Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/mm/memory.c |
| 3 | * |
| 4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds |
| 5 | */ |
| 6 | |
| 7 | /* |
| 8 | * demand-loading started 01.12.91 - seems it is high on the list of |
| 9 | * things wanted, and it should be easy to implement. - Linus |
| 10 | */ |
| 11 | |
| 12 | /* |
| 13 | * Ok, demand-loading was easy, shared pages a little bit tricker. Shared |
| 14 | * pages started 02.12.91, seems to work. - Linus. |
| 15 | * |
| 16 | * Tested sharing by executing about 30 /bin/sh: under the old kernel it |
| 17 | * would have taken more than the 6M I have free, but it worked well as |
| 18 | * far as I could see. |
| 19 | * |
| 20 | * Also corrected some "invalidate()"s - I wasn't doing enough of them. |
| 21 | */ |
| 22 | |
| 23 | /* |
| 24 | * Real VM (paging to/from disk) started 18.12.91. Much more work and |
| 25 | * thought has to go into this. Oh, well.. |
| 26 | * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. |
| 27 | * Found it. Everything seems to work now. |
| 28 | * 20.12.91 - Ok, making the swap-device changeable like the root. |
| 29 | */ |
| 30 | |
| 31 | /* |
| 32 | * 05.04.94 - Multi-page memory management added for v1.1. |
| 33 | * Idea by Alex Bligh (alex@cconcepts.co.uk) |
| 34 | * |
| 35 | * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG |
| 36 | * (Gerhard.Wichert@pdb.siemens.de) |
| 37 | * |
| 38 | * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) |
| 39 | */ |
| 40 | |
| 41 | #include <linux/kernel_stat.h> |
| 42 | #include <linux/mm.h> |
| 43 | #include <linux/hugetlb.h> |
| 44 | #include <linux/mman.h> |
| 45 | #include <linux/swap.h> |
| 46 | #include <linux/highmem.h> |
| 47 | #include <linux/pagemap.h> |
| 48 | #include <linux/rmap.h> |
| 49 | #include <linux/module.h> |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 50 | #include <linux/delayacct.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 51 | #include <linux/init.h> |
Peter Zijlstra | edc79b2 | 2006-09-25 23:30:58 -0700 | [diff] [blame] | 52 | #include <linux/writeback.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 53 | #include <linux/memcontrol.h> |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 54 | #include <linux/mmu_notifier.h> |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 55 | #include <linux/kallsyms.h> |
| 56 | #include <linux/swapops.h> |
| 57 | #include <linux/elf.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | |
| 59 | #include <asm/pgalloc.h> |
| 60 | #include <asm/uaccess.h> |
| 61 | #include <asm/tlb.h> |
| 62 | #include <asm/tlbflush.h> |
| 63 | #include <asm/pgtable.h> |
| 64 | |
Jan Beulich | 42b7772 | 2008-07-23 21:27:10 -0700 | [diff] [blame] | 65 | #include "internal.h" |
| 66 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 67 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
| 69 | unsigned long max_mapnr; |
| 70 | struct page *mem_map; |
| 71 | |
| 72 | EXPORT_SYMBOL(max_mapnr); |
| 73 | EXPORT_SYMBOL(mem_map); |
| 74 | #endif |
| 75 | |
| 76 | unsigned long num_physpages; |
| 77 | /* |
| 78 | * A number of key systems in x86 including ioremap() rely on the assumption |
| 79 | * that high_memory defines the upper bound on direct map memory, then end |
| 80 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and |
| 81 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL |
| 82 | * and ZONE_HIGHMEM. |
| 83 | */ |
| 84 | void * high_memory; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 85 | |
| 86 | EXPORT_SYMBOL(num_physpages); |
| 87 | EXPORT_SYMBOL(high_memory); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 88 | |
Ingo Molnar | 32a9323 | 2008-02-06 22:39:44 +0100 | [diff] [blame] | 89 | /* |
| 90 | * Randomize the address space (stacks, mmaps, brk, etc.). |
| 91 | * |
| 92 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, |
| 93 | * as ancient (libc5 based) binaries can segfault. ) |
| 94 | */ |
| 95 | int randomize_va_space __read_mostly = |
| 96 | #ifdef CONFIG_COMPAT_BRK |
| 97 | 1; |
| 98 | #else |
| 99 | 2; |
| 100 | #endif |
Andi Kleen | a62eaf1 | 2006-02-16 23:41:58 +0100 | [diff] [blame] | 101 | |
| 102 | static int __init disable_randmaps(char *s) |
| 103 | { |
| 104 | randomize_va_space = 0; |
OGAWA Hirofumi | 9b41046 | 2006-03-31 02:30:33 -0800 | [diff] [blame] | 105 | return 1; |
Andi Kleen | a62eaf1 | 2006-02-16 23:41:58 +0100 | [diff] [blame] | 106 | } |
| 107 | __setup("norandmaps", disable_randmaps); |
| 108 | |
| 109 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 110 | /* |
| 111 | * If a p?d_bad entry is found while walking page tables, report |
| 112 | * the error, before resetting entry to p?d_none. Usually (but |
| 113 | * very seldom) called out from the p?d_none_or_clear_bad macros. |
| 114 | */ |
| 115 | |
| 116 | void pgd_clear_bad(pgd_t *pgd) |
| 117 | { |
| 118 | pgd_ERROR(*pgd); |
| 119 | pgd_clear(pgd); |
| 120 | } |
| 121 | |
| 122 | void pud_clear_bad(pud_t *pud) |
| 123 | { |
| 124 | pud_ERROR(*pud); |
| 125 | pud_clear(pud); |
| 126 | } |
| 127 | |
| 128 | void pmd_clear_bad(pmd_t *pmd) |
| 129 | { |
| 130 | pmd_ERROR(*pmd); |
| 131 | pmd_clear(pmd); |
| 132 | } |
| 133 | |
| 134 | /* |
| 135 | * Note: this doesn't free the actual pages themselves. That |
| 136 | * has been handled earlier when unmapping all the memory regions. |
| 137 | */ |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 138 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 139 | { |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 140 | pgtable_t token = pmd_pgtable(*pmd); |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 141 | pmd_clear(pmd); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 142 | pte_free_tlb(tlb, token); |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 143 | tlb->mm->nr_ptes--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | } |
| 145 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 146 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
| 147 | unsigned long addr, unsigned long end, |
| 148 | unsigned long floor, unsigned long ceiling) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | { |
| 150 | pmd_t *pmd; |
| 151 | unsigned long next; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 152 | unsigned long start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 154 | start = addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | pmd = pmd_offset(pud, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 156 | do { |
| 157 | next = pmd_addr_end(addr, end); |
| 158 | if (pmd_none_or_clear_bad(pmd)) |
| 159 | continue; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 160 | free_pte_range(tlb, pmd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | } while (pmd++, addr = next, addr != end); |
| 162 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 163 | start &= PUD_MASK; |
| 164 | if (start < floor) |
| 165 | return; |
| 166 | if (ceiling) { |
| 167 | ceiling &= PUD_MASK; |
| 168 | if (!ceiling) |
| 169 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | } |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 171 | if (end - 1 > ceiling - 1) |
| 172 | return; |
| 173 | |
| 174 | pmd = pmd_offset(pud, start); |
| 175 | pud_clear(pud); |
| 176 | pmd_free_tlb(tlb, pmd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | } |
| 178 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 179 | static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, |
| 180 | unsigned long addr, unsigned long end, |
| 181 | unsigned long floor, unsigned long ceiling) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | { |
| 183 | pud_t *pud; |
| 184 | unsigned long next; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 185 | unsigned long start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 187 | start = addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | pud = pud_offset(pgd, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 189 | do { |
| 190 | next = pud_addr_end(addr, end); |
| 191 | if (pud_none_or_clear_bad(pud)) |
| 192 | continue; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 193 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | } while (pud++, addr = next, addr != end); |
| 195 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 196 | start &= PGDIR_MASK; |
| 197 | if (start < floor) |
| 198 | return; |
| 199 | if (ceiling) { |
| 200 | ceiling &= PGDIR_MASK; |
| 201 | if (!ceiling) |
| 202 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 203 | } |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 204 | if (end - 1 > ceiling - 1) |
| 205 | return; |
| 206 | |
| 207 | pud = pud_offset(pgd, start); |
| 208 | pgd_clear(pgd); |
| 209 | pud_free_tlb(tlb, pud); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | } |
| 211 | |
| 212 | /* |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 213 | * This function frees user-level page tables of a process. |
| 214 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 215 | * Must be called with pagetable lock held. |
| 216 | */ |
Jan Beulich | 42b7772 | 2008-07-23 21:27:10 -0700 | [diff] [blame] | 217 | void free_pgd_range(struct mmu_gather *tlb, |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 218 | unsigned long addr, unsigned long end, |
| 219 | unsigned long floor, unsigned long ceiling) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 220 | { |
| 221 | pgd_t *pgd; |
| 222 | unsigned long next; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 223 | unsigned long start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 224 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 225 | /* |
| 226 | * The next few lines have given us lots of grief... |
| 227 | * |
| 228 | * Why are we testing PMD* at this top level? Because often |
| 229 | * there will be no work to do at all, and we'd prefer not to |
| 230 | * go all the way down to the bottom just to discover that. |
| 231 | * |
| 232 | * Why all these "- 1"s? Because 0 represents both the bottom |
| 233 | * of the address space and the top of it (using -1 for the |
| 234 | * top wouldn't help much: the masks would do the wrong thing). |
| 235 | * The rule is that addr 0 and floor 0 refer to the bottom of |
| 236 | * the address space, but end 0 and ceiling 0 refer to the top |
| 237 | * Comparisons need to use "end - 1" and "ceiling - 1" (though |
| 238 | * that end 0 case should be mythical). |
| 239 | * |
| 240 | * Wherever addr is brought up or ceiling brought down, we must |
| 241 | * be careful to reject "the opposite 0" before it confuses the |
| 242 | * subsequent tests. But what about where end is brought down |
| 243 | * by PMD_SIZE below? no, end can't go down to 0 there. |
| 244 | * |
| 245 | * Whereas we round start (addr) and ceiling down, by different |
| 246 | * masks at different levels, in order to test whether a table |
| 247 | * now has no other vmas using it, so can be freed, we don't |
| 248 | * bother to round floor or end up - the tests don't need that. |
| 249 | */ |
| 250 | |
| 251 | addr &= PMD_MASK; |
| 252 | if (addr < floor) { |
| 253 | addr += PMD_SIZE; |
| 254 | if (!addr) |
| 255 | return; |
| 256 | } |
| 257 | if (ceiling) { |
| 258 | ceiling &= PMD_MASK; |
| 259 | if (!ceiling) |
| 260 | return; |
| 261 | } |
| 262 | if (end - 1 > ceiling - 1) |
| 263 | end -= PMD_SIZE; |
| 264 | if (addr > end - 1) |
| 265 | return; |
| 266 | |
| 267 | start = addr; |
Jan Beulich | 42b7772 | 2008-07-23 21:27:10 -0700 | [diff] [blame] | 268 | pgd = pgd_offset(tlb->mm, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 269 | do { |
| 270 | next = pgd_addr_end(addr, end); |
| 271 | if (pgd_none_or_clear_bad(pgd)) |
| 272 | continue; |
Jan Beulich | 42b7772 | 2008-07-23 21:27:10 -0700 | [diff] [blame] | 273 | free_pud_range(tlb, pgd, addr, next, floor, ceiling); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 274 | } while (pgd++, addr = next, addr != end); |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 275 | } |
| 276 | |
Jan Beulich | 42b7772 | 2008-07-23 21:27:10 -0700 | [diff] [blame] | 277 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 278 | unsigned long floor, unsigned long ceiling) |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 279 | { |
| 280 | while (vma) { |
| 281 | struct vm_area_struct *next = vma->vm_next; |
| 282 | unsigned long addr = vma->vm_start; |
| 283 | |
Hugh Dickins | 8f4f8c1 | 2005-10-29 18:16:29 -0700 | [diff] [blame] | 284 | /* |
| 285 | * Hide vma from rmap and vmtruncate before freeing pgtables |
| 286 | */ |
| 287 | anon_vma_unlink(vma); |
| 288 | unlink_file_vma(vma); |
| 289 | |
David Gibson | 9da61ae | 2006-03-22 00:08:57 -0800 | [diff] [blame] | 290 | if (is_vm_hugetlb_page(vma)) { |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 291 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 292 | floor, next? next->vm_start: ceiling); |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 293 | } else { |
| 294 | /* |
| 295 | * Optimization: gather nearby vmas into one call down |
| 296 | */ |
| 297 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE |
David Gibson | 4866920 | 2006-03-22 00:08:58 -0800 | [diff] [blame] | 298 | && !is_vm_hugetlb_page(next)) { |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 299 | vma = next; |
| 300 | next = vma->vm_next; |
Hugh Dickins | 8f4f8c1 | 2005-10-29 18:16:29 -0700 | [diff] [blame] | 301 | anon_vma_unlink(vma); |
| 302 | unlink_file_vma(vma); |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 303 | } |
| 304 | free_pgd_range(tlb, addr, vma->vm_end, |
| 305 | floor, next? next->vm_start: ceiling); |
| 306 | } |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 307 | vma = next; |
| 308 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 309 | } |
| 310 | |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 311 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 312 | { |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 313 | pgtable_t new = pte_alloc_one(mm, address); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 314 | if (!new) |
| 315 | return -ENOMEM; |
| 316 | |
Nick Piggin | 362a61a | 2008-05-14 06:37:36 +0200 | [diff] [blame] | 317 | /* |
| 318 | * Ensure all pte setup (eg. pte page lock and page clearing) are |
| 319 | * visible before the pte is made visible to other CPUs by being |
| 320 | * put into page tables. |
| 321 | * |
| 322 | * The other side of the story is the pointer chasing in the page |
| 323 | * table walking code (when walking the page table without locking; |
| 324 | * ie. most of the time). Fortunately, these data accesses consist |
| 325 | * of a chain of data-dependent loads, meaning most CPUs (alpha |
| 326 | * being the notable exception) will already guarantee loads are |
| 327 | * seen in-order. See the alpha page table accessors for the |
| 328 | * smp_read_barrier_depends() barriers in page table walking code. |
| 329 | */ |
| 330 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ |
| 331 | |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 332 | spin_lock(&mm->page_table_lock); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 333 | if (!pmd_present(*pmd)) { /* Has another populated it ? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | mm->nr_ptes++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 335 | pmd_populate(mm, pmd, new); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 336 | new = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 337 | } |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 338 | spin_unlock(&mm->page_table_lock); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 339 | if (new) |
| 340 | pte_free(mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 341 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 342 | } |
| 343 | |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 344 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 345 | { |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 346 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
| 347 | if (!new) |
| 348 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 | |
Nick Piggin | 362a61a | 2008-05-14 06:37:36 +0200 | [diff] [blame] | 350 | smp_wmb(); /* See comment in __pte_alloc */ |
| 351 | |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 352 | spin_lock(&init_mm.page_table_lock); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 353 | if (!pmd_present(*pmd)) { /* Has another populated it ? */ |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 354 | pmd_populate_kernel(&init_mm, pmd, new); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 355 | new = NULL; |
| 356 | } |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 357 | spin_unlock(&init_mm.page_table_lock); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 358 | if (new) |
| 359 | pte_free_kernel(&init_mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 360 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 361 | } |
| 362 | |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 363 | static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss) |
| 364 | { |
| 365 | if (file_rss) |
| 366 | add_mm_counter(mm, file_rss, file_rss); |
| 367 | if (anon_rss) |
| 368 | add_mm_counter(mm, anon_rss, anon_rss); |
| 369 | } |
| 370 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 371 | /* |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 372 | * This function is called to print an error when a bad pte |
| 373 | * is found. For example, we might have a PFN-mapped pte in |
| 374 | * a region that doesn't allow it. |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 375 | * |
| 376 | * The calling function must still handle the error. |
| 377 | */ |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 378 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
| 379 | pte_t pte, struct page *page) |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 380 | { |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 381 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
| 382 | pud_t *pud = pud_offset(pgd, addr); |
| 383 | pmd_t *pmd = pmd_offset(pud, addr); |
| 384 | struct address_space *mapping; |
| 385 | pgoff_t index; |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 386 | static unsigned long resume; |
| 387 | static unsigned long nr_shown; |
| 388 | static unsigned long nr_unshown; |
| 389 | |
| 390 | /* |
| 391 | * Allow a burst of 60 reports, then keep quiet for that minute; |
| 392 | * or allow a steady drip of one report per second. |
| 393 | */ |
| 394 | if (nr_shown == 60) { |
| 395 | if (time_before(jiffies, resume)) { |
| 396 | nr_unshown++; |
| 397 | return; |
| 398 | } |
| 399 | if (nr_unshown) { |
Hugh Dickins | 1e9e636 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 400 | printk(KERN_ALERT |
| 401 | "BUG: Bad page map: %lu messages suppressed\n", |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 402 | nr_unshown); |
| 403 | nr_unshown = 0; |
| 404 | } |
| 405 | nr_shown = 0; |
| 406 | } |
| 407 | if (nr_shown++ == 0) |
| 408 | resume = jiffies + 60 * HZ; |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 409 | |
| 410 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; |
| 411 | index = linear_page_index(vma, addr); |
| 412 | |
Hugh Dickins | 1e9e636 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 413 | printk(KERN_ALERT |
| 414 | "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 415 | current->comm, |
| 416 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); |
| 417 | if (page) { |
Hugh Dickins | 1e9e636 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 418 | printk(KERN_ALERT |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 419 | "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n", |
| 420 | page, (void *)page->flags, page_count(page), |
| 421 | page_mapcount(page), page->mapping, page->index); |
| 422 | } |
Hugh Dickins | 1e9e636 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 423 | printk(KERN_ALERT |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 424 | "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
| 425 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); |
| 426 | /* |
| 427 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y |
| 428 | */ |
| 429 | if (vma->vm_ops) |
Hugh Dickins | 1e9e636 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 430 | print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n", |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 431 | (unsigned long)vma->vm_ops->fault); |
| 432 | if (vma->vm_file && vma->vm_file->f_op) |
Hugh Dickins | 1e9e636 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 433 | print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n", |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 434 | (unsigned long)vma->vm_file->f_op->mmap); |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 435 | dump_stack(); |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 436 | add_taint(TAINT_BAD_PAGE); |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 437 | } |
| 438 | |
Linus Torvalds | 6712117 | 2005-12-11 20:38:17 -0800 | [diff] [blame] | 439 | static inline int is_cow_mapping(unsigned int flags) |
| 440 | { |
| 441 | return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; |
| 442 | } |
| 443 | |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 444 | /* |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 445 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 446 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 447 | * "Special" mappings do not wish to be associated with a "struct page" (either |
| 448 | * it doesn't exist, or it exists but they don't want to touch it). In this |
| 449 | * case, NULL is returned here. "Normal" mappings do have a struct page. |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 450 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 451 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
| 452 | * pte bit, in which case this function is trivial. Secondly, an architecture |
| 453 | * may not have a spare pte bit, which requires a more complicated scheme, |
| 454 | * described below. |
| 455 | * |
| 456 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a |
| 457 | * special mapping (even if there are underlying and valid "struct pages"). |
| 458 | * COWed pages of a VM_PFNMAP are always normal. |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 459 | * |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 460 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
| 461 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 462 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
| 463 | * mapping will always honor the rule |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 464 | * |
| 465 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) |
| 466 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 467 | * And for normal mappings this is false. |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 468 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 469 | * This restricts such mappings to be a linear translation from virtual address |
| 470 | * to pfn. To get around this restriction, we allow arbitrary mappings so long |
| 471 | * as the vma is not a COW mapping; in that case, we know that all ptes are |
| 472 | * special (because none can have been COWed). |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 473 | * |
| 474 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 475 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
| 476 | * |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 477 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct |
| 478 | * page" backing, however the difference is that _all_ pages with a struct |
| 479 | * page (that is, those where pfn_valid is true) are refcounted and considered |
| 480 | * normal pages by the VM. The disadvantage is that pages are refcounted |
| 481 | * (which can be slower and simply not an option for some PFNMAP users). The |
| 482 | * advantage is that we don't have to follow the strict linearity rule of |
| 483 | * PFNMAP mappings in order to support COWable mappings. |
| 484 | * |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 485 | */ |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 486 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
| 487 | # define HAVE_PTE_SPECIAL 1 |
| 488 | #else |
| 489 | # define HAVE_PTE_SPECIAL 0 |
| 490 | #endif |
| 491 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
| 492 | pte_t pte) |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 493 | { |
Hugh Dickins | 22b31ee | 2009-01-06 14:40:09 -0800 | [diff] [blame] | 494 | unsigned long pfn = pte_pfn(pte); |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 495 | |
| 496 | if (HAVE_PTE_SPECIAL) { |
Hugh Dickins | 22b31ee | 2009-01-06 14:40:09 -0800 | [diff] [blame] | 497 | if (likely(!pte_special(pte))) |
| 498 | goto check_pfn; |
| 499 | if (!(vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))) |
| 500 | print_bad_pte(vma, addr, pte, NULL); |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 501 | return NULL; |
| 502 | } |
| 503 | |
| 504 | /* !HAVE_PTE_SPECIAL case follows: */ |
| 505 | |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 506 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
| 507 | if (vma->vm_flags & VM_MIXEDMAP) { |
| 508 | if (!pfn_valid(pfn)) |
| 509 | return NULL; |
| 510 | goto out; |
| 511 | } else { |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 512 | unsigned long off; |
| 513 | off = (addr - vma->vm_start) >> PAGE_SHIFT; |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 514 | if (pfn == vma->vm_pgoff + off) |
| 515 | return NULL; |
| 516 | if (!is_cow_mapping(vma->vm_flags)) |
| 517 | return NULL; |
| 518 | } |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 519 | } |
| 520 | |
Hugh Dickins | 22b31ee | 2009-01-06 14:40:09 -0800 | [diff] [blame] | 521 | check_pfn: |
| 522 | if (unlikely(pfn > highest_memmap_pfn)) { |
| 523 | print_bad_pte(vma, addr, pte, NULL); |
| 524 | return NULL; |
| 525 | } |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 526 | |
| 527 | /* |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 528 | * NOTE! We still have PageReserved() pages in the page tables. |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 529 | * eg. VDSO mappings can cause them to exist. |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 530 | */ |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 531 | out: |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 532 | return pfn_to_page(pfn); |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 533 | } |
| 534 | |
| 535 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | * copy one vm_area from one task to the other. Assumes the page tables |
| 537 | * already present in the new task to be cleared in the whole range |
| 538 | * covered by this vma. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 539 | */ |
| 540 | |
Hugh Dickins | 8c10376 | 2005-10-29 18:16:13 -0700 | [diff] [blame] | 541 | static inline void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 542 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 543 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
Hugh Dickins | 8c10376 | 2005-10-29 18:16:13 -0700 | [diff] [blame] | 544 | unsigned long addr, int *rss) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 545 | { |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 546 | unsigned long vm_flags = vma->vm_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 547 | pte_t pte = *src_pte; |
| 548 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 549 | |
| 550 | /* pte contains position in swap or file, so copy. */ |
| 551 | if (unlikely(!pte_present(pte))) { |
| 552 | if (!pte_file(pte)) { |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 553 | swp_entry_t entry = pte_to_swp_entry(pte); |
| 554 | |
| 555 | swap_duplicate(entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 556 | /* make sure dst_mm is on swapoff's mmlist. */ |
| 557 | if (unlikely(list_empty(&dst_mm->mmlist))) { |
| 558 | spin_lock(&mmlist_lock); |
Hugh Dickins | f412ac0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 559 | if (list_empty(&dst_mm->mmlist)) |
| 560 | list_add(&dst_mm->mmlist, |
| 561 | &src_mm->mmlist); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | spin_unlock(&mmlist_lock); |
| 563 | } |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 564 | if (is_write_migration_entry(entry) && |
| 565 | is_cow_mapping(vm_flags)) { |
| 566 | /* |
| 567 | * COW mappings require pages in both parent |
| 568 | * and child to be set to read. |
| 569 | */ |
| 570 | make_migration_entry_read(&entry); |
| 571 | pte = swp_entry_to_pte(entry); |
| 572 | set_pte_at(src_mm, addr, src_pte, pte); |
| 573 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | } |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 575 | goto out_set_pte; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 576 | } |
| 577 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 578 | /* |
| 579 | * If it's a COW mapping, write protect it both |
| 580 | * in the parent and the child |
| 581 | */ |
Linus Torvalds | 6712117 | 2005-12-11 20:38:17 -0800 | [diff] [blame] | 582 | if (is_cow_mapping(vm_flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 583 | ptep_set_wrprotect(src_mm, addr, src_pte); |
Zachary Amsden | 3dc9079 | 2006-09-30 23:29:30 -0700 | [diff] [blame] | 584 | pte = pte_wrprotect(pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 585 | } |
| 586 | |
| 587 | /* |
| 588 | * If it's a shared mapping, mark it clean in |
| 589 | * the child |
| 590 | */ |
| 591 | if (vm_flags & VM_SHARED) |
| 592 | pte = pte_mkclean(pte); |
| 593 | pte = pte_mkold(pte); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 594 | |
| 595 | page = vm_normal_page(vma, addr, pte); |
| 596 | if (page) { |
| 597 | get_page(page); |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 598 | page_dup_rmap(page, vma, addr); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 599 | rss[!!PageAnon(page)]++; |
| 600 | } |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 601 | |
| 602 | out_set_pte: |
| 603 | set_pte_at(dst_mm, addr, dst_pte, pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 604 | } |
| 605 | |
| 606 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
| 607 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
| 608 | unsigned long addr, unsigned long end) |
| 609 | { |
| 610 | pte_t *src_pte, *dst_pte; |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 611 | spinlock_t *src_ptl, *dst_ptl; |
Hugh Dickins | e040f21 | 2005-10-29 18:15:53 -0700 | [diff] [blame] | 612 | int progress = 0; |
Hugh Dickins | 8c10376 | 2005-10-29 18:16:13 -0700 | [diff] [blame] | 613 | int rss[2]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 614 | |
| 615 | again: |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 616 | rss[1] = rss[0] = 0; |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 617 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 618 | if (!dst_pte) |
| 619 | return -ENOMEM; |
| 620 | src_pte = pte_offset_map_nested(src_pmd, addr); |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 621 | src_ptl = pte_lockptr(src_mm, src_pmd); |
Ingo Molnar | f20dc5f | 2006-07-03 00:25:08 -0700 | [diff] [blame] | 622 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 623 | arch_enter_lazy_mmu_mode(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 624 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | do { |
| 626 | /* |
| 627 | * We are holding two locks at this point - either of them |
| 628 | * could generate latencies in another task on another CPU. |
| 629 | */ |
Hugh Dickins | e040f21 | 2005-10-29 18:15:53 -0700 | [diff] [blame] | 630 | if (progress >= 32) { |
| 631 | progress = 0; |
| 632 | if (need_resched() || |
Nick Piggin | 95c354f | 2008-01-30 13:31:20 +0100 | [diff] [blame] | 633 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
Hugh Dickins | e040f21 | 2005-10-29 18:15:53 -0700 | [diff] [blame] | 634 | break; |
| 635 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | if (pte_none(*src_pte)) { |
| 637 | progress++; |
| 638 | continue; |
| 639 | } |
Hugh Dickins | 8c10376 | 2005-10-29 18:16:13 -0700 | [diff] [blame] | 640 | copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | progress += 8; |
| 642 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 643 | |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 644 | arch_leave_lazy_mmu_mode(); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 645 | spin_unlock(src_ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 646 | pte_unmap_nested(src_pte - 1); |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 647 | add_mm_rss(dst_mm, rss[0], rss[1]); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 648 | pte_unmap_unlock(dst_pte - 1, dst_ptl); |
| 649 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 650 | if (addr != end) |
| 651 | goto again; |
| 652 | return 0; |
| 653 | } |
| 654 | |
| 655 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
| 656 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, |
| 657 | unsigned long addr, unsigned long end) |
| 658 | { |
| 659 | pmd_t *src_pmd, *dst_pmd; |
| 660 | unsigned long next; |
| 661 | |
| 662 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); |
| 663 | if (!dst_pmd) |
| 664 | return -ENOMEM; |
| 665 | src_pmd = pmd_offset(src_pud, addr); |
| 666 | do { |
| 667 | next = pmd_addr_end(addr, end); |
| 668 | if (pmd_none_or_clear_bad(src_pmd)) |
| 669 | continue; |
| 670 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, |
| 671 | vma, addr, next)) |
| 672 | return -ENOMEM; |
| 673 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); |
| 674 | return 0; |
| 675 | } |
| 676 | |
| 677 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
| 678 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, |
| 679 | unsigned long addr, unsigned long end) |
| 680 | { |
| 681 | pud_t *src_pud, *dst_pud; |
| 682 | unsigned long next; |
| 683 | |
| 684 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); |
| 685 | if (!dst_pud) |
| 686 | return -ENOMEM; |
| 687 | src_pud = pud_offset(src_pgd, addr); |
| 688 | do { |
| 689 | next = pud_addr_end(addr, end); |
| 690 | if (pud_none_or_clear_bad(src_pud)) |
| 691 | continue; |
| 692 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, |
| 693 | vma, addr, next)) |
| 694 | return -ENOMEM; |
| 695 | } while (dst_pud++, src_pud++, addr = next, addr != end); |
| 696 | return 0; |
| 697 | } |
| 698 | |
| 699 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
| 700 | struct vm_area_struct *vma) |
| 701 | { |
| 702 | pgd_t *src_pgd, *dst_pgd; |
| 703 | unsigned long next; |
| 704 | unsigned long addr = vma->vm_start; |
| 705 | unsigned long end = vma->vm_end; |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 706 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 707 | |
Nick Piggin | d992895 | 2005-08-28 16:49:11 +1000 | [diff] [blame] | 708 | /* |
| 709 | * Don't copy ptes where a page fault will fill them correctly. |
| 710 | * Fork becomes much lighter when there are big shared or private |
| 711 | * readonly mappings. The tradeoff is that copy_page_range is more |
| 712 | * efficient than faulting. |
| 713 | */ |
Linus Torvalds | 4d7672b | 2005-12-16 10:21:23 -0800 | [diff] [blame] | 714 | if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) { |
Nick Piggin | d992895 | 2005-08-28 16:49:11 +1000 | [diff] [blame] | 715 | if (!vma->anon_vma) |
| 716 | return 0; |
| 717 | } |
| 718 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 719 | if (is_vm_hugetlb_page(vma)) |
| 720 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); |
| 721 | |
venkatesh.pallipadi@intel.com | 34801ba | 2008-12-19 13:47:29 -0800 | [diff] [blame] | 722 | if (unlikely(is_pfn_mapping(vma))) { |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 723 | /* |
| 724 | * We do not free on error cases below as remove_vma |
| 725 | * gets called on error from higher level routine |
| 726 | */ |
| 727 | ret = track_pfn_vma_copy(vma); |
| 728 | if (ret) |
| 729 | return ret; |
| 730 | } |
| 731 | |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 732 | /* |
| 733 | * We need to invalidate the secondary MMU mappings only when |
| 734 | * there could be a permission downgrade on the ptes of the |
| 735 | * parent mm. And a permission downgrade will only happen if |
| 736 | * is_cow_mapping() returns true. |
| 737 | */ |
| 738 | if (is_cow_mapping(vma->vm_flags)) |
| 739 | mmu_notifier_invalidate_range_start(src_mm, addr, end); |
| 740 | |
| 741 | ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | dst_pgd = pgd_offset(dst_mm, addr); |
| 743 | src_pgd = pgd_offset(src_mm, addr); |
| 744 | do { |
| 745 | next = pgd_addr_end(addr, end); |
| 746 | if (pgd_none_or_clear_bad(src_pgd)) |
| 747 | continue; |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 748 | if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, |
| 749 | vma, addr, next))) { |
| 750 | ret = -ENOMEM; |
| 751 | break; |
| 752 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 753 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 754 | |
| 755 | if (is_cow_mapping(vma->vm_flags)) |
| 756 | mmu_notifier_invalidate_range_end(src_mm, |
| 757 | vma->vm_start, end); |
| 758 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 759 | } |
| 760 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 761 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 762 | struct vm_area_struct *vma, pmd_t *pmd, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 763 | unsigned long addr, unsigned long end, |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 764 | long *zap_work, struct zap_details *details) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 765 | { |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 766 | struct mm_struct *mm = tlb->mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | pte_t *pte; |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 768 | spinlock_t *ptl; |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 769 | int file_rss = 0; |
| 770 | int anon_rss = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 771 | |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 772 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 773 | arch_enter_lazy_mmu_mode(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 774 | do { |
| 775 | pte_t ptent = *pte; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 776 | if (pte_none(ptent)) { |
| 777 | (*zap_work)--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 778 | continue; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 779 | } |
Hugh Dickins | 6f5e6b9 | 2006-03-16 23:04:09 -0800 | [diff] [blame] | 780 | |
| 781 | (*zap_work) -= PAGE_SIZE; |
| 782 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 783 | if (pte_present(ptent)) { |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 784 | struct page *page; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 785 | |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 786 | page = vm_normal_page(vma, addr, ptent); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 787 | if (unlikely(details) && page) { |
| 788 | /* |
| 789 | * unmap_shared_mapping_pages() wants to |
| 790 | * invalidate cache without truncating: |
| 791 | * unmap shared but keep private pages. |
| 792 | */ |
| 793 | if (details->check_mapping && |
| 794 | details->check_mapping != page->mapping) |
| 795 | continue; |
| 796 | /* |
| 797 | * Each page->index must be checked when |
| 798 | * invalidating or truncating nonlinear. |
| 799 | */ |
| 800 | if (details->nonlinear_vma && |
| 801 | (page->index < details->first_index || |
| 802 | page->index > details->last_index)) |
| 803 | continue; |
| 804 | } |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 805 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
Zachary Amsden | a600388 | 2005-09-03 15:55:04 -0700 | [diff] [blame] | 806 | tlb->fullmm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 807 | tlb_remove_tlb_entry(tlb, pte, addr); |
| 808 | if (unlikely(!page)) |
| 809 | continue; |
| 810 | if (unlikely(details) && details->nonlinear_vma |
| 811 | && linear_page_index(details->nonlinear_vma, |
| 812 | addr) != page->index) |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 813 | set_pte_at(mm, addr, pte, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 814 | pgoff_to_pte(page->index)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 815 | if (PageAnon(page)) |
Hugh Dickins | 86d912f | 2005-10-29 18:16:14 -0700 | [diff] [blame] | 816 | anon_rss--; |
Hugh Dickins | 6237bcd | 2005-10-29 18:15:54 -0700 | [diff] [blame] | 817 | else { |
| 818 | if (pte_dirty(ptent)) |
| 819 | set_page_dirty(page); |
Johannes Weiner | 4917e5d | 2009-01-06 14:39:17 -0800 | [diff] [blame] | 820 | if (pte_young(ptent) && |
| 821 | likely(!VM_SequentialReadHint(vma))) |
Nick Piggin | bf3f3bc | 2009-01-06 14:38:55 -0800 | [diff] [blame] | 822 | mark_page_accessed(page); |
Hugh Dickins | 86d912f | 2005-10-29 18:16:14 -0700 | [diff] [blame] | 823 | file_rss--; |
Hugh Dickins | 6237bcd | 2005-10-29 18:15:54 -0700 | [diff] [blame] | 824 | } |
Hugh Dickins | edc315f | 2009-01-06 14:40:11 -0800 | [diff] [blame] | 825 | page_remove_rmap(page); |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 826 | if (unlikely(page_mapcount(page) < 0)) |
| 827 | print_bad_pte(vma, addr, ptent, page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 828 | tlb_remove_page(tlb, page); |
| 829 | continue; |
| 830 | } |
| 831 | /* |
| 832 | * If details->check_mapping, we leave swap entries; |
| 833 | * if details->nonlinear_vma, we leave file entries. |
| 834 | */ |
| 835 | if (unlikely(details)) |
| 836 | continue; |
Hugh Dickins | 2509ef2 | 2009-01-06 14:40:10 -0800 | [diff] [blame] | 837 | if (pte_file(ptent)) { |
| 838 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) |
| 839 | print_bad_pte(vma, addr, ptent, NULL); |
| 840 | } else if |
| 841 | (unlikely(!free_swap_and_cache(pte_to_swp_entry(ptent)))) |
| 842 | print_bad_pte(vma, addr, ptent, NULL); |
Zachary Amsden | 9888a1c | 2006-09-30 23:29:31 -0700 | [diff] [blame] | 843 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 844 | } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0)); |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 845 | |
Hugh Dickins | 86d912f | 2005-10-29 18:16:14 -0700 | [diff] [blame] | 846 | add_mm_rss(mm, file_rss, anon_rss); |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 847 | arch_leave_lazy_mmu_mode(); |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 848 | pte_unmap_unlock(pte - 1, ptl); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 849 | |
| 850 | return addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 851 | } |
| 852 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 853 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 854 | struct vm_area_struct *vma, pud_t *pud, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 855 | unsigned long addr, unsigned long end, |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 856 | long *zap_work, struct zap_details *details) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 857 | { |
| 858 | pmd_t *pmd; |
| 859 | unsigned long next; |
| 860 | |
| 861 | pmd = pmd_offset(pud, addr); |
| 862 | do { |
| 863 | next = pmd_addr_end(addr, end); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 864 | if (pmd_none_or_clear_bad(pmd)) { |
| 865 | (*zap_work)--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 866 | continue; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 867 | } |
| 868 | next = zap_pte_range(tlb, vma, pmd, addr, next, |
| 869 | zap_work, details); |
| 870 | } while (pmd++, addr = next, (addr != end && *zap_work > 0)); |
| 871 | |
| 872 | return addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 873 | } |
| 874 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 875 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 876 | struct vm_area_struct *vma, pgd_t *pgd, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 877 | unsigned long addr, unsigned long end, |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 878 | long *zap_work, struct zap_details *details) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 879 | { |
| 880 | pud_t *pud; |
| 881 | unsigned long next; |
| 882 | |
| 883 | pud = pud_offset(pgd, addr); |
| 884 | do { |
| 885 | next = pud_addr_end(addr, end); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 886 | if (pud_none_or_clear_bad(pud)) { |
| 887 | (*zap_work)--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 888 | continue; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 889 | } |
| 890 | next = zap_pmd_range(tlb, vma, pud, addr, next, |
| 891 | zap_work, details); |
| 892 | } while (pud++, addr = next, (addr != end && *zap_work > 0)); |
| 893 | |
| 894 | return addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 895 | } |
| 896 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 897 | static unsigned long unmap_page_range(struct mmu_gather *tlb, |
| 898 | struct vm_area_struct *vma, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 899 | unsigned long addr, unsigned long end, |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 900 | long *zap_work, struct zap_details *details) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 901 | { |
| 902 | pgd_t *pgd; |
| 903 | unsigned long next; |
| 904 | |
| 905 | if (details && !details->check_mapping && !details->nonlinear_vma) |
| 906 | details = NULL; |
| 907 | |
| 908 | BUG_ON(addr >= end); |
| 909 | tlb_start_vma(tlb, vma); |
| 910 | pgd = pgd_offset(vma->vm_mm, addr); |
| 911 | do { |
| 912 | next = pgd_addr_end(addr, end); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 913 | if (pgd_none_or_clear_bad(pgd)) { |
| 914 | (*zap_work)--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 915 | continue; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 916 | } |
| 917 | next = zap_pud_range(tlb, vma, pgd, addr, next, |
| 918 | zap_work, details); |
| 919 | } while (pgd++, addr = next, (addr != end && *zap_work > 0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 920 | tlb_end_vma(tlb, vma); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 921 | |
| 922 | return addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 923 | } |
| 924 | |
| 925 | #ifdef CONFIG_PREEMPT |
| 926 | # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE) |
| 927 | #else |
| 928 | /* No preempt: go for improved straight-line efficiency */ |
| 929 | # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE) |
| 930 | #endif |
| 931 | |
| 932 | /** |
| 933 | * unmap_vmas - unmap a range of memory covered by a list of vma's |
| 934 | * @tlbp: address of the caller's struct mmu_gather |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 935 | * @vma: the starting vma |
| 936 | * @start_addr: virtual address at which to start unmapping |
| 937 | * @end_addr: virtual address at which to end unmapping |
| 938 | * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here |
| 939 | * @details: details of nonlinear truncation or shared cache invalidation |
| 940 | * |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 941 | * Returns the end address of the unmapping (restart addr if interrupted). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 942 | * |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 943 | * Unmap all pages in the vma list. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 944 | * |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 945 | * We aim to not hold locks for too long (for scheduling latency reasons). |
| 946 | * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 947 | * return the ending mmu_gather to the caller. |
| 948 | * |
| 949 | * Only addresses between `start' and `end' will be unmapped. |
| 950 | * |
| 951 | * The VMA list must be sorted in ascending virtual address order. |
| 952 | * |
| 953 | * unmap_vmas() assumes that the caller will flush the whole unmapped address |
| 954 | * range after unmap_vmas() returns. So the only responsibility here is to |
| 955 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() |
| 956 | * drops the lock and schedules. |
| 957 | */ |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 958 | unsigned long unmap_vmas(struct mmu_gather **tlbp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 959 | struct vm_area_struct *vma, unsigned long start_addr, |
| 960 | unsigned long end_addr, unsigned long *nr_accounted, |
| 961 | struct zap_details *details) |
| 962 | { |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 963 | long zap_work = ZAP_BLOCK_SIZE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 964 | unsigned long tlb_start = 0; /* For tlb_finish_mmu */ |
| 965 | int tlb_start_valid = 0; |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 966 | unsigned long start = start_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 967 | spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL; |
Hugh Dickins | 4d6ddfa | 2005-10-29 18:16:02 -0700 | [diff] [blame] | 968 | int fullmm = (*tlbp)->fullmm; |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 969 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 970 | |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 971 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 972 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | unsigned long end; |
| 974 | |
| 975 | start = max(vma->vm_start, start_addr); |
| 976 | if (start >= vma->vm_end) |
| 977 | continue; |
| 978 | end = min(vma->vm_end, end_addr); |
| 979 | if (end <= vma->vm_start) |
| 980 | continue; |
| 981 | |
| 982 | if (vma->vm_flags & VM_ACCOUNT) |
| 983 | *nr_accounted += (end - start) >> PAGE_SHIFT; |
| 984 | |
venkatesh.pallipadi@intel.com | 34801ba | 2008-12-19 13:47:29 -0800 | [diff] [blame] | 985 | if (unlikely(is_pfn_mapping(vma))) |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 986 | untrack_pfn_vma(vma, 0, 0); |
| 987 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 988 | while (start != end) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 989 | if (!tlb_start_valid) { |
| 990 | tlb_start = start; |
| 991 | tlb_start_valid = 1; |
| 992 | } |
| 993 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 994 | if (unlikely(is_vm_hugetlb_page(vma))) { |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 995 | /* |
| 996 | * It is undesirable to test vma->vm_file as it |
| 997 | * should be non-null for valid hugetlb area. |
| 998 | * However, vm_file will be NULL in the error |
| 999 | * cleanup path of do_mmap_pgoff. When |
| 1000 | * hugetlbfs ->mmap method fails, |
| 1001 | * do_mmap_pgoff() nullifies vma->vm_file |
| 1002 | * before calling this function to clean up. |
| 1003 | * Since no pte has actually been setup, it is |
| 1004 | * safe to do nothing in this case. |
| 1005 | */ |
| 1006 | if (vma->vm_file) { |
| 1007 | unmap_hugepage_range(vma, start, end, NULL); |
| 1008 | zap_work -= (end - start) / |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1009 | pages_per_huge_page(hstate_vma(vma)); |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 1010 | } |
| 1011 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1012 | start = end; |
| 1013 | } else |
| 1014 | start = unmap_page_range(*tlbp, vma, |
| 1015 | start, end, &zap_work, details); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1016 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1017 | if (zap_work > 0) { |
| 1018 | BUG_ON(start != end); |
| 1019 | break; |
| 1020 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1021 | |
| 1022 | tlb_finish_mmu(*tlbp, tlb_start, start); |
| 1023 | |
| 1024 | if (need_resched() || |
Nick Piggin | 95c354f | 2008-01-30 13:31:20 +0100 | [diff] [blame] | 1025 | (i_mmap_lock && spin_needbreak(i_mmap_lock))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1026 | if (i_mmap_lock) { |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 1027 | *tlbp = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1028 | goto out; |
| 1029 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1030 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1031 | } |
| 1032 | |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 1033 | *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | tlb_start_valid = 0; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1035 | zap_work = ZAP_BLOCK_SIZE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1036 | } |
| 1037 | } |
| 1038 | out: |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1039 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 1040 | return start; /* which is now the end (or restart) address */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1041 | } |
| 1042 | |
| 1043 | /** |
| 1044 | * zap_page_range - remove user pages in a given range |
| 1045 | * @vma: vm_area_struct holding the applicable pages |
| 1046 | * @address: starting address of pages to zap |
| 1047 | * @size: number of bytes to zap |
| 1048 | * @details: details of nonlinear truncation or shared cache invalidation |
| 1049 | */ |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 1050 | unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1051 | unsigned long size, struct zap_details *details) |
| 1052 | { |
| 1053 | struct mm_struct *mm = vma->vm_mm; |
| 1054 | struct mmu_gather *tlb; |
| 1055 | unsigned long end = address + size; |
| 1056 | unsigned long nr_accounted = 0; |
| 1057 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1058 | lru_add_drain(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1059 | tlb = tlb_gather_mmu(mm, 0); |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 1060 | update_hiwater_rss(mm); |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 1061 | end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details); |
| 1062 | if (tlb) |
| 1063 | tlb_finish_mmu(tlb, address, end); |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 1064 | return end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1065 | } |
| 1066 | |
Jack Steiner | c627f9c | 2008-07-29 22:33:53 -0700 | [diff] [blame] | 1067 | /** |
| 1068 | * zap_vma_ptes - remove ptes mapping the vma |
| 1069 | * @vma: vm_area_struct holding ptes to be zapped |
| 1070 | * @address: starting address of pages to zap |
| 1071 | * @size: number of bytes to zap |
| 1072 | * |
| 1073 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. |
| 1074 | * |
| 1075 | * The entire address range must be fully contained within the vma. |
| 1076 | * |
| 1077 | * Returns 0 if successful. |
| 1078 | */ |
| 1079 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, |
| 1080 | unsigned long size) |
| 1081 | { |
| 1082 | if (address < vma->vm_start || address + size > vma->vm_end || |
| 1083 | !(vma->vm_flags & VM_PFNMAP)) |
| 1084 | return -1; |
| 1085 | zap_page_range(vma, address, size, NULL); |
| 1086 | return 0; |
| 1087 | } |
| 1088 | EXPORT_SYMBOL_GPL(zap_vma_ptes); |
| 1089 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1090 | /* |
| 1091 | * Do a quick page-table lookup for a single page. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1092 | */ |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1093 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1094 | unsigned int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1095 | { |
| 1096 | pgd_t *pgd; |
| 1097 | pud_t *pud; |
| 1098 | pmd_t *pmd; |
| 1099 | pte_t *ptep, pte; |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1100 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1101 | struct page *page; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1102 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1103 | |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1104 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); |
| 1105 | if (!IS_ERR(page)) { |
| 1106 | BUG_ON(flags & FOLL_GET); |
| 1107 | goto out; |
| 1108 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1109 | |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1110 | page = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1111 | pgd = pgd_offset(mm, address); |
| 1112 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1113 | goto no_page_table; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1114 | |
| 1115 | pud = pud_offset(pgd, address); |
Andi Kleen | ceb8687 | 2008-07-23 21:27:50 -0700 | [diff] [blame] | 1116 | if (pud_none(*pud)) |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1117 | goto no_page_table; |
Andi Kleen | ceb8687 | 2008-07-23 21:27:50 -0700 | [diff] [blame] | 1118 | if (pud_huge(*pud)) { |
| 1119 | BUG_ON(flags & FOLL_GET); |
| 1120 | page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE); |
| 1121 | goto out; |
| 1122 | } |
| 1123 | if (unlikely(pud_bad(*pud))) |
| 1124 | goto no_page_table; |
| 1125 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1126 | pmd = pmd_offset(pud, address); |
Hugh Dickins | aeed5fc | 2008-05-06 20:49:23 +0100 | [diff] [blame] | 1127 | if (pmd_none(*pmd)) |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1128 | goto no_page_table; |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1129 | if (pmd_huge(*pmd)) { |
| 1130 | BUG_ON(flags & FOLL_GET); |
| 1131 | page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); |
| 1132 | goto out; |
| 1133 | } |
Hugh Dickins | aeed5fc | 2008-05-06 20:49:23 +0100 | [diff] [blame] | 1134 | if (unlikely(pmd_bad(*pmd))) |
| 1135 | goto no_page_table; |
| 1136 | |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1137 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1138 | |
| 1139 | pte = *ptep; |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1140 | if (!pte_present(pte)) |
Linus Torvalds | 89f5b7d | 2008-06-20 11:18:25 -0700 | [diff] [blame] | 1141 | goto no_page; |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1142 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
| 1143 | goto unlock; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1144 | page = vm_normal_page(vma, address, pte); |
| 1145 | if (unlikely(!page)) |
Linus Torvalds | 89f5b7d | 2008-06-20 11:18:25 -0700 | [diff] [blame] | 1146 | goto bad_page; |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1147 | |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1148 | if (flags & FOLL_GET) |
| 1149 | get_page(page); |
| 1150 | if (flags & FOLL_TOUCH) { |
| 1151 | if ((flags & FOLL_WRITE) && |
| 1152 | !pte_dirty(pte) && !PageDirty(page)) |
| 1153 | set_page_dirty(page); |
KOSAKI Motohiro | bd775c4 | 2009-03-31 15:19:37 -0700 | [diff] [blame] | 1154 | /* |
| 1155 | * pte_mkyoung() would be more correct here, but atomic care |
| 1156 | * is needed to avoid losing the dirty bit: it is easier to use |
| 1157 | * mark_page_accessed(). |
| 1158 | */ |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1159 | mark_page_accessed(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1160 | } |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1161 | unlock: |
| 1162 | pte_unmap_unlock(ptep, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1163 | out: |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1164 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1165 | |
Linus Torvalds | 89f5b7d | 2008-06-20 11:18:25 -0700 | [diff] [blame] | 1166 | bad_page: |
| 1167 | pte_unmap_unlock(ptep, ptl); |
| 1168 | return ERR_PTR(-EFAULT); |
| 1169 | |
| 1170 | no_page: |
| 1171 | pte_unmap_unlock(ptep, ptl); |
| 1172 | if (!pte_none(pte)) |
| 1173 | return page; |
| 1174 | /* Fall through to ZERO_PAGE handling */ |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1175 | no_page_table: |
| 1176 | /* |
| 1177 | * When core dumping an enormous anonymous area that nobody |
| 1178 | * has touched so far, we don't want to allocate page tables. |
| 1179 | */ |
| 1180 | if (flags & FOLL_ANON) { |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 1181 | page = ZERO_PAGE(0); |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1182 | if (flags & FOLL_GET) |
| 1183 | get_page(page); |
| 1184 | BUG_ON(flags & FOLL_WRITE); |
| 1185 | } |
| 1186 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1187 | } |
| 1188 | |
Linus Torvalds | 672ca28 | 2008-06-23 11:21:37 -0700 | [diff] [blame] | 1189 | /* Can we do the FOLL_ANON optimization? */ |
| 1190 | static inline int use_zero_page(struct vm_area_struct *vma) |
| 1191 | { |
| 1192 | /* |
| 1193 | * We don't want to optimize FOLL_ANON for make_pages_present() |
| 1194 | * when it tries to page in a VM_LOCKED region. As to VM_SHARED, |
| 1195 | * we want to get the page from the page tables to make sure |
| 1196 | * that we serialize and update with any other user of that |
| 1197 | * mapping. |
| 1198 | */ |
| 1199 | if (vma->vm_flags & (VM_LOCKED | VM_SHARED)) |
| 1200 | return 0; |
| 1201 | /* |
Nick Piggin | 0d71d10a | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 1202 | * And if we have a fault routine, it's not an anonymous region. |
Linus Torvalds | 672ca28 | 2008-06-23 11:21:37 -0700 | [diff] [blame] | 1203 | */ |
Nick Piggin | 0d71d10a | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 1204 | return !vma->vm_ops || !vma->vm_ops->fault; |
Linus Torvalds | 672ca28 | 2008-06-23 11:21:37 -0700 | [diff] [blame] | 1205 | } |
| 1206 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1207 | |
| 1208 | |
| 1209 | int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
| 1210 | unsigned long start, int len, int flags, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1211 | struct page **pages, struct vm_area_struct **vmas) |
| 1212 | { |
| 1213 | int i; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1214 | unsigned int vm_flags = 0; |
| 1215 | int write = !!(flags & GUP_FLAGS_WRITE); |
| 1216 | int force = !!(flags & GUP_FLAGS_FORCE); |
| 1217 | int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS); |
Ying Han | 4779280 | 2009-01-06 14:40:18 -0800 | [diff] [blame] | 1218 | int ignore_sigkill = !!(flags & GUP_FLAGS_IGNORE_SIGKILL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1219 | |
Jonathan Corbet | 900cf08 | 2008-02-11 16:17:33 -0700 | [diff] [blame] | 1220 | if (len <= 0) |
| 1221 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1222 | /* |
| 1223 | * Require read or write permissions. |
| 1224 | * If 'force' is set, we only require the "MAY" flags. |
| 1225 | */ |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1226 | vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); |
| 1227 | vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1228 | i = 0; |
| 1229 | |
| 1230 | do { |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1231 | struct vm_area_struct *vma; |
| 1232 | unsigned int foll_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1233 | |
| 1234 | vma = find_extend_vma(mm, start); |
| 1235 | if (!vma && in_gate_area(tsk, start)) { |
| 1236 | unsigned long pg = start & PAGE_MASK; |
| 1237 | struct vm_area_struct *gate_vma = get_gate_vma(tsk); |
| 1238 | pgd_t *pgd; |
| 1239 | pud_t *pud; |
| 1240 | pmd_t *pmd; |
| 1241 | pte_t *pte; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1242 | |
| 1243 | /* user gate pages are read-only */ |
| 1244 | if (!ignore && write) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1245 | return i ? : -EFAULT; |
| 1246 | if (pg > TASK_SIZE) |
| 1247 | pgd = pgd_offset_k(pg); |
| 1248 | else |
| 1249 | pgd = pgd_offset_gate(mm, pg); |
| 1250 | BUG_ON(pgd_none(*pgd)); |
| 1251 | pud = pud_offset(pgd, pg); |
| 1252 | BUG_ON(pud_none(*pud)); |
| 1253 | pmd = pmd_offset(pud, pg); |
Hugh Dickins | 690dbe1 | 2005-08-01 21:11:42 -0700 | [diff] [blame] | 1254 | if (pmd_none(*pmd)) |
| 1255 | return i ? : -EFAULT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1256 | pte = pte_offset_map(pmd, pg); |
Hugh Dickins | 690dbe1 | 2005-08-01 21:11:42 -0700 | [diff] [blame] | 1257 | if (pte_none(*pte)) { |
| 1258 | pte_unmap(pte); |
| 1259 | return i ? : -EFAULT; |
| 1260 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1261 | if (pages) { |
Nick Piggin | fa2a455 | 2005-11-29 18:43:17 +1100 | [diff] [blame] | 1262 | struct page *page = vm_normal_page(gate_vma, start, *pte); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1263 | pages[i] = page; |
| 1264 | if (page) |
| 1265 | get_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1266 | } |
| 1267 | pte_unmap(pte); |
| 1268 | if (vmas) |
| 1269 | vmas[i] = gate_vma; |
| 1270 | i++; |
| 1271 | start += PAGE_SIZE; |
| 1272 | len--; |
| 1273 | continue; |
| 1274 | } |
| 1275 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1276 | if (!vma || |
| 1277 | (vma->vm_flags & (VM_IO | VM_PFNMAP)) || |
| 1278 | (!ignore && !(vm_flags & vma->vm_flags))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1279 | return i ? : -EFAULT; |
| 1280 | |
| 1281 | if (is_vm_hugetlb_page(vma)) { |
| 1282 | i = follow_hugetlb_page(mm, vma, pages, vmas, |
Adam Litke | 5b23dbe | 2007-11-14 16:59:33 -0800 | [diff] [blame] | 1283 | &start, &len, i, write); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1284 | continue; |
| 1285 | } |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1286 | |
| 1287 | foll_flags = FOLL_TOUCH; |
| 1288 | if (pages) |
| 1289 | foll_flags |= FOLL_GET; |
Linus Torvalds | 672ca28 | 2008-06-23 11:21:37 -0700 | [diff] [blame] | 1290 | if (!write && use_zero_page(vma)) |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1291 | foll_flags |= FOLL_ANON; |
| 1292 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1293 | do { |
Hugh Dickins | 08ef472 | 2005-06-21 17:15:10 -0700 | [diff] [blame] | 1294 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1295 | |
Ethan Solomita | 462e00c | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 1296 | /* |
Ying Han | 4779280 | 2009-01-06 14:40:18 -0800 | [diff] [blame] | 1297 | * If we have a pending SIGKILL, don't keep faulting |
| 1298 | * pages and potentially allocating memory, unless |
| 1299 | * current is handling munlock--e.g., on exit. In |
| 1300 | * that case, we are not allocating memory. Rather, |
| 1301 | * we're only unlocking already resident/mapped pages. |
Ethan Solomita | 462e00c | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 1302 | */ |
Ying Han | 4779280 | 2009-01-06 14:40:18 -0800 | [diff] [blame] | 1303 | if (unlikely(!ignore_sigkill && |
| 1304 | fatal_signal_pending(current))) |
| 1305 | return i ? i : -ERESTARTSYS; |
Ethan Solomita | 462e00c | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 1306 | |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1307 | if (write) |
| 1308 | foll_flags |= FOLL_WRITE; |
| 1309 | |
| 1310 | cond_resched(); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1311 | while (!(page = follow_page(vma, start, foll_flags))) { |
Linus Torvalds | a68d2eb | 2005-08-03 10:07:09 -0700 | [diff] [blame] | 1312 | int ret; |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 1313 | ret = handle_mm_fault(mm, vma, start, |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1314 | foll_flags & FOLL_WRITE); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 1315 | if (ret & VM_FAULT_ERROR) { |
| 1316 | if (ret & VM_FAULT_OOM) |
| 1317 | return i ? i : -ENOMEM; |
| 1318 | else if (ret & VM_FAULT_SIGBUS) |
| 1319 | return i ? i : -EFAULT; |
| 1320 | BUG(); |
| 1321 | } |
| 1322 | if (ret & VM_FAULT_MAJOR) |
| 1323 | tsk->maj_flt++; |
| 1324 | else |
| 1325 | tsk->min_flt++; |
| 1326 | |
Linus Torvalds | a68d2eb | 2005-08-03 10:07:09 -0700 | [diff] [blame] | 1327 | /* |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 1328 | * The VM_FAULT_WRITE bit tells us that |
| 1329 | * do_wp_page has broken COW when necessary, |
| 1330 | * even if maybe_mkwrite decided not to set |
| 1331 | * pte_write. We can thus safely do subsequent |
Hugh Dickins | 878b63a | 2009-01-06 14:39:32 -0800 | [diff] [blame] | 1332 | * page lookups as if they were reads. But only |
| 1333 | * do so when looping for pte_write is futile: |
| 1334 | * in some cases userspace may also be wanting |
| 1335 | * to write to the gotten user page, which a |
| 1336 | * read fault here might prevent (a readonly |
| 1337 | * page might get reCOWed by userspace write). |
Linus Torvalds | a68d2eb | 2005-08-03 10:07:09 -0700 | [diff] [blame] | 1338 | */ |
Hugh Dickins | 878b63a | 2009-01-06 14:39:32 -0800 | [diff] [blame] | 1339 | if ((ret & VM_FAULT_WRITE) && |
| 1340 | !(vma->vm_flags & VM_WRITE)) |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1341 | foll_flags &= ~FOLL_WRITE; |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 1342 | |
Benjamin Herrenschmidt | 7f7bbbe | 2006-10-06 00:43:53 -0700 | [diff] [blame] | 1343 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1344 | } |
Linus Torvalds | 89f5b7d | 2008-06-20 11:18:25 -0700 | [diff] [blame] | 1345 | if (IS_ERR(page)) |
| 1346 | return i ? i : PTR_ERR(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1347 | if (pages) { |
Hugh Dickins | 08ef472 | 2005-06-21 17:15:10 -0700 | [diff] [blame] | 1348 | pages[i] = page; |
James Bottomley | 03beb07 | 2006-03-26 01:36:57 -0800 | [diff] [blame] | 1349 | |
Russell King | a6f36be | 2006-12-30 22:24:19 +0000 | [diff] [blame] | 1350 | flush_anon_page(vma, page, start); |
Hugh Dickins | 08ef472 | 2005-06-21 17:15:10 -0700 | [diff] [blame] | 1351 | flush_dcache_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1352 | } |
| 1353 | if (vmas) |
| 1354 | vmas[i] = vma; |
| 1355 | i++; |
| 1356 | start += PAGE_SIZE; |
| 1357 | len--; |
Hugh Dickins | 08ef472 | 2005-06-21 17:15:10 -0700 | [diff] [blame] | 1358 | } while (len && start < vma->vm_end); |
Hugh Dickins | 08ef472 | 2005-06-21 17:15:10 -0700 | [diff] [blame] | 1359 | } while (len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1360 | return i; |
| 1361 | } |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1362 | |
| 1363 | int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
| 1364 | unsigned long start, int len, int write, int force, |
| 1365 | struct page **pages, struct vm_area_struct **vmas) |
| 1366 | { |
| 1367 | int flags = 0; |
| 1368 | |
| 1369 | if (write) |
| 1370 | flags |= GUP_FLAGS_WRITE; |
| 1371 | if (force) |
| 1372 | flags |= GUP_FLAGS_FORCE; |
| 1373 | |
| 1374 | return __get_user_pages(tsk, mm, |
| 1375 | start, len, flags, |
| 1376 | pages, vmas); |
| 1377 | } |
| 1378 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1379 | EXPORT_SYMBOL(get_user_pages); |
| 1380 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 1381 | pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr, |
| 1382 | spinlock_t **ptl) |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1383 | { |
| 1384 | pgd_t * pgd = pgd_offset(mm, addr); |
| 1385 | pud_t * pud = pud_alloc(mm, pgd, addr); |
| 1386 | if (pud) { |
Trond Myklebust | 49c91fb | 2005-11-29 19:27:22 -0500 | [diff] [blame] | 1387 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1388 | if (pmd) |
| 1389 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
| 1390 | } |
| 1391 | return NULL; |
| 1392 | } |
| 1393 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1394 | /* |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1395 | * This is the old fallback for page remapping. |
| 1396 | * |
| 1397 | * For historical reasons, it only allows reserved pages. Only |
| 1398 | * old drivers should use this, and they needed to mark their |
| 1399 | * pages reserved for the old functions anyway. |
| 1400 | */ |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1401 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
| 1402 | struct page *page, pgprot_t prot) |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1403 | { |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1404 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1405 | int retval; |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1406 | pte_t *pte; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1407 | spinlock_t *ptl; |
| 1408 | |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1409 | retval = -EINVAL; |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1410 | if (PageAnon(page)) |
KAMEZAWA Hiroyuki | 5b4e655 | 2008-10-18 20:28:10 -0700 | [diff] [blame] | 1411 | goto out; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1412 | retval = -ENOMEM; |
| 1413 | flush_dcache_page(page); |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1414 | pte = get_locked_pte(mm, addr, &ptl); |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1415 | if (!pte) |
KAMEZAWA Hiroyuki | 5b4e655 | 2008-10-18 20:28:10 -0700 | [diff] [blame] | 1416 | goto out; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1417 | retval = -EBUSY; |
| 1418 | if (!pte_none(*pte)) |
| 1419 | goto out_unlock; |
| 1420 | |
| 1421 | /* Ok, finally just insert the thing.. */ |
| 1422 | get_page(page); |
| 1423 | inc_mm_counter(mm, file_rss); |
| 1424 | page_add_file_rmap(page); |
| 1425 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); |
| 1426 | |
| 1427 | retval = 0; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1428 | pte_unmap_unlock(pte, ptl); |
| 1429 | return retval; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1430 | out_unlock: |
| 1431 | pte_unmap_unlock(pte, ptl); |
| 1432 | out: |
| 1433 | return retval; |
| 1434 | } |
| 1435 | |
Rolf Eike Beer | bfa5bf6 | 2006-09-25 23:31:22 -0700 | [diff] [blame] | 1436 | /** |
| 1437 | * vm_insert_page - insert single page into user vma |
| 1438 | * @vma: user vma to map to |
| 1439 | * @addr: target user address of this page |
| 1440 | * @page: source kernel page |
| 1441 | * |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1442 | * This allows drivers to insert individual pages they've allocated |
| 1443 | * into a user vma. |
| 1444 | * |
| 1445 | * The page has to be a nice clean _individual_ kernel allocation. |
| 1446 | * If you allocate a compound page, you need to have marked it as |
| 1447 | * such (__GFP_COMP), or manually just split the page up yourself |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 1448 | * (see split_page()). |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1449 | * |
| 1450 | * NOTE! Traditionally this was done with "remap_pfn_range()" which |
| 1451 | * took an arbitrary page protection parameter. This doesn't allow |
| 1452 | * that. Your vma protection will have to be set up correctly, which |
| 1453 | * means that if you want a shared writable mapping, you'd better |
| 1454 | * ask for a shared writable mapping! |
| 1455 | * |
| 1456 | * The page does not need to be reserved. |
| 1457 | */ |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1458 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
| 1459 | struct page *page) |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1460 | { |
| 1461 | if (addr < vma->vm_start || addr >= vma->vm_end) |
| 1462 | return -EFAULT; |
| 1463 | if (!page_count(page)) |
| 1464 | return -EINVAL; |
Linus Torvalds | 4d7672b | 2005-12-16 10:21:23 -0800 | [diff] [blame] | 1465 | vma->vm_flags |= VM_INSERTPAGE; |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1466 | return insert_page(vma, addr, page, vma->vm_page_prot); |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1467 | } |
Linus Torvalds | e3c3374 | 2005-12-03 20:48:11 -0800 | [diff] [blame] | 1468 | EXPORT_SYMBOL(vm_insert_page); |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1469 | |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1470 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
| 1471 | unsigned long pfn, pgprot_t prot) |
| 1472 | { |
| 1473 | struct mm_struct *mm = vma->vm_mm; |
| 1474 | int retval; |
| 1475 | pte_t *pte, entry; |
| 1476 | spinlock_t *ptl; |
| 1477 | |
| 1478 | retval = -ENOMEM; |
| 1479 | pte = get_locked_pte(mm, addr, &ptl); |
| 1480 | if (!pte) |
| 1481 | goto out; |
| 1482 | retval = -EBUSY; |
| 1483 | if (!pte_none(*pte)) |
| 1484 | goto out_unlock; |
| 1485 | |
| 1486 | /* Ok, finally just insert the thing.. */ |
| 1487 | entry = pte_mkspecial(pfn_pte(pfn, prot)); |
| 1488 | set_pte_at(mm, addr, pte, entry); |
| 1489 | update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */ |
| 1490 | |
| 1491 | retval = 0; |
| 1492 | out_unlock: |
| 1493 | pte_unmap_unlock(pte, ptl); |
| 1494 | out: |
| 1495 | return retval; |
| 1496 | } |
| 1497 | |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1498 | /** |
| 1499 | * vm_insert_pfn - insert single pfn into user vma |
| 1500 | * @vma: user vma to map to |
| 1501 | * @addr: target user address of this page |
| 1502 | * @pfn: source kernel pfn |
| 1503 | * |
| 1504 | * Similar to vm_inert_page, this allows drivers to insert individual pages |
| 1505 | * they've allocated into a user vma. Same comments apply. |
| 1506 | * |
| 1507 | * This function should only be called from a vm_ops->fault handler, and |
| 1508 | * in that case the handler should return NULL. |
Nick Piggin | 0d71d10a | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 1509 | * |
| 1510 | * vma cannot be a COW mapping. |
| 1511 | * |
| 1512 | * As this is called only for pages that do not currently exist, we |
| 1513 | * do not need to flush old virtual caches or the TLB. |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1514 | */ |
| 1515 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1516 | unsigned long pfn) |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1517 | { |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1518 | int ret; |
venkatesh.pallipadi@intel.com | e4b866e | 2009-01-09 16:13:11 -0800 | [diff] [blame] | 1519 | pgprot_t pgprot = vma->vm_page_prot; |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 1520 | /* |
| 1521 | * Technically, architectures with pte_special can avoid all these |
| 1522 | * restrictions (same for remap_pfn_range). However we would like |
| 1523 | * consistency in testing and feature parity among all, so we should |
| 1524 | * try to keep these invariants in place for everybody. |
| 1525 | */ |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 1526 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
| 1527 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == |
| 1528 | (VM_PFNMAP|VM_MIXEDMAP)); |
| 1529 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); |
| 1530 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1531 | |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1532 | if (addr < vma->vm_start || addr >= vma->vm_end) |
| 1533 | return -EFAULT; |
venkatesh.pallipadi@intel.com | e4b866e | 2009-01-09 16:13:11 -0800 | [diff] [blame] | 1534 | if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE)) |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1535 | return -EINVAL; |
| 1536 | |
venkatesh.pallipadi@intel.com | e4b866e | 2009-01-09 16:13:11 -0800 | [diff] [blame] | 1537 | ret = insert_pfn(vma, addr, pfn, pgprot); |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1538 | |
| 1539 | if (ret) |
| 1540 | untrack_pfn_vma(vma, pfn, PAGE_SIZE); |
| 1541 | |
| 1542 | return ret; |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1543 | } |
| 1544 | EXPORT_SYMBOL(vm_insert_pfn); |
| 1545 | |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1546 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
| 1547 | unsigned long pfn) |
| 1548 | { |
| 1549 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); |
| 1550 | |
| 1551 | if (addr < vma->vm_start || addr >= vma->vm_end) |
| 1552 | return -EFAULT; |
| 1553 | |
| 1554 | /* |
| 1555 | * If we don't have pte special, then we have to use the pfn_valid() |
| 1556 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* |
| 1557 | * refcount the page if pfn_valid is true (hence insert_page rather |
| 1558 | * than insert_pfn). |
| 1559 | */ |
| 1560 | if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { |
| 1561 | struct page *page; |
| 1562 | |
| 1563 | page = pfn_to_page(pfn); |
| 1564 | return insert_page(vma, addr, page, vma->vm_page_prot); |
| 1565 | } |
| 1566 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); |
| 1567 | } |
| 1568 | EXPORT_SYMBOL(vm_insert_mixed); |
| 1569 | |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1570 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1571 | * maps a range of physical memory into the requested pages. the old |
| 1572 | * mappings are removed. any references to nonexistent pages results |
| 1573 | * in null mappings (currently treated as "copy-on-access") |
| 1574 | */ |
| 1575 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, |
| 1576 | unsigned long addr, unsigned long end, |
| 1577 | unsigned long pfn, pgprot_t prot) |
| 1578 | { |
| 1579 | pte_t *pte; |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 1580 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1581 | |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 1582 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1583 | if (!pte) |
| 1584 | return -ENOMEM; |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 1585 | arch_enter_lazy_mmu_mode(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1586 | do { |
| 1587 | BUG_ON(!pte_none(*pte)); |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 1588 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1589 | pfn++; |
| 1590 | } while (pte++, addr += PAGE_SIZE, addr != end); |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 1591 | arch_leave_lazy_mmu_mode(); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 1592 | pte_unmap_unlock(pte - 1, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1593 | return 0; |
| 1594 | } |
| 1595 | |
| 1596 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, |
| 1597 | unsigned long addr, unsigned long end, |
| 1598 | unsigned long pfn, pgprot_t prot) |
| 1599 | { |
| 1600 | pmd_t *pmd; |
| 1601 | unsigned long next; |
| 1602 | |
| 1603 | pfn -= addr >> PAGE_SHIFT; |
| 1604 | pmd = pmd_alloc(mm, pud, addr); |
| 1605 | if (!pmd) |
| 1606 | return -ENOMEM; |
| 1607 | do { |
| 1608 | next = pmd_addr_end(addr, end); |
| 1609 | if (remap_pte_range(mm, pmd, addr, next, |
| 1610 | pfn + (addr >> PAGE_SHIFT), prot)) |
| 1611 | return -ENOMEM; |
| 1612 | } while (pmd++, addr = next, addr != end); |
| 1613 | return 0; |
| 1614 | } |
| 1615 | |
| 1616 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, |
| 1617 | unsigned long addr, unsigned long end, |
| 1618 | unsigned long pfn, pgprot_t prot) |
| 1619 | { |
| 1620 | pud_t *pud; |
| 1621 | unsigned long next; |
| 1622 | |
| 1623 | pfn -= addr >> PAGE_SHIFT; |
| 1624 | pud = pud_alloc(mm, pgd, addr); |
| 1625 | if (!pud) |
| 1626 | return -ENOMEM; |
| 1627 | do { |
| 1628 | next = pud_addr_end(addr, end); |
| 1629 | if (remap_pmd_range(mm, pud, addr, next, |
| 1630 | pfn + (addr >> PAGE_SHIFT), prot)) |
| 1631 | return -ENOMEM; |
| 1632 | } while (pud++, addr = next, addr != end); |
| 1633 | return 0; |
| 1634 | } |
| 1635 | |
Rolf Eike Beer | bfa5bf6 | 2006-09-25 23:31:22 -0700 | [diff] [blame] | 1636 | /** |
| 1637 | * remap_pfn_range - remap kernel memory to userspace |
| 1638 | * @vma: user vma to map to |
| 1639 | * @addr: target user address to start at |
| 1640 | * @pfn: physical address of kernel memory |
| 1641 | * @size: size of map area |
| 1642 | * @prot: page protection flags for this mapping |
| 1643 | * |
| 1644 | * Note: this is only safe if the mm semaphore is held when called. |
| 1645 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1646 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
| 1647 | unsigned long pfn, unsigned long size, pgprot_t prot) |
| 1648 | { |
| 1649 | pgd_t *pgd; |
| 1650 | unsigned long next; |
Hugh Dickins | 2d15cab | 2005-06-25 14:54:33 -0700 | [diff] [blame] | 1651 | unsigned long end = addr + PAGE_ALIGN(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1652 | struct mm_struct *mm = vma->vm_mm; |
| 1653 | int err; |
| 1654 | |
| 1655 | /* |
| 1656 | * Physically remapped pages are special. Tell the |
| 1657 | * rest of the world about it: |
| 1658 | * VM_IO tells people not to look at these pages |
| 1659 | * (accesses can have side effects). |
Hugh Dickins | 0b14c17 | 2005-11-21 21:32:15 -0800 | [diff] [blame] | 1660 | * VM_RESERVED is specified all over the place, because |
| 1661 | * in 2.4 it kept swapout's vma scan off this vma; but |
| 1662 | * in 2.6 the LRU scan won't even find its pages, so this |
| 1663 | * flag means no more than count its pages in reserved_vm, |
| 1664 | * and omit it from core dump, even when VM_IO turned off. |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1665 | * VM_PFNMAP tells the core MM that the base pages are just |
| 1666 | * raw PFN mappings, and do not have a "struct page" associated |
| 1667 | * with them. |
Linus Torvalds | fb155c1 | 2005-12-11 19:46:02 -0800 | [diff] [blame] | 1668 | * |
| 1669 | * There's a horrible special case to handle copy-on-write |
| 1670 | * behaviour that some programs depend on. We mark the "original" |
| 1671 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1672 | */ |
Pallipadi, Venkatesh | 4bb9c5c | 2009-03-12 17:45:27 -0700 | [diff] [blame] | 1673 | if (addr == vma->vm_start && end == vma->vm_end) { |
Linus Torvalds | fb155c1 | 2005-12-11 19:46:02 -0800 | [diff] [blame] | 1674 | vma->vm_pgoff = pfn; |
Pallipadi, Venkatesh | 895791d | 2009-03-13 16:35:44 -0700 | [diff] [blame] | 1675 | vma->vm_flags |= VM_PFN_AT_MMAP; |
Pallipadi, Venkatesh | 4bb9c5c | 2009-03-12 17:45:27 -0700 | [diff] [blame] | 1676 | } else if (is_cow_mapping(vma->vm_flags)) |
venkatesh.pallipadi@intel.com | 3c8bb73 | 2008-12-18 11:41:27 -0800 | [diff] [blame] | 1677 | return -EINVAL; |
Linus Torvalds | fb155c1 | 2005-12-11 19:46:02 -0800 | [diff] [blame] | 1678 | |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1679 | vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1680 | |
venkatesh.pallipadi@intel.com | e4b866e | 2009-01-09 16:13:11 -0800 | [diff] [blame] | 1681 | err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size)); |
venkatesh.pallipadi@intel.com | a367061 | 2009-01-09 16:13:09 -0800 | [diff] [blame] | 1682 | if (err) { |
| 1683 | /* |
| 1684 | * To indicate that track_pfn related cleanup is not |
| 1685 | * needed from higher level routine calling unmap_vmas |
| 1686 | */ |
| 1687 | vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP); |
Pallipadi, Venkatesh | 895791d | 2009-03-13 16:35:44 -0700 | [diff] [blame] | 1688 | vma->vm_flags &= ~VM_PFN_AT_MMAP; |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1689 | return -EINVAL; |
venkatesh.pallipadi@intel.com | a367061 | 2009-01-09 16:13:09 -0800 | [diff] [blame] | 1690 | } |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1691 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1692 | BUG_ON(addr >= end); |
| 1693 | pfn -= addr >> PAGE_SHIFT; |
| 1694 | pgd = pgd_offset(mm, addr); |
| 1695 | flush_cache_range(vma, addr, end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1696 | do { |
| 1697 | next = pgd_addr_end(addr, end); |
| 1698 | err = remap_pud_range(mm, pgd, addr, next, |
| 1699 | pfn + (addr >> PAGE_SHIFT), prot); |
| 1700 | if (err) |
| 1701 | break; |
| 1702 | } while (pgd++, addr = next, addr != end); |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1703 | |
| 1704 | if (err) |
| 1705 | untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size)); |
| 1706 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1707 | return err; |
| 1708 | } |
| 1709 | EXPORT_SYMBOL(remap_pfn_range); |
| 1710 | |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1711 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
| 1712 | unsigned long addr, unsigned long end, |
| 1713 | pte_fn_t fn, void *data) |
| 1714 | { |
| 1715 | pte_t *pte; |
| 1716 | int err; |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 1717 | pgtable_t token; |
Borislav Petkov | 9490991 | 2007-05-06 14:49:17 -0700 | [diff] [blame] | 1718 | spinlock_t *uninitialized_var(ptl); |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1719 | |
| 1720 | pte = (mm == &init_mm) ? |
| 1721 | pte_alloc_kernel(pmd, addr) : |
| 1722 | pte_alloc_map_lock(mm, pmd, addr, &ptl); |
| 1723 | if (!pte) |
| 1724 | return -ENOMEM; |
| 1725 | |
| 1726 | BUG_ON(pmd_huge(*pmd)); |
| 1727 | |
Jeremy Fitzhardinge | 38e0edb | 2009-01-06 14:39:21 -0800 | [diff] [blame] | 1728 | arch_enter_lazy_mmu_mode(); |
| 1729 | |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 1730 | token = pmd_pgtable(*pmd); |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1731 | |
| 1732 | do { |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 1733 | err = fn(pte, token, addr, data); |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1734 | if (err) |
| 1735 | break; |
| 1736 | } while (pte++, addr += PAGE_SIZE, addr != end); |
| 1737 | |
Jeremy Fitzhardinge | 38e0edb | 2009-01-06 14:39:21 -0800 | [diff] [blame] | 1738 | arch_leave_lazy_mmu_mode(); |
| 1739 | |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1740 | if (mm != &init_mm) |
| 1741 | pte_unmap_unlock(pte-1, ptl); |
| 1742 | return err; |
| 1743 | } |
| 1744 | |
| 1745 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, |
| 1746 | unsigned long addr, unsigned long end, |
| 1747 | pte_fn_t fn, void *data) |
| 1748 | { |
| 1749 | pmd_t *pmd; |
| 1750 | unsigned long next; |
| 1751 | int err; |
| 1752 | |
Andi Kleen | ceb8687 | 2008-07-23 21:27:50 -0700 | [diff] [blame] | 1753 | BUG_ON(pud_huge(*pud)); |
| 1754 | |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1755 | pmd = pmd_alloc(mm, pud, addr); |
| 1756 | if (!pmd) |
| 1757 | return -ENOMEM; |
| 1758 | do { |
| 1759 | next = pmd_addr_end(addr, end); |
| 1760 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); |
| 1761 | if (err) |
| 1762 | break; |
| 1763 | } while (pmd++, addr = next, addr != end); |
| 1764 | return err; |
| 1765 | } |
| 1766 | |
| 1767 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, |
| 1768 | unsigned long addr, unsigned long end, |
| 1769 | pte_fn_t fn, void *data) |
| 1770 | { |
| 1771 | pud_t *pud; |
| 1772 | unsigned long next; |
| 1773 | int err; |
| 1774 | |
| 1775 | pud = pud_alloc(mm, pgd, addr); |
| 1776 | if (!pud) |
| 1777 | return -ENOMEM; |
| 1778 | do { |
| 1779 | next = pud_addr_end(addr, end); |
| 1780 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); |
| 1781 | if (err) |
| 1782 | break; |
| 1783 | } while (pud++, addr = next, addr != end); |
| 1784 | return err; |
| 1785 | } |
| 1786 | |
| 1787 | /* |
| 1788 | * Scan a region of virtual memory, filling in page tables as necessary |
| 1789 | * and calling a provided function on each leaf page table. |
| 1790 | */ |
| 1791 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, |
| 1792 | unsigned long size, pte_fn_t fn, void *data) |
| 1793 | { |
| 1794 | pgd_t *pgd; |
| 1795 | unsigned long next; |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1796 | unsigned long start = addr, end = addr + size; |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1797 | int err; |
| 1798 | |
| 1799 | BUG_ON(addr >= end); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1800 | mmu_notifier_invalidate_range_start(mm, start, end); |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1801 | pgd = pgd_offset(mm, addr); |
| 1802 | do { |
| 1803 | next = pgd_addr_end(addr, end); |
| 1804 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); |
| 1805 | if (err) |
| 1806 | break; |
| 1807 | } while (pgd++, addr = next, addr != end); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1808 | mmu_notifier_invalidate_range_end(mm, start, end); |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1809 | return err; |
| 1810 | } |
| 1811 | EXPORT_SYMBOL_GPL(apply_to_page_range); |
| 1812 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1813 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1814 | * handle_pte_fault chooses page fault handler according to an entry |
| 1815 | * which was read non-atomically. Before making any commitment, on |
| 1816 | * those architectures or configurations (e.g. i386 with PAE) which |
| 1817 | * might give a mix of unmatched parts, do_swap_page and do_file_page |
| 1818 | * must check under lock before unmapping the pte and proceeding |
| 1819 | * (but do_wp_page is only called after already making such a check; |
| 1820 | * and do_anonymous_page and do_no_page can safely check later on). |
| 1821 | */ |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 1822 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1823 | pte_t *page_table, pte_t orig_pte) |
| 1824 | { |
| 1825 | int same = 1; |
| 1826 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) |
| 1827 | if (sizeof(pte_t) > sizeof(unsigned long)) { |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 1828 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
| 1829 | spin_lock(ptl); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1830 | same = pte_same(*page_table, orig_pte); |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 1831 | spin_unlock(ptl); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1832 | } |
| 1833 | #endif |
| 1834 | pte_unmap(page_table); |
| 1835 | return same; |
| 1836 | } |
| 1837 | |
| 1838 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1839 | * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when |
| 1840 | * servicing faults for write access. In the normal case, do always want |
| 1841 | * pte_mkwrite. But get_user_pages can cause write faults for mappings |
| 1842 | * that do not have writing enabled, when used by access_process_vm. |
| 1843 | */ |
| 1844 | static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma) |
| 1845 | { |
| 1846 | if (likely(vma->vm_flags & VM_WRITE)) |
| 1847 | pte = pte_mkwrite(pte); |
| 1848 | return pte; |
| 1849 | } |
| 1850 | |
Atsushi Nemoto | 9de455b | 2006-12-12 17:14:55 +0000 | [diff] [blame] | 1851 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1852 | { |
| 1853 | /* |
| 1854 | * If the source page was a PFN mapping, we don't have |
| 1855 | * a "struct page" for it. We do a best-effort copy by |
| 1856 | * just copying from the original user address. If that |
| 1857 | * fails, we just zero-fill it. Live with it. |
| 1858 | */ |
| 1859 | if (unlikely(!src)) { |
| 1860 | void *kaddr = kmap_atomic(dst, KM_USER0); |
Linus Torvalds | 5d2a2db | 2005-11-29 14:07:55 -0800 | [diff] [blame] | 1861 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
| 1862 | |
| 1863 | /* |
| 1864 | * This really shouldn't fail, because the page is there |
| 1865 | * in the page tables. But it might just be unreadable, |
| 1866 | * in which case we just give up and fill the result with |
| 1867 | * zeroes. |
| 1868 | */ |
| 1869 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1870 | memset(kaddr, 0, PAGE_SIZE); |
| 1871 | kunmap_atomic(kaddr, KM_USER0); |
Dmitriy Monakhov | c4ec7b0 | 2006-10-19 23:29:08 -0700 | [diff] [blame] | 1872 | flush_dcache_page(dst); |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 1873 | } else |
| 1874 | copy_user_highpage(dst, src, va, vma); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1875 | } |
| 1876 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1877 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1878 | * This routine handles present pages, when users try to write |
| 1879 | * to a shared page. It is done by copying the page to a new address |
| 1880 | * and decrementing the shared-page counter for the old page. |
| 1881 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1882 | * Note that this routine assumes that the protection checks have been |
| 1883 | * done by the caller (the low-level page fault routine in most cases). |
| 1884 | * Thus we can safely just mark it writable once we've done any necessary |
| 1885 | * COW. |
| 1886 | * |
| 1887 | * We also mark the page dirty at this point even though the page will |
| 1888 | * change only once the write actually happens. This avoids a few races, |
| 1889 | * and potentially makes it more efficient. |
| 1890 | * |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1891 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 1892 | * but allow concurrent faults), with pte both mapped and locked. |
| 1893 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1894 | */ |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 1895 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, |
| 1896 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1897 | spinlock_t *ptl, pte_t orig_pte) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1898 | { |
Hugh Dickins | e5bbe4d | 2005-11-29 16:54:51 +0000 | [diff] [blame] | 1899 | struct page *old_page, *new_page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1900 | pte_t entry; |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 1901 | int reuse = 0, ret = 0; |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 1902 | int page_mkwrite = 0; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1903 | struct page *dirty_page = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1904 | |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1905 | old_page = vm_normal_page(vma, address, orig_pte); |
Peter Zijlstra | 251b97f | 2008-07-04 09:59:24 -0700 | [diff] [blame] | 1906 | if (!old_page) { |
| 1907 | /* |
| 1908 | * VM_MIXEDMAP !pfn_valid() case |
| 1909 | * |
| 1910 | * We should not cow pages in a shared writeable mapping. |
| 1911 | * Just mark the pages writable as we can't do any dirty |
| 1912 | * accounting on raw pfn maps. |
| 1913 | */ |
| 1914 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
| 1915 | (VM_WRITE|VM_SHARED)) |
| 1916 | goto reuse; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1917 | goto gotten; |
Peter Zijlstra | 251b97f | 2008-07-04 09:59:24 -0700 | [diff] [blame] | 1918 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1919 | |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1920 | /* |
Peter Zijlstra | ee6a645 | 2006-09-25 23:31:00 -0700 | [diff] [blame] | 1921 | * Take out anonymous pages first, anonymous shared vmas are |
| 1922 | * not dirty accountable. |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1923 | */ |
Peter Zijlstra | ee6a645 | 2006-09-25 23:31:00 -0700 | [diff] [blame] | 1924 | if (PageAnon(old_page)) { |
Hugh Dickins | ab967d8 | 2009-01-06 14:39:33 -0800 | [diff] [blame] | 1925 | if (!trylock_page(old_page)) { |
| 1926 | page_cache_get(old_page); |
| 1927 | pte_unmap_unlock(page_table, ptl); |
| 1928 | lock_page(old_page); |
| 1929 | page_table = pte_offset_map_lock(mm, pmd, address, |
| 1930 | &ptl); |
| 1931 | if (!pte_same(*page_table, orig_pte)) { |
| 1932 | unlock_page(old_page); |
| 1933 | page_cache_release(old_page); |
| 1934 | goto unlock; |
| 1935 | } |
| 1936 | page_cache_release(old_page); |
Peter Zijlstra | ee6a645 | 2006-09-25 23:31:00 -0700 | [diff] [blame] | 1937 | } |
Hugh Dickins | 7b1fe59 | 2009-01-06 14:39:34 -0800 | [diff] [blame] | 1938 | reuse = reuse_swap_page(old_page); |
Hugh Dickins | ab967d8 | 2009-01-06 14:39:33 -0800 | [diff] [blame] | 1939 | unlock_page(old_page); |
Peter Zijlstra | ee6a645 | 2006-09-25 23:31:00 -0700 | [diff] [blame] | 1940 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1941 | (VM_WRITE|VM_SHARED))) { |
Peter Zijlstra | ee6a645 | 2006-09-25 23:31:00 -0700 | [diff] [blame] | 1942 | /* |
| 1943 | * Only catch write-faults on shared writable pages, |
| 1944 | * read-only shared pages can get COWed by |
| 1945 | * get_user_pages(.write=1, .force=1). |
| 1946 | */ |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1947 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
Nick Piggin | c2ec175 | 2009-03-31 15:23:21 -0700 | [diff] [blame] | 1948 | struct vm_fault vmf; |
| 1949 | int tmp; |
| 1950 | |
| 1951 | vmf.virtual_address = (void __user *)(address & |
| 1952 | PAGE_MASK); |
| 1953 | vmf.pgoff = old_page->index; |
| 1954 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
| 1955 | vmf.page = old_page; |
| 1956 | |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1957 | /* |
| 1958 | * Notify the address space that the page is about to |
| 1959 | * become writable so that it can prohibit this or wait |
| 1960 | * for the page to get into an appropriate state. |
| 1961 | * |
| 1962 | * We do this without the lock held, so that it can |
| 1963 | * sleep if it needs to. |
| 1964 | */ |
| 1965 | page_cache_get(old_page); |
| 1966 | pte_unmap_unlock(page_table, ptl); |
| 1967 | |
Nick Piggin | c2ec175 | 2009-03-31 15:23:21 -0700 | [diff] [blame] | 1968 | tmp = vma->vm_ops->page_mkwrite(vma, &vmf); |
| 1969 | if (unlikely(tmp & |
| 1970 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
| 1971 | ret = tmp; |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1972 | goto unwritable_page; |
Nick Piggin | c2ec175 | 2009-03-31 15:23:21 -0700 | [diff] [blame] | 1973 | } |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 1974 | if (unlikely(!(tmp & VM_FAULT_LOCKED))) { |
| 1975 | lock_page(old_page); |
| 1976 | if (!old_page->mapping) { |
| 1977 | ret = 0; /* retry the fault */ |
| 1978 | unlock_page(old_page); |
| 1979 | goto unwritable_page; |
| 1980 | } |
| 1981 | } else |
| 1982 | VM_BUG_ON(!PageLocked(old_page)); |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1983 | |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1984 | /* |
| 1985 | * Since we dropped the lock we need to revalidate |
| 1986 | * the PTE as someone else may have changed it. If |
| 1987 | * they did, we just return, as we can count on the |
| 1988 | * MMU to tell us if they didn't also make it writable. |
| 1989 | */ |
| 1990 | page_table = pte_offset_map_lock(mm, pmd, address, |
| 1991 | &ptl); |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 1992 | if (!pte_same(*page_table, orig_pte)) { |
| 1993 | unlock_page(old_page); |
| 1994 | page_cache_release(old_page); |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1995 | goto unlock; |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 1996 | } |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 1997 | |
| 1998 | page_mkwrite = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1999 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2000 | dirty_page = old_page; |
| 2001 | get_page(dirty_page); |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2002 | reuse = 1; |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2003 | } |
| 2004 | |
| 2005 | if (reuse) { |
Peter Zijlstra | 251b97f | 2008-07-04 09:59:24 -0700 | [diff] [blame] | 2006 | reuse: |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2007 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
| 2008 | entry = pte_mkyoung(orig_pte); |
| 2009 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
KAMEZAWA Hiroyuki | 954ffcb | 2007-10-16 01:25:44 -0700 | [diff] [blame] | 2010 | if (ptep_set_access_flags(vma, address, page_table, entry,1)) |
Benjamin Herrenschmidt | 8dab524 | 2007-06-16 10:16:12 -0700 | [diff] [blame] | 2011 | update_mmu_cache(vma, address, entry); |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2012 | ret |= VM_FAULT_WRITE; |
| 2013 | goto unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2014 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2015 | |
| 2016 | /* |
| 2017 | * Ok, we need to copy. Oh, well.. |
| 2018 | */ |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 2019 | page_cache_get(old_page); |
Hugh Dickins | 920fc35 | 2005-11-21 21:32:17 -0800 | [diff] [blame] | 2020 | gotten: |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2021 | pte_unmap_unlock(page_table, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2022 | |
| 2023 | if (unlikely(anon_vma_prepare(vma))) |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2024 | goto oom; |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2025 | VM_BUG_ON(old_page == ZERO_PAGE(0)); |
| 2026 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); |
| 2027 | if (!new_page) |
| 2028 | goto oom; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 2029 | /* |
| 2030 | * Don't let another task, with possibly unlocked vma, |
| 2031 | * keep the mlocked page. |
| 2032 | */ |
Carsten Otte | ab92661 | 2009-02-04 15:12:16 -0800 | [diff] [blame] | 2033 | if ((vma->vm_flags & VM_LOCKED) && old_page) { |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 2034 | lock_page(old_page); /* for LRU manipulation */ |
| 2035 | clear_page_mlock(old_page); |
| 2036 | unlock_page(old_page); |
| 2037 | } |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2038 | cow_user_page(new_page, old_page, address, vma); |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 2039 | __SetPageUptodate(new_page); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2040 | |
KAMEZAWA Hiroyuki | 2c26fdd | 2009-01-07 18:08:10 -0800 | [diff] [blame] | 2041 | if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL)) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2042 | goto oom_free_new; |
| 2043 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2044 | /* |
| 2045 | * Re-check the pte - we dropped the lock |
| 2046 | */ |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2047 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2048 | if (likely(pte_same(*page_table, orig_pte))) { |
Hugh Dickins | 920fc35 | 2005-11-21 21:32:17 -0800 | [diff] [blame] | 2049 | if (old_page) { |
Hugh Dickins | 920fc35 | 2005-11-21 21:32:17 -0800 | [diff] [blame] | 2050 | if (!PageAnon(old_page)) { |
| 2051 | dec_mm_counter(mm, file_rss); |
| 2052 | inc_mm_counter(mm, anon_rss); |
| 2053 | } |
| 2054 | } else |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 2055 | inc_mm_counter(mm, anon_rss); |
Ben Collins | eca3513 | 2005-11-29 11:45:26 -0800 | [diff] [blame] | 2056 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2057 | entry = mk_pte(new_page, vma->vm_page_prot); |
| 2058 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
Siddha, Suresh B | 4ce072f | 2006-09-29 01:58:42 -0700 | [diff] [blame] | 2059 | /* |
| 2060 | * Clear the pte entry and flush it first, before updating the |
| 2061 | * pte with the new entry. This will avoid a race condition |
| 2062 | * seen in the presence of one thread doing SMC and another |
| 2063 | * thread doing COW. |
| 2064 | */ |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 2065 | ptep_clear_flush_notify(vma, address, page_table); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 2066 | page_add_new_anon_rmap(new_page, vma, address); |
Lee Schermerhorn | 64d6519 | 2008-10-18 20:26:52 -0700 | [diff] [blame] | 2067 | set_pte_at(mm, address, page_table, entry); |
| 2068 | update_mmu_cache(vma, address, entry); |
Nick Piggin | 945754a | 2008-06-23 14:30:30 +0200 | [diff] [blame] | 2069 | if (old_page) { |
| 2070 | /* |
| 2071 | * Only after switching the pte to the new page may |
| 2072 | * we remove the mapcount here. Otherwise another |
| 2073 | * process may come and find the rmap count decremented |
| 2074 | * before the pte is switched to the new page, and |
| 2075 | * "reuse" the old page writing into it while our pte |
| 2076 | * here still points into it and can be read by other |
| 2077 | * threads. |
| 2078 | * |
| 2079 | * The critical issue is to order this |
| 2080 | * page_remove_rmap with the ptp_clear_flush above. |
| 2081 | * Those stores are ordered by (if nothing else,) |
| 2082 | * the barrier present in the atomic_add_negative |
| 2083 | * in page_remove_rmap. |
| 2084 | * |
| 2085 | * Then the TLB flush in ptep_clear_flush ensures that |
| 2086 | * no process can access the old page before the |
| 2087 | * decremented mapcount is visible. And the old page |
| 2088 | * cannot be reused until after the decremented |
| 2089 | * mapcount is visible. So transitively, TLBs to |
| 2090 | * old page will be flushed before it can be reused. |
| 2091 | */ |
Hugh Dickins | edc315f | 2009-01-06 14:40:11 -0800 | [diff] [blame] | 2092 | page_remove_rmap(old_page); |
Nick Piggin | 945754a | 2008-06-23 14:30:30 +0200 | [diff] [blame] | 2093 | } |
| 2094 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2095 | /* Free the old page.. */ |
| 2096 | new_page = old_page; |
Nick Piggin | f33ea7f | 2005-08-03 20:24:01 +1000 | [diff] [blame] | 2097 | ret |= VM_FAULT_WRITE; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2098 | } else |
| 2099 | mem_cgroup_uncharge_page(new_page); |
| 2100 | |
Hugh Dickins | 920fc35 | 2005-11-21 21:32:17 -0800 | [diff] [blame] | 2101 | if (new_page) |
| 2102 | page_cache_release(new_page); |
| 2103 | if (old_page) |
| 2104 | page_cache_release(old_page); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2105 | unlock: |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2106 | pte_unmap_unlock(page_table, ptl); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2107 | if (dirty_page) { |
Nick Piggin | 7935289 | 2007-07-19 01:47:22 -0700 | [diff] [blame] | 2108 | /* |
| 2109 | * Yes, Virginia, this is actually required to prevent a race |
| 2110 | * with clear_page_dirty_for_io() from clearing the page dirty |
| 2111 | * bit after it clear all dirty ptes, but before a racing |
| 2112 | * do_wp_page installs a dirty pte. |
| 2113 | * |
| 2114 | * do_no_page is protected similarly. |
| 2115 | */ |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 2116 | if (!page_mkwrite) { |
| 2117 | wait_on_page_locked(dirty_page); |
| 2118 | set_page_dirty_balance(dirty_page, page_mkwrite); |
| 2119 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2120 | put_page(dirty_page); |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 2121 | if (page_mkwrite) { |
| 2122 | struct address_space *mapping = dirty_page->mapping; |
| 2123 | |
| 2124 | set_page_dirty(dirty_page); |
| 2125 | unlock_page(dirty_page); |
| 2126 | page_cache_release(dirty_page); |
| 2127 | if (mapping) { |
| 2128 | /* |
| 2129 | * Some device drivers do not set page.mapping |
| 2130 | * but still dirty their pages |
| 2131 | */ |
| 2132 | balance_dirty_pages_ratelimited(mapping); |
| 2133 | } |
| 2134 | } |
| 2135 | |
| 2136 | /* file_update_time outside page_lock */ |
| 2137 | if (vma->vm_file) |
| 2138 | file_update_time(vma->vm_file); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2139 | } |
Nick Piggin | f33ea7f | 2005-08-03 20:24:01 +1000 | [diff] [blame] | 2140 | return ret; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2141 | oom_free_new: |
Hugh Dickins | 6dbf6d3 | 2008-03-04 14:29:04 -0800 | [diff] [blame] | 2142 | page_cache_release(new_page); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2143 | oom: |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 2144 | if (old_page) { |
| 2145 | if (page_mkwrite) { |
| 2146 | unlock_page(old_page); |
| 2147 | page_cache_release(old_page); |
| 2148 | } |
Hugh Dickins | 920fc35 | 2005-11-21 21:32:17 -0800 | [diff] [blame] | 2149 | page_cache_release(old_page); |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 2150 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2151 | return VM_FAULT_OOM; |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2152 | |
| 2153 | unwritable_page: |
| 2154 | page_cache_release(old_page); |
Nick Piggin | c2ec175 | 2009-03-31 15:23:21 -0700 | [diff] [blame] | 2155 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2156 | } |
| 2157 | |
| 2158 | /* |
| 2159 | * Helper functions for unmap_mapping_range(). |
| 2160 | * |
| 2161 | * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __ |
| 2162 | * |
| 2163 | * We have to restart searching the prio_tree whenever we drop the lock, |
| 2164 | * since the iterator is only valid while the lock is held, and anyway |
| 2165 | * a later vma might be split and reinserted earlier while lock dropped. |
| 2166 | * |
| 2167 | * The list of nonlinear vmas could be handled more efficiently, using |
| 2168 | * a placeholder, but handle it in the same way until a need is shown. |
| 2169 | * It is important to search the prio_tree before nonlinear list: a vma |
| 2170 | * may become nonlinear and be shifted from prio_tree to nonlinear list |
| 2171 | * while the lock is dropped; but never shifted from list to prio_tree. |
| 2172 | * |
| 2173 | * In order to make forward progress despite restarting the search, |
| 2174 | * vm_truncate_count is used to mark a vma as now dealt with, so we can |
| 2175 | * quickly skip it next time around. Since the prio_tree search only |
| 2176 | * shows us those vmas affected by unmapping the range in question, we |
| 2177 | * can't efficiently keep all vmas in step with mapping->truncate_count: |
| 2178 | * so instead reset them all whenever it wraps back to 0 (then go to 1). |
| 2179 | * mapping->truncate_count and vma->vm_truncate_count are protected by |
| 2180 | * i_mmap_lock. |
| 2181 | * |
| 2182 | * In order to make forward progress despite repeatedly restarting some |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 2183 | * large vma, note the restart_addr from unmap_vmas when it breaks out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2184 | * and restart from that address when we reach that vma again. It might |
| 2185 | * have been split or merged, shrunk or extended, but never shifted: so |
| 2186 | * restart_addr remains valid so long as it remains in the vma's range. |
| 2187 | * unmap_mapping_range forces truncate_count to leap over page-aligned |
| 2188 | * values so we can save vma's restart_addr in its truncate_count field. |
| 2189 | */ |
| 2190 | #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK)) |
| 2191 | |
| 2192 | static void reset_vma_truncate_counts(struct address_space *mapping) |
| 2193 | { |
| 2194 | struct vm_area_struct *vma; |
| 2195 | struct prio_tree_iter iter; |
| 2196 | |
| 2197 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX) |
| 2198 | vma->vm_truncate_count = 0; |
| 2199 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
| 2200 | vma->vm_truncate_count = 0; |
| 2201 | } |
| 2202 | |
| 2203 | static int unmap_mapping_range_vma(struct vm_area_struct *vma, |
| 2204 | unsigned long start_addr, unsigned long end_addr, |
| 2205 | struct zap_details *details) |
| 2206 | { |
| 2207 | unsigned long restart_addr; |
| 2208 | int need_break; |
| 2209 | |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2210 | /* |
| 2211 | * files that support invalidating or truncating portions of the |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2212 | * file from under mmaped areas must have their ->fault function |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2213 | * return a locked page (and set VM_FAULT_LOCKED in the return). |
| 2214 | * This provides synchronisation against concurrent unmapping here. |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2215 | */ |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2216 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2217 | again: |
| 2218 | restart_addr = vma->vm_truncate_count; |
| 2219 | if (is_restart_addr(restart_addr) && start_addr < restart_addr) { |
| 2220 | start_addr = restart_addr; |
| 2221 | if (start_addr >= end_addr) { |
| 2222 | /* Top of vma has been split off since last time */ |
| 2223 | vma->vm_truncate_count = details->truncate_count; |
| 2224 | return 0; |
| 2225 | } |
| 2226 | } |
| 2227 | |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 2228 | restart_addr = zap_page_range(vma, start_addr, |
| 2229 | end_addr - start_addr, details); |
Nick Piggin | 95c354f | 2008-01-30 13:31:20 +0100 | [diff] [blame] | 2230 | need_break = need_resched() || spin_needbreak(details->i_mmap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2231 | |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 2232 | if (restart_addr >= end_addr) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2233 | /* We have now completed this vma: mark it so */ |
| 2234 | vma->vm_truncate_count = details->truncate_count; |
| 2235 | if (!need_break) |
| 2236 | return 0; |
| 2237 | } else { |
| 2238 | /* Note restart_addr in vma's truncate_count field */ |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 2239 | vma->vm_truncate_count = restart_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2240 | if (!need_break) |
| 2241 | goto again; |
| 2242 | } |
| 2243 | |
| 2244 | spin_unlock(details->i_mmap_lock); |
| 2245 | cond_resched(); |
| 2246 | spin_lock(details->i_mmap_lock); |
| 2247 | return -EINTR; |
| 2248 | } |
| 2249 | |
| 2250 | static inline void unmap_mapping_range_tree(struct prio_tree_root *root, |
| 2251 | struct zap_details *details) |
| 2252 | { |
| 2253 | struct vm_area_struct *vma; |
| 2254 | struct prio_tree_iter iter; |
| 2255 | pgoff_t vba, vea, zba, zea; |
| 2256 | |
| 2257 | restart: |
| 2258 | vma_prio_tree_foreach(vma, &iter, root, |
| 2259 | details->first_index, details->last_index) { |
| 2260 | /* Skip quickly over those we have already dealt with */ |
| 2261 | if (vma->vm_truncate_count == details->truncate_count) |
| 2262 | continue; |
| 2263 | |
| 2264 | vba = vma->vm_pgoff; |
| 2265 | vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; |
| 2266 | /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ |
| 2267 | zba = details->first_index; |
| 2268 | if (zba < vba) |
| 2269 | zba = vba; |
| 2270 | zea = details->last_index; |
| 2271 | if (zea > vea) |
| 2272 | zea = vea; |
| 2273 | |
| 2274 | if (unmap_mapping_range_vma(vma, |
| 2275 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
| 2276 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, |
| 2277 | details) < 0) |
| 2278 | goto restart; |
| 2279 | } |
| 2280 | } |
| 2281 | |
| 2282 | static inline void unmap_mapping_range_list(struct list_head *head, |
| 2283 | struct zap_details *details) |
| 2284 | { |
| 2285 | struct vm_area_struct *vma; |
| 2286 | |
| 2287 | /* |
| 2288 | * In nonlinear VMAs there is no correspondence between virtual address |
| 2289 | * offset and file offset. So we must perform an exhaustive search |
| 2290 | * across *all* the pages in each nonlinear VMA, not just the pages |
| 2291 | * whose virtual address lies outside the file truncation point. |
| 2292 | */ |
| 2293 | restart: |
| 2294 | list_for_each_entry(vma, head, shared.vm_set.list) { |
| 2295 | /* Skip quickly over those we have already dealt with */ |
| 2296 | if (vma->vm_truncate_count == details->truncate_count) |
| 2297 | continue; |
| 2298 | details->nonlinear_vma = vma; |
| 2299 | if (unmap_mapping_range_vma(vma, vma->vm_start, |
| 2300 | vma->vm_end, details) < 0) |
| 2301 | goto restart; |
| 2302 | } |
| 2303 | } |
| 2304 | |
| 2305 | /** |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 2306 | * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file. |
Martin Waitz | 3d41088 | 2005-06-23 22:05:21 -0700 | [diff] [blame] | 2307 | * @mapping: the address space containing mmaps to be unmapped. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2308 | * @holebegin: byte in first page to unmap, relative to the start of |
| 2309 | * the underlying file. This will be rounded down to a PAGE_SIZE |
| 2310 | * boundary. Note that this is different from vmtruncate(), which |
| 2311 | * must keep the partial page. In contrast, we must get rid of |
| 2312 | * partial pages. |
| 2313 | * @holelen: size of prospective hole in bytes. This will be rounded |
| 2314 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the |
| 2315 | * end of the file. |
| 2316 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; |
| 2317 | * but 0 when invalidating pagecache, don't throw away private data. |
| 2318 | */ |
| 2319 | void unmap_mapping_range(struct address_space *mapping, |
| 2320 | loff_t const holebegin, loff_t const holelen, int even_cows) |
| 2321 | { |
| 2322 | struct zap_details details; |
| 2323 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
| 2324 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 2325 | |
| 2326 | /* Check for overflow. */ |
| 2327 | if (sizeof(holelen) > sizeof(hlen)) { |
| 2328 | long long holeend = |
| 2329 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 2330 | if (holeend & ~(long long)ULONG_MAX) |
| 2331 | hlen = ULONG_MAX - hba + 1; |
| 2332 | } |
| 2333 | |
| 2334 | details.check_mapping = even_cows? NULL: mapping; |
| 2335 | details.nonlinear_vma = NULL; |
| 2336 | details.first_index = hba; |
| 2337 | details.last_index = hba + hlen - 1; |
| 2338 | if (details.last_index < details.first_index) |
| 2339 | details.last_index = ULONG_MAX; |
| 2340 | details.i_mmap_lock = &mapping->i_mmap_lock; |
| 2341 | |
| 2342 | spin_lock(&mapping->i_mmap_lock); |
| 2343 | |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2344 | /* Protect against endless unmapping loops */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2345 | mapping->truncate_count++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2346 | if (unlikely(is_restart_addr(mapping->truncate_count))) { |
| 2347 | if (mapping->truncate_count == 0) |
| 2348 | reset_vma_truncate_counts(mapping); |
| 2349 | mapping->truncate_count++; |
| 2350 | } |
| 2351 | details.truncate_count = mapping->truncate_count; |
| 2352 | |
| 2353 | if (unlikely(!prio_tree_empty(&mapping->i_mmap))) |
| 2354 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
| 2355 | if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) |
| 2356 | unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); |
| 2357 | spin_unlock(&mapping->i_mmap_lock); |
| 2358 | } |
| 2359 | EXPORT_SYMBOL(unmap_mapping_range); |
| 2360 | |
Rolf Eike Beer | bfa5bf6 | 2006-09-25 23:31:22 -0700 | [diff] [blame] | 2361 | /** |
| 2362 | * vmtruncate - unmap mappings "freed" by truncate() syscall |
| 2363 | * @inode: inode of the file used |
| 2364 | * @offset: file offset to start truncating |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2365 | * |
| 2366 | * NOTE! We have to be ready to update the memory sharing |
| 2367 | * between the file and the memory map for a potential last |
| 2368 | * incomplete page. Ugly, but necessary. |
| 2369 | */ |
| 2370 | int vmtruncate(struct inode * inode, loff_t offset) |
| 2371 | { |
Christoph Hellwig | 61d5048 | 2008-02-04 22:28:56 -0800 | [diff] [blame] | 2372 | if (inode->i_size < offset) { |
| 2373 | unsigned long limit; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2374 | |
Christoph Hellwig | 61d5048 | 2008-02-04 22:28:56 -0800 | [diff] [blame] | 2375 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; |
| 2376 | if (limit != RLIM_INFINITY && offset > limit) |
| 2377 | goto out_sig; |
| 2378 | if (offset > inode->i_sb->s_maxbytes) |
| 2379 | goto out_big; |
| 2380 | i_size_write(inode, offset); |
| 2381 | } else { |
| 2382 | struct address_space *mapping = inode->i_mapping; |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2383 | |
Christoph Hellwig | 61d5048 | 2008-02-04 22:28:56 -0800 | [diff] [blame] | 2384 | /* |
| 2385 | * truncation of in-use swapfiles is disallowed - it would |
| 2386 | * cause subsequent swapout to scribble on the now-freed |
| 2387 | * blocks. |
| 2388 | */ |
| 2389 | if (IS_SWAPFILE(inode)) |
| 2390 | return -ETXTBSY; |
| 2391 | i_size_write(inode, offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2392 | |
Christoph Hellwig | 61d5048 | 2008-02-04 22:28:56 -0800 | [diff] [blame] | 2393 | /* |
| 2394 | * unmap_mapping_range is called twice, first simply for |
| 2395 | * efficiency so that truncate_inode_pages does fewer |
| 2396 | * single-page unmaps. However after this first call, and |
| 2397 | * before truncate_inode_pages finishes, it is possible for |
| 2398 | * private pages to be COWed, which remain after |
| 2399 | * truncate_inode_pages finishes, hence the second |
| 2400 | * unmap_mapping_range call must be made for correctness. |
| 2401 | */ |
| 2402 | unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); |
| 2403 | truncate_inode_pages(mapping, offset); |
| 2404 | unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); |
| 2405 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2406 | |
Al Viro | acfa438 | 2008-12-04 10:06:33 -0500 | [diff] [blame] | 2407 | if (inode->i_op->truncate) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2408 | inode->i_op->truncate(inode); |
| 2409 | return 0; |
Christoph Hellwig | 61d5048 | 2008-02-04 22:28:56 -0800 | [diff] [blame] | 2410 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2411 | out_sig: |
| 2412 | send_sig(SIGXFSZ, current, 0); |
| 2413 | out_big: |
| 2414 | return -EFBIG; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2415 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2416 | EXPORT_SYMBOL(vmtruncate); |
| 2417 | |
Badari Pulavarty | f6b3ec2 | 2006-01-06 00:10:38 -0800 | [diff] [blame] | 2418 | int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end) |
| 2419 | { |
| 2420 | struct address_space *mapping = inode->i_mapping; |
| 2421 | |
| 2422 | /* |
| 2423 | * If the underlying filesystem is not going to provide |
| 2424 | * a way to truncate a range of blocks (punch a hole) - |
| 2425 | * we should return failure right now. |
| 2426 | */ |
Al Viro | acfa438 | 2008-12-04 10:06:33 -0500 | [diff] [blame] | 2427 | if (!inode->i_op->truncate_range) |
Badari Pulavarty | f6b3ec2 | 2006-01-06 00:10:38 -0800 | [diff] [blame] | 2428 | return -ENOSYS; |
| 2429 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 2430 | mutex_lock(&inode->i_mutex); |
Badari Pulavarty | f6b3ec2 | 2006-01-06 00:10:38 -0800 | [diff] [blame] | 2431 | down_write(&inode->i_alloc_sem); |
| 2432 | unmap_mapping_range(mapping, offset, (end - offset), 1); |
| 2433 | truncate_inode_pages_range(mapping, offset, end); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2434 | unmap_mapping_range(mapping, offset, (end - offset), 1); |
Badari Pulavarty | f6b3ec2 | 2006-01-06 00:10:38 -0800 | [diff] [blame] | 2435 | inode->i_op->truncate_range(inode, offset, end); |
| 2436 | up_write(&inode->i_alloc_sem); |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 2437 | mutex_unlock(&inode->i_mutex); |
Badari Pulavarty | f6b3ec2 | 2006-01-06 00:10:38 -0800 | [diff] [blame] | 2438 | |
| 2439 | return 0; |
| 2440 | } |
Badari Pulavarty | f6b3ec2 | 2006-01-06 00:10:38 -0800 | [diff] [blame] | 2441 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2442 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2443 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 2444 | * but allow concurrent faults), and pte mapped but not yet locked. |
| 2445 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2446 | */ |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2447 | static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, |
| 2448 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
| 2449 | int write_access, pte_t orig_pte) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2450 | { |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2451 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2452 | struct page *page; |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2453 | swp_entry_t entry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2454 | pte_t pte; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2455 | struct mem_cgroup *ptr = NULL; |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2456 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2457 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 2458 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2459 | goto out; |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2460 | |
| 2461 | entry = pte_to_swp_entry(orig_pte); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 2462 | if (is_migration_entry(entry)) { |
| 2463 | migration_entry_wait(mm, pmd, address); |
| 2464 | goto out; |
| 2465 | } |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 2466 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2467 | page = lookup_swap_cache(entry); |
| 2468 | if (!page) { |
Ashwin Chaugule | 098fe65 | 2006-12-06 20:31:54 -0800 | [diff] [blame] | 2469 | grab_swap_token(); /* Contend for token _before_ read-in */ |
Hugh Dickins | 02098fe | 2008-02-04 22:28:42 -0800 | [diff] [blame] | 2470 | page = swapin_readahead(entry, |
| 2471 | GFP_HIGHUSER_MOVABLE, vma, address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2472 | if (!page) { |
| 2473 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2474 | * Back out if somebody else faulted in this pte |
| 2475 | * while we released the pte lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2476 | */ |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2477 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2478 | if (likely(pte_same(*page_table, orig_pte))) |
| 2479 | ret = VM_FAULT_OOM; |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 2480 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2481 | goto unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2482 | } |
| 2483 | |
| 2484 | /* Had to read the page from swap area: Major fault */ |
| 2485 | ret = VM_FAULT_MAJOR; |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2486 | count_vm_event(PGMAJFAULT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2487 | } |
| 2488 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2489 | lock_page(page); |
Balbir Singh | 20a1022 | 2007-11-14 17:00:33 -0800 | [diff] [blame] | 2490 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2491 | |
KAMEZAWA Hiroyuki | 2c26fdd | 2009-01-07 18:08:10 -0800 | [diff] [blame] | 2492 | if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) { |
KAMEZAWA Hiroyuki | 073e587 | 2008-10-18 20:28:08 -0700 | [diff] [blame] | 2493 | ret = VM_FAULT_OOM; |
Johannes Weiner | bc43f75 | 2009-04-30 15:08:08 -0700 | [diff] [blame] | 2494 | goto out_page; |
KAMEZAWA Hiroyuki | 073e587 | 2008-10-18 20:28:08 -0700 | [diff] [blame] | 2495 | } |
| 2496 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2497 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2498 | * Back out if somebody else already faulted in this pte. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2499 | */ |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2500 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
Hugh Dickins | 9e9bef0 | 2005-10-29 18:16:15 -0700 | [diff] [blame] | 2501 | if (unlikely(!pte_same(*page_table, orig_pte))) |
Kirill Korotaev | b810748 | 2005-05-16 21:53:50 -0700 | [diff] [blame] | 2502 | goto out_nomap; |
Kirill Korotaev | b810748 | 2005-05-16 21:53:50 -0700 | [diff] [blame] | 2503 | |
| 2504 | if (unlikely(!PageUptodate(page))) { |
| 2505 | ret = VM_FAULT_SIGBUS; |
| 2506 | goto out_nomap; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2507 | } |
| 2508 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2509 | /* |
| 2510 | * The page isn't present yet, go ahead with the fault. |
| 2511 | * |
| 2512 | * Be careful about the sequence of operations here. |
| 2513 | * To get its accounting right, reuse_swap_page() must be called |
| 2514 | * while the page is counted on swap but not yet in mapcount i.e. |
| 2515 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() |
| 2516 | * must be called after the swap_free(), or it will never succeed. |
KAMEZAWA Hiroyuki | 03f3c43 | 2009-01-07 18:08:31 -0800 | [diff] [blame] | 2517 | * Because delete_from_swap_page() may be called by reuse_swap_page(), |
| 2518 | * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry |
| 2519 | * in page->private. In this case, a record in swap_cgroup is silently |
| 2520 | * discarded at swap_free(). |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2521 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2522 | |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 2523 | inc_mm_counter(mm, anon_rss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2524 | pte = mk_pte(page, vma->vm_page_prot); |
Hugh Dickins | 7b1fe59 | 2009-01-06 14:39:34 -0800 | [diff] [blame] | 2525 | if (write_access && reuse_swap_page(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2526 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
| 2527 | write_access = 0; |
| 2528 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2529 | flush_icache_page(vma, page); |
| 2530 | set_pte_at(mm, address, page_table, pte); |
| 2531 | page_add_anon_rmap(page, vma, address); |
KAMEZAWA Hiroyuki | 03f3c43 | 2009-01-07 18:08:31 -0800 | [diff] [blame] | 2532 | /* It's better to call commit-charge after rmap is established */ |
| 2533 | mem_cgroup_commit_charge_swapin(page, ptr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2534 | |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 2535 | swap_free(entry); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 2536 | if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 2537 | try_to_free_swap(page); |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 2538 | unlock_page(page); |
| 2539 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2540 | if (write_access) { |
Hugh Dickins | 61469f1 | 2008-03-04 14:29:04 -0800 | [diff] [blame] | 2541 | ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); |
| 2542 | if (ret & VM_FAULT_ERROR) |
| 2543 | ret &= VM_FAULT_ERROR; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2544 | goto out; |
| 2545 | } |
| 2546 | |
| 2547 | /* No need to invalidate - it was non-present before */ |
| 2548 | update_mmu_cache(vma, address, pte); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2549 | unlock: |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2550 | pte_unmap_unlock(page_table, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2551 | out: |
| 2552 | return ret; |
Kirill Korotaev | b810748 | 2005-05-16 21:53:50 -0700 | [diff] [blame] | 2553 | out_nomap: |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2554 | mem_cgroup_cancel_charge_swapin(ptr); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2555 | pte_unmap_unlock(page_table, ptl); |
Johannes Weiner | bc43f75 | 2009-04-30 15:08:08 -0700 | [diff] [blame] | 2556 | out_page: |
Kirill Korotaev | b810748 | 2005-05-16 21:53:50 -0700 | [diff] [blame] | 2557 | unlock_page(page); |
| 2558 | page_cache_release(page); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2559 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2560 | } |
| 2561 | |
| 2562 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2563 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 2564 | * but allow concurrent faults), and pte mapped but not yet locked. |
| 2565 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2566 | */ |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2567 | static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, |
| 2568 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
| 2569 | int write_access) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2570 | { |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2571 | struct page *page; |
| 2572 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2573 | pte_t entry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2574 | |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2575 | /* Allocate our own private page. */ |
| 2576 | pte_unmap(page_table); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2577 | |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2578 | if (unlikely(anon_vma_prepare(vma))) |
| 2579 | goto oom; |
| 2580 | page = alloc_zeroed_user_highpage_movable(vma, address); |
| 2581 | if (!page) |
| 2582 | goto oom; |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 2583 | __SetPageUptodate(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2584 | |
KAMEZAWA Hiroyuki | 2c26fdd | 2009-01-07 18:08:10 -0800 | [diff] [blame] | 2585 | if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2586 | goto oom_free_page; |
| 2587 | |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2588 | entry = mk_pte(page, vma->vm_page_prot); |
| 2589 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2590 | |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2591 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
| 2592 | if (!pte_none(*page_table)) |
| 2593 | goto release; |
| 2594 | inc_mm_counter(mm, anon_rss); |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2595 | page_add_new_anon_rmap(page, vma, address); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2596 | set_pte_at(mm, address, page_table, entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2597 | |
| 2598 | /* No need to invalidate - it was non-present before */ |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2599 | update_mmu_cache(vma, address, entry); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2600 | unlock: |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2601 | pte_unmap_unlock(page_table, ptl); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2602 | return 0; |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2603 | release: |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2604 | mem_cgroup_uncharge_page(page); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2605 | page_cache_release(page); |
| 2606 | goto unlock; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2607 | oom_free_page: |
Hugh Dickins | 6dbf6d3 | 2008-03-04 14:29:04 -0800 | [diff] [blame] | 2608 | page_cache_release(page); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2609 | oom: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2610 | return VM_FAULT_OOM; |
| 2611 | } |
| 2612 | |
| 2613 | /* |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2614 | * __do_fault() tries to create a new page mapping. It aggressively |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2615 | * tries to share with existing pages, but makes a separate copy if |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2616 | * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid |
| 2617 | * the next page fault. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2618 | * |
| 2619 | * As this is called only for pages that do not currently exist, we |
| 2620 | * do not need to flush old virtual caches or the TLB. |
| 2621 | * |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2622 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
Hugh Dickins | 16abfa0 | 2007-10-04 16:56:06 +0100 | [diff] [blame] | 2623 | * but allow concurrent faults), and pte neither mapped nor locked. |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2624 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2625 | */ |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2626 | static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
Hugh Dickins | 16abfa0 | 2007-10-04 16:56:06 +0100 | [diff] [blame] | 2627 | unsigned long address, pmd_t *pmd, |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2628 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2629 | { |
Hugh Dickins | 16abfa0 | 2007-10-04 16:56:06 +0100 | [diff] [blame] | 2630 | pte_t *page_table; |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2631 | spinlock_t *ptl; |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2632 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2633 | pte_t entry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2634 | int anon = 0; |
KAMEZAWA Hiroyuki | 5b4e655 | 2008-10-18 20:28:10 -0700 | [diff] [blame] | 2635 | int charged = 0; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2636 | struct page *dirty_page = NULL; |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2637 | struct vm_fault vmf; |
| 2638 | int ret; |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 2639 | int page_mkwrite = 0; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2640 | |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2641 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); |
| 2642 | vmf.pgoff = pgoff; |
| 2643 | vmf.flags = flags; |
| 2644 | vmf.page = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2645 | |
Nick Piggin | 3c18ddd | 2008-04-28 02:12:10 -0700 | [diff] [blame] | 2646 | ret = vma->vm_ops->fault(vma, &vmf); |
| 2647 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
| 2648 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2649 | |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2650 | /* |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2651 | * For consistency in subsequent calls, make the faulted page always |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2652 | * locked. |
| 2653 | */ |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2654 | if (unlikely(!(ret & VM_FAULT_LOCKED))) |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2655 | lock_page(vmf.page); |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2656 | else |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2657 | VM_BUG_ON(!PageLocked(vmf.page)); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2658 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2659 | /* |
| 2660 | * Should we do an early C-O-W break? |
| 2661 | */ |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2662 | page = vmf.page; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2663 | if (flags & FAULT_FLAG_WRITE) { |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2664 | if (!(vma->vm_flags & VM_SHARED)) { |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2665 | anon = 1; |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2666 | if (unlikely(anon_vma_prepare(vma))) { |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2667 | ret = VM_FAULT_OOM; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2668 | goto out; |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2669 | } |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2670 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, |
| 2671 | vma, address); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2672 | if (!page) { |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2673 | ret = VM_FAULT_OOM; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2674 | goto out; |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2675 | } |
KAMEZAWA Hiroyuki | 2c26fdd | 2009-01-07 18:08:10 -0800 | [diff] [blame] | 2676 | if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) { |
KAMEZAWA Hiroyuki | 5b4e655 | 2008-10-18 20:28:10 -0700 | [diff] [blame] | 2677 | ret = VM_FAULT_OOM; |
| 2678 | page_cache_release(page); |
| 2679 | goto out; |
| 2680 | } |
| 2681 | charged = 1; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 2682 | /* |
| 2683 | * Don't let another task, with possibly unlocked vma, |
| 2684 | * keep the mlocked page. |
| 2685 | */ |
| 2686 | if (vma->vm_flags & VM_LOCKED) |
| 2687 | clear_page_mlock(vmf.page); |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2688 | copy_user_highpage(page, vmf.page, address, vma); |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 2689 | __SetPageUptodate(page); |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2690 | } else { |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2691 | /* |
| 2692 | * If the page will be shareable, see if the backing |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2693 | * address space wants to know that the page is about |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2694 | * to become writable |
| 2695 | */ |
Mark Fasheh | 6967614 | 2007-07-19 01:47:00 -0700 | [diff] [blame] | 2696 | if (vma->vm_ops->page_mkwrite) { |
Nick Piggin | c2ec175 | 2009-03-31 15:23:21 -0700 | [diff] [blame] | 2697 | int tmp; |
| 2698 | |
Mark Fasheh | 6967614 | 2007-07-19 01:47:00 -0700 | [diff] [blame] | 2699 | unlock_page(page); |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 2700 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
Nick Piggin | c2ec175 | 2009-03-31 15:23:21 -0700 | [diff] [blame] | 2701 | tmp = vma->vm_ops->page_mkwrite(vma, &vmf); |
| 2702 | if (unlikely(tmp & |
| 2703 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
| 2704 | ret = tmp; |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 2705 | goto unwritable_page; |
Mark Fasheh | 6967614 | 2007-07-19 01:47:00 -0700 | [diff] [blame] | 2706 | } |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 2707 | if (unlikely(!(tmp & VM_FAULT_LOCKED))) { |
| 2708 | lock_page(page); |
| 2709 | if (!page->mapping) { |
| 2710 | ret = 0; /* retry the fault */ |
| 2711 | unlock_page(page); |
| 2712 | goto unwritable_page; |
| 2713 | } |
| 2714 | } else |
| 2715 | VM_BUG_ON(!PageLocked(page)); |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 2716 | page_mkwrite = 1; |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2717 | } |
| 2718 | } |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2719 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2720 | } |
| 2721 | |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2722 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2723 | |
| 2724 | /* |
| 2725 | * This silly early PAGE_DIRTY setting removes a race |
| 2726 | * due to the bad i386 page protection. But it's valid |
| 2727 | * for other architectures too. |
| 2728 | * |
| 2729 | * Note that if write_access is true, we either now have |
| 2730 | * an exclusive copy of the page, or this is a shared mapping, |
| 2731 | * so we can make it writable and dirty to avoid having to |
| 2732 | * handle that later. |
| 2733 | */ |
| 2734 | /* Only go through if we didn't race with anybody else... */ |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2735 | if (likely(pte_same(*page_table, orig_pte))) { |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2736 | flush_icache_page(vma, page); |
| 2737 | entry = mk_pte(page, vma->vm_page_prot); |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2738 | if (flags & FAULT_FLAG_WRITE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2739 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2740 | if (anon) { |
Lee Schermerhorn | 64d6519 | 2008-10-18 20:26:52 -0700 | [diff] [blame] | 2741 | inc_mm_counter(mm, anon_rss); |
Lee Schermerhorn | 64d6519 | 2008-10-18 20:26:52 -0700 | [diff] [blame] | 2742 | page_add_new_anon_rmap(page, vma, address); |
Hugh Dickins | f57e88a | 2005-11-21 21:32:19 -0800 | [diff] [blame] | 2743 | } else { |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 2744 | inc_mm_counter(mm, file_rss); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2745 | page_add_file_rmap(page); |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2746 | if (flags & FAULT_FLAG_WRITE) { |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2747 | dirty_page = page; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2748 | get_page(dirty_page); |
| 2749 | } |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 2750 | } |
Lee Schermerhorn | 64d6519 | 2008-10-18 20:26:52 -0700 | [diff] [blame] | 2751 | set_pte_at(mm, address, page_table, entry); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2752 | |
| 2753 | /* no need to invalidate: a not-present page won't be cached */ |
| 2754 | update_mmu_cache(vma, address, entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2755 | } else { |
KAMEZAWA Hiroyuki | 5b4e655 | 2008-10-18 20:28:10 -0700 | [diff] [blame] | 2756 | if (charged) |
| 2757 | mem_cgroup_uncharge_page(page); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2758 | if (anon) |
| 2759 | page_cache_release(page); |
| 2760 | else |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2761 | anon = 1; /* no anon but release faulted_page */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2762 | } |
| 2763 | |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2764 | pte_unmap_unlock(page_table, ptl); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2765 | |
| 2766 | out: |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 2767 | if (dirty_page) { |
| 2768 | struct address_space *mapping = page->mapping; |
| 2769 | |
| 2770 | if (set_page_dirty(dirty_page)) |
| 2771 | page_mkwrite = 1; |
| 2772 | unlock_page(dirty_page); |
| 2773 | put_page(dirty_page); |
| 2774 | if (page_mkwrite && mapping) { |
| 2775 | /* |
| 2776 | * Some device drivers do not set page.mapping but still |
| 2777 | * dirty their pages |
| 2778 | */ |
| 2779 | balance_dirty_pages_ratelimited(mapping); |
| 2780 | } |
| 2781 | |
| 2782 | /* file_update_time outside page_lock */ |
Anton Salikhmetov | 8f7b3d1 | 2008-01-23 02:21:18 +0300 | [diff] [blame] | 2783 | if (vma->vm_file) |
| 2784 | file_update_time(vma->vm_file); |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 2785 | } else { |
| 2786 | unlock_page(vmf.page); |
| 2787 | if (anon) |
| 2788 | page_cache_release(vmf.page); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2789 | } |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2790 | |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2791 | return ret; |
Nick Piggin | b827e49 | 2009-04-30 15:08:16 -0700 | [diff] [blame] | 2792 | |
| 2793 | unwritable_page: |
| 2794 | page_cache_release(page); |
| 2795 | return ret; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2796 | } |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2797 | |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2798 | static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
| 2799 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
| 2800 | int write_access, pte_t orig_pte) |
| 2801 | { |
| 2802 | pgoff_t pgoff = (((address & PAGE_MASK) |
Dean Nelson | 0da7e01 | 2007-10-16 01:24:45 -0700 | [diff] [blame] | 2803 | - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2804 | unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0); |
| 2805 | |
Hugh Dickins | 16abfa0 | 2007-10-04 16:56:06 +0100 | [diff] [blame] | 2806 | pte_unmap(page_table); |
| 2807 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2808 | } |
| 2809 | |
Jes Sorensen | f4b8180 | 2006-09-27 01:50:10 -0700 | [diff] [blame] | 2810 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2811 | * Fault of a previously existing named mapping. Repopulate the pte |
| 2812 | * from the encoded file_pte if possible. This enables swappable |
| 2813 | * nonlinear vmas. |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2814 | * |
| 2815 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 2816 | * but allow concurrent faults), and pte mapped but not yet locked. |
| 2817 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2818 | */ |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2819 | static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2820 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
| 2821 | int write_access, pte_t orig_pte) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2822 | { |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2823 | unsigned int flags = FAULT_FLAG_NONLINEAR | |
| 2824 | (write_access ? FAULT_FLAG_WRITE : 0); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2825 | pgoff_t pgoff; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2826 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 2827 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2828 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2829 | |
Hugh Dickins | 2509ef2 | 2009-01-06 14:40:10 -0800 | [diff] [blame] | 2830 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) { |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2831 | /* |
| 2832 | * Page table corrupted: show pte and kill process. |
| 2833 | */ |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 2834 | print_bad_pte(vma, address, orig_pte, NULL); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2835 | return VM_FAULT_OOM; |
| 2836 | } |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2837 | |
| 2838 | pgoff = pte_to_pgoff(orig_pte); |
Hugh Dickins | 16abfa0 | 2007-10-04 16:56:06 +0100 | [diff] [blame] | 2839 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2840 | } |
| 2841 | |
| 2842 | /* |
| 2843 | * These routines also need to handle stuff like marking pages dirty |
| 2844 | * and/or accessed for architectures that don't do it in hardware (most |
| 2845 | * RISC architectures). The early dirtying is also good on the i386. |
| 2846 | * |
| 2847 | * There is also a hook called "update_mmu_cache()" that architectures |
| 2848 | * with external mmu caches can use to update those (ie the Sparc or |
| 2849 | * PowerPC hashed page tables that act as extended TLBs). |
| 2850 | * |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2851 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 2852 | * but allow concurrent faults), and pte mapped but not yet locked. |
| 2853 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2854 | */ |
| 2855 | static inline int handle_pte_fault(struct mm_struct *mm, |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2856 | struct vm_area_struct *vma, unsigned long address, |
| 2857 | pte_t *pte, pmd_t *pmd, int write_access) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2858 | { |
| 2859 | pte_t entry; |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2860 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2861 | |
Benjamin Herrenschmidt | 8dab524 | 2007-06-16 10:16:12 -0700 | [diff] [blame] | 2862 | entry = *pte; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2863 | if (!pte_present(entry)) { |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2864 | if (pte_none(entry)) { |
Jes Sorensen | f4b8180 | 2006-09-27 01:50:10 -0700 | [diff] [blame] | 2865 | if (vma->vm_ops) { |
Nick Piggin | 3c18ddd | 2008-04-28 02:12:10 -0700 | [diff] [blame] | 2866 | if (likely(vma->vm_ops->fault)) |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2867 | return do_linear_fault(mm, vma, address, |
| 2868 | pte, pmd, write_access, entry); |
Jes Sorensen | f4b8180 | 2006-09-27 01:50:10 -0700 | [diff] [blame] | 2869 | } |
| 2870 | return do_anonymous_page(mm, vma, address, |
| 2871 | pte, pmd, write_access); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2872 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2873 | if (pte_file(entry)) |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2874 | return do_nonlinear_fault(mm, vma, address, |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2875 | pte, pmd, write_access, entry); |
| 2876 | return do_swap_page(mm, vma, address, |
| 2877 | pte, pmd, write_access, entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2878 | } |
| 2879 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 2880 | ptl = pte_lockptr(mm, pmd); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2881 | spin_lock(ptl); |
| 2882 | if (unlikely(!pte_same(*pte, entry))) |
| 2883 | goto unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2884 | if (write_access) { |
| 2885 | if (!pte_write(entry)) |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2886 | return do_wp_page(mm, vma, address, |
| 2887 | pte, pmd, ptl, entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2888 | entry = pte_mkdirty(entry); |
| 2889 | } |
| 2890 | entry = pte_mkyoung(entry); |
Benjamin Herrenschmidt | 8dab524 | 2007-06-16 10:16:12 -0700 | [diff] [blame] | 2891 | if (ptep_set_access_flags(vma, address, pte, entry, write_access)) { |
Andrea Arcangeli | 1a44e14 | 2005-10-29 18:16:48 -0700 | [diff] [blame] | 2892 | update_mmu_cache(vma, address, entry); |
Andrea Arcangeli | 1a44e14 | 2005-10-29 18:16:48 -0700 | [diff] [blame] | 2893 | } else { |
| 2894 | /* |
| 2895 | * This is needed only for protection faults but the arch code |
| 2896 | * is not yet telling us if this is a protection fault or not. |
| 2897 | * This still avoids useless tlb flushes for .text page faults |
| 2898 | * with threads. |
| 2899 | */ |
| 2900 | if (write_access) |
| 2901 | flush_tlb_page(vma, address); |
| 2902 | } |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2903 | unlock: |
| 2904 | pte_unmap_unlock(pte, ptl); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2905 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2906 | } |
| 2907 | |
| 2908 | /* |
| 2909 | * By the time we get here, we already hold the mm semaphore |
| 2910 | */ |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2911 | int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2912 | unsigned long address, int write_access) |
| 2913 | { |
| 2914 | pgd_t *pgd; |
| 2915 | pud_t *pud; |
| 2916 | pmd_t *pmd; |
| 2917 | pte_t *pte; |
| 2918 | |
| 2919 | __set_current_state(TASK_RUNNING); |
| 2920 | |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2921 | count_vm_event(PGFAULT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2922 | |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2923 | if (unlikely(is_vm_hugetlb_page(vma))) |
| 2924 | return hugetlb_fault(mm, vma, address, write_access); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2925 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2926 | pgd = pgd_offset(mm, address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2927 | pud = pud_alloc(mm, pgd, address); |
| 2928 | if (!pud) |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2929 | return VM_FAULT_OOM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2930 | pmd = pmd_alloc(mm, pud, address); |
| 2931 | if (!pmd) |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2932 | return VM_FAULT_OOM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2933 | pte = pte_alloc_map(mm, pmd, address); |
| 2934 | if (!pte) |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2935 | return VM_FAULT_OOM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2936 | |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2937 | return handle_pte_fault(mm, vma, address, pte, pmd, write_access); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2938 | } |
| 2939 | |
| 2940 | #ifndef __PAGETABLE_PUD_FOLDED |
| 2941 | /* |
| 2942 | * Allocate page upper directory. |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 2943 | * We've already handled the fast-path in-line. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2944 | */ |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2945 | int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2946 | { |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2947 | pud_t *new = pud_alloc_one(mm, address); |
| 2948 | if (!new) |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2949 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2950 | |
Nick Piggin | 362a61a | 2008-05-14 06:37:36 +0200 | [diff] [blame] | 2951 | smp_wmb(); /* See comment in __pte_alloc */ |
| 2952 | |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 2953 | spin_lock(&mm->page_table_lock); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2954 | if (pgd_present(*pgd)) /* Another has populated it */ |
Benjamin Herrenschmidt | 5e54197 | 2008-02-04 22:29:14 -0800 | [diff] [blame] | 2955 | pud_free(mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2956 | else |
| 2957 | pgd_populate(mm, pgd, new); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2958 | spin_unlock(&mm->page_table_lock); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2959 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2960 | } |
| 2961 | #endif /* __PAGETABLE_PUD_FOLDED */ |
| 2962 | |
| 2963 | #ifndef __PAGETABLE_PMD_FOLDED |
| 2964 | /* |
| 2965 | * Allocate page middle directory. |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 2966 | * We've already handled the fast-path in-line. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2967 | */ |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2968 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2969 | { |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2970 | pmd_t *new = pmd_alloc_one(mm, address); |
| 2971 | if (!new) |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2972 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2973 | |
Nick Piggin | 362a61a | 2008-05-14 06:37:36 +0200 | [diff] [blame] | 2974 | smp_wmb(); /* See comment in __pte_alloc */ |
| 2975 | |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 2976 | spin_lock(&mm->page_table_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2977 | #ifndef __ARCH_HAS_4LEVEL_HACK |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2978 | if (pud_present(*pud)) /* Another has populated it */ |
Benjamin Herrenschmidt | 5e54197 | 2008-02-04 22:29:14 -0800 | [diff] [blame] | 2979 | pmd_free(mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2980 | else |
| 2981 | pud_populate(mm, pud, new); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2982 | #else |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2983 | if (pgd_present(*pud)) /* Another has populated it */ |
Benjamin Herrenschmidt | 5e54197 | 2008-02-04 22:29:14 -0800 | [diff] [blame] | 2984 | pmd_free(mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2985 | else |
| 2986 | pgd_populate(mm, pud, new); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2987 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2988 | spin_unlock(&mm->page_table_lock); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2989 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2990 | } |
| 2991 | #endif /* __PAGETABLE_PMD_FOLDED */ |
| 2992 | |
| 2993 | int make_pages_present(unsigned long addr, unsigned long end) |
| 2994 | { |
| 2995 | int ret, len, write; |
| 2996 | struct vm_area_struct * vma; |
| 2997 | |
| 2998 | vma = find_vma(current->mm, addr); |
| 2999 | if (!vma) |
KOSAKI Motohiro | a477097 | 2008-08-04 13:41:14 -0700 | [diff] [blame] | 3000 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3001 | write = (vma->vm_flags & VM_WRITE) != 0; |
Eric Sesterhenn | 5bcb28b | 2006-03-26 18:30:52 +0200 | [diff] [blame] | 3002 | BUG_ON(addr >= end); |
| 3003 | BUG_ON(end > vma->vm_end); |
Rolf Eike Beer | 68e116a | 2007-07-15 23:38:03 -0700 | [diff] [blame] | 3004 | len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3005 | ret = get_user_pages(current, current->mm, addr, |
| 3006 | len, write, 0, NULL, NULL); |
Lee Schermerhorn | c11d69d | 2008-10-18 20:26:56 -0700 | [diff] [blame] | 3007 | if (ret < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3008 | return ret; |
Lee Schermerhorn | 9978ad5 | 2008-10-18 20:26:56 -0700 | [diff] [blame] | 3009 | return ret == len ? 0 : -EFAULT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3010 | } |
| 3011 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3012 | #if !defined(__HAVE_ARCH_GATE_AREA) |
| 3013 | |
| 3014 | #if defined(AT_SYSINFO_EHDR) |
Adrian Bunk | 5ce7852 | 2005-09-10 00:26:28 -0700 | [diff] [blame] | 3015 | static struct vm_area_struct gate_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3016 | |
| 3017 | static int __init gate_vma_init(void) |
| 3018 | { |
| 3019 | gate_vma.vm_mm = NULL; |
| 3020 | gate_vma.vm_start = FIXADDR_USER_START; |
| 3021 | gate_vma.vm_end = FIXADDR_USER_END; |
Roland McGrath | b6558c4 | 2007-01-26 00:56:47 -0800 | [diff] [blame] | 3022 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; |
| 3023 | gate_vma.vm_page_prot = __P101; |
Roland McGrath | f47aef5 | 2007-01-26 00:56:49 -0800 | [diff] [blame] | 3024 | /* |
| 3025 | * Make sure the vDSO gets into every core dump. |
| 3026 | * Dumping its contents makes post-mortem fully interpretable later |
| 3027 | * without matching up the same kernel and hardware config to see |
| 3028 | * what PC values meant. |
| 3029 | */ |
| 3030 | gate_vma.vm_flags |= VM_ALWAYSDUMP; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3031 | return 0; |
| 3032 | } |
| 3033 | __initcall(gate_vma_init); |
| 3034 | #endif |
| 3035 | |
| 3036 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) |
| 3037 | { |
| 3038 | #ifdef AT_SYSINFO_EHDR |
| 3039 | return &gate_vma; |
| 3040 | #else |
| 3041 | return NULL; |
| 3042 | #endif |
| 3043 | } |
| 3044 | |
| 3045 | int in_gate_area_no_task(unsigned long addr) |
| 3046 | { |
| 3047 | #ifdef AT_SYSINFO_EHDR |
| 3048 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) |
| 3049 | return 1; |
| 3050 | #endif |
| 3051 | return 0; |
| 3052 | } |
| 3053 | |
| 3054 | #endif /* __HAVE_ARCH_GATE_AREA */ |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3055 | |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3056 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3057 | int follow_phys(struct vm_area_struct *vma, |
| 3058 | unsigned long address, unsigned int flags, |
| 3059 | unsigned long *prot, resource_size_t *phys) |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3060 | { |
| 3061 | pgd_t *pgd; |
| 3062 | pud_t *pud; |
| 3063 | pmd_t *pmd; |
| 3064 | pte_t *ptep, pte; |
| 3065 | spinlock_t *ptl; |
| 3066 | resource_size_t phys_addr = 0; |
| 3067 | struct mm_struct *mm = vma->vm_mm; |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3068 | int ret = -EINVAL; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3069 | |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3070 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
| 3071 | goto out; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3072 | |
| 3073 | pgd = pgd_offset(mm, address); |
| 3074 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3075 | goto out; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3076 | |
| 3077 | pud = pud_offset(pgd, address); |
| 3078 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3079 | goto out; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3080 | |
| 3081 | pmd = pmd_offset(pud, address); |
| 3082 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3083 | goto out; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3084 | |
| 3085 | /* We cannot handle huge page PFN maps. Luckily they don't exist. */ |
| 3086 | if (pmd_huge(*pmd)) |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3087 | goto out; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3088 | |
| 3089 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
| 3090 | if (!ptep) |
| 3091 | goto out; |
| 3092 | |
| 3093 | pte = *ptep; |
| 3094 | if (!pte_present(pte)) |
| 3095 | goto unlock; |
| 3096 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
| 3097 | goto unlock; |
| 3098 | phys_addr = pte_pfn(pte); |
| 3099 | phys_addr <<= PAGE_SHIFT; /* Shift here to avoid overflow on PAE */ |
| 3100 | |
| 3101 | *prot = pgprot_val(pte_pgprot(pte)); |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3102 | *phys = phys_addr; |
| 3103 | ret = 0; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3104 | |
| 3105 | unlock: |
| 3106 | pte_unmap_unlock(ptep, ptl); |
| 3107 | out: |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3108 | return ret; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3109 | } |
| 3110 | |
| 3111 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, |
| 3112 | void *buf, int len, int write) |
| 3113 | { |
| 3114 | resource_size_t phys_addr; |
| 3115 | unsigned long prot = 0; |
KOSAKI Motohiro | 2bc7273 | 2009-01-06 14:39:43 -0800 | [diff] [blame] | 3116 | void __iomem *maddr; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3117 | int offset = addr & (PAGE_SIZE-1); |
| 3118 | |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3119 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3120 | return -EINVAL; |
| 3121 | |
| 3122 | maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot); |
| 3123 | if (write) |
| 3124 | memcpy_toio(maddr + offset, buf, len); |
| 3125 | else |
| 3126 | memcpy_fromio(buf, maddr + offset, len); |
| 3127 | iounmap(maddr); |
| 3128 | |
| 3129 | return len; |
| 3130 | } |
| 3131 | #endif |
| 3132 | |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3133 | /* |
| 3134 | * Access another process' address space. |
| 3135 | * Source/target buffer must be kernel space, |
| 3136 | * Do not walk the page table directly, use get_user_pages |
| 3137 | */ |
| 3138 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) |
| 3139 | { |
| 3140 | struct mm_struct *mm; |
| 3141 | struct vm_area_struct *vma; |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3142 | void *old_buf = buf; |
| 3143 | |
| 3144 | mm = get_task_mm(tsk); |
| 3145 | if (!mm) |
| 3146 | return 0; |
| 3147 | |
| 3148 | down_read(&mm->mmap_sem); |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 3149 | /* ignore errors, just check how much was successfully transferred */ |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3150 | while (len) { |
| 3151 | int bytes, ret, offset; |
| 3152 | void *maddr; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3153 | struct page *page = NULL; |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3154 | |
| 3155 | ret = get_user_pages(tsk, mm, addr, 1, |
| 3156 | write, 1, &page, &vma); |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3157 | if (ret <= 0) { |
| 3158 | /* |
| 3159 | * Check if this is a VM_IO | VM_PFNMAP VMA, which |
| 3160 | * we can access using slightly different code. |
| 3161 | */ |
| 3162 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
| 3163 | vma = find_vma(mm, addr); |
| 3164 | if (!vma) |
| 3165 | break; |
| 3166 | if (vma->vm_ops && vma->vm_ops->access) |
| 3167 | ret = vma->vm_ops->access(vma, addr, buf, |
| 3168 | len, write); |
| 3169 | if (ret <= 0) |
| 3170 | #endif |
| 3171 | break; |
| 3172 | bytes = ret; |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3173 | } else { |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3174 | bytes = len; |
| 3175 | offset = addr & (PAGE_SIZE-1); |
| 3176 | if (bytes > PAGE_SIZE-offset) |
| 3177 | bytes = PAGE_SIZE-offset; |
| 3178 | |
| 3179 | maddr = kmap(page); |
| 3180 | if (write) { |
| 3181 | copy_to_user_page(vma, page, addr, |
| 3182 | maddr + offset, buf, bytes); |
| 3183 | set_page_dirty_lock(page); |
| 3184 | } else { |
| 3185 | copy_from_user_page(vma, page, addr, |
| 3186 | buf, maddr + offset, bytes); |
| 3187 | } |
| 3188 | kunmap(page); |
| 3189 | page_cache_release(page); |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3190 | } |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3191 | len -= bytes; |
| 3192 | buf += bytes; |
| 3193 | addr += bytes; |
| 3194 | } |
| 3195 | up_read(&mm->mmap_sem); |
| 3196 | mmput(mm); |
| 3197 | |
| 3198 | return buf - old_buf; |
| 3199 | } |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 3200 | |
| 3201 | /* |
| 3202 | * Print the name of a VMA. |
| 3203 | */ |
| 3204 | void print_vma_addr(char *prefix, unsigned long ip) |
| 3205 | { |
| 3206 | struct mm_struct *mm = current->mm; |
| 3207 | struct vm_area_struct *vma; |
| 3208 | |
Ingo Molnar | e8bff74 | 2008-02-13 20:21:06 +0100 | [diff] [blame] | 3209 | /* |
| 3210 | * Do not print if we are in atomic |
| 3211 | * contexts (in exception stacks, etc.): |
| 3212 | */ |
| 3213 | if (preempt_count()) |
| 3214 | return; |
| 3215 | |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 3216 | down_read(&mm->mmap_sem); |
| 3217 | vma = find_vma(mm, ip); |
| 3218 | if (vma && vma->vm_file) { |
| 3219 | struct file *f = vma->vm_file; |
| 3220 | char *buf = (char *)__get_free_page(GFP_KERNEL); |
| 3221 | if (buf) { |
| 3222 | char *p, *s; |
| 3223 | |
Jan Blunck | cf28b48 | 2008-02-14 19:38:44 -0800 | [diff] [blame] | 3224 | p = d_path(&f->f_path, buf, PAGE_SIZE); |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 3225 | if (IS_ERR(p)) |
| 3226 | p = "?"; |
| 3227 | s = strrchr(p, '/'); |
| 3228 | if (s) |
| 3229 | p = s+1; |
| 3230 | printk("%s%s[%lx+%lx]", prefix, p, |
| 3231 | vma->vm_start, |
| 3232 | vma->vm_end - vma->vm_start); |
| 3233 | free_page((unsigned long)buf); |
| 3234 | } |
| 3235 | } |
| 3236 | up_read(¤t->mm->mmap_sem); |
| 3237 | } |
Nick Piggin | 3ee1afa | 2008-09-10 13:37:17 +0200 | [diff] [blame] | 3238 | |
| 3239 | #ifdef CONFIG_PROVE_LOCKING |
| 3240 | void might_fault(void) |
| 3241 | { |
Peter Zijlstra | 95156f0 | 2009-01-12 13:02:11 +0100 | [diff] [blame] | 3242 | /* |
| 3243 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while |
| 3244 | * holding the mmap_sem, this is safe because kernel memory doesn't |
| 3245 | * get paged out, therefore we'll never actually fault, and the |
| 3246 | * below annotations will generate false positives. |
| 3247 | */ |
| 3248 | if (segment_eq(get_fs(), KERNEL_DS)) |
| 3249 | return; |
| 3250 | |
Nick Piggin | 3ee1afa | 2008-09-10 13:37:17 +0200 | [diff] [blame] | 3251 | might_sleep(); |
| 3252 | /* |
| 3253 | * it would be nicer only to annotate paths which are not under |
| 3254 | * pagefault_disable, however that requires a larger audit and |
| 3255 | * providing helpers like get_user_atomic. |
| 3256 | */ |
| 3257 | if (!in_atomic() && current->mm) |
| 3258 | might_lock_read(¤t->mm->mmap_sem); |
| 3259 | } |
| 3260 | EXPORT_SYMBOL(might_fault); |
| 3261 | #endif |