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> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | |
| 55 | #include <asm/pgalloc.h> |
| 56 | #include <asm/uaccess.h> |
| 57 | #include <asm/tlb.h> |
| 58 | #include <asm/tlbflush.h> |
| 59 | #include <asm/pgtable.h> |
| 60 | |
| 61 | #include <linux/swapops.h> |
| 62 | #include <linux/elf.h> |
| 63 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 64 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
| 66 | unsigned long max_mapnr; |
| 67 | struct page *mem_map; |
| 68 | |
| 69 | EXPORT_SYMBOL(max_mapnr); |
| 70 | EXPORT_SYMBOL(mem_map); |
| 71 | #endif |
| 72 | |
| 73 | unsigned long num_physpages; |
| 74 | /* |
| 75 | * A number of key systems in x86 including ioremap() rely on the assumption |
| 76 | * that high_memory defines the upper bound on direct map memory, then end |
| 77 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and |
| 78 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL |
| 79 | * and ZONE_HIGHMEM. |
| 80 | */ |
| 81 | void * high_memory; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | |
| 83 | EXPORT_SYMBOL(num_physpages); |
| 84 | EXPORT_SYMBOL(high_memory); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 85 | |
Ingo Molnar | 32a9323 | 2008-02-06 22:39:44 +0100 | [diff] [blame] | 86 | /* |
| 87 | * Randomize the address space (stacks, mmaps, brk, etc.). |
| 88 | * |
| 89 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, |
| 90 | * as ancient (libc5 based) binaries can segfault. ) |
| 91 | */ |
| 92 | int randomize_va_space __read_mostly = |
| 93 | #ifdef CONFIG_COMPAT_BRK |
| 94 | 1; |
| 95 | #else |
| 96 | 2; |
| 97 | #endif |
Andi Kleen | a62eaf1 | 2006-02-16 23:41:58 +0100 | [diff] [blame] | 98 | |
| 99 | static int __init disable_randmaps(char *s) |
| 100 | { |
| 101 | randomize_va_space = 0; |
OGAWA Hirofumi | 9b41046 | 2006-03-31 02:30:33 -0800 | [diff] [blame] | 102 | return 1; |
Andi Kleen | a62eaf1 | 2006-02-16 23:41:58 +0100 | [diff] [blame] | 103 | } |
| 104 | __setup("norandmaps", disable_randmaps); |
| 105 | |
| 106 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | /* |
| 108 | * If a p?d_bad entry is found while walking page tables, report |
| 109 | * the error, before resetting entry to p?d_none. Usually (but |
| 110 | * very seldom) called out from the p?d_none_or_clear_bad macros. |
| 111 | */ |
| 112 | |
| 113 | void pgd_clear_bad(pgd_t *pgd) |
| 114 | { |
| 115 | pgd_ERROR(*pgd); |
| 116 | pgd_clear(pgd); |
| 117 | } |
| 118 | |
| 119 | void pud_clear_bad(pud_t *pud) |
| 120 | { |
| 121 | pud_ERROR(*pud); |
| 122 | pud_clear(pud); |
| 123 | } |
| 124 | |
| 125 | void pmd_clear_bad(pmd_t *pmd) |
| 126 | { |
| 127 | pmd_ERROR(*pmd); |
| 128 | pmd_clear(pmd); |
| 129 | } |
| 130 | |
| 131 | /* |
| 132 | * Note: this doesn't free the actual pages themselves. That |
| 133 | * has been handled earlier when unmapping all the memory regions. |
| 134 | */ |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 135 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 136 | { |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 137 | pgtable_t token = pmd_pgtable(*pmd); |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 138 | pmd_clear(pmd); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 139 | pte_free_tlb(tlb, token); |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 140 | tlb->mm->nr_ptes--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | } |
| 142 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 143 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
| 144 | unsigned long addr, unsigned long end, |
| 145 | unsigned long floor, unsigned long ceiling) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 146 | { |
| 147 | pmd_t *pmd; |
| 148 | unsigned long next; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 149 | unsigned long start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 150 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 151 | start = addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | pmd = pmd_offset(pud, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | do { |
| 154 | next = pmd_addr_end(addr, end); |
| 155 | if (pmd_none_or_clear_bad(pmd)) |
| 156 | continue; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 157 | free_pte_range(tlb, pmd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 158 | } while (pmd++, addr = next, addr != end); |
| 159 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 160 | start &= PUD_MASK; |
| 161 | if (start < floor) |
| 162 | return; |
| 163 | if (ceiling) { |
| 164 | ceiling &= PUD_MASK; |
| 165 | if (!ceiling) |
| 166 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | } |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 168 | if (end - 1 > ceiling - 1) |
| 169 | return; |
| 170 | |
| 171 | pmd = pmd_offset(pud, start); |
| 172 | pud_clear(pud); |
| 173 | pmd_free_tlb(tlb, pmd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 174 | } |
| 175 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 176 | static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, |
| 177 | unsigned long addr, unsigned long end, |
| 178 | unsigned long floor, unsigned long ceiling) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 179 | { |
| 180 | pud_t *pud; |
| 181 | unsigned long next; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 182 | unsigned long start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 183 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 184 | start = addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | pud = pud_offset(pgd, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | do { |
| 187 | next = pud_addr_end(addr, end); |
| 188 | if (pud_none_or_clear_bad(pud)) |
| 189 | continue; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 190 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 191 | } while (pud++, addr = next, addr != end); |
| 192 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 193 | start &= PGDIR_MASK; |
| 194 | if (start < floor) |
| 195 | return; |
| 196 | if (ceiling) { |
| 197 | ceiling &= PGDIR_MASK; |
| 198 | if (!ceiling) |
| 199 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | } |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 201 | if (end - 1 > ceiling - 1) |
| 202 | return; |
| 203 | |
| 204 | pud = pud_offset(pgd, start); |
| 205 | pgd_clear(pgd); |
| 206 | pud_free_tlb(tlb, pud); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 207 | } |
| 208 | |
| 209 | /* |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 210 | * This function frees user-level page tables of a process. |
| 211 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 212 | * Must be called with pagetable lock held. |
| 213 | */ |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 214 | void free_pgd_range(struct mmu_gather **tlb, |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 215 | unsigned long addr, unsigned long end, |
| 216 | unsigned long floor, unsigned long ceiling) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 217 | { |
| 218 | pgd_t *pgd; |
| 219 | unsigned long next; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 220 | unsigned long start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 221 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 222 | /* |
| 223 | * The next few lines have given us lots of grief... |
| 224 | * |
| 225 | * Why are we testing PMD* at this top level? Because often |
| 226 | * there will be no work to do at all, and we'd prefer not to |
| 227 | * go all the way down to the bottom just to discover that. |
| 228 | * |
| 229 | * Why all these "- 1"s? Because 0 represents both the bottom |
| 230 | * of the address space and the top of it (using -1 for the |
| 231 | * top wouldn't help much: the masks would do the wrong thing). |
| 232 | * The rule is that addr 0 and floor 0 refer to the bottom of |
| 233 | * the address space, but end 0 and ceiling 0 refer to the top |
| 234 | * Comparisons need to use "end - 1" and "ceiling - 1" (though |
| 235 | * that end 0 case should be mythical). |
| 236 | * |
| 237 | * Wherever addr is brought up or ceiling brought down, we must |
| 238 | * be careful to reject "the opposite 0" before it confuses the |
| 239 | * subsequent tests. But what about where end is brought down |
| 240 | * by PMD_SIZE below? no, end can't go down to 0 there. |
| 241 | * |
| 242 | * Whereas we round start (addr) and ceiling down, by different |
| 243 | * masks at different levels, in order to test whether a table |
| 244 | * now has no other vmas using it, so can be freed, we don't |
| 245 | * bother to round floor or end up - the tests don't need that. |
| 246 | */ |
| 247 | |
| 248 | addr &= PMD_MASK; |
| 249 | if (addr < floor) { |
| 250 | addr += PMD_SIZE; |
| 251 | if (!addr) |
| 252 | return; |
| 253 | } |
| 254 | if (ceiling) { |
| 255 | ceiling &= PMD_MASK; |
| 256 | if (!ceiling) |
| 257 | return; |
| 258 | } |
| 259 | if (end - 1 > ceiling - 1) |
| 260 | end -= PMD_SIZE; |
| 261 | if (addr > end - 1) |
| 262 | return; |
| 263 | |
| 264 | start = addr; |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 265 | pgd = pgd_offset((*tlb)->mm, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | do { |
| 267 | next = pgd_addr_end(addr, end); |
| 268 | if (pgd_none_or_clear_bad(pgd)) |
| 269 | continue; |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 270 | free_pud_range(*tlb, pgd, addr, next, floor, ceiling); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 271 | } while (pgd++, addr = next, addr != end); |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 272 | } |
| 273 | |
| 274 | void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma, |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 275 | unsigned long floor, unsigned long ceiling) |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 276 | { |
| 277 | while (vma) { |
| 278 | struct vm_area_struct *next = vma->vm_next; |
| 279 | unsigned long addr = vma->vm_start; |
| 280 | |
Hugh Dickins | 8f4f8c1 | 2005-10-29 18:16:29 -0700 | [diff] [blame] | 281 | /* |
| 282 | * Hide vma from rmap and vmtruncate before freeing pgtables |
| 283 | */ |
| 284 | anon_vma_unlink(vma); |
| 285 | unlink_file_vma(vma); |
| 286 | |
David Gibson | 9da61ae | 2006-03-22 00:08:57 -0800 | [diff] [blame] | 287 | if (is_vm_hugetlb_page(vma)) { |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 288 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 289 | floor, next? next->vm_start: ceiling); |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 290 | } else { |
| 291 | /* |
| 292 | * Optimization: gather nearby vmas into one call down |
| 293 | */ |
| 294 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE |
David Gibson | 4866920 | 2006-03-22 00:08:58 -0800 | [diff] [blame] | 295 | && !is_vm_hugetlb_page(next)) { |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 296 | vma = next; |
| 297 | next = vma->vm_next; |
Hugh Dickins | 8f4f8c1 | 2005-10-29 18:16:29 -0700 | [diff] [blame] | 298 | anon_vma_unlink(vma); |
| 299 | unlink_file_vma(vma); |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 300 | } |
| 301 | free_pgd_range(tlb, addr, vma->vm_end, |
| 302 | floor, next? next->vm_start: ceiling); |
| 303 | } |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 304 | vma = next; |
| 305 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 306 | } |
| 307 | |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 308 | 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] | 309 | { |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 310 | pgtable_t new = pte_alloc_one(mm, address); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 311 | if (!new) |
| 312 | return -ENOMEM; |
| 313 | |
Nick Piggin | 362a61a | 2008-05-14 06:37:36 +0200 | [diff] [blame] | 314 | /* |
| 315 | * Ensure all pte setup (eg. pte page lock and page clearing) are |
| 316 | * visible before the pte is made visible to other CPUs by being |
| 317 | * put into page tables. |
| 318 | * |
| 319 | * The other side of the story is the pointer chasing in the page |
| 320 | * table walking code (when walking the page table without locking; |
| 321 | * ie. most of the time). Fortunately, these data accesses consist |
| 322 | * of a chain of data-dependent loads, meaning most CPUs (alpha |
| 323 | * being the notable exception) will already guarantee loads are |
| 324 | * seen in-order. See the alpha page table accessors for the |
| 325 | * smp_read_barrier_depends() barriers in page table walking code. |
| 326 | */ |
| 327 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ |
| 328 | |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 329 | spin_lock(&mm->page_table_lock); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 330 | if (!pmd_present(*pmd)) { /* Has another populated it ? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 331 | mm->nr_ptes++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | pmd_populate(mm, pmd, new); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 333 | new = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | } |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 335 | spin_unlock(&mm->page_table_lock); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 336 | if (new) |
| 337 | pte_free(mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 338 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | } |
| 340 | |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 341 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 342 | { |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 343 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
| 344 | if (!new) |
| 345 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 346 | |
Nick Piggin | 362a61a | 2008-05-14 06:37:36 +0200 | [diff] [blame] | 347 | smp_wmb(); /* See comment in __pte_alloc */ |
| 348 | |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 349 | spin_lock(&init_mm.page_table_lock); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 350 | if (!pmd_present(*pmd)) { /* Has another populated it ? */ |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 351 | pmd_populate_kernel(&init_mm, pmd, new); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 352 | new = NULL; |
| 353 | } |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 354 | spin_unlock(&init_mm.page_table_lock); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 355 | if (new) |
| 356 | pte_free_kernel(&init_mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 357 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 358 | } |
| 359 | |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 360 | static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss) |
| 361 | { |
| 362 | if (file_rss) |
| 363 | add_mm_counter(mm, file_rss, file_rss); |
| 364 | if (anon_rss) |
| 365 | add_mm_counter(mm, anon_rss, anon_rss); |
| 366 | } |
| 367 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | /* |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 369 | * This function is called to print an error when a bad pte |
| 370 | * is found. For example, we might have a PFN-mapped pte in |
| 371 | * a region that doesn't allow it. |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 372 | * |
| 373 | * The calling function must still handle the error. |
| 374 | */ |
| 375 | void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr) |
| 376 | { |
| 377 | printk(KERN_ERR "Bad pte = %08llx, process = %s, " |
| 378 | "vm_flags = %lx, vaddr = %lx\n", |
| 379 | (long long)pte_val(pte), |
| 380 | (vma->vm_mm == current->mm ? current->comm : "???"), |
| 381 | vma->vm_flags, vaddr); |
| 382 | dump_stack(); |
| 383 | } |
| 384 | |
Linus Torvalds | 6712117 | 2005-12-11 20:38:17 -0800 | [diff] [blame] | 385 | static inline int is_cow_mapping(unsigned int flags) |
| 386 | { |
| 387 | return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; |
| 388 | } |
| 389 | |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 390 | /* |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 391 | * 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] | 392 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 393 | * "Special" mappings do not wish to be associated with a "struct page" (either |
| 394 | * it doesn't exist, or it exists but they don't want to touch it). In this |
| 395 | * case, NULL is returned here. "Normal" mappings do have a struct page. |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 396 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 397 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
| 398 | * pte bit, in which case this function is trivial. Secondly, an architecture |
| 399 | * may not have a spare pte bit, which requires a more complicated scheme, |
| 400 | * described below. |
| 401 | * |
| 402 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a |
| 403 | * special mapping (even if there are underlying and valid "struct pages"). |
| 404 | * COWed pages of a VM_PFNMAP are always normal. |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 405 | * |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 406 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
| 407 | * 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] | 408 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
| 409 | * mapping will always honor the rule |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 410 | * |
| 411 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) |
| 412 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 413 | * And for normal mappings this is false. |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 414 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 415 | * This restricts such mappings to be a linear translation from virtual address |
| 416 | * to pfn. To get around this restriction, we allow arbitrary mappings so long |
| 417 | * as the vma is not a COW mapping; in that case, we know that all ptes are |
| 418 | * special (because none can have been COWed). |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 419 | * |
| 420 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 421 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
| 422 | * |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 423 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct |
| 424 | * page" backing, however the difference is that _all_ pages with a struct |
| 425 | * page (that is, those where pfn_valid is true) are refcounted and considered |
| 426 | * normal pages by the VM. The disadvantage is that pages are refcounted |
| 427 | * (which can be slower and simply not an option for some PFNMAP users). The |
| 428 | * advantage is that we don't have to follow the strict linearity rule of |
| 429 | * PFNMAP mappings in order to support COWable mappings. |
| 430 | * |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 431 | */ |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 432 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
| 433 | # define HAVE_PTE_SPECIAL 1 |
| 434 | #else |
| 435 | # define HAVE_PTE_SPECIAL 0 |
| 436 | #endif |
| 437 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
| 438 | pte_t pte) |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 439 | { |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 440 | unsigned long pfn; |
| 441 | |
| 442 | if (HAVE_PTE_SPECIAL) { |
| 443 | if (likely(!pte_special(pte))) { |
| 444 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); |
| 445 | return pte_page(pte); |
| 446 | } |
| 447 | VM_BUG_ON(!(vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))); |
| 448 | return NULL; |
| 449 | } |
| 450 | |
| 451 | /* !HAVE_PTE_SPECIAL case follows: */ |
| 452 | |
| 453 | pfn = pte_pfn(pte); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 454 | |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 455 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
| 456 | if (vma->vm_flags & VM_MIXEDMAP) { |
| 457 | if (!pfn_valid(pfn)) |
| 458 | return NULL; |
| 459 | goto out; |
| 460 | } else { |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 461 | unsigned long off; |
| 462 | off = (addr - vma->vm_start) >> PAGE_SHIFT; |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 463 | if (pfn == vma->vm_pgoff + off) |
| 464 | return NULL; |
| 465 | if (!is_cow_mapping(vma->vm_flags)) |
| 466 | return NULL; |
| 467 | } |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 468 | } |
| 469 | |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 470 | VM_BUG_ON(!pfn_valid(pfn)); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 471 | |
| 472 | /* |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 473 | * NOTE! We still have PageReserved() pages in the page tables. |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 474 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 475 | * eg. VDSO mappings can cause them to exist. |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 476 | */ |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 477 | out: |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 478 | return pfn_to_page(pfn); |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 479 | } |
| 480 | |
| 481 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 482 | * copy one vm_area from one task to the other. Assumes the page tables |
| 483 | * already present in the new task to be cleared in the whole range |
| 484 | * covered by this vma. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 485 | */ |
| 486 | |
Hugh Dickins | 8c10376 | 2005-10-29 18:16:13 -0700 | [diff] [blame] | 487 | static inline void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 488 | 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] | 489 | 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] | 490 | unsigned long addr, int *rss) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 491 | { |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 492 | unsigned long vm_flags = vma->vm_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 493 | pte_t pte = *src_pte; |
| 494 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 495 | |
| 496 | /* pte contains position in swap or file, so copy. */ |
| 497 | if (unlikely(!pte_present(pte))) { |
| 498 | if (!pte_file(pte)) { |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 499 | swp_entry_t entry = pte_to_swp_entry(pte); |
| 500 | |
| 501 | swap_duplicate(entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 502 | /* make sure dst_mm is on swapoff's mmlist. */ |
| 503 | if (unlikely(list_empty(&dst_mm->mmlist))) { |
| 504 | spin_lock(&mmlist_lock); |
Hugh Dickins | f412ac0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 505 | if (list_empty(&dst_mm->mmlist)) |
| 506 | list_add(&dst_mm->mmlist, |
| 507 | &src_mm->mmlist); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 508 | spin_unlock(&mmlist_lock); |
| 509 | } |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 510 | if (is_write_migration_entry(entry) && |
| 511 | is_cow_mapping(vm_flags)) { |
| 512 | /* |
| 513 | * COW mappings require pages in both parent |
| 514 | * and child to be set to read. |
| 515 | */ |
| 516 | make_migration_entry_read(&entry); |
| 517 | pte = swp_entry_to_pte(entry); |
| 518 | set_pte_at(src_mm, addr, src_pte, pte); |
| 519 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 520 | } |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 521 | goto out_set_pte; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 522 | } |
| 523 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 524 | /* |
| 525 | * If it's a COW mapping, write protect it both |
| 526 | * in the parent and the child |
| 527 | */ |
Linus Torvalds | 6712117 | 2005-12-11 20:38:17 -0800 | [diff] [blame] | 528 | if (is_cow_mapping(vm_flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 529 | ptep_set_wrprotect(src_mm, addr, src_pte); |
Zachary Amsden | 3dc9079 | 2006-09-30 23:29:30 -0700 | [diff] [blame] | 530 | pte = pte_wrprotect(pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 531 | } |
| 532 | |
| 533 | /* |
| 534 | * If it's a shared mapping, mark it clean in |
| 535 | * the child |
| 536 | */ |
| 537 | if (vm_flags & VM_SHARED) |
| 538 | pte = pte_mkclean(pte); |
| 539 | pte = pte_mkold(pte); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 540 | |
| 541 | page = vm_normal_page(vma, addr, pte); |
| 542 | if (page) { |
| 543 | get_page(page); |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 544 | page_dup_rmap(page, vma, addr); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 545 | rss[!!PageAnon(page)]++; |
| 546 | } |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 547 | |
| 548 | out_set_pte: |
| 549 | set_pte_at(dst_mm, addr, dst_pte, pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | } |
| 551 | |
| 552 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
| 553 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
| 554 | unsigned long addr, unsigned long end) |
| 555 | { |
| 556 | pte_t *src_pte, *dst_pte; |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 557 | spinlock_t *src_ptl, *dst_ptl; |
Hugh Dickins | e040f21 | 2005-10-29 18:15:53 -0700 | [diff] [blame] | 558 | int progress = 0; |
Hugh Dickins | 8c10376 | 2005-10-29 18:16:13 -0700 | [diff] [blame] | 559 | int rss[2]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 560 | |
| 561 | again: |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 562 | rss[1] = rss[0] = 0; |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 563 | 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] | 564 | if (!dst_pte) |
| 565 | return -ENOMEM; |
| 566 | src_pte = pte_offset_map_nested(src_pmd, addr); |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 567 | src_ptl = pte_lockptr(src_mm, src_pmd); |
Ingo Molnar | f20dc5f | 2006-07-03 00:25:08 -0700 | [diff] [blame] | 568 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 569 | arch_enter_lazy_mmu_mode(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 571 | do { |
| 572 | /* |
| 573 | * We are holding two locks at this point - either of them |
| 574 | * could generate latencies in another task on another CPU. |
| 575 | */ |
Hugh Dickins | e040f21 | 2005-10-29 18:15:53 -0700 | [diff] [blame] | 576 | if (progress >= 32) { |
| 577 | progress = 0; |
| 578 | if (need_resched() || |
Nick Piggin | 95c354f | 2008-01-30 13:31:20 +0100 | [diff] [blame] | 579 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
Hugh Dickins | e040f21 | 2005-10-29 18:15:53 -0700 | [diff] [blame] | 580 | break; |
| 581 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 582 | if (pte_none(*src_pte)) { |
| 583 | progress++; |
| 584 | continue; |
| 585 | } |
Hugh Dickins | 8c10376 | 2005-10-29 18:16:13 -0700 | [diff] [blame] | 586 | 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] | 587 | progress += 8; |
| 588 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 589 | |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 590 | arch_leave_lazy_mmu_mode(); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 591 | spin_unlock(src_ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 592 | pte_unmap_nested(src_pte - 1); |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 593 | add_mm_rss(dst_mm, rss[0], rss[1]); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 594 | pte_unmap_unlock(dst_pte - 1, dst_ptl); |
| 595 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 596 | if (addr != end) |
| 597 | goto again; |
| 598 | return 0; |
| 599 | } |
| 600 | |
| 601 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
| 602 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, |
| 603 | unsigned long addr, unsigned long end) |
| 604 | { |
| 605 | pmd_t *src_pmd, *dst_pmd; |
| 606 | unsigned long next; |
| 607 | |
| 608 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); |
| 609 | if (!dst_pmd) |
| 610 | return -ENOMEM; |
| 611 | src_pmd = pmd_offset(src_pud, addr); |
| 612 | do { |
| 613 | next = pmd_addr_end(addr, end); |
| 614 | if (pmd_none_or_clear_bad(src_pmd)) |
| 615 | continue; |
| 616 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, |
| 617 | vma, addr, next)) |
| 618 | return -ENOMEM; |
| 619 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); |
| 620 | return 0; |
| 621 | } |
| 622 | |
| 623 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
| 624 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, |
| 625 | unsigned long addr, unsigned long end) |
| 626 | { |
| 627 | pud_t *src_pud, *dst_pud; |
| 628 | unsigned long next; |
| 629 | |
| 630 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); |
| 631 | if (!dst_pud) |
| 632 | return -ENOMEM; |
| 633 | src_pud = pud_offset(src_pgd, addr); |
| 634 | do { |
| 635 | next = pud_addr_end(addr, end); |
| 636 | if (pud_none_or_clear_bad(src_pud)) |
| 637 | continue; |
| 638 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, |
| 639 | vma, addr, next)) |
| 640 | return -ENOMEM; |
| 641 | } while (dst_pud++, src_pud++, addr = next, addr != end); |
| 642 | return 0; |
| 643 | } |
| 644 | |
| 645 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
| 646 | struct vm_area_struct *vma) |
| 647 | { |
| 648 | pgd_t *src_pgd, *dst_pgd; |
| 649 | unsigned long next; |
| 650 | unsigned long addr = vma->vm_start; |
| 651 | unsigned long end = vma->vm_end; |
| 652 | |
Nick Piggin | d992895 | 2005-08-28 16:49:11 +1000 | [diff] [blame] | 653 | /* |
| 654 | * Don't copy ptes where a page fault will fill them correctly. |
| 655 | * Fork becomes much lighter when there are big shared or private |
| 656 | * readonly mappings. The tradeoff is that copy_page_range is more |
| 657 | * efficient than faulting. |
| 658 | */ |
Linus Torvalds | 4d7672b | 2005-12-16 10:21:23 -0800 | [diff] [blame] | 659 | if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) { |
Nick Piggin | d992895 | 2005-08-28 16:49:11 +1000 | [diff] [blame] | 660 | if (!vma->anon_vma) |
| 661 | return 0; |
| 662 | } |
| 663 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | if (is_vm_hugetlb_page(vma)) |
| 665 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); |
| 666 | |
| 667 | dst_pgd = pgd_offset(dst_mm, addr); |
| 668 | src_pgd = pgd_offset(src_mm, addr); |
| 669 | do { |
| 670 | next = pgd_addr_end(addr, end); |
| 671 | if (pgd_none_or_clear_bad(src_pgd)) |
| 672 | continue; |
| 673 | if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, |
| 674 | vma, addr, next)) |
| 675 | return -ENOMEM; |
| 676 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
| 677 | return 0; |
| 678 | } |
| 679 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 680 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 681 | struct vm_area_struct *vma, pmd_t *pmd, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 682 | unsigned long addr, unsigned long end, |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 683 | long *zap_work, struct zap_details *details) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 684 | { |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 685 | struct mm_struct *mm = tlb->mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 686 | pte_t *pte; |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 687 | spinlock_t *ptl; |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 688 | int file_rss = 0; |
| 689 | int anon_rss = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 691 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 692 | arch_enter_lazy_mmu_mode(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | do { |
| 694 | pte_t ptent = *pte; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 695 | if (pte_none(ptent)) { |
| 696 | (*zap_work)--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 697 | continue; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 698 | } |
Hugh Dickins | 6f5e6b9 | 2006-03-16 23:04:09 -0800 | [diff] [blame] | 699 | |
| 700 | (*zap_work) -= PAGE_SIZE; |
| 701 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 702 | if (pte_present(ptent)) { |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 703 | struct page *page; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 704 | |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 705 | page = vm_normal_page(vma, addr, ptent); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 706 | if (unlikely(details) && page) { |
| 707 | /* |
| 708 | * unmap_shared_mapping_pages() wants to |
| 709 | * invalidate cache without truncating: |
| 710 | * unmap shared but keep private pages. |
| 711 | */ |
| 712 | if (details->check_mapping && |
| 713 | details->check_mapping != page->mapping) |
| 714 | continue; |
| 715 | /* |
| 716 | * Each page->index must be checked when |
| 717 | * invalidating or truncating nonlinear. |
| 718 | */ |
| 719 | if (details->nonlinear_vma && |
| 720 | (page->index < details->first_index || |
| 721 | page->index > details->last_index)) |
| 722 | continue; |
| 723 | } |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 724 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
Zachary Amsden | a600388 | 2005-09-03 15:55:04 -0700 | [diff] [blame] | 725 | tlb->fullmm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 726 | tlb_remove_tlb_entry(tlb, pte, addr); |
| 727 | if (unlikely(!page)) |
| 728 | continue; |
| 729 | if (unlikely(details) && details->nonlinear_vma |
| 730 | && linear_page_index(details->nonlinear_vma, |
| 731 | addr) != page->index) |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 732 | set_pte_at(mm, addr, pte, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 733 | pgoff_to_pte(page->index)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 734 | if (PageAnon(page)) |
Hugh Dickins | 86d912f | 2005-10-29 18:16:14 -0700 | [diff] [blame] | 735 | anon_rss--; |
Hugh Dickins | 6237bcd | 2005-10-29 18:15:54 -0700 | [diff] [blame] | 736 | else { |
| 737 | if (pte_dirty(ptent)) |
| 738 | set_page_dirty(page); |
| 739 | if (pte_young(ptent)) |
Ken Chen | daa88c8 | 2007-02-10 01:43:18 -0800 | [diff] [blame] | 740 | SetPageReferenced(page); |
Hugh Dickins | 86d912f | 2005-10-29 18:16:14 -0700 | [diff] [blame] | 741 | file_rss--; |
Hugh Dickins | 6237bcd | 2005-10-29 18:15:54 -0700 | [diff] [blame] | 742 | } |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 743 | page_remove_rmap(page, vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 744 | tlb_remove_page(tlb, page); |
| 745 | continue; |
| 746 | } |
| 747 | /* |
| 748 | * If details->check_mapping, we leave swap entries; |
| 749 | * if details->nonlinear_vma, we leave file entries. |
| 750 | */ |
| 751 | if (unlikely(details)) |
| 752 | continue; |
| 753 | if (!pte_file(ptent)) |
| 754 | free_swap_and_cache(pte_to_swp_entry(ptent)); |
Zachary Amsden | 9888a1c | 2006-09-30 23:29:31 -0700 | [diff] [blame] | 755 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 756 | } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0)); |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 757 | |
Hugh Dickins | 86d912f | 2005-10-29 18:16:14 -0700 | [diff] [blame] | 758 | add_mm_rss(mm, file_rss, anon_rss); |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 759 | arch_leave_lazy_mmu_mode(); |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 760 | pte_unmap_unlock(pte - 1, ptl); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 761 | |
| 762 | return addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 763 | } |
| 764 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 765 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 766 | struct vm_area_struct *vma, pud_t *pud, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | unsigned long addr, unsigned long end, |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 768 | long *zap_work, struct zap_details *details) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 769 | { |
| 770 | pmd_t *pmd; |
| 771 | unsigned long next; |
| 772 | |
| 773 | pmd = pmd_offset(pud, addr); |
| 774 | do { |
| 775 | next = pmd_addr_end(addr, end); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 776 | if (pmd_none_or_clear_bad(pmd)) { |
| 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 | } |
| 780 | next = zap_pte_range(tlb, vma, pmd, addr, next, |
| 781 | zap_work, details); |
| 782 | } while (pmd++, addr = next, (addr != end && *zap_work > 0)); |
| 783 | |
| 784 | return addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 785 | } |
| 786 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 787 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 788 | struct vm_area_struct *vma, pgd_t *pgd, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 789 | unsigned long addr, unsigned long end, |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 790 | long *zap_work, struct zap_details *details) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 791 | { |
| 792 | pud_t *pud; |
| 793 | unsigned long next; |
| 794 | |
| 795 | pud = pud_offset(pgd, addr); |
| 796 | do { |
| 797 | next = pud_addr_end(addr, end); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 798 | if (pud_none_or_clear_bad(pud)) { |
| 799 | (*zap_work)--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 800 | continue; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 801 | } |
| 802 | next = zap_pmd_range(tlb, vma, pud, addr, next, |
| 803 | zap_work, details); |
| 804 | } while (pud++, addr = next, (addr != end && *zap_work > 0)); |
| 805 | |
| 806 | return addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 807 | } |
| 808 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 809 | static unsigned long unmap_page_range(struct mmu_gather *tlb, |
| 810 | struct vm_area_struct *vma, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 811 | unsigned long addr, unsigned long end, |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 812 | long *zap_work, struct zap_details *details) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 813 | { |
| 814 | pgd_t *pgd; |
| 815 | unsigned long next; |
| 816 | |
| 817 | if (details && !details->check_mapping && !details->nonlinear_vma) |
| 818 | details = NULL; |
| 819 | |
| 820 | BUG_ON(addr >= end); |
| 821 | tlb_start_vma(tlb, vma); |
| 822 | pgd = pgd_offset(vma->vm_mm, addr); |
| 823 | do { |
| 824 | next = pgd_addr_end(addr, end); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 825 | if (pgd_none_or_clear_bad(pgd)) { |
| 826 | (*zap_work)--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 827 | continue; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 828 | } |
| 829 | next = zap_pud_range(tlb, vma, pgd, addr, next, |
| 830 | zap_work, details); |
| 831 | } while (pgd++, addr = next, (addr != end && *zap_work > 0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 832 | tlb_end_vma(tlb, vma); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 833 | |
| 834 | return addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 835 | } |
| 836 | |
| 837 | #ifdef CONFIG_PREEMPT |
| 838 | # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE) |
| 839 | #else |
| 840 | /* No preempt: go for improved straight-line efficiency */ |
| 841 | # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE) |
| 842 | #endif |
| 843 | |
| 844 | /** |
| 845 | * unmap_vmas - unmap a range of memory covered by a list of vma's |
| 846 | * @tlbp: address of the caller's struct mmu_gather |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 847 | * @vma: the starting vma |
| 848 | * @start_addr: virtual address at which to start unmapping |
| 849 | * @end_addr: virtual address at which to end unmapping |
| 850 | * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here |
| 851 | * @details: details of nonlinear truncation or shared cache invalidation |
| 852 | * |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 853 | * Returns the end address of the unmapping (restart addr if interrupted). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 854 | * |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 855 | * Unmap all pages in the vma list. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 856 | * |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 857 | * We aim to not hold locks for too long (for scheduling latency reasons). |
| 858 | * 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] | 859 | * return the ending mmu_gather to the caller. |
| 860 | * |
| 861 | * Only addresses between `start' and `end' will be unmapped. |
| 862 | * |
| 863 | * The VMA list must be sorted in ascending virtual address order. |
| 864 | * |
| 865 | * unmap_vmas() assumes that the caller will flush the whole unmapped address |
| 866 | * range after unmap_vmas() returns. So the only responsibility here is to |
| 867 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() |
| 868 | * drops the lock and schedules. |
| 869 | */ |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 870 | unsigned long unmap_vmas(struct mmu_gather **tlbp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 871 | struct vm_area_struct *vma, unsigned long start_addr, |
| 872 | unsigned long end_addr, unsigned long *nr_accounted, |
| 873 | struct zap_details *details) |
| 874 | { |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 875 | long zap_work = ZAP_BLOCK_SIZE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 876 | unsigned long tlb_start = 0; /* For tlb_finish_mmu */ |
| 877 | int tlb_start_valid = 0; |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 878 | unsigned long start = start_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 879 | spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL; |
Hugh Dickins | 4d6ddfa | 2005-10-29 18:16:02 -0700 | [diff] [blame] | 880 | int fullmm = (*tlbp)->fullmm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 881 | |
| 882 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 883 | unsigned long end; |
| 884 | |
| 885 | start = max(vma->vm_start, start_addr); |
| 886 | if (start >= vma->vm_end) |
| 887 | continue; |
| 888 | end = min(vma->vm_end, end_addr); |
| 889 | if (end <= vma->vm_start) |
| 890 | continue; |
| 891 | |
| 892 | if (vma->vm_flags & VM_ACCOUNT) |
| 893 | *nr_accounted += (end - start) >> PAGE_SHIFT; |
| 894 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 895 | while (start != end) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 896 | if (!tlb_start_valid) { |
| 897 | tlb_start = start; |
| 898 | tlb_start_valid = 1; |
| 899 | } |
| 900 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 901 | if (unlikely(is_vm_hugetlb_page(vma))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 902 | unmap_hugepage_range(vma, start, end); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 903 | zap_work -= (end - start) / |
| 904 | (HPAGE_SIZE / PAGE_SIZE); |
| 905 | start = end; |
| 906 | } else |
| 907 | start = unmap_page_range(*tlbp, vma, |
| 908 | start, end, &zap_work, details); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 909 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 910 | if (zap_work > 0) { |
| 911 | BUG_ON(start != end); |
| 912 | break; |
| 913 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 914 | |
| 915 | tlb_finish_mmu(*tlbp, tlb_start, start); |
| 916 | |
| 917 | if (need_resched() || |
Nick Piggin | 95c354f | 2008-01-30 13:31:20 +0100 | [diff] [blame] | 918 | (i_mmap_lock && spin_needbreak(i_mmap_lock))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 919 | if (i_mmap_lock) { |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 920 | *tlbp = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 921 | goto out; |
| 922 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 923 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 924 | } |
| 925 | |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 926 | *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 927 | tlb_start_valid = 0; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 928 | zap_work = ZAP_BLOCK_SIZE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 929 | } |
| 930 | } |
| 931 | out: |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 932 | return start; /* which is now the end (or restart) address */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 933 | } |
| 934 | |
| 935 | /** |
| 936 | * zap_page_range - remove user pages in a given range |
| 937 | * @vma: vm_area_struct holding the applicable pages |
| 938 | * @address: starting address of pages to zap |
| 939 | * @size: number of bytes to zap |
| 940 | * @details: details of nonlinear truncation or shared cache invalidation |
| 941 | */ |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 942 | 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] | 943 | unsigned long size, struct zap_details *details) |
| 944 | { |
| 945 | struct mm_struct *mm = vma->vm_mm; |
| 946 | struct mmu_gather *tlb; |
| 947 | unsigned long end = address + size; |
| 948 | unsigned long nr_accounted = 0; |
| 949 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 950 | lru_add_drain(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 951 | tlb = tlb_gather_mmu(mm, 0); |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 952 | update_hiwater_rss(mm); |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 953 | end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details); |
| 954 | if (tlb) |
| 955 | tlb_finish_mmu(tlb, address, end); |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 956 | return end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 957 | } |
| 958 | |
| 959 | /* |
| 960 | * Do a quick page-table lookup for a single page. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 961 | */ |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 962 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 963 | unsigned int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 964 | { |
| 965 | pgd_t *pgd; |
| 966 | pud_t *pud; |
| 967 | pmd_t *pmd; |
| 968 | pte_t *ptep, pte; |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 969 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 970 | struct page *page; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 971 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 972 | |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 973 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); |
| 974 | if (!IS_ERR(page)) { |
| 975 | BUG_ON(flags & FOLL_GET); |
| 976 | goto out; |
| 977 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 978 | |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 979 | page = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 980 | pgd = pgd_offset(mm, address); |
| 981 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 982 | goto no_page_table; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 983 | |
| 984 | pud = pud_offset(pgd, address); |
| 985 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 986 | goto no_page_table; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 987 | |
| 988 | pmd = pmd_offset(pud, address); |
Hugh Dickins | aeed5fc | 2008-05-06 20:49:23 +0100 | [diff] [blame] | 989 | if (pmd_none(*pmd)) |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 990 | goto no_page_table; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 991 | |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 992 | if (pmd_huge(*pmd)) { |
| 993 | BUG_ON(flags & FOLL_GET); |
| 994 | page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); |
| 995 | goto out; |
| 996 | } |
| 997 | |
Hugh Dickins | aeed5fc | 2008-05-06 20:49:23 +0100 | [diff] [blame] | 998 | if (unlikely(pmd_bad(*pmd))) |
| 999 | goto no_page_table; |
| 1000 | |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1001 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1002 | |
| 1003 | pte = *ptep; |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1004 | if (!pte_present(pte)) |
Linus Torvalds | 89f5b7d | 2008-06-20 11:18:25 -0700 | [diff] [blame] | 1005 | goto no_page; |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1006 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
| 1007 | goto unlock; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1008 | page = vm_normal_page(vma, address, pte); |
| 1009 | if (unlikely(!page)) |
Linus Torvalds | 89f5b7d | 2008-06-20 11:18:25 -0700 | [diff] [blame] | 1010 | goto bad_page; |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1011 | |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1012 | if (flags & FOLL_GET) |
| 1013 | get_page(page); |
| 1014 | if (flags & FOLL_TOUCH) { |
| 1015 | if ((flags & FOLL_WRITE) && |
| 1016 | !pte_dirty(pte) && !PageDirty(page)) |
| 1017 | set_page_dirty(page); |
| 1018 | mark_page_accessed(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1019 | } |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1020 | unlock: |
| 1021 | pte_unmap_unlock(ptep, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1022 | out: |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1023 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1024 | |
Linus Torvalds | 89f5b7d | 2008-06-20 11:18:25 -0700 | [diff] [blame] | 1025 | bad_page: |
| 1026 | pte_unmap_unlock(ptep, ptl); |
| 1027 | return ERR_PTR(-EFAULT); |
| 1028 | |
| 1029 | no_page: |
| 1030 | pte_unmap_unlock(ptep, ptl); |
| 1031 | if (!pte_none(pte)) |
| 1032 | return page; |
| 1033 | /* Fall through to ZERO_PAGE handling */ |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1034 | no_page_table: |
| 1035 | /* |
| 1036 | * When core dumping an enormous anonymous area that nobody |
| 1037 | * has touched so far, we don't want to allocate page tables. |
| 1038 | */ |
| 1039 | if (flags & FOLL_ANON) { |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 1040 | page = ZERO_PAGE(0); |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1041 | if (flags & FOLL_GET) |
| 1042 | get_page(page); |
| 1043 | BUG_ON(flags & FOLL_WRITE); |
| 1044 | } |
| 1045 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1046 | } |
| 1047 | |
Linus Torvalds | 672ca28 | 2008-06-23 11:21:37 -0700 | [diff] [blame] | 1048 | /* Can we do the FOLL_ANON optimization? */ |
| 1049 | static inline int use_zero_page(struct vm_area_struct *vma) |
| 1050 | { |
| 1051 | /* |
| 1052 | * We don't want to optimize FOLL_ANON for make_pages_present() |
| 1053 | * when it tries to page in a VM_LOCKED region. As to VM_SHARED, |
| 1054 | * we want to get the page from the page tables to make sure |
| 1055 | * that we serialize and update with any other user of that |
| 1056 | * mapping. |
| 1057 | */ |
| 1058 | if (vma->vm_flags & (VM_LOCKED | VM_SHARED)) |
| 1059 | return 0; |
| 1060 | /* |
Nick Piggin | 0d71d10a | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 1061 | * 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] | 1062 | */ |
Nick Piggin | 0d71d10a | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 1063 | return !vma->vm_ops || !vma->vm_ops->fault; |
Linus Torvalds | 672ca28 | 2008-06-23 11:21:37 -0700 | [diff] [blame] | 1064 | } |
| 1065 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1066 | int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
| 1067 | unsigned long start, int len, int write, int force, |
| 1068 | struct page **pages, struct vm_area_struct **vmas) |
| 1069 | { |
| 1070 | int i; |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1071 | unsigned int vm_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1072 | |
Jonathan Corbet | 900cf08 | 2008-02-11 16:17:33 -0700 | [diff] [blame] | 1073 | if (len <= 0) |
| 1074 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1075 | /* |
| 1076 | * Require read or write permissions. |
| 1077 | * If 'force' is set, we only require the "MAY" flags. |
| 1078 | */ |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1079 | vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); |
| 1080 | vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1081 | i = 0; |
| 1082 | |
| 1083 | do { |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1084 | struct vm_area_struct *vma; |
| 1085 | unsigned int foll_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1086 | |
| 1087 | vma = find_extend_vma(mm, start); |
| 1088 | if (!vma && in_gate_area(tsk, start)) { |
| 1089 | unsigned long pg = start & PAGE_MASK; |
| 1090 | struct vm_area_struct *gate_vma = get_gate_vma(tsk); |
| 1091 | pgd_t *pgd; |
| 1092 | pud_t *pud; |
| 1093 | pmd_t *pmd; |
| 1094 | pte_t *pte; |
| 1095 | if (write) /* user gate pages are read-only */ |
| 1096 | return i ? : -EFAULT; |
| 1097 | if (pg > TASK_SIZE) |
| 1098 | pgd = pgd_offset_k(pg); |
| 1099 | else |
| 1100 | pgd = pgd_offset_gate(mm, pg); |
| 1101 | BUG_ON(pgd_none(*pgd)); |
| 1102 | pud = pud_offset(pgd, pg); |
| 1103 | BUG_ON(pud_none(*pud)); |
| 1104 | pmd = pmd_offset(pud, pg); |
Hugh Dickins | 690dbe1 | 2005-08-01 21:11:42 -0700 | [diff] [blame] | 1105 | if (pmd_none(*pmd)) |
| 1106 | return i ? : -EFAULT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1107 | pte = pte_offset_map(pmd, pg); |
Hugh Dickins | 690dbe1 | 2005-08-01 21:11:42 -0700 | [diff] [blame] | 1108 | if (pte_none(*pte)) { |
| 1109 | pte_unmap(pte); |
| 1110 | return i ? : -EFAULT; |
| 1111 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1112 | if (pages) { |
Nick Piggin | fa2a455 | 2005-11-29 18:43:17 +1100 | [diff] [blame] | 1113 | struct page *page = vm_normal_page(gate_vma, start, *pte); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1114 | pages[i] = page; |
| 1115 | if (page) |
| 1116 | get_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1117 | } |
| 1118 | pte_unmap(pte); |
| 1119 | if (vmas) |
| 1120 | vmas[i] = gate_vma; |
| 1121 | i++; |
| 1122 | start += PAGE_SIZE; |
| 1123 | len--; |
| 1124 | continue; |
| 1125 | } |
| 1126 | |
Linus Torvalds | 1ff8038 | 2005-12-12 16:24:33 -0800 | [diff] [blame] | 1127 | if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP)) |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1128 | || !(vm_flags & vma->vm_flags)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1129 | return i ? : -EFAULT; |
| 1130 | |
| 1131 | if (is_vm_hugetlb_page(vma)) { |
| 1132 | i = follow_hugetlb_page(mm, vma, pages, vmas, |
Adam Litke | 5b23dbe | 2007-11-14 16:59:33 -0800 | [diff] [blame] | 1133 | &start, &len, i, write); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1134 | continue; |
| 1135 | } |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1136 | |
| 1137 | foll_flags = FOLL_TOUCH; |
| 1138 | if (pages) |
| 1139 | foll_flags |= FOLL_GET; |
Linus Torvalds | 672ca28 | 2008-06-23 11:21:37 -0700 | [diff] [blame] | 1140 | if (!write && use_zero_page(vma)) |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1141 | foll_flags |= FOLL_ANON; |
| 1142 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1143 | do { |
Hugh Dickins | 08ef472 | 2005-06-21 17:15:10 -0700 | [diff] [blame] | 1144 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1145 | |
Ethan Solomita | 462e00c | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 1146 | /* |
| 1147 | * If tsk is ooming, cut off its access to large memory |
| 1148 | * allocations. It has a pending SIGKILL, but it can't |
| 1149 | * be processed until returning to user space. |
| 1150 | */ |
| 1151 | if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE))) |
Oleg Nesterov | 7a36a75 | 2008-07-04 09:59:28 -0700 | [diff] [blame] | 1152 | return i ? i : -ENOMEM; |
Ethan Solomita | 462e00c | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 1153 | |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1154 | if (write) |
| 1155 | foll_flags |= FOLL_WRITE; |
| 1156 | |
| 1157 | cond_resched(); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1158 | while (!(page = follow_page(vma, start, foll_flags))) { |
Linus Torvalds | a68d2eb | 2005-08-03 10:07:09 -0700 | [diff] [blame] | 1159 | int ret; |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 1160 | ret = handle_mm_fault(mm, vma, start, |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1161 | foll_flags & FOLL_WRITE); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 1162 | if (ret & VM_FAULT_ERROR) { |
| 1163 | if (ret & VM_FAULT_OOM) |
| 1164 | return i ? i : -ENOMEM; |
| 1165 | else if (ret & VM_FAULT_SIGBUS) |
| 1166 | return i ? i : -EFAULT; |
| 1167 | BUG(); |
| 1168 | } |
| 1169 | if (ret & VM_FAULT_MAJOR) |
| 1170 | tsk->maj_flt++; |
| 1171 | else |
| 1172 | tsk->min_flt++; |
| 1173 | |
Linus Torvalds | a68d2eb | 2005-08-03 10:07:09 -0700 | [diff] [blame] | 1174 | /* |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 1175 | * The VM_FAULT_WRITE bit tells us that |
| 1176 | * do_wp_page has broken COW when necessary, |
| 1177 | * even if maybe_mkwrite decided not to set |
| 1178 | * pte_write. We can thus safely do subsequent |
| 1179 | * page lookups as if they were reads. |
Linus Torvalds | a68d2eb | 2005-08-03 10:07:09 -0700 | [diff] [blame] | 1180 | */ |
| 1181 | if (ret & VM_FAULT_WRITE) |
Hugh Dickins | deceb6c | 2005-10-29 18:16:33 -0700 | [diff] [blame] | 1182 | foll_flags &= ~FOLL_WRITE; |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 1183 | |
Benjamin Herrenschmidt | 7f7bbbe | 2006-10-06 00:43:53 -0700 | [diff] [blame] | 1184 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1185 | } |
Linus Torvalds | 89f5b7d | 2008-06-20 11:18:25 -0700 | [diff] [blame] | 1186 | if (IS_ERR(page)) |
| 1187 | return i ? i : PTR_ERR(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1188 | if (pages) { |
Hugh Dickins | 08ef472 | 2005-06-21 17:15:10 -0700 | [diff] [blame] | 1189 | pages[i] = page; |
James Bottomley | 03beb07 | 2006-03-26 01:36:57 -0800 | [diff] [blame] | 1190 | |
Russell King | a6f36be | 2006-12-30 22:24:19 +0000 | [diff] [blame] | 1191 | flush_anon_page(vma, page, start); |
Hugh Dickins | 08ef472 | 2005-06-21 17:15:10 -0700 | [diff] [blame] | 1192 | flush_dcache_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1193 | } |
| 1194 | if (vmas) |
| 1195 | vmas[i] = vma; |
| 1196 | i++; |
| 1197 | start += PAGE_SIZE; |
| 1198 | len--; |
Hugh Dickins | 08ef472 | 2005-06-21 17:15:10 -0700 | [diff] [blame] | 1199 | } while (len && start < vma->vm_end); |
Hugh Dickins | 08ef472 | 2005-06-21 17:15:10 -0700 | [diff] [blame] | 1200 | } while (len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1201 | return i; |
| 1202 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1203 | EXPORT_SYMBOL(get_user_pages); |
| 1204 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 1205 | pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr, |
| 1206 | spinlock_t **ptl) |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1207 | { |
| 1208 | pgd_t * pgd = pgd_offset(mm, addr); |
| 1209 | pud_t * pud = pud_alloc(mm, pgd, addr); |
| 1210 | if (pud) { |
Trond Myklebust | 49c91fb | 2005-11-29 19:27:22 -0500 | [diff] [blame] | 1211 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1212 | if (pmd) |
| 1213 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
| 1214 | } |
| 1215 | return NULL; |
| 1216 | } |
| 1217 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1218 | /* |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1219 | * This is the old fallback for page remapping. |
| 1220 | * |
| 1221 | * For historical reasons, it only allows reserved pages. Only |
| 1222 | * old drivers should use this, and they needed to mark their |
| 1223 | * pages reserved for the old functions anyway. |
| 1224 | */ |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1225 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
| 1226 | struct page *page, pgprot_t prot) |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1227 | { |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1228 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1229 | int retval; |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1230 | pte_t *pte; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1231 | spinlock_t *ptl; |
| 1232 | |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame] | 1233 | retval = mem_cgroup_charge(page, mm, GFP_KERNEL); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1234 | if (retval) |
| 1235 | goto out; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1236 | |
| 1237 | retval = -EINVAL; |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1238 | if (PageAnon(page)) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1239 | goto out_uncharge; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1240 | retval = -ENOMEM; |
| 1241 | flush_dcache_page(page); |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1242 | pte = get_locked_pte(mm, addr, &ptl); |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1243 | if (!pte) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1244 | goto out_uncharge; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1245 | retval = -EBUSY; |
| 1246 | if (!pte_none(*pte)) |
| 1247 | goto out_unlock; |
| 1248 | |
| 1249 | /* Ok, finally just insert the thing.. */ |
| 1250 | get_page(page); |
| 1251 | inc_mm_counter(mm, file_rss); |
| 1252 | page_add_file_rmap(page); |
| 1253 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); |
| 1254 | |
| 1255 | retval = 0; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1256 | pte_unmap_unlock(pte, ptl); |
| 1257 | return retval; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1258 | out_unlock: |
| 1259 | pte_unmap_unlock(pte, ptl); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1260 | out_uncharge: |
| 1261 | mem_cgroup_uncharge_page(page); |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1262 | out: |
| 1263 | return retval; |
| 1264 | } |
| 1265 | |
Rolf Eike Beer | bfa5bf6 | 2006-09-25 23:31:22 -0700 | [diff] [blame] | 1266 | /** |
| 1267 | * vm_insert_page - insert single page into user vma |
| 1268 | * @vma: user vma to map to |
| 1269 | * @addr: target user address of this page |
| 1270 | * @page: source kernel page |
| 1271 | * |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1272 | * This allows drivers to insert individual pages they've allocated |
| 1273 | * into a user vma. |
| 1274 | * |
| 1275 | * The page has to be a nice clean _individual_ kernel allocation. |
| 1276 | * If you allocate a compound page, you need to have marked it as |
| 1277 | * such (__GFP_COMP), or manually just split the page up yourself |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 1278 | * (see split_page()). |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1279 | * |
| 1280 | * NOTE! Traditionally this was done with "remap_pfn_range()" which |
| 1281 | * took an arbitrary page protection parameter. This doesn't allow |
| 1282 | * that. Your vma protection will have to be set up correctly, which |
| 1283 | * means that if you want a shared writable mapping, you'd better |
| 1284 | * ask for a shared writable mapping! |
| 1285 | * |
| 1286 | * The page does not need to be reserved. |
| 1287 | */ |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1288 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
| 1289 | struct page *page) |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1290 | { |
| 1291 | if (addr < vma->vm_start || addr >= vma->vm_end) |
| 1292 | return -EFAULT; |
| 1293 | if (!page_count(page)) |
| 1294 | return -EINVAL; |
Linus Torvalds | 4d7672b | 2005-12-16 10:21:23 -0800 | [diff] [blame] | 1295 | vma->vm_flags |= VM_INSERTPAGE; |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1296 | return insert_page(vma, addr, page, vma->vm_page_prot); |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1297 | } |
Linus Torvalds | e3c3374 | 2005-12-03 20:48:11 -0800 | [diff] [blame] | 1298 | EXPORT_SYMBOL(vm_insert_page); |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1299 | |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1300 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
| 1301 | unsigned long pfn, pgprot_t prot) |
| 1302 | { |
| 1303 | struct mm_struct *mm = vma->vm_mm; |
| 1304 | int retval; |
| 1305 | pte_t *pte, entry; |
| 1306 | spinlock_t *ptl; |
| 1307 | |
| 1308 | retval = -ENOMEM; |
| 1309 | pte = get_locked_pte(mm, addr, &ptl); |
| 1310 | if (!pte) |
| 1311 | goto out; |
| 1312 | retval = -EBUSY; |
| 1313 | if (!pte_none(*pte)) |
| 1314 | goto out_unlock; |
| 1315 | |
| 1316 | /* Ok, finally just insert the thing.. */ |
| 1317 | entry = pte_mkspecial(pfn_pte(pfn, prot)); |
| 1318 | set_pte_at(mm, addr, pte, entry); |
| 1319 | update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */ |
| 1320 | |
| 1321 | retval = 0; |
| 1322 | out_unlock: |
| 1323 | pte_unmap_unlock(pte, ptl); |
| 1324 | out: |
| 1325 | return retval; |
| 1326 | } |
| 1327 | |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1328 | /** |
| 1329 | * vm_insert_pfn - insert single pfn into user vma |
| 1330 | * @vma: user vma to map to |
| 1331 | * @addr: target user address of this page |
| 1332 | * @pfn: source kernel pfn |
| 1333 | * |
| 1334 | * Similar to vm_inert_page, this allows drivers to insert individual pages |
| 1335 | * they've allocated into a user vma. Same comments apply. |
| 1336 | * |
| 1337 | * This function should only be called from a vm_ops->fault handler, and |
| 1338 | * in that case the handler should return NULL. |
Nick Piggin | 0d71d10a | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 1339 | * |
| 1340 | * vma cannot be a COW mapping. |
| 1341 | * |
| 1342 | * As this is called only for pages that do not currently exist, we |
| 1343 | * do not need to flush old virtual caches or the TLB. |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1344 | */ |
| 1345 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1346 | unsigned long pfn) |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1347 | { |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 1348 | /* |
| 1349 | * Technically, architectures with pte_special can avoid all these |
| 1350 | * restrictions (same for remap_pfn_range). However we would like |
| 1351 | * consistency in testing and feature parity among all, so we should |
| 1352 | * try to keep these invariants in place for everybody. |
| 1353 | */ |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 1354 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
| 1355 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == |
| 1356 | (VM_PFNMAP|VM_MIXEDMAP)); |
| 1357 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); |
| 1358 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1359 | |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1360 | if (addr < vma->vm_start || addr >= vma->vm_end) |
| 1361 | return -EFAULT; |
| 1362 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1363 | } |
| 1364 | EXPORT_SYMBOL(vm_insert_pfn); |
| 1365 | |
Nick Piggin | 423bad6 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1366 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
| 1367 | unsigned long pfn) |
| 1368 | { |
| 1369 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); |
| 1370 | |
| 1371 | if (addr < vma->vm_start || addr >= vma->vm_end) |
| 1372 | return -EFAULT; |
| 1373 | |
| 1374 | /* |
| 1375 | * If we don't have pte special, then we have to use the pfn_valid() |
| 1376 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* |
| 1377 | * refcount the page if pfn_valid is true (hence insert_page rather |
| 1378 | * than insert_pfn). |
| 1379 | */ |
| 1380 | if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { |
| 1381 | struct page *page; |
| 1382 | |
| 1383 | page = pfn_to_page(pfn); |
| 1384 | return insert_page(vma, addr, page, vma->vm_page_prot); |
| 1385 | } |
| 1386 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); |
| 1387 | } |
| 1388 | EXPORT_SYMBOL(vm_insert_mixed); |
| 1389 | |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1390 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1391 | * maps a range of physical memory into the requested pages. the old |
| 1392 | * mappings are removed. any references to nonexistent pages results |
| 1393 | * in null mappings (currently treated as "copy-on-access") |
| 1394 | */ |
| 1395 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, |
| 1396 | unsigned long addr, unsigned long end, |
| 1397 | unsigned long pfn, pgprot_t prot) |
| 1398 | { |
| 1399 | pte_t *pte; |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 1400 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1401 | |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 1402 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1403 | if (!pte) |
| 1404 | return -ENOMEM; |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 1405 | arch_enter_lazy_mmu_mode(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1406 | do { |
| 1407 | BUG_ON(!pte_none(*pte)); |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 1408 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1409 | pfn++; |
| 1410 | } while (pte++, addr += PAGE_SIZE, addr != end); |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 1411 | arch_leave_lazy_mmu_mode(); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 1412 | pte_unmap_unlock(pte - 1, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1413 | return 0; |
| 1414 | } |
| 1415 | |
| 1416 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, |
| 1417 | unsigned long addr, unsigned long end, |
| 1418 | unsigned long pfn, pgprot_t prot) |
| 1419 | { |
| 1420 | pmd_t *pmd; |
| 1421 | unsigned long next; |
| 1422 | |
| 1423 | pfn -= addr >> PAGE_SHIFT; |
| 1424 | pmd = pmd_alloc(mm, pud, addr); |
| 1425 | if (!pmd) |
| 1426 | return -ENOMEM; |
| 1427 | do { |
| 1428 | next = pmd_addr_end(addr, end); |
| 1429 | if (remap_pte_range(mm, pmd, addr, next, |
| 1430 | pfn + (addr >> PAGE_SHIFT), prot)) |
| 1431 | return -ENOMEM; |
| 1432 | } while (pmd++, addr = next, addr != end); |
| 1433 | return 0; |
| 1434 | } |
| 1435 | |
| 1436 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, |
| 1437 | unsigned long addr, unsigned long end, |
| 1438 | unsigned long pfn, pgprot_t prot) |
| 1439 | { |
| 1440 | pud_t *pud; |
| 1441 | unsigned long next; |
| 1442 | |
| 1443 | pfn -= addr >> PAGE_SHIFT; |
| 1444 | pud = pud_alloc(mm, pgd, addr); |
| 1445 | if (!pud) |
| 1446 | return -ENOMEM; |
| 1447 | do { |
| 1448 | next = pud_addr_end(addr, end); |
| 1449 | if (remap_pmd_range(mm, pud, addr, next, |
| 1450 | pfn + (addr >> PAGE_SHIFT), prot)) |
| 1451 | return -ENOMEM; |
| 1452 | } while (pud++, addr = next, addr != end); |
| 1453 | return 0; |
| 1454 | } |
| 1455 | |
Rolf Eike Beer | bfa5bf6 | 2006-09-25 23:31:22 -0700 | [diff] [blame] | 1456 | /** |
| 1457 | * remap_pfn_range - remap kernel memory to userspace |
| 1458 | * @vma: user vma to map to |
| 1459 | * @addr: target user address to start at |
| 1460 | * @pfn: physical address of kernel memory |
| 1461 | * @size: size of map area |
| 1462 | * @prot: page protection flags for this mapping |
| 1463 | * |
| 1464 | * Note: this is only safe if the mm semaphore is held when called. |
| 1465 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1466 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
| 1467 | unsigned long pfn, unsigned long size, pgprot_t prot) |
| 1468 | { |
| 1469 | pgd_t *pgd; |
| 1470 | unsigned long next; |
Hugh Dickins | 2d15cab | 2005-06-25 14:54:33 -0700 | [diff] [blame] | 1471 | unsigned long end = addr + PAGE_ALIGN(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1472 | struct mm_struct *mm = vma->vm_mm; |
| 1473 | int err; |
| 1474 | |
| 1475 | /* |
| 1476 | * Physically remapped pages are special. Tell the |
| 1477 | * rest of the world about it: |
| 1478 | * VM_IO tells people not to look at these pages |
| 1479 | * (accesses can have side effects). |
Hugh Dickins | 0b14c17 | 2005-11-21 21:32:15 -0800 | [diff] [blame] | 1480 | * VM_RESERVED is specified all over the place, because |
| 1481 | * in 2.4 it kept swapout's vma scan off this vma; but |
| 1482 | * in 2.6 the LRU scan won't even find its pages, so this |
| 1483 | * flag means no more than count its pages in reserved_vm, |
| 1484 | * and omit it from core dump, even when VM_IO turned off. |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1485 | * VM_PFNMAP tells the core MM that the base pages are just |
| 1486 | * raw PFN mappings, and do not have a "struct page" associated |
| 1487 | * with them. |
Linus Torvalds | fb155c1 | 2005-12-11 19:46:02 -0800 | [diff] [blame] | 1488 | * |
| 1489 | * There's a horrible special case to handle copy-on-write |
| 1490 | * behaviour that some programs depend on. We mark the "original" |
| 1491 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1492 | */ |
Linus Torvalds | 6712117 | 2005-12-11 20:38:17 -0800 | [diff] [blame] | 1493 | if (is_cow_mapping(vma->vm_flags)) { |
Linus Torvalds | fb155c1 | 2005-12-11 19:46:02 -0800 | [diff] [blame] | 1494 | if (addr != vma->vm_start || end != vma->vm_end) |
Linus Torvalds | 7fc7e2e | 2005-12-11 19:57:52 -0800 | [diff] [blame] | 1495 | return -EINVAL; |
Linus Torvalds | fb155c1 | 2005-12-11 19:46:02 -0800 | [diff] [blame] | 1496 | vma->vm_pgoff = pfn; |
| 1497 | } |
| 1498 | |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1499 | vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1500 | |
| 1501 | BUG_ON(addr >= end); |
| 1502 | pfn -= addr >> PAGE_SHIFT; |
| 1503 | pgd = pgd_offset(mm, addr); |
| 1504 | flush_cache_range(vma, addr, end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1505 | do { |
| 1506 | next = pgd_addr_end(addr, end); |
| 1507 | err = remap_pud_range(mm, pgd, addr, next, |
| 1508 | pfn + (addr >> PAGE_SHIFT), prot); |
| 1509 | if (err) |
| 1510 | break; |
| 1511 | } while (pgd++, addr = next, addr != end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1512 | return err; |
| 1513 | } |
| 1514 | EXPORT_SYMBOL(remap_pfn_range); |
| 1515 | |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1516 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
| 1517 | unsigned long addr, unsigned long end, |
| 1518 | pte_fn_t fn, void *data) |
| 1519 | { |
| 1520 | pte_t *pte; |
| 1521 | int err; |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 1522 | pgtable_t token; |
Borislav Petkov | 9490991 | 2007-05-06 14:49:17 -0700 | [diff] [blame] | 1523 | spinlock_t *uninitialized_var(ptl); |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1524 | |
| 1525 | pte = (mm == &init_mm) ? |
| 1526 | pte_alloc_kernel(pmd, addr) : |
| 1527 | pte_alloc_map_lock(mm, pmd, addr, &ptl); |
| 1528 | if (!pte) |
| 1529 | return -ENOMEM; |
| 1530 | |
| 1531 | BUG_ON(pmd_huge(*pmd)); |
| 1532 | |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 1533 | token = pmd_pgtable(*pmd); |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1534 | |
| 1535 | do { |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 1536 | err = fn(pte, token, addr, data); |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1537 | if (err) |
| 1538 | break; |
| 1539 | } while (pte++, addr += PAGE_SIZE, addr != end); |
| 1540 | |
| 1541 | if (mm != &init_mm) |
| 1542 | pte_unmap_unlock(pte-1, ptl); |
| 1543 | return err; |
| 1544 | } |
| 1545 | |
| 1546 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, |
| 1547 | unsigned long addr, unsigned long end, |
| 1548 | pte_fn_t fn, void *data) |
| 1549 | { |
| 1550 | pmd_t *pmd; |
| 1551 | unsigned long next; |
| 1552 | int err; |
| 1553 | |
| 1554 | pmd = pmd_alloc(mm, pud, addr); |
| 1555 | if (!pmd) |
| 1556 | return -ENOMEM; |
| 1557 | do { |
| 1558 | next = pmd_addr_end(addr, end); |
| 1559 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); |
| 1560 | if (err) |
| 1561 | break; |
| 1562 | } while (pmd++, addr = next, addr != end); |
| 1563 | return err; |
| 1564 | } |
| 1565 | |
| 1566 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, |
| 1567 | unsigned long addr, unsigned long end, |
| 1568 | pte_fn_t fn, void *data) |
| 1569 | { |
| 1570 | pud_t *pud; |
| 1571 | unsigned long next; |
| 1572 | int err; |
| 1573 | |
| 1574 | pud = pud_alloc(mm, pgd, addr); |
| 1575 | if (!pud) |
| 1576 | return -ENOMEM; |
| 1577 | do { |
| 1578 | next = pud_addr_end(addr, end); |
| 1579 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); |
| 1580 | if (err) |
| 1581 | break; |
| 1582 | } while (pud++, addr = next, addr != end); |
| 1583 | return err; |
| 1584 | } |
| 1585 | |
| 1586 | /* |
| 1587 | * Scan a region of virtual memory, filling in page tables as necessary |
| 1588 | * and calling a provided function on each leaf page table. |
| 1589 | */ |
| 1590 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, |
| 1591 | unsigned long size, pte_fn_t fn, void *data) |
| 1592 | { |
| 1593 | pgd_t *pgd; |
| 1594 | unsigned long next; |
| 1595 | unsigned long end = addr + size; |
| 1596 | int err; |
| 1597 | |
| 1598 | BUG_ON(addr >= end); |
| 1599 | pgd = pgd_offset(mm, addr); |
| 1600 | do { |
| 1601 | next = pgd_addr_end(addr, end); |
| 1602 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); |
| 1603 | if (err) |
| 1604 | break; |
| 1605 | } while (pgd++, addr = next, addr != end); |
| 1606 | return err; |
| 1607 | } |
| 1608 | EXPORT_SYMBOL_GPL(apply_to_page_range); |
| 1609 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1610 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1611 | * handle_pte_fault chooses page fault handler according to an entry |
| 1612 | * which was read non-atomically. Before making any commitment, on |
| 1613 | * those architectures or configurations (e.g. i386 with PAE) which |
| 1614 | * might give a mix of unmatched parts, do_swap_page and do_file_page |
| 1615 | * must check under lock before unmapping the pte and proceeding |
| 1616 | * (but do_wp_page is only called after already making such a check; |
| 1617 | * and do_anonymous_page and do_no_page can safely check later on). |
| 1618 | */ |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 1619 | 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] | 1620 | pte_t *page_table, pte_t orig_pte) |
| 1621 | { |
| 1622 | int same = 1; |
| 1623 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) |
| 1624 | if (sizeof(pte_t) > sizeof(unsigned long)) { |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 1625 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
| 1626 | spin_lock(ptl); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1627 | same = pte_same(*page_table, orig_pte); |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 1628 | spin_unlock(ptl); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1629 | } |
| 1630 | #endif |
| 1631 | pte_unmap(page_table); |
| 1632 | return same; |
| 1633 | } |
| 1634 | |
| 1635 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1636 | * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when |
| 1637 | * servicing faults for write access. In the normal case, do always want |
| 1638 | * pte_mkwrite. But get_user_pages can cause write faults for mappings |
| 1639 | * that do not have writing enabled, when used by access_process_vm. |
| 1640 | */ |
| 1641 | static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma) |
| 1642 | { |
| 1643 | if (likely(vma->vm_flags & VM_WRITE)) |
| 1644 | pte = pte_mkwrite(pte); |
| 1645 | return pte; |
| 1646 | } |
| 1647 | |
Atsushi Nemoto | 9de455b | 2006-12-12 17:14:55 +0000 | [diff] [blame] | 1648 | 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] | 1649 | { |
| 1650 | /* |
| 1651 | * If the source page was a PFN mapping, we don't have |
| 1652 | * a "struct page" for it. We do a best-effort copy by |
| 1653 | * just copying from the original user address. If that |
| 1654 | * fails, we just zero-fill it. Live with it. |
| 1655 | */ |
| 1656 | if (unlikely(!src)) { |
| 1657 | void *kaddr = kmap_atomic(dst, KM_USER0); |
Linus Torvalds | 5d2a2db | 2005-11-29 14:07:55 -0800 | [diff] [blame] | 1658 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
| 1659 | |
| 1660 | /* |
| 1661 | * This really shouldn't fail, because the page is there |
| 1662 | * in the page tables. But it might just be unreadable, |
| 1663 | * in which case we just give up and fill the result with |
| 1664 | * zeroes. |
| 1665 | */ |
| 1666 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1667 | memset(kaddr, 0, PAGE_SIZE); |
| 1668 | kunmap_atomic(kaddr, KM_USER0); |
Dmitriy Monakhov | c4ec7b0 | 2006-10-19 23:29:08 -0700 | [diff] [blame] | 1669 | flush_dcache_page(dst); |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 1670 | } else |
| 1671 | copy_user_highpage(dst, src, va, vma); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1672 | } |
| 1673 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1674 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1675 | * This routine handles present pages, when users try to write |
| 1676 | * to a shared page. It is done by copying the page to a new address |
| 1677 | * and decrementing the shared-page counter for the old page. |
| 1678 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1679 | * Note that this routine assumes that the protection checks have been |
| 1680 | * done by the caller (the low-level page fault routine in most cases). |
| 1681 | * Thus we can safely just mark it writable once we've done any necessary |
| 1682 | * COW. |
| 1683 | * |
| 1684 | * We also mark the page dirty at this point even though the page will |
| 1685 | * change only once the write actually happens. This avoids a few races, |
| 1686 | * and potentially makes it more efficient. |
| 1687 | * |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1688 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 1689 | * but allow concurrent faults), with pte both mapped and locked. |
| 1690 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1691 | */ |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 1692 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, |
| 1693 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1694 | spinlock_t *ptl, pte_t orig_pte) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1695 | { |
Hugh Dickins | e5bbe4d | 2005-11-29 16:54:51 +0000 | [diff] [blame] | 1696 | struct page *old_page, *new_page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1697 | pte_t entry; |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 1698 | int reuse = 0, ret = 0; |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 1699 | int page_mkwrite = 0; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1700 | struct page *dirty_page = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1701 | |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1702 | old_page = vm_normal_page(vma, address, orig_pte); |
Peter Zijlstra | 251b97f | 2008-07-04 09:59:24 -0700 | [diff] [blame] | 1703 | if (!old_page) { |
| 1704 | /* |
| 1705 | * VM_MIXEDMAP !pfn_valid() case |
| 1706 | * |
| 1707 | * We should not cow pages in a shared writeable mapping. |
| 1708 | * Just mark the pages writable as we can't do any dirty |
| 1709 | * accounting on raw pfn maps. |
| 1710 | */ |
| 1711 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
| 1712 | (VM_WRITE|VM_SHARED)) |
| 1713 | goto reuse; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1714 | goto gotten; |
Peter Zijlstra | 251b97f | 2008-07-04 09:59:24 -0700 | [diff] [blame] | 1715 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1716 | |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1717 | /* |
Peter Zijlstra | ee6a645 | 2006-09-25 23:31:00 -0700 | [diff] [blame] | 1718 | * Take out anonymous pages first, anonymous shared vmas are |
| 1719 | * not dirty accountable. |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1720 | */ |
Peter Zijlstra | ee6a645 | 2006-09-25 23:31:00 -0700 | [diff] [blame] | 1721 | if (PageAnon(old_page)) { |
| 1722 | if (!TestSetPageLocked(old_page)) { |
| 1723 | reuse = can_share_swap_page(old_page); |
| 1724 | unlock_page(old_page); |
| 1725 | } |
| 1726 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1727 | (VM_WRITE|VM_SHARED))) { |
Peter Zijlstra | ee6a645 | 2006-09-25 23:31:00 -0700 | [diff] [blame] | 1728 | /* |
| 1729 | * Only catch write-faults on shared writable pages, |
| 1730 | * read-only shared pages can get COWed by |
| 1731 | * get_user_pages(.write=1, .force=1). |
| 1732 | */ |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1733 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
| 1734 | /* |
| 1735 | * Notify the address space that the page is about to |
| 1736 | * become writable so that it can prohibit this or wait |
| 1737 | * for the page to get into an appropriate state. |
| 1738 | * |
| 1739 | * We do this without the lock held, so that it can |
| 1740 | * sleep if it needs to. |
| 1741 | */ |
| 1742 | page_cache_get(old_page); |
| 1743 | pte_unmap_unlock(page_table, ptl); |
| 1744 | |
| 1745 | if (vma->vm_ops->page_mkwrite(vma, old_page) < 0) |
| 1746 | goto unwritable_page; |
| 1747 | |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1748 | /* |
| 1749 | * Since we dropped the lock we need to revalidate |
| 1750 | * the PTE as someone else may have changed it. If |
| 1751 | * they did, we just return, as we can count on the |
| 1752 | * MMU to tell us if they didn't also make it writable. |
| 1753 | */ |
| 1754 | page_table = pte_offset_map_lock(mm, pmd, address, |
| 1755 | &ptl); |
Hugh Dickins | c3704ce | 2007-02-10 01:43:00 -0800 | [diff] [blame] | 1756 | page_cache_release(old_page); |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1757 | if (!pte_same(*page_table, orig_pte)) |
| 1758 | goto unlock; |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 1759 | |
| 1760 | page_mkwrite = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1761 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1762 | dirty_page = old_page; |
| 1763 | get_page(dirty_page); |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1764 | reuse = 1; |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1765 | } |
| 1766 | |
| 1767 | if (reuse) { |
Peter Zijlstra | 251b97f | 2008-07-04 09:59:24 -0700 | [diff] [blame] | 1768 | reuse: |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1769 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
| 1770 | entry = pte_mkyoung(orig_pte); |
| 1771 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
KAMEZAWA Hiroyuki | 954ffcb | 2007-10-16 01:25:44 -0700 | [diff] [blame] | 1772 | if (ptep_set_access_flags(vma, address, page_table, entry,1)) |
Benjamin Herrenschmidt | 8dab524 | 2007-06-16 10:16:12 -0700 | [diff] [blame] | 1773 | update_mmu_cache(vma, address, entry); |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1774 | ret |= VM_FAULT_WRITE; |
| 1775 | goto unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1776 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1777 | |
| 1778 | /* |
| 1779 | * Ok, we need to copy. Oh, well.. |
| 1780 | */ |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 1781 | page_cache_get(old_page); |
Hugh Dickins | 920fc35 | 2005-11-21 21:32:17 -0800 | [diff] [blame] | 1782 | gotten: |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1783 | pte_unmap_unlock(page_table, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1784 | |
| 1785 | if (unlikely(anon_vma_prepare(vma))) |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 1786 | goto oom; |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 1787 | VM_BUG_ON(old_page == ZERO_PAGE(0)); |
| 1788 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); |
| 1789 | if (!new_page) |
| 1790 | goto oom; |
| 1791 | cow_user_page(new_page, old_page, address, vma); |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 1792 | __SetPageUptodate(new_page); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 1793 | |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame] | 1794 | if (mem_cgroup_charge(new_page, mm, GFP_KERNEL)) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1795 | goto oom_free_new; |
| 1796 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1797 | /* |
| 1798 | * Re-check the pte - we dropped the lock |
| 1799 | */ |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1800 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 1801 | if (likely(pte_same(*page_table, orig_pte))) { |
Hugh Dickins | 920fc35 | 2005-11-21 21:32:17 -0800 | [diff] [blame] | 1802 | if (old_page) { |
Hugh Dickins | 920fc35 | 2005-11-21 21:32:17 -0800 | [diff] [blame] | 1803 | if (!PageAnon(old_page)) { |
| 1804 | dec_mm_counter(mm, file_rss); |
| 1805 | inc_mm_counter(mm, anon_rss); |
| 1806 | } |
| 1807 | } else |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 1808 | inc_mm_counter(mm, anon_rss); |
Ben Collins | eca3513 | 2005-11-29 11:45:26 -0800 | [diff] [blame] | 1809 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 1810 | entry = mk_pte(new_page, vma->vm_page_prot); |
| 1811 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
Siddha, Suresh B | 4ce072f | 2006-09-29 01:58:42 -0700 | [diff] [blame] | 1812 | /* |
| 1813 | * Clear the pte entry and flush it first, before updating the |
| 1814 | * pte with the new entry. This will avoid a race condition |
| 1815 | * seen in the presence of one thread doing SMC and another |
| 1816 | * thread doing COW. |
| 1817 | */ |
| 1818 | ptep_clear_flush(vma, address, page_table); |
| 1819 | set_pte_at(mm, address, page_table, entry); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 1820 | update_mmu_cache(vma, address, entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1821 | lru_cache_add_active(new_page); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 1822 | page_add_new_anon_rmap(new_page, vma, address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1823 | |
Nick Piggin | 945754a | 2008-06-23 14:30:30 +0200 | [diff] [blame] | 1824 | if (old_page) { |
| 1825 | /* |
| 1826 | * Only after switching the pte to the new page may |
| 1827 | * we remove the mapcount here. Otherwise another |
| 1828 | * process may come and find the rmap count decremented |
| 1829 | * before the pte is switched to the new page, and |
| 1830 | * "reuse" the old page writing into it while our pte |
| 1831 | * here still points into it and can be read by other |
| 1832 | * threads. |
| 1833 | * |
| 1834 | * The critical issue is to order this |
| 1835 | * page_remove_rmap with the ptp_clear_flush above. |
| 1836 | * Those stores are ordered by (if nothing else,) |
| 1837 | * the barrier present in the atomic_add_negative |
| 1838 | * in page_remove_rmap. |
| 1839 | * |
| 1840 | * Then the TLB flush in ptep_clear_flush ensures that |
| 1841 | * no process can access the old page before the |
| 1842 | * decremented mapcount is visible. And the old page |
| 1843 | * cannot be reused until after the decremented |
| 1844 | * mapcount is visible. So transitively, TLBs to |
| 1845 | * old page will be flushed before it can be reused. |
| 1846 | */ |
| 1847 | page_remove_rmap(old_page, vma); |
| 1848 | } |
| 1849 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1850 | /* Free the old page.. */ |
| 1851 | new_page = old_page; |
Nick Piggin | f33ea7f | 2005-08-03 20:24:01 +1000 | [diff] [blame] | 1852 | ret |= VM_FAULT_WRITE; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1853 | } else |
| 1854 | mem_cgroup_uncharge_page(new_page); |
| 1855 | |
Hugh Dickins | 920fc35 | 2005-11-21 21:32:17 -0800 | [diff] [blame] | 1856 | if (new_page) |
| 1857 | page_cache_release(new_page); |
| 1858 | if (old_page) |
| 1859 | page_cache_release(old_page); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 1860 | unlock: |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1861 | pte_unmap_unlock(page_table, ptl); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1862 | if (dirty_page) { |
Anton Salikhmetov | 8f7b3d1 | 2008-01-23 02:21:18 +0300 | [diff] [blame] | 1863 | if (vma->vm_file) |
| 1864 | file_update_time(vma->vm_file); |
| 1865 | |
Nick Piggin | 7935289 | 2007-07-19 01:47:22 -0700 | [diff] [blame] | 1866 | /* |
| 1867 | * Yes, Virginia, this is actually required to prevent a race |
| 1868 | * with clear_page_dirty_for_io() from clearing the page dirty |
| 1869 | * bit after it clear all dirty ptes, but before a racing |
| 1870 | * do_wp_page installs a dirty pte. |
| 1871 | * |
| 1872 | * do_no_page is protected similarly. |
| 1873 | */ |
| 1874 | wait_on_page_locked(dirty_page); |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 1875 | set_page_dirty_balance(dirty_page, page_mkwrite); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1876 | put_page(dirty_page); |
| 1877 | } |
Nick Piggin | f33ea7f | 2005-08-03 20:24:01 +1000 | [diff] [blame] | 1878 | return ret; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1879 | oom_free_new: |
Hugh Dickins | 6dbf6d3 | 2008-03-04 14:29:04 -0800 | [diff] [blame] | 1880 | page_cache_release(new_page); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 1881 | oom: |
Hugh Dickins | 920fc35 | 2005-11-21 21:32:17 -0800 | [diff] [blame] | 1882 | if (old_page) |
| 1883 | page_cache_release(old_page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1884 | return VM_FAULT_OOM; |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 1885 | |
| 1886 | unwritable_page: |
| 1887 | page_cache_release(old_page); |
| 1888 | return VM_FAULT_SIGBUS; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1889 | } |
| 1890 | |
| 1891 | /* |
| 1892 | * Helper functions for unmap_mapping_range(). |
| 1893 | * |
| 1894 | * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __ |
| 1895 | * |
| 1896 | * We have to restart searching the prio_tree whenever we drop the lock, |
| 1897 | * since the iterator is only valid while the lock is held, and anyway |
| 1898 | * a later vma might be split and reinserted earlier while lock dropped. |
| 1899 | * |
| 1900 | * The list of nonlinear vmas could be handled more efficiently, using |
| 1901 | * a placeholder, but handle it in the same way until a need is shown. |
| 1902 | * It is important to search the prio_tree before nonlinear list: a vma |
| 1903 | * may become nonlinear and be shifted from prio_tree to nonlinear list |
| 1904 | * while the lock is dropped; but never shifted from list to prio_tree. |
| 1905 | * |
| 1906 | * In order to make forward progress despite restarting the search, |
| 1907 | * vm_truncate_count is used to mark a vma as now dealt with, so we can |
| 1908 | * quickly skip it next time around. Since the prio_tree search only |
| 1909 | * shows us those vmas affected by unmapping the range in question, we |
| 1910 | * can't efficiently keep all vmas in step with mapping->truncate_count: |
| 1911 | * so instead reset them all whenever it wraps back to 0 (then go to 1). |
| 1912 | * mapping->truncate_count and vma->vm_truncate_count are protected by |
| 1913 | * i_mmap_lock. |
| 1914 | * |
| 1915 | * In order to make forward progress despite repeatedly restarting some |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 1916 | * 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] | 1917 | * and restart from that address when we reach that vma again. It might |
| 1918 | * have been split or merged, shrunk or extended, but never shifted: so |
| 1919 | * restart_addr remains valid so long as it remains in the vma's range. |
| 1920 | * unmap_mapping_range forces truncate_count to leap over page-aligned |
| 1921 | * values so we can save vma's restart_addr in its truncate_count field. |
| 1922 | */ |
| 1923 | #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK)) |
| 1924 | |
| 1925 | static void reset_vma_truncate_counts(struct address_space *mapping) |
| 1926 | { |
| 1927 | struct vm_area_struct *vma; |
| 1928 | struct prio_tree_iter iter; |
| 1929 | |
| 1930 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX) |
| 1931 | vma->vm_truncate_count = 0; |
| 1932 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
| 1933 | vma->vm_truncate_count = 0; |
| 1934 | } |
| 1935 | |
| 1936 | static int unmap_mapping_range_vma(struct vm_area_struct *vma, |
| 1937 | unsigned long start_addr, unsigned long end_addr, |
| 1938 | struct zap_details *details) |
| 1939 | { |
| 1940 | unsigned long restart_addr; |
| 1941 | int need_break; |
| 1942 | |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 1943 | /* |
| 1944 | * files that support invalidating or truncating portions of the |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 1945 | * file from under mmaped areas must have their ->fault function |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 1946 | * return a locked page (and set VM_FAULT_LOCKED in the return). |
| 1947 | * This provides synchronisation against concurrent unmapping here. |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 1948 | */ |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 1949 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1950 | again: |
| 1951 | restart_addr = vma->vm_truncate_count; |
| 1952 | if (is_restart_addr(restart_addr) && start_addr < restart_addr) { |
| 1953 | start_addr = restart_addr; |
| 1954 | if (start_addr >= end_addr) { |
| 1955 | /* Top of vma has been split off since last time */ |
| 1956 | vma->vm_truncate_count = details->truncate_count; |
| 1957 | return 0; |
| 1958 | } |
| 1959 | } |
| 1960 | |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 1961 | restart_addr = zap_page_range(vma, start_addr, |
| 1962 | end_addr - start_addr, details); |
Nick Piggin | 95c354f | 2008-01-30 13:31:20 +0100 | [diff] [blame] | 1963 | need_break = need_resched() || spin_needbreak(details->i_mmap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1964 | |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 1965 | if (restart_addr >= end_addr) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1966 | /* We have now completed this vma: mark it so */ |
| 1967 | vma->vm_truncate_count = details->truncate_count; |
| 1968 | if (!need_break) |
| 1969 | return 0; |
| 1970 | } else { |
| 1971 | /* Note restart_addr in vma's truncate_count field */ |
Hugh Dickins | ee39b37 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 1972 | vma->vm_truncate_count = restart_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1973 | if (!need_break) |
| 1974 | goto again; |
| 1975 | } |
| 1976 | |
| 1977 | spin_unlock(details->i_mmap_lock); |
| 1978 | cond_resched(); |
| 1979 | spin_lock(details->i_mmap_lock); |
| 1980 | return -EINTR; |
| 1981 | } |
| 1982 | |
| 1983 | static inline void unmap_mapping_range_tree(struct prio_tree_root *root, |
| 1984 | struct zap_details *details) |
| 1985 | { |
| 1986 | struct vm_area_struct *vma; |
| 1987 | struct prio_tree_iter iter; |
| 1988 | pgoff_t vba, vea, zba, zea; |
| 1989 | |
| 1990 | restart: |
| 1991 | vma_prio_tree_foreach(vma, &iter, root, |
| 1992 | details->first_index, details->last_index) { |
| 1993 | /* Skip quickly over those we have already dealt with */ |
| 1994 | if (vma->vm_truncate_count == details->truncate_count) |
| 1995 | continue; |
| 1996 | |
| 1997 | vba = vma->vm_pgoff; |
| 1998 | vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; |
| 1999 | /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ |
| 2000 | zba = details->first_index; |
| 2001 | if (zba < vba) |
| 2002 | zba = vba; |
| 2003 | zea = details->last_index; |
| 2004 | if (zea > vea) |
| 2005 | zea = vea; |
| 2006 | |
| 2007 | if (unmap_mapping_range_vma(vma, |
| 2008 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
| 2009 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, |
| 2010 | details) < 0) |
| 2011 | goto restart; |
| 2012 | } |
| 2013 | } |
| 2014 | |
| 2015 | static inline void unmap_mapping_range_list(struct list_head *head, |
| 2016 | struct zap_details *details) |
| 2017 | { |
| 2018 | struct vm_area_struct *vma; |
| 2019 | |
| 2020 | /* |
| 2021 | * In nonlinear VMAs there is no correspondence between virtual address |
| 2022 | * offset and file offset. So we must perform an exhaustive search |
| 2023 | * across *all* the pages in each nonlinear VMA, not just the pages |
| 2024 | * whose virtual address lies outside the file truncation point. |
| 2025 | */ |
| 2026 | restart: |
| 2027 | list_for_each_entry(vma, head, shared.vm_set.list) { |
| 2028 | /* Skip quickly over those we have already dealt with */ |
| 2029 | if (vma->vm_truncate_count == details->truncate_count) |
| 2030 | continue; |
| 2031 | details->nonlinear_vma = vma; |
| 2032 | if (unmap_mapping_range_vma(vma, vma->vm_start, |
| 2033 | vma->vm_end, details) < 0) |
| 2034 | goto restart; |
| 2035 | } |
| 2036 | } |
| 2037 | |
| 2038 | /** |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 2039 | * 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] | 2040 | * @mapping: the address space containing mmaps to be unmapped. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2041 | * @holebegin: byte in first page to unmap, relative to the start of |
| 2042 | * the underlying file. This will be rounded down to a PAGE_SIZE |
| 2043 | * boundary. Note that this is different from vmtruncate(), which |
| 2044 | * must keep the partial page. In contrast, we must get rid of |
| 2045 | * partial pages. |
| 2046 | * @holelen: size of prospective hole in bytes. This will be rounded |
| 2047 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the |
| 2048 | * end of the file. |
| 2049 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; |
| 2050 | * but 0 when invalidating pagecache, don't throw away private data. |
| 2051 | */ |
| 2052 | void unmap_mapping_range(struct address_space *mapping, |
| 2053 | loff_t const holebegin, loff_t const holelen, int even_cows) |
| 2054 | { |
| 2055 | struct zap_details details; |
| 2056 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
| 2057 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 2058 | |
| 2059 | /* Check for overflow. */ |
| 2060 | if (sizeof(holelen) > sizeof(hlen)) { |
| 2061 | long long holeend = |
| 2062 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 2063 | if (holeend & ~(long long)ULONG_MAX) |
| 2064 | hlen = ULONG_MAX - hba + 1; |
| 2065 | } |
| 2066 | |
| 2067 | details.check_mapping = even_cows? NULL: mapping; |
| 2068 | details.nonlinear_vma = NULL; |
| 2069 | details.first_index = hba; |
| 2070 | details.last_index = hba + hlen - 1; |
| 2071 | if (details.last_index < details.first_index) |
| 2072 | details.last_index = ULONG_MAX; |
| 2073 | details.i_mmap_lock = &mapping->i_mmap_lock; |
| 2074 | |
| 2075 | spin_lock(&mapping->i_mmap_lock); |
| 2076 | |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2077 | /* Protect against endless unmapping loops */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2078 | mapping->truncate_count++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2079 | if (unlikely(is_restart_addr(mapping->truncate_count))) { |
| 2080 | if (mapping->truncate_count == 0) |
| 2081 | reset_vma_truncate_counts(mapping); |
| 2082 | mapping->truncate_count++; |
| 2083 | } |
| 2084 | details.truncate_count = mapping->truncate_count; |
| 2085 | |
| 2086 | if (unlikely(!prio_tree_empty(&mapping->i_mmap))) |
| 2087 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
| 2088 | if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) |
| 2089 | unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); |
| 2090 | spin_unlock(&mapping->i_mmap_lock); |
| 2091 | } |
| 2092 | EXPORT_SYMBOL(unmap_mapping_range); |
| 2093 | |
Rolf Eike Beer | bfa5bf6 | 2006-09-25 23:31:22 -0700 | [diff] [blame] | 2094 | /** |
| 2095 | * vmtruncate - unmap mappings "freed" by truncate() syscall |
| 2096 | * @inode: inode of the file used |
| 2097 | * @offset: file offset to start truncating |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2098 | * |
| 2099 | * NOTE! We have to be ready to update the memory sharing |
| 2100 | * between the file and the memory map for a potential last |
| 2101 | * incomplete page. Ugly, but necessary. |
| 2102 | */ |
| 2103 | int vmtruncate(struct inode * inode, loff_t offset) |
| 2104 | { |
Christoph Hellwig | 61d5048 | 2008-02-04 22:28:56 -0800 | [diff] [blame] | 2105 | if (inode->i_size < offset) { |
| 2106 | unsigned long limit; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2107 | |
Christoph Hellwig | 61d5048 | 2008-02-04 22:28:56 -0800 | [diff] [blame] | 2108 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; |
| 2109 | if (limit != RLIM_INFINITY && offset > limit) |
| 2110 | goto out_sig; |
| 2111 | if (offset > inode->i_sb->s_maxbytes) |
| 2112 | goto out_big; |
| 2113 | i_size_write(inode, offset); |
| 2114 | } else { |
| 2115 | struct address_space *mapping = inode->i_mapping; |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2116 | |
Christoph Hellwig | 61d5048 | 2008-02-04 22:28:56 -0800 | [diff] [blame] | 2117 | /* |
| 2118 | * truncation of in-use swapfiles is disallowed - it would |
| 2119 | * cause subsequent swapout to scribble on the now-freed |
| 2120 | * blocks. |
| 2121 | */ |
| 2122 | if (IS_SWAPFILE(inode)) |
| 2123 | return -ETXTBSY; |
| 2124 | i_size_write(inode, offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2125 | |
Christoph Hellwig | 61d5048 | 2008-02-04 22:28:56 -0800 | [diff] [blame] | 2126 | /* |
| 2127 | * unmap_mapping_range is called twice, first simply for |
| 2128 | * efficiency so that truncate_inode_pages does fewer |
| 2129 | * single-page unmaps. However after this first call, and |
| 2130 | * before truncate_inode_pages finishes, it is possible for |
| 2131 | * private pages to be COWed, which remain after |
| 2132 | * truncate_inode_pages finishes, hence the second |
| 2133 | * unmap_mapping_range call must be made for correctness. |
| 2134 | */ |
| 2135 | unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); |
| 2136 | truncate_inode_pages(mapping, offset); |
| 2137 | unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); |
| 2138 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2139 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2140 | if (inode->i_op && inode->i_op->truncate) |
| 2141 | inode->i_op->truncate(inode); |
| 2142 | return 0; |
Christoph Hellwig | 61d5048 | 2008-02-04 22:28:56 -0800 | [diff] [blame] | 2143 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2144 | out_sig: |
| 2145 | send_sig(SIGXFSZ, current, 0); |
| 2146 | out_big: |
| 2147 | return -EFBIG; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2148 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2149 | EXPORT_SYMBOL(vmtruncate); |
| 2150 | |
Badari Pulavarty | f6b3ec2 | 2006-01-06 00:10:38 -0800 | [diff] [blame] | 2151 | int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end) |
| 2152 | { |
| 2153 | struct address_space *mapping = inode->i_mapping; |
| 2154 | |
| 2155 | /* |
| 2156 | * If the underlying filesystem is not going to provide |
| 2157 | * a way to truncate a range of blocks (punch a hole) - |
| 2158 | * we should return failure right now. |
| 2159 | */ |
| 2160 | if (!inode->i_op || !inode->i_op->truncate_range) |
| 2161 | return -ENOSYS; |
| 2162 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 2163 | mutex_lock(&inode->i_mutex); |
Badari Pulavarty | f6b3ec2 | 2006-01-06 00:10:38 -0800 | [diff] [blame] | 2164 | down_write(&inode->i_alloc_sem); |
| 2165 | unmap_mapping_range(mapping, offset, (end - offset), 1); |
| 2166 | truncate_inode_pages_range(mapping, offset, end); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2167 | unmap_mapping_range(mapping, offset, (end - offset), 1); |
Badari Pulavarty | f6b3ec2 | 2006-01-06 00:10:38 -0800 | [diff] [blame] | 2168 | inode->i_op->truncate_range(inode, offset, end); |
| 2169 | up_write(&inode->i_alloc_sem); |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 2170 | mutex_unlock(&inode->i_mutex); |
Badari Pulavarty | f6b3ec2 | 2006-01-06 00:10:38 -0800 | [diff] [blame] | 2171 | |
| 2172 | return 0; |
| 2173 | } |
Badari Pulavarty | f6b3ec2 | 2006-01-06 00:10:38 -0800 | [diff] [blame] | 2174 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2175 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2176 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 2177 | * but allow concurrent faults), and pte mapped but not yet locked. |
| 2178 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2179 | */ |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2180 | static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, |
| 2181 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
| 2182 | int write_access, pte_t orig_pte) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2183 | { |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2184 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2185 | struct page *page; |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2186 | swp_entry_t entry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2187 | pte_t pte; |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2188 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2189 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 2190 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2191 | goto out; |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2192 | |
| 2193 | entry = pte_to_swp_entry(orig_pte); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 2194 | if (is_migration_entry(entry)) { |
| 2195 | migration_entry_wait(mm, pmd, address); |
| 2196 | goto out; |
| 2197 | } |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 2198 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2199 | page = lookup_swap_cache(entry); |
| 2200 | if (!page) { |
Ashwin Chaugule | 098fe65 | 2006-12-06 20:31:54 -0800 | [diff] [blame] | 2201 | grab_swap_token(); /* Contend for token _before_ read-in */ |
Hugh Dickins | 02098fe | 2008-02-04 22:28:42 -0800 | [diff] [blame] | 2202 | page = swapin_readahead(entry, |
| 2203 | GFP_HIGHUSER_MOVABLE, vma, address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2204 | if (!page) { |
| 2205 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2206 | * Back out if somebody else faulted in this pte |
| 2207 | * while we released the pte lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2208 | */ |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2209 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2210 | if (likely(pte_same(*page_table, orig_pte))) |
| 2211 | ret = VM_FAULT_OOM; |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 2212 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2213 | goto unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2214 | } |
| 2215 | |
| 2216 | /* Had to read the page from swap area: Major fault */ |
| 2217 | ret = VM_FAULT_MAJOR; |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2218 | count_vm_event(PGMAJFAULT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2219 | } |
| 2220 | |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame] | 2221 | if (mem_cgroup_charge(page, mm, GFP_KERNEL)) { |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2222 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
| 2223 | ret = VM_FAULT_OOM; |
| 2224 | goto out; |
| 2225 | } |
| 2226 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2227 | mark_page_accessed(page); |
| 2228 | lock_page(page); |
Balbir Singh | 20a1022 | 2007-11-14 17:00:33 -0800 | [diff] [blame] | 2229 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2230 | |
| 2231 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2232 | * Back out if somebody else already faulted in this pte. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2233 | */ |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2234 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
Hugh Dickins | 9e9bef0 | 2005-10-29 18:16:15 -0700 | [diff] [blame] | 2235 | if (unlikely(!pte_same(*page_table, orig_pte))) |
Kirill Korotaev | b810748 | 2005-05-16 21:53:50 -0700 | [diff] [blame] | 2236 | goto out_nomap; |
Kirill Korotaev | b810748 | 2005-05-16 21:53:50 -0700 | [diff] [blame] | 2237 | |
| 2238 | if (unlikely(!PageUptodate(page))) { |
| 2239 | ret = VM_FAULT_SIGBUS; |
| 2240 | goto out_nomap; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2241 | } |
| 2242 | |
| 2243 | /* The page isn't present yet, go ahead with the fault. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2244 | |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 2245 | inc_mm_counter(mm, anon_rss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2246 | pte = mk_pte(page, vma->vm_page_prot); |
| 2247 | if (write_access && can_share_swap_page(page)) { |
| 2248 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
| 2249 | write_access = 0; |
| 2250 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2251 | |
| 2252 | flush_icache_page(vma, page); |
| 2253 | set_pte_at(mm, address, page_table, pte); |
| 2254 | page_add_anon_rmap(page, vma, address); |
| 2255 | |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 2256 | swap_free(entry); |
| 2257 | if (vm_swap_full()) |
| 2258 | remove_exclusive_swap_page(page); |
| 2259 | unlock_page(page); |
| 2260 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2261 | if (write_access) { |
Hugh Dickins | 61469f1 | 2008-03-04 14:29:04 -0800 | [diff] [blame] | 2262 | ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); |
| 2263 | if (ret & VM_FAULT_ERROR) |
| 2264 | ret &= VM_FAULT_ERROR; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2265 | goto out; |
| 2266 | } |
| 2267 | |
| 2268 | /* No need to invalidate - it was non-present before */ |
| 2269 | update_mmu_cache(vma, address, pte); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2270 | unlock: |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2271 | pte_unmap_unlock(page_table, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2272 | out: |
| 2273 | return ret; |
Kirill Korotaev | b810748 | 2005-05-16 21:53:50 -0700 | [diff] [blame] | 2274 | out_nomap: |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2275 | mem_cgroup_uncharge_page(page); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2276 | pte_unmap_unlock(page_table, ptl); |
Kirill Korotaev | b810748 | 2005-05-16 21:53:50 -0700 | [diff] [blame] | 2277 | unlock_page(page); |
| 2278 | page_cache_release(page); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2279 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2280 | } |
| 2281 | |
| 2282 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2283 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 2284 | * but allow concurrent faults), and pte mapped but not yet locked. |
| 2285 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2286 | */ |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2287 | static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, |
| 2288 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
| 2289 | int write_access) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2290 | { |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2291 | struct page *page; |
| 2292 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2293 | pte_t entry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2294 | |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2295 | /* Allocate our own private page. */ |
| 2296 | pte_unmap(page_table); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2297 | |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2298 | if (unlikely(anon_vma_prepare(vma))) |
| 2299 | goto oom; |
| 2300 | page = alloc_zeroed_user_highpage_movable(vma, address); |
| 2301 | if (!page) |
| 2302 | goto oom; |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 2303 | __SetPageUptodate(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2304 | |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame] | 2305 | if (mem_cgroup_charge(page, mm, GFP_KERNEL)) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2306 | goto oom_free_page; |
| 2307 | |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2308 | entry = mk_pte(page, vma->vm_page_prot); |
| 2309 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2310 | |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2311 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
| 2312 | if (!pte_none(*page_table)) |
| 2313 | goto release; |
| 2314 | inc_mm_counter(mm, anon_rss); |
| 2315 | lru_cache_add_active(page); |
| 2316 | page_add_new_anon_rmap(page, vma, address); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2317 | set_pte_at(mm, address, page_table, entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2318 | |
| 2319 | /* No need to invalidate - it was non-present before */ |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2320 | update_mmu_cache(vma, address, entry); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2321 | unlock: |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2322 | pte_unmap_unlock(page_table, ptl); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2323 | return 0; |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2324 | release: |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2325 | mem_cgroup_uncharge_page(page); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2326 | page_cache_release(page); |
| 2327 | goto unlock; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2328 | oom_free_page: |
Hugh Dickins | 6dbf6d3 | 2008-03-04 14:29:04 -0800 | [diff] [blame] | 2329 | page_cache_release(page); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2330 | oom: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2331 | return VM_FAULT_OOM; |
| 2332 | } |
| 2333 | |
| 2334 | /* |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2335 | * __do_fault() tries to create a new page mapping. It aggressively |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2336 | * tries to share with existing pages, but makes a separate copy if |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2337 | * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid |
| 2338 | * the next page fault. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2339 | * |
| 2340 | * As this is called only for pages that do not currently exist, we |
| 2341 | * do not need to flush old virtual caches or the TLB. |
| 2342 | * |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2343 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
Hugh Dickins | 16abfa0 | 2007-10-04 16:56:06 +0100 | [diff] [blame] | 2344 | * but allow concurrent faults), and pte neither mapped nor locked. |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2345 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2346 | */ |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2347 | 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] | 2348 | unsigned long address, pmd_t *pmd, |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2349 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2350 | { |
Hugh Dickins | 16abfa0 | 2007-10-04 16:56:06 +0100 | [diff] [blame] | 2351 | pte_t *page_table; |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2352 | spinlock_t *ptl; |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2353 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2354 | pte_t entry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2355 | int anon = 0; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2356 | struct page *dirty_page = NULL; |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2357 | struct vm_fault vmf; |
| 2358 | int ret; |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 2359 | int page_mkwrite = 0; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2360 | |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2361 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); |
| 2362 | vmf.pgoff = pgoff; |
| 2363 | vmf.flags = flags; |
| 2364 | vmf.page = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2365 | |
Nick Piggin | 3c18ddd | 2008-04-28 02:12:10 -0700 | [diff] [blame] | 2366 | ret = vma->vm_ops->fault(vma, &vmf); |
| 2367 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
| 2368 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2369 | |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2370 | /* |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2371 | * For consistency in subsequent calls, make the faulted page always |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2372 | * locked. |
| 2373 | */ |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2374 | if (unlikely(!(ret & VM_FAULT_LOCKED))) |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2375 | lock_page(vmf.page); |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2376 | else |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2377 | VM_BUG_ON(!PageLocked(vmf.page)); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2378 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2379 | /* |
| 2380 | * Should we do an early C-O-W break? |
| 2381 | */ |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2382 | page = vmf.page; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2383 | if (flags & FAULT_FLAG_WRITE) { |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2384 | if (!(vma->vm_flags & VM_SHARED)) { |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2385 | anon = 1; |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2386 | if (unlikely(anon_vma_prepare(vma))) { |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2387 | ret = VM_FAULT_OOM; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2388 | goto out; |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2389 | } |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2390 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, |
| 2391 | vma, address); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2392 | if (!page) { |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2393 | ret = VM_FAULT_OOM; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2394 | goto out; |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2395 | } |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2396 | copy_user_highpage(page, vmf.page, address, vma); |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 2397 | __SetPageUptodate(page); |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2398 | } else { |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2399 | /* |
| 2400 | * If the page will be shareable, see if the backing |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2401 | * address space wants to know that the page is about |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2402 | * to become writable |
| 2403 | */ |
Mark Fasheh | 6967614 | 2007-07-19 01:47:00 -0700 | [diff] [blame] | 2404 | if (vma->vm_ops->page_mkwrite) { |
| 2405 | unlock_page(page); |
| 2406 | if (vma->vm_ops->page_mkwrite(vma, page) < 0) { |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2407 | ret = VM_FAULT_SIGBUS; |
| 2408 | anon = 1; /* no anon but release vmf.page */ |
Mark Fasheh | 6967614 | 2007-07-19 01:47:00 -0700 | [diff] [blame] | 2409 | goto out_unlocked; |
| 2410 | } |
| 2411 | lock_page(page); |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2412 | /* |
| 2413 | * XXX: this is not quite right (racy vs |
| 2414 | * invalidate) to unlock and relock the page |
| 2415 | * like this, however a better fix requires |
| 2416 | * reworking page_mkwrite locking API, which |
| 2417 | * is better done later. |
| 2418 | */ |
| 2419 | if (!page->mapping) { |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2420 | ret = 0; |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2421 | anon = 1; /* no anon but release vmf.page */ |
| 2422 | goto out; |
| 2423 | } |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 2424 | page_mkwrite = 1; |
David Howells | 9637a5e | 2006-06-23 02:03:43 -0700 | [diff] [blame] | 2425 | } |
| 2426 | } |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2427 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2428 | } |
| 2429 | |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame] | 2430 | if (mem_cgroup_charge(page, mm, GFP_KERNEL)) { |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2431 | ret = VM_FAULT_OOM; |
| 2432 | goto out; |
| 2433 | } |
| 2434 | |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2435 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2436 | |
| 2437 | /* |
| 2438 | * This silly early PAGE_DIRTY setting removes a race |
| 2439 | * due to the bad i386 page protection. But it's valid |
| 2440 | * for other architectures too. |
| 2441 | * |
| 2442 | * Note that if write_access is true, we either now have |
| 2443 | * an exclusive copy of the page, or this is a shared mapping, |
| 2444 | * so we can make it writable and dirty to avoid having to |
| 2445 | * handle that later. |
| 2446 | */ |
| 2447 | /* Only go through if we didn't race with anybody else... */ |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2448 | if (likely(pte_same(*page_table, orig_pte))) { |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2449 | flush_icache_page(vma, page); |
| 2450 | entry = mk_pte(page, vma->vm_page_prot); |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2451 | if (flags & FAULT_FLAG_WRITE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2452 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
| 2453 | set_pte_at(mm, address, page_table, entry); |
| 2454 | if (anon) { |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2455 | inc_mm_counter(mm, anon_rss); |
| 2456 | lru_cache_add_active(page); |
| 2457 | page_add_new_anon_rmap(page, vma, address); |
Hugh Dickins | f57e88a | 2005-11-21 21:32:19 -0800 | [diff] [blame] | 2458 | } else { |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 2459 | inc_mm_counter(mm, file_rss); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2460 | page_add_file_rmap(page); |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2461 | if (flags & FAULT_FLAG_WRITE) { |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2462 | dirty_page = page; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2463 | get_page(dirty_page); |
| 2464 | } |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 2465 | } |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2466 | |
| 2467 | /* no need to invalidate: a not-present page won't be cached */ |
| 2468 | update_mmu_cache(vma, address, entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2469 | } else { |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2470 | mem_cgroup_uncharge_page(page); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2471 | if (anon) |
| 2472 | page_cache_release(page); |
| 2473 | else |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2474 | anon = 1; /* no anon but release faulted_page */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2475 | } |
| 2476 | |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2477 | pte_unmap_unlock(page_table, ptl); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2478 | |
| 2479 | out: |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2480 | unlock_page(vmf.page); |
Mark Fasheh | 6967614 | 2007-07-19 01:47:00 -0700 | [diff] [blame] | 2481 | out_unlocked: |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2482 | if (anon) |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2483 | page_cache_release(vmf.page); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2484 | else if (dirty_page) { |
Anton Salikhmetov | 8f7b3d1 | 2008-01-23 02:21:18 +0300 | [diff] [blame] | 2485 | if (vma->vm_file) |
| 2486 | file_update_time(vma->vm_file); |
| 2487 | |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 2488 | set_page_dirty_balance(dirty_page, page_mkwrite); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2489 | put_page(dirty_page); |
| 2490 | } |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2491 | |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2492 | return ret; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2493 | } |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 2494 | |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2495 | static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
| 2496 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
| 2497 | int write_access, pte_t orig_pte) |
| 2498 | { |
| 2499 | pgoff_t pgoff = (((address & PAGE_MASK) |
Dean Nelson | 0da7e01 | 2007-10-16 01:24:45 -0700 | [diff] [blame] | 2500 | - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2501 | unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0); |
| 2502 | |
Hugh Dickins | 16abfa0 | 2007-10-04 16:56:06 +0100 | [diff] [blame] | 2503 | pte_unmap(page_table); |
| 2504 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2505 | } |
| 2506 | |
Jes Sorensen | f4b8180 | 2006-09-27 01:50:10 -0700 | [diff] [blame] | 2507 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2508 | * Fault of a previously existing named mapping. Repopulate the pte |
| 2509 | * from the encoded file_pte if possible. This enables swappable |
| 2510 | * nonlinear vmas. |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2511 | * |
| 2512 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 2513 | * but allow concurrent faults), and pte mapped but not yet locked. |
| 2514 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2515 | */ |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2516 | 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] | 2517 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
| 2518 | int write_access, pte_t orig_pte) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2519 | { |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2520 | unsigned int flags = FAULT_FLAG_NONLINEAR | |
| 2521 | (write_access ? FAULT_FLAG_WRITE : 0); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2522 | pgoff_t pgoff; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2523 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 2524 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2525 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2526 | |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2527 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR) || |
| 2528 | !(vma->vm_flags & VM_CAN_NONLINEAR))) { |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2529 | /* |
| 2530 | * Page table corrupted: show pte and kill process. |
| 2531 | */ |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 2532 | print_bad_pte(vma, orig_pte, address); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2533 | return VM_FAULT_OOM; |
| 2534 | } |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2535 | |
| 2536 | pgoff = pte_to_pgoff(orig_pte); |
Hugh Dickins | 16abfa0 | 2007-10-04 16:56:06 +0100 | [diff] [blame] | 2537 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2538 | } |
| 2539 | |
| 2540 | /* |
| 2541 | * These routines also need to handle stuff like marking pages dirty |
| 2542 | * and/or accessed for architectures that don't do it in hardware (most |
| 2543 | * RISC architectures). The early dirtying is also good on the i386. |
| 2544 | * |
| 2545 | * There is also a hook called "update_mmu_cache()" that architectures |
| 2546 | * with external mmu caches can use to update those (ie the Sparc or |
| 2547 | * PowerPC hashed page tables that act as extended TLBs). |
| 2548 | * |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2549 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 2550 | * but allow concurrent faults), and pte mapped but not yet locked. |
| 2551 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2552 | */ |
| 2553 | static inline int handle_pte_fault(struct mm_struct *mm, |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2554 | struct vm_area_struct *vma, unsigned long address, |
| 2555 | pte_t *pte, pmd_t *pmd, int write_access) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2556 | { |
| 2557 | pte_t entry; |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2558 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2559 | |
Benjamin Herrenschmidt | 8dab524 | 2007-06-16 10:16:12 -0700 | [diff] [blame] | 2560 | entry = *pte; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2561 | if (!pte_present(entry)) { |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2562 | if (pte_none(entry)) { |
Jes Sorensen | f4b8180 | 2006-09-27 01:50:10 -0700 | [diff] [blame] | 2563 | if (vma->vm_ops) { |
Nick Piggin | 3c18ddd | 2008-04-28 02:12:10 -0700 | [diff] [blame] | 2564 | if (likely(vma->vm_ops->fault)) |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 2565 | return do_linear_fault(mm, vma, address, |
| 2566 | pte, pmd, write_access, entry); |
Jes Sorensen | f4b8180 | 2006-09-27 01:50:10 -0700 | [diff] [blame] | 2567 | } |
| 2568 | return do_anonymous_page(mm, vma, address, |
| 2569 | pte, pmd, write_access); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2570 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2571 | if (pte_file(entry)) |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2572 | return do_nonlinear_fault(mm, vma, address, |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2573 | pte, pmd, write_access, entry); |
| 2574 | return do_swap_page(mm, vma, address, |
| 2575 | pte, pmd, write_access, entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2576 | } |
| 2577 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 2578 | ptl = pte_lockptr(mm, pmd); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2579 | spin_lock(ptl); |
| 2580 | if (unlikely(!pte_same(*pte, entry))) |
| 2581 | goto unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2582 | if (write_access) { |
| 2583 | if (!pte_write(entry)) |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2584 | return do_wp_page(mm, vma, address, |
| 2585 | pte, pmd, ptl, entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2586 | entry = pte_mkdirty(entry); |
| 2587 | } |
| 2588 | entry = pte_mkyoung(entry); |
Benjamin Herrenschmidt | 8dab524 | 2007-06-16 10:16:12 -0700 | [diff] [blame] | 2589 | if (ptep_set_access_flags(vma, address, pte, entry, write_access)) { |
Andrea Arcangeli | 1a44e14 | 2005-10-29 18:16:48 -0700 | [diff] [blame] | 2590 | update_mmu_cache(vma, address, entry); |
Andrea Arcangeli | 1a44e14 | 2005-10-29 18:16:48 -0700 | [diff] [blame] | 2591 | } else { |
| 2592 | /* |
| 2593 | * This is needed only for protection faults but the arch code |
| 2594 | * is not yet telling us if this is a protection fault or not. |
| 2595 | * This still avoids useless tlb flushes for .text page faults |
| 2596 | * with threads. |
| 2597 | */ |
| 2598 | if (write_access) |
| 2599 | flush_tlb_page(vma, address); |
| 2600 | } |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2601 | unlock: |
| 2602 | pte_unmap_unlock(pte, ptl); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2603 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2604 | } |
| 2605 | |
| 2606 | /* |
| 2607 | * By the time we get here, we already hold the mm semaphore |
| 2608 | */ |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2609 | 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] | 2610 | unsigned long address, int write_access) |
| 2611 | { |
| 2612 | pgd_t *pgd; |
| 2613 | pud_t *pud; |
| 2614 | pmd_t *pmd; |
| 2615 | pte_t *pte; |
| 2616 | |
| 2617 | __set_current_state(TASK_RUNNING); |
| 2618 | |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2619 | count_vm_event(PGFAULT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2620 | |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2621 | if (unlikely(is_vm_hugetlb_page(vma))) |
| 2622 | return hugetlb_fault(mm, vma, address, write_access); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2623 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2624 | pgd = pgd_offset(mm, address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2625 | pud = pud_alloc(mm, pgd, address); |
| 2626 | if (!pud) |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2627 | return VM_FAULT_OOM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2628 | pmd = pmd_alloc(mm, pud, address); |
| 2629 | if (!pmd) |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2630 | return VM_FAULT_OOM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2631 | pte = pte_alloc_map(mm, pmd, address); |
| 2632 | if (!pte) |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2633 | return VM_FAULT_OOM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2634 | |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2635 | return handle_pte_fault(mm, vma, address, pte, pmd, write_access); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2636 | } |
| 2637 | |
| 2638 | #ifndef __PAGETABLE_PUD_FOLDED |
| 2639 | /* |
| 2640 | * Allocate page upper directory. |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 2641 | * We've already handled the fast-path in-line. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2642 | */ |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2643 | 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] | 2644 | { |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2645 | pud_t *new = pud_alloc_one(mm, address); |
| 2646 | if (!new) |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2647 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2648 | |
Nick Piggin | 362a61a | 2008-05-14 06:37:36 +0200 | [diff] [blame] | 2649 | smp_wmb(); /* See comment in __pte_alloc */ |
| 2650 | |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 2651 | spin_lock(&mm->page_table_lock); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2652 | if (pgd_present(*pgd)) /* Another has populated it */ |
Benjamin Herrenschmidt | 5e54197 | 2008-02-04 22:29:14 -0800 | [diff] [blame] | 2653 | pud_free(mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2654 | else |
| 2655 | pgd_populate(mm, pgd, new); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2656 | spin_unlock(&mm->page_table_lock); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2657 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2658 | } |
| 2659 | #endif /* __PAGETABLE_PUD_FOLDED */ |
| 2660 | |
| 2661 | #ifndef __PAGETABLE_PMD_FOLDED |
| 2662 | /* |
| 2663 | * Allocate page middle directory. |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 2664 | * We've already handled the fast-path in-line. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2665 | */ |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2666 | 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] | 2667 | { |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2668 | pmd_t *new = pmd_alloc_one(mm, address); |
| 2669 | if (!new) |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2670 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2671 | |
Nick Piggin | 362a61a | 2008-05-14 06:37:36 +0200 | [diff] [blame] | 2672 | smp_wmb(); /* See comment in __pte_alloc */ |
| 2673 | |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 2674 | spin_lock(&mm->page_table_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2675 | #ifndef __ARCH_HAS_4LEVEL_HACK |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2676 | if (pud_present(*pud)) /* Another has populated it */ |
Benjamin Herrenschmidt | 5e54197 | 2008-02-04 22:29:14 -0800 | [diff] [blame] | 2677 | pmd_free(mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2678 | else |
| 2679 | pud_populate(mm, pud, new); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2680 | #else |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2681 | if (pgd_present(*pud)) /* Another has populated it */ |
Benjamin Herrenschmidt | 5e54197 | 2008-02-04 22:29:14 -0800 | [diff] [blame] | 2682 | pmd_free(mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2683 | else |
| 2684 | pgd_populate(mm, pud, new); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2685 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2686 | spin_unlock(&mm->page_table_lock); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 2687 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2688 | } |
| 2689 | #endif /* __PAGETABLE_PMD_FOLDED */ |
| 2690 | |
| 2691 | int make_pages_present(unsigned long addr, unsigned long end) |
| 2692 | { |
| 2693 | int ret, len, write; |
| 2694 | struct vm_area_struct * vma; |
| 2695 | |
| 2696 | vma = find_vma(current->mm, addr); |
| 2697 | if (!vma) |
| 2698 | return -1; |
| 2699 | write = (vma->vm_flags & VM_WRITE) != 0; |
Eric Sesterhenn | 5bcb28b | 2006-03-26 18:30:52 +0200 | [diff] [blame] | 2700 | BUG_ON(addr >= end); |
| 2701 | BUG_ON(end > vma->vm_end); |
Rolf Eike Beer | 68e116a | 2007-07-15 23:38:03 -0700 | [diff] [blame] | 2702 | len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2703 | ret = get_user_pages(current, current->mm, addr, |
| 2704 | len, write, 0, NULL, NULL); |
| 2705 | if (ret < 0) |
| 2706 | return ret; |
| 2707 | return ret == len ? 0 : -1; |
| 2708 | } |
| 2709 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2710 | #if !defined(__HAVE_ARCH_GATE_AREA) |
| 2711 | |
| 2712 | #if defined(AT_SYSINFO_EHDR) |
Adrian Bunk | 5ce7852 | 2005-09-10 00:26:28 -0700 | [diff] [blame] | 2713 | static struct vm_area_struct gate_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2714 | |
| 2715 | static int __init gate_vma_init(void) |
| 2716 | { |
| 2717 | gate_vma.vm_mm = NULL; |
| 2718 | gate_vma.vm_start = FIXADDR_USER_START; |
| 2719 | gate_vma.vm_end = FIXADDR_USER_END; |
Roland McGrath | b6558c4 | 2007-01-26 00:56:47 -0800 | [diff] [blame] | 2720 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; |
| 2721 | gate_vma.vm_page_prot = __P101; |
Roland McGrath | f47aef5 | 2007-01-26 00:56:49 -0800 | [diff] [blame] | 2722 | /* |
| 2723 | * Make sure the vDSO gets into every core dump. |
| 2724 | * Dumping its contents makes post-mortem fully interpretable later |
| 2725 | * without matching up the same kernel and hardware config to see |
| 2726 | * what PC values meant. |
| 2727 | */ |
| 2728 | gate_vma.vm_flags |= VM_ALWAYSDUMP; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2729 | return 0; |
| 2730 | } |
| 2731 | __initcall(gate_vma_init); |
| 2732 | #endif |
| 2733 | |
| 2734 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) |
| 2735 | { |
| 2736 | #ifdef AT_SYSINFO_EHDR |
| 2737 | return &gate_vma; |
| 2738 | #else |
| 2739 | return NULL; |
| 2740 | #endif |
| 2741 | } |
| 2742 | |
| 2743 | int in_gate_area_no_task(unsigned long addr) |
| 2744 | { |
| 2745 | #ifdef AT_SYSINFO_EHDR |
| 2746 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) |
| 2747 | return 1; |
| 2748 | #endif |
| 2749 | return 0; |
| 2750 | } |
| 2751 | |
| 2752 | #endif /* __HAVE_ARCH_GATE_AREA */ |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 2753 | |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 2754 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
| 2755 | static resource_size_t follow_phys(struct vm_area_struct *vma, |
| 2756 | unsigned long address, unsigned int flags, |
| 2757 | unsigned long *prot) |
| 2758 | { |
| 2759 | pgd_t *pgd; |
| 2760 | pud_t *pud; |
| 2761 | pmd_t *pmd; |
| 2762 | pte_t *ptep, pte; |
| 2763 | spinlock_t *ptl; |
| 2764 | resource_size_t phys_addr = 0; |
| 2765 | struct mm_struct *mm = vma->vm_mm; |
| 2766 | |
| 2767 | VM_BUG_ON(!(vma->vm_flags & (VM_IO | VM_PFNMAP))); |
| 2768 | |
| 2769 | pgd = pgd_offset(mm, address); |
| 2770 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) |
| 2771 | goto no_page_table; |
| 2772 | |
| 2773 | pud = pud_offset(pgd, address); |
| 2774 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
| 2775 | goto no_page_table; |
| 2776 | |
| 2777 | pmd = pmd_offset(pud, address); |
| 2778 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) |
| 2779 | goto no_page_table; |
| 2780 | |
| 2781 | /* We cannot handle huge page PFN maps. Luckily they don't exist. */ |
| 2782 | if (pmd_huge(*pmd)) |
| 2783 | goto no_page_table; |
| 2784 | |
| 2785 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
| 2786 | if (!ptep) |
| 2787 | goto out; |
| 2788 | |
| 2789 | pte = *ptep; |
| 2790 | if (!pte_present(pte)) |
| 2791 | goto unlock; |
| 2792 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
| 2793 | goto unlock; |
| 2794 | phys_addr = pte_pfn(pte); |
| 2795 | phys_addr <<= PAGE_SHIFT; /* Shift here to avoid overflow on PAE */ |
| 2796 | |
| 2797 | *prot = pgprot_val(pte_pgprot(pte)); |
| 2798 | |
| 2799 | unlock: |
| 2800 | pte_unmap_unlock(ptep, ptl); |
| 2801 | out: |
| 2802 | return phys_addr; |
| 2803 | no_page_table: |
| 2804 | return 0; |
| 2805 | } |
| 2806 | |
| 2807 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, |
| 2808 | void *buf, int len, int write) |
| 2809 | { |
| 2810 | resource_size_t phys_addr; |
| 2811 | unsigned long prot = 0; |
| 2812 | void *maddr; |
| 2813 | int offset = addr & (PAGE_SIZE-1); |
| 2814 | |
| 2815 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
| 2816 | return -EINVAL; |
| 2817 | |
| 2818 | phys_addr = follow_phys(vma, addr, write, &prot); |
| 2819 | |
| 2820 | if (!phys_addr) |
| 2821 | return -EINVAL; |
| 2822 | |
| 2823 | maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot); |
| 2824 | if (write) |
| 2825 | memcpy_toio(maddr + offset, buf, len); |
| 2826 | else |
| 2827 | memcpy_fromio(buf, maddr + offset, len); |
| 2828 | iounmap(maddr); |
| 2829 | |
| 2830 | return len; |
| 2831 | } |
| 2832 | #endif |
| 2833 | |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 2834 | /* |
| 2835 | * Access another process' address space. |
| 2836 | * Source/target buffer must be kernel space, |
| 2837 | * Do not walk the page table directly, use get_user_pages |
| 2838 | */ |
| 2839 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) |
| 2840 | { |
| 2841 | struct mm_struct *mm; |
| 2842 | struct vm_area_struct *vma; |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 2843 | void *old_buf = buf; |
| 2844 | |
| 2845 | mm = get_task_mm(tsk); |
| 2846 | if (!mm) |
| 2847 | return 0; |
| 2848 | |
| 2849 | down_read(&mm->mmap_sem); |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 2850 | /* ignore errors, just check how much was successfully transferred */ |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 2851 | while (len) { |
| 2852 | int bytes, ret, offset; |
| 2853 | void *maddr; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 2854 | struct page *page = NULL; |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 2855 | |
| 2856 | ret = get_user_pages(tsk, mm, addr, 1, |
| 2857 | write, 1, &page, &vma); |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 2858 | if (ret <= 0) { |
| 2859 | /* |
| 2860 | * Check if this is a VM_IO | VM_PFNMAP VMA, which |
| 2861 | * we can access using slightly different code. |
| 2862 | */ |
| 2863 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
| 2864 | vma = find_vma(mm, addr); |
| 2865 | if (!vma) |
| 2866 | break; |
| 2867 | if (vma->vm_ops && vma->vm_ops->access) |
| 2868 | ret = vma->vm_ops->access(vma, addr, buf, |
| 2869 | len, write); |
| 2870 | if (ret <= 0) |
| 2871 | #endif |
| 2872 | break; |
| 2873 | bytes = ret; |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 2874 | } else { |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 2875 | bytes = len; |
| 2876 | offset = addr & (PAGE_SIZE-1); |
| 2877 | if (bytes > PAGE_SIZE-offset) |
| 2878 | bytes = PAGE_SIZE-offset; |
| 2879 | |
| 2880 | maddr = kmap(page); |
| 2881 | if (write) { |
| 2882 | copy_to_user_page(vma, page, addr, |
| 2883 | maddr + offset, buf, bytes); |
| 2884 | set_page_dirty_lock(page); |
| 2885 | } else { |
| 2886 | copy_from_user_page(vma, page, addr, |
| 2887 | buf, maddr + offset, bytes); |
| 2888 | } |
| 2889 | kunmap(page); |
| 2890 | page_cache_release(page); |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 2891 | } |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 2892 | len -= bytes; |
| 2893 | buf += bytes; |
| 2894 | addr += bytes; |
| 2895 | } |
| 2896 | up_read(&mm->mmap_sem); |
| 2897 | mmput(mm); |
| 2898 | |
| 2899 | return buf - old_buf; |
| 2900 | } |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 2901 | |
| 2902 | /* |
| 2903 | * Print the name of a VMA. |
| 2904 | */ |
| 2905 | void print_vma_addr(char *prefix, unsigned long ip) |
| 2906 | { |
| 2907 | struct mm_struct *mm = current->mm; |
| 2908 | struct vm_area_struct *vma; |
| 2909 | |
Ingo Molnar | e8bff74 | 2008-02-13 20:21:06 +0100 | [diff] [blame] | 2910 | /* |
| 2911 | * Do not print if we are in atomic |
| 2912 | * contexts (in exception stacks, etc.): |
| 2913 | */ |
| 2914 | if (preempt_count()) |
| 2915 | return; |
| 2916 | |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 2917 | down_read(&mm->mmap_sem); |
| 2918 | vma = find_vma(mm, ip); |
| 2919 | if (vma && vma->vm_file) { |
| 2920 | struct file *f = vma->vm_file; |
| 2921 | char *buf = (char *)__get_free_page(GFP_KERNEL); |
| 2922 | if (buf) { |
| 2923 | char *p, *s; |
| 2924 | |
Jan Blunck | cf28b48 | 2008-02-14 19:38:44 -0800 | [diff] [blame] | 2925 | p = d_path(&f->f_path, buf, PAGE_SIZE); |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 2926 | if (IS_ERR(p)) |
| 2927 | p = "?"; |
| 2928 | s = strrchr(p, '/'); |
| 2929 | if (s) |
| 2930 | p = s+1; |
| 2931 | printk("%s%s[%lx+%lx]", prefix, p, |
| 2932 | vma->vm_start, |
| 2933 | vma->vm_end - vma->vm_start); |
| 2934 | free_page((unsigned long)buf); |
| 2935 | } |
| 2936 | } |
| 2937 | up_read(¤t->mm->mmap_sem); |
| 2938 | } |