_page(old_page); 2593e15f8c01SMichel Lespinasse unlock_page(old_page); 2594e15f8c01SMichel Lespinasse } 2595e15f8c01SMichel Lespinasse page_cache_release(old_page); 2596e15f8c01SMichel Lespinasse } 2597f33ea7f4SNick Piggin return ret; 25988a9f3ccdSBalbir Singh oom_free_new: 25996dbf6d3bSHugh Dickins page_cache_release(new_page); 260065500d23SHugh Dickins oom: 2601b827e496SNick Piggin if (old_page) { 2602b827e496SNick Piggin if (page_mkwrite) { 2603b827e496SNick Piggin unlock_page(old_page); 26041da177e4SLinus Torvalds page_cache_release(old_page); 2605b827e496SNick Piggin } 2606b827e496SNick Piggin page_cache_release(old_page); 2607b827e496SNick Piggin } 26081da177e4SLinus Torvalds return VM_FAULT_OOM; 26099637a5efSDavid Howells 26109637a5efSDavid Howells unwritable_page: 26119637a5efSDavid Howells page_cache_release(old_page); 2612c2ec175cSNick Piggin return ret; 26131da177e4SLinus Torvalds } 26141da177e4SLinus Torvalds 26151da177e4SLinus Torvalds /* 26161da177e4SLinus Torvalds * Helper functions for unmap_mapping_range(). 26171da177e4SLinus Torvalds * 26181da177e4SLinus Torvalds * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __ 26191da177e4SLinus Torvalds * 26201da177e4SLinus Torvalds * We have to restart searching the prio_tree whenever we drop the lock, 26211da177e4SLinus Torvalds * since the iterator is only valid while the lock is held, and anyway 26221da177e4SLinus Torvalds * a later vma might be split and reinserted earlier while lock dropped. 26231da177e4SLinus Torvalds * 26241da177e4SLinus Torvalds * The list of nonlinear vmas could be handled more efficiently, using 26251da177e4SLinus Torvalds * a placeholder, but handle it in the same way until a need is shown. 26261da177e4SLinus Torvalds * It is important to search the prio_tree before nonlinear list: a vma 26271da177e4SLinus Torvalds * may become nonlinear and be shifted from prio_tree to nonlinear list 26281da177e4SLinus Torvalds * while the lock is dropped; but never shifted from list to prio_tree. 26291da177e4SLinus Torvalds * 26301da177e4SLinus Torvalds * In order to make forward progress despite restarting the search, 26311da177e4SLinus Torvalds * vm_truncate_count is used to mark a vma as now dealt with, so we can 26321da177e4SLinus Torvalds * quickly skip it next time around. Since the prio_tree search only 26331da177e4SLinus Torvalds * shows us those vmas affected by unmapping the range in question, we 26341da177e4SLinus Torvalds * can't efficiently keep all vmas in step with mapping->truncate_count: 26351da177e4SLinus Torvalds * so instead reset them all whenever it wraps back to 0 (then go to 1). 26361da177e4SLinus Torvalds * mapping->truncate_count and vma->vm_truncate_count are protected by 26371da177e4SLinus Torvalds * i_mmap_lock. 26381da177e4SLinus Torvalds * 26391da177e4SLinus Torvalds * In order to make forward progress despite repeatedly restarting some 2640ee39b37bSHugh Dickins * large vma, note the restart_addr from unmap_vmas when it breaks out: 26411da177e4SLinus Torvalds * and restart from that address when we reach that vma again. It might 26421da177e4SLinus Torvalds * have been split or merged, shrunk or extended, but never shifted: so 26431da177e4SLinus Torvalds * restart_addr remains valid so long as it remains in the vma's range. 26441da177e4SLinus Torvalds * unmap_mapping_range forces truncate_count to leap over page-aligned 26451da177e4SLinus Torvalds * values so we can save vma's restart_addr in its truncate_count field. 26461da177e4SLinus Torvalds */ 26471da177e4SLinus Torvalds #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK)) 26481da177e4SLinus Torvalds 26491da177e4SLinus Torvalds static void reset_vma_truncate_counts(struct address_space *mapping) 26501da177e4SLinus Torvalds { 26511da177e4SLinus Torvalds struct vm_area_struct *vma; 26521da177e4SLinus Torvalds struct prio_tree_iter iter; 26531da177e4SLinus Torvalds 26541da177e4SLinus Torvalds vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX) 26551da177e4SLinus Torvalds vma->vm_truncate_count = 0; 26561da177e4SLinus Torvalds list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) 26571da177e4SLinus Torvalds vma->vm_truncate_count = 0; 26581da177e4SLinus Torvalds } 26591da177e4SLinus Torvalds 26601da177e4SLinus Torvalds static int unmap_mapping_range_vma(struct vm_area_struct *vma, 26611da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 26621da177e4SLinus Torvalds struct zap_details *details) 26631da177e4SLinus Torvalds { 26641da177e4SLinus Torvalds unsigned long restart_addr; 26651da177e4SLinus Torvalds int need_break; 26661da177e4SLinus Torvalds 2667d00806b1SNick Piggin /* 2668d00806b1SNick Piggin * files that support invalidating or truncating portions of the 2669d0217ac0SNick Piggin * file from under mmaped areas must have their ->fault function 267083c54070SNick Piggin * return a locked page (and set VM_FAULT_LOCKED in the return). 267183c54070SNick Piggin * This provides synchronisation against concurrent unmapping here. 2672d00806b1SNick Piggin */ 2673d00806b1SNick Piggin 26741da177e4SLinus Torvalds again: 26751da177e4SLinus Torvalds restart_addr = vma->vm_truncate_count; 26761da177e4SLinus Torvalds if (is_restart_addr(restart_addr) && start_addr < restart_addr) { 26771da177e4SLinus Torvalds start_addr = restart_addr; 26781da177e4SLinus Torvalds if (start_addr >= end_addr) { 26791da177e4SLinus Torvalds /* Top of vma has been split off since last time */ 26801da177e4SLinus Torvalds vma->vm_truncate_count = details->truncate_count; 26811da177e4SLinus Torvalds return 0; 26821da177e4SLinus Torvalds } 26831da177e4SLinus Torvalds } 26841da177e4SLinus Torvalds 2685ee39b37bSHugh Dickins restart_addr = zap_page_range(vma, start_addr, 2686ee39b37bSHugh Dickins end_addr - start_addr, details); 268795c354feSNick Piggin need_break = need_resched() || spin_needbreak(details->i_mmap_lock); 26881da177e4SLinus Torvalds 2689ee39b37bSHugh Dickins if (restart_addr >= end_addr) { 26901da177e4SLinus Torvalds /* We have now completed this vma: mark it so */ 26911da177e4SLinus Torvalds vma->vm_truncate_count = details->truncate_count; 26921da177e4SLinus Torvalds if (!need_break) 26931da177e4SLinus Torvalds return 0; 26941da177e4SLinus Torvalds } else { 26951da177e4SLinus Torvalds /* Note restart_addr in vma's truncate_count field */ 2696ee39b37bSHugh Dickins vma->vm_truncate_count = restart_addr; 26971da177e4SLinus Torvalds if (!need_break) 26981da177e4SLinus Torvalds goto again; 26991da177e4SLinus Torvalds } 27001da177e4SLinus Torvalds 27011da177e4SLinus Torvalds spin_unlock(details->i_mmap_lock); 27021da177e4SLinus Torvalds cond_resched(); 27031da177e4SLinus Torvalds spin_lock(details->i_mmap_lock); 27041da177e4SLinus Torvalds return -EINTR; 27051da177e4SLinus Torvalds } 27061da177e4SLinus Torvalds 27071da177e4SLinus Torvalds static inline void unmap_mapping_range_tree(struct prio_tree_root *root, 27081da177e4SLinus Torvalds struct zap_details *details) 27091da177e4SLinus Torvalds { 27101da177e4SLinus Torvalds struct vm_area_struct *vma; 27111da177e4SLinus Torvalds struct prio_tree_iter iter; 27121da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 27131da177e4SLinus Torvalds 27141da177e4SLinus Torvalds restart: 27151da177e4SLinus Torvalds vma_prio_tree_foreach(vma, &iter, root, 27161da177e4SLinus Torvalds details->first_index, details->last_index) { 27171da177e4SLinus Torvalds /* Skip quickly over those we have already dealt with */ 27181da177e4SLinus Torvalds if (vma->vm_truncate_count == details->truncate_count) 27191da177e4SLinus Torvalds continue; 27201da177e4SLinus Torvalds 27211da177e4SLinus Torvalds vba = vma->vm_pgoff; 27221da177e4SLinus Torvalds vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; 27231da177e4SLinus Torvalds /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ 27241da177e4SLinus Torvalds zba = details->first_index; 27251da177e4SLinus Torvalds if (zba < vba) 27261da177e4SLinus Torvalds zba = vba; 27271da177e4SLinus Torvalds zea = details->last_index; 27281da177e4SLinus Torvalds if (zea > vea) 27291da177e4SLinus Torvalds zea = vea; 27301da177e4SLinus Torvalds 27311da177e4SLinus Torvalds if (unmap_mapping_range_vma(vma, 27321da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 27331da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 27341da177e4SLinus Torvalds details) < 0) 27351da177e4SLinus Torvalds goto restart; 27361da177e4SLinus Torvalds } 27371da177e4SLinus Torvalds } 27381da177e4SLinus Torvalds 27391da177e4SLinus Torvalds static inline void unmap_mapping_range_list(struct list_head *head, 27401da177e4SLinus Torvalds struct zap_details *details) 27411da177e4SLinus Torvalds { 27421da177e4SLinus Torvalds struct vm_area_struct *vma; 27431da177e4SLinus Torvalds 27441da177e4SLinus Torvalds /* 27451da177e4SLinus Torvalds * In nonlinear VMAs there is no correspondence between virtual address 27461da177e4SLinus Torvalds * offset and file offset. So we must perform an exhaustive search 27471da177e4SLinus Torvalds * across *all* the pages in each nonlinear VMA, not just the pages 27481da177e4SLinus Torvalds * whose virtual address lies outside the file truncation point. 27491da177e4SLinus Torvalds */ 27501da177e4SLinus Torvalds restart: 27511da177e4SLinus Torvalds list_for_each_entry(vma, head, shared.vm_set.list) { 27521da177e4SLinus Torvalds /* Skip quickly over those we have already dealt with */ 27531da177e4SLinus Torvalds if (vma->vm_truncate_count == details->truncate_count) 27541da177e4SLinus Torvalds continue; 27551da177e4SLinus Torvalds details->nonlinear_vma = vma; 27561da177e4SLinus Torvalds if (unmap_mapping_range_vma(vma, vma->vm_start, 27571da177e4SLinus Torvalds vma->vm_end, details) < 0) 27581da177e4SLinus Torvalds goto restart; 27591da177e4SLinus Torvalds } 27601da177e4SLinus Torvalds } 27611da177e4SLinus Torvalds 27621da177e4SLinus Torvalds /** 276372fd4a35SRobert P. J. Day * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file. 27643d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 27651da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 27661da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 276725d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 27681da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 27691da177e4SLinus Torvalds * partial pages. 27701da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 27711da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 27721da177e4SLinus Torvalds * end of the file. 27731da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 27741da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 27751da177e4SLinus Torvalds */ 27761da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 27771da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 27781da177e4SLinus Torvalds { 27791da177e4SLinus Torvalds struct zap_details details; 27801da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 27811da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 27821da177e4SLinus Torvalds 27831da177e4SLinus Torvalds /* Check for overflow. */ 27841da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 27851da177e4SLinus Torvalds long long holeend = 27861da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 27871da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 27881da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 27891da177e4SLinus Torvalds } 27901da177e4SLinus Torvalds 27911da177e4SLinus Torvalds details.check_mapping = even_cows? NULL: mapping; 27921da177e4SLinus Torvalds details.nonlinear_vma = NULL; 27931da177e4SLinus Torvalds details.first_index = hba; 27941da177e4SLinus Torvalds details.last_index = hba + hlen - 1; 27951da177e4SLinus Torvalds if (details.last_index < details.first_index) 27961da177e4SLinus Torvalds details.last_index = ULONG_MAX; 27971da177e4SLinus Torvalds details.i_mmap_lock = &mapping->i_mmap_lock; 27981da177e4SLinus Torvalds 27992aa15890SMiklos Szeredi mutex_lock(&mapping->unmap_mutex); 28001da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 28011da177e4SLinus Torvalds 2802d00806b1SNick Piggin /* Protect against endless unmapping loops */ 28031da177e4SLinus Torvalds mapping->truncate_count++; 28041da177e4SLinus Torvalds if (unlikely(is_restart_addr(mapping->truncate_count))) { 28051da177e4SLinus Torvalds if (mapping->truncate_count == 0) 28061da177e4SLinus Torvalds reset_vma_truncate_counts(mapping); 28071da177e4SLinus Torvalds mapping->truncate_count++; 28081da177e4SLinus Torvalds } 28091da177e4SLinus Torvalds details.truncate_count = mapping->truncate_count; 28101da177e4SLinus Torvalds 28111da177e4SLinus Torvalds if (unlikely(!prio_tree_empty(&mapping->i_mmap))) 28121da177e4SLinus Torvalds unmap_mapping_range_tree(&mapping->i_mmap, &details); 28131da177e4SLinus Torvalds if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) 28141da177e4SLinus Torvalds unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); 28151da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 28162aa15890SMiklos Szeredi mutex_unlock(&mapping->unmap_mutex); 28171da177e4SLinus Torvalds } 28181da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 28191da177e4SLinus Torvalds 2820f6b3ec23SBadari Pulavarty int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end) 2821f6b3ec23SBadari Pulavarty { 2822f6b3ec23SBadari Pulavarty struct address_space *mapping = inode->i_mapping; 2823f6b3ec23SBadari Pulavarty 2824f6b3ec23SBadari Pulavarty /* 2825f6b3ec23SBadari Pulavarty * If the underlying filesystem is not going to provide 2826f6b3ec23SBadari Pulavarty * a way to truncate a range of blocks (punch a hole) - 2827f6b3ec23SBadari Pulavarty * we should return failure right now. 2828f6b3ec23SBadari Pulavarty */ 2829acfa4380SAl Viro if (!inode->i_op->truncate_range) 2830f6b3ec23SBadari Pulavarty return -ENOSYS; 2831f6b3ec23SBadari Pulavarty 28321b1dcc1bSJes Sorensen mutex_lock(&inode->i_mutex); 2833f6b3ec23SBadari Pulavarty down_write(&inode->i_alloc_sem); 2834f6b3ec23SBadari Pulavarty unmap_mapping_range(mapping, offset, (end - offset), 1); 2835f6b3ec23SBadari Pulavarty truncate_inode_pages_range(mapping, offset, end); 2836d00806b1SNick Piggin unmap_mapping_range(mapping, offset, (end - offset), 1); 2837f6b3ec23SBadari Pulavarty inode->i_op->truncate_range(inode, offset, end); 2838f6b3ec23SBadari Pulavarty up_write(&inode->i_alloc_sem); 28391b1dcc1bSJes Sorensen mutex_unlock(&inode->i_mutex); 2840f6b3ec23SBadari Pulavarty 2841f6b3ec23SBadari Pulavarty return 0; 2842f6b3ec23SBadari Pulavarty } 2843f6b3ec23SBadari Pulavarty 28441da177e4SLinus Torvalds /* 28458f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 28468f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 28478f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 28481da177e4SLinus Torvalds */ 284965500d23SHugh Dickins static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, 285065500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 285130c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 28521da177e4SLinus Torvalds { 28538f4e2101SHugh Dickins spinlock_t *ptl; 28544969c119SAndrea Arcangeli struct page *page, *swapcache = NULL; 285565500d23SHugh Dickins swp_entry_t entry; 28561da177e4SLinus Torvalds pte_t pte; 2857d065bd81SMichel Lespinasse int locked; 285856039efaSKAMEZAWA Hiroyuki struct mem_cgroup *ptr; 2859ad8c2ee8SRik van Riel int exclusive = 0; 286083c54070SNick Piggin int ret = 0; 28611da177e4SLinus Torvalds 28624c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 28638f4e2101SHugh Dickins goto out; 286465500d23SHugh Dickins 286565500d23SHugh Dickins entry = pte_to_swp_entry(orig_pte); 2866d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 28670697212aSChristoph Lameter if (is_migration_entry(entry)) { 28680697212aSChristoph Lameter migration_entry_wait(mm, pmd, address); 2869d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 2870d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 2871d1737fdbSAndi Kleen } else { 2872d1737fdbSAndi Kleen print_bad_pte(vma, address, orig_pte, NULL); 2873d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 2874d1737fdbSAndi Kleen } 28750697212aSChristoph Lameter goto out; 28760697212aSChristoph Lameter } 28770ff92245SShailabh Nagar delayacct_set_flag(DELAYACCT_PF_SWAPIN); 28781da177e4SLinus Torvalds page = lookup_swap_cache(entry); 28791da177e4SLinus Torvalds if (!page) { 2880a5c9b696SHugh Dickins grab_swap_token(mm); /* Contend for token _before_ read-in */ 288102098feaSHugh Dickins page = swapin_readahead(entry, 288202098feaSHugh Dickins GFP_HIGHUSER_MOVABLE, vma, address); 28831da177e4SLinus Torvalds if (!page) { 28841da177e4SLinus Torvalds /* 28858f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 28868f4e2101SHugh Dickins * while we released the pte lock. 28871da177e4SLinus Torvalds */ 28888f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 28891da177e4SLinus Torvalds if (likely(pte_same(*page_table, orig_pte))) 28901da177e4SLinus Torvalds ret = VM_FAULT_OOM; 28910ff92245SShailabh Nagar delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 289265500d23SHugh Dickins goto unlock; 28931da177e4SLinus Torvalds } 28941da177e4SLinus Torvalds 28951da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 28961da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 2897f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 2898d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 289971f72525SWu Fengguang /* 290071f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 290171f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 290271f72525SWu Fengguang */ 2903d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 2904d1737fdbSAndi Kleen delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 29054779cb31SAndi Kleen goto out_release; 29061da177e4SLinus Torvalds } 29071da177e4SLinus Torvalds 2908d065bd81SMichel Lespinasse locked = lock_page_or_retry(page, mm, flags); 290920a1022dSBalbir Singh delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 2910d065bd81SMichel Lespinasse if (!locked) { 2911d065bd81SMichel Lespinasse ret |= VM_FAULT_RETRY; 2912d065bd81SMichel Lespinasse goto out_release; 2913d065bd81SMichel Lespinasse } 29141da177e4SLinus Torvalds 29154969c119SAndrea Arcangeli /* 291631c4a3d3SHugh Dickins * Make sure try_to_free_swap or reuse_swap_page or swapoff did not 291731c4a3d3SHugh Dickins * release the swapcache from under us. The page pin, and pte_same 291831c4a3d3SHugh Dickins * test below, are not enough to exclude that. Even if it is still 291931c4a3d3SHugh Dickins * swapcache, we need to check that the page's swap has not changed. 29204969c119SAndrea Arcangeli */ 292131c4a3d3SHugh Dickins if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) 29224969c119SAndrea Arcangeli goto out_page; 29234969c119SAndrea Arcangeli 29244969c119SAndrea Arcangeli if (ksm_might_need_to_copy(page, vma, address)) { 29254969c119SAndrea Arcangeli swapcache = page; 29264969c119SAndrea Arcangeli page = ksm_does_need_to_copy(page, vma, address); 29274969c119SAndrea Arcangeli 29284969c119SAndrea Arcangeli if (unlikely(!page)) { 29295ad64688SHugh Dickins ret = VM_FAULT_OOM; 29304969c119SAndrea Arcangeli page = swapcache; 29314969c119SAndrea Arcangeli swapcache = NULL; 29324969c119SAndrea Arcangeli goto out_page; 29334969c119SAndrea Arcangeli } 29345ad64688SHugh Dickins } 29355ad64688SHugh Dickins 29362c26fdd7SKAMEZAWA Hiroyuki if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) { 2937073e587eSKAMEZAWA Hiroyuki ret = VM_FAULT_OOM; 2938bc43f75cSJohannes Weiner goto out_page; 2939073e587eSKAMEZAWA Hiroyuki } 2940073e587eSKAMEZAWA Hiroyuki 29411da177e4SLinus Torvalds /* 29428f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 29431da177e4SLinus Torvalds */ 29448f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 29459e9bef07SHugh Dickins if (unlikely(!pte_same(*page_table, orig_pte))) 2946b8107480SKirill Korotaev goto out_nomap; 2947b8107480SKirill Korotaev 2948b8107480SKirill Korotaev if (unlikely(!PageUptodate(page))) { 2949b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 2950b8107480SKirill Korotaev goto out_nomap; 29511da177e4SLinus Torvalds } 29521da177e4SLinus Torvalds 29538c7c6e34SKAMEZAWA Hiroyuki /* 29548c7c6e34SKAMEZAWA Hiroyuki * The page isn't present yet, go ahead with the fault. 29558c7c6e34SKAMEZAWA Hiroyuki * 29568c7c6e34SKAMEZAWA Hiroyuki * Be careful about the sequence of operations here. 29578c7c6e34SKAMEZAWA Hiroyuki * To get its accounting right, reuse_swap_page() must be called 29588c7c6e34SKAMEZAWA Hiroyuki * while the page is counted on swap but not yet in mapcount i.e. 29598c7c6e34SKAMEZAWA Hiroyuki * before page_add_anon_rmap() and swap_free(); try_to_free_swap() 29608c7c6e34SKAMEZAWA Hiroyuki * must be called after the swap_free(), or it will never succeed. 296103f3c433SKAMEZAWA Hiroyuki * Because delete_from_swap_page() may be called by reuse_swap_page(), 296203f3c433SKAMEZAWA Hiroyuki * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry 296303f3c433SKAMEZAWA Hiroyuki * in page->private. In this case, a record in swap_cgroup is silently 296403f3c433SKAMEZAWA Hiroyuki * discarded at swap_free(). 29658c7c6e34SKAMEZAWA Hiroyuki */ 29661da177e4SLinus Torvalds 296734e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 2968b084d435SKAMEZAWA Hiroyuki dec_mm_counter_fast(mm, MM_SWAPENTS); 29691da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 297030c9f3a9SLinus Torvalds if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { 29711da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 297230c9f3a9SLinus Torvalds flags &= ~FAULT_FLAG_WRITE; 29739a5b489bSAndrea Arcangeli ret |= VM_FAULT_WRITE; 2974ad8c2ee8SRik van Riel exclusive = 1; 29751da177e4SLinus Torvalds } 29761da177e4SLinus Torvalds flush_icache_page(vma, page); 29771da177e4SLinus Torvalds set_pte_at(mm, address, page_table, pte); 2978ad8c2ee8SRik van Riel do_page_add_anon_rmap(page, vma, address, exclusive); 297903f3c433SKAMEZAWA Hiroyuki /* It's better to call commit-charge after rmap is established */ 298003f3c433SKAMEZAWA Hiroyuki mem_cgroup_commit_charge_swapin(page, ptr); 29811da177e4SLinus Torvalds 2982c475a8abSHugh Dickins swap_free(entry); 2983b291f000SNick Piggin if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) 2984a2c43eedSHugh Dickins try_to_free_swap(page); 2985c475a8abSHugh Dickins unlock_page(page); 29864969c119SAndrea Arcangeli if (swapcache) { 29874969c119SAndrea Arcangeli /* 29884969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 29894969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 29904969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 29914969c119SAndrea Arcangeli * further safety release the lock after the swap_free 29924969c119SAndrea Arcangeli * so that the swap count won't change under a 29934969c119SAndrea Arcangeli * parallel locked swapcache. 29944969c119SAndrea Arcangeli */ 29954969c119SAndrea Arcangeli unlock_page(swapcache); 29964969c119SAndrea Arcangeli page_cache_release(swapcache); 29974969c119SAndrea Arcangeli } 2998c475a8abSHugh Dickins 299930c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 300061469f1dSHugh Dickins ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); 300161469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 300261469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 30031da177e4SLinus Torvalds goto out; 30041da177e4SLinus Torvalds } 30051da177e4SLinus Torvalds 30061da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 30074b3073e1SRussell King update_mmu_cache(vma, address, page_table); 300865500d23SHugh Dickins unlock: 30098f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 30101da177e4SLinus Torvalds out: 30111da177e4SLinus Torvalds return ret; 3012b8107480SKirill Korotaev out_nomap: 30137a81b88cSKAMEZAWA Hiroyuki mem_cgroup_cancel_charge_swapin(ptr); 30148f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 3015bc43f75cSJohannes Weiner out_page: 3016b8107480SKirill Korotaev unlock_page(page); 30174779cb31SAndi Kleen out_release: 3018b8107480SKirill Korotaev page_cache_release(page); 30194969c119SAndrea Arcangeli if (swapcache) { 30204969c119SAndrea Arcangeli unlock_page(swapcache); 30214969c119SAndrea Arcangeli page_cache_release(swapcache); 30224969c119SAndrea Arcangeli } 302365500d23SHugh Dickins return ret; 30241da177e4SLinus Torvalds } 30251da177e4SLinus Torvalds 30261da177e4SLinus Torvalds /* 30278ca3eb08SLuck, Tony * This is like a special single-page "expand_{down|up}wards()", 30288ca3eb08SLuck, Tony * except we must first make sure that 'address{-|+}PAGE_SIZE' 3029320b2b8dSLinus Torvalds * doesn't hit another vma. 3030320b2b8dSLinus Torvalds */ 3031320b2b8dSLinus Torvalds static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) 3032320b2b8dSLinus Torvalds { 3033320b2b8dSLinus Torvalds address &= PAGE_MASK; 3034320b2b8dSLinus Torvalds if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { 30350e8e50e2SLinus Torvalds struct vm_area_struct *prev = vma->vm_prev; 3036320b2b8dSLinus Torvalds 30370e8e50e2SLinus Torvalds /* 30380e8e50e2SLinus Torvalds * Is there a mapping abutting this one below? 30390e8e50e2SLinus Torvalds * 30400e8e50e2SLinus Torvalds * That's only ok if it's the same stack mapping 30410e8e50e2SLinus Torvalds * that has gotten split.. 30420e8e50e2SLinus Torvalds */ 30430e8e50e2SLinus Torvalds if (prev && prev->vm_end == address) 30440e8e50e2SLinus Torvalds return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; 30450e8e50e2SLinus Torvalds 3046d05f3169SMichal Hocko expand_downwards(vma, address - PAGE_SIZE); 3047320b2b8dSLinus Torvalds } 30488ca3eb08SLuck, Tony if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { 30498ca3eb08SLuck, Tony struct vm_area_struct *next = vma->vm_next; 30508ca3eb08SLuck, Tony 30518ca3eb08SLuck, Tony /* As VM_GROWSDOWN but s/below/above/ */ 30528ca3eb08SLuck, Tony if (next && next->vm_start == address + PAGE_SIZE) 30538ca3eb08SLuck, Tony return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; 30548ca3eb08SLuck, Tony 30558ca3eb08SLuck, Tony expand_upwards(vma, address + PAGE_SIZE); 30568ca3eb08SLuck, Tony } 3057320b2b8dSLinus Torvalds return 0; 3058320b2b8dSLinus Torvalds } 3059320b2b8dSLinus Torvalds 3060320b2b8dSLinus Torvalds /* 30618f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 30628f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 30638f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 30641da177e4SLinus Torvalds */ 306565500d23SHugh Dickins static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, 306665500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 306730c9f3a9SLinus Torvalds unsigned int flags) 30681da177e4SLinus Torvalds { 30698f4e2101SHugh Dickins struct page *page; 30708f4e2101SHugh Dickins spinlock_t *ptl; 30711da177e4SLinus Torvalds pte_t entry; 30721da177e4SLinus Torvalds 30735528f913SLinus Torvalds pte_unmap(page_table); 3074320b2b8dSLinus Torvalds 307511ac5524SLinus Torvalds /* Check if we need to add a guard page to the stack */ 307611ac5524SLinus Torvalds if (check_stack_guard_page(vma, address) < 0) 307711ac5524SLinus Torvalds return VM_FAULT_SIGBUS; 307811ac5524SLinus Torvalds 307911ac5524SLinus Torvalds /* Use the zero-page for reads */ 308062eede62SHugh Dickins if (!(flags & FAULT_FLAG_WRITE)) { 308162eede62SHugh Dickins entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), 308262eede62SHugh Dickins vma->vm_page_prot)); 308311ac5524SLinus Torvalds page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 3084a13ea5b7SHugh Dickins if (!pte_none(*page_table)) 3085a13ea5b7SHugh Dickins goto unlock; 3086a13ea5b7SHugh Dickins goto setpte; 3087a13ea5b7SHugh Dickins } 3088a13ea5b7SHugh Dickins 30891da177e4SLinus Torvalds /* Allocate our own private page. */ 30901da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 309165500d23SHugh Dickins goto oom; 3092769848c0SMel Gorman page = alloc_zeroed_user_highpage_movable(vma, address); 30931da177e4SLinus Torvalds if (!page) 309465500d23SHugh Dickins goto oom; 30950ed361deSNick Piggin __SetPageUptodate(page); 30961da177e4SLinus Torvalds 30972c26fdd7SKAMEZAWA Hiroyuki if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) 30988a9f3ccdSBalbir Singh goto oom_free_page; 30998a9f3ccdSBalbir Singh 310065500d23SHugh Dickins entry = mk_pte(page, vma->vm_page_prot); 31011ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 31021ac0cb5dSHugh Dickins entry = pte_mkwrite(pte_mkdirty(entry)); 31038f4e2101SHugh Dickins 31048f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 31051c2fb7a4SAndrea Arcangeli if (!pte_none(*page_table)) 31068f4e2101SHugh Dickins goto release; 31079ba69294SHugh Dickins 310834e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 31099617d95eSNick Piggin page_add_new_anon_rmap(page, vma, address); 3110a13ea5b7SHugh Dickins setpte: 311165500d23SHugh Dickins set_pte_at(mm, address, page_table, entry); 31121da177e4SLinus Torvalds 31131da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 31144b3073e1SRussell King update_mmu_cache(vma, address, page_table); 311565500d23SHugh Dickins unlock: 31168f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 311783c54070SNick Piggin return 0; 31188f4e2101SHugh Dickins release: 31198a9f3ccdSBalbir Singh mem_cgroup_uncharge_page(page); 31208f4e2101SHugh Dickins page_cache_release(page); 31218f4e2101SHugh Dickins goto unlock; 31228a9f3ccdSBalbir Singh oom_free_page: 31236dbf6d3bSHugh Dickins page_cache_release(page); 312465500d23SHugh Dickins oom: 31251da177e4SLinus Torvalds return VM_FAULT_OOM; 31261da177e4SLinus Torvalds } 31271da177e4SLinus Torvalds 31281da177e4SLinus Torvalds /* 312954cb8821SNick Piggin * __do_fault() tries to create a new page mapping. It aggressively 31301da177e4SLinus Torvalds * tries to share with existing pages, but makes a separate copy if 313154cb8821SNick Piggin * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid 313254cb8821SNick Piggin * the next page fault. 31331da177e4SLinus Torvalds * 31341da177e4SLinus Torvalds * As this is called only for pages that do not currently exist, we 31351da177e4SLinus Torvalds * do not need to flush old virtual caches or the TLB. 31361da177e4SLinus Torvalds * 31378f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 313816abfa08SHugh Dickins * but allow concurrent faults), and pte neither mapped nor locked. 31398f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 31401da177e4SLinus Torvalds */ 314154cb8821SNick Piggin static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma, 314216abfa08SHugh Dickins unsigned long address, pmd_t *pmd, 314354cb8821SNick Piggin pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 31441da177e4SLinus Torvalds { 314516abfa08SHugh Dickins pte_t *page_table; 31468f4e2101SHugh Dickins spinlock_t *ptl; 3147d0217ac0SNick Piggin struct page *page; 31481da177e4SLinus Torvalds pte_t entry; 31491da177e4SLinus Torvalds int anon = 0; 31505b4e655eSKAMEZAWA Hiroyuki int charged = 0; 3151d08b3851SPeter Zijlstra struct page *dirty_page = NULL; 3152d0217ac0SNick Piggin struct vm_fault vmf; 3153d0217ac0SNick Piggin int ret; 3154a200ee18SPeter Zijlstra int page_mkwrite = 0; 315554cb8821SNick Piggin 3156d0217ac0SNick Piggin vmf.virtual_address = (void __user *)(address & PAGE_MASK); 3157d0217ac0SNick Piggin vmf.pgoff = pgoff; 3158d0217ac0SNick Piggin vmf.flags = flags; 3159d0217ac0SNick Piggin vmf.page = NULL; 31601da177e4SLinus Torvalds 3161d0217ac0SNick Piggin ret = vma->vm_ops->fault(vma, &vmf); 3162d065bd81SMichel Lespinasse if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | 3163d065bd81SMichel Lespinasse VM_FAULT_RETRY))) 316483c54070SNick Piggin return ret; 31651da177e4SLinus Torvalds 3166a3b947eaSAndi Kleen if (unlikely(PageHWPoison(vmf.page))) { 3167a3b947eaSAndi Kleen if (ret & VM_FAULT_LOCKED) 3168a3b947eaSAndi Kleen unlock_page(vmf.page); 3169a3b947eaSAndi Kleen return VM_FAULT_HWPOISON; 3170a3b947eaSAndi Kleen } 3171a3b947eaSAndi Kleen 3172d00806b1SNick Piggin /* 3173d0217ac0SNick Piggin * For consistency in subsequent calls, make the faulted page always 3174d00806b1SNick Piggin * locked. 3175d00806b1SNick Piggin */ 317683c54070SNick Piggin if (unlikely(!(ret & VM_FAULT_LOCKED))) 3177d0217ac0SNick Piggin lock_page(vmf.page); 317854cb8821SNick Piggin else 3179d0217ac0SNick Piggin VM_BUG_ON(!PageLocked(vmf.page)); 3180d00806b1SNick Piggin 31811da177e4SLinus Torvalds /* 31821da177e4SLinus Torvalds * Should we do an early C-O-W break? 31831da177e4SLinus Torvalds */ 3184d0217ac0SNick Piggin page = vmf.page; 318554cb8821SNick Piggin if (flags & FAULT_FLAG_WRITE) { 31869637a5efSDavid Howells if (!(vma->vm_flags & VM_SHARED)) { 318754cb8821SNick Piggin anon = 1; 3188d00806b1SNick Piggin if (unlikely(anon_vma_prepare(vma))) { 3189d0217ac0SNick Piggin ret = VM_FAULT_OOM; 319054cb8821SNick Piggin goto out; 3191d00806b1SNick Piggin } 319283c54070SNick Piggin page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, 319383c54070SNick Piggin vma, address); 3194d00806b1SNick Piggin if (!page) { 3195d0217ac0SNick Piggin ret = VM_FAULT_OOM; 319654cb8821SNick Piggin goto out; 3197d00806b1SNick Piggin } 31982c26fdd7SKAMEZAWA Hiroyuki if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) { 31995b4e655eSKAMEZAWA Hiroyuki ret = VM_FAULT_OOM; 32005b4e655eSKAMEZAWA Hiroyuki page_cache_release(page); 32015b4e655eSKAMEZAWA Hiroyuki goto out; 32025b4e655eSKAMEZAWA Hiroyuki } 32035b4e655eSKAMEZAWA Hiroyuki charged = 1; 3204d0217ac0SNick Piggin copy_user_highpage(page, vmf.page, address, vma); 32050ed361deSNick Piggin __SetPageUptodate(page); 32069637a5efSDavid Howells } else { 320754cb8821SNick Piggin /* 320854cb8821SNick Piggin * If the page will be shareable, see if the backing 32099637a5efSDavid Howells * address space wants to know that the page is about 321054cb8821SNick Piggin * to become writable 321154cb8821SNick Piggin */ 321269676147SMark Fasheh if (vma->vm_ops->page_mkwrite) { 3213c2ec175cSNick Piggin int tmp; 3214c2ec175cSNick Piggin 321569676147SMark Fasheh unlock_page(page); 3216b827e496SNick Piggin vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 3217c2ec175cSNick Piggin tmp = vma->vm_ops->page_mkwrite(vma, &vmf); 3218c2ec175cSNick Piggin if (unlikely(tmp & 3219c2ec175cSNick Piggin (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { 3220c2ec175cSNick Piggin ret = tmp; 3221b827e496SNick Piggin goto unwritable_page; 322269676147SMark Fasheh } 3223b827e496SNick Piggin if (unlikely(!(tmp & VM_FAULT_LOCKED))) { 322469676147SMark Fasheh lock_page(page); 3225d0217ac0SNick Piggin if (!page->mapping) { 3226b827e496SNick Piggin ret = 0; /* retry the fault */ 3227b827e496SNick Piggin unlock_page(page); 3228b827e496SNick Piggin goto unwritable_page; 3229d0217ac0SNick Piggin } 3230b827e496SNick Piggin } else 3231b827e496SNick Piggin VM_BUG_ON(!PageLocked(page)); 3232a200ee18SPeter Zijlstra page_mkwrite = 1; 32339637a5efSDavid Howells } 32349637a5efSDavid Howells } 323554cb8821SNick Piggin 32361da177e4SLinus Torvalds } 32371da177e4SLinus Torvalds 32388f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 32391da177e4SLinus Torvalds 32401da177e4SLinus Torvalds /* 32411da177e4SLinus Torvalds * This silly early PAGE_DIRTY setting removes a race 32421da177e4SLinus Torvalds * due to the bad i386 page protection. But it's valid 32431da177e4SLinus Torvalds * for other architectures too. 32441da177e4SLinus Torvalds * 324530c9f3a9SLinus Torvalds * Note that if FAULT_FLAG_WRITE is set, we either now have 32461da177e4SLinus Torvalds * an exclusive copy of the page, or this is a shared mapping, 32471da177e4SLinus Torvalds * so we can make it writable and dirty to avoid having to 32481da177e4SLinus Torvalds * handle that later. 32491da177e4SLinus Torvalds */ 32501da177e4SLinus Torvalds /* Only go through if we didn't race with anybody else... */ 32511c2fb7a4SAndrea Arcangeli if (likely(pte_same(*page_table, orig_pte))) { 3252d00806b1SNick Piggin flush_icache_page(vma, page); 3253d00806b1SNick Piggin entry = mk_pte(page, vma->vm_page_prot); 325454cb8821SNick Piggin if (flags & FAULT_FLAG_WRITE) 32551da177e4SLinus Torvalds entry = maybe_mkwrite(pte_mkdirty(entry), vma); 32561da177e4SLinus Torvalds if (anon) { 325734e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 3258d00806b1SNick Piggin page_add_new_anon_rmap(page, vma, address); 3259f57e88a8SHugh Dickins } else { 326034e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_FILEPAGES); 3261d00806b1SNick Piggin page_add_file_rmap(page); 326254cb8821SNick Piggin if (flags & FAULT_FLAG_WRITE) { 3263d00806b1SNick Piggin dirty_page = page; 3264d08b3851SPeter Zijlstra get_page(dirty_page); 3265d08b3851SPeter Zijlstra } 32664294621fSHugh Dickins } 326764d6519dSLee Schermerhorn set_pte_at(mm, address, page_table, entry); 32681da177e4SLinus Torvalds 3269d00806b1SNick Piggin /* no need to invalidate: a not-present page won't be cached */ 32704b3073e1SRussell King update_mmu_cache(vma, address, page_table); 3271d00806b1SNick Piggin } else { 32725b4e655eSKAMEZAWA Hiroyuki if (charged) 32738a9f3ccdSBalbir Singh mem_cgroup_uncharge_page(page); 3274d00806b1SNick Piggin if (anon) 3275d00806b1SNick Piggin page_cache_release(page); 3276d00806b1SNick Piggin else 327754cb8821SNick Piggin anon = 1; /* no anon but release faulted_page */ 3278d00806b1SNick Piggin } 3279d00806b1SNick Piggin 32808f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 3281d00806b1SNick Piggin 3282d00806b1SNick Piggin out: 3283b827e496SNick Piggin if (dirty_page) { 3284b827e496SNick Piggin struct address_space *mapping = page->mapping; 32858f7b3d15SAnton Salikhmetov 3286b827e496SNick Piggin if (set_page_dirty(dirty_page)) 3287b827e496SNick Piggin page_mkwrite = 1; 3288b827e496SNick Piggin unlock_page(dirty_page); 3289d08b3851SPeter Zijlstra put_page(dirty_page); 3290b827e496SNick Piggin if (page_mkwrite && mapping) { 3291b827e496SNick Piggin /* 3292b827e496SNick Piggin * Some device drivers do not set page.mapping but still 3293b827e496SNick Piggin * dirty their pages 3294b827e496SNick Piggin */ 3295b827e496SNick Piggin balance_dirty_pages_ratelimited(mapping); 3296d08b3851SPeter Zijlstra } 3297d00806b1SNick Piggin 3298b827e496SNick Piggin /* file_update_time outside page_lock */ 3299b827e496SNick Piggin if (vma->vm_file) 3300b827e496SNick Piggin file_update_time(vma->vm_file); 3301b827e496SNick Piggin } else { 3302b827e496SNick Piggin unlock_page(vmf.page); 3303b827e496SNick Piggin if (anon) 3304b827e496SNick Piggin page_cache_release(vmf.page); 3305b827e496SNick Piggin } 3306b827e496SNick Piggin 3307b827e496SNick Piggin return ret; 3308b827e496SNick Piggin 3309b827e496SNick Piggin unwritable_page: 3310b827e496SNick Piggin page_cache_release(page); 331183c54070SNick Piggin return ret; 331254cb8821SNick Piggin } 3313d00806b1SNick Piggin 331454cb8821SNick Piggin static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, 331554cb8821SNick Piggin unsigned long address, pte_t *page_table, pmd_t *pmd, 331630c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 331754cb8821SNick Piggin { 331854cb8821SNick Piggin pgoff_t pgoff = (((address & PAGE_MASK) 33190da7e01fSDean Nelson - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 332054cb8821SNick Piggin 332116abfa08SHugh Dickins pte_unmap(page_table); 332216abfa08SHugh Dickins return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); 332354cb8821SNick Piggin } 332454cb8821SNick Piggin 3325f4b81804SJes Sorensen /* 33261da177e4SLinus Torvalds * Fault of a previously existing named mapping. Repopulate the pte 33271da177e4SLinus Torvalds * from the encoded file_pte if possible. This enables swappable 33281da177e4SLinus Torvalds * nonlinear vmas. 33298f4e2101SHugh Dickins * 33308f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 33318f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 33328f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 33331da177e4SLinus Torvalds */ 3334d0217ac0SNick Piggin static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, 333565500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 333630c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 33371da177e4SLinus Torvalds { 333865500d23SHugh Dickins pgoff_t pgoff; 33391da177e4SLinus Torvalds 334030c9f3a9SLinus Torvalds flags |= FAULT_FLAG_NONLINEAR; 334130c9f3a9SLinus Torvalds 33424c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 334383c54070SNick Piggin return 0; 33441da177e4SLinus Torvalds 33452509ef26SHugh Dickins if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) { 334665500d23SHugh Dickins /* 334765500d23SHugh Dickins * Page table corrupted: show pte and kill process. 334865500d23SHugh Dickins */ 33493dc14741SHugh Dickins print_bad_pte(vma, address, orig_pte, NULL); 3350d99be1a8SHugh Dickins return VM_FAULT_SIGBUS; 335165500d23SHugh Dickins } 335265500d23SHugh Dickins 335365500d23SHugh Dickins pgoff = pte_to_pgoff(orig_pte); 335416abfa08SHugh Dickins return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); 33551da177e4SLinus Torvalds } 33561da177e4SLinus Torvalds 33571da177e4SLinus Torvalds /* 33581da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 33591da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 33601da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 33611da177e4SLinus Torvalds * 33621da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 33631da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 33641da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 33651da177e4SLinus Torvalds * 3366c74df32cSHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 3367c74df32cSHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 3368c74df32cSHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 33691da177e4SLinus Torvalds */ 337071e3aac0SAndrea Arcangeli int handle_pte_fault(struct mm_struct *mm, 33711da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long address, 337230c9f3a9SLinus Torvalds pte_t *pte, pmd_t *pmd, unsigned int flags) 33731da177e4SLinus Torvalds { 33741da177e4SLinus Torvalds pte_t entry; 33758f4e2101SHugh Dickins spinlock_t *ptl; 33761da177e4SLinus Torvalds 33778dab5241SBenjamin Herrenschmidt entry = *pte; 33781da177e4SLinus Torvalds if (!pte_present(entry)) { 337965500d23SHugh Dickins if (pte_none(entry)) { 3380f4b81804SJes Sorensen if (vma->vm_ops) { 33813c18ddd1SNick Piggin if (likely(vma->vm_ops->fault)) 338254cb8821SNick Piggin return do_linear_fault(mm, vma, address, 338330c9f3a9SLinus Torvalds pte, pmd, flags, entry); 3384f4b81804SJes Sorensen } 3385f4b81804SJes Sorensen return do_anonymous_page(mm, vma, address, 338630c9f3a9SLinus Torvalds pte, pmd, flags); 338765500d23SHugh Dickins } 33881da177e4SLinus Torvalds if (pte_file(entry)) 3389d0217ac0SNick Piggin return do_nonlinear_fault(mm, vma, address, 339030c9f3a9SLinus Torvalds pte, pmd, flags, entry); 339165500d23SHugh Dickins return do_swap_page(mm, vma, address, 339230c9f3a9SLinus Torvalds pte, pmd, flags, entry); 33931da177e4SLinus Torvalds } 33941da177e4SLinus Torvalds 33954c21e2f2SHugh Dickins ptl = pte_lockptr(mm, pmd); 33968f4e2101SHugh Dickins spin_lock(ptl); 33978f4e2101SHugh Dickins if (unlikely(!pte_same(*pte, entry))) 33988f4e2101SHugh Dickins goto unlock; 339930c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 34001da177e4SLinus Torvalds if (!pte_write(entry)) 34018f4e2101SHugh Dickins return do_wp_page(mm, vma, address, 34028f4e2101SHugh Dickins pte, pmd, ptl, entry); 34031da177e4SLinus Torvalds entry = pte_mkdirty(entry); 34041da177e4SLinus Torvalds } 34051da177e4SLinus Torvalds entry = pte_mkyoung(entry); 340630c9f3a9SLinus Torvalds if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { 34074b3073e1SRussell King update_mmu_cache(vma, address, pte); 34081a44e149SAndrea Arcangeli } else { 34091a44e149SAndrea Arcangeli /* 34101a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 34111a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 34121a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 34131a44e149SAndrea Arcangeli * with threads. 34141a44e149SAndrea Arcangeli */ 341530c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) 341661c77326SShaohua Li flush_tlb_fix_spurious_fault(vma, address); 34171a44e149SAndrea Arcangeli } 34188f4e2101SHugh Dickins unlock: 34198f4e2101SHugh Dickins pte_unmap_unlock(pte, ptl); 342083c54070SNick Piggin return 0; 34211da177e4SLinus Torvalds } 34221da177e4SLinus Torvalds 34231da177e4SLinus Torvalds /* 34241da177e4SLinus Torvalds * By the time we get here, we already hold the mm semaphore 34251da177e4SLinus Torvalds */ 342683c54070SNick Piggin int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3427d06063ccSLinus Torvalds unsigned long address, unsigned int flags) 34281da177e4SLinus Torvalds { 34291da177e4SLinus Torvalds pgd_t *pgd; 34301da177e4SLinus Torvalds pud_t *pud; 34311da177e4SLinus Torvalds pmd_t *pmd; 34321da177e4SLinus Torvalds pte_t *pte; 34331da177e4SLinus Torvalds 34341da177e4SLinus Torvalds __set_current_state(TASK_RUNNING); 34351da177e4SLinus Torvalds 3436f8891e5eSChristoph Lameter count_vm_event(PGFAULT); 34371da177e4SLinus Torvalds 343834e55232SKAMEZAWA Hiroyuki /* do counter updates before entering really critical section. */ 343934e55232SKAMEZAWA Hiroyuki check_sync_rss_stat(current); 344034e55232SKAMEZAWA Hiroyuki 3441ac9b9c66SHugh Dickins if (unlikely(is_vm_hugetlb_page(vma))) 344230c9f3a9SLinus Torvalds return hugetlb_fault(mm, vma, address, flags); 34431da177e4SLinus Torvalds 34441da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 34451da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, address); 34461da177e4SLinus Torvalds if (!pud) 3447c74df32cSHugh Dickins return VM_FAULT_OOM; 34481da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, address); 34491da177e4SLinus Torvalds if (!pmd) 3450c74df32cSHugh Dickins return VM_FAULT_OOM; 345171e3aac0SAndrea Arcangeli if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) { 345271e3aac0SAndrea Arcangeli if (!vma->vm_ops) 345371e3aac0SAndrea Arcangeli return do_huge_pmd_anonymous_page(mm, vma, address, 345471e3aac0SAndrea Arcangeli pmd, flags); 345571e3aac0SAndrea Arcangeli } else { 345671e3aac0SAndrea Arcangeli pmd_t orig_pmd = *pmd; 345771e3aac0SAndrea Arcangeli barrier(); 345871e3aac0SAndrea Arcangeli if (pmd_trans_huge(orig_pmd)) { 345971e3aac0SAndrea Arcangeli if (flags & FAULT_FLAG_WRITE && 346071e3aac0SAndrea Arcangeli !pmd_write(orig_pmd) && 346171e3aac0SAndrea Arcangeli !pmd_trans_splitting(orig_pmd)) 346271e3aac0SAndrea Arcangeli return do_huge_pmd_wp_page(mm, vma, address, 346371e3aac0SAndrea Arcangeli pmd, orig_pmd); 346471e3aac0SAndrea Arcangeli return 0; 346571e3aac0SAndrea Arcangeli } 346671e3aac0SAndrea Arcangeli } 346771e3aac0SAndrea Arcangeli 346871e3aac0SAndrea Arcangeli /* 346971e3aac0SAndrea Arcangeli * Use __pte_alloc instead of pte_alloc_map, because we can't 347071e3aac0SAndrea Arcangeli * run pte_offset_map on the pmd, if an huge pmd could 347171e3aac0SAndrea Arcangeli * materialize from under us from a different thread. 347271e3aac0SAndrea Arcangeli */ 3473cc03638dSMel Gorman if (unlikely(pmd_none(*pmd)) && __pte_alloc(mm, vma, pmd, address)) 3474c74df32cSHugh Dickins return VM_FAULT_OOM; 347571e3aac0SAndrea Arcangeli /* if an huge pmd materialized from under us just retry later */ 347671e3aac0SAndrea Arcangeli if (unlikely(pmd_trans_huge(*pmd))) 347771e3aac0SAndrea Arcangeli return 0; 347871e3aac0SAndrea Arcangeli /* 347971e3aac0SAndrea Arcangeli * A regular pmd is established and it can't morph into a huge pmd 348071e3aac0SAndrea Arcangeli * from under us anymore at this point because we hold the mmap_sem 348171e3aac0SAndrea Arcangeli * read mode and khugepaged takes it in write mode. So now it's 348271e3aac0SAndrea Arcangeli * safe to run pte_offset_map(). 348371e3aac0SAndrea Arcangeli */ 348471e3aac0SAndrea Arcangeli pte = pte_offset_map(pmd, address); 34851da177e4SLinus Torvalds 348630c9f3a9SLinus Torvalds return handle_pte_fault(mm, vma, address, pte, pmd, flags); 34871da177e4SLinus Torvalds } 34881da177e4SLinus Torvalds 34891da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 34901da177e4SLinus Torvalds /* 34911da177e4SLinus Torvalds * Allocate page upper directory. 3492872fec16SHugh Dickins * We've already handled the fast-path in-line. 34931da177e4SLinus Torvalds */ 34941bb3630eSHugh Dickins int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 34951da177e4SLinus Torvalds { 3496c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 3497c74df32cSHugh Dickins if (!new) 34981bb3630eSHugh Dickins return -ENOMEM; 34991da177e4SLinus Torvalds 3500362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3501362a61adSNick Piggin 3502872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 35031bb3630eSHugh Dickins if (pgd_present(*pgd)) /* Another has populated it */ 35045e541973SBenjamin Herrenschmidt pud_free(mm, new); 35051bb3630eSHugh Dickins else 35061da177e4SLinus Torvalds pgd_populate(mm, pgd, new); 3507872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 35081bb3630eSHugh Dickins return 0; 35091da177e4SLinus Torvalds } 35101da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 35111da177e4SLinus Torvalds 35121da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 35131da177e4SLinus Torvalds /* 35141da177e4SLinus Torvalds * Allocate page middle directory. 3515872fec16SHugh Dickins * We've already handled the fast-path in-line. 35161da177e4SLinus Torvalds */ 35171bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 35181da177e4SLinus Torvalds { 3519c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 3520c74df32cSHugh Dickins if (!new) 35211bb3630eSHugh Dickins return -ENOMEM; 35221da177e4SLinus Torvalds 3523362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3524362a61adSNick Piggin 3525872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 35261da177e4SLinus Torvalds #ifndef __ARCH_HAS_4LEVEL_HACK 35271bb3630eSHugh Dickins if (pud_present(*pud)) /* Another has populated it */ 35285e541973SBenjamin Herrenschmidt pmd_free(mm, new); 35291bb3630eSHugh Dickins else 35301da177e4SLinus Torvalds pud_populate(mm, pud, new); 35311da177e4SLinus Torvalds #else 35321bb3630eSHugh Dickins if (pgd_present(*pud)) /* Another has populated it */ 35335e541973SBenjamin Herrenschmidt pmd_free(mm, new); 35341bb3630eSHugh Dickins else 35351da177e4SLinus Torvalds pgd_populate(mm, pud, new); 35361da177e4SLinus Torvalds #endif /* __ARCH_HAS_4LEVEL_HACK */ 3537872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 35381bb3630eSHugh Dickins return 0; 35391da177e4SLinus Torvalds } 35401da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 35411da177e4SLinus Torvalds 35421da177e4SLinus Torvalds int make_pages_present(unsigned long addr, unsigned long end) 35431da177e4SLinus Torvalds { 35441da177e4SLinus Torvalds int ret, len, write; 35451da177e4SLinus Torvalds struct vm_area_struct * vma; 35461da177e4SLinus Torvalds 35471da177e4SLinus Torvalds vma = find_vma(current->mm, addr); 35481da177e4SLinus Torvalds if (!vma) 3549a477097dSKOSAKI Motohiro return -ENOMEM; 35505ecfda04SMichel Lespinasse /* 35515ecfda04SMichel Lespinasse * We want to touch writable mappings with a write fault in order 35525ecfda04SMichel Lespinasse * to break COW, except for shared mappings because these don't COW 35535ecfda04SMichel Lespinasse * and we would not want to dirty them for nothing. 35545ecfda04SMichel Lespinasse */ 35555ecfda04SMichel Lespinasse write = (vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE; 35565bcb28b1SEric Sesterhenn BUG_ON(addr >= end); 35575bcb28b1SEric Sesterhenn BUG_ON(end > vma->vm_end); 355868e116a3SRolf Eike Beer len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE; 35591da177e4SLinus Torvalds ret = get_user_pages(current, current->mm, addr, 35601da177e4SLinus Torvalds len, write, 0, NULL, NULL); 3561c11d69d8SLee Schermerhorn if (ret < 0) 35621da177e4SLinus Torvalds return ret; 35639978ad58SLee Schermerhorn return ret == len ? 0 : -EFAULT; 35641da177e4SLinus Torvalds } 35651da177e4SLinus Torvalds 35661da177e4SLinus Torvalds #if !defined(__HAVE_ARCH_GATE_AREA) 35671da177e4SLinus Torvalds 35681da177e4SLinus Torvalds #if defined(AT_SYSINFO_EHDR) 35695ce7852cSAdrian Bunk static struct vm_area_struct gate_vma; 35701da177e4SLinus Torvalds 35711da177e4SLinus Torvalds static int __init gate_vma_init(void) 35721da177e4SLinus Torvalds { 35731da177e4SLinus Torvalds gate_vma.vm_mm = NULL; 35741da177e4SLinus Torvalds gate_vma.vm_start = FIXADDR_USER_START; 35751da177e4SLinus Torvalds gate_vma.vm_end = FIXADDR_USER_END; 3576b6558c4aSRoland McGrath gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; 3577b6558c4aSRoland McGrath gate_vma.vm_page_prot = __P101; 3578f47aef55SRoland McGrath /* 3579f47aef55SRoland McGrath * Make sure the vDSO gets into every core dump. 3580f47aef55SRoland McGrath * Dumping its contents makes post-mortem fully interpretable later 3581f47aef55SRoland McGrath * without matching up the same kernel and hardware config to see 3582f47aef55SRoland McGrath * what PC values meant. 3583f47aef55SRoland McGrath */ 3584f47aef55SRoland McGrath gate_vma.vm_flags |= VM_ALWAYSDUMP; 35851da177e4SLinus Torvalds return 0; 35861da177e4SLinus Torvalds } 35871da177e4SLinus Torvalds __initcall(gate_vma_init); 35881da177e4SLinus Torvalds #endif 35891da177e4SLinus Torvalds 359031db58b3SStephen Wilson struct vm_area_struct *get_gate_vma(struct mm_struct *mm) 35911da177e4SLinus Torvalds { 35921da177e4SLinus Torvalds #ifdef AT_SYSINFO_EHDR 35931da177e4SLinus Torvalds return &gate_vma; 35941da177e4SLinus Torvalds #else 35951da177e4SLinus Torvalds return NULL; 35961da177e4SLinus Torvalds #endif 35971da177e4SLinus Torvalds } 35981da177e4SLinus Torvalds 3599cae5d390SStephen Wilson int in_gate_area_no_mm(unsigned long addr) 36001da177e4SLinus Torvalds { 36011da177e4SLinus Torvalds #ifdef AT_SYSINFO_EHDR 36021da177e4SLinus Torvalds if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) 36031da177e4SLinus Torvalds return 1; 36041da177e4SLinus Torvalds #endif 36051da177e4SLinus Torvalds return 0; 36061da177e4SLinus Torvalds } 36071da177e4SLinus Torvalds 36081da177e4SLinus Torvalds #endif /* __HAVE_ARCH_GATE_AREA */ 36090ec76a11SDavid Howells 36101b36ba81SNamhyung Kim static int __follow_pte(struct mm_struct *mm, unsigned long address, 3611f8ad0f49SJohannes Weiner pte_t **ptepp, spinlock_t **ptlp) 3612f8ad0f49SJohannes Weiner { 3613f8ad0f49SJohannes Weiner pgd_t *pgd; 3614f8ad0f49SJohannes Weiner pud_t *pud; 3615f8ad0f49SJohannes Weiner pmd_t *pmd; 3616f8ad0f49SJohannes Weiner pte_t *ptep; 3617f8ad0f49SJohannes Weiner 3618f8ad0f49SJohannes Weiner pgd = pgd_offset(mm, address); 3619f8ad0f49SJohannes Weiner if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 3620f8ad0f49SJohannes Weiner goto out; 3621f8ad0f49SJohannes Weiner 3622f8ad0f49SJohannes Weiner pud = pud_offset(pgd, address); 3623f8ad0f49SJohannes Weiner if (pud_none(*pud) || unlikely(pud_bad(*pud))) 3624f8ad0f49SJohannes Weiner goto out; 3625f8ad0f49SJohannes Weiner 3626f8ad0f49SJohannes Weiner pmd = pmd_offset(pud, address); 3627f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 3628f8ad0f49SJohannes Weiner if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) 3629f8ad0f49SJohannes Weiner goto out; 3630f8ad0f49SJohannes Weiner 3631f8ad0f49SJohannes Weiner /* We cannot handle huge page PFN maps. Luckily they don't exist. */ 3632f8ad0f49SJohannes Weiner if (pmd_huge(*pmd)) 3633f8ad0f49SJohannes Weiner goto out; 3634f8ad0f49SJohannes Weiner 3635f8ad0f49SJohannes Weiner ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 3636f8ad0f49SJohannes Weiner if (!ptep) 3637f8ad0f49SJohannes Weiner goto out; 3638f8ad0f49SJohannes Weiner if (!pte_present(*ptep)) 3639f8ad0f49SJohannes Weiner goto unlock; 3640f8ad0f49SJohannes Weiner *ptepp = ptep; 3641f8ad0f49SJohannes Weiner return 0; 3642f8ad0f49SJohannes Weiner unlock: 3643f8ad0f49SJohannes Weiner pte_unmap_unlock(ptep, *ptlp); 3644f8ad0f49SJohannes Weiner out: 3645f8ad0f49SJohannes Weiner return -EINVAL; 3646f8ad0f49SJohannes Weiner } 3647f8ad0f49SJohannes Weiner 36481b36ba81SNamhyung Kim static inline int follow_pte(struct mm_struct *mm, unsigned long address, 36491b36ba81SNamhyung Kim pte_t **ptepp, spinlock_t **ptlp) 36501b36ba81SNamhyung Kim { 36511b36ba81SNamhyung Kim int res; 36521b36ba81SNamhyung Kim 36531b36ba81SNamhyung Kim /* (void) is needed to make gcc happy */ 36541b36ba81SNamhyung Kim (void) __cond_lock(*ptlp, 36551b36ba81SNamhyung Kim !(res = __follow_pte(mm, address, ptepp, ptlp))); 36561b36ba81SNamhyung Kim return res; 36571b36ba81SNamhyung Kim } 36581b36ba81SNamhyung Kim 36593b6748e2SJohannes Weiner /** 36603b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 36613b6748e2SJohannes Weiner * @vma: memory mapping 36623b6748e2SJohannes Weiner * @address: user virtual address 36633b6748e2SJohannes Weiner * @pfn: location to store found PFN 36643b6748e2SJohannes Weiner * 36653b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 36663b6748e2SJohannes Weiner * 36673b6748e2SJohannes Weiner * Returns zero and the pfn at @pfn on success, -ve otherwise. 36683b6748e2SJohannes Weiner */ 36693b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 36703b6748e2SJohannes Weiner unsigned long *pfn) 36713b6748e2SJohannes Weiner { 36723b6748e2SJohannes Weiner int ret = -EINVAL; 36733b6748e2SJohannes Weiner spinlock_t *ptl; 36743b6748e2SJohannes Weiner pte_t *ptep; 36753b6748e2SJohannes Weiner 36763b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 36773b6748e2SJohannes Weiner return ret; 36783b6748e2SJohannes Weiner 36793b6748e2SJohannes Weiner ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 36803b6748e2SJohannes Weiner if (ret) 36813b6748e2SJohannes Weiner return ret; 36823b6748e2SJohannes Weiner *pfn = pte_pfn(*ptep); 36833b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 36843b6748e2SJohannes Weiner return 0; 36853b6748e2SJohannes Weiner } 36863b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 36873b6748e2SJohannes Weiner 368828b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 3689d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 369028b2ee20SRik van Riel unsigned long address, unsigned int flags, 3691d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 369228b2ee20SRik van Riel { 369303668a4dSJohannes Weiner int ret = -EINVAL; 369428b2ee20SRik van Riel pte_t *ptep, pte; 369528b2ee20SRik van Riel spinlock_t *ptl; 369628b2ee20SRik van Riel 3697d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 3698d87fe660Svenkatesh.pallipadi@intel.com goto out; 369928b2ee20SRik van Riel 370003668a4dSJohannes Weiner if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 3701d87fe660Svenkatesh.pallipadi@intel.com goto out; 370228b2ee20SRik van Riel pte = *ptep; 370303668a4dSJohannes Weiner 370428b2ee20SRik van Riel if ((flags & FOLL_WRITE) && !pte_write(pte)) 370528b2ee20SRik van Riel goto unlock; 370628b2ee20SRik van Riel 370728b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 370803668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 370928b2ee20SRik van Riel 371003668a4dSJohannes Weiner ret = 0; 371128b2ee20SRik van Riel unlock: 371228b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 371328b2ee20SRik van Riel out: 3714d87fe660Svenkatesh.pallipadi@intel.com return ret; 371528b2ee20SRik van Riel } 371628b2ee20SRik van Riel 371728b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 371828b2ee20SRik van Riel void *buf, int len, int write) 371928b2ee20SRik van Riel { 372028b2ee20SRik van Riel resource_size_t phys_addr; 372128b2ee20SRik van Riel unsigned long prot = 0; 37222bc7273bSKOSAKI Motohiro void __iomem *maddr; 372328b2ee20SRik van Riel int offset = addr & (PAGE_SIZE-1); 372428b2ee20SRik van Riel 3725d87fe660Svenkatesh.pallipadi@intel.com if (follow_phys(vma, addr, write, &prot, &phys_addr)) 372628b2ee20SRik van Riel return -EINVAL; 372728b2ee20SRik van Riel 372828b2ee20SRik van Riel maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot); 372928b2ee20SRik van Riel if (write) 373028b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 373128b2ee20SRik van Riel else 373228b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 373328b2ee20SRik van Riel iounmap(maddr); 373428b2ee20SRik van Riel 373528b2ee20SRik van Riel return len; 373628b2ee20SRik van Riel } 373728b2ee20SRik van Riel #endif 373828b2ee20SRik van Riel 37390ec76a11SDavid Howells /* 3740206cb636SStephen Wilson * Access another process' address space as given in mm. If non-NULL, use the 3741206cb636SStephen Wilson * given task for page fault accounting. 37420ec76a11SDavid Howells */ 3743206cb636SStephen Wilson static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 3744206cb636SStephen Wilson unsigned long addr, void *buf, int len, int write) 37450ec76a11SDavid Howells { 37460ec76a11SDavid Howells struct vm_area_struct *vma; 37470ec76a11SDavid Howells void *old_buf = buf; 37480ec76a11SDavid Howells 37490ec76a11SDavid Howells down_read(&mm->mmap_sem); 3750183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 37510ec76a11SDavid Howells while (len) { 37520ec76a11SDavid Howells int bytes, ret, offset; 37530ec76a11SDavid Howells void *maddr; 375428b2ee20SRik van Riel struct page *page = NULL; 37550ec76a11SDavid Howells 37560ec76a11SDavid Howells ret = get_user_pages(tsk, mm, addr, 1, 37570ec76a11SDavid Howells write, 1, &page, &vma); 375828b2ee20SRik van Riel if (ret <= 0) { 375928b2ee20SRik van Riel /* 376028b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 376128b2ee20SRik van Riel * we can access using slightly different code. 376228b2ee20SRik van Riel */ 376328b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 376428b2ee20SRik van Riel vma = find_vma(mm, addr); 3765fe936dfcSMichael Ellerman if (!vma || vma->vm_start > addr) 37660ec76a11SDavid Howells break; 376728b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 376828b2ee20SRik van Riel ret = vma->vm_ops->access(vma, addr, buf, 376928b2ee20SRik van Riel len, write); 377028b2ee20SRik van Riel if (ret <= 0) 377128b2ee20SRik van Riel #endif 377228b2ee20SRik van Riel break; 377328b2ee20SRik van Riel bytes = ret; 377428b2ee20SRik van Riel } else { 37750ec76a11SDavid Howells bytes = len; 37760ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 37770ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 37780ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 37790ec76a11SDavid Howells 37800ec76a11SDavid Howells maddr = kmap(page); 37810ec76a11SDavid Howells if (write) { 37820ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 37830ec76a11SDavid Howells maddr + offset, buf, bytes); 37840ec76a11SDavid Howells set_page_dirty_lock(page); 37850ec76a11SDavid Howells } else { 37860ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 37870ec76a11SDavid Howells buf, maddr + offset, bytes); 37880ec76a11SDavid Howells } 37890ec76a11SDavid Howells kunmap(page); 37900ec76a11SDavid Howells page_cache_release(page); 379128b2ee20SRik van Riel } 37920ec76a11SDavid Howells len -= bytes; 37930ec76a11SDavid Howells buf += bytes; 37940ec76a11SDavid Howells addr += bytes; 37950ec76a11SDavid Howells } 37960ec76a11SDavid Howells up_read(&mm->mmap_sem); 37970ec76a11SDavid Howells 37980ec76a11SDavid Howells return buf - old_buf; 37990ec76a11SDavid Howells } 380003252919SAndi Kleen 38015ddd36b9SStephen Wilson /** 3802ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 38035ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 38045ddd36b9SStephen Wilson * @addr: start address to access 38055ddd36b9SStephen Wilson * @buf: source or destination buffer 38065ddd36b9SStephen Wilson * @len: number of bytes to transfer 38075ddd36b9SStephen Wilson * @write: whether the access is a write 38085ddd36b9SStephen Wilson * 38095ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 38105ddd36b9SStephen Wilson */ 38115ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 38125ddd36b9SStephen Wilson void *buf, int len, int write) 38135ddd36b9SStephen Wilson { 38145ddd36b9SStephen Wilson return __access_remote_vm(NULL, mm, addr, buf, len, write); 38155ddd36b9SStephen Wilson } 38165ddd36b9SStephen Wilson 381703252919SAndi Kleen /* 3818206cb636SStephen Wilson * Access another process' address space. 3819206cb636SStephen Wilson * Source/target buffer must be kernel space, 3820206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 3821206cb636SStephen Wilson */ 3822206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 3823206cb636SStephen Wilson void *buf, int len, int write) 3824206cb636SStephen Wilson { 3825206cb636SStephen Wilson struct mm_struct *mm; 3826206cb636SStephen Wilson int ret; 3827206cb636SStephen Wilson 3828206cb636SStephen Wilson mm = get_task_mm(tsk); 3829206cb636SStephen Wilson if (!mm) 3830206cb636SStephen Wilson return 0; 3831206cb636SStephen Wilson 3832206cb636SStephen Wilson ret = __access_remote_vm(tsk, mm, addr, buf, len, write); 3833206cb636SStephen Wilson mmput(mm); 3834206cb636SStephen Wilson 3835206cb636SStephen Wilson return ret; 3836206cb636SStephen Wilson } 3837206cb636SStephen Wilson 383803252919SAndi Kleen /* 383903252919SAndi Kleen * Print the name of a VMA. 384003252919SAndi Kleen */ 384103252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 384203252919SAndi Kleen { 384303252919SAndi Kleen struct mm_struct *mm = current->mm; 384403252919SAndi Kleen struct vm_area_struct *vma; 384503252919SAndi Kleen 3846e8bff74aSIngo Molnar /* 3847e8bff74aSIngo Molnar * Do not print if we are in atomic 3848e8bff74aSIngo Molnar * contexts (in exception stacks, etc.): 3849e8bff74aSIngo Molnar */ 3850e8bff74aSIngo Molnar if (preempt_count()) 3851e8bff74aSIngo Molnar return; 3852e8bff74aSIngo Molnar 385303252919SAndi Kleen down_read(&mm->mmap_sem); 385403252919SAndi Kleen vma = find_vma(mm, ip); 385503252919SAndi Kleen if (vma && vma->vm_file) { 385603252919SAndi Kleen struct file *f = vma->vm_file; 385703252919SAndi Kleen char *buf = (char *)__get_free_page(GFP_KERNEL); 385803252919SAndi Kleen if (buf) { 385903252919SAndi Kleen char *p, *s; 386003252919SAndi Kleen 3861cf28b486SJan Blunck p = d_path(&f->f_path, buf, PAGE_SIZE); 386203252919SAndi Kleen if (IS_ERR(p)) 386303252919SAndi Kleen p = "?"; 386403252919SAndi Kleen s = strrchr(p, '/'); 386503252919SAndi Kleen if (s) 386603252919SAndi Kleen p = s+1; 386703252919SAndi Kleen printk("%s%s[%lx+%lx]", prefix, p, 386803252919SAndi Kleen vma->vm_start, 386903252919SAndi Kleen vma->vm_end - vma->vm_start); 387003252919SAndi Kleen free_page((unsigned long)buf); 387103252919SAndi Kleen } 387203252919SAndi Kleen } 387303252919SAndi Kleen up_read(¤t->mm->mmap_sem); 387403252919SAndi Kleen } 38753ee1afa3SNick Piggin 38763ee1afa3SNick Piggin #ifdef CONFIG_PROVE_LOCKING 38773ee1afa3SNick Piggin void might_fault(void) 38783ee1afa3SNick Piggin { 387995156f00SPeter Zijlstra /* 388095156f00SPeter Zijlstra * Some code (nfs/sunrpc) uses socket ops on kernel memory while 388195156f00SPeter Zijlstra * holding the mmap_sem, this is safe because kernel memory doesn't 388295156f00SPeter Zijlstra * get paged out, therefore we'll never actually fault, and the 388395156f00SPeter Zijlstra * below annotations will generate false positives. 388495156f00SPeter Zijlstra */ 388595156f00SPeter Zijlstra if (segment_eq(get_fs(), KERNEL_DS)) 388695156f00SPeter Zijlstra return; 388795156f00SPeter Zijlstra 38883ee1afa3SNick Piggin might_sleep(); 38893ee1afa3SNick Piggin /* 38903ee1afa3SNick Piggin * it would be nicer only to annotate paths which are not under 38913ee1afa3SNick Piggin * pagefault_disable, however that requires a larger audit and 38923ee1afa3SNick Piggin * providing helpers like get_user_atomic. 38933ee1afa3SNick Piggin */ 38943ee1afa3SNick Piggin if (!in_atomic() && current->mm) 38953ee1afa3SNick Piggin might_lock_read(¤t->mm->mmap_sem); 38963ee1afa3SNick Piggin } 38973ee1afa3SNick Piggin EXPORT_SYMBOL(might_fault); 38983ee1afa3SNick Piggin #endif 389947ad8475SAndrea Arcangeli 390047ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 390147ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 390247ad8475SAndrea Arcangeli unsigned long addr, 390347ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 390447ad8475SAndrea Arcangeli { 390547ad8475SAndrea Arcangeli int i; 390647ad8475SAndrea Arcangeli struct page *p = page; 390747ad8475SAndrea Arcangeli 390847ad8475SAndrea Arcangeli might_sleep(); 390947ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; 391047ad8475SAndrea Arcangeli i++, p = mem_map_next(p, page, i)) { 391147ad8475SAndrea Arcangeli cond_resched(); 391247ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 391347ad8475SAndrea Arcangeli } 391447ad8475SAndrea Arcangeli } 391547ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 391647ad8475SAndrea Arcangeli unsigned long addr, unsigned int pages_per_huge_page) 391747ad8475SAndrea Arcangeli { 391847ad8475SAndrea Arcangeli int i; 391947ad8475SAndrea Arcangeli 392047ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 392147ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 392247ad8475SAndrea Arcangeli return; 392347ad8475SAndrea Arcangeli } 392447ad8475SAndrea Arcangeli 392547ad8475SAndrea Arcangeli might_sleep(); 392647ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 392747ad8475SAndrea Arcangeli cond_resched(); 392847ad8475SAndrea Arcangeli clear_user_highpage(page + i, addr + i * PAGE_SIZE); 392947ad8475SAndrea Arcangeli } 393047ad8475SAndrea Arcangeli } 393147ad8475SAndrea Arcangeli 393247ad8475SAndrea Arcangeli static void copy_user_gigantic_page(struct page *dst, struct page *src, 393347ad8475SAndrea Arcangeli unsigned long addr, 393447ad8475SAndrea Arcangeli struct vm_area_struct *vma, 393547ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 393647ad8475SAndrea Arcangeli { 393747ad8475SAndrea Arcangeli int i; 393847ad8475SAndrea Arcangeli struct page *dst_base = dst; 393947ad8475SAndrea Arcangeli struct page *src_base = src; 394047ad8475SAndrea Arcangeli 394147ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; ) { 394247ad8475SAndrea Arcangeli cond_resched(); 394347ad8475SAndrea Arcangeli copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); 394447ad8475SAndrea Arcangeli 394547ad8475SAndrea Arcangeli i++; 394647ad8475SAndrea Arcangeli dst = mem_map_next(dst, dst_base, i); 394747ad8475SAndrea Arcangeli src = mem_map_next(src, src_base, i); 394847ad8475SAndrea Arcangeli } 394947ad8475SAndrea Arcangeli } 395047ad8475SAndrea Arcangeli 395147ad8475SAndrea Arcangeli void copy_user_huge_page(struct page *dst, struct page *src, 395247ad8475SAndrea Arcangeli unsigned long addr, struct vm_area_struct *vma, 395347ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 395447ad8475SAndrea Arcangeli { 395547ad8475SAndrea Arcangeli int i; 395647ad8475SAndrea Arcangeli 395747ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 395847ad8475SAndrea Arcangeli copy_user_gigantic_page(dst, src, addr, vma, 395947ad8475SAndrea Arcangeli pages_per_huge_page); 396047ad8475SAndrea Arcangeli return; 396147ad8475SAndrea Arcangeli } 396247ad8475SAndrea Arcangeli 396347ad8475SAndrea Arcangeli might_sleep(); 396447ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 396547ad8475SAndrea Arcangeli cond_resched(); 396647ad8475SAndrea Arcangeli copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); 396747ad8475SAndrea Arcangeli } 396847ad8475SAndrea Arcangeli } 396947ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 3970