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> 50b95f1b31SPaul Gortmaker #include <linux/export.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> 604daae3b4SMel Gorman #include <linux/migrate.h> 611da177e4SLinus Torvalds 626952b61dSAlexey Dobriyan #include <asm/io.h> 631da177e4SLinus Torvalds #include <asm/pgalloc.h> 641da177e4SLinus Torvalds #include <asm/uaccess.h> 651da177e4SLinus Torvalds #include <asm/tlb.h> 661da177e4SLinus Torvalds #include <asm/tlbflush.h> 671da177e4SLinus Torvalds #include <asm/pgtable.h> 681da177e4SLinus Torvalds 6942b77728SJan Beulich #include "internal.h" 7042b77728SJan Beulich 71d41dee36SAndy Whitcroft #ifndef CONFIG_NEED_MULTIPLE_NODES 721da177e4SLinus Torvalds /* use the per-pgdat data instead for discontigmem - mbligh */ 731da177e4SLinus Torvalds unsigned long max_mapnr; 741da177e4SLinus Torvalds struct page *mem_map; 751da177e4SLinus Torvalds 761da177e4SLinus Torvalds EXPORT_SYMBOL(max_mapnr); 771da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 781da177e4SLinus Torvalds #endif 791da177e4SLinus Torvalds 801da177e4SLinus Torvalds unsigned long num_physpages; 811da177e4SLinus Torvalds /* 821da177e4SLinus Torvalds * A number of key systems in x86 including ioremap() rely on the assumption 831da177e4SLinus Torvalds * that high_memory defines the upper bound on direct map memory, then end 841da177e4SLinus Torvalds * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and 851da177e4SLinus Torvalds * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL 861da177e4SLinus Torvalds * and ZONE_HIGHMEM. 871da177e4SLinus Torvalds */ 881da177e4SLinus Torvalds void * high_memory; 891da177e4SLinus Torvalds 901da177e4SLinus Torvalds EXPORT_SYMBOL(num_physpages); 911da177e4SLinus Torvalds EXPORT_SYMBOL(high_memory); 921da177e4SLinus Torvalds 9332a93233SIngo Molnar /* 9432a93233SIngo Molnar * Randomize the address space (stacks, mmaps, brk, etc.). 9532a93233SIngo Molnar * 9632a93233SIngo Molnar * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, 9732a93233SIngo Molnar * as ancient (libc5 based) binaries can segfault. ) 9832a93233SIngo Molnar */ 9932a93233SIngo Molnar int randomize_va_space __read_mostly = 10032a93233SIngo Molnar #ifdef CONFIG_COMPAT_BRK 10132a93233SIngo Molnar 1; 10232a93233SIngo Molnar #else 10332a93233SIngo Molnar 2; 10432a93233SIngo Molnar #endif 105a62eaf15SAndi Kleen 106a62eaf15SAndi Kleen static int __init disable_randmaps(char *s) 107a62eaf15SAndi Kleen { 108a62eaf15SAndi Kleen randomize_va_space = 0; 1099b41046cSOGAWA Hirofumi return 1; 110a62eaf15SAndi Kleen } 111a62eaf15SAndi Kleen __setup("norandmaps", disable_randmaps); 112a62eaf15SAndi Kleen 11362eede62SHugh Dickins unsigned long zero_pfn __read_mostly; 11403f6462aSHugh Dickins unsigned long highest_memmap_pfn __read_mostly; 115a13ea5b7SHugh Dickins 116a13ea5b7SHugh Dickins /* 117a13ea5b7SHugh Dickins * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() 118a13ea5b7SHugh Dickins */ 119a13ea5b7SHugh Dickins static int __init init_zero_pfn(void) 120a13ea5b7SHugh Dickins { 121a13ea5b7SHugh Dickins zero_pfn = page_to_pfn(ZERO_PAGE(0)); 122a13ea5b7SHugh Dickins return 0; 123a13ea5b7SHugh Dickins } 124a13ea5b7SHugh Dickins core_initcall(init_zero_pfn); 125a62eaf15SAndi Kleen 126d559db08SKAMEZAWA Hiroyuki 12734e55232SKAMEZAWA Hiroyuki #if defined(SPLIT_RSS_COUNTING) 12834e55232SKAMEZAWA Hiroyuki 129ea48cf78SDavid Rientjes void sync_mm_rss(struct mm_struct *mm) 13034e55232SKAMEZAWA Hiroyuki { 13134e55232SKAMEZAWA Hiroyuki int i; 13234e55232SKAMEZAWA Hiroyuki 13334e55232SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) { 13405af2e10SDavid Rientjes if (current->rss_stat.count[i]) { 13505af2e10SDavid Rientjes add_mm_counter(mm, i, current->rss_stat.count[i]); 13605af2e10SDavid Rientjes current->rss_stat.count[i] = 0; 13734e55232SKAMEZAWA Hiroyuki } 13834e55232SKAMEZAWA Hiroyuki } 13905af2e10SDavid Rientjes current->rss_stat.events = 0; 14034e55232SKAMEZAWA Hiroyuki } 14134e55232SKAMEZAWA Hiroyuki 14234e55232SKAMEZAWA Hiroyuki static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) 14334e55232SKAMEZAWA Hiroyuki { 14434e55232SKAMEZAWA Hiroyuki struct task_struct *task = current; 14534e55232SKAMEZAWA Hiroyuki 14634e55232SKAMEZAWA Hiroyuki if (likely(task->mm == mm)) 14734e55232SKAMEZAWA Hiroyuki task->rss_stat.count[member] += val; 14834e55232SKAMEZAWA Hiroyuki else 14934e55232SKAMEZAWA Hiroyuki add_mm_counter(mm, member, val); 15034e55232SKAMEZAWA Hiroyuki } 15134e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) 15234e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) 15334e55232SKAMEZAWA Hiroyuki 15434e55232SKAMEZAWA Hiroyuki /* sync counter once per 64 page faults */ 15534e55232SKAMEZAWA Hiroyuki #define TASK_RSS_EVENTS_THRESH (64) 15634e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 15734e55232SKAMEZAWA Hiroyuki { 15834e55232SKAMEZAWA Hiroyuki if (unlikely(task != current)) 15934e55232SKAMEZAWA Hiroyuki return; 16034e55232SKAMEZAWA Hiroyuki if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) 161ea48cf78SDavid Rientjes sync_mm_rss(task->mm); 16234e55232SKAMEZAWA Hiroyuki } 1639547d01bSPeter Zijlstra #else /* SPLIT_RSS_COUNTING */ 16434e55232SKAMEZAWA Hiroyuki 16534e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) 16634e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) 16734e55232SKAMEZAWA Hiroyuki 16834e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 16934e55232SKAMEZAWA Hiroyuki { 17034e55232SKAMEZAWA Hiroyuki } 17134e55232SKAMEZAWA Hiroyuki 1729547d01bSPeter Zijlstra #endif /* SPLIT_RSS_COUNTING */ 1739547d01bSPeter Zijlstra 1749547d01bSPeter Zijlstra #ifdef HAVE_GENERIC_MMU_GATHER 1759547d01bSPeter Zijlstra 1769547d01bSPeter Zijlstra static int tlb_next_batch(struct mmu_gather *tlb) 1779547d01bSPeter Zijlstra { 1789547d01bSPeter Zijlstra struct mmu_gather_batch *batch; 1799547d01bSPeter Zijlstra 1809547d01bSPeter Zijlstra batch = tlb->active; 1819547d01bSPeter Zijlstra if (batch->next) { 1829547d01bSPeter Zijlstra tlb->active = batch->next; 1839547d01bSPeter Zijlstra return 1; 1849547d01bSPeter Zijlstra } 1859547d01bSPeter Zijlstra 1869547d01bSPeter Zijlstra batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); 1879547d01bSPeter Zijlstra if (!batch) 1889547d01bSPeter Zijlstra return 0; 1899547d01bSPeter Zijlstra 1909547d01bSPeter Zijlstra batch->next = NULL; 1919547d01bSPeter Zijlstra batch->nr = 0; 1929547d01bSPeter Zijlstra batch->max = MAX_GATHER_BATCH; 1939547d01bSPeter Zijlstra 1949547d01bSPeter Zijlstra tlb->active->next = batch; 1959547d01bSPeter Zijlstra tlb->active = batch; 1969547d01bSPeter Zijlstra 1979547d01bSPeter Zijlstra return 1; 1989547d01bSPeter Zijlstra } 1999547d01bSPeter Zijlstra 2009547d01bSPeter Zijlstra /* tlb_gather_mmu 2019547d01bSPeter Zijlstra * Called to initialize an (on-stack) mmu_gather structure for page-table 2029547d01bSPeter Zijlstra * tear-down from @mm. The @fullmm argument is used when @mm is without 2039547d01bSPeter Zijlstra * users and we're going to destroy the full address space (exit/execve). 2049547d01bSPeter Zijlstra */ 2059547d01bSPeter Zijlstra void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, bool fullmm) 2069547d01bSPeter Zijlstra { 2079547d01bSPeter Zijlstra tlb->mm = mm; 2089547d01bSPeter Zijlstra 2099547d01bSPeter Zijlstra tlb->fullmm = fullmm; 210597e1c35SAlex Shi tlb->start = -1UL; 211597e1c35SAlex Shi tlb->end = 0; 2129547d01bSPeter Zijlstra tlb->need_flush = 0; 2139547d01bSPeter Zijlstra tlb->fast_mode = (num_possible_cpus() == 1); 2149547d01bSPeter Zijlstra tlb->local.next = NULL; 2159547d01bSPeter Zijlstra tlb->local.nr = 0; 2169547d01bSPeter Zijlstra tlb->local.max = ARRAY_SIZE(tlb->__pages); 2179547d01bSPeter Zijlstra tlb->active = &tlb->local; 2189547d01bSPeter Zijlstra 2199547d01bSPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 2209547d01bSPeter Zijlstra tlb->batch = NULL; 22134e55232SKAMEZAWA Hiroyuki #endif 2229547d01bSPeter Zijlstra } 2239547d01bSPeter Zijlstra 2249547d01bSPeter Zijlstra void tlb_flush_mmu(struct mmu_gather *tlb) 2259547d01bSPeter Zijlstra { 2269547d01bSPeter Zijlstra struct mmu_gather_batch *batch; 2279547d01bSPeter Zijlstra 2289547d01bSPeter Zijlstra if (!tlb->need_flush) 2299547d01bSPeter Zijlstra return; 2309547d01bSPeter Zijlstra tlb->need_flush = 0; 2319547d01bSPeter Zijlstra tlb_flush(tlb); 2329547d01bSPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 2339547d01bSPeter Zijlstra tlb_table_flush(tlb); 2349547d01bSPeter Zijlstra #endif 2359547d01bSPeter Zijlstra 2369547d01bSPeter Zijlstra if (tlb_fast_mode(tlb)) 2379547d01bSPeter Zijlstra return; 2389547d01bSPeter Zijlstra 2399547d01bSPeter Zijlstra for (batch = &tlb->local; batch; batch = batch->next) { 2409547d01bSPeter Zijlstra free_pages_and_swap_cache(batch->pages, batch->nr); 2419547d01bSPeter Zijlstra batch->nr = 0; 2429547d01bSPeter Zijlstra } 2439547d01bSPeter Zijlstra tlb->active = &tlb->local; 2449547d01bSPeter Zijlstra } 2459547d01bSPeter Zijlstra 2469547d01bSPeter Zijlstra /* tlb_finish_mmu 2479547d01bSPeter Zijlstra * Called at the end of the shootdown operation to free up any resources 2489547d01bSPeter Zijlstra * that were required. 2499547d01bSPeter Zijlstra */ 2509547d01bSPeter Zijlstra void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end) 2519547d01bSPeter Zijlstra { 2529547d01bSPeter Zijlstra struct mmu_gather_batch *batch, *next; 2539547d01bSPeter Zijlstra 254597e1c35SAlex Shi tlb->start = start; 255597e1c35SAlex Shi tlb->end = end; 2569547d01bSPeter Zijlstra tlb_flush_mmu(tlb); 2579547d01bSPeter Zijlstra 2589547d01bSPeter Zijlstra /* keep the page table cache within bounds */ 2599547d01bSPeter Zijlstra check_pgt_cache(); 2609547d01bSPeter Zijlstra 2619547d01bSPeter Zijlstra for (batch = tlb->local.next; batch; batch = next) { 2629547d01bSPeter Zijlstra next = batch->next; 2639547d01bSPeter Zijlstra free_pages((unsigned long)batch, 0); 2649547d01bSPeter Zijlstra } 2659547d01bSPeter Zijlstra tlb->local.next = NULL; 2669547d01bSPeter Zijlstra } 2679547d01bSPeter Zijlstra 2689547d01bSPeter Zijlstra /* __tlb_remove_page 2699547d01bSPeter Zijlstra * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while 2709547d01bSPeter Zijlstra * handling the additional races in SMP caused by other CPUs caching valid 2719547d01bSPeter Zijlstra * mappings in their TLBs. Returns the number of free page slots left. 2729547d01bSPeter Zijlstra * When out of page slots we must call tlb_flush_mmu(). 2739547d01bSPeter Zijlstra */ 2749547d01bSPeter Zijlstra int __tlb_remove_page(struct mmu_gather *tlb, struct page *page) 2759547d01bSPeter Zijlstra { 2769547d01bSPeter Zijlstra struct mmu_gather_batch *batch; 2779547d01bSPeter Zijlstra 278f21760b1SShaohua Li VM_BUG_ON(!tlb->need_flush); 2799547d01bSPeter Zijlstra 2809547d01bSPeter Zijlstra if (tlb_fast_mode(tlb)) { 2819547d01bSPeter Zijlstra free_page_and_swap_cache(page); 2829547d01bSPeter Zijlstra return 1; /* avoid calling tlb_flush_mmu() */ 2839547d01bSPeter Zijlstra } 2849547d01bSPeter Zijlstra 2859547d01bSPeter Zijlstra batch = tlb->active; 2869547d01bSPeter Zijlstra batch->pages[batch->nr++] = page; 2879547d01bSPeter Zijlstra if (batch->nr == batch->max) { 2889547d01bSPeter Zijlstra if (!tlb_next_batch(tlb)) 2899547d01bSPeter Zijlstra return 0; 2900b43c3aaSShaohua Li batch = tlb->active; 2919547d01bSPeter Zijlstra } 2929547d01bSPeter Zijlstra VM_BUG_ON(batch->nr > batch->max); 2939547d01bSPeter Zijlstra 2949547d01bSPeter Zijlstra return batch->max - batch->nr; 2959547d01bSPeter Zijlstra } 2969547d01bSPeter Zijlstra 2979547d01bSPeter Zijlstra #endif /* HAVE_GENERIC_MMU_GATHER */ 29834e55232SKAMEZAWA Hiroyuki 29926723911SPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 30026723911SPeter Zijlstra 30126723911SPeter Zijlstra /* 30226723911SPeter Zijlstra * See the comment near struct mmu_table_batch. 30326723911SPeter Zijlstra */ 30426723911SPeter Zijlstra 30526723911SPeter Zijlstra static void tlb_remove_table_smp_sync(void *arg) 30626723911SPeter Zijlstra { 30726723911SPeter Zijlstra /* Simply deliver the interrupt */ 30826723911SPeter Zijlstra } 30926723911SPeter Zijlstra 31026723911SPeter Zijlstra static void tlb_remove_table_one(void *table) 31126723911SPeter Zijlstra { 31226723911SPeter Zijlstra /* 31326723911SPeter Zijlstra * This isn't an RCU grace period and hence the page-tables cannot be 31426723911SPeter Zijlstra * assumed to be actually RCU-freed. 31526723911SPeter Zijlstra * 31626723911SPeter Zijlstra * It is however sufficient for software page-table walkers that rely on 31726723911SPeter Zijlstra * IRQ disabling. See the comment near struct mmu_table_batch. 31826723911SPeter Zijlstra */ 31926723911SPeter Zijlstra smp_call_function(tlb_remove_table_smp_sync, NULL, 1); 32026723911SPeter Zijlstra __tlb_remove_table(table); 32126723911SPeter Zijlstra } 32226723911SPeter Zijlstra 32326723911SPeter Zijlstra static void tlb_remove_table_rcu(struct rcu_head *head) 32426723911SPeter Zijlstra { 32526723911SPeter Zijlstra struct mmu_table_batch *batch; 32626723911SPeter Zijlstra int i; 32726723911SPeter Zijlstra 32826723911SPeter Zijlstra batch = container_of(head, struct mmu_table_batch, rcu); 32926723911SPeter Zijlstra 33026723911SPeter Zijlstra for (i = 0; i < batch->nr; i++) 33126723911SPeter Zijlstra __tlb_remove_table(batch->tables[i]); 33226723911SPeter Zijlstra 33326723911SPeter Zijlstra free_page((unsigned long)batch); 33426723911SPeter Zijlstra } 33526723911SPeter Zijlstra 33626723911SPeter Zijlstra void tlb_table_flush(struct mmu_gather *tlb) 33726723911SPeter Zijlstra { 33826723911SPeter Zijlstra struct mmu_table_batch **batch = &tlb->batch; 33926723911SPeter Zijlstra 34026723911SPeter Zijlstra if (*batch) { 34126723911SPeter Zijlstra call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); 34226723911SPeter Zijlstra *batch = NULL; 34326723911SPeter Zijlstra } 34426723911SPeter Zijlstra } 34526723911SPeter Zijlstra 34626723911SPeter Zijlstra void tlb_remove_table(struct mmu_gather *tlb, void *table) 34726723911SPeter Zijlstra { 34826723911SPeter Zijlstra struct mmu_table_batch **batch = &tlb->batch; 34926723911SPeter Zijlstra 35026723911SPeter Zijlstra tlb->need_flush = 1; 35126723911SPeter Zijlstra 35226723911SPeter Zijlstra /* 35326723911SPeter Zijlstra * When there's less then two users of this mm there cannot be a 35426723911SPeter Zijlstra * concurrent page-table walk. 35526723911SPeter Zijlstra */ 35626723911SPeter Zijlstra if (atomic_read(&tlb->mm->mm_users) < 2) { 35726723911SPeter Zijlstra __tlb_remove_table(table); 35826723911SPeter Zijlstra return; 35926723911SPeter Zijlstra } 36026723911SPeter Zijlstra 36126723911SPeter Zijlstra if (*batch == NULL) { 36226723911SPeter Zijlstra *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); 36326723911SPeter Zijlstra if (*batch == NULL) { 36426723911SPeter Zijlstra tlb_remove_table_one(table); 36526723911SPeter Zijlstra return; 36626723911SPeter Zijlstra } 36726723911SPeter Zijlstra (*batch)->nr = 0; 36826723911SPeter Zijlstra } 36926723911SPeter Zijlstra (*batch)->tables[(*batch)->nr++] = table; 37026723911SPeter Zijlstra if ((*batch)->nr == MAX_TABLE_BATCH) 37126723911SPeter Zijlstra tlb_table_flush(tlb); 37226723911SPeter Zijlstra } 37326723911SPeter Zijlstra 3749547d01bSPeter Zijlstra #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ 37526723911SPeter Zijlstra 3761da177e4SLinus Torvalds /* 3771da177e4SLinus Torvalds * If a p?d_bad entry is found while walking page tables, report 3781da177e4SLinus Torvalds * the error, before resetting entry to p?d_none. Usually (but 3791da177e4SLinus Torvalds * very seldom) called out from the p?d_none_or_clear_bad macros. 3801da177e4SLinus Torvalds */ 3811da177e4SLinus Torvalds 3821da177e4SLinus Torvalds void pgd_clear_bad(pgd_t *pgd) 3831da177e4SLinus Torvalds { 3841da177e4SLinus Torvalds pgd_ERROR(*pgd); 3851da177e4SLinus Torvalds pgd_clear(pgd); 3861da177e4SLinus Torvalds } 3871da177e4SLinus Torvalds 3881da177e4SLinus Torvalds void pud_clear_bad(pud_t *pud) 3891da177e4SLinus Torvalds { 3901da177e4SLinus Torvalds pud_ERROR(*pud); 3911da177e4SLinus Torvalds pud_clear(pud); 3921da177e4SLinus Torvalds } 3931da177e4SLinus Torvalds 3941da177e4SLinus Torvalds void pmd_clear_bad(pmd_t *pmd) 3951da177e4SLinus Torvalds { 3961da177e4SLinus Torvalds pmd_ERROR(*pmd); 3971da177e4SLinus Torvalds pmd_clear(pmd); 3981da177e4SLinus Torvalds } 3991da177e4SLinus Torvalds 4001da177e4SLinus Torvalds /* 4011da177e4SLinus Torvalds * Note: this doesn't free the actual pages themselves. That 4021da177e4SLinus Torvalds * has been handled earlier when unmapping all the memory regions. 4031da177e4SLinus Torvalds */ 4049e1b32caSBenjamin Herrenschmidt static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, 4059e1b32caSBenjamin Herrenschmidt unsigned long addr) 4061da177e4SLinus Torvalds { 4072f569afdSMartin Schwidefsky pgtable_t token = pmd_pgtable(*pmd); 4081da177e4SLinus Torvalds pmd_clear(pmd); 4099e1b32caSBenjamin Herrenschmidt pte_free_tlb(tlb, token, addr); 4101da177e4SLinus Torvalds tlb->mm->nr_ptes--; 4111da177e4SLinus Torvalds } 4121da177e4SLinus Torvalds 413e0da382cSHugh Dickins static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, 414e0da382cSHugh Dickins unsigned long addr, unsigned long end, 415e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4161da177e4SLinus Torvalds { 4171da177e4SLinus Torvalds pmd_t *pmd; 4181da177e4SLinus Torvalds unsigned long next; 419e0da382cSHugh Dickins unsigned long start; 4201da177e4SLinus Torvalds 421e0da382cSHugh Dickins start = addr; 4221da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 4231da177e4SLinus Torvalds do { 4241da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 4251da177e4SLinus Torvalds if (pmd_none_or_clear_bad(pmd)) 4261da177e4SLinus Torvalds continue; 4279e1b32caSBenjamin Herrenschmidt free_pte_range(tlb, pmd, addr); 4281da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 4291da177e4SLinus Torvalds 430e0da382cSHugh Dickins start &= PUD_MASK; 431e0da382cSHugh Dickins if (start < floor) 432e0da382cSHugh Dickins return; 433e0da382cSHugh Dickins if (ceiling) { 434e0da382cSHugh Dickins ceiling &= PUD_MASK; 435e0da382cSHugh Dickins if (!ceiling) 436e0da382cSHugh Dickins return; 4371da177e4SLinus Torvalds } 438e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 439e0da382cSHugh Dickins return; 440e0da382cSHugh Dickins 441e0da382cSHugh Dickins pmd = pmd_offset(pud, start); 442e0da382cSHugh Dickins pud_clear(pud); 4439e1b32caSBenjamin Herrenschmidt pmd_free_tlb(tlb, pmd, start); 4441da177e4SLinus Torvalds } 4451da177e4SLinus Torvalds 446e0da382cSHugh Dickins static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, 447e0da382cSHugh Dickins unsigned long addr, unsigned long end, 448e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4491da177e4SLinus Torvalds { 4501da177e4SLinus Torvalds pud_t *pud; 4511da177e4SLinus Torvalds unsigned long next; 452e0da382cSHugh Dickins unsigned long start; 4531da177e4SLinus Torvalds 454e0da382cSHugh Dickins start = addr; 4551da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 4561da177e4SLinus Torvalds do { 4571da177e4SLinus Torvalds next = pud_addr_end(addr, end); 4581da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 4591da177e4SLinus Torvalds continue; 460e0da382cSHugh Dickins free_pmd_range(tlb, pud, addr, next, floor, ceiling); 4611da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 4621da177e4SLinus Torvalds 463e0da382cSHugh Dickins start &= PGDIR_MASK; 464e0da382cSHugh Dickins if (start < floor) 465e0da382cSHugh Dickins return; 466e0da382cSHugh Dickins if (ceiling) { 467e0da382cSHugh Dickins ceiling &= PGDIR_MASK; 468e0da382cSHugh Dickins if (!ceiling) 469e0da382cSHugh Dickins return; 4701da177e4SLinus Torvalds } 471e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 472e0da382cSHugh Dickins return; 473e0da382cSHugh Dickins 474e0da382cSHugh Dickins pud = pud_offset(pgd, start); 475e0da382cSHugh Dickins pgd_clear(pgd); 4769e1b32caSBenjamin Herrenschmidt pud_free_tlb(tlb, pud, start); 4771da177e4SLinus Torvalds } 4781da177e4SLinus Torvalds 4791da177e4SLinus Torvalds /* 480e0da382cSHugh Dickins * This function frees user-level page tables of a process. 481e0da382cSHugh Dickins * 4821da177e4SLinus Torvalds * Must be called with pagetable lock held. 4831da177e4SLinus Torvalds */ 48442b77728SJan Beulich void free_pgd_range(struct mmu_gather *tlb, 485e0da382cSHugh Dickins unsigned long addr, unsigned long end, 486e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4871da177e4SLinus Torvalds { 4881da177e4SLinus Torvalds pgd_t *pgd; 4891da177e4SLinus Torvalds unsigned long next; 4901da177e4SLinus Torvalds 491e0da382cSHugh Dickins /* 492e0da382cSHugh Dickins * The next few lines have given us lots of grief... 493e0da382cSHugh Dickins * 494e0da382cSHugh Dickins * Why are we testing PMD* at this top level? Because often 495e0da382cSHugh Dickins * there will be no work to do at all, and we'd prefer not to 496e0da382cSHugh Dickins * go all the way down to the bottom just to discover that. 497e0da382cSHugh Dickins * 498e0da382cSHugh Dickins * Why all these "- 1"s? Because 0 represents both the bottom 499e0da382cSHugh Dickins * of the address space and the top of it (using -1 for the 500e0da382cSHugh Dickins * top wouldn't help much: the masks would do the wrong thing). 501e0da382cSHugh Dickins * The rule is that addr 0 and floor 0 refer to the bottom of 502e0da382cSHugh Dickins * the address space, but end 0 and ceiling 0 refer to the top 503e0da382cSHugh Dickins * Comparisons need to use "end - 1" and "ceiling - 1" (though 504e0da382cSHugh Dickins * that end 0 case should be mythical). 505e0da382cSHugh Dickins * 506e0da382cSHugh Dickins * Wherever addr is brought up or ceiling brought down, we must 507e0da382cSHugh Dickins * be careful to reject "the opposite 0" before it confuses the 508e0da382cSHugh Dickins * subsequent tests. But what about where end is brought down 509e0da382cSHugh Dickins * by PMD_SIZE below? no, end can't go down to 0 there. 510e0da382cSHugh Dickins * 511e0da382cSHugh Dickins * Whereas we round start (addr) and ceiling down, by different 512e0da382cSHugh Dickins * masks at different levels, in order to test whether a table 513e0da382cSHugh Dickins * now has no other vmas using it, so can be freed, we don't 514e0da382cSHugh Dickins * bother to round floor or end up - the tests don't need that. 515e0da382cSHugh Dickins */ 516e0da382cSHugh Dickins 517e0da382cSHugh Dickins addr &= PMD_MASK; 518e0da382cSHugh Dickins if (addr < floor) { 519e0da382cSHugh Dickins addr += PMD_SIZE; 520e0da382cSHugh Dickins if (!addr) 521e0da382cSHugh Dickins return; 522e0da382cSHugh Dickins } 523e0da382cSHugh Dickins if (ceiling) { 524e0da382cSHugh Dickins ceiling &= PMD_MASK; 525e0da382cSHugh Dickins if (!ceiling) 526e0da382cSHugh Dickins return; 527e0da382cSHugh Dickins } 528e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 529e0da382cSHugh Dickins end -= PMD_SIZE; 530e0da382cSHugh Dickins if (addr > end - 1) 531e0da382cSHugh Dickins return; 532e0da382cSHugh Dickins 53342b77728SJan Beulich pgd = pgd_offset(tlb->mm, addr); 5341da177e4SLinus Torvalds do { 5351da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 5361da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 5371da177e4SLinus Torvalds continue; 53842b77728SJan Beulich free_pud_range(tlb, pgd, addr, next, floor, ceiling); 5391da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 540e0da382cSHugh Dickins } 541e0da382cSHugh Dickins 54242b77728SJan Beulich void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, 543e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 544e0da382cSHugh Dickins { 545e0da382cSHugh Dickins while (vma) { 546e0da382cSHugh Dickins struct vm_area_struct *next = vma->vm_next; 547e0da382cSHugh Dickins unsigned long addr = vma->vm_start; 548e0da382cSHugh Dickins 5498f4f8c16SHugh Dickins /* 55025d9e2d1Snpiggin@suse.de * Hide vma from rmap and truncate_pagecache before freeing 55125d9e2d1Snpiggin@suse.de * pgtables 5528f4f8c16SHugh Dickins */ 5535beb4930SRik van Riel unlink_anon_vmas(vma); 5548f4f8c16SHugh Dickins unlink_file_vma(vma); 5558f4f8c16SHugh Dickins 5569da61aefSDavid Gibson if (is_vm_hugetlb_page(vma)) { 5573bf5ee95SHugh Dickins hugetlb_free_pgd_range(tlb, addr, vma->vm_end, 5583bf5ee95SHugh Dickins floor, next? next->vm_start: ceiling); 5593bf5ee95SHugh Dickins } else { 5603bf5ee95SHugh Dickins /* 5613bf5ee95SHugh Dickins * Optimization: gather nearby vmas into one call down 5623bf5ee95SHugh Dickins */ 5633bf5ee95SHugh Dickins while (next && next->vm_start <= vma->vm_end + PMD_SIZE 5644866920bSDavid Gibson && !is_vm_hugetlb_page(next)) { 565e0da382cSHugh Dickins vma = next; 566e0da382cSHugh Dickins next = vma->vm_next; 5675beb4930SRik van Riel unlink_anon_vmas(vma); 5688f4f8c16SHugh Dickins unlink_file_vma(vma); 569e0da382cSHugh Dickins } 5703bf5ee95SHugh Dickins free_pgd_range(tlb, addr, vma->vm_end, 571e0da382cSHugh Dickins floor, next? next->vm_start: ceiling); 5723bf5ee95SHugh Dickins } 573e0da382cSHugh Dickins vma = next; 574e0da382cSHugh Dickins } 5751da177e4SLinus Torvalds } 5761da177e4SLinus Torvalds 5778ac1f832SAndrea Arcangeli int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma, 5788ac1f832SAndrea Arcangeli pmd_t *pmd, unsigned long address) 5791da177e4SLinus Torvalds { 5802f569afdSMartin Schwidefsky pgtable_t new = pte_alloc_one(mm, address); 5818ac1f832SAndrea Arcangeli int wait_split_huge_page; 5821da177e4SLinus Torvalds if (!new) 5831bb3630eSHugh Dickins return -ENOMEM; 5841bb3630eSHugh Dickins 585362a61adSNick Piggin /* 586362a61adSNick Piggin * Ensure all pte setup (eg. pte page lock and page clearing) are 587362a61adSNick Piggin * visible before the pte is made visible to other CPUs by being 588362a61adSNick Piggin * put into page tables. 589362a61adSNick Piggin * 590362a61adSNick Piggin * The other side of the story is the pointer chasing in the page 591362a61adSNick Piggin * table walking code (when walking the page table without locking; 592362a61adSNick Piggin * ie. most of the time). Fortunately, these data accesses consist 593362a61adSNick Piggin * of a chain of data-dependent loads, meaning most CPUs (alpha 594362a61adSNick Piggin * being the notable exception) will already guarantee loads are 595362a61adSNick Piggin * seen in-order. See the alpha page table accessors for the 596362a61adSNick Piggin * smp_read_barrier_depends() barriers in page table walking code. 597362a61adSNick Piggin */ 598362a61adSNick Piggin smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ 599362a61adSNick Piggin 600c74df32cSHugh Dickins spin_lock(&mm->page_table_lock); 6018ac1f832SAndrea Arcangeli wait_split_huge_page = 0; 6028ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 6031da177e4SLinus Torvalds mm->nr_ptes++; 6041da177e4SLinus Torvalds pmd_populate(mm, pmd, new); 6052f569afdSMartin Schwidefsky new = NULL; 6068ac1f832SAndrea Arcangeli } else if (unlikely(pmd_trans_splitting(*pmd))) 6078ac1f832SAndrea Arcangeli wait_split_huge_page = 1; 608c74df32cSHugh Dickins spin_unlock(&mm->page_table_lock); 6092f569afdSMartin Schwidefsky if (new) 6102f569afdSMartin Schwidefsky pte_free(mm, new); 6118ac1f832SAndrea Arcangeli if (wait_split_huge_page) 6128ac1f832SAndrea Arcangeli wait_split_huge_page(vma->anon_vma, pmd); 6131bb3630eSHugh Dickins return 0; 6141da177e4SLinus Torvalds } 6151da177e4SLinus Torvalds 6161bb3630eSHugh Dickins int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) 6171da177e4SLinus Torvalds { 6181bb3630eSHugh Dickins pte_t *new = pte_alloc_one_kernel(&init_mm, address); 6191da177e4SLinus Torvalds if (!new) 6201bb3630eSHugh Dickins return -ENOMEM; 6211da177e4SLinus Torvalds 622362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 623362a61adSNick Piggin 624872fec16SHugh Dickins spin_lock(&init_mm.page_table_lock); 6258ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 626872fec16SHugh Dickins pmd_populate_kernel(&init_mm, pmd, new); 6272f569afdSMartin Schwidefsky new = NULL; 6288ac1f832SAndrea Arcangeli } else 6298ac1f832SAndrea Arcangeli VM_BUG_ON(pmd_trans_splitting(*pmd)); 630872fec16SHugh Dickins spin_unlock(&init_mm.page_table_lock); 6312f569afdSMartin Schwidefsky if (new) 6322f569afdSMartin Schwidefsky pte_free_kernel(&init_mm, new); 6331bb3630eSHugh Dickins return 0; 6341da177e4SLinus Torvalds } 6351da177e4SLinus Torvalds 636d559db08SKAMEZAWA Hiroyuki static inline void init_rss_vec(int *rss) 637ae859762SHugh Dickins { 638d559db08SKAMEZAWA Hiroyuki memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); 639d559db08SKAMEZAWA Hiroyuki } 640d559db08SKAMEZAWA Hiroyuki 641d559db08SKAMEZAWA Hiroyuki static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) 642d559db08SKAMEZAWA Hiroyuki { 643d559db08SKAMEZAWA Hiroyuki int i; 644d559db08SKAMEZAWA Hiroyuki 64534e55232SKAMEZAWA Hiroyuki if (current->mm == mm) 64605af2e10SDavid Rientjes sync_mm_rss(mm); 647d559db08SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) 648d559db08SKAMEZAWA Hiroyuki if (rss[i]) 649d559db08SKAMEZAWA Hiroyuki add_mm_counter(mm, i, rss[i]); 650ae859762SHugh Dickins } 651ae859762SHugh Dickins 6521da177e4SLinus Torvalds /* 6536aab341eSLinus Torvalds * This function is called to print an error when a bad pte 6546aab341eSLinus Torvalds * is found. For example, we might have a PFN-mapped pte in 6556aab341eSLinus Torvalds * a region that doesn't allow it. 656b5810039SNick Piggin * 657b5810039SNick Piggin * The calling function must still handle the error. 658b5810039SNick Piggin */ 6593dc14741SHugh Dickins static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, 6603dc14741SHugh Dickins pte_t pte, struct page *page) 661b5810039SNick Piggin { 6623dc14741SHugh Dickins pgd_t *pgd = pgd_offset(vma->vm_mm, addr); 6633dc14741SHugh Dickins pud_t *pud = pud_offset(pgd, addr); 6643dc14741SHugh Dickins pmd_t *pmd = pmd_offset(pud, addr); 6653dc14741SHugh Dickins struct address_space *mapping; 6663dc14741SHugh Dickins pgoff_t index; 667d936cf9bSHugh Dickins static unsigned long resume; 668d936cf9bSHugh Dickins static unsigned long nr_shown; 669d936cf9bSHugh Dickins static unsigned long nr_unshown; 670d936cf9bSHugh Dickins 671d936cf9bSHugh Dickins /* 672d936cf9bSHugh Dickins * Allow a burst of 60 reports, then keep quiet for that minute; 673d936cf9bSHugh Dickins * or allow a steady drip of one report per second. 674d936cf9bSHugh Dickins */ 675d936cf9bSHugh Dickins if (nr_shown == 60) { 676d936cf9bSHugh Dickins if (time_before(jiffies, resume)) { 677d936cf9bSHugh Dickins nr_unshown++; 678d936cf9bSHugh Dickins return; 679d936cf9bSHugh Dickins } 680d936cf9bSHugh Dickins if (nr_unshown) { 6811e9e6365SHugh Dickins printk(KERN_ALERT 6821e9e6365SHugh Dickins "BUG: Bad page map: %lu messages suppressed\n", 683d936cf9bSHugh Dickins nr_unshown); 684d936cf9bSHugh Dickins nr_unshown = 0; 685d936cf9bSHugh Dickins } 686d936cf9bSHugh Dickins nr_shown = 0; 687d936cf9bSHugh Dickins } 688d936cf9bSHugh Dickins if (nr_shown++ == 0) 689d936cf9bSHugh Dickins resume = jiffies + 60 * HZ; 6903dc14741SHugh Dickins 6913dc14741SHugh Dickins mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; 6923dc14741SHugh Dickins index = linear_page_index(vma, addr); 6933dc14741SHugh Dickins 6941e9e6365SHugh Dickins printk(KERN_ALERT 6951e9e6365SHugh Dickins "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", 6963dc14741SHugh Dickins current->comm, 6973dc14741SHugh Dickins (long long)pte_val(pte), (long long)pmd_val(*pmd)); 698718a3821SWu Fengguang if (page) 699718a3821SWu Fengguang dump_page(page); 7001e9e6365SHugh Dickins printk(KERN_ALERT 7013dc14741SHugh Dickins "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", 7023dc14741SHugh Dickins (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); 7033dc14741SHugh Dickins /* 7043dc14741SHugh Dickins * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y 7053dc14741SHugh Dickins */ 7063dc14741SHugh Dickins if (vma->vm_ops) 7071e9e6365SHugh Dickins print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n", 7083dc14741SHugh Dickins (unsigned long)vma->vm_ops->fault); 7093dc14741SHugh Dickins if (vma->vm_file && vma->vm_file->f_op) 7101e9e6365SHugh Dickins print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n", 7113dc14741SHugh Dickins (unsigned long)vma->vm_file->f_op->mmap); 712b5810039SNick Piggin dump_stack(); 7133dc14741SHugh Dickins add_taint(TAINT_BAD_PAGE); 714b5810039SNick Piggin } 715b5810039SNick Piggin 7162ec74c3eSSagi Grimberg static inline bool is_cow_mapping(vm_flags_t flags) 71767121172SLinus Torvalds { 71867121172SLinus Torvalds return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; 71967121172SLinus Torvalds } 72067121172SLinus Torvalds 72162eede62SHugh Dickins #ifndef is_zero_pfn 72262eede62SHugh Dickins static inline int is_zero_pfn(unsigned long pfn) 72362eede62SHugh Dickins { 72462eede62SHugh Dickins return pfn == zero_pfn; 72562eede62SHugh Dickins } 72662eede62SHugh Dickins #endif 72762eede62SHugh Dickins 72862eede62SHugh Dickins #ifndef my_zero_pfn 72962eede62SHugh Dickins static inline unsigned long my_zero_pfn(unsigned long addr) 73062eede62SHugh Dickins { 73162eede62SHugh Dickins return zero_pfn; 73262eede62SHugh Dickins } 73362eede62SHugh Dickins #endif 73462eede62SHugh Dickins 735b5810039SNick Piggin /* 7367e675137SNick Piggin * vm_normal_page -- This function gets the "struct page" associated with a pte. 7376aab341eSLinus Torvalds * 7387e675137SNick Piggin * "Special" mappings do not wish to be associated with a "struct page" (either 7397e675137SNick Piggin * it doesn't exist, or it exists but they don't want to touch it). In this 7407e675137SNick Piggin * case, NULL is returned here. "Normal" mappings do have a struct page. 741b379d790SJared Hulbert * 7427e675137SNick Piggin * There are 2 broad cases. Firstly, an architecture may define a pte_special() 7437e675137SNick Piggin * pte bit, in which case this function is trivial. Secondly, an architecture 7447e675137SNick Piggin * may not have a spare pte bit, which requires a more complicated scheme, 7457e675137SNick Piggin * described below. 7467e675137SNick Piggin * 7477e675137SNick Piggin * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a 7487e675137SNick Piggin * special mapping (even if there are underlying and valid "struct pages"). 7497e675137SNick Piggin * COWed pages of a VM_PFNMAP are always normal. 7506aab341eSLinus Torvalds * 751b379d790SJared Hulbert * The way we recognize COWed pages within VM_PFNMAP mappings is through the 752b379d790SJared Hulbert * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit 7537e675137SNick Piggin * set, and the vm_pgoff will point to the first PFN mapped: thus every special 7547e675137SNick Piggin * mapping will always honor the rule 7556aab341eSLinus Torvalds * 7566aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 7576aab341eSLinus Torvalds * 7587e675137SNick Piggin * And for normal mappings this is false. 759b379d790SJared Hulbert * 7607e675137SNick Piggin * This restricts such mappings to be a linear translation from virtual address 7617e675137SNick Piggin * to pfn. To get around this restriction, we allow arbitrary mappings so long 7627e675137SNick Piggin * as the vma is not a COW mapping; in that case, we know that all ptes are 7637e675137SNick Piggin * special (because none can have been COWed). 764b379d790SJared Hulbert * 765b379d790SJared Hulbert * 7667e675137SNick Piggin * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. 7677e675137SNick Piggin * 768b379d790SJared Hulbert * VM_MIXEDMAP mappings can likewise contain memory with or without "struct 769b379d790SJared Hulbert * page" backing, however the difference is that _all_ pages with a struct 770b379d790SJared Hulbert * page (that is, those where pfn_valid is true) are refcounted and considered 771b379d790SJared Hulbert * normal pages by the VM. The disadvantage is that pages are refcounted 772b379d790SJared Hulbert * (which can be slower and simply not an option for some PFNMAP users). The 773b379d790SJared Hulbert * advantage is that we don't have to follow the strict linearity rule of 774b379d790SJared Hulbert * PFNMAP mappings in order to support COWable mappings. 775b379d790SJared Hulbert * 776ee498ed7SHugh Dickins */ 7777e675137SNick Piggin #ifdef __HAVE_ARCH_PTE_SPECIAL 7787e675137SNick Piggin # define HAVE_PTE_SPECIAL 1 7797e675137SNick Piggin #else 7807e675137SNick Piggin # define HAVE_PTE_SPECIAL 0 7817e675137SNick Piggin #endif 7827e675137SNick Piggin struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 7837e675137SNick Piggin pte_t pte) 784ee498ed7SHugh Dickins { 78522b31eecSHugh Dickins unsigned long pfn = pte_pfn(pte); 7867e675137SNick Piggin 7877e675137SNick Piggin if (HAVE_PTE_SPECIAL) { 78822b31eecSHugh Dickins if (likely(!pte_special(pte))) 78922b31eecSHugh Dickins goto check_pfn; 790a13ea5b7SHugh Dickins if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 791a13ea5b7SHugh Dickins return NULL; 79262eede62SHugh Dickins if (!is_zero_pfn(pfn)) 79322b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 7947e675137SNick Piggin return NULL; 7957e675137SNick Piggin } 7967e675137SNick Piggin 7977e675137SNick Piggin /* !HAVE_PTE_SPECIAL case follows: */ 7987e675137SNick Piggin 799b379d790SJared Hulbert if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 800b379d790SJared Hulbert if (vma->vm_flags & VM_MIXEDMAP) { 801b379d790SJared Hulbert if (!pfn_valid(pfn)) 802b379d790SJared Hulbert return NULL; 803b379d790SJared Hulbert goto out; 804b379d790SJared Hulbert } else { 8057e675137SNick Piggin unsigned long off; 8067e675137SNick Piggin off = (addr - vma->vm_start) >> PAGE_SHIFT; 8076aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 8086aab341eSLinus Torvalds return NULL; 80967121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 810fb155c16SLinus Torvalds return NULL; 8116aab341eSLinus Torvalds } 812b379d790SJared Hulbert } 8136aab341eSLinus Torvalds 81462eede62SHugh Dickins if (is_zero_pfn(pfn)) 81562eede62SHugh Dickins return NULL; 81622b31eecSHugh Dickins check_pfn: 81722b31eecSHugh Dickins if (unlikely(pfn > highest_memmap_pfn)) { 81822b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 81922b31eecSHugh Dickins return NULL; 82022b31eecSHugh Dickins } 8216aab341eSLinus Torvalds 8226aab341eSLinus Torvalds /* 8237e675137SNick Piggin * NOTE! We still have PageReserved() pages in the page tables. 8247e675137SNick Piggin * eg. VDSO mappings can cause them to exist. 8256aab341eSLinus Torvalds */ 826b379d790SJared Hulbert out: 8276aab341eSLinus Torvalds return pfn_to_page(pfn); 828ee498ed7SHugh Dickins } 829ee498ed7SHugh Dickins 830ee498ed7SHugh Dickins /* 8311da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 8321da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 8331da177e4SLinus Torvalds * covered by this vma. 8341da177e4SLinus Torvalds */ 8351da177e4SLinus Torvalds 836570a335bSHugh Dickins static inline unsigned long 8371da177e4SLinus Torvalds copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 838b5810039SNick Piggin pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, 8398c103762SHugh Dickins unsigned long addr, int *rss) 8401da177e4SLinus Torvalds { 841b5810039SNick Piggin unsigned long vm_flags = vma->vm_flags; 8421da177e4SLinus Torvalds pte_t pte = *src_pte; 8431da177e4SLinus Torvalds struct page *page; 8441da177e4SLinus Torvalds 8451da177e4SLinus Torvalds /* pte contains position in swap or file, so copy. */ 8461da177e4SLinus Torvalds if (unlikely(!pte_present(pte))) { 8471da177e4SLinus Torvalds if (!pte_file(pte)) { 8480697212aSChristoph Lameter swp_entry_t entry = pte_to_swp_entry(pte); 8490697212aSChristoph Lameter 850570a335bSHugh Dickins if (swap_duplicate(entry) < 0) 851570a335bSHugh Dickins return entry.val; 852570a335bSHugh Dickins 8531da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 8541da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 8551da177e4SLinus Torvalds spin_lock(&mmlist_lock); 856f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 857f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 858f412ac08SHugh Dickins &src_mm->mmlist); 8591da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 8601da177e4SLinus Torvalds } 861b084d435SKAMEZAWA Hiroyuki if (likely(!non_swap_entry(entry))) 862b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]++; 8639f9f1acdSKonstantin Khlebnikov else if (is_migration_entry(entry)) { 8649f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 8659f9f1acdSKonstantin Khlebnikov 8669f9f1acdSKonstantin Khlebnikov if (PageAnon(page)) 8679f9f1acdSKonstantin Khlebnikov rss[MM_ANONPAGES]++; 8689f9f1acdSKonstantin Khlebnikov else 8699f9f1acdSKonstantin Khlebnikov rss[MM_FILEPAGES]++; 8709f9f1acdSKonstantin Khlebnikov 8719f9f1acdSKonstantin Khlebnikov if (is_write_migration_entry(entry) && 8720697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 8730697212aSChristoph Lameter /* 8749f9f1acdSKonstantin Khlebnikov * COW mappings require pages in both 8759f9f1acdSKonstantin Khlebnikov * parent and child to be set to read. 8760697212aSChristoph Lameter */ 8770697212aSChristoph Lameter make_migration_entry_read(&entry); 8780697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 8790697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 8800697212aSChristoph Lameter } 8811da177e4SLinus Torvalds } 8829f9f1acdSKonstantin Khlebnikov } 883ae859762SHugh Dickins goto out_set_pte; 8841da177e4SLinus Torvalds } 8851da177e4SLinus Torvalds 8861da177e4SLinus Torvalds /* 8871da177e4SLinus Torvalds * If it's a COW mapping, write protect it both 8881da177e4SLinus Torvalds * in the parent and the child 8891da177e4SLinus Torvalds */ 89067121172SLinus Torvalds if (is_cow_mapping(vm_flags)) { 8911da177e4SLinus Torvalds ptep_set_wrprotect(src_mm, addr, src_pte); 8923dc90795SZachary Amsden pte = pte_wrprotect(pte); 8931da177e4SLinus Torvalds } 8941da177e4SLinus Torvalds 8951da177e4SLinus Torvalds /* 8961da177e4SLinus Torvalds * If it's a shared mapping, mark it clean in 8971da177e4SLinus Torvalds * the child 8981da177e4SLinus Torvalds */ 8991da177e4SLinus Torvalds if (vm_flags & VM_SHARED) 9001da177e4SLinus Torvalds pte = pte_mkclean(pte); 9011da177e4SLinus Torvalds pte = pte_mkold(pte); 9026aab341eSLinus Torvalds 9036aab341eSLinus Torvalds page = vm_normal_page(vma, addr, pte); 9046aab341eSLinus Torvalds if (page) { 9051da177e4SLinus Torvalds get_page(page); 90621333b2bSHugh Dickins page_dup_rmap(page); 907d559db08SKAMEZAWA Hiroyuki if (PageAnon(page)) 908d559db08SKAMEZAWA Hiroyuki rss[MM_ANONPAGES]++; 909d559db08SKAMEZAWA Hiroyuki else 910d559db08SKAMEZAWA Hiroyuki rss[MM_FILEPAGES]++; 9116aab341eSLinus Torvalds } 912ae859762SHugh Dickins 913ae859762SHugh Dickins out_set_pte: 914ae859762SHugh Dickins set_pte_at(dst_mm, addr, dst_pte, pte); 915570a335bSHugh Dickins return 0; 9161da177e4SLinus Torvalds } 9171da177e4SLinus Torvalds 91871e3aac0SAndrea Arcangeli int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9191da177e4SLinus Torvalds pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, 9201da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9211da177e4SLinus Torvalds { 922c36987e2SDaisuke Nishimura pte_t *orig_src_pte, *orig_dst_pte; 9231da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 924c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 925e040f218SHugh Dickins int progress = 0; 926d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 927570a335bSHugh Dickins swp_entry_t entry = (swp_entry_t){0}; 9281da177e4SLinus Torvalds 9291da177e4SLinus Torvalds again: 930d559db08SKAMEZAWA Hiroyuki init_rss_vec(rss); 931d559db08SKAMEZAWA Hiroyuki 932c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 9331da177e4SLinus Torvalds if (!dst_pte) 9341da177e4SLinus Torvalds return -ENOMEM; 935ece0e2b6SPeter Zijlstra src_pte = pte_offset_map(src_pmd, addr); 9364c21e2f2SHugh Dickins src_ptl = pte_lockptr(src_mm, src_pmd); 937f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 938c36987e2SDaisuke Nishimura orig_src_pte = src_pte; 939c36987e2SDaisuke Nishimura orig_dst_pte = dst_pte; 9406606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 9411da177e4SLinus Torvalds 9421da177e4SLinus Torvalds do { 9431da177e4SLinus Torvalds /* 9441da177e4SLinus Torvalds * We are holding two locks at this point - either of them 9451da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 9461da177e4SLinus Torvalds */ 947e040f218SHugh Dickins if (progress >= 32) { 948e040f218SHugh Dickins progress = 0; 949e040f218SHugh Dickins if (need_resched() || 95095c354feSNick Piggin spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) 9511da177e4SLinus Torvalds break; 952e040f218SHugh Dickins } 9531da177e4SLinus Torvalds if (pte_none(*src_pte)) { 9541da177e4SLinus Torvalds progress++; 9551da177e4SLinus Torvalds continue; 9561da177e4SLinus Torvalds } 957570a335bSHugh Dickins entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, 958570a335bSHugh Dickins vma, addr, rss); 959570a335bSHugh Dickins if (entry.val) 960570a335bSHugh Dickins break; 9611da177e4SLinus Torvalds progress += 8; 9621da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 9631da177e4SLinus Torvalds 9646606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 965c74df32cSHugh Dickins spin_unlock(src_ptl); 966ece0e2b6SPeter Zijlstra pte_unmap(orig_src_pte); 967d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(dst_mm, rss); 968c36987e2SDaisuke Nishimura pte_unmap_unlock(orig_dst_pte, dst_ptl); 969c74df32cSHugh Dickins cond_resched(); 970570a335bSHugh Dickins 971570a335bSHugh Dickins if (entry.val) { 972570a335bSHugh Dickins if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) 973570a335bSHugh Dickins return -ENOMEM; 974570a335bSHugh Dickins progress = 0; 975570a335bSHugh Dickins } 9761da177e4SLinus Torvalds if (addr != end) 9771da177e4SLinus Torvalds goto again; 9781da177e4SLinus Torvalds return 0; 9791da177e4SLinus Torvalds } 9801da177e4SLinus Torvalds 9811da177e4SLinus Torvalds static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9821da177e4SLinus Torvalds pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, 9831da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9841da177e4SLinus Torvalds { 9851da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 9861da177e4SLinus Torvalds unsigned long next; 9871da177e4SLinus Torvalds 9881da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 9891da177e4SLinus Torvalds if (!dst_pmd) 9901da177e4SLinus Torvalds return -ENOMEM; 9911da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 9921da177e4SLinus Torvalds do { 9931da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 99471e3aac0SAndrea Arcangeli if (pmd_trans_huge(*src_pmd)) { 99571e3aac0SAndrea Arcangeli int err; 99614d1a55cSAndrea Arcangeli VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); 99771e3aac0SAndrea Arcangeli err = copy_huge_pmd(dst_mm, src_mm, 99871e3aac0SAndrea Arcangeli dst_pmd, src_pmd, addr, vma); 99971e3aac0SAndrea Arcangeli if (err == -ENOMEM) 100071e3aac0SAndrea Arcangeli return -ENOMEM; 100171e3aac0SAndrea Arcangeli if (!err) 100271e3aac0SAndrea Arcangeli continue; 100371e3aac0SAndrea Arcangeli /* fall through */ 100471e3aac0SAndrea Arcangeli } 10051da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 10061da177e4SLinus Torvalds continue; 10071da177e4SLinus Torvalds if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, 10081da177e4SLinus Torvalds vma, addr, next)) 10091da177e4SLinus Torvalds return -ENOMEM; 10101da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 10111da177e4SLinus Torvalds return 0; 10121da177e4SLinus Torvalds } 10131da177e4SLinus Torvalds 10141da177e4SLinus Torvalds static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 10151da177e4SLinus Torvalds pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, 10161da177e4SLinus Torvalds unsigned long addr, unsigned long end) 10171da177e4SLinus Torvalds { 10181da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 10191da177e4SLinus Torvalds unsigned long next; 10201da177e4SLinus Torvalds 10211da177e4SLinus Torvalds dst_pud = pud_alloc(dst_mm, dst_pgd, addr); 10221da177e4SLinus Torvalds if (!dst_pud) 10231da177e4SLinus Torvalds return -ENOMEM; 10241da177e4SLinus Torvalds src_pud = pud_offset(src_pgd, addr); 10251da177e4SLinus Torvalds do { 10261da177e4SLinus Torvalds next = pud_addr_end(addr, end); 10271da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 10281da177e4SLinus Torvalds continue; 10291da177e4SLinus Torvalds if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, 10301da177e4SLinus Torvalds vma, addr, next)) 10311da177e4SLinus Torvalds return -ENOMEM; 10321da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 10331da177e4SLinus Torvalds return 0; 10341da177e4SLinus Torvalds } 10351da177e4SLinus Torvalds 10361da177e4SLinus Torvalds int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 10371da177e4SLinus Torvalds struct vm_area_struct *vma) 10381da177e4SLinus Torvalds { 10391da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 10401da177e4SLinus Torvalds unsigned long next; 10411da177e4SLinus Torvalds unsigned long addr = vma->vm_start; 10421da177e4SLinus Torvalds unsigned long end = vma->vm_end; 10432ec74c3eSSagi Grimberg unsigned long mmun_start; /* For mmu_notifiers */ 10442ec74c3eSSagi Grimberg unsigned long mmun_end; /* For mmu_notifiers */ 10452ec74c3eSSagi Grimberg bool is_cow; 1046cddb8a5cSAndrea Arcangeli int ret; 10471da177e4SLinus Torvalds 1048d992895bSNick Piggin /* 1049d992895bSNick Piggin * Don't copy ptes where a page fault will fill them correctly. 1050d992895bSNick Piggin * Fork becomes much lighter when there are big shared or private 1051d992895bSNick Piggin * readonly mappings. The tradeoff is that copy_page_range is more 1052d992895bSNick Piggin * efficient than faulting. 1053d992895bSNick Piggin */ 10544b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & (VM_HUGETLB | VM_NONLINEAR | 10554b6e1e37SKonstantin Khlebnikov VM_PFNMAP | VM_MIXEDMAP))) { 1056d992895bSNick Piggin if (!vma->anon_vma) 1057d992895bSNick Piggin return 0; 1058d992895bSNick Piggin } 1059d992895bSNick Piggin 10601da177e4SLinus Torvalds if (is_vm_hugetlb_page(vma)) 10611da177e4SLinus Torvalds return copy_hugetlb_page_range(dst_mm, src_mm, vma); 10621da177e4SLinus Torvalds 1063b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) { 10642ab64037Svenkatesh.pallipadi@intel.com /* 10652ab64037Svenkatesh.pallipadi@intel.com * We do not free on error cases below as remove_vma 10662ab64037Svenkatesh.pallipadi@intel.com * gets called on error from higher level routine 10672ab64037Svenkatesh.pallipadi@intel.com */ 10685180da41SSuresh Siddha ret = track_pfn_copy(vma); 10692ab64037Svenkatesh.pallipadi@intel.com if (ret) 10702ab64037Svenkatesh.pallipadi@intel.com return ret; 10712ab64037Svenkatesh.pallipadi@intel.com } 10722ab64037Svenkatesh.pallipadi@intel.com 1073cddb8a5cSAndrea Arcangeli /* 1074cddb8a5cSAndrea Arcangeli * We need to invalidate the secondary MMU mappings only when 1075cddb8a5cSAndrea Arcangeli * there could be a permission downgrade on the ptes of the 1076cddb8a5cSAndrea Arcangeli * parent mm. And a permission downgrade will only happen if 1077cddb8a5cSAndrea Arcangeli * is_cow_mapping() returns true. 1078cddb8a5cSAndrea Arcangeli */ 10792ec74c3eSSagi Grimberg is_cow = is_cow_mapping(vma->vm_flags); 10802ec74c3eSSagi Grimberg mmun_start = addr; 10812ec74c3eSSagi Grimberg mmun_end = end; 10822ec74c3eSSagi Grimberg if (is_cow) 10832ec74c3eSSagi Grimberg mmu_notifier_invalidate_range_start(src_mm, mmun_start, 10842ec74c3eSSagi Grimberg mmun_end); 1085cddb8a5cSAndrea Arcangeli 1086cddb8a5cSAndrea Arcangeli ret = 0; 10871da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 10881da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 10891da177e4SLinus Torvalds do { 10901da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 10911da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 10921da177e4SLinus Torvalds continue; 1093cddb8a5cSAndrea Arcangeli if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, 1094cddb8a5cSAndrea Arcangeli vma, addr, next))) { 1095cddb8a5cSAndrea Arcangeli ret = -ENOMEM; 1096cddb8a5cSAndrea Arcangeli break; 1097cddb8a5cSAndrea Arcangeli } 10981da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 1099cddb8a5cSAndrea Arcangeli 11002ec74c3eSSagi Grimberg if (is_cow) 11012ec74c3eSSagi Grimberg mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); 1102cddb8a5cSAndrea Arcangeli return ret; 11031da177e4SLinus Torvalds } 11041da177e4SLinus Torvalds 110551c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 1106b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 11071da177e4SLinus Torvalds unsigned long addr, unsigned long end, 110897a89413SPeter Zijlstra struct zap_details *details) 11091da177e4SLinus Torvalds { 1110b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 1111d16dfc55SPeter Zijlstra int force_flush = 0; 1112d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 111397a89413SPeter Zijlstra spinlock_t *ptl; 11145f1a1907SSteven Rostedt pte_t *start_pte; 111597a89413SPeter Zijlstra pte_t *pte; 1116d559db08SKAMEZAWA Hiroyuki 1117d16dfc55SPeter Zijlstra again: 1118e303297eSPeter Zijlstra init_rss_vec(rss); 11195f1a1907SSteven Rostedt start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 11205f1a1907SSteven Rostedt pte = start_pte; 11216606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 11221da177e4SLinus Torvalds do { 11231da177e4SLinus Torvalds pte_t ptent = *pte; 112451c6f666SRobin Holt if (pte_none(ptent)) { 11251da177e4SLinus Torvalds continue; 112651c6f666SRobin Holt } 112751c6f666SRobin Holt 11286f5e6b9eSHugh Dickins if (pte_present(ptent)) { 11296f5e6b9eSHugh Dickins struct page *page; 11306f5e6b9eSHugh Dickins 11316aab341eSLinus Torvalds page = vm_normal_page(vma, addr, ptent); 11321da177e4SLinus Torvalds if (unlikely(details) && page) { 11331da177e4SLinus Torvalds /* 11341da177e4SLinus Torvalds * unmap_shared_mapping_pages() wants to 11351da177e4SLinus Torvalds * invalidate cache without truncating: 11361da177e4SLinus Torvalds * unmap shared but keep private pages. 11371da177e4SLinus Torvalds */ 11381da177e4SLinus Torvalds if (details->check_mapping && 11391da177e4SLinus Torvalds details->check_mapping != page->mapping) 11401da177e4SLinus Torvalds continue; 11411da177e4SLinus Torvalds /* 11421da177e4SLinus Torvalds * Each page->index must be checked when 11431da177e4SLinus Torvalds * invalidating or truncating nonlinear. 11441da177e4SLinus Torvalds */ 11451da177e4SLinus Torvalds if (details->nonlinear_vma && 11461da177e4SLinus Torvalds (page->index < details->first_index || 11471da177e4SLinus Torvalds page->index > details->last_index)) 11481da177e4SLinus Torvalds continue; 11491da177e4SLinus Torvalds } 1150b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 1151a600388dSZachary Amsden tlb->fullmm); 11521da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 11531da177e4SLinus Torvalds if (unlikely(!page)) 11541da177e4SLinus Torvalds continue; 11551da177e4SLinus Torvalds if (unlikely(details) && details->nonlinear_vma 11561da177e4SLinus Torvalds && linear_page_index(details->nonlinear_vma, 11571da177e4SLinus Torvalds addr) != page->index) 1158b5810039SNick Piggin set_pte_at(mm, addr, pte, 11591da177e4SLinus Torvalds pgoff_to_pte(page->index)); 11601da177e4SLinus Torvalds if (PageAnon(page)) 1161d559db08SKAMEZAWA Hiroyuki rss[MM_ANONPAGES]--; 11626237bcd9SHugh Dickins else { 11636237bcd9SHugh Dickins if (pte_dirty(ptent)) 11646237bcd9SHugh Dickins set_page_dirty(page); 11654917e5d0SJohannes Weiner if (pte_young(ptent) && 11664917e5d0SJohannes Weiner likely(!VM_SequentialReadHint(vma))) 1167bf3f3bc5SNick Piggin mark_page_accessed(page); 1168d559db08SKAMEZAWA Hiroyuki rss[MM_FILEPAGES]--; 11696237bcd9SHugh Dickins } 1170edc315fdSHugh Dickins page_remove_rmap(page); 11713dc14741SHugh Dickins if (unlikely(page_mapcount(page) < 0)) 11723dc14741SHugh Dickins print_bad_pte(vma, addr, ptent, page); 1173d16dfc55SPeter Zijlstra force_flush = !__tlb_remove_page(tlb, page); 1174d16dfc55SPeter Zijlstra if (force_flush) 1175d16dfc55SPeter Zijlstra break; 11761da177e4SLinus Torvalds continue; 11771da177e4SLinus Torvalds } 11781da177e4SLinus Torvalds /* 11791da177e4SLinus Torvalds * If details->check_mapping, we leave swap entries; 11801da177e4SLinus Torvalds * if details->nonlinear_vma, we leave file entries. 11811da177e4SLinus Torvalds */ 11821da177e4SLinus Torvalds if (unlikely(details)) 11831da177e4SLinus Torvalds continue; 11842509ef26SHugh Dickins if (pte_file(ptent)) { 11852509ef26SHugh Dickins if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) 11862509ef26SHugh Dickins print_bad_pte(vma, addr, ptent, NULL); 1187b084d435SKAMEZAWA Hiroyuki } else { 1188b084d435SKAMEZAWA Hiroyuki swp_entry_t entry = pte_to_swp_entry(ptent); 1189b084d435SKAMEZAWA Hiroyuki 1190b084d435SKAMEZAWA Hiroyuki if (!non_swap_entry(entry)) 1191b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 11929f9f1acdSKonstantin Khlebnikov else if (is_migration_entry(entry)) { 11939f9f1acdSKonstantin Khlebnikov struct page *page; 11949f9f1acdSKonstantin Khlebnikov 11959f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 11969f9f1acdSKonstantin Khlebnikov 11979f9f1acdSKonstantin Khlebnikov if (PageAnon(page)) 11989f9f1acdSKonstantin Khlebnikov rss[MM_ANONPAGES]--; 11999f9f1acdSKonstantin Khlebnikov else 12009f9f1acdSKonstantin Khlebnikov rss[MM_FILEPAGES]--; 12019f9f1acdSKonstantin Khlebnikov } 1202b084d435SKAMEZAWA Hiroyuki if (unlikely(!free_swap_and_cache(entry))) 12032509ef26SHugh Dickins print_bad_pte(vma, addr, ptent, NULL); 1204b084d435SKAMEZAWA Hiroyuki } 12059888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 120697a89413SPeter Zijlstra } while (pte++, addr += PAGE_SIZE, addr != end); 1207ae859762SHugh Dickins 1208d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 12096606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 12105f1a1907SSteven Rostedt pte_unmap_unlock(start_pte, ptl); 121151c6f666SRobin Holt 1212d16dfc55SPeter Zijlstra /* 1213d16dfc55SPeter Zijlstra * mmu_gather ran out of room to batch pages, we break out of 1214d16dfc55SPeter Zijlstra * the PTE lock to avoid doing the potential expensive TLB invalidate 1215d16dfc55SPeter Zijlstra * and page-free while holding it. 1216d16dfc55SPeter Zijlstra */ 1217d16dfc55SPeter Zijlstra if (force_flush) { 1218d16dfc55SPeter Zijlstra force_flush = 0; 1219597e1c35SAlex Shi 1220597e1c35SAlex Shi #ifdef HAVE_GENERIC_MMU_GATHER 1221597e1c35SAlex Shi tlb->start = addr; 1222597e1c35SAlex Shi tlb->end = end; 1223597e1c35SAlex Shi #endif 1224d16dfc55SPeter Zijlstra tlb_flush_mmu(tlb); 1225d16dfc55SPeter Zijlstra if (addr != end) 1226d16dfc55SPeter Zijlstra goto again; 1227d16dfc55SPeter Zijlstra } 1228d16dfc55SPeter Zijlstra 122951c6f666SRobin Holt return addr; 12301da177e4SLinus Torvalds } 12311da177e4SLinus Torvalds 123251c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1233b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 12341da177e4SLinus Torvalds unsigned long addr, unsigned long end, 123597a89413SPeter Zijlstra struct zap_details *details) 12361da177e4SLinus Torvalds { 12371da177e4SLinus Torvalds pmd_t *pmd; 12381da177e4SLinus Torvalds unsigned long next; 12391da177e4SLinus Torvalds 12401da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 12411da177e4SLinus Torvalds do { 12421da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 124371e3aac0SAndrea Arcangeli if (pmd_trans_huge(*pmd)) { 124414d1a55cSAndrea Arcangeli if (next - addr != HPAGE_PMD_SIZE) { 1245e0897d75SDavid Rientjes #ifdef CONFIG_DEBUG_VM 1246e0897d75SDavid Rientjes if (!rwsem_is_locked(&tlb->mm->mmap_sem)) { 1247e0897d75SDavid Rientjes pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n", 1248e0897d75SDavid Rientjes __func__, addr, end, 1249e0897d75SDavid Rientjes vma->vm_start, 1250e0897d75SDavid Rientjes vma->vm_end); 1251e0897d75SDavid Rientjes BUG(); 1252e0897d75SDavid Rientjes } 1253e0897d75SDavid Rientjes #endif 125471e3aac0SAndrea Arcangeli split_huge_page_pmd(vma->vm_mm, pmd); 1255f21760b1SShaohua Li } else if (zap_huge_pmd(tlb, vma, pmd, addr)) 12561a5a9906SAndrea Arcangeli goto next; 125771e3aac0SAndrea Arcangeli /* fall through */ 125871e3aac0SAndrea Arcangeli } 12591a5a9906SAndrea Arcangeli /* 12601a5a9906SAndrea Arcangeli * Here there can be other concurrent MADV_DONTNEED or 12611a5a9906SAndrea Arcangeli * trans huge page faults running, and if the pmd is 12621a5a9906SAndrea Arcangeli * none or trans huge it can change under us. This is 12631a5a9906SAndrea Arcangeli * because MADV_DONTNEED holds the mmap_sem in read 12641a5a9906SAndrea Arcangeli * mode. 12651a5a9906SAndrea Arcangeli */ 12661a5a9906SAndrea Arcangeli if (pmd_none_or_trans_huge_or_clear_bad(pmd)) 12671a5a9906SAndrea Arcangeli goto next; 126897a89413SPeter Zijlstra next = zap_pte_range(tlb, vma, pmd, addr, next, details); 12691a5a9906SAndrea Arcangeli next: 127097a89413SPeter Zijlstra cond_resched(); 127197a89413SPeter Zijlstra } while (pmd++, addr = next, addr != end); 127251c6f666SRobin Holt 127351c6f666SRobin Holt return addr; 12741da177e4SLinus Torvalds } 12751da177e4SLinus Torvalds 127651c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1277b5810039SNick Piggin struct vm_area_struct *vma, pgd_t *pgd, 12781da177e4SLinus Torvalds unsigned long addr, unsigned long end, 127997a89413SPeter Zijlstra struct zap_details *details) 12801da177e4SLinus Torvalds { 12811da177e4SLinus Torvalds pud_t *pud; 12821da177e4SLinus Torvalds unsigned long next; 12831da177e4SLinus Torvalds 12841da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 12851da177e4SLinus Torvalds do { 12861da177e4SLinus Torvalds next = pud_addr_end(addr, end); 128797a89413SPeter Zijlstra if (pud_none_or_clear_bad(pud)) 12881da177e4SLinus Torvalds continue; 128997a89413SPeter Zijlstra next = zap_pmd_range(tlb, vma, pud, addr, next, details); 129097a89413SPeter Zijlstra } while (pud++, addr = next, addr != end); 129151c6f666SRobin Holt 129251c6f666SRobin Holt return addr; 12931da177e4SLinus Torvalds } 12941da177e4SLinus Torvalds 1295038c7aa1SAl Viro static void unmap_page_range(struct mmu_gather *tlb, 129651c6f666SRobin Holt struct vm_area_struct *vma, 12971da177e4SLinus Torvalds unsigned long addr, unsigned long end, 129897a89413SPeter Zijlstra struct zap_details *details) 12991da177e4SLinus Torvalds { 13001da177e4SLinus Torvalds pgd_t *pgd; 13011da177e4SLinus Torvalds unsigned long next; 13021da177e4SLinus Torvalds 13031da177e4SLinus Torvalds if (details && !details->check_mapping && !details->nonlinear_vma) 13041da177e4SLinus Torvalds details = NULL; 13051da177e4SLinus Torvalds 13061da177e4SLinus Torvalds BUG_ON(addr >= end); 1307569b846dSKAMEZAWA Hiroyuki mem_cgroup_uncharge_start(); 13081da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 13091da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 13101da177e4SLinus Torvalds do { 13111da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 131297a89413SPeter Zijlstra if (pgd_none_or_clear_bad(pgd)) 13131da177e4SLinus Torvalds continue; 131497a89413SPeter Zijlstra next = zap_pud_range(tlb, vma, pgd, addr, next, details); 131597a89413SPeter Zijlstra } while (pgd++, addr = next, addr != end); 13161da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 1317569b846dSKAMEZAWA Hiroyuki mem_cgroup_uncharge_end(); 13181da177e4SLinus Torvalds } 13191da177e4SLinus Torvalds 1320f5cc4eefSAl Viro 1321f5cc4eefSAl Viro static void unmap_single_vma(struct mmu_gather *tlb, 13221da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 13234f74d2c8SLinus Torvalds unsigned long end_addr, 13241da177e4SLinus Torvalds struct zap_details *details) 13251da177e4SLinus Torvalds { 1326f5cc4eefSAl Viro unsigned long start = max(vma->vm_start, start_addr); 13271da177e4SLinus Torvalds unsigned long end; 13281da177e4SLinus Torvalds 13291da177e4SLinus Torvalds if (start >= vma->vm_end) 1330f5cc4eefSAl Viro return; 13311da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 13321da177e4SLinus Torvalds if (end <= vma->vm_start) 1333f5cc4eefSAl Viro return; 13341da177e4SLinus Torvalds 1335cbc91f71SSrikar Dronamraju if (vma->vm_file) 1336cbc91f71SSrikar Dronamraju uprobe_munmap(vma, start, end); 1337cbc91f71SSrikar Dronamraju 1338b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) 13395180da41SSuresh Siddha untrack_pfn(vma, 0, 0); 13402ab64037Svenkatesh.pallipadi@intel.com 13418b2a1238SAl Viro if (start != end) { 134251c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1343a137e1ccSAndi Kleen /* 1344a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1345a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1346a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 1347a137e1ccSAndi Kleen * cleanup path of do_mmap_pgoff. When 1348a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 1349a137e1ccSAndi Kleen * do_mmap_pgoff() nullifies vma->vm_file 1350a137e1ccSAndi Kleen * before calling this function to clean up. 1351a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1352a137e1ccSAndi Kleen * safe to do nothing in this case. 1353a137e1ccSAndi Kleen */ 135424669e58SAneesh Kumar K.V if (vma->vm_file) { 135524669e58SAneesh Kumar K.V mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); 1356d833352aSMel Gorman __unmap_hugepage_range_final(tlb, vma, start, end, NULL); 135724669e58SAneesh Kumar K.V mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); 135824669e58SAneesh Kumar K.V } 135951c6f666SRobin Holt } else 1360038c7aa1SAl Viro unmap_page_range(tlb, vma, start, end, details); 136197a89413SPeter Zijlstra } 136251c6f666SRobin Holt } 13631da177e4SLinus Torvalds 1364f5cc4eefSAl Viro /** 1365f5cc4eefSAl Viro * unmap_vmas - unmap a range of memory covered by a list of vma's 1366f5cc4eefSAl Viro * @tlb: address of the caller's struct mmu_gather 1367f5cc4eefSAl Viro * @vma: the starting vma 1368f5cc4eefSAl Viro * @start_addr: virtual address at which to start unmapping 1369f5cc4eefSAl Viro * @end_addr: virtual address at which to end unmapping 1370f5cc4eefSAl Viro * 1371f5cc4eefSAl Viro * Unmap all pages in the vma list. 1372f5cc4eefSAl Viro * 1373f5cc4eefSAl Viro * Only addresses between `start' and `end' will be unmapped. 1374f5cc4eefSAl Viro * 1375f5cc4eefSAl Viro * The VMA list must be sorted in ascending virtual address order. 1376f5cc4eefSAl Viro * 1377f5cc4eefSAl Viro * unmap_vmas() assumes that the caller will flush the whole unmapped address 1378f5cc4eefSAl Viro * range after unmap_vmas() returns. So the only responsibility here is to 1379f5cc4eefSAl Viro * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 1380f5cc4eefSAl Viro * drops the lock and schedules. 1381f5cc4eefSAl Viro */ 1382f5cc4eefSAl Viro void unmap_vmas(struct mmu_gather *tlb, 1383f5cc4eefSAl Viro struct vm_area_struct *vma, unsigned long start_addr, 13844f74d2c8SLinus Torvalds unsigned long end_addr) 1385f5cc4eefSAl Viro { 1386f5cc4eefSAl Viro struct mm_struct *mm = vma->vm_mm; 1387f5cc4eefSAl Viro 1388f5cc4eefSAl Viro mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); 1389f5cc4eefSAl Viro for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) 13904f74d2c8SLinus Torvalds unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); 1391cddb8a5cSAndrea Arcangeli mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); 13921da177e4SLinus Torvalds } 13931da177e4SLinus Torvalds 13941da177e4SLinus Torvalds /** 13951da177e4SLinus Torvalds * zap_page_range - remove user pages in a given range 13961da177e4SLinus Torvalds * @vma: vm_area_struct holding the applicable pages 1397eb4546bbSRandy Dunlap * @start: starting address of pages to zap 13981da177e4SLinus Torvalds * @size: number of bytes to zap 13991da177e4SLinus Torvalds * @details: details of nonlinear truncation or shared cache invalidation 1400f5cc4eefSAl Viro * 1401f5cc4eefSAl Viro * Caller must protect the VMA list 14021da177e4SLinus Torvalds */ 14037e027b14SLinus Torvalds void zap_page_range(struct vm_area_struct *vma, unsigned long start, 14041da177e4SLinus Torvalds unsigned long size, struct zap_details *details) 14051da177e4SLinus Torvalds { 14061da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 1407d16dfc55SPeter Zijlstra struct mmu_gather tlb; 14087e027b14SLinus Torvalds unsigned long end = start + size; 14091da177e4SLinus Torvalds 14101da177e4SLinus Torvalds lru_add_drain(); 1411d16dfc55SPeter Zijlstra tlb_gather_mmu(&tlb, mm, 0); 1412365e9c87SHugh Dickins update_hiwater_rss(mm); 14137e027b14SLinus Torvalds mmu_notifier_invalidate_range_start(mm, start, end); 14147e027b14SLinus Torvalds for ( ; vma && vma->vm_start < end; vma = vma->vm_next) 14154f74d2c8SLinus Torvalds unmap_single_vma(&tlb, vma, start, end, details); 14167e027b14SLinus Torvalds mmu_notifier_invalidate_range_end(mm, start, end); 14177e027b14SLinus Torvalds tlb_finish_mmu(&tlb, start, end); 14181da177e4SLinus Torvalds } 14191da177e4SLinus Torvalds 1420c627f9ccSJack Steiner /** 1421f5cc4eefSAl Viro * zap_page_range_single - remove user pages in a given range 1422f5cc4eefSAl Viro * @vma: vm_area_struct holding the applicable pages 1423f5cc4eefSAl Viro * @address: starting address of pages to zap 1424f5cc4eefSAl Viro * @size: number of bytes to zap 1425f5cc4eefSAl Viro * @details: details of nonlinear truncation or shared cache invalidation 1426f5cc4eefSAl Viro * 1427f5cc4eefSAl Viro * The range must fit into one VMA. 1428f5cc4eefSAl Viro */ 1429f5cc4eefSAl Viro static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 1430f5cc4eefSAl Viro unsigned long size, struct zap_details *details) 1431f5cc4eefSAl Viro { 1432f5cc4eefSAl Viro struct mm_struct *mm = vma->vm_mm; 1433f5cc4eefSAl Viro struct mmu_gather tlb; 1434f5cc4eefSAl Viro unsigned long end = address + size; 1435f5cc4eefSAl Viro 1436f5cc4eefSAl Viro lru_add_drain(); 1437f5cc4eefSAl Viro tlb_gather_mmu(&tlb, mm, 0); 1438f5cc4eefSAl Viro update_hiwater_rss(mm); 1439f5cc4eefSAl Viro mmu_notifier_invalidate_range_start(mm, address, end); 14404f74d2c8SLinus Torvalds unmap_single_vma(&tlb, vma, address, end, details); 1441f5cc4eefSAl Viro mmu_notifier_invalidate_range_end(mm, address, end); 1442f5cc4eefSAl Viro tlb_finish_mmu(&tlb, address, end); 14431da177e4SLinus Torvalds } 14441da177e4SLinus Torvalds 1445c627f9ccSJack Steiner /** 1446c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1447c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1448c627f9ccSJack Steiner * @address: starting address of pages to zap 1449c627f9ccSJack Steiner * @size: number of bytes to zap 1450c627f9ccSJack Steiner * 1451c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1452c627f9ccSJack Steiner * 1453c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1454c627f9ccSJack Steiner * 1455c627f9ccSJack Steiner * Returns 0 if successful. 1456c627f9ccSJack Steiner */ 1457c627f9ccSJack Steiner int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1458c627f9ccSJack Steiner unsigned long size) 1459c627f9ccSJack Steiner { 1460c627f9ccSJack Steiner if (address < vma->vm_start || address + size > vma->vm_end || 1461c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 1462c627f9ccSJack Steiner return -1; 1463f5cc4eefSAl Viro zap_page_range_single(vma, address, size, NULL); 1464c627f9ccSJack Steiner return 0; 1465c627f9ccSJack Steiner } 1466c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1467c627f9ccSJack Steiner 1468142762bdSJohannes Weiner /** 1469142762bdSJohannes Weiner * follow_page - look up a page descriptor from a user-virtual address 1470142762bdSJohannes Weiner * @vma: vm_area_struct mapping @address 1471142762bdSJohannes Weiner * @address: virtual address to look up 1472142762bdSJohannes Weiner * @flags: flags modifying lookup behaviour 1473142762bdSJohannes Weiner * 1474142762bdSJohannes Weiner * @flags can have FOLL_ flags set, defined in <linux/mm.h> 1475142762bdSJohannes Weiner * 1476142762bdSJohannes Weiner * Returns the mapped (struct page *), %NULL if no mapping exists, or 1477142762bdSJohannes Weiner * an error pointer if there is a mapping to something not represented 1478142762bdSJohannes Weiner * by a page descriptor (see also vm_normal_page()). 14791da177e4SLinus Torvalds */ 14806aab341eSLinus Torvalds struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 1481deceb6cdSHugh Dickins unsigned int flags) 14821da177e4SLinus Torvalds { 14831da177e4SLinus Torvalds pgd_t *pgd; 14841da177e4SLinus Torvalds pud_t *pud; 14851da177e4SLinus Torvalds pmd_t *pmd; 14861da177e4SLinus Torvalds pte_t *ptep, pte; 1487deceb6cdSHugh Dickins spinlock_t *ptl; 14881da177e4SLinus Torvalds struct page *page; 14896aab341eSLinus Torvalds struct mm_struct *mm = vma->vm_mm; 14901da177e4SLinus Torvalds 1491deceb6cdSHugh Dickins page = follow_huge_addr(mm, address, flags & FOLL_WRITE); 1492deceb6cdSHugh Dickins if (!IS_ERR(page)) { 1493deceb6cdSHugh Dickins BUG_ON(flags & FOLL_GET); 1494deceb6cdSHugh Dickins goto out; 1495deceb6cdSHugh Dickins } 14961da177e4SLinus Torvalds 1497deceb6cdSHugh Dickins page = NULL; 14981da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 14991da177e4SLinus Torvalds if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 1500deceb6cdSHugh Dickins goto no_page_table; 15011da177e4SLinus Torvalds 15021da177e4SLinus Torvalds pud = pud_offset(pgd, address); 1503ceb86879SAndi Kleen if (pud_none(*pud)) 1504ceb86879SAndi Kleen goto no_page_table; 15058a07651eSHugh Dickins if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) { 1506ceb86879SAndi Kleen BUG_ON(flags & FOLL_GET); 1507ceb86879SAndi Kleen page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE); 1508ceb86879SAndi Kleen goto out; 1509ceb86879SAndi Kleen } 1510ceb86879SAndi Kleen if (unlikely(pud_bad(*pud))) 1511deceb6cdSHugh Dickins goto no_page_table; 15121da177e4SLinus Torvalds 15131da177e4SLinus Torvalds pmd = pmd_offset(pud, address); 1514aeed5fceSHugh Dickins if (pmd_none(*pmd)) 1515deceb6cdSHugh Dickins goto no_page_table; 151671e3aac0SAndrea Arcangeli if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) { 1517deceb6cdSHugh Dickins BUG_ON(flags & FOLL_GET); 1518deceb6cdSHugh Dickins page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); 1519deceb6cdSHugh Dickins goto out; 1520deceb6cdSHugh Dickins } 15210b9d7052SAndrea Arcangeli if ((flags & FOLL_NUMA) && pmd_numa(*pmd)) 15220b9d7052SAndrea Arcangeli goto no_page_table; 152371e3aac0SAndrea Arcangeli if (pmd_trans_huge(*pmd)) { 1524500d65d4SAndrea Arcangeli if (flags & FOLL_SPLIT) { 1525500d65d4SAndrea Arcangeli split_huge_page_pmd(mm, pmd); 1526500d65d4SAndrea Arcangeli goto split_fallthrough; 1527500d65d4SAndrea Arcangeli } 152871e3aac0SAndrea Arcangeli spin_lock(&mm->page_table_lock); 152971e3aac0SAndrea Arcangeli if (likely(pmd_trans_huge(*pmd))) { 153071e3aac0SAndrea Arcangeli if (unlikely(pmd_trans_splitting(*pmd))) { 153171e3aac0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 153271e3aac0SAndrea Arcangeli wait_split_huge_page(vma->anon_vma, pmd); 153371e3aac0SAndrea Arcangeli } else { 1534b676b293SDavid Rientjes page = follow_trans_huge_pmd(vma, address, 153571e3aac0SAndrea Arcangeli pmd, flags); 153671e3aac0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 153771e3aac0SAndrea Arcangeli goto out; 153871e3aac0SAndrea Arcangeli } 153971e3aac0SAndrea Arcangeli } else 154071e3aac0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 154171e3aac0SAndrea Arcangeli /* fall through */ 154271e3aac0SAndrea Arcangeli } 1543500d65d4SAndrea Arcangeli split_fallthrough: 1544aeed5fceSHugh Dickins if (unlikely(pmd_bad(*pmd))) 1545aeed5fceSHugh Dickins goto no_page_table; 1546aeed5fceSHugh Dickins 1547deceb6cdSHugh Dickins ptep = pte_offset_map_lock(mm, pmd, address, &ptl); 15481da177e4SLinus Torvalds 15491da177e4SLinus Torvalds pte = *ptep; 1550deceb6cdSHugh Dickins if (!pte_present(pte)) 155189f5b7daSLinus Torvalds goto no_page; 15520b9d7052SAndrea Arcangeli if ((flags & FOLL_NUMA) && pte_numa(pte)) 15530b9d7052SAndrea Arcangeli goto no_page; 1554deceb6cdSHugh Dickins if ((flags & FOLL_WRITE) && !pte_write(pte)) 1555deceb6cdSHugh Dickins goto unlock; 1556a13ea5b7SHugh Dickins 15576aab341eSLinus Torvalds page = vm_normal_page(vma, address, pte); 1558a13ea5b7SHugh Dickins if (unlikely(!page)) { 1559a13ea5b7SHugh Dickins if ((flags & FOLL_DUMP) || 156062eede62SHugh Dickins !is_zero_pfn(pte_pfn(pte))) 156189f5b7daSLinus Torvalds goto bad_page; 1562a13ea5b7SHugh Dickins page = pte_page(pte); 1563a13ea5b7SHugh Dickins } 1564deceb6cdSHugh Dickins 1565deceb6cdSHugh Dickins if (flags & FOLL_GET) 156670b50f94SAndrea Arcangeli get_page_foll(page); 1567deceb6cdSHugh Dickins if (flags & FOLL_TOUCH) { 1568deceb6cdSHugh Dickins if ((flags & FOLL_WRITE) && 1569deceb6cdSHugh Dickins !pte_dirty(pte) && !PageDirty(page)) 1570f33ea7f4SNick Piggin set_page_dirty(page); 1571bd775c42SKOSAKI Motohiro /* 1572bd775c42SKOSAKI Motohiro * pte_mkyoung() would be more correct here, but atomic care 1573bd775c42SKOSAKI Motohiro * is needed to avoid losing the dirty bit: it is easier to use 1574bd775c42SKOSAKI Motohiro * mark_page_accessed(). 1575bd775c42SKOSAKI Motohiro */ 15761da177e4SLinus Torvalds mark_page_accessed(page); 15771da177e4SLinus Torvalds } 1578a1fde08cSLinus Torvalds if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) { 1579110d74a9SMichel Lespinasse /* 1580110d74a9SMichel Lespinasse * The preliminary mapping check is mainly to avoid the 1581110d74a9SMichel Lespinasse * pointless overhead of lock_page on the ZERO_PAGE 1582110d74a9SMichel Lespinasse * which might bounce very badly if there is contention. 1583110d74a9SMichel Lespinasse * 1584110d74a9SMichel Lespinasse * If the page is already locked, we don't need to 1585110d74a9SMichel Lespinasse * handle it now - vmscan will handle it later if and 1586110d74a9SMichel Lespinasse * when it attempts to reclaim the page. 1587110d74a9SMichel Lespinasse */ 1588110d74a9SMichel Lespinasse if (page->mapping && trylock_page(page)) { 1589110d74a9SMichel Lespinasse lru_add_drain(); /* push cached pages to LRU */ 1590110d74a9SMichel Lespinasse /* 1591e6c509f8SHugh Dickins * Because we lock page here, and migration is 1592e6c509f8SHugh Dickins * blocked by the pte's page reference, and we 1593e6c509f8SHugh Dickins * know the page is still mapped, we don't even 1594e6c509f8SHugh Dickins * need to check for file-cache page truncation. 1595110d74a9SMichel Lespinasse */ 1596110d74a9SMichel Lespinasse mlock_vma_page(page); 1597110d74a9SMichel Lespinasse unlock_page(page); 1598110d74a9SMichel Lespinasse } 1599110d74a9SMichel Lespinasse } 1600deceb6cdSHugh Dickins unlock: 1601deceb6cdSHugh Dickins pte_unmap_unlock(ptep, ptl); 16021da177e4SLinus Torvalds out: 1603deceb6cdSHugh Dickins return page; 1604deceb6cdSHugh Dickins 160589f5b7daSLinus Torvalds bad_page: 160689f5b7daSLinus Torvalds pte_unmap_unlock(ptep, ptl); 160789f5b7daSLinus Torvalds return ERR_PTR(-EFAULT); 160889f5b7daSLinus Torvalds 160989f5b7daSLinus Torvalds no_page: 161089f5b7daSLinus Torvalds pte_unmap_unlock(ptep, ptl); 161189f5b7daSLinus Torvalds if (!pte_none(pte)) 161289f5b7daSLinus Torvalds return page; 16138e4b9a60SHugh Dickins 1614deceb6cdSHugh Dickins no_page_table: 1615deceb6cdSHugh Dickins /* 1616deceb6cdSHugh Dickins * When core dumping an enormous anonymous area that nobody 16178e4b9a60SHugh Dickins * has touched so far, we don't want to allocate unnecessary pages or 16188e4b9a60SHugh Dickins * page tables. Return error instead of NULL to skip handle_mm_fault, 16198e4b9a60SHugh Dickins * then get_dump_page() will return NULL to leave a hole in the dump. 16208e4b9a60SHugh Dickins * But we can only make this optimization where a hole would surely 16218e4b9a60SHugh Dickins * be zero-filled if handle_mm_fault() actually did handle it. 1622deceb6cdSHugh Dickins */ 16238e4b9a60SHugh Dickins if ((flags & FOLL_DUMP) && 16248e4b9a60SHugh Dickins (!vma->vm_ops || !vma->vm_ops->fault)) 16258e4b9a60SHugh Dickins return ERR_PTR(-EFAULT); 1626deceb6cdSHugh Dickins return page; 16271da177e4SLinus Torvalds } 16281da177e4SLinus Torvalds 162995042f9eSLinus Torvalds static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr) 163095042f9eSLinus Torvalds { 1631a09a79f6SMikulas Patocka return stack_guard_page_start(vma, addr) || 1632a09a79f6SMikulas Patocka stack_guard_page_end(vma, addr+PAGE_SIZE); 163395042f9eSLinus Torvalds } 163495042f9eSLinus Torvalds 16350014bd99SHuang Ying /** 16360014bd99SHuang Ying * __get_user_pages() - pin user pages in memory 16370014bd99SHuang Ying * @tsk: task_struct of target task 16380014bd99SHuang Ying * @mm: mm_struct of target mm 16390014bd99SHuang Ying * @start: starting user address 16400014bd99SHuang Ying * @nr_pages: number of pages from start to pin 16410014bd99SHuang Ying * @gup_flags: flags modifying pin behaviour 16420014bd99SHuang Ying * @pages: array that receives pointers to the pages pinned. 16430014bd99SHuang Ying * Should be at least nr_pages long. Or NULL, if caller 16440014bd99SHuang Ying * only intends to ensure the pages are faulted in. 16450014bd99SHuang Ying * @vmas: array of pointers to vmas corresponding to each page. 16460014bd99SHuang Ying * Or NULL if the caller does not require them. 16470014bd99SHuang Ying * @nonblocking: whether waiting for disk IO or mmap_sem contention 16480014bd99SHuang Ying * 16490014bd99SHuang Ying * Returns number of pages pinned. This may be fewer than the number 16500014bd99SHuang Ying * requested. If nr_pages is 0 or negative, returns 0. If no pages 16510014bd99SHuang Ying * were pinned, returns -errno. Each page returned must be released 16520014bd99SHuang Ying * with a put_page() call when it is finished with. vmas will only 16530014bd99SHuang Ying * remain valid while mmap_sem is held. 16540014bd99SHuang Ying * 16550014bd99SHuang Ying * Must be called with mmap_sem held for read or write. 16560014bd99SHuang Ying * 16570014bd99SHuang Ying * __get_user_pages walks a process's page tables and takes a reference to 16580014bd99SHuang Ying * each struct page that each user address corresponds to at a given 16590014bd99SHuang Ying * instant. That is, it takes the page that would be accessed if a user 16600014bd99SHuang Ying * thread accesses the given user virtual address at that instant. 16610014bd99SHuang Ying * 16620014bd99SHuang Ying * This does not guarantee that the page exists in the user mappings when 16630014bd99SHuang Ying * __get_user_pages returns, and there may even be a completely different 16640014bd99SHuang Ying * page there in some cases (eg. if mmapped pagecache has been invalidated 16650014bd99SHuang Ying * and subsequently re faulted). However it does guarantee that the page 16660014bd99SHuang Ying * won't be freed completely. And mostly callers simply care that the page 16670014bd99SHuang Ying * contains data that was valid *at some point in time*. Typically, an IO 16680014bd99SHuang Ying * or similar operation cannot guarantee anything stronger anyway because 16690014bd99SHuang Ying * locks can't be held over the syscall boundary. 16700014bd99SHuang Ying * 16710014bd99SHuang Ying * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If 16720014bd99SHuang Ying * the page is written to, set_page_dirty (or set_page_dirty_lock, as 16730014bd99SHuang Ying * appropriate) must be called after the page is finished with, and 16740014bd99SHuang Ying * before put_page is called. 16750014bd99SHuang Ying * 16760014bd99SHuang Ying * If @nonblocking != NULL, __get_user_pages will not wait for disk IO 16770014bd99SHuang Ying * or mmap_sem contention, and if waiting is needed to pin all pages, 16780014bd99SHuang Ying * *@nonblocking will be set to 0. 16790014bd99SHuang Ying * 16800014bd99SHuang Ying * In most cases, get_user_pages or get_user_pages_fast should be used 16810014bd99SHuang Ying * instead of __get_user_pages. __get_user_pages should be used only if 16820014bd99SHuang Ying * you need some special @gup_flags. 16830014bd99SHuang Ying */ 1684b291f000SNick Piggin int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 168558fa879eSHugh Dickins unsigned long start, int nr_pages, unsigned int gup_flags, 168653a7706dSMichel Lespinasse struct page **pages, struct vm_area_struct **vmas, 168753a7706dSMichel Lespinasse int *nonblocking) 16881da177e4SLinus Torvalds { 16891da177e4SLinus Torvalds int i; 169058fa879eSHugh Dickins unsigned long vm_flags; 16911da177e4SLinus Torvalds 16929d73777eSPeter Zijlstra if (nr_pages <= 0) 1693900cf086SJonathan Corbet return 0; 169458fa879eSHugh Dickins 169558fa879eSHugh Dickins VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET)); 169658fa879eSHugh Dickins 16971da177e4SLinus Torvalds /* 16981da177e4SLinus Torvalds * Require read or write permissions. 169958fa879eSHugh Dickins * If FOLL_FORCE is set, we only require the "MAY" flags. 17001da177e4SLinus Torvalds */ 170158fa879eSHugh Dickins vm_flags = (gup_flags & FOLL_WRITE) ? 170258fa879eSHugh Dickins (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 170358fa879eSHugh Dickins vm_flags &= (gup_flags & FOLL_FORCE) ? 170458fa879eSHugh Dickins (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 17050b9d7052SAndrea Arcangeli 17060b9d7052SAndrea Arcangeli /* 17070b9d7052SAndrea Arcangeli * If FOLL_FORCE and FOLL_NUMA are both set, handle_mm_fault 17080b9d7052SAndrea Arcangeli * would be called on PROT_NONE ranges. We must never invoke 17090b9d7052SAndrea Arcangeli * handle_mm_fault on PROT_NONE ranges or the NUMA hinting 17100b9d7052SAndrea Arcangeli * page faults would unprotect the PROT_NONE ranges if 17110b9d7052SAndrea Arcangeli * _PAGE_NUMA and _PAGE_PROTNONE are sharing the same pte/pmd 17120b9d7052SAndrea Arcangeli * bitflag. So to avoid that, don't set FOLL_NUMA if 17130b9d7052SAndrea Arcangeli * FOLL_FORCE is set. 17140b9d7052SAndrea Arcangeli */ 17150b9d7052SAndrea Arcangeli if (!(gup_flags & FOLL_FORCE)) 17160b9d7052SAndrea Arcangeli gup_flags |= FOLL_NUMA; 17170b9d7052SAndrea Arcangeli 17181da177e4SLinus Torvalds i = 0; 17191da177e4SLinus Torvalds 17201da177e4SLinus Torvalds do { 17211da177e4SLinus Torvalds struct vm_area_struct *vma; 17221da177e4SLinus Torvalds 17231da177e4SLinus Torvalds vma = find_extend_vma(mm, start); 1724e7f22e20SStephen Wilson if (!vma && in_gate_area(mm, start)) { 17251da177e4SLinus Torvalds unsigned long pg = start & PAGE_MASK; 17261da177e4SLinus Torvalds pgd_t *pgd; 17271da177e4SLinus Torvalds pud_t *pud; 17281da177e4SLinus Torvalds pmd_t *pmd; 17291da177e4SLinus Torvalds pte_t *pte; 1730b291f000SNick Piggin 1731b291f000SNick Piggin /* user gate pages are read-only */ 173258fa879eSHugh Dickins if (gup_flags & FOLL_WRITE) 17331da177e4SLinus Torvalds return i ? : -EFAULT; 17341da177e4SLinus Torvalds if (pg > TASK_SIZE) 17351da177e4SLinus Torvalds pgd = pgd_offset_k(pg); 17361da177e4SLinus Torvalds else 17371da177e4SLinus Torvalds pgd = pgd_offset_gate(mm, pg); 17381da177e4SLinus Torvalds BUG_ON(pgd_none(*pgd)); 17391da177e4SLinus Torvalds pud = pud_offset(pgd, pg); 17401da177e4SLinus Torvalds BUG_ON(pud_none(*pud)); 17411da177e4SLinus Torvalds pmd = pmd_offset(pud, pg); 1742690dbe1cSHugh Dickins if (pmd_none(*pmd)) 1743690dbe1cSHugh Dickins return i ? : -EFAULT; 1744f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 17451da177e4SLinus Torvalds pte = pte_offset_map(pmd, pg); 1746690dbe1cSHugh Dickins if (pte_none(*pte)) { 1747690dbe1cSHugh Dickins pte_unmap(pte); 1748690dbe1cSHugh Dickins return i ? : -EFAULT; 1749690dbe1cSHugh Dickins } 175095042f9eSLinus Torvalds vma = get_gate_vma(mm); 17511da177e4SLinus Torvalds if (pages) { 1752de51257aSHugh Dickins struct page *page; 1753de51257aSHugh Dickins 175495042f9eSLinus Torvalds page = vm_normal_page(vma, start, *pte); 1755de51257aSHugh Dickins if (!page) { 1756de51257aSHugh Dickins if (!(gup_flags & FOLL_DUMP) && 1757de51257aSHugh Dickins is_zero_pfn(pte_pfn(*pte))) 1758de51257aSHugh Dickins page = pte_page(*pte); 1759de51257aSHugh Dickins else { 1760de51257aSHugh Dickins pte_unmap(pte); 1761de51257aSHugh Dickins return i ? : -EFAULT; 1762de51257aSHugh Dickins } 1763de51257aSHugh Dickins } 17646aab341eSLinus Torvalds pages[i] = page; 17656aab341eSLinus Torvalds get_page(page); 17661da177e4SLinus Torvalds } 17671da177e4SLinus Torvalds pte_unmap(pte); 176895042f9eSLinus Torvalds goto next_page; 17691da177e4SLinus Torvalds } 17701da177e4SLinus Torvalds 1771b291f000SNick Piggin if (!vma || 1772b291f000SNick Piggin (vma->vm_flags & (VM_IO | VM_PFNMAP)) || 17731c3aff1cSHugh Dickins !(vm_flags & vma->vm_flags)) 17741da177e4SLinus Torvalds return i ? : -EFAULT; 17751da177e4SLinus Torvalds 17762a15efc9SHugh Dickins if (is_vm_hugetlb_page(vma)) { 17772a15efc9SHugh Dickins i = follow_hugetlb_page(mm, vma, pages, vmas, 177858fa879eSHugh Dickins &start, &nr_pages, i, gup_flags); 17792a15efc9SHugh Dickins continue; 17802a15efc9SHugh Dickins } 1781deceb6cdSHugh Dickins 17821da177e4SLinus Torvalds do { 178308ef4729SHugh Dickins struct page *page; 178458fa879eSHugh Dickins unsigned int foll_flags = gup_flags; 17851da177e4SLinus Torvalds 1786462e00ccSEthan Solomita /* 17874779280dSYing Han * If we have a pending SIGKILL, don't keep faulting 17881c3aff1cSHugh Dickins * pages and potentially allocating memory. 1789462e00ccSEthan Solomita */ 17901c3aff1cSHugh Dickins if (unlikely(fatal_signal_pending(current))) 17914779280dSYing Han return i ? i : -ERESTARTSYS; 1792462e00ccSEthan Solomita 1793deceb6cdSHugh Dickins cond_resched(); 17946aab341eSLinus Torvalds while (!(page = follow_page(vma, start, foll_flags))) { 1795a68d2ebcSLinus Torvalds int ret; 179653a7706dSMichel Lespinasse unsigned int fault_flags = 0; 179753a7706dSMichel Lespinasse 1798a09a79f6SMikulas Patocka /* For mlock, just skip the stack guard page. */ 1799a09a79f6SMikulas Patocka if (foll_flags & FOLL_MLOCK) { 1800a09a79f6SMikulas Patocka if (stack_guard_page(vma, start)) 1801a09a79f6SMikulas Patocka goto next_page; 1802a09a79f6SMikulas Patocka } 180353a7706dSMichel Lespinasse if (foll_flags & FOLL_WRITE) 180453a7706dSMichel Lespinasse fault_flags |= FAULT_FLAG_WRITE; 180553a7706dSMichel Lespinasse if (nonblocking) 180653a7706dSMichel Lespinasse fault_flags |= FAULT_FLAG_ALLOW_RETRY; 1807318b275fSGleb Natapov if (foll_flags & FOLL_NOWAIT) 1808318b275fSGleb Natapov fault_flags |= (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT); 1809d06063ccSLinus Torvalds 1810d26ed650SHugh Dickins ret = handle_mm_fault(mm, vma, start, 181153a7706dSMichel Lespinasse fault_flags); 1812d26ed650SHugh Dickins 181383c54070SNick Piggin if (ret & VM_FAULT_ERROR) { 181483c54070SNick Piggin if (ret & VM_FAULT_OOM) 181583c54070SNick Piggin return i ? i : -ENOMEM; 181669ebb83eSHuang Ying if (ret & (VM_FAULT_HWPOISON | 181769ebb83eSHuang Ying VM_FAULT_HWPOISON_LARGE)) { 181869ebb83eSHuang Ying if (i) 181969ebb83eSHuang Ying return i; 182069ebb83eSHuang Ying else if (gup_flags & FOLL_HWPOISON) 182169ebb83eSHuang Ying return -EHWPOISON; 182269ebb83eSHuang Ying else 182369ebb83eSHuang Ying return -EFAULT; 182469ebb83eSHuang Ying } 182569ebb83eSHuang Ying if (ret & VM_FAULT_SIGBUS) 182683c54070SNick Piggin return i ? i : -EFAULT; 182783c54070SNick Piggin BUG(); 182883c54070SNick Piggin } 1829e7f22e20SStephen Wilson 1830e7f22e20SStephen Wilson if (tsk) { 183183c54070SNick Piggin if (ret & VM_FAULT_MAJOR) 183283c54070SNick Piggin tsk->maj_flt++; 183383c54070SNick Piggin else 183483c54070SNick Piggin tsk->min_flt++; 1835e7f22e20SStephen Wilson } 183683c54070SNick Piggin 183753a7706dSMichel Lespinasse if (ret & VM_FAULT_RETRY) { 1838318b275fSGleb Natapov if (nonblocking) 183953a7706dSMichel Lespinasse *nonblocking = 0; 184053a7706dSMichel Lespinasse return i; 184153a7706dSMichel Lespinasse } 184253a7706dSMichel Lespinasse 1843f33ea7f4SNick Piggin /* 184483c54070SNick Piggin * The VM_FAULT_WRITE bit tells us that 184583c54070SNick Piggin * do_wp_page has broken COW when necessary, 184683c54070SNick Piggin * even if maybe_mkwrite decided not to set 184783c54070SNick Piggin * pte_write. We can thus safely do subsequent 1848878b63acSHugh Dickins * page lookups as if they were reads. But only 1849878b63acSHugh Dickins * do so when looping for pte_write is futile: 1850878b63acSHugh Dickins * in some cases userspace may also be wanting 1851878b63acSHugh Dickins * to write to the gotten user page, which a 1852878b63acSHugh Dickins * read fault here might prevent (a readonly 1853878b63acSHugh Dickins * page might get reCOWed by userspace write). 1854f33ea7f4SNick Piggin */ 1855878b63acSHugh Dickins if ((ret & VM_FAULT_WRITE) && 1856878b63acSHugh Dickins !(vma->vm_flags & VM_WRITE)) 1857deceb6cdSHugh Dickins foll_flags &= ~FOLL_WRITE; 1858a68d2ebcSLinus Torvalds 18597f7bbbe5SBenjamin Herrenschmidt cond_resched(); 18601da177e4SLinus Torvalds } 186189f5b7daSLinus Torvalds if (IS_ERR(page)) 186289f5b7daSLinus Torvalds return i ? i : PTR_ERR(page); 18631da177e4SLinus Torvalds if (pages) { 186408ef4729SHugh Dickins pages[i] = page; 186503beb076SJames Bottomley 1866a6f36be3SRussell King flush_anon_page(vma, page, start); 186708ef4729SHugh Dickins flush_dcache_page(page); 18681da177e4SLinus Torvalds } 186995042f9eSLinus Torvalds next_page: 18701da177e4SLinus Torvalds if (vmas) 18711da177e4SLinus Torvalds vmas[i] = vma; 18721da177e4SLinus Torvalds i++; 18731da177e4SLinus Torvalds start += PAGE_SIZE; 18749d73777eSPeter Zijlstra nr_pages--; 18759d73777eSPeter Zijlstra } while (nr_pages && start < vma->vm_end); 18769d73777eSPeter Zijlstra } while (nr_pages); 18771da177e4SLinus Torvalds return i; 18781da177e4SLinus Torvalds } 18790014bd99SHuang Ying EXPORT_SYMBOL(__get_user_pages); 1880b291f000SNick Piggin 18812efaca92SBenjamin Herrenschmidt /* 18822efaca92SBenjamin Herrenschmidt * fixup_user_fault() - manually resolve a user page fault 18832efaca92SBenjamin Herrenschmidt * @tsk: the task_struct to use for page fault accounting, or 18842efaca92SBenjamin Herrenschmidt * NULL if faults are not to be recorded. 18852efaca92SBenjamin Herrenschmidt * @mm: mm_struct of target mm 18862efaca92SBenjamin Herrenschmidt * @address: user address 18872efaca92SBenjamin Herrenschmidt * @fault_flags:flags to pass down to handle_mm_fault() 18882efaca92SBenjamin Herrenschmidt * 18892efaca92SBenjamin Herrenschmidt * This is meant to be called in the specific scenario where for locking reasons 18902efaca92SBenjamin Herrenschmidt * we try to access user memory in atomic context (within a pagefault_disable() 18912efaca92SBenjamin Herrenschmidt * section), this returns -EFAULT, and we want to resolve the user fault before 18922efaca92SBenjamin Herrenschmidt * trying again. 18932efaca92SBenjamin Herrenschmidt * 18942efaca92SBenjamin Herrenschmidt * Typically this is meant to be used by the futex code. 18952efaca92SBenjamin Herrenschmidt * 18962efaca92SBenjamin Herrenschmidt * The main difference with get_user_pages() is that this function will 18972efaca92SBenjamin Herrenschmidt * unconditionally call handle_mm_fault() which will in turn perform all the 18982efaca92SBenjamin Herrenschmidt * necessary SW fixup of the dirty and young bits in the PTE, while 18992efaca92SBenjamin Herrenschmidt * handle_mm_fault() only guarantees to update these in the struct page. 19002efaca92SBenjamin Herrenschmidt * 19012efaca92SBenjamin Herrenschmidt * This is important for some architectures where those bits also gate the 19022efaca92SBenjamin Herrenschmidt * access permission to the page because they are maintained in software. On 19032efaca92SBenjamin Herrenschmidt * such architectures, gup() will not be enough to make a subsequent access 19042efaca92SBenjamin Herrenschmidt * succeed. 19052efaca92SBenjamin Herrenschmidt * 19062efaca92SBenjamin Herrenschmidt * This should be called with the mm_sem held for read. 19072efaca92SBenjamin Herrenschmidt */ 19082efaca92SBenjamin Herrenschmidt int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm, 19092efaca92SBenjamin Herrenschmidt unsigned long address, unsigned int fault_flags) 19102efaca92SBenjamin Herrenschmidt { 19112efaca92SBenjamin Herrenschmidt struct vm_area_struct *vma; 19122efaca92SBenjamin Herrenschmidt int ret; 19132efaca92SBenjamin Herrenschmidt 19142efaca92SBenjamin Herrenschmidt vma = find_extend_vma(mm, address); 19152efaca92SBenjamin Herrenschmidt if (!vma || address < vma->vm_start) 19162efaca92SBenjamin Herrenschmidt return -EFAULT; 19172efaca92SBenjamin Herrenschmidt 19182efaca92SBenjamin Herrenschmidt ret = handle_mm_fault(mm, vma, address, fault_flags); 19192efaca92SBenjamin Herrenschmidt if (ret & VM_FAULT_ERROR) { 19202efaca92SBenjamin Herrenschmidt if (ret & VM_FAULT_OOM) 19212efaca92SBenjamin Herrenschmidt return -ENOMEM; 19222efaca92SBenjamin Herrenschmidt if (ret & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE)) 19232efaca92SBenjamin Herrenschmidt return -EHWPOISON; 19242efaca92SBenjamin Herrenschmidt if (ret & VM_FAULT_SIGBUS) 19252efaca92SBenjamin Herrenschmidt return -EFAULT; 19262efaca92SBenjamin Herrenschmidt BUG(); 19272efaca92SBenjamin Herrenschmidt } 19282efaca92SBenjamin Herrenschmidt if (tsk) { 19292efaca92SBenjamin Herrenschmidt if (ret & VM_FAULT_MAJOR) 19302efaca92SBenjamin Herrenschmidt tsk->maj_flt++; 19312efaca92SBenjamin Herrenschmidt else 19322efaca92SBenjamin Herrenschmidt tsk->min_flt++; 19332efaca92SBenjamin Herrenschmidt } 19342efaca92SBenjamin Herrenschmidt return 0; 19352efaca92SBenjamin Herrenschmidt } 19362efaca92SBenjamin Herrenschmidt 19372efaca92SBenjamin Herrenschmidt /* 1938d2bf6be8SNick Piggin * get_user_pages() - pin user pages in memory 1939e7f22e20SStephen Wilson * @tsk: the task_struct to use for page fault accounting, or 1940e7f22e20SStephen Wilson * NULL if faults are not to be recorded. 1941d2bf6be8SNick Piggin * @mm: mm_struct of target mm 1942d2bf6be8SNick Piggin * @start: starting user address 19439d73777eSPeter Zijlstra * @nr_pages: number of pages from start to pin 1944d2bf6be8SNick Piggin * @write: whether pages will be written to by the caller 1945d2bf6be8SNick Piggin * @force: whether to force write access even if user mapping is 1946d2bf6be8SNick Piggin * readonly. This will result in the page being COWed even 1947d2bf6be8SNick Piggin * in MAP_SHARED mappings. You do not want this. 1948d2bf6be8SNick Piggin * @pages: array that receives pointers to the pages pinned. 1949d2bf6be8SNick Piggin * Should be at least nr_pages long. Or NULL, if caller 1950d2bf6be8SNick Piggin * only intends to ensure the pages are faulted in. 1951d2bf6be8SNick Piggin * @vmas: array of pointers to vmas corresponding to each page. 1952d2bf6be8SNick Piggin * Or NULL if the caller does not require them. 1953d2bf6be8SNick Piggin * 1954d2bf6be8SNick Piggin * Returns number of pages pinned. This may be fewer than the number 19559d73777eSPeter Zijlstra * requested. If nr_pages is 0 or negative, returns 0. If no pages 1956d2bf6be8SNick Piggin * were pinned, returns -errno. Each page returned must be released 1957d2bf6be8SNick Piggin * with a put_page() call when it is finished with. vmas will only 1958d2bf6be8SNick Piggin * remain valid while mmap_sem is held. 1959d2bf6be8SNick Piggin * 1960d2bf6be8SNick Piggin * Must be called with mmap_sem held for read or write. 1961d2bf6be8SNick Piggin * 1962d2bf6be8SNick Piggin * get_user_pages walks a process's page tables and takes a reference to 1963d2bf6be8SNick Piggin * each struct page that each user address corresponds to at a given 1964d2bf6be8SNick Piggin * instant. That is, it takes the page that would be accessed if a user 1965d2bf6be8SNick Piggin * thread accesses the given user virtual address at that instant. 1966d2bf6be8SNick Piggin * 1967d2bf6be8SNick Piggin * This does not guarantee that the page exists in the user mappings when 1968d2bf6be8SNick Piggin * get_user_pages returns, and there may even be a completely different 1969d2bf6be8SNick Piggin * page there in some cases (eg. if mmapped pagecache has been invalidated 1970d2bf6be8SNick Piggin * and subsequently re faulted). However it does guarantee that the page 1971d2bf6be8SNick Piggin * won't be freed completely. And mostly callers simply care that the page 1972d2bf6be8SNick Piggin * contains data that was valid *at some point in time*. Typically, an IO 1973d2bf6be8SNick Piggin * or similar operation cannot guarantee anything stronger anyway because 1974d2bf6be8SNick Piggin * locks can't be held over the syscall boundary. 1975d2bf6be8SNick Piggin * 1976d2bf6be8SNick Piggin * If write=0, the page must not be written to. If the page is written to, 1977d2bf6be8SNick Piggin * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called 1978d2bf6be8SNick Piggin * after the page is finished with, and before put_page is called. 1979d2bf6be8SNick Piggin * 1980d2bf6be8SNick Piggin * get_user_pages is typically used for fewer-copy IO operations, to get a 1981d2bf6be8SNick Piggin * handle on the memory by some means other than accesses via the user virtual 1982d2bf6be8SNick Piggin * addresses. The pages may be submitted for DMA to devices or accessed via 1983d2bf6be8SNick Piggin * their kernel linear mapping (via the kmap APIs). Care should be taken to 1984d2bf6be8SNick Piggin * use the correct cache flushing APIs. 1985d2bf6be8SNick Piggin * 1986d2bf6be8SNick Piggin * See also get_user_pages_fast, for performance critical applications. 1987d2bf6be8SNick Piggin */ 1988b291f000SNick Piggin int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 19899d73777eSPeter Zijlstra unsigned long start, int nr_pages, int write, int force, 1990b291f000SNick Piggin struct page **pages, struct vm_area_struct **vmas) 1991b291f000SNick Piggin { 199258fa879eSHugh Dickins int flags = FOLL_TOUCH; 1993b291f000SNick Piggin 199458fa879eSHugh Dickins if (pages) 199558fa879eSHugh Dickins flags |= FOLL_GET; 1996b291f000SNick Piggin if (write) 199758fa879eSHugh Dickins flags |= FOLL_WRITE; 1998b291f000SNick Piggin if (force) 199958fa879eSHugh Dickins flags |= FOLL_FORCE; 2000b291f000SNick Piggin 200153a7706dSMichel Lespinasse return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, 200253a7706dSMichel Lespinasse NULL); 2003b291f000SNick Piggin } 20041da177e4SLinus Torvalds EXPORT_SYMBOL(get_user_pages); 20051da177e4SLinus Torvalds 2006f3e8fccdSHugh Dickins /** 2007f3e8fccdSHugh Dickins * get_dump_page() - pin user page in memory while writing it to core dump 2008f3e8fccdSHugh Dickins * @addr: user address 2009f3e8fccdSHugh Dickins * 2010f3e8fccdSHugh Dickins * Returns struct page pointer of user page pinned for dump, 2011f3e8fccdSHugh Dickins * to be freed afterwards by page_cache_release() or put_page(). 2012f3e8fccdSHugh Dickins * 2013f3e8fccdSHugh Dickins * Returns NULL on any kind of failure - a hole must then be inserted into 2014f3e8fccdSHugh Dickins * the corefile, to preserve alignment with its headers; and also returns 2015f3e8fccdSHugh Dickins * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - 2016f3e8fccdSHugh Dickins * allowing a hole to be left in the corefile to save diskspace. 2017f3e8fccdSHugh Dickins * 2018f3e8fccdSHugh Dickins * Called without mmap_sem, but after all other threads have been killed. 2019f3e8fccdSHugh Dickins */ 2020f3e8fccdSHugh Dickins #ifdef CONFIG_ELF_CORE 2021f3e8fccdSHugh Dickins struct page *get_dump_page(unsigned long addr) 2022f3e8fccdSHugh Dickins { 2023f3e8fccdSHugh Dickins struct vm_area_struct *vma; 2024f3e8fccdSHugh Dickins struct page *page; 2025f3e8fccdSHugh Dickins 2026f3e8fccdSHugh Dickins if (__get_user_pages(current, current->mm, addr, 1, 202753a7706dSMichel Lespinasse FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma, 202853a7706dSMichel Lespinasse NULL) < 1) 2029f3e8fccdSHugh Dickins return NULL; 2030f3e8fccdSHugh Dickins flush_cache_page(vma, addr, page_to_pfn(page)); 2031f3e8fccdSHugh Dickins return page; 2032f3e8fccdSHugh Dickins } 2033f3e8fccdSHugh Dickins #endif /* CONFIG_ELF_CORE */ 2034f3e8fccdSHugh Dickins 203525ca1d6cSNamhyung Kim pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 2036920c7a5dSHarvey Harrison spinlock_t **ptl) 2037c9cfcddfSLinus Torvalds { 2038c9cfcddfSLinus Torvalds pgd_t * pgd = pgd_offset(mm, addr); 2039c9cfcddfSLinus Torvalds pud_t * pud = pud_alloc(mm, pgd, addr); 2040c9cfcddfSLinus Torvalds if (pud) { 204149c91fb0STrond Myklebust pmd_t * pmd = pmd_alloc(mm, pud, addr); 2042f66055abSAndrea Arcangeli if (pmd) { 2043f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 2044c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 2045c9cfcddfSLinus Torvalds } 2046f66055abSAndrea Arcangeli } 2047c9cfcddfSLinus Torvalds return NULL; 2048c9cfcddfSLinus Torvalds } 2049c9cfcddfSLinus Torvalds 20501da177e4SLinus Torvalds /* 2051238f58d8SLinus Torvalds * This is the old fallback for page remapping. 2052238f58d8SLinus Torvalds * 2053238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 2054238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 2055238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 2056238f58d8SLinus Torvalds */ 2057423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 2058423bad60SNick Piggin struct page *page, pgprot_t prot) 2059238f58d8SLinus Torvalds { 2060423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 2061238f58d8SLinus Torvalds int retval; 2062238f58d8SLinus Torvalds pte_t *pte; 2063238f58d8SLinus Torvalds spinlock_t *ptl; 2064238f58d8SLinus Torvalds 2065238f58d8SLinus Torvalds retval = -EINVAL; 2066a145dd41SLinus Torvalds if (PageAnon(page)) 20675b4e655eSKAMEZAWA Hiroyuki goto out; 2068238f58d8SLinus Torvalds retval = -ENOMEM; 2069238f58d8SLinus Torvalds flush_dcache_page(page); 2070c9cfcddfSLinus Torvalds pte = get_locked_pte(mm, addr, &ptl); 2071238f58d8SLinus Torvalds if (!pte) 20725b4e655eSKAMEZAWA Hiroyuki goto out; 2073238f58d8SLinus Torvalds retval = -EBUSY; 2074238f58d8SLinus Torvalds if (!pte_none(*pte)) 2075238f58d8SLinus Torvalds goto out_unlock; 2076238f58d8SLinus Torvalds 2077238f58d8SLinus Torvalds /* Ok, finally just insert the thing.. */ 2078238f58d8SLinus Torvalds get_page(page); 207934e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_FILEPAGES); 2080238f58d8SLinus Torvalds page_add_file_rmap(page); 2081238f58d8SLinus Torvalds set_pte_at(mm, addr, pte, mk_pte(page, prot)); 2082238f58d8SLinus Torvalds 2083238f58d8SLinus Torvalds retval = 0; 20848a9f3ccdSBalbir Singh pte_unmap_unlock(pte, ptl); 20858a9f3ccdSBalbir Singh return retval; 2086238f58d8SLinus Torvalds out_unlock: 2087238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 2088238f58d8SLinus Torvalds out: 2089238f58d8SLinus Torvalds return retval; 2090238f58d8SLinus Torvalds } 2091238f58d8SLinus Torvalds 2092bfa5bf6dSRolf Eike Beer /** 2093bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 2094bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 2095bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 2096bfa5bf6dSRolf Eike Beer * @page: source kernel page 2097bfa5bf6dSRolf Eike Beer * 2098a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 2099a145dd41SLinus Torvalds * into a user vma. 2100a145dd41SLinus Torvalds * 2101a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 2102a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 2103a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 21048dfcc9baSNick Piggin * (see split_page()). 2105a145dd41SLinus Torvalds * 2106a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 2107a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 2108a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 2109a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 2110a145dd41SLinus Torvalds * ask for a shared writable mapping! 2111a145dd41SLinus Torvalds * 2112a145dd41SLinus Torvalds * The page does not need to be reserved. 21134b6e1e37SKonstantin Khlebnikov * 21144b6e1e37SKonstantin Khlebnikov * Usually this function is called from f_op->mmap() handler 21154b6e1e37SKonstantin Khlebnikov * under mm->mmap_sem write-lock, so it can change vma->vm_flags. 21164b6e1e37SKonstantin Khlebnikov * Caller must set VM_MIXEDMAP on vma if it wants to call this 21174b6e1e37SKonstantin Khlebnikov * function from other places, for example from page-fault handler. 2118a145dd41SLinus Torvalds */ 2119423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 2120423bad60SNick Piggin struct page *page) 2121a145dd41SLinus Torvalds { 2122a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 2123a145dd41SLinus Torvalds return -EFAULT; 2124a145dd41SLinus Torvalds if (!page_count(page)) 2125a145dd41SLinus Torvalds return -EINVAL; 21264b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & VM_MIXEDMAP)) { 21274b6e1e37SKonstantin Khlebnikov BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); 21284b6e1e37SKonstantin Khlebnikov BUG_ON(vma->vm_flags & VM_PFNMAP); 21294b6e1e37SKonstantin Khlebnikov vma->vm_flags |= VM_MIXEDMAP; 21304b6e1e37SKonstantin Khlebnikov } 2131423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 2132a145dd41SLinus Torvalds } 2133e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 2134a145dd41SLinus Torvalds 2135423bad60SNick Piggin static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2136423bad60SNick Piggin unsigned long pfn, pgprot_t prot) 2137423bad60SNick Piggin { 2138423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 2139423bad60SNick Piggin int retval; 2140423bad60SNick Piggin pte_t *pte, entry; 2141423bad60SNick Piggin spinlock_t *ptl; 2142423bad60SNick Piggin 2143423bad60SNick Piggin retval = -ENOMEM; 2144423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 2145423bad60SNick Piggin if (!pte) 2146423bad60SNick Piggin goto out; 2147423bad60SNick Piggin retval = -EBUSY; 2148423bad60SNick Piggin if (!pte_none(*pte)) 2149423bad60SNick Piggin goto out_unlock; 2150423bad60SNick Piggin 2151423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 2152423bad60SNick Piggin entry = pte_mkspecial(pfn_pte(pfn, prot)); 2153423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 21544b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 2155423bad60SNick Piggin 2156423bad60SNick Piggin retval = 0; 2157423bad60SNick Piggin out_unlock: 2158423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 2159423bad60SNick Piggin out: 2160423bad60SNick Piggin return retval; 2161423bad60SNick Piggin } 2162423bad60SNick Piggin 2163e0dc0d8fSNick Piggin /** 2164e0dc0d8fSNick Piggin * vm_insert_pfn - insert single pfn into user vma 2165e0dc0d8fSNick Piggin * @vma: user vma to map to 2166e0dc0d8fSNick Piggin * @addr: target user address of this page 2167e0dc0d8fSNick Piggin * @pfn: source kernel pfn 2168e0dc0d8fSNick Piggin * 2169c462f179SRobert P. J. Day * Similar to vm_insert_page, this allows drivers to insert individual pages 2170e0dc0d8fSNick Piggin * they've allocated into a user vma. Same comments apply. 2171e0dc0d8fSNick Piggin * 2172e0dc0d8fSNick Piggin * This function should only be called from a vm_ops->fault handler, and 2173e0dc0d8fSNick Piggin * in that case the handler should return NULL. 21740d71d10aSNick Piggin * 21750d71d10aSNick Piggin * vma cannot be a COW mapping. 21760d71d10aSNick Piggin * 21770d71d10aSNick Piggin * As this is called only for pages that do not currently exist, we 21780d71d10aSNick Piggin * do not need to flush old virtual caches or the TLB. 2179e0dc0d8fSNick Piggin */ 2180e0dc0d8fSNick Piggin int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2181e0dc0d8fSNick Piggin unsigned long pfn) 2182e0dc0d8fSNick Piggin { 21832ab64037Svenkatesh.pallipadi@intel.com int ret; 2184e4b866edSvenkatesh.pallipadi@intel.com pgprot_t pgprot = vma->vm_page_prot; 21857e675137SNick Piggin /* 21867e675137SNick Piggin * Technically, architectures with pte_special can avoid all these 21877e675137SNick Piggin * restrictions (same for remap_pfn_range). However we would like 21887e675137SNick Piggin * consistency in testing and feature parity among all, so we should 21897e675137SNick Piggin * try to keep these invariants in place for everybody. 21907e675137SNick Piggin */ 2191b379d790SJared Hulbert BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 2192b379d790SJared Hulbert BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 2193b379d790SJared Hulbert (VM_PFNMAP|VM_MIXEDMAP)); 2194b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 2195b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 2196e0dc0d8fSNick Piggin 2197423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 2198423bad60SNick Piggin return -EFAULT; 21995180da41SSuresh Siddha if (track_pfn_insert(vma, &pgprot, pfn)) 22002ab64037Svenkatesh.pallipadi@intel.com return -EINVAL; 22012ab64037Svenkatesh.pallipadi@intel.com 2202e4b866edSvenkatesh.pallipadi@intel.com ret = insert_pfn(vma, addr, pfn, pgprot); 22032ab64037Svenkatesh.pallipadi@intel.com 22042ab64037Svenkatesh.pallipadi@intel.com return ret; 2205e0dc0d8fSNick Piggin } 2206e0dc0d8fSNick Piggin EXPORT_SYMBOL(vm_insert_pfn); 2207e0dc0d8fSNick Piggin 2208423bad60SNick Piggin int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 2209423bad60SNick Piggin unsigned long pfn) 2210423bad60SNick Piggin { 2211423bad60SNick Piggin BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); 2212423bad60SNick Piggin 2213423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 2214423bad60SNick Piggin return -EFAULT; 2215423bad60SNick Piggin 2216423bad60SNick Piggin /* 2217423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 2218423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 2219423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 222062eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 222162eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 2222423bad60SNick Piggin */ 2223423bad60SNick Piggin if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { 2224423bad60SNick Piggin struct page *page; 2225423bad60SNick Piggin 2226423bad60SNick Piggin page = pfn_to_page(pfn); 2227423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 2228423bad60SNick Piggin } 2229423bad60SNick Piggin return insert_pfn(vma, addr, pfn, vma->vm_page_prot); 2230423bad60SNick Piggin } 2231423bad60SNick Piggin EXPORT_SYMBOL(vm_insert_mixed); 2232423bad60SNick Piggin 2233a145dd41SLinus Torvalds /* 22341da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 22351da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 22361da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 22371da177e4SLinus Torvalds */ 22381da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 22391da177e4SLinus Torvalds unsigned long addr, unsigned long end, 22401da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 22411da177e4SLinus Torvalds { 22421da177e4SLinus Torvalds pte_t *pte; 2243c74df32cSHugh Dickins spinlock_t *ptl; 22441da177e4SLinus Torvalds 2245c74df32cSHugh Dickins pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 22461da177e4SLinus Torvalds if (!pte) 22471da177e4SLinus Torvalds return -ENOMEM; 22486606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 22491da177e4SLinus Torvalds do { 22501da177e4SLinus Torvalds BUG_ON(!pte_none(*pte)); 22517e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 22521da177e4SLinus Torvalds pfn++; 22531da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 22546606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 2255c74df32cSHugh Dickins pte_unmap_unlock(pte - 1, ptl); 22561da177e4SLinus Torvalds return 0; 22571da177e4SLinus Torvalds } 22581da177e4SLinus Torvalds 22591da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 22601da177e4SLinus Torvalds unsigned long addr, unsigned long end, 22611da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 22621da177e4SLinus Torvalds { 22631da177e4SLinus Torvalds pmd_t *pmd; 22641da177e4SLinus Torvalds unsigned long next; 22651da177e4SLinus Torvalds 22661da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 22671da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 22681da177e4SLinus Torvalds if (!pmd) 22691da177e4SLinus Torvalds return -ENOMEM; 2270f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 22711da177e4SLinus Torvalds do { 22721da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 22731da177e4SLinus Torvalds if (remap_pte_range(mm, pmd, addr, next, 22741da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 22751da177e4SLinus Torvalds return -ENOMEM; 22761da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 22771da177e4SLinus Torvalds return 0; 22781da177e4SLinus Torvalds } 22791da177e4SLinus Torvalds 22801da177e4SLinus Torvalds static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, 22811da177e4SLinus Torvalds unsigned long addr, unsigned long end, 22821da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 22831da177e4SLinus Torvalds { 22841da177e4SLinus Torvalds pud_t *pud; 22851da177e4SLinus Torvalds unsigned long next; 22861da177e4SLinus Torvalds 22871da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 22881da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, addr); 22891da177e4SLinus Torvalds if (!pud) 22901da177e4SLinus Torvalds return -ENOMEM; 22911da177e4SLinus Torvalds do { 22921da177e4SLinus Torvalds next = pud_addr_end(addr, end); 22931da177e4SLinus Torvalds if (remap_pmd_range(mm, pud, addr, next, 22941da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 22951da177e4SLinus Torvalds return -ENOMEM; 22961da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 22971da177e4SLinus Torvalds return 0; 22981da177e4SLinus Torvalds } 22991da177e4SLinus Torvalds 2300bfa5bf6dSRolf Eike Beer /** 2301bfa5bf6dSRolf Eike Beer * remap_pfn_range - remap kernel memory to userspace 2302bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 2303bfa5bf6dSRolf Eike Beer * @addr: target user address to start at 2304bfa5bf6dSRolf Eike Beer * @pfn: physical address of kernel memory 2305bfa5bf6dSRolf Eike Beer * @size: size of map area 2306bfa5bf6dSRolf Eike Beer * @prot: page protection flags for this mapping 2307bfa5bf6dSRolf Eike Beer * 2308bfa5bf6dSRolf Eike Beer * Note: this is only safe if the mm semaphore is held when called. 2309bfa5bf6dSRolf Eike Beer */ 23101da177e4SLinus Torvalds int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 23111da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 23121da177e4SLinus Torvalds { 23131da177e4SLinus Torvalds pgd_t *pgd; 23141da177e4SLinus Torvalds unsigned long next; 23152d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 23161da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 23171da177e4SLinus Torvalds int err; 23181da177e4SLinus Torvalds 23191da177e4SLinus Torvalds /* 23201da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 23211da177e4SLinus Torvalds * rest of the world about it: 23221da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 23231da177e4SLinus Torvalds * (accesses can have side effects). 23246aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 23256aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 23266aab341eSLinus Torvalds * with them. 2327314e51b9SKonstantin Khlebnikov * VM_DONTEXPAND 2328314e51b9SKonstantin Khlebnikov * Disable vma merging and expanding with mremap(). 2329314e51b9SKonstantin Khlebnikov * VM_DONTDUMP 2330314e51b9SKonstantin Khlebnikov * Omit vma from core dump, even when VM_IO turned off. 2331fb155c16SLinus Torvalds * 2332fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 2333fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 2334fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 2335b3b9c293SKonstantin Khlebnikov * See vm_normal_page() for details. 23361da177e4SLinus Torvalds */ 2337b3b9c293SKonstantin Khlebnikov if (is_cow_mapping(vma->vm_flags)) { 2338b3b9c293SKonstantin Khlebnikov if (addr != vma->vm_start || end != vma->vm_end) 2339b3b9c293SKonstantin Khlebnikov return -EINVAL; 23406aab341eSLinus Torvalds vma->vm_pgoff = pfn; 2341b3b9c293SKonstantin Khlebnikov } 2342b3b9c293SKonstantin Khlebnikov 2343b3b9c293SKonstantin Khlebnikov err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); 2344b3b9c293SKonstantin Khlebnikov if (err) 23453c8bb73aSvenkatesh.pallipadi@intel.com return -EINVAL; 2346fb155c16SLinus Torvalds 2347314e51b9SKonstantin Khlebnikov vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 23481da177e4SLinus Torvalds 23491da177e4SLinus Torvalds BUG_ON(addr >= end); 23501da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 23511da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 23521da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 23531da177e4SLinus Torvalds do { 23541da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 23551da177e4SLinus Torvalds err = remap_pud_range(mm, pgd, addr, next, 23561da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 23571da177e4SLinus Torvalds if (err) 23581da177e4SLinus Torvalds break; 23591da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 23602ab64037Svenkatesh.pallipadi@intel.com 23612ab64037Svenkatesh.pallipadi@intel.com if (err) 23625180da41SSuresh Siddha untrack_pfn(vma, pfn, PAGE_ALIGN(size)); 23632ab64037Svenkatesh.pallipadi@intel.com 23641da177e4SLinus Torvalds return err; 23651da177e4SLinus Torvalds } 23661da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 23671da177e4SLinus Torvalds 2368aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 2369aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2370aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 2371aee16b3cSJeremy Fitzhardinge { 2372aee16b3cSJeremy Fitzhardinge pte_t *pte; 2373aee16b3cSJeremy Fitzhardinge int err; 23742f569afdSMartin Schwidefsky pgtable_t token; 237594909914SBorislav Petkov spinlock_t *uninitialized_var(ptl); 2376aee16b3cSJeremy Fitzhardinge 2377aee16b3cSJeremy Fitzhardinge pte = (mm == &init_mm) ? 2378aee16b3cSJeremy Fitzhardinge pte_alloc_kernel(pmd, addr) : 2379aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 2380aee16b3cSJeremy Fitzhardinge if (!pte) 2381aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2382aee16b3cSJeremy Fitzhardinge 2383aee16b3cSJeremy Fitzhardinge BUG_ON(pmd_huge(*pmd)); 2384aee16b3cSJeremy Fitzhardinge 238538e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 238638e0edb1SJeremy Fitzhardinge 23872f569afdSMartin Schwidefsky token = pmd_pgtable(*pmd); 2388aee16b3cSJeremy Fitzhardinge 2389aee16b3cSJeremy Fitzhardinge do { 2390c36987e2SDaisuke Nishimura err = fn(pte++, token, addr, data); 2391aee16b3cSJeremy Fitzhardinge if (err) 2392aee16b3cSJeremy Fitzhardinge break; 2393c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 2394aee16b3cSJeremy Fitzhardinge 239538e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 239638e0edb1SJeremy Fitzhardinge 2397aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 2398aee16b3cSJeremy Fitzhardinge pte_unmap_unlock(pte-1, ptl); 2399aee16b3cSJeremy Fitzhardinge return err; 2400aee16b3cSJeremy Fitzhardinge } 2401aee16b3cSJeremy Fitzhardinge 2402aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 2403aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2404aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 2405aee16b3cSJeremy Fitzhardinge { 2406aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 2407aee16b3cSJeremy Fitzhardinge unsigned long next; 2408aee16b3cSJeremy Fitzhardinge int err; 2409aee16b3cSJeremy Fitzhardinge 2410ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 2411ceb86879SAndi Kleen 2412aee16b3cSJeremy Fitzhardinge pmd = pmd_alloc(mm, pud, addr); 2413aee16b3cSJeremy Fitzhardinge if (!pmd) 2414aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2415aee16b3cSJeremy Fitzhardinge do { 2416aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 2417aee16b3cSJeremy Fitzhardinge err = apply_to_pte_range(mm, pmd, addr, next, fn, data); 2418aee16b3cSJeremy Fitzhardinge if (err) 2419aee16b3cSJeremy Fitzhardinge break; 2420aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 2421aee16b3cSJeremy Fitzhardinge return err; 2422aee16b3cSJeremy Fitzhardinge } 2423aee16b3cSJeremy Fitzhardinge 2424aee16b3cSJeremy Fitzhardinge static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, 2425aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2426aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 2427aee16b3cSJeremy Fitzhardinge { 2428aee16b3cSJeremy Fitzhardinge pud_t *pud; 2429aee16b3cSJeremy Fitzhardinge unsigned long next; 2430aee16b3cSJeremy Fitzhardinge int err; 2431aee16b3cSJeremy Fitzhardinge 2432aee16b3cSJeremy Fitzhardinge pud = pud_alloc(mm, pgd, addr); 2433aee16b3cSJeremy Fitzhardinge if (!pud) 2434aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2435aee16b3cSJeremy Fitzhardinge do { 2436aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 2437aee16b3cSJeremy Fitzhardinge err = apply_to_pmd_range(mm, pud, addr, next, fn, data); 2438aee16b3cSJeremy Fitzhardinge if (err) 2439aee16b3cSJeremy Fitzhardinge break; 2440aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 2441aee16b3cSJeremy Fitzhardinge return err; 2442aee16b3cSJeremy Fitzhardinge } 2443aee16b3cSJeremy Fitzhardinge 2444aee16b3cSJeremy Fitzhardinge /* 2445aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 2446aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 2447aee16b3cSJeremy Fitzhardinge */ 2448aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2449aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 2450aee16b3cSJeremy Fitzhardinge { 2451aee16b3cSJeremy Fitzhardinge pgd_t *pgd; 2452aee16b3cSJeremy Fitzhardinge unsigned long next; 245357250a5bSJeremy Fitzhardinge unsigned long end = addr + size; 2454aee16b3cSJeremy Fitzhardinge int err; 2455aee16b3cSJeremy Fitzhardinge 2456aee16b3cSJeremy Fitzhardinge BUG_ON(addr >= end); 2457aee16b3cSJeremy Fitzhardinge pgd = pgd_offset(mm, addr); 2458aee16b3cSJeremy Fitzhardinge do { 2459aee16b3cSJeremy Fitzhardinge next = pgd_addr_end(addr, end); 2460aee16b3cSJeremy Fitzhardinge err = apply_to_pud_range(mm, pgd, addr, next, fn, data); 2461aee16b3cSJeremy Fitzhardinge if (err) 2462aee16b3cSJeremy Fitzhardinge break; 2463aee16b3cSJeremy Fitzhardinge } while (pgd++, addr = next, addr != end); 246457250a5bSJeremy Fitzhardinge 2465aee16b3cSJeremy Fitzhardinge return err; 2466aee16b3cSJeremy Fitzhardinge } 2467aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 2468aee16b3cSJeremy Fitzhardinge 24691da177e4SLinus Torvalds /* 24708f4e2101SHugh Dickins * handle_pte_fault chooses page fault handler according to an entry 24718f4e2101SHugh Dickins * which was read non-atomically. Before making any commitment, on 24728f4e2101SHugh Dickins * those architectures or configurations (e.g. i386 with PAE) which 2473a335b2e1SRyota Ozaki * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault 24748f4e2101SHugh Dickins * must check under lock before unmapping the pte and proceeding 24758f4e2101SHugh Dickins * (but do_wp_page is only called after already making such a check; 2476a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 24778f4e2101SHugh Dickins */ 24784c21e2f2SHugh Dickins static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, 24798f4e2101SHugh Dickins pte_t *page_table, pte_t orig_pte) 24808f4e2101SHugh Dickins { 24818f4e2101SHugh Dickins int same = 1; 24828f4e2101SHugh Dickins #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) 24838f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 24844c21e2f2SHugh Dickins spinlock_t *ptl = pte_lockptr(mm, pmd); 24854c21e2f2SHugh Dickins spin_lock(ptl); 24868f4e2101SHugh Dickins same = pte_same(*page_table, orig_pte); 24874c21e2f2SHugh Dickins spin_unlock(ptl); 24888f4e2101SHugh Dickins } 24898f4e2101SHugh Dickins #endif 24908f4e2101SHugh Dickins pte_unmap(page_table); 24918f4e2101SHugh Dickins return same; 24928f4e2101SHugh Dickins } 24938f4e2101SHugh Dickins 24949de455b2SAtsushi Nemoto static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) 24956aab341eSLinus Torvalds { 24966aab341eSLinus Torvalds /* 24976aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 24986aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 24996aab341eSLinus Torvalds * just copying from the original user address. If that 25006aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 25016aab341eSLinus Torvalds */ 25026aab341eSLinus Torvalds if (unlikely(!src)) { 25039b04c5feSCong Wang void *kaddr = kmap_atomic(dst); 25045d2a2dbbSLinus Torvalds void __user *uaddr = (void __user *)(va & PAGE_MASK); 25055d2a2dbbSLinus Torvalds 25065d2a2dbbSLinus Torvalds /* 25075d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 25085d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 25095d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 25105d2a2dbbSLinus Torvalds * zeroes. 25115d2a2dbbSLinus Torvalds */ 25125d2a2dbbSLinus Torvalds if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) 25133ecb01dfSJan Beulich clear_page(kaddr); 25149b04c5feSCong Wang kunmap_atomic(kaddr); 2515c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 25160ed361deSNick Piggin } else 25179de455b2SAtsushi Nemoto copy_user_highpage(dst, src, va, vma); 25186aab341eSLinus Torvalds } 25196aab341eSLinus Torvalds 25201da177e4SLinus Torvalds /* 25211da177e4SLinus Torvalds * This routine handles present pages, when users try to write 25221da177e4SLinus Torvalds * to a shared page. It is done by copying the page to a new address 25231da177e4SLinus Torvalds * and decrementing the shared-page counter for the old page. 25241da177e4SLinus Torvalds * 25251da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 25261da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 25271da177e4SLinus Torvalds * Thus we can safely just mark it writable once we've done any necessary 25281da177e4SLinus Torvalds * COW. 25291da177e4SLinus Torvalds * 25301da177e4SLinus Torvalds * We also mark the page dirty at this point even though the page will 25311da177e4SLinus Torvalds * change only once the write actually happens. This avoids a few races, 25321da177e4SLinus Torvalds * and potentially makes it more efficient. 25331da177e4SLinus Torvalds * 25348f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 25358f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 25368f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 25371da177e4SLinus Torvalds */ 25381da177e4SLinus Torvalds static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, 253965500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 25408f4e2101SHugh Dickins spinlock_t *ptl, pte_t orig_pte) 2541e6219ec8SNamhyung Kim __releases(ptl) 25421da177e4SLinus Torvalds { 25432ec74c3eSSagi Grimberg struct page *old_page, *new_page = NULL; 25441da177e4SLinus Torvalds pte_t entry; 2545b009c024SMichel Lespinasse int ret = 0; 2546a200ee18SPeter Zijlstra int page_mkwrite = 0; 2547d08b3851SPeter Zijlstra struct page *dirty_page = NULL; 25481756954cSDavid Rientjes unsigned long mmun_start = 0; /* For mmu_notifiers */ 25491756954cSDavid Rientjes unsigned long mmun_end = 0; /* For mmu_notifiers */ 25501da177e4SLinus Torvalds 25516aab341eSLinus Torvalds old_page = vm_normal_page(vma, address, orig_pte); 2552251b97f5SPeter Zijlstra if (!old_page) { 2553251b97f5SPeter Zijlstra /* 2554251b97f5SPeter Zijlstra * VM_MIXEDMAP !pfn_valid() case 2555251b97f5SPeter Zijlstra * 2556251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 2557251b97f5SPeter Zijlstra * Just mark the pages writable as we can't do any dirty 2558251b97f5SPeter Zijlstra * accounting on raw pfn maps. 2559251b97f5SPeter Zijlstra */ 2560251b97f5SPeter Zijlstra if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2561251b97f5SPeter Zijlstra (VM_WRITE|VM_SHARED)) 2562251b97f5SPeter Zijlstra goto reuse; 2563920fc356SHugh Dickins goto gotten; 2564251b97f5SPeter Zijlstra } 25651da177e4SLinus Torvalds 2566d08b3851SPeter Zijlstra /* 2567ee6a6457SPeter Zijlstra * Take out anonymous pages first, anonymous shared vmas are 2568ee6a6457SPeter Zijlstra * not dirty accountable. 2569d08b3851SPeter Zijlstra */ 25709a840895SHugh Dickins if (PageAnon(old_page) && !PageKsm(old_page)) { 2571ab967d86SHugh Dickins if (!trylock_page(old_page)) { 2572ab967d86SHugh Dickins page_cache_get(old_page); 2573ab967d86SHugh Dickins pte_unmap_unlock(page_table, ptl); 2574ab967d86SHugh Dickins lock_page(old_page); 2575ab967d86SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, 2576ab967d86SHugh Dickins &ptl); 2577ab967d86SHugh Dickins if (!pte_same(*page_table, orig_pte)) { 2578ab967d86SHugh Dickins unlock_page(old_page); 2579ab967d86SHugh Dickins goto unlock; 2580ab967d86SHugh Dickins } 2581ab967d86SHugh Dickins page_cache_release(old_page); 2582ab967d86SHugh Dickins } 2583b009c024SMichel Lespinasse if (reuse_swap_page(old_page)) { 2584c44b6743SRik van Riel /* 2585c44b6743SRik van Riel * The page is all ours. Move it to our anon_vma so 2586c44b6743SRik van Riel * the rmap code will not search our parent or siblings. 2587c44b6743SRik van Riel * Protected against the rmap code by the page lock. 2588c44b6743SRik van Riel */ 2589c44b6743SRik van Riel page_move_anon_rmap(old_page, vma, address); 2590ee6a6457SPeter Zijlstra unlock_page(old_page); 2591b009c024SMichel Lespinasse goto reuse; 2592b009c024SMichel Lespinasse } 2593b009c024SMichel Lespinasse unlock_page(old_page); 2594ee6a6457SPeter Zijlstra } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2595d08b3851SPeter Zijlstra (VM_WRITE|VM_SHARED))) { 2596ee6a6457SPeter Zijlstra /* 2597ee6a6457SPeter Zijlstra * Only catch write-faults on shared writable pages, 2598ee6a6457SPeter Zijlstra * read-only shared pages can get COWed by 2599ee6a6457SPeter Zijlstra * get_user_pages(.write=1, .force=1). 2600ee6a6457SPeter Zijlstra */ 26019637a5efSDavid Howells if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 2602c2ec175cSNick Piggin struct vm_fault vmf; 2603c2ec175cSNick Piggin int tmp; 2604c2ec175cSNick Piggin 2605c2ec175cSNick Piggin vmf.virtual_address = (void __user *)(address & 2606c2ec175cSNick Piggin PAGE_MASK); 2607c2ec175cSNick Piggin vmf.pgoff = old_page->index; 2608c2ec175cSNick Piggin vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 2609c2ec175cSNick Piggin vmf.page = old_page; 2610c2ec175cSNick Piggin 26119637a5efSDavid Howells /* 26129637a5efSDavid Howells * Notify the address space that the page is about to 26139637a5efSDavid Howells * become writable so that it can prohibit this or wait 26149637a5efSDavid Howells * for the page to get into an appropriate state. 26159637a5efSDavid Howells * 26169637a5efSDavid Howells * We do this without the lock held, so that it can 26179637a5efSDavid Howells * sleep if it needs to. 26189637a5efSDavid Howells */ 26199637a5efSDavid Howells page_cache_get(old_page); 26209637a5efSDavid Howells pte_unmap_unlock(page_table, ptl); 26219637a5efSDavid Howells 2622c2ec175cSNick Piggin tmp = vma->vm_ops->page_mkwrite(vma, &vmf); 2623c2ec175cSNick Piggin if (unlikely(tmp & 2624c2ec175cSNick Piggin (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { 2625c2ec175cSNick Piggin ret = tmp; 26269637a5efSDavid Howells goto unwritable_page; 2627c2ec175cSNick Piggin } 2628b827e496SNick Piggin if (unlikely(!(tmp & VM_FAULT_LOCKED))) { 2629b827e496SNick Piggin lock_page(old_page); 2630b827e496SNick Piggin if (!old_page->mapping) { 2631b827e496SNick Piggin ret = 0; /* retry the fault */ 2632b827e496SNick Piggin unlock_page(old_page); 2633b827e496SNick Piggin goto unwritable_page; 2634b827e496SNick Piggin } 2635b827e496SNick Piggin } else 2636b827e496SNick Piggin VM_BUG_ON(!PageLocked(old_page)); 26379637a5efSDavid Howells 26389637a5efSDavid Howells /* 26399637a5efSDavid Howells * Since we dropped the lock we need to revalidate 26409637a5efSDavid Howells * the PTE as someone else may have changed it. If 26419637a5efSDavid Howells * they did, we just return, as we can count on the 26429637a5efSDavid Howells * MMU to tell us if they didn't also make it writable. 26439637a5efSDavid Howells */ 26449637a5efSDavid Howells page_table = pte_offset_map_lock(mm, pmd, address, 26459637a5efSDavid Howells &ptl); 2646b827e496SNick Piggin if (!pte_same(*page_table, orig_pte)) { 2647b827e496SNick Piggin unlock_page(old_page); 26489637a5efSDavid Howells goto unlock; 2649b827e496SNick Piggin } 2650a200ee18SPeter Zijlstra 2651a200ee18SPeter Zijlstra page_mkwrite = 1; 26529637a5efSDavid Howells } 2653d08b3851SPeter Zijlstra dirty_page = old_page; 2654d08b3851SPeter Zijlstra get_page(dirty_page); 26559637a5efSDavid Howells 2656251b97f5SPeter Zijlstra reuse: 2657eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 265865500d23SHugh Dickins entry = pte_mkyoung(orig_pte); 265965500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2660954ffcb3SKAMEZAWA Hiroyuki if (ptep_set_access_flags(vma, address, page_table, entry,1)) 26614b3073e1SRussell King update_mmu_cache(vma, address, page_table); 266272ddc8f7SMichel Lespinasse pte_unmap_unlock(page_table, ptl); 266365500d23SHugh Dickins ret |= VM_FAULT_WRITE; 266472ddc8f7SMichel Lespinasse 266572ddc8f7SMichel Lespinasse if (!dirty_page) 266672ddc8f7SMichel Lespinasse return ret; 266772ddc8f7SMichel Lespinasse 266872ddc8f7SMichel Lespinasse /* 266972ddc8f7SMichel Lespinasse * Yes, Virginia, this is actually required to prevent a race 267072ddc8f7SMichel Lespinasse * with clear_page_dirty_for_io() from clearing the page dirty 267172ddc8f7SMichel Lespinasse * bit after it clear all dirty ptes, but before a racing 267272ddc8f7SMichel Lespinasse * do_wp_page installs a dirty pte. 267372ddc8f7SMichel Lespinasse * 2674a335b2e1SRyota Ozaki * __do_fault is protected similarly. 267572ddc8f7SMichel Lespinasse */ 267672ddc8f7SMichel Lespinasse if (!page_mkwrite) { 267772ddc8f7SMichel Lespinasse wait_on_page_locked(dirty_page); 267872ddc8f7SMichel Lespinasse set_page_dirty_balance(dirty_page, page_mkwrite); 267941c4d25fSJan Kara /* file_update_time outside page_lock */ 268041c4d25fSJan Kara if (vma->vm_file) 268141c4d25fSJan Kara file_update_time(vma->vm_file); 268272ddc8f7SMichel Lespinasse } 268372ddc8f7SMichel Lespinasse put_page(dirty_page); 268472ddc8f7SMichel Lespinasse if (page_mkwrite) { 268572ddc8f7SMichel Lespinasse struct address_space *mapping = dirty_page->mapping; 268672ddc8f7SMichel Lespinasse 268772ddc8f7SMichel Lespinasse set_page_dirty(dirty_page); 268872ddc8f7SMichel Lespinasse unlock_page(dirty_page); 268972ddc8f7SMichel Lespinasse page_cache_release(dirty_page); 269072ddc8f7SMichel Lespinasse if (mapping) { 269172ddc8f7SMichel Lespinasse /* 269272ddc8f7SMichel Lespinasse * Some device drivers do not set page.mapping 269372ddc8f7SMichel Lespinasse * but still dirty their pages 269472ddc8f7SMichel Lespinasse */ 269572ddc8f7SMichel Lespinasse balance_dirty_pages_ratelimited(mapping); 269672ddc8f7SMichel Lespinasse } 269772ddc8f7SMichel Lespinasse } 269872ddc8f7SMichel Lespinasse 269972ddc8f7SMichel Lespinasse return ret; 27001da177e4SLinus Torvalds } 27011da177e4SLinus Torvalds 27021da177e4SLinus Torvalds /* 27031da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 27041da177e4SLinus Torvalds */ 27051da177e4SLinus Torvalds page_cache_get(old_page); 2706920fc356SHugh Dickins gotten: 27078f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 27081da177e4SLinus Torvalds 27091da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 271065500d23SHugh Dickins goto oom; 2711a13ea5b7SHugh Dickins 271262eede62SHugh Dickins if (is_zero_pfn(pte_pfn(orig_pte))) { 2713a13ea5b7SHugh Dickins new_page = alloc_zeroed_user_highpage_movable(vma, address); 2714a13ea5b7SHugh Dickins if (!new_page) 2715a13ea5b7SHugh Dickins goto oom; 2716a13ea5b7SHugh Dickins } else { 2717769848c0SMel Gorman new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 27181da177e4SLinus Torvalds if (!new_page) 271965500d23SHugh Dickins goto oom; 2720a13ea5b7SHugh Dickins cow_user_page(new_page, old_page, address, vma); 2721a13ea5b7SHugh Dickins } 2722a13ea5b7SHugh Dickins __SetPageUptodate(new_page); 2723a13ea5b7SHugh Dickins 27242c26fdd7SKAMEZAWA Hiroyuki if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL)) 27258a9f3ccdSBalbir Singh goto oom_free_new; 27268a9f3ccdSBalbir Singh 27276bdb913fSHaggai Eran mmun_start = address & PAGE_MASK; 27281756954cSDavid Rientjes mmun_end = mmun_start + PAGE_SIZE; 27296bdb913fSHaggai Eran mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); 27306bdb913fSHaggai Eran 27311da177e4SLinus Torvalds /* 27321da177e4SLinus Torvalds * Re-check the pte - we dropped the lock 27331da177e4SLinus Torvalds */ 27348f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 273565500d23SHugh Dickins if (likely(pte_same(*page_table, orig_pte))) { 2736920fc356SHugh Dickins if (old_page) { 27374294621fSHugh Dickins if (!PageAnon(old_page)) { 273834e55232SKAMEZAWA Hiroyuki dec_mm_counter_fast(mm, MM_FILEPAGES); 273934e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 27404294621fSHugh Dickins } 2741920fc356SHugh Dickins } else 274234e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 2743eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 274465500d23SHugh Dickins entry = mk_pte(new_page, vma->vm_page_prot); 274565500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 27464ce072f1SSiddha, Suresh B /* 27474ce072f1SSiddha, Suresh B * Clear the pte entry and flush it first, before updating the 27484ce072f1SSiddha, Suresh B * pte with the new entry. This will avoid a race condition 27494ce072f1SSiddha, Suresh B * seen in the presence of one thread doing SMC and another 27504ce072f1SSiddha, Suresh B * thread doing COW. 27514ce072f1SSiddha, Suresh B */ 2752828502d3SIzik Eidus ptep_clear_flush(vma, address, page_table); 27539617d95eSNick Piggin page_add_new_anon_rmap(new_page, vma, address); 2754828502d3SIzik Eidus /* 2755828502d3SIzik Eidus * We call the notify macro here because, when using secondary 2756828502d3SIzik Eidus * mmu page tables (such as kvm shadow page tables), we want the 2757828502d3SIzik Eidus * new page to be mapped directly into the secondary page table. 2758828502d3SIzik Eidus */ 2759828502d3SIzik Eidus set_pte_at_notify(mm, address, page_table, entry); 27604b3073e1SRussell King update_mmu_cache(vma, address, page_table); 2761945754a1SNick Piggin if (old_page) { 2762945754a1SNick Piggin /* 2763945754a1SNick Piggin * Only after switching the pte to the new page may 2764945754a1SNick Piggin * we remove the mapcount here. Otherwise another 2765945754a1SNick Piggin * process may come and find the rmap count decremented 2766945754a1SNick Piggin * before the pte is switched to the new page, and 2767945754a1SNick Piggin * "reuse" the old page writing into it while our pte 2768945754a1SNick Piggin * here still points into it and can be read by other 2769945754a1SNick Piggin * threads. 2770945754a1SNick Piggin * 2771945754a1SNick Piggin * The critical issue is to order this 2772945754a1SNick Piggin * page_remove_rmap with the ptp_clear_flush above. 2773945754a1SNick Piggin * Those stores are ordered by (if nothing else,) 2774945754a1SNick Piggin * the barrier present in the atomic_add_negative 2775945754a1SNick Piggin * in page_remove_rmap. 2776945754a1SNick Piggin * 2777945754a1SNick Piggin * Then the TLB flush in ptep_clear_flush ensures that 2778945754a1SNick Piggin * no process can access the old page before the 2779945754a1SNick Piggin * decremented mapcount is visible. And the old page 2780945754a1SNick Piggin * cannot be reused until after the decremented 2781945754a1SNick Piggin * mapcount is visible. So transitively, TLBs to 2782945754a1SNick Piggin * old page will be flushed before it can be reused. 2783945754a1SNick Piggin */ 2784edc315fdSHugh Dickins page_remove_rmap(old_page); 2785945754a1SNick Piggin } 2786945754a1SNick Piggin 27871da177e4SLinus Torvalds /* Free the old page.. */ 27881da177e4SLinus Torvalds new_page = old_page; 2789f33ea7f4SNick Piggin ret |= VM_FAULT_WRITE; 27908a9f3ccdSBalbir Singh } else 27918a9f3ccdSBalbir Singh mem_cgroup_uncharge_page(new_page); 27928a9f3ccdSBalbir Singh 27936bdb913fSHaggai Eran if (new_page) 27946bdb913fSHaggai Eran page_cache_release(new_page); 279565500d23SHugh Dickins unlock: 27968f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 27971756954cSDavid Rientjes if (mmun_end > mmun_start) 27986bdb913fSHaggai Eran mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); 2799e15f8c01SMichel Lespinasse if (old_page) { 2800e15f8c01SMichel Lespinasse /* 2801e15f8c01SMichel Lespinasse * Don't let another task, with possibly unlocked vma, 2802e15f8c01SMichel Lespinasse * keep the mlocked page. 2803e15f8c01SMichel Lespinasse */ 2804e15f8c01SMichel Lespinasse if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) { 2805e15f8c01SMichel Lespinasse lock_page(old_page); /* LRU manipulation */ 2806e15f8c01SMichel Lespinasse munlock_vma_page(old_page); 2807e15f8c01SMichel Lespinasse unlock_page(old_page); 2808e15f8c01SMichel Lespinasse } 2809e15f8c01SMichel Lespinasse page_cache_release(old_page); 2810e15f8c01SMichel Lespinasse } 2811f33ea7f4SNick Piggin return ret; 28128a9f3ccdSBalbir Singh oom_free_new: 28136dbf6d3bSHugh Dickins page_cache_release(new_page); 281465500d23SHugh Dickins oom: 2815b827e496SNick Piggin if (old_page) { 2816b827e496SNick Piggin if (page_mkwrite) { 2817b827e496SNick Piggin unlock_page(old_page); 28181da177e4SLinus Torvalds page_cache_release(old_page); 2819b827e496SNick Piggin } 2820b827e496SNick Piggin page_cache_release(old_page); 2821b827e496SNick Piggin } 28221da177e4SLinus Torvalds return VM_FAULT_OOM; 28239637a5efSDavid Howells 28249637a5efSDavid Howells unwritable_page: 28259637a5efSDavid Howells page_cache_release(old_page); 2826c2ec175cSNick Piggin return ret; 28271da177e4SLinus Torvalds } 28281da177e4SLinus Torvalds 282997a89413SPeter Zijlstra static void unmap_mapping_range_vma(struct vm_area_struct *vma, 28301da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 28311da177e4SLinus Torvalds struct zap_details *details) 28321da177e4SLinus Torvalds { 2833f5cc4eefSAl Viro zap_page_range_single(vma, start_addr, end_addr - start_addr, details); 28341da177e4SLinus Torvalds } 28351da177e4SLinus Torvalds 28366b2dbba8SMichel Lespinasse static inline void unmap_mapping_range_tree(struct rb_root *root, 28371da177e4SLinus Torvalds struct zap_details *details) 28381da177e4SLinus Torvalds { 28391da177e4SLinus Torvalds struct vm_area_struct *vma; 28401da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 28411da177e4SLinus Torvalds 28426b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, root, 28431da177e4SLinus Torvalds details->first_index, details->last_index) { 28441da177e4SLinus Torvalds 28451da177e4SLinus Torvalds vba = vma->vm_pgoff; 28461da177e4SLinus Torvalds vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; 28471da177e4SLinus Torvalds /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ 28481da177e4SLinus Torvalds zba = details->first_index; 28491da177e4SLinus Torvalds if (zba < vba) 28501da177e4SLinus Torvalds zba = vba; 28511da177e4SLinus Torvalds zea = details->last_index; 28521da177e4SLinus Torvalds if (zea > vea) 28531da177e4SLinus Torvalds zea = vea; 28541da177e4SLinus Torvalds 285597a89413SPeter Zijlstra unmap_mapping_range_vma(vma, 28561da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 28571da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 285897a89413SPeter Zijlstra details); 28591da177e4SLinus Torvalds } 28601da177e4SLinus Torvalds } 28611da177e4SLinus Torvalds 28621da177e4SLinus Torvalds static inline void unmap_mapping_range_list(struct list_head *head, 28631da177e4SLinus Torvalds struct zap_details *details) 28641da177e4SLinus Torvalds { 28651da177e4SLinus Torvalds struct vm_area_struct *vma; 28661da177e4SLinus Torvalds 28671da177e4SLinus Torvalds /* 28681da177e4SLinus Torvalds * In nonlinear VMAs there is no correspondence between virtual address 28691da177e4SLinus Torvalds * offset and file offset. So we must perform an exhaustive search 28701da177e4SLinus Torvalds * across *all* the pages in each nonlinear VMA, not just the pages 28711da177e4SLinus Torvalds * whose virtual address lies outside the file truncation point. 28721da177e4SLinus Torvalds */ 28736b2dbba8SMichel Lespinasse list_for_each_entry(vma, head, shared.nonlinear) { 28741da177e4SLinus Torvalds details->nonlinear_vma = vma; 287597a89413SPeter Zijlstra unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details); 28761da177e4SLinus Torvalds } 28771da177e4SLinus Torvalds } 28781da177e4SLinus Torvalds 28791da177e4SLinus Torvalds /** 288072fd4a35SRobert 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. 28813d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 28821da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 28831da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 288425d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 28851da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 28861da177e4SLinus Torvalds * partial pages. 28871da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 28881da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 28891da177e4SLinus Torvalds * end of the file. 28901da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 28911da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 28921da177e4SLinus Torvalds */ 28931da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 28941da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 28951da177e4SLinus Torvalds { 28961da177e4SLinus Torvalds struct zap_details details; 28971da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 28981da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 28991da177e4SLinus Torvalds 29001da177e4SLinus Torvalds /* Check for overflow. */ 29011da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 29021da177e4SLinus Torvalds long long holeend = 29031da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 29041da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 29051da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 29061da177e4SLinus Torvalds } 29071da177e4SLinus Torvalds 29081da177e4SLinus Torvalds details.check_mapping = even_cows? NULL: mapping; 29091da177e4SLinus Torvalds details.nonlinear_vma = NULL; 29101da177e4SLinus Torvalds details.first_index = hba; 29111da177e4SLinus Torvalds details.last_index = hba + hlen - 1; 29121da177e4SLinus Torvalds if (details.last_index < details.first_index) 29131da177e4SLinus Torvalds details.last_index = ULONG_MAX; 29141da177e4SLinus Torvalds 291597a89413SPeter Zijlstra 29163d48ae45SPeter Zijlstra mutex_lock(&mapping->i_mmap_mutex); 29176b2dbba8SMichel Lespinasse if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) 29181da177e4SLinus Torvalds unmap_mapping_range_tree(&mapping->i_mmap, &details); 29191da177e4SLinus Torvalds if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) 29201da177e4SLinus Torvalds unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); 29213d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 29221da177e4SLinus Torvalds } 29231da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 29241da177e4SLinus Torvalds 29251da177e4SLinus Torvalds /* 29268f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 29278f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 29288f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 29291da177e4SLinus Torvalds */ 293065500d23SHugh Dickins static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, 293165500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 293230c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 29331da177e4SLinus Torvalds { 29348f4e2101SHugh Dickins spinlock_t *ptl; 29354969c119SAndrea Arcangeli struct page *page, *swapcache = NULL; 293665500d23SHugh Dickins swp_entry_t entry; 29371da177e4SLinus Torvalds pte_t pte; 2938d065bd81SMichel Lespinasse int locked; 293956039efaSKAMEZAWA Hiroyuki struct mem_cgroup *ptr; 2940ad8c2ee8SRik van Riel int exclusive = 0; 294183c54070SNick Piggin int ret = 0; 29421da177e4SLinus Torvalds 29434c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 29448f4e2101SHugh Dickins goto out; 294565500d23SHugh Dickins 294665500d23SHugh Dickins entry = pte_to_swp_entry(orig_pte); 2947d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 29480697212aSChristoph Lameter if (is_migration_entry(entry)) { 29490697212aSChristoph Lameter migration_entry_wait(mm, pmd, address); 2950d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 2951d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 2952d1737fdbSAndi Kleen } else { 2953d1737fdbSAndi Kleen print_bad_pte(vma, address, orig_pte, NULL); 2954d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 2955d1737fdbSAndi Kleen } 29560697212aSChristoph Lameter goto out; 29570697212aSChristoph Lameter } 29580ff92245SShailabh Nagar delayacct_set_flag(DELAYACCT_PF_SWAPIN); 29591da177e4SLinus Torvalds page = lookup_swap_cache(entry); 29601da177e4SLinus Torvalds if (!page) { 296102098feaSHugh Dickins page = swapin_readahead(entry, 296202098feaSHugh Dickins GFP_HIGHUSER_MOVABLE, vma, address); 29631da177e4SLinus Torvalds if (!page) { 29641da177e4SLinus Torvalds /* 29658f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 29668f4e2101SHugh Dickins * while we released the pte lock. 29671da177e4SLinus Torvalds */ 29688f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 29691da177e4SLinus Torvalds if (likely(pte_same(*page_table, orig_pte))) 29701da177e4SLinus Torvalds ret = VM_FAULT_OOM; 29710ff92245SShailabh Nagar delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 297265500d23SHugh Dickins goto unlock; 29731da177e4SLinus Torvalds } 29741da177e4SLinus Torvalds 29751da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 29761da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 2977f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 2978456f998eSYing Han mem_cgroup_count_vm_event(mm, PGMAJFAULT); 2979d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 298071f72525SWu Fengguang /* 298171f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 298271f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 298371f72525SWu Fengguang */ 2984d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 2985d1737fdbSAndi Kleen delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 29864779cb31SAndi Kleen goto out_release; 29871da177e4SLinus Torvalds } 29881da177e4SLinus Torvalds 2989d065bd81SMichel Lespinasse locked = lock_page_or_retry(page, mm, flags); 2990e709ffd6SRik van Riel 299120a1022dSBalbir Singh delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 2992d065bd81SMichel Lespinasse if (!locked) { 2993d065bd81SMichel Lespinasse ret |= VM_FAULT_RETRY; 2994d065bd81SMichel Lespinasse goto out_release; 2995d065bd81SMichel Lespinasse } 29961da177e4SLinus Torvalds 29974969c119SAndrea Arcangeli /* 299831c4a3d3SHugh Dickins * Make sure try_to_free_swap or reuse_swap_page or swapoff did not 299931c4a3d3SHugh Dickins * release the swapcache from under us. The page pin, and pte_same 300031c4a3d3SHugh Dickins * test below, are not enough to exclude that. Even if it is still 300131c4a3d3SHugh Dickins * swapcache, we need to check that the page's swap has not changed. 30024969c119SAndrea Arcangeli */ 300331c4a3d3SHugh Dickins if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) 30044969c119SAndrea Arcangeli goto out_page; 30054969c119SAndrea Arcangeli 30064969c119SAndrea Arcangeli if (ksm_might_need_to_copy(page, vma, address)) { 30074969c119SAndrea Arcangeli swapcache = page; 30084969c119SAndrea Arcangeli page = ksm_does_need_to_copy(page, vma, address); 30094969c119SAndrea Arcangeli 30104969c119SAndrea Arcangeli if (unlikely(!page)) { 30115ad64688SHugh Dickins ret = VM_FAULT_OOM; 30124969c119SAndrea Arcangeli page = swapcache; 30134969c119SAndrea Arcangeli swapcache = NULL; 30144969c119SAndrea Arcangeli goto out_page; 30154969c119SAndrea Arcangeli } 30165ad64688SHugh Dickins } 30175ad64688SHugh Dickins 30182c26fdd7SKAMEZAWA Hiroyuki if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) { 3019073e587eSKAMEZAWA Hiroyuki ret = VM_FAULT_OOM; 3020bc43f75cSJohannes Weiner goto out_page; 3021073e587eSKAMEZAWA Hiroyuki } 3022073e587eSKAMEZAWA Hiroyuki 30231da177e4SLinus Torvalds /* 30248f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 30251da177e4SLinus Torvalds */ 30268f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 30279e9bef07SHugh Dickins if (unlikely(!pte_same(*page_table, orig_pte))) 3028b8107480SKirill Korotaev goto out_nomap; 3029b8107480SKirill Korotaev 3030b8107480SKirill Korotaev if (unlikely(!PageUptodate(page))) { 3031b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 3032b8107480SKirill Korotaev goto out_nomap; 30331da177e4SLinus Torvalds } 30341da177e4SLinus Torvalds 30358c7c6e34SKAMEZAWA Hiroyuki /* 30368c7c6e34SKAMEZAWA Hiroyuki * The page isn't present yet, go ahead with the fault. 30378c7c6e34SKAMEZAWA Hiroyuki * 30388c7c6e34SKAMEZAWA Hiroyuki * Be careful about the sequence of operations here. 30398c7c6e34SKAMEZAWA Hiroyuki * To get its accounting right, reuse_swap_page() must be called 30408c7c6e34SKAMEZAWA Hiroyuki * while the page is counted on swap but not yet in mapcount i.e. 30418c7c6e34SKAMEZAWA Hiroyuki * before page_add_anon_rmap() and swap_free(); try_to_free_swap() 30428c7c6e34SKAMEZAWA Hiroyuki * must be called after the swap_free(), or it will never succeed. 304303f3c433SKAMEZAWA Hiroyuki * Because delete_from_swap_page() may be called by reuse_swap_page(), 304403f3c433SKAMEZAWA Hiroyuki * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry 304503f3c433SKAMEZAWA Hiroyuki * in page->private. In this case, a record in swap_cgroup is silently 304603f3c433SKAMEZAWA Hiroyuki * discarded at swap_free(). 30478c7c6e34SKAMEZAWA Hiroyuki */ 30481da177e4SLinus Torvalds 304934e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 3050b084d435SKAMEZAWA Hiroyuki dec_mm_counter_fast(mm, MM_SWAPENTS); 30511da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 305230c9f3a9SLinus Torvalds if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { 30531da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 305430c9f3a9SLinus Torvalds flags &= ~FAULT_FLAG_WRITE; 30559a5b489bSAndrea Arcangeli ret |= VM_FAULT_WRITE; 3056ad8c2ee8SRik van Riel exclusive = 1; 30571da177e4SLinus Torvalds } 30581da177e4SLinus Torvalds flush_icache_page(vma, page); 30591da177e4SLinus Torvalds set_pte_at(mm, address, page_table, pte); 3060ad8c2ee8SRik van Riel do_page_add_anon_rmap(page, vma, address, exclusive); 306103f3c433SKAMEZAWA Hiroyuki /* It's better to call commit-charge after rmap is established */ 306203f3c433SKAMEZAWA Hiroyuki mem_cgroup_commit_charge_swapin(page, ptr); 30631da177e4SLinus Torvalds 3064c475a8abSHugh Dickins swap_free(entry); 3065b291f000SNick Piggin if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) 3066a2c43eedSHugh Dickins try_to_free_swap(page); 3067c475a8abSHugh Dickins unlock_page(page); 30684969c119SAndrea Arcangeli if (swapcache) { 30694969c119SAndrea Arcangeli /* 30704969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 30714969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 30724969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 30734969c119SAndrea Arcangeli * further safety release the lock after the swap_free 30744969c119SAndrea Arcangeli * so that the swap count won't change under a 30754969c119SAndrea Arcangeli * parallel locked swapcache. 30764969c119SAndrea Arcangeli */ 30774969c119SAndrea Arcangeli unlock_page(swapcache); 30784969c119SAndrea Arcangeli page_cache_release(swapcache); 30794969c119SAndrea Arcangeli } 3080c475a8abSHugh Dickins 308130c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 308261469f1dSHugh Dickins ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); 308361469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 308461469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 30851da177e4SLinus Torvalds goto out; 30861da177e4SLinus Torvalds } 30871da177e4SLinus Torvalds 30881da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 30894b3073e1SRussell King update_mmu_cache(vma, address, page_table); 309065500d23SHugh Dickins unlock: 30918f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 30921da177e4SLinus Torvalds out: 30931da177e4SLinus Torvalds return ret; 3094b8107480SKirill Korotaev out_nomap: 30957a81b88cSKAMEZAWA Hiroyuki mem_cgroup_cancel_charge_swapin(ptr); 30968f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 3097bc43f75cSJohannes Weiner out_page: 3098b8107480SKirill Korotaev unlock_page(page); 30994779cb31SAndi Kleen out_release: 3100b8107480SKirill Korotaev page_cache_release(page); 31014969c119SAndrea Arcangeli if (swapcache) { 31024969c119SAndrea Arcangeli unlock_page(swapcache); 31034969c119SAndrea Arcangeli page_cache_release(swapcache); 31044969c119SAndrea Arcangeli } 310565500d23SHugh Dickins return ret; 31061da177e4SLinus Torvalds } 31071da177e4SLinus Torvalds 31081da177e4SLinus Torvalds /* 31098ca3eb08SLuck, Tony * This is like a special single-page "expand_{down|up}wards()", 31108ca3eb08SLuck, Tony * except we must first make sure that 'address{-|+}PAGE_SIZE' 3111320b2b8dSLinus Torvalds * doesn't hit another vma. 3112320b2b8dSLinus Torvalds */ 3113320b2b8dSLinus Torvalds static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) 3114320b2b8dSLinus Torvalds { 3115320b2b8dSLinus Torvalds address &= PAGE_MASK; 3116320b2b8dSLinus Torvalds if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { 31170e8e50e2SLinus Torvalds struct vm_area_struct *prev = vma->vm_prev; 3118320b2b8dSLinus Torvalds 31190e8e50e2SLinus Torvalds /* 31200e8e50e2SLinus Torvalds * Is there a mapping abutting this one below? 31210e8e50e2SLinus Torvalds * 31220e8e50e2SLinus Torvalds * That's only ok if it's the same stack mapping 31230e8e50e2SLinus Torvalds * that has gotten split.. 31240e8e50e2SLinus Torvalds */ 31250e8e50e2SLinus Torvalds if (prev && prev->vm_end == address) 31260e8e50e2SLinus Torvalds return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; 31270e8e50e2SLinus Torvalds 3128d05f3169SMichal Hocko expand_downwards(vma, address - PAGE_SIZE); 3129320b2b8dSLinus Torvalds } 31308ca3eb08SLuck, Tony if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { 31318ca3eb08SLuck, Tony struct vm_area_struct *next = vma->vm_next; 31328ca3eb08SLuck, Tony 31338ca3eb08SLuck, Tony /* As VM_GROWSDOWN but s/below/above/ */ 31348ca3eb08SLuck, Tony if (next && next->vm_start == address + PAGE_SIZE) 31358ca3eb08SLuck, Tony return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; 31368ca3eb08SLuck, Tony 31378ca3eb08SLuck, Tony expand_upwards(vma, address + PAGE_SIZE); 31388ca3eb08SLuck, Tony } 3139320b2b8dSLinus Torvalds return 0; 3140320b2b8dSLinus Torvalds } 3141320b2b8dSLinus Torvalds 3142320b2b8dSLinus Torvalds /* 31438f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 31448f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 31458f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 31461da177e4SLinus Torvalds */ 314765500d23SHugh Dickins static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, 314865500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 314930c9f3a9SLinus Torvalds unsigned int flags) 31501da177e4SLinus Torvalds { 31518f4e2101SHugh Dickins struct page *page; 31528f4e2101SHugh Dickins spinlock_t *ptl; 31531da177e4SLinus Torvalds pte_t entry; 31541da177e4SLinus Torvalds 31555528f913SLinus Torvalds pte_unmap(page_table); 3156320b2b8dSLinus Torvalds 315711ac5524SLinus Torvalds /* Check if we need to add a guard page to the stack */ 315811ac5524SLinus Torvalds if (check_stack_guard_page(vma, address) < 0) 315911ac5524SLinus Torvalds return VM_FAULT_SIGBUS; 316011ac5524SLinus Torvalds 316111ac5524SLinus Torvalds /* Use the zero-page for reads */ 316262eede62SHugh Dickins if (!(flags & FAULT_FLAG_WRITE)) { 316362eede62SHugh Dickins entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), 316462eede62SHugh Dickins vma->vm_page_prot)); 316511ac5524SLinus Torvalds page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 3166a13ea5b7SHugh Dickins if (!pte_none(*page_table)) 3167a13ea5b7SHugh Dickins goto unlock; 3168a13ea5b7SHugh Dickins goto setpte; 3169a13ea5b7SHugh Dickins } 3170a13ea5b7SHugh Dickins 31711da177e4SLinus Torvalds /* Allocate our own private page. */ 31721da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 317365500d23SHugh Dickins goto oom; 3174769848c0SMel Gorman page = alloc_zeroed_user_highpage_movable(vma, address); 31751da177e4SLinus Torvalds if (!page) 317665500d23SHugh Dickins goto oom; 31770ed361deSNick Piggin __SetPageUptodate(page); 31781da177e4SLinus Torvalds 31792c26fdd7SKAMEZAWA Hiroyuki if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) 31808a9f3ccdSBalbir Singh goto oom_free_page; 31818a9f3ccdSBalbir Singh 318265500d23SHugh Dickins entry = mk_pte(page, vma->vm_page_prot); 31831ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 31841ac0cb5dSHugh Dickins entry = pte_mkwrite(pte_mkdirty(entry)); 31858f4e2101SHugh Dickins 31868f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 31871c2fb7a4SAndrea Arcangeli if (!pte_none(*page_table)) 31888f4e2101SHugh Dickins goto release; 31899ba69294SHugh Dickins 319034e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 31919617d95eSNick Piggin page_add_new_anon_rmap(page, vma, address); 3192a13ea5b7SHugh Dickins setpte: 319365500d23SHugh Dickins set_pte_at(mm, address, page_table, entry); 31941da177e4SLinus Torvalds 31951da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 31964b3073e1SRussell King update_mmu_cache(vma, address, page_table); 319765500d23SHugh Dickins unlock: 31988f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 319983c54070SNick Piggin return 0; 32008f4e2101SHugh Dickins release: 32018a9f3ccdSBalbir Singh mem_cgroup_uncharge_page(page); 32028f4e2101SHugh Dickins page_cache_release(page); 32038f4e2101SHugh Dickins goto unlock; 32048a9f3ccdSBalbir Singh oom_free_page: 32056dbf6d3bSHugh Dickins page_cache_release(page); 320665500d23SHugh Dickins oom: 32071da177e4SLinus Torvalds return VM_FAULT_OOM; 32081da177e4SLinus Torvalds } 32091da177e4SLinus Torvalds 32101da177e4SLinus Torvalds /* 321154cb8821SNick Piggin * __do_fault() tries to create a new page mapping. It aggressively 32121da177e4SLinus Torvalds * tries to share with existing pages, but makes a separate copy if 321354cb8821SNick Piggin * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid 321454cb8821SNick Piggin * the next page fault. 32151da177e4SLinus Torvalds * 32161da177e4SLinus Torvalds * As this is called only for pages that do not currently exist, we 32171da177e4SLinus Torvalds * do not need to flush old virtual caches or the TLB. 32181da177e4SLinus Torvalds * 32198f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 322016abfa08SHugh Dickins * but allow concurrent faults), and pte neither mapped nor locked. 32218f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 32221da177e4SLinus Torvalds */ 322354cb8821SNick Piggin static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma, 322416abfa08SHugh Dickins unsigned long address, pmd_t *pmd, 322554cb8821SNick Piggin pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 32261da177e4SLinus Torvalds { 322716abfa08SHugh Dickins pte_t *page_table; 32288f4e2101SHugh Dickins spinlock_t *ptl; 3229d0217ac0SNick Piggin struct page *page; 32301d65f86dSKAMEZAWA Hiroyuki struct page *cow_page; 32311da177e4SLinus Torvalds pte_t entry; 32321da177e4SLinus Torvalds int anon = 0; 3233d08b3851SPeter Zijlstra struct page *dirty_page = NULL; 3234d0217ac0SNick Piggin struct vm_fault vmf; 3235d0217ac0SNick Piggin int ret; 3236a200ee18SPeter Zijlstra int page_mkwrite = 0; 323754cb8821SNick Piggin 32381d65f86dSKAMEZAWA Hiroyuki /* 32391d65f86dSKAMEZAWA Hiroyuki * If we do COW later, allocate page befor taking lock_page() 32401d65f86dSKAMEZAWA Hiroyuki * on the file cache page. This will reduce lock holding time. 32411d65f86dSKAMEZAWA Hiroyuki */ 32421d65f86dSKAMEZAWA Hiroyuki if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) { 32431d65f86dSKAMEZAWA Hiroyuki 32441d65f86dSKAMEZAWA Hiroyuki if (unlikely(anon_vma_prepare(vma))) 32451d65f86dSKAMEZAWA Hiroyuki return VM_FAULT_OOM; 32461d65f86dSKAMEZAWA Hiroyuki 32471d65f86dSKAMEZAWA Hiroyuki cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 32481d65f86dSKAMEZAWA Hiroyuki if (!cow_page) 32491d65f86dSKAMEZAWA Hiroyuki return VM_FAULT_OOM; 32501d65f86dSKAMEZAWA Hiroyuki 32511d65f86dSKAMEZAWA Hiroyuki if (mem_cgroup_newpage_charge(cow_page, mm, GFP_KERNEL)) { 32521d65f86dSKAMEZAWA Hiroyuki page_cache_release(cow_page); 32531d65f86dSKAMEZAWA Hiroyuki return VM_FAULT_OOM; 32541d65f86dSKAMEZAWA Hiroyuki } 32551d65f86dSKAMEZAWA Hiroyuki } else 32561d65f86dSKAMEZAWA Hiroyuki cow_page = NULL; 32571d65f86dSKAMEZAWA Hiroyuki 3258d0217ac0SNick Piggin vmf.virtual_address = (void __user *)(address & PAGE_MASK); 3259d0217ac0SNick Piggin vmf.pgoff = pgoff; 3260d0217ac0SNick Piggin vmf.flags = flags; 3261d0217ac0SNick Piggin vmf.page = NULL; 32621da177e4SLinus Torvalds 3263d0217ac0SNick Piggin ret = vma->vm_ops->fault(vma, &vmf); 3264d065bd81SMichel Lespinasse if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | 3265d065bd81SMichel Lespinasse VM_FAULT_RETRY))) 32661d65f86dSKAMEZAWA Hiroyuki goto uncharge_out; 32671da177e4SLinus Torvalds 3268a3b947eaSAndi Kleen if (unlikely(PageHWPoison(vmf.page))) { 3269a3b947eaSAndi Kleen if (ret & VM_FAULT_LOCKED) 3270a3b947eaSAndi Kleen unlock_page(vmf.page); 32711d65f86dSKAMEZAWA Hiroyuki ret = VM_FAULT_HWPOISON; 32721d65f86dSKAMEZAWA Hiroyuki goto uncharge_out; 3273a3b947eaSAndi Kleen } 3274a3b947eaSAndi Kleen 3275d00806b1SNick Piggin /* 3276d0217ac0SNick Piggin * For consistency in subsequent calls, make the faulted page always 3277d00806b1SNick Piggin * locked. 3278d00806b1SNick Piggin */ 327983c54070SNick Piggin if (unlikely(!(ret & VM_FAULT_LOCKED))) 3280d0217ac0SNick Piggin lock_page(vmf.page); 328154cb8821SNick Piggin else 3282d0217ac0SNick Piggin VM_BUG_ON(!PageLocked(vmf.page)); 3283d00806b1SNick Piggin 32841da177e4SLinus Torvalds /* 32851da177e4SLinus Torvalds * Should we do an early C-O-W break? 32861da177e4SLinus Torvalds */ 3287d0217ac0SNick Piggin page = vmf.page; 328854cb8821SNick Piggin if (flags & FAULT_FLAG_WRITE) { 32899637a5efSDavid Howells if (!(vma->vm_flags & VM_SHARED)) { 32901d65f86dSKAMEZAWA Hiroyuki page = cow_page; 329154cb8821SNick Piggin anon = 1; 3292d0217ac0SNick Piggin copy_user_highpage(page, vmf.page, address, vma); 32930ed361deSNick Piggin __SetPageUptodate(page); 32949637a5efSDavid Howells } else { 329554cb8821SNick Piggin /* 329654cb8821SNick Piggin * If the page will be shareable, see if the backing 32979637a5efSDavid Howells * address space wants to know that the page is about 329854cb8821SNick Piggin * to become writable 329954cb8821SNick Piggin */ 330069676147SMark Fasheh if (vma->vm_ops->page_mkwrite) { 3301c2ec175cSNick Piggin int tmp; 3302c2ec175cSNick Piggin 330369676147SMark Fasheh unlock_page(page); 3304b827e496SNick Piggin vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 3305c2ec175cSNick Piggin tmp = vma->vm_ops->page_mkwrite(vma, &vmf); 3306c2ec175cSNick Piggin if (unlikely(tmp & 3307c2ec175cSNick Piggin (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { 3308c2ec175cSNick Piggin ret = tmp; 3309b827e496SNick Piggin goto unwritable_page; 331069676147SMark Fasheh } 3311b827e496SNick Piggin if (unlikely(!(tmp & VM_FAULT_LOCKED))) { 331269676147SMark Fasheh lock_page(page); 3313d0217ac0SNick Piggin if (!page->mapping) { 3314b827e496SNick Piggin ret = 0; /* retry the fault */ 3315b827e496SNick Piggin unlock_page(page); 3316b827e496SNick Piggin goto unwritable_page; 3317d0217ac0SNick Piggin } 3318b827e496SNick Piggin } else 3319b827e496SNick Piggin VM_BUG_ON(!PageLocked(page)); 3320a200ee18SPeter Zijlstra page_mkwrite = 1; 33219637a5efSDavid Howells } 33229637a5efSDavid Howells } 332354cb8821SNick Piggin 33241da177e4SLinus Torvalds } 33251da177e4SLinus Torvalds 33268f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 33271da177e4SLinus Torvalds 33281da177e4SLinus Torvalds /* 33291da177e4SLinus Torvalds * This silly early PAGE_DIRTY setting removes a race 33301da177e4SLinus Torvalds * due to the bad i386 page protection. But it's valid 33311da177e4SLinus Torvalds * for other architectures too. 33321da177e4SLinus Torvalds * 333330c9f3a9SLinus Torvalds * Note that if FAULT_FLAG_WRITE is set, we either now have 33341da177e4SLinus Torvalds * an exclusive copy of the page, or this is a shared mapping, 33351da177e4SLinus Torvalds * so we can make it writable and dirty to avoid having to 33361da177e4SLinus Torvalds * handle that later. 33371da177e4SLinus Torvalds */ 33381da177e4SLinus Torvalds /* Only go through if we didn't race with anybody else... */ 33391c2fb7a4SAndrea Arcangeli if (likely(pte_same(*page_table, orig_pte))) { 3340d00806b1SNick Piggin flush_icache_page(vma, page); 3341d00806b1SNick Piggin entry = mk_pte(page, vma->vm_page_prot); 334254cb8821SNick Piggin if (flags & FAULT_FLAG_WRITE) 33431da177e4SLinus Torvalds entry = maybe_mkwrite(pte_mkdirty(entry), vma); 33441da177e4SLinus Torvalds if (anon) { 334534e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 3346d00806b1SNick Piggin page_add_new_anon_rmap(page, vma, address); 3347f57e88a8SHugh Dickins } else { 334834e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_FILEPAGES); 3349d00806b1SNick Piggin page_add_file_rmap(page); 335054cb8821SNick Piggin if (flags & FAULT_FLAG_WRITE) { 3351d00806b1SNick Piggin dirty_page = page; 3352d08b3851SPeter Zijlstra get_page(dirty_page); 3353d08b3851SPeter Zijlstra } 33544294621fSHugh Dickins } 335564d6519dSLee Schermerhorn set_pte_at(mm, address, page_table, entry); 33561da177e4SLinus Torvalds 3357d00806b1SNick Piggin /* no need to invalidate: a not-present page won't be cached */ 33584b3073e1SRussell King update_mmu_cache(vma, address, page_table); 3359d00806b1SNick Piggin } else { 33601d65f86dSKAMEZAWA Hiroyuki if (cow_page) 33611d65f86dSKAMEZAWA Hiroyuki mem_cgroup_uncharge_page(cow_page); 3362d00806b1SNick Piggin if (anon) 3363d00806b1SNick Piggin page_cache_release(page); 3364d00806b1SNick Piggin else 336554cb8821SNick Piggin anon = 1; /* no anon but release faulted_page */ 3366d00806b1SNick Piggin } 3367d00806b1SNick Piggin 33688f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 3369d00806b1SNick Piggin 3370b827e496SNick Piggin if (dirty_page) { 3371b827e496SNick Piggin struct address_space *mapping = page->mapping; 337241c4d25fSJan Kara int dirtied = 0; 33738f7b3d15SAnton Salikhmetov 3374b827e496SNick Piggin if (set_page_dirty(dirty_page)) 337541c4d25fSJan Kara dirtied = 1; 3376b827e496SNick Piggin unlock_page(dirty_page); 3377d08b3851SPeter Zijlstra put_page(dirty_page); 337841c4d25fSJan Kara if ((dirtied || page_mkwrite) && mapping) { 3379b827e496SNick Piggin /* 3380b827e496SNick Piggin * Some device drivers do not set page.mapping but still 3381b827e496SNick Piggin * dirty their pages 3382b827e496SNick Piggin */ 3383b827e496SNick Piggin balance_dirty_pages_ratelimited(mapping); 3384d08b3851SPeter Zijlstra } 3385d00806b1SNick Piggin 3386b827e496SNick Piggin /* file_update_time outside page_lock */ 338741c4d25fSJan Kara if (vma->vm_file && !page_mkwrite) 3388b827e496SNick Piggin file_update_time(vma->vm_file); 3389b827e496SNick Piggin } else { 3390b827e496SNick Piggin unlock_page(vmf.page); 3391b827e496SNick Piggin if (anon) 3392b827e496SNick Piggin page_cache_release(vmf.page); 3393b827e496SNick Piggin } 3394b827e496SNick Piggin 3395b827e496SNick Piggin return ret; 3396b827e496SNick Piggin 3397b827e496SNick Piggin unwritable_page: 3398b827e496SNick Piggin page_cache_release(page); 339983c54070SNick Piggin return ret; 34001d65f86dSKAMEZAWA Hiroyuki uncharge_out: 34011d65f86dSKAMEZAWA Hiroyuki /* fs's fault handler get error */ 34021d65f86dSKAMEZAWA Hiroyuki if (cow_page) { 34031d65f86dSKAMEZAWA Hiroyuki mem_cgroup_uncharge_page(cow_page); 34041d65f86dSKAMEZAWA Hiroyuki page_cache_release(cow_page); 34051d65f86dSKAMEZAWA Hiroyuki } 34061d65f86dSKAMEZAWA Hiroyuki return ret; 340754cb8821SNick Piggin } 3408d00806b1SNick Piggin 340954cb8821SNick Piggin static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, 341054cb8821SNick Piggin unsigned long address, pte_t *page_table, pmd_t *pmd, 341130c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 341254cb8821SNick Piggin { 341354cb8821SNick Piggin pgoff_t pgoff = (((address & PAGE_MASK) 34140da7e01fSDean Nelson - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 341554cb8821SNick Piggin 341616abfa08SHugh Dickins pte_unmap(page_table); 341716abfa08SHugh Dickins return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); 341854cb8821SNick Piggin } 341954cb8821SNick Piggin 3420f4b81804SJes Sorensen /* 34211da177e4SLinus Torvalds * Fault of a previously existing named mapping. Repopulate the pte 34221da177e4SLinus Torvalds * from the encoded file_pte if possible. This enables swappable 34231da177e4SLinus Torvalds * nonlinear vmas. 34248f4e2101SHugh Dickins * 34258f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 34268f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 34278f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 34281da177e4SLinus Torvalds */ 3429d0217ac0SNick Piggin static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, 343065500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 343130c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 34321da177e4SLinus Torvalds { 343365500d23SHugh Dickins pgoff_t pgoff; 34341da177e4SLinus Torvalds 343530c9f3a9SLinus Torvalds flags |= FAULT_FLAG_NONLINEAR; 343630c9f3a9SLinus Torvalds 34374c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 343883c54070SNick Piggin return 0; 34391da177e4SLinus Torvalds 34402509ef26SHugh Dickins if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) { 344165500d23SHugh Dickins /* 344265500d23SHugh Dickins * Page table corrupted: show pte and kill process. 344365500d23SHugh Dickins */ 34443dc14741SHugh Dickins print_bad_pte(vma, address, orig_pte, NULL); 3445d99be1a8SHugh Dickins return VM_FAULT_SIGBUS; 344665500d23SHugh Dickins } 344765500d23SHugh Dickins 344865500d23SHugh Dickins pgoff = pte_to_pgoff(orig_pte); 344916abfa08SHugh Dickins return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); 34501da177e4SLinus Torvalds } 34511da177e4SLinus Torvalds 3452*9532fec1SMel Gorman int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, 3453*9532fec1SMel Gorman unsigned long addr, int current_nid) 3454*9532fec1SMel Gorman { 3455*9532fec1SMel Gorman get_page(page); 3456*9532fec1SMel Gorman 3457*9532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS); 3458*9532fec1SMel Gorman if (current_nid == numa_node_id()) 3459*9532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); 3460*9532fec1SMel Gorman 3461*9532fec1SMel Gorman return mpol_misplaced(page, vma, addr); 3462*9532fec1SMel Gorman } 3463*9532fec1SMel Gorman 3464d10e63f2SMel Gorman int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma, 3465d10e63f2SMel Gorman unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd) 3466d10e63f2SMel Gorman { 34674daae3b4SMel Gorman struct page *page = NULL; 3468d10e63f2SMel Gorman spinlock_t *ptl; 3469cbee9f88SPeter Zijlstra int current_nid = -1; 3470cbee9f88SPeter Zijlstra int target_nid; 3471d10e63f2SMel Gorman 3472d10e63f2SMel Gorman /* 3473d10e63f2SMel Gorman * The "pte" at this point cannot be used safely without 3474d10e63f2SMel Gorman * validation through pte_unmap_same(). It's of NUMA type but 3475d10e63f2SMel Gorman * the pfn may be screwed if the read is non atomic. 3476d10e63f2SMel Gorman * 3477d10e63f2SMel Gorman * ptep_modify_prot_start is not called as this is clearing 3478d10e63f2SMel Gorman * the _PAGE_NUMA bit and it is not really expected that there 3479d10e63f2SMel Gorman * would be concurrent hardware modifications to the PTE. 3480d10e63f2SMel Gorman */ 3481d10e63f2SMel Gorman ptl = pte_lockptr(mm, pmd); 3482d10e63f2SMel Gorman spin_lock(ptl); 34834daae3b4SMel Gorman if (unlikely(!pte_same(*ptep, pte))) { 34844daae3b4SMel Gorman pte_unmap_unlock(ptep, ptl); 34854daae3b4SMel Gorman goto out; 34864daae3b4SMel Gorman } 34874daae3b4SMel Gorman 3488d10e63f2SMel Gorman pte = pte_mknonnuma(pte); 3489d10e63f2SMel Gorman set_pte_at(mm, addr, ptep, pte); 3490d10e63f2SMel Gorman update_mmu_cache(vma, addr, ptep); 3491d10e63f2SMel Gorman 3492d10e63f2SMel Gorman page = vm_normal_page(vma, addr, pte); 3493d10e63f2SMel Gorman if (!page) { 3494d10e63f2SMel Gorman pte_unmap_unlock(ptep, ptl); 3495d10e63f2SMel Gorman return 0; 3496d10e63f2SMel Gorman } 3497d10e63f2SMel Gorman 34984daae3b4SMel Gorman current_nid = page_to_nid(page); 3499*9532fec1SMel Gorman target_nid = numa_migrate_prep(page, vma, addr, current_nid); 3500d10e63f2SMel Gorman pte_unmap_unlock(ptep, ptl); 35014daae3b4SMel Gorman if (target_nid == -1) { 35024daae3b4SMel Gorman /* 35034daae3b4SMel Gorman * Account for the fault against the current node if it not 35044daae3b4SMel Gorman * being replaced regardless of where the page is located. 35054daae3b4SMel Gorman */ 35064daae3b4SMel Gorman current_nid = numa_node_id(); 35074daae3b4SMel Gorman put_page(page); 35084daae3b4SMel Gorman goto out; 35094daae3b4SMel Gorman } 35104daae3b4SMel Gorman 35114daae3b4SMel Gorman /* Migrate to the requested node */ 35124daae3b4SMel Gorman if (migrate_misplaced_page(page, target_nid)) 35134daae3b4SMel Gorman current_nid = target_nid; 35144daae3b4SMel Gorman 35154daae3b4SMel Gorman out: 3516*9532fec1SMel Gorman if (current_nid != -1) 3517cbee9f88SPeter Zijlstra task_numa_fault(current_nid, 1); 3518d10e63f2SMel Gorman return 0; 3519d10e63f2SMel Gorman } 3520d10e63f2SMel Gorman 3521d10e63f2SMel Gorman /* NUMA hinting page fault entry point for regular pmds */ 3522d10e63f2SMel Gorman #ifdef CONFIG_NUMA_BALANCING 3523d10e63f2SMel Gorman static int do_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma, 3524d10e63f2SMel Gorman unsigned long addr, pmd_t *pmdp) 3525d10e63f2SMel Gorman { 3526d10e63f2SMel Gorman pmd_t pmd; 3527d10e63f2SMel Gorman pte_t *pte, *orig_pte; 3528d10e63f2SMel Gorman unsigned long _addr = addr & PMD_MASK; 3529d10e63f2SMel Gorman unsigned long offset; 3530d10e63f2SMel Gorman spinlock_t *ptl; 3531d10e63f2SMel Gorman bool numa = false; 353203c5a6e1SMel Gorman int local_nid = numa_node_id(); 3533d10e63f2SMel Gorman 3534d10e63f2SMel Gorman spin_lock(&mm->page_table_lock); 3535d10e63f2SMel Gorman pmd = *pmdp; 3536d10e63f2SMel Gorman if (pmd_numa(pmd)) { 3537d10e63f2SMel Gorman set_pmd_at(mm, _addr, pmdp, pmd_mknonnuma(pmd)); 3538d10e63f2SMel Gorman numa = true; 3539d10e63f2SMel Gorman } 3540d10e63f2SMel Gorman spin_unlock(&mm->page_table_lock); 3541d10e63f2SMel Gorman 3542d10e63f2SMel Gorman if (!numa) 3543d10e63f2SMel Gorman return 0; 3544d10e63f2SMel Gorman 3545d10e63f2SMel Gorman /* we're in a page fault so some vma must be in the range */ 3546d10e63f2SMel Gorman BUG_ON(!vma); 3547d10e63f2SMel Gorman BUG_ON(vma->vm_start >= _addr + PMD_SIZE); 3548d10e63f2SMel Gorman offset = max(_addr, vma->vm_start) & ~PMD_MASK; 3549d10e63f2SMel Gorman VM_BUG_ON(offset >= PMD_SIZE); 3550d10e63f2SMel Gorman orig_pte = pte = pte_offset_map_lock(mm, pmdp, _addr, &ptl); 3551d10e63f2SMel Gorman pte += offset >> PAGE_SHIFT; 3552d10e63f2SMel Gorman for (addr = _addr + offset; addr < _addr + PMD_SIZE; pte++, addr += PAGE_SIZE) { 3553d10e63f2SMel Gorman pte_t pteval = *pte; 3554d10e63f2SMel Gorman struct page *page; 3555*9532fec1SMel Gorman int curr_nid = local_nid; 3556*9532fec1SMel Gorman int target_nid; 3557d10e63f2SMel Gorman if (!pte_present(pteval)) 3558d10e63f2SMel Gorman continue; 3559d10e63f2SMel Gorman if (!pte_numa(pteval)) 3560d10e63f2SMel Gorman continue; 3561d10e63f2SMel Gorman if (addr >= vma->vm_end) { 3562d10e63f2SMel Gorman vma = find_vma(mm, addr); 3563d10e63f2SMel Gorman /* there's a pte present so there must be a vma */ 3564d10e63f2SMel Gorman BUG_ON(!vma); 3565d10e63f2SMel Gorman BUG_ON(addr < vma->vm_start); 3566d10e63f2SMel Gorman } 3567d10e63f2SMel Gorman if (pte_numa(pteval)) { 3568d10e63f2SMel Gorman pteval = pte_mknonnuma(pteval); 3569d10e63f2SMel Gorman set_pte_at(mm, addr, pte, pteval); 3570d10e63f2SMel Gorman } 3571d10e63f2SMel Gorman page = vm_normal_page(vma, addr, pteval); 3572d10e63f2SMel Gorman if (unlikely(!page)) 3573d10e63f2SMel Gorman continue; 3574cbee9f88SPeter Zijlstra /* only check non-shared pages */ 3575cbee9f88SPeter Zijlstra if (unlikely(page_mapcount(page) != 1)) 3576cbee9f88SPeter Zijlstra continue; 3577*9532fec1SMel Gorman 3578*9532fec1SMel Gorman /* 3579*9532fec1SMel Gorman * Note that the NUMA fault is later accounted to either 3580*9532fec1SMel Gorman * the node that is currently running or where the page is 3581*9532fec1SMel Gorman * migrated to. 3582*9532fec1SMel Gorman */ 3583*9532fec1SMel Gorman curr_nid = local_nid; 3584*9532fec1SMel Gorman target_nid = numa_migrate_prep(page, vma, addr, 3585*9532fec1SMel Gorman page_to_nid(page)); 3586*9532fec1SMel Gorman if (target_nid == -1) { 3587*9532fec1SMel Gorman put_page(page); 3588*9532fec1SMel Gorman continue; 3589*9532fec1SMel Gorman } 3590*9532fec1SMel Gorman 3591*9532fec1SMel Gorman /* Migrate to the requested node */ 3592cbee9f88SPeter Zijlstra pte_unmap_unlock(pte, ptl); 3593*9532fec1SMel Gorman if (migrate_misplaced_page(page, target_nid)) 3594*9532fec1SMel Gorman curr_nid = target_nid; 3595cbee9f88SPeter Zijlstra task_numa_fault(curr_nid, 1); 3596cbee9f88SPeter Zijlstra 3597cbee9f88SPeter Zijlstra pte = pte_offset_map_lock(mm, pmdp, addr, &ptl); 3598d10e63f2SMel Gorman } 3599d10e63f2SMel Gorman pte_unmap_unlock(orig_pte, ptl); 3600d10e63f2SMel Gorman 3601d10e63f2SMel Gorman return 0; 3602d10e63f2SMel Gorman } 3603d10e63f2SMel Gorman #else 3604d10e63f2SMel Gorman static int do_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma, 3605d10e63f2SMel Gorman unsigned long addr, pmd_t *pmdp) 3606d10e63f2SMel Gorman { 3607d10e63f2SMel Gorman BUG(); 3608d10e63f2SMel Gorman } 3609d10e63f2SMel Gorman #endif /* CONFIG_NUMA_BALANCING */ 3610d10e63f2SMel Gorman 36111da177e4SLinus Torvalds /* 36121da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 36131da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 36141da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 36151da177e4SLinus Torvalds * 36161da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 36171da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 36181da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 36191da177e4SLinus Torvalds * 3620c74df32cSHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 3621c74df32cSHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 3622c74df32cSHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 36231da177e4SLinus Torvalds */ 362471e3aac0SAndrea Arcangeli int handle_pte_fault(struct mm_struct *mm, 36251da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long address, 362630c9f3a9SLinus Torvalds pte_t *pte, pmd_t *pmd, unsigned int flags) 36271da177e4SLinus Torvalds { 36281da177e4SLinus Torvalds pte_t entry; 36298f4e2101SHugh Dickins spinlock_t *ptl; 36301da177e4SLinus Torvalds 36318dab5241SBenjamin Herrenschmidt entry = *pte; 36321da177e4SLinus Torvalds if (!pte_present(entry)) { 363365500d23SHugh Dickins if (pte_none(entry)) { 3634f4b81804SJes Sorensen if (vma->vm_ops) { 36353c18ddd1SNick Piggin if (likely(vma->vm_ops->fault)) 363654cb8821SNick Piggin return do_linear_fault(mm, vma, address, 363730c9f3a9SLinus Torvalds pte, pmd, flags, entry); 3638f4b81804SJes Sorensen } 3639f4b81804SJes Sorensen return do_anonymous_page(mm, vma, address, 364030c9f3a9SLinus Torvalds pte, pmd, flags); 364165500d23SHugh Dickins } 36421da177e4SLinus Torvalds if (pte_file(entry)) 3643d0217ac0SNick Piggin return do_nonlinear_fault(mm, vma, address, 364430c9f3a9SLinus Torvalds pte, pmd, flags, entry); 364565500d23SHugh Dickins return do_swap_page(mm, vma, address, 364630c9f3a9SLinus Torvalds pte, pmd, flags, entry); 36471da177e4SLinus Torvalds } 36481da177e4SLinus Torvalds 3649d10e63f2SMel Gorman if (pte_numa(entry)) 3650d10e63f2SMel Gorman return do_numa_page(mm, vma, address, entry, pte, pmd); 3651d10e63f2SMel Gorman 36524c21e2f2SHugh Dickins ptl = pte_lockptr(mm, pmd); 36538f4e2101SHugh Dickins spin_lock(ptl); 36548f4e2101SHugh Dickins if (unlikely(!pte_same(*pte, entry))) 36558f4e2101SHugh Dickins goto unlock; 365630c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 36571da177e4SLinus Torvalds if (!pte_write(entry)) 36588f4e2101SHugh Dickins return do_wp_page(mm, vma, address, 36598f4e2101SHugh Dickins pte, pmd, ptl, entry); 36601da177e4SLinus Torvalds entry = pte_mkdirty(entry); 36611da177e4SLinus Torvalds } 36621da177e4SLinus Torvalds entry = pte_mkyoung(entry); 366330c9f3a9SLinus Torvalds if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { 36644b3073e1SRussell King update_mmu_cache(vma, address, pte); 36651a44e149SAndrea Arcangeli } else { 36661a44e149SAndrea Arcangeli /* 36671a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 36681a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 36691a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 36701a44e149SAndrea Arcangeli * with threads. 36711a44e149SAndrea Arcangeli */ 367230c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) 367361c77326SShaohua Li flush_tlb_fix_spurious_fault(vma, address); 36741a44e149SAndrea Arcangeli } 36758f4e2101SHugh Dickins unlock: 36768f4e2101SHugh Dickins pte_unmap_unlock(pte, ptl); 367783c54070SNick Piggin return 0; 36781da177e4SLinus Torvalds } 36791da177e4SLinus Torvalds 36801da177e4SLinus Torvalds /* 36811da177e4SLinus Torvalds * By the time we get here, we already hold the mm semaphore 36821da177e4SLinus Torvalds */ 368383c54070SNick Piggin int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3684d06063ccSLinus Torvalds unsigned long address, unsigned int flags) 36851da177e4SLinus Torvalds { 36861da177e4SLinus Torvalds pgd_t *pgd; 36871da177e4SLinus Torvalds pud_t *pud; 36881da177e4SLinus Torvalds pmd_t *pmd; 36891da177e4SLinus Torvalds pte_t *pte; 36901da177e4SLinus Torvalds 36911da177e4SLinus Torvalds __set_current_state(TASK_RUNNING); 36921da177e4SLinus Torvalds 3693f8891e5eSChristoph Lameter count_vm_event(PGFAULT); 3694456f998eSYing Han mem_cgroup_count_vm_event(mm, PGFAULT); 36951da177e4SLinus Torvalds 369634e55232SKAMEZAWA Hiroyuki /* do counter updates before entering really critical section. */ 369734e55232SKAMEZAWA Hiroyuki check_sync_rss_stat(current); 369834e55232SKAMEZAWA Hiroyuki 3699ac9b9c66SHugh Dickins if (unlikely(is_vm_hugetlb_page(vma))) 370030c9f3a9SLinus Torvalds return hugetlb_fault(mm, vma, address, flags); 37011da177e4SLinus Torvalds 37021f1d06c3SDavid Rientjes retry: 37031da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 37041da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, address); 37051da177e4SLinus Torvalds if (!pud) 3706c74df32cSHugh Dickins return VM_FAULT_OOM; 37071da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, address); 37081da177e4SLinus Torvalds if (!pmd) 3709c74df32cSHugh Dickins return VM_FAULT_OOM; 371071e3aac0SAndrea Arcangeli if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) { 371171e3aac0SAndrea Arcangeli if (!vma->vm_ops) 371271e3aac0SAndrea Arcangeli return do_huge_pmd_anonymous_page(mm, vma, address, 371371e3aac0SAndrea Arcangeli pmd, flags); 371471e3aac0SAndrea Arcangeli } else { 371571e3aac0SAndrea Arcangeli pmd_t orig_pmd = *pmd; 37161f1d06c3SDavid Rientjes int ret; 37171f1d06c3SDavid Rientjes 371871e3aac0SAndrea Arcangeli barrier(); 371971e3aac0SAndrea Arcangeli if (pmd_trans_huge(orig_pmd)) { 3720d10e63f2SMel Gorman if (pmd_numa(*pmd)) 37214daae3b4SMel Gorman return do_huge_pmd_numa_page(mm, vma, address, 3722d10e63f2SMel Gorman orig_pmd, pmd); 3723d10e63f2SMel Gorman 3724d10e63f2SMel Gorman if ((flags & FAULT_FLAG_WRITE) && !pmd_write(orig_pmd)) { 37251f1d06c3SDavid Rientjes ret = do_huge_pmd_wp_page(mm, vma, address, pmd, 37261f1d06c3SDavid Rientjes orig_pmd); 37271f1d06c3SDavid Rientjes /* 37281f1d06c3SDavid Rientjes * If COW results in an oom, the huge pmd will 37291f1d06c3SDavid Rientjes * have been split, so retry the fault on the 37301f1d06c3SDavid Rientjes * pte for a smaller charge. 37311f1d06c3SDavid Rientjes */ 37321f1d06c3SDavid Rientjes if (unlikely(ret & VM_FAULT_OOM)) 37331f1d06c3SDavid Rientjes goto retry; 37341f1d06c3SDavid Rientjes return ret; 37351f1d06c3SDavid Rientjes } 3736d10e63f2SMel Gorman 373771e3aac0SAndrea Arcangeli return 0; 373871e3aac0SAndrea Arcangeli } 373971e3aac0SAndrea Arcangeli } 374071e3aac0SAndrea Arcangeli 3741d10e63f2SMel Gorman if (pmd_numa(*pmd)) 3742d10e63f2SMel Gorman return do_pmd_numa_page(mm, vma, address, pmd); 3743d10e63f2SMel Gorman 374471e3aac0SAndrea Arcangeli /* 374571e3aac0SAndrea Arcangeli * Use __pte_alloc instead of pte_alloc_map, because we can't 374671e3aac0SAndrea Arcangeli * run pte_offset_map on the pmd, if an huge pmd could 374771e3aac0SAndrea Arcangeli * materialize from under us from a different thread. 374871e3aac0SAndrea Arcangeli */ 37494fd01770SMel Gorman if (unlikely(pmd_none(*pmd)) && 37504fd01770SMel Gorman unlikely(__pte_alloc(mm, vma, pmd, address))) 3751c74df32cSHugh Dickins return VM_FAULT_OOM; 375271e3aac0SAndrea Arcangeli /* if an huge pmd materialized from under us just retry later */ 375371e3aac0SAndrea Arcangeli if (unlikely(pmd_trans_huge(*pmd))) 375471e3aac0SAndrea Arcangeli return 0; 375571e3aac0SAndrea Arcangeli /* 375671e3aac0SAndrea Arcangeli * A regular pmd is established and it can't morph into a huge pmd 375771e3aac0SAndrea Arcangeli * from under us anymore at this point because we hold the mmap_sem 375871e3aac0SAndrea Arcangeli * read mode and khugepaged takes it in write mode. So now it's 375971e3aac0SAndrea Arcangeli * safe to run pte_offset_map(). 376071e3aac0SAndrea Arcangeli */ 376171e3aac0SAndrea Arcangeli pte = pte_offset_map(pmd, address); 37621da177e4SLinus Torvalds 376330c9f3a9SLinus Torvalds return handle_pte_fault(mm, vma, address, pte, pmd, flags); 37641da177e4SLinus Torvalds } 37651da177e4SLinus Torvalds 37661da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 37671da177e4SLinus Torvalds /* 37681da177e4SLinus Torvalds * Allocate page upper directory. 3769872fec16SHugh Dickins * We've already handled the fast-path in-line. 37701da177e4SLinus Torvalds */ 37711bb3630eSHugh Dickins int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 37721da177e4SLinus Torvalds { 3773c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 3774c74df32cSHugh Dickins if (!new) 37751bb3630eSHugh Dickins return -ENOMEM; 37761da177e4SLinus Torvalds 3777362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3778362a61adSNick Piggin 3779872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 37801bb3630eSHugh Dickins if (pgd_present(*pgd)) /* Another has populated it */ 37815e541973SBenjamin Herrenschmidt pud_free(mm, new); 37821bb3630eSHugh Dickins else 37831da177e4SLinus Torvalds pgd_populate(mm, pgd, new); 3784872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 37851bb3630eSHugh Dickins return 0; 37861da177e4SLinus Torvalds } 37871da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 37881da177e4SLinus Torvalds 37891da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 37901da177e4SLinus Torvalds /* 37911da177e4SLinus Torvalds * Allocate page middle directory. 3792872fec16SHugh Dickins * We've already handled the fast-path in-line. 37931da177e4SLinus Torvalds */ 37941bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 37951da177e4SLinus Torvalds { 3796c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 3797c74df32cSHugh Dickins if (!new) 37981bb3630eSHugh Dickins return -ENOMEM; 37991da177e4SLinus Torvalds 3800362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3801362a61adSNick Piggin 3802872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 38031da177e4SLinus Torvalds #ifndef __ARCH_HAS_4LEVEL_HACK 38041bb3630eSHugh Dickins if (pud_present(*pud)) /* Another has populated it */ 38055e541973SBenjamin Herrenschmidt pmd_free(mm, new); 38061bb3630eSHugh Dickins else 38071da177e4SLinus Torvalds pud_populate(mm, pud, new); 38081da177e4SLinus Torvalds #else 38091bb3630eSHugh Dickins if (pgd_present(*pud)) /* Another has populated it */ 38105e541973SBenjamin Herrenschmidt pmd_free(mm, new); 38111bb3630eSHugh Dickins else 38121da177e4SLinus Torvalds pgd_populate(mm, pud, new); 38131da177e4SLinus Torvalds #endif /* __ARCH_HAS_4LEVEL_HACK */ 3814872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 38151bb3630eSHugh Dickins return 0; 38161da177e4SLinus Torvalds } 38171da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 38181da177e4SLinus Torvalds 38191da177e4SLinus Torvalds int make_pages_present(unsigned long addr, unsigned long end) 38201da177e4SLinus Torvalds { 38211da177e4SLinus Torvalds int ret, len, write; 38221da177e4SLinus Torvalds struct vm_area_struct * vma; 38231da177e4SLinus Torvalds 38241da177e4SLinus Torvalds vma = find_vma(current->mm, addr); 38251da177e4SLinus Torvalds if (!vma) 3826a477097dSKOSAKI Motohiro return -ENOMEM; 38275ecfda04SMichel Lespinasse /* 38285ecfda04SMichel Lespinasse * We want to touch writable mappings with a write fault in order 38295ecfda04SMichel Lespinasse * to break COW, except for shared mappings because these don't COW 38305ecfda04SMichel Lespinasse * and we would not want to dirty them for nothing. 38315ecfda04SMichel Lespinasse */ 38325ecfda04SMichel Lespinasse write = (vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE; 38335bcb28b1SEric Sesterhenn BUG_ON(addr >= end); 38345bcb28b1SEric Sesterhenn BUG_ON(end > vma->vm_end); 383568e116a3SRolf Eike Beer len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE; 38361da177e4SLinus Torvalds ret = get_user_pages(current, current->mm, addr, 38371da177e4SLinus Torvalds len, write, 0, NULL, NULL); 3838c11d69d8SLee Schermerhorn if (ret < 0) 38391da177e4SLinus Torvalds return ret; 38409978ad58SLee Schermerhorn return ret == len ? 0 : -EFAULT; 38411da177e4SLinus Torvalds } 38421da177e4SLinus Torvalds 38431da177e4SLinus Torvalds #if !defined(__HAVE_ARCH_GATE_AREA) 38441da177e4SLinus Torvalds 38451da177e4SLinus Torvalds #if defined(AT_SYSINFO_EHDR) 38465ce7852cSAdrian Bunk static struct vm_area_struct gate_vma; 38471da177e4SLinus Torvalds 38481da177e4SLinus Torvalds static int __init gate_vma_init(void) 38491da177e4SLinus Torvalds { 38501da177e4SLinus Torvalds gate_vma.vm_mm = NULL; 38511da177e4SLinus Torvalds gate_vma.vm_start = FIXADDR_USER_START; 38521da177e4SLinus Torvalds gate_vma.vm_end = FIXADDR_USER_END; 3853b6558c4aSRoland McGrath gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; 3854b6558c4aSRoland McGrath gate_vma.vm_page_prot = __P101; 3855909af768SJason Baron 38561da177e4SLinus Torvalds return 0; 38571da177e4SLinus Torvalds } 38581da177e4SLinus Torvalds __initcall(gate_vma_init); 38591da177e4SLinus Torvalds #endif 38601da177e4SLinus Torvalds 386131db58b3SStephen Wilson struct vm_area_struct *get_gate_vma(struct mm_struct *mm) 38621da177e4SLinus Torvalds { 38631da177e4SLinus Torvalds #ifdef AT_SYSINFO_EHDR 38641da177e4SLinus Torvalds return &gate_vma; 38651da177e4SLinus Torvalds #else 38661da177e4SLinus Torvalds return NULL; 38671da177e4SLinus Torvalds #endif 38681da177e4SLinus Torvalds } 38691da177e4SLinus Torvalds 3870cae5d390SStephen Wilson int in_gate_area_no_mm(unsigned long addr) 38711da177e4SLinus Torvalds { 38721da177e4SLinus Torvalds #ifdef AT_SYSINFO_EHDR 38731da177e4SLinus Torvalds if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) 38741da177e4SLinus Torvalds return 1; 38751da177e4SLinus Torvalds #endif 38761da177e4SLinus Torvalds return 0; 38771da177e4SLinus Torvalds } 38781da177e4SLinus Torvalds 38791da177e4SLinus Torvalds #endif /* __HAVE_ARCH_GATE_AREA */ 38800ec76a11SDavid Howells 38811b36ba81SNamhyung Kim static int __follow_pte(struct mm_struct *mm, unsigned long address, 3882f8ad0f49SJohannes Weiner pte_t **ptepp, spinlock_t **ptlp) 3883f8ad0f49SJohannes Weiner { 3884f8ad0f49SJohannes Weiner pgd_t *pgd; 3885f8ad0f49SJohannes Weiner pud_t *pud; 3886f8ad0f49SJohannes Weiner pmd_t *pmd; 3887f8ad0f49SJohannes Weiner pte_t *ptep; 3888f8ad0f49SJohannes Weiner 3889f8ad0f49SJohannes Weiner pgd = pgd_offset(mm, address); 3890f8ad0f49SJohannes Weiner if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 3891f8ad0f49SJohannes Weiner goto out; 3892f8ad0f49SJohannes Weiner 3893f8ad0f49SJohannes Weiner pud = pud_offset(pgd, address); 3894f8ad0f49SJohannes Weiner if (pud_none(*pud) || unlikely(pud_bad(*pud))) 3895f8ad0f49SJohannes Weiner goto out; 3896f8ad0f49SJohannes Weiner 3897f8ad0f49SJohannes Weiner pmd = pmd_offset(pud, address); 3898f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 3899f8ad0f49SJohannes Weiner if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) 3900f8ad0f49SJohannes Weiner goto out; 3901f8ad0f49SJohannes Weiner 3902f8ad0f49SJohannes Weiner /* We cannot handle huge page PFN maps. Luckily they don't exist. */ 3903f8ad0f49SJohannes Weiner if (pmd_huge(*pmd)) 3904f8ad0f49SJohannes Weiner goto out; 3905f8ad0f49SJohannes Weiner 3906f8ad0f49SJohannes Weiner ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 3907f8ad0f49SJohannes Weiner if (!ptep) 3908f8ad0f49SJohannes Weiner goto out; 3909f8ad0f49SJohannes Weiner if (!pte_present(*ptep)) 3910f8ad0f49SJohannes Weiner goto unlock; 3911f8ad0f49SJohannes Weiner *ptepp = ptep; 3912f8ad0f49SJohannes Weiner return 0; 3913f8ad0f49SJohannes Weiner unlock: 3914f8ad0f49SJohannes Weiner pte_unmap_unlock(ptep, *ptlp); 3915f8ad0f49SJohannes Weiner out: 3916f8ad0f49SJohannes Weiner return -EINVAL; 3917f8ad0f49SJohannes Weiner } 3918f8ad0f49SJohannes Weiner 39191b36ba81SNamhyung Kim static inline int follow_pte(struct mm_struct *mm, unsigned long address, 39201b36ba81SNamhyung Kim pte_t **ptepp, spinlock_t **ptlp) 39211b36ba81SNamhyung Kim { 39221b36ba81SNamhyung Kim int res; 39231b36ba81SNamhyung Kim 39241b36ba81SNamhyung Kim /* (void) is needed to make gcc happy */ 39251b36ba81SNamhyung Kim (void) __cond_lock(*ptlp, 39261b36ba81SNamhyung Kim !(res = __follow_pte(mm, address, ptepp, ptlp))); 39271b36ba81SNamhyung Kim return res; 39281b36ba81SNamhyung Kim } 39291b36ba81SNamhyung Kim 39303b6748e2SJohannes Weiner /** 39313b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 39323b6748e2SJohannes Weiner * @vma: memory mapping 39333b6748e2SJohannes Weiner * @address: user virtual address 39343b6748e2SJohannes Weiner * @pfn: location to store found PFN 39353b6748e2SJohannes Weiner * 39363b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 39373b6748e2SJohannes Weiner * 39383b6748e2SJohannes Weiner * Returns zero and the pfn at @pfn on success, -ve otherwise. 39393b6748e2SJohannes Weiner */ 39403b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 39413b6748e2SJohannes Weiner unsigned long *pfn) 39423b6748e2SJohannes Weiner { 39433b6748e2SJohannes Weiner int ret = -EINVAL; 39443b6748e2SJohannes Weiner spinlock_t *ptl; 39453b6748e2SJohannes Weiner pte_t *ptep; 39463b6748e2SJohannes Weiner 39473b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 39483b6748e2SJohannes Weiner return ret; 39493b6748e2SJohannes Weiner 39503b6748e2SJohannes Weiner ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 39513b6748e2SJohannes Weiner if (ret) 39523b6748e2SJohannes Weiner return ret; 39533b6748e2SJohannes Weiner *pfn = pte_pfn(*ptep); 39543b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 39553b6748e2SJohannes Weiner return 0; 39563b6748e2SJohannes Weiner } 39573b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 39583b6748e2SJohannes Weiner 395928b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 3960d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 396128b2ee20SRik van Riel unsigned long address, unsigned int flags, 3962d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 396328b2ee20SRik van Riel { 396403668a4dSJohannes Weiner int ret = -EINVAL; 396528b2ee20SRik van Riel pte_t *ptep, pte; 396628b2ee20SRik van Riel spinlock_t *ptl; 396728b2ee20SRik van Riel 3968d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 3969d87fe660Svenkatesh.pallipadi@intel.com goto out; 397028b2ee20SRik van Riel 397103668a4dSJohannes Weiner if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 3972d87fe660Svenkatesh.pallipadi@intel.com goto out; 397328b2ee20SRik van Riel pte = *ptep; 397403668a4dSJohannes Weiner 397528b2ee20SRik van Riel if ((flags & FOLL_WRITE) && !pte_write(pte)) 397628b2ee20SRik van Riel goto unlock; 397728b2ee20SRik van Riel 397828b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 397903668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 398028b2ee20SRik van Riel 398103668a4dSJohannes Weiner ret = 0; 398228b2ee20SRik van Riel unlock: 398328b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 398428b2ee20SRik van Riel out: 3985d87fe660Svenkatesh.pallipadi@intel.com return ret; 398628b2ee20SRik van Riel } 398728b2ee20SRik van Riel 398828b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 398928b2ee20SRik van Riel void *buf, int len, int write) 399028b2ee20SRik van Riel { 399128b2ee20SRik van Riel resource_size_t phys_addr; 399228b2ee20SRik van Riel unsigned long prot = 0; 39932bc7273bSKOSAKI Motohiro void __iomem *maddr; 399428b2ee20SRik van Riel int offset = addr & (PAGE_SIZE-1); 399528b2ee20SRik van Riel 3996d87fe660Svenkatesh.pallipadi@intel.com if (follow_phys(vma, addr, write, &prot, &phys_addr)) 399728b2ee20SRik van Riel return -EINVAL; 399828b2ee20SRik van Riel 399928b2ee20SRik van Riel maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot); 400028b2ee20SRik van Riel if (write) 400128b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 400228b2ee20SRik van Riel else 400328b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 400428b2ee20SRik van Riel iounmap(maddr); 400528b2ee20SRik van Riel 400628b2ee20SRik van Riel return len; 400728b2ee20SRik van Riel } 400828b2ee20SRik van Riel #endif 400928b2ee20SRik van Riel 40100ec76a11SDavid Howells /* 4011206cb636SStephen Wilson * Access another process' address space as given in mm. If non-NULL, use the 4012206cb636SStephen Wilson * given task for page fault accounting. 40130ec76a11SDavid Howells */ 4014206cb636SStephen Wilson static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 4015206cb636SStephen Wilson unsigned long addr, void *buf, int len, int write) 40160ec76a11SDavid Howells { 40170ec76a11SDavid Howells struct vm_area_struct *vma; 40180ec76a11SDavid Howells void *old_buf = buf; 40190ec76a11SDavid Howells 40200ec76a11SDavid Howells down_read(&mm->mmap_sem); 4021183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 40220ec76a11SDavid Howells while (len) { 40230ec76a11SDavid Howells int bytes, ret, offset; 40240ec76a11SDavid Howells void *maddr; 402528b2ee20SRik van Riel struct page *page = NULL; 40260ec76a11SDavid Howells 40270ec76a11SDavid Howells ret = get_user_pages(tsk, mm, addr, 1, 40280ec76a11SDavid Howells write, 1, &page, &vma); 402928b2ee20SRik van Riel if (ret <= 0) { 403028b2ee20SRik van Riel /* 403128b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 403228b2ee20SRik van Riel * we can access using slightly different code. 403328b2ee20SRik van Riel */ 403428b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 403528b2ee20SRik van Riel vma = find_vma(mm, addr); 4036fe936dfcSMichael Ellerman if (!vma || vma->vm_start > addr) 40370ec76a11SDavid Howells break; 403828b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 403928b2ee20SRik van Riel ret = vma->vm_ops->access(vma, addr, buf, 404028b2ee20SRik van Riel len, write); 404128b2ee20SRik van Riel if (ret <= 0) 404228b2ee20SRik van Riel #endif 404328b2ee20SRik van Riel break; 404428b2ee20SRik van Riel bytes = ret; 404528b2ee20SRik van Riel } else { 40460ec76a11SDavid Howells bytes = len; 40470ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 40480ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 40490ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 40500ec76a11SDavid Howells 40510ec76a11SDavid Howells maddr = kmap(page); 40520ec76a11SDavid Howells if (write) { 40530ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 40540ec76a11SDavid Howells maddr + offset, buf, bytes); 40550ec76a11SDavid Howells set_page_dirty_lock(page); 40560ec76a11SDavid Howells } else { 40570ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 40580ec76a11SDavid Howells buf, maddr + offset, bytes); 40590ec76a11SDavid Howells } 40600ec76a11SDavid Howells kunmap(page); 40610ec76a11SDavid Howells page_cache_release(page); 406228b2ee20SRik van Riel } 40630ec76a11SDavid Howells len -= bytes; 40640ec76a11SDavid Howells buf += bytes; 40650ec76a11SDavid Howells addr += bytes; 40660ec76a11SDavid Howells } 40670ec76a11SDavid Howells up_read(&mm->mmap_sem); 40680ec76a11SDavid Howells 40690ec76a11SDavid Howells return buf - old_buf; 40700ec76a11SDavid Howells } 407103252919SAndi Kleen 40725ddd36b9SStephen Wilson /** 4073ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 40745ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 40755ddd36b9SStephen Wilson * @addr: start address to access 40765ddd36b9SStephen Wilson * @buf: source or destination buffer 40775ddd36b9SStephen Wilson * @len: number of bytes to transfer 40785ddd36b9SStephen Wilson * @write: whether the access is a write 40795ddd36b9SStephen Wilson * 40805ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 40815ddd36b9SStephen Wilson */ 40825ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 40835ddd36b9SStephen Wilson void *buf, int len, int write) 40845ddd36b9SStephen Wilson { 40855ddd36b9SStephen Wilson return __access_remote_vm(NULL, mm, addr, buf, len, write); 40865ddd36b9SStephen Wilson } 40875ddd36b9SStephen Wilson 408803252919SAndi Kleen /* 4089206cb636SStephen Wilson * Access another process' address space. 4090206cb636SStephen Wilson * Source/target buffer must be kernel space, 4091206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 4092206cb636SStephen Wilson */ 4093206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 4094206cb636SStephen Wilson void *buf, int len, int write) 4095206cb636SStephen Wilson { 4096206cb636SStephen Wilson struct mm_struct *mm; 4097206cb636SStephen Wilson int ret; 4098206cb636SStephen Wilson 4099206cb636SStephen Wilson mm = get_task_mm(tsk); 4100206cb636SStephen Wilson if (!mm) 4101206cb636SStephen Wilson return 0; 4102206cb636SStephen Wilson 4103206cb636SStephen Wilson ret = __access_remote_vm(tsk, mm, addr, buf, len, write); 4104206cb636SStephen Wilson mmput(mm); 4105206cb636SStephen Wilson 4106206cb636SStephen Wilson return ret; 4107206cb636SStephen Wilson } 4108206cb636SStephen Wilson 410903252919SAndi Kleen /* 411003252919SAndi Kleen * Print the name of a VMA. 411103252919SAndi Kleen */ 411203252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 411303252919SAndi Kleen { 411403252919SAndi Kleen struct mm_struct *mm = current->mm; 411503252919SAndi Kleen struct vm_area_struct *vma; 411603252919SAndi Kleen 4117e8bff74aSIngo Molnar /* 4118e8bff74aSIngo Molnar * Do not print if we are in atomic 4119e8bff74aSIngo Molnar * contexts (in exception stacks, etc.): 4120e8bff74aSIngo Molnar */ 4121e8bff74aSIngo Molnar if (preempt_count()) 4122e8bff74aSIngo Molnar return; 4123e8bff74aSIngo Molnar 412403252919SAndi Kleen down_read(&mm->mmap_sem); 412503252919SAndi Kleen vma = find_vma(mm, ip); 412603252919SAndi Kleen if (vma && vma->vm_file) { 412703252919SAndi Kleen struct file *f = vma->vm_file; 412803252919SAndi Kleen char *buf = (char *)__get_free_page(GFP_KERNEL); 412903252919SAndi Kleen if (buf) { 413003252919SAndi Kleen char *p, *s; 413103252919SAndi Kleen 4132cf28b486SJan Blunck p = d_path(&f->f_path, buf, PAGE_SIZE); 413303252919SAndi Kleen if (IS_ERR(p)) 413403252919SAndi Kleen p = "?"; 413503252919SAndi Kleen s = strrchr(p, '/'); 413603252919SAndi Kleen if (s) 413703252919SAndi Kleen p = s+1; 413803252919SAndi Kleen printk("%s%s[%lx+%lx]", prefix, p, 413903252919SAndi Kleen vma->vm_start, 414003252919SAndi Kleen vma->vm_end - vma->vm_start); 414103252919SAndi Kleen free_page((unsigned long)buf); 414203252919SAndi Kleen } 414303252919SAndi Kleen } 414451a07e50SJeff Liu up_read(&mm->mmap_sem); 414503252919SAndi Kleen } 41463ee1afa3SNick Piggin 41473ee1afa3SNick Piggin #ifdef CONFIG_PROVE_LOCKING 41483ee1afa3SNick Piggin void might_fault(void) 41493ee1afa3SNick Piggin { 415095156f00SPeter Zijlstra /* 415195156f00SPeter Zijlstra * Some code (nfs/sunrpc) uses socket ops on kernel memory while 415295156f00SPeter Zijlstra * holding the mmap_sem, this is safe because kernel memory doesn't 415395156f00SPeter Zijlstra * get paged out, therefore we'll never actually fault, and the 415495156f00SPeter Zijlstra * below annotations will generate false positives. 415595156f00SPeter Zijlstra */ 415695156f00SPeter Zijlstra if (segment_eq(get_fs(), KERNEL_DS)) 415795156f00SPeter Zijlstra return; 415895156f00SPeter Zijlstra 41593ee1afa3SNick Piggin might_sleep(); 41603ee1afa3SNick Piggin /* 41613ee1afa3SNick Piggin * it would be nicer only to annotate paths which are not under 41623ee1afa3SNick Piggin * pagefault_disable, however that requires a larger audit and 41633ee1afa3SNick Piggin * providing helpers like get_user_atomic. 41643ee1afa3SNick Piggin */ 41653ee1afa3SNick Piggin if (!in_atomic() && current->mm) 41663ee1afa3SNick Piggin might_lock_read(¤t->mm->mmap_sem); 41673ee1afa3SNick Piggin } 41683ee1afa3SNick Piggin EXPORT_SYMBOL(might_fault); 41693ee1afa3SNick Piggin #endif 417047ad8475SAndrea Arcangeli 417147ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 417247ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 417347ad8475SAndrea Arcangeli unsigned long addr, 417447ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 417547ad8475SAndrea Arcangeli { 417647ad8475SAndrea Arcangeli int i; 417747ad8475SAndrea Arcangeli struct page *p = page; 417847ad8475SAndrea Arcangeli 417947ad8475SAndrea Arcangeli might_sleep(); 418047ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; 418147ad8475SAndrea Arcangeli i++, p = mem_map_next(p, page, i)) { 418247ad8475SAndrea Arcangeli cond_resched(); 418347ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 418447ad8475SAndrea Arcangeli } 418547ad8475SAndrea Arcangeli } 418647ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 418747ad8475SAndrea Arcangeli unsigned long addr, unsigned int pages_per_huge_page) 418847ad8475SAndrea Arcangeli { 418947ad8475SAndrea Arcangeli int i; 419047ad8475SAndrea Arcangeli 419147ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 419247ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 419347ad8475SAndrea Arcangeli return; 419447ad8475SAndrea Arcangeli } 419547ad8475SAndrea Arcangeli 419647ad8475SAndrea Arcangeli might_sleep(); 419747ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 419847ad8475SAndrea Arcangeli cond_resched(); 419947ad8475SAndrea Arcangeli clear_user_highpage(page + i, addr + i * PAGE_SIZE); 420047ad8475SAndrea Arcangeli } 420147ad8475SAndrea Arcangeli } 420247ad8475SAndrea Arcangeli 420347ad8475SAndrea Arcangeli static void copy_user_gigantic_page(struct page *dst, struct page *src, 420447ad8475SAndrea Arcangeli unsigned long addr, 420547ad8475SAndrea Arcangeli struct vm_area_struct *vma, 420647ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 420747ad8475SAndrea Arcangeli { 420847ad8475SAndrea Arcangeli int i; 420947ad8475SAndrea Arcangeli struct page *dst_base = dst; 421047ad8475SAndrea Arcangeli struct page *src_base = src; 421147ad8475SAndrea Arcangeli 421247ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; ) { 421347ad8475SAndrea Arcangeli cond_resched(); 421447ad8475SAndrea Arcangeli copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); 421547ad8475SAndrea Arcangeli 421647ad8475SAndrea Arcangeli i++; 421747ad8475SAndrea Arcangeli dst = mem_map_next(dst, dst_base, i); 421847ad8475SAndrea Arcangeli src = mem_map_next(src, src_base, i); 421947ad8475SAndrea Arcangeli } 422047ad8475SAndrea Arcangeli } 422147ad8475SAndrea Arcangeli 422247ad8475SAndrea Arcangeli void copy_user_huge_page(struct page *dst, struct page *src, 422347ad8475SAndrea Arcangeli unsigned long addr, struct vm_area_struct *vma, 422447ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 422547ad8475SAndrea Arcangeli { 422647ad8475SAndrea Arcangeli int i; 422747ad8475SAndrea Arcangeli 422847ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 422947ad8475SAndrea Arcangeli copy_user_gigantic_page(dst, src, addr, vma, 423047ad8475SAndrea Arcangeli pages_per_huge_page); 423147ad8475SAndrea Arcangeli return; 423247ad8475SAndrea Arcangeli } 423347ad8475SAndrea Arcangeli 423447ad8475SAndrea Arcangeli might_sleep(); 423547ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 423647ad8475SAndrea Arcangeli cond_resched(); 423747ad8475SAndrea Arcangeli copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); 423847ad8475SAndrea Arcangeli } 423947ad8475SAndrea Arcangeli } 424047ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 4241