Commit | Line | Data |
---|---|---|
1da177e4 LT |
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> | |
9a840895 | 48 | #include <linux/ksm.h> |
1da177e4 LT |
49 | #include <linux/rmap.h> |
50 | #include <linux/module.h> | |
0ff92245 | 51 | #include <linux/delayacct.h> |
1da177e4 | 52 | #include <linux/init.h> |
edc79b2a | 53 | #include <linux/writeback.h> |
8a9f3ccd | 54 | #include <linux/memcontrol.h> |
cddb8a5c | 55 | #include <linux/mmu_notifier.h> |
3dc14741 HD |
56 | #include <linux/kallsyms.h> |
57 | #include <linux/swapops.h> | |
58 | #include <linux/elf.h> | |
5a0e3ad6 | 59 | #include <linux/gfp.h> |
1da177e4 | 60 | |
6952b61d | 61 | #include <asm/io.h> |
1da177e4 LT |
62 | #include <asm/pgalloc.h> |
63 | #include <asm/uaccess.h> | |
64 | #include <asm/tlb.h> | |
65 | #include <asm/tlbflush.h> | |
66 | #include <asm/pgtable.h> | |
67 | ||
42b77728 JB |
68 | #include "internal.h" |
69 | ||
d41dee36 | 70 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
71 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
72 | unsigned long max_mapnr; | |
73 | struct page *mem_map; | |
74 | ||
75 | EXPORT_SYMBOL(max_mapnr); | |
76 | EXPORT_SYMBOL(mem_map); | |
77 | #endif | |
78 | ||
79 | unsigned long num_physpages; | |
80 | /* | |
81 | * A number of key systems in x86 including ioremap() rely on the assumption | |
82 | * that high_memory defines the upper bound on direct map memory, then end | |
83 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
84 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
85 | * and ZONE_HIGHMEM. | |
86 | */ | |
87 | void * high_memory; | |
1da177e4 LT |
88 | |
89 | EXPORT_SYMBOL(num_physpages); | |
90 | EXPORT_SYMBOL(high_memory); | |
1da177e4 | 91 | |
32a93233 IM |
92 | /* |
93 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
94 | * | |
95 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
96 | * as ancient (libc5 based) binaries can segfault. ) | |
97 | */ | |
98 | int randomize_va_space __read_mostly = | |
99 | #ifdef CONFIG_COMPAT_BRK | |
100 | 1; | |
101 | #else | |
102 | 2; | |
103 | #endif | |
a62eaf15 AK |
104 | |
105 | static int __init disable_randmaps(char *s) | |
106 | { | |
107 | randomize_va_space = 0; | |
9b41046c | 108 | return 1; |
a62eaf15 AK |
109 | } |
110 | __setup("norandmaps", disable_randmaps); | |
111 | ||
62eede62 | 112 | unsigned long zero_pfn __read_mostly; |
03f6462a | 113 | unsigned long highest_memmap_pfn __read_mostly; |
a13ea5b7 HD |
114 | |
115 | /* | |
116 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
117 | */ | |
118 | static int __init init_zero_pfn(void) | |
119 | { | |
120 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
121 | return 0; | |
122 | } | |
123 | core_initcall(init_zero_pfn); | |
a62eaf15 | 124 | |
d559db08 | 125 | |
34e55232 KH |
126 | #if defined(SPLIT_RSS_COUNTING) |
127 | ||
a3a2e76c | 128 | static void __sync_task_rss_stat(struct task_struct *task, struct mm_struct *mm) |
34e55232 KH |
129 | { |
130 | int i; | |
131 | ||
132 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
133 | if (task->rss_stat.count[i]) { | |
134 | add_mm_counter(mm, i, task->rss_stat.count[i]); | |
135 | task->rss_stat.count[i] = 0; | |
136 | } | |
137 | } | |
138 | task->rss_stat.events = 0; | |
139 | } | |
140 | ||
141 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
142 | { | |
143 | struct task_struct *task = current; | |
144 | ||
145 | if (likely(task->mm == mm)) | |
146 | task->rss_stat.count[member] += val; | |
147 | else | |
148 | add_mm_counter(mm, member, val); | |
149 | } | |
150 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
151 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
152 | ||
153 | /* sync counter once per 64 page faults */ | |
154 | #define TASK_RSS_EVENTS_THRESH (64) | |
155 | static void check_sync_rss_stat(struct task_struct *task) | |
156 | { | |
157 | if (unlikely(task != current)) | |
158 | return; | |
159 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
160 | __sync_task_rss_stat(task, task->mm); | |
161 | } | |
162 | ||
163 | unsigned long get_mm_counter(struct mm_struct *mm, int member) | |
164 | { | |
165 | long val = 0; | |
166 | ||
167 | /* | |
168 | * Don't use task->mm here...for avoiding to use task_get_mm().. | |
169 | * The caller must guarantee task->mm is not invalid. | |
170 | */ | |
171 | val = atomic_long_read(&mm->rss_stat.count[member]); | |
172 | /* | |
173 | * counter is updated in asynchronous manner and may go to minus. | |
174 | * But it's never be expected number for users. | |
175 | */ | |
176 | if (val < 0) | |
177 | return 0; | |
178 | return (unsigned long)val; | |
179 | } | |
180 | ||
181 | void sync_mm_rss(struct task_struct *task, struct mm_struct *mm) | |
182 | { | |
183 | __sync_task_rss_stat(task, mm); | |
184 | } | |
185 | #else | |
186 | ||
187 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
188 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
189 | ||
190 | static void check_sync_rss_stat(struct task_struct *task) | |
191 | { | |
192 | } | |
193 | ||
34e55232 KH |
194 | #endif |
195 | ||
1da177e4 LT |
196 | /* |
197 | * If a p?d_bad entry is found while walking page tables, report | |
198 | * the error, before resetting entry to p?d_none. Usually (but | |
199 | * very seldom) called out from the p?d_none_or_clear_bad macros. | |
200 | */ | |
201 | ||
202 | void pgd_clear_bad(pgd_t *pgd) | |
203 | { | |
204 | pgd_ERROR(*pgd); | |
205 | pgd_clear(pgd); | |
206 | } | |
207 | ||
208 | void pud_clear_bad(pud_t *pud) | |
209 | { | |
210 | pud_ERROR(*pud); | |
211 | pud_clear(pud); | |
212 | } | |
213 | ||
214 | void pmd_clear_bad(pmd_t *pmd) | |
215 | { | |
216 | pmd_ERROR(*pmd); | |
217 | pmd_clear(pmd); | |
218 | } | |
219 | ||
220 | /* | |
221 | * Note: this doesn't free the actual pages themselves. That | |
222 | * has been handled earlier when unmapping all the memory regions. | |
223 | */ | |
9e1b32ca BH |
224 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
225 | unsigned long addr) | |
1da177e4 | 226 | { |
2f569afd | 227 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 228 | pmd_clear(pmd); |
9e1b32ca | 229 | pte_free_tlb(tlb, token, addr); |
e0da382c | 230 | tlb->mm->nr_ptes--; |
1da177e4 LT |
231 | } |
232 | ||
e0da382c HD |
233 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
234 | unsigned long addr, unsigned long end, | |
235 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
236 | { |
237 | pmd_t *pmd; | |
238 | unsigned long next; | |
e0da382c | 239 | unsigned long start; |
1da177e4 | 240 | |
e0da382c | 241 | start = addr; |
1da177e4 | 242 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
243 | do { |
244 | next = pmd_addr_end(addr, end); | |
245 | if (pmd_none_or_clear_bad(pmd)) | |
246 | continue; | |
9e1b32ca | 247 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
248 | } while (pmd++, addr = next, addr != end); |
249 | ||
e0da382c HD |
250 | start &= PUD_MASK; |
251 | if (start < floor) | |
252 | return; | |
253 | if (ceiling) { | |
254 | ceiling &= PUD_MASK; | |
255 | if (!ceiling) | |
256 | return; | |
1da177e4 | 257 | } |
e0da382c HD |
258 | if (end - 1 > ceiling - 1) |
259 | return; | |
260 | ||
261 | pmd = pmd_offset(pud, start); | |
262 | pud_clear(pud); | |
9e1b32ca | 263 | pmd_free_tlb(tlb, pmd, start); |
1da177e4 LT |
264 | } |
265 | ||
e0da382c HD |
266 | static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, |
267 | unsigned long addr, unsigned long end, | |
268 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
269 | { |
270 | pud_t *pud; | |
271 | unsigned long next; | |
e0da382c | 272 | unsigned long start; |
1da177e4 | 273 | |
e0da382c | 274 | start = addr; |
1da177e4 | 275 | pud = pud_offset(pgd, addr); |
1da177e4 LT |
276 | do { |
277 | next = pud_addr_end(addr, end); | |
278 | if (pud_none_or_clear_bad(pud)) | |
279 | continue; | |
e0da382c | 280 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
281 | } while (pud++, addr = next, addr != end); |
282 | ||
e0da382c HD |
283 | start &= PGDIR_MASK; |
284 | if (start < floor) | |
285 | return; | |
286 | if (ceiling) { | |
287 | ceiling &= PGDIR_MASK; | |
288 | if (!ceiling) | |
289 | return; | |
1da177e4 | 290 | } |
e0da382c HD |
291 | if (end - 1 > ceiling - 1) |
292 | return; | |
293 | ||
294 | pud = pud_offset(pgd, start); | |
295 | pgd_clear(pgd); | |
9e1b32ca | 296 | pud_free_tlb(tlb, pud, start); |
1da177e4 LT |
297 | } |
298 | ||
299 | /* | |
e0da382c HD |
300 | * This function frees user-level page tables of a process. |
301 | * | |
1da177e4 LT |
302 | * Must be called with pagetable lock held. |
303 | */ | |
42b77728 | 304 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
305 | unsigned long addr, unsigned long end, |
306 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
307 | { |
308 | pgd_t *pgd; | |
309 | unsigned long next; | |
e0da382c HD |
310 | unsigned long start; |
311 | ||
312 | /* | |
313 | * The next few lines have given us lots of grief... | |
314 | * | |
315 | * Why are we testing PMD* at this top level? Because often | |
316 | * there will be no work to do at all, and we'd prefer not to | |
317 | * go all the way down to the bottom just to discover that. | |
318 | * | |
319 | * Why all these "- 1"s? Because 0 represents both the bottom | |
320 | * of the address space and the top of it (using -1 for the | |
321 | * top wouldn't help much: the masks would do the wrong thing). | |
322 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
323 | * the address space, but end 0 and ceiling 0 refer to the top | |
324 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
325 | * that end 0 case should be mythical). | |
326 | * | |
327 | * Wherever addr is brought up or ceiling brought down, we must | |
328 | * be careful to reject "the opposite 0" before it confuses the | |
329 | * subsequent tests. But what about where end is brought down | |
330 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
331 | * | |
332 | * Whereas we round start (addr) and ceiling down, by different | |
333 | * masks at different levels, in order to test whether a table | |
334 | * now has no other vmas using it, so can be freed, we don't | |
335 | * bother to round floor or end up - the tests don't need that. | |
336 | */ | |
1da177e4 | 337 | |
e0da382c HD |
338 | addr &= PMD_MASK; |
339 | if (addr < floor) { | |
340 | addr += PMD_SIZE; | |
341 | if (!addr) | |
342 | return; | |
343 | } | |
344 | if (ceiling) { | |
345 | ceiling &= PMD_MASK; | |
346 | if (!ceiling) | |
347 | return; | |
348 | } | |
349 | if (end - 1 > ceiling - 1) | |
350 | end -= PMD_SIZE; | |
351 | if (addr > end - 1) | |
352 | return; | |
353 | ||
354 | start = addr; | |
42b77728 | 355 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
356 | do { |
357 | next = pgd_addr_end(addr, end); | |
358 | if (pgd_none_or_clear_bad(pgd)) | |
359 | continue; | |
42b77728 | 360 | free_pud_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 361 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
362 | } |
363 | ||
42b77728 | 364 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 365 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
366 | { |
367 | while (vma) { | |
368 | struct vm_area_struct *next = vma->vm_next; | |
369 | unsigned long addr = vma->vm_start; | |
370 | ||
8f4f8c16 | 371 | /* |
25d9e2d1 | 372 | * Hide vma from rmap and truncate_pagecache before freeing |
373 | * pgtables | |
8f4f8c16 | 374 | */ |
5beb4930 | 375 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
376 | unlink_file_vma(vma); |
377 | ||
9da61aef | 378 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 379 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
e0da382c | 380 | floor, next? next->vm_start: ceiling); |
3bf5ee95 HD |
381 | } else { |
382 | /* | |
383 | * Optimization: gather nearby vmas into one call down | |
384 | */ | |
385 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 386 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
387 | vma = next; |
388 | next = vma->vm_next; | |
5beb4930 | 389 | unlink_anon_vmas(vma); |
8f4f8c16 | 390 | unlink_file_vma(vma); |
3bf5ee95 HD |
391 | } |
392 | free_pgd_range(tlb, addr, vma->vm_end, | |
393 | floor, next? next->vm_start: ceiling); | |
394 | } | |
e0da382c HD |
395 | vma = next; |
396 | } | |
1da177e4 LT |
397 | } |
398 | ||
1bb3630e | 399 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
1da177e4 | 400 | { |
2f569afd | 401 | pgtable_t new = pte_alloc_one(mm, address); |
1bb3630e HD |
402 | if (!new) |
403 | return -ENOMEM; | |
404 | ||
362a61ad NP |
405 | /* |
406 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
407 | * visible before the pte is made visible to other CPUs by being | |
408 | * put into page tables. | |
409 | * | |
410 | * The other side of the story is the pointer chasing in the page | |
411 | * table walking code (when walking the page table without locking; | |
412 | * ie. most of the time). Fortunately, these data accesses consist | |
413 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
414 | * being the notable exception) will already guarantee loads are | |
415 | * seen in-order. See the alpha page table accessors for the | |
416 | * smp_read_barrier_depends() barriers in page table walking code. | |
417 | */ | |
418 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
419 | ||
c74df32c | 420 | spin_lock(&mm->page_table_lock); |
2f569afd | 421 | if (!pmd_present(*pmd)) { /* Has another populated it ? */ |
1da177e4 | 422 | mm->nr_ptes++; |
1da177e4 | 423 | pmd_populate(mm, pmd, new); |
2f569afd | 424 | new = NULL; |
1da177e4 | 425 | } |
c74df32c | 426 | spin_unlock(&mm->page_table_lock); |
2f569afd MS |
427 | if (new) |
428 | pte_free(mm, new); | |
1bb3630e | 429 | return 0; |
1da177e4 LT |
430 | } |
431 | ||
1bb3630e | 432 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 433 | { |
1bb3630e HD |
434 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
435 | if (!new) | |
436 | return -ENOMEM; | |
437 | ||
362a61ad NP |
438 | smp_wmb(); /* See comment in __pte_alloc */ |
439 | ||
1bb3630e | 440 | spin_lock(&init_mm.page_table_lock); |
2f569afd | 441 | if (!pmd_present(*pmd)) { /* Has another populated it ? */ |
1bb3630e | 442 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd MS |
443 | new = NULL; |
444 | } | |
1bb3630e | 445 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
446 | if (new) |
447 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 448 | return 0; |
1da177e4 LT |
449 | } |
450 | ||
d559db08 KH |
451 | static inline void init_rss_vec(int *rss) |
452 | { | |
453 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
454 | } | |
455 | ||
456 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 457 | { |
d559db08 KH |
458 | int i; |
459 | ||
34e55232 KH |
460 | if (current->mm == mm) |
461 | sync_mm_rss(current, mm); | |
d559db08 KH |
462 | for (i = 0; i < NR_MM_COUNTERS; i++) |
463 | if (rss[i]) | |
464 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
465 | } |
466 | ||
b5810039 | 467 | /* |
6aab341e LT |
468 | * This function is called to print an error when a bad pte |
469 | * is found. For example, we might have a PFN-mapped pte in | |
470 | * a region that doesn't allow it. | |
b5810039 NP |
471 | * |
472 | * The calling function must still handle the error. | |
473 | */ | |
3dc14741 HD |
474 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
475 | pte_t pte, struct page *page) | |
b5810039 | 476 | { |
3dc14741 HD |
477 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
478 | pud_t *pud = pud_offset(pgd, addr); | |
479 | pmd_t *pmd = pmd_offset(pud, addr); | |
480 | struct address_space *mapping; | |
481 | pgoff_t index; | |
d936cf9b HD |
482 | static unsigned long resume; |
483 | static unsigned long nr_shown; | |
484 | static unsigned long nr_unshown; | |
485 | ||
486 | /* | |
487 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
488 | * or allow a steady drip of one report per second. | |
489 | */ | |
490 | if (nr_shown == 60) { | |
491 | if (time_before(jiffies, resume)) { | |
492 | nr_unshown++; | |
493 | return; | |
494 | } | |
495 | if (nr_unshown) { | |
1e9e6365 HD |
496 | printk(KERN_ALERT |
497 | "BUG: Bad page map: %lu messages suppressed\n", | |
d936cf9b HD |
498 | nr_unshown); |
499 | nr_unshown = 0; | |
500 | } | |
501 | nr_shown = 0; | |
502 | } | |
503 | if (nr_shown++ == 0) | |
504 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
505 | |
506 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
507 | index = linear_page_index(vma, addr); | |
508 | ||
1e9e6365 HD |
509 | printk(KERN_ALERT |
510 | "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", | |
3dc14741 HD |
511 | current->comm, |
512 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 WF |
513 | if (page) |
514 | dump_page(page); | |
1e9e6365 | 515 | printk(KERN_ALERT |
3dc14741 HD |
516 | "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
517 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
518 | /* | |
519 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y | |
520 | */ | |
521 | if (vma->vm_ops) | |
1e9e6365 | 522 | print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n", |
3dc14741 HD |
523 | (unsigned long)vma->vm_ops->fault); |
524 | if (vma->vm_file && vma->vm_file->f_op) | |
1e9e6365 | 525 | print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n", |
3dc14741 | 526 | (unsigned long)vma->vm_file->f_op->mmap); |
b5810039 | 527 | dump_stack(); |
3dc14741 | 528 | add_taint(TAINT_BAD_PAGE); |
b5810039 NP |
529 | } |
530 | ||
67121172 LT |
531 | static inline int is_cow_mapping(unsigned int flags) |
532 | { | |
533 | return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; | |
534 | } | |
535 | ||
62eede62 HD |
536 | #ifndef is_zero_pfn |
537 | static inline int is_zero_pfn(unsigned long pfn) | |
538 | { | |
539 | return pfn == zero_pfn; | |
540 | } | |
541 | #endif | |
542 | ||
543 | #ifndef my_zero_pfn | |
544 | static inline unsigned long my_zero_pfn(unsigned long addr) | |
545 | { | |
546 | return zero_pfn; | |
547 | } | |
548 | #endif | |
549 | ||
ee498ed7 | 550 | /* |
7e675137 | 551 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 552 | * |
7e675137 NP |
553 | * "Special" mappings do not wish to be associated with a "struct page" (either |
554 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
555 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 556 | * |
7e675137 NP |
557 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
558 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
559 | * may not have a spare pte bit, which requires a more complicated scheme, | |
560 | * described below. | |
561 | * | |
562 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
563 | * special mapping (even if there are underlying and valid "struct pages"). | |
564 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 565 | * |
b379d790 JH |
566 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
567 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
568 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
569 | * mapping will always honor the rule | |
6aab341e LT |
570 | * |
571 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
572 | * | |
7e675137 NP |
573 | * And for normal mappings this is false. |
574 | * | |
575 | * This restricts such mappings to be a linear translation from virtual address | |
576 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
577 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
578 | * special (because none can have been COWed). | |
b379d790 | 579 | * |
b379d790 | 580 | * |
7e675137 | 581 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
582 | * |
583 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
584 | * page" backing, however the difference is that _all_ pages with a struct | |
585 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
586 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
587 | * (which can be slower and simply not an option for some PFNMAP users). The | |
588 | * advantage is that we don't have to follow the strict linearity rule of | |
589 | * PFNMAP mappings in order to support COWable mappings. | |
590 | * | |
ee498ed7 | 591 | */ |
7e675137 NP |
592 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
593 | # define HAVE_PTE_SPECIAL 1 | |
594 | #else | |
595 | # define HAVE_PTE_SPECIAL 0 | |
596 | #endif | |
597 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, | |
598 | pte_t pte) | |
ee498ed7 | 599 | { |
22b31eec | 600 | unsigned long pfn = pte_pfn(pte); |
7e675137 NP |
601 | |
602 | if (HAVE_PTE_SPECIAL) { | |
22b31eec HD |
603 | if (likely(!pte_special(pte))) |
604 | goto check_pfn; | |
a13ea5b7 HD |
605 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
606 | return NULL; | |
62eede62 | 607 | if (!is_zero_pfn(pfn)) |
22b31eec | 608 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
609 | return NULL; |
610 | } | |
611 | ||
612 | /* !HAVE_PTE_SPECIAL case follows: */ | |
613 | ||
b379d790 JH |
614 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
615 | if (vma->vm_flags & VM_MIXEDMAP) { | |
616 | if (!pfn_valid(pfn)) | |
617 | return NULL; | |
618 | goto out; | |
619 | } else { | |
7e675137 NP |
620 | unsigned long off; |
621 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
622 | if (pfn == vma->vm_pgoff + off) |
623 | return NULL; | |
624 | if (!is_cow_mapping(vma->vm_flags)) | |
625 | return NULL; | |
626 | } | |
6aab341e LT |
627 | } |
628 | ||
62eede62 HD |
629 | if (is_zero_pfn(pfn)) |
630 | return NULL; | |
22b31eec HD |
631 | check_pfn: |
632 | if (unlikely(pfn > highest_memmap_pfn)) { | |
633 | print_bad_pte(vma, addr, pte, NULL); | |
634 | return NULL; | |
635 | } | |
6aab341e LT |
636 | |
637 | /* | |
7e675137 | 638 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 639 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 640 | */ |
b379d790 | 641 | out: |
6aab341e | 642 | return pfn_to_page(pfn); |
ee498ed7 HD |
643 | } |
644 | ||
1da177e4 LT |
645 | /* |
646 | * copy one vm_area from one task to the other. Assumes the page tables | |
647 | * already present in the new task to be cleared in the whole range | |
648 | * covered by this vma. | |
1da177e4 LT |
649 | */ |
650 | ||
570a335b | 651 | static inline unsigned long |
1da177e4 | 652 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 653 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 654 | unsigned long addr, int *rss) |
1da177e4 | 655 | { |
b5810039 | 656 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
657 | pte_t pte = *src_pte; |
658 | struct page *page; | |
1da177e4 LT |
659 | |
660 | /* pte contains position in swap or file, so copy. */ | |
661 | if (unlikely(!pte_present(pte))) { | |
662 | if (!pte_file(pte)) { | |
0697212a CL |
663 | swp_entry_t entry = pte_to_swp_entry(pte); |
664 | ||
570a335b HD |
665 | if (swap_duplicate(entry) < 0) |
666 | return entry.val; | |
667 | ||
1da177e4 LT |
668 | /* make sure dst_mm is on swapoff's mmlist. */ |
669 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
670 | spin_lock(&mmlist_lock); | |
f412ac08 HD |
671 | if (list_empty(&dst_mm->mmlist)) |
672 | list_add(&dst_mm->mmlist, | |
673 | &src_mm->mmlist); | |
1da177e4 LT |
674 | spin_unlock(&mmlist_lock); |
675 | } | |
b084d435 KH |
676 | if (likely(!non_swap_entry(entry))) |
677 | rss[MM_SWAPENTS]++; | |
678 | else if (is_write_migration_entry(entry) && | |
0697212a CL |
679 | is_cow_mapping(vm_flags)) { |
680 | /* | |
681 | * COW mappings require pages in both parent | |
682 | * and child to be set to read. | |
683 | */ | |
684 | make_migration_entry_read(&entry); | |
685 | pte = swp_entry_to_pte(entry); | |
686 | set_pte_at(src_mm, addr, src_pte, pte); | |
687 | } | |
1da177e4 | 688 | } |
ae859762 | 689 | goto out_set_pte; |
1da177e4 LT |
690 | } |
691 | ||
1da177e4 LT |
692 | /* |
693 | * If it's a COW mapping, write protect it both | |
694 | * in the parent and the child | |
695 | */ | |
67121172 | 696 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 697 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 698 | pte = pte_wrprotect(pte); |
1da177e4 LT |
699 | } |
700 | ||
701 | /* | |
702 | * If it's a shared mapping, mark it clean in | |
703 | * the child | |
704 | */ | |
705 | if (vm_flags & VM_SHARED) | |
706 | pte = pte_mkclean(pte); | |
707 | pte = pte_mkold(pte); | |
6aab341e LT |
708 | |
709 | page = vm_normal_page(vma, addr, pte); | |
710 | if (page) { | |
711 | get_page(page); | |
21333b2b | 712 | page_dup_rmap(page); |
d559db08 KH |
713 | if (PageAnon(page)) |
714 | rss[MM_ANONPAGES]++; | |
715 | else | |
716 | rss[MM_FILEPAGES]++; | |
6aab341e | 717 | } |
ae859762 HD |
718 | |
719 | out_set_pte: | |
720 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 721 | return 0; |
1da177e4 LT |
722 | } |
723 | ||
724 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
725 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, | |
726 | unsigned long addr, unsigned long end) | |
727 | { | |
c36987e2 | 728 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 729 | pte_t *src_pte, *dst_pte; |
c74df32c | 730 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 731 | int progress = 0; |
d559db08 | 732 | int rss[NR_MM_COUNTERS]; |
570a335b | 733 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
734 | |
735 | again: | |
d559db08 KH |
736 | init_rss_vec(rss); |
737 | ||
c74df32c | 738 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
739 | if (!dst_pte) |
740 | return -ENOMEM; | |
741 | src_pte = pte_offset_map_nested(src_pmd, addr); | |
4c21e2f2 | 742 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 743 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
744 | orig_src_pte = src_pte; |
745 | orig_dst_pte = dst_pte; | |
6606c3e0 | 746 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 747 | |
1da177e4 LT |
748 | do { |
749 | /* | |
750 | * We are holding two locks at this point - either of them | |
751 | * could generate latencies in another task on another CPU. | |
752 | */ | |
e040f218 HD |
753 | if (progress >= 32) { |
754 | progress = 0; | |
755 | if (need_resched() || | |
95c354fe | 756 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
757 | break; |
758 | } | |
1da177e4 LT |
759 | if (pte_none(*src_pte)) { |
760 | progress++; | |
761 | continue; | |
762 | } | |
570a335b HD |
763 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
764 | vma, addr, rss); | |
765 | if (entry.val) | |
766 | break; | |
1da177e4 LT |
767 | progress += 8; |
768 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 769 | |
6606c3e0 | 770 | arch_leave_lazy_mmu_mode(); |
c74df32c | 771 | spin_unlock(src_ptl); |
c36987e2 | 772 | pte_unmap_nested(orig_src_pte); |
d559db08 | 773 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 774 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 775 | cond_resched(); |
570a335b HD |
776 | |
777 | if (entry.val) { | |
778 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
779 | return -ENOMEM; | |
780 | progress = 0; | |
781 | } | |
1da177e4 LT |
782 | if (addr != end) |
783 | goto again; | |
784 | return 0; | |
785 | } | |
786 | ||
787 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
788 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
789 | unsigned long addr, unsigned long end) | |
790 | { | |
791 | pmd_t *src_pmd, *dst_pmd; | |
792 | unsigned long next; | |
793 | ||
794 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
795 | if (!dst_pmd) | |
796 | return -ENOMEM; | |
797 | src_pmd = pmd_offset(src_pud, addr); | |
798 | do { | |
799 | next = pmd_addr_end(addr, end); | |
800 | if (pmd_none_or_clear_bad(src_pmd)) | |
801 | continue; | |
802 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
803 | vma, addr, next)) | |
804 | return -ENOMEM; | |
805 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
806 | return 0; | |
807 | } | |
808 | ||
809 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
810 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
811 | unsigned long addr, unsigned long end) | |
812 | { | |
813 | pud_t *src_pud, *dst_pud; | |
814 | unsigned long next; | |
815 | ||
816 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); | |
817 | if (!dst_pud) | |
818 | return -ENOMEM; | |
819 | src_pud = pud_offset(src_pgd, addr); | |
820 | do { | |
821 | next = pud_addr_end(addr, end); | |
822 | if (pud_none_or_clear_bad(src_pud)) | |
823 | continue; | |
824 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
825 | vma, addr, next)) | |
826 | return -ENOMEM; | |
827 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
828 | return 0; | |
829 | } | |
830 | ||
831 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
832 | struct vm_area_struct *vma) | |
833 | { | |
834 | pgd_t *src_pgd, *dst_pgd; | |
835 | unsigned long next; | |
836 | unsigned long addr = vma->vm_start; | |
837 | unsigned long end = vma->vm_end; | |
cddb8a5c | 838 | int ret; |
1da177e4 | 839 | |
d992895b NP |
840 | /* |
841 | * Don't copy ptes where a page fault will fill them correctly. | |
842 | * Fork becomes much lighter when there are big shared or private | |
843 | * readonly mappings. The tradeoff is that copy_page_range is more | |
844 | * efficient than faulting. | |
845 | */ | |
4d7672b4 | 846 | if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) { |
d992895b NP |
847 | if (!vma->anon_vma) |
848 | return 0; | |
849 | } | |
850 | ||
1da177e4 LT |
851 | if (is_vm_hugetlb_page(vma)) |
852 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
853 | ||
34801ba9 | 854 | if (unlikely(is_pfn_mapping(vma))) { |
2ab64037 | 855 | /* |
856 | * We do not free on error cases below as remove_vma | |
857 | * gets called on error from higher level routine | |
858 | */ | |
859 | ret = track_pfn_vma_copy(vma); | |
860 | if (ret) | |
861 | return ret; | |
862 | } | |
863 | ||
cddb8a5c AA |
864 | /* |
865 | * We need to invalidate the secondary MMU mappings only when | |
866 | * there could be a permission downgrade on the ptes of the | |
867 | * parent mm. And a permission downgrade will only happen if | |
868 | * is_cow_mapping() returns true. | |
869 | */ | |
870 | if (is_cow_mapping(vma->vm_flags)) | |
871 | mmu_notifier_invalidate_range_start(src_mm, addr, end); | |
872 | ||
873 | ret = 0; | |
1da177e4 LT |
874 | dst_pgd = pgd_offset(dst_mm, addr); |
875 | src_pgd = pgd_offset(src_mm, addr); | |
876 | do { | |
877 | next = pgd_addr_end(addr, end); | |
878 | if (pgd_none_or_clear_bad(src_pgd)) | |
879 | continue; | |
cddb8a5c AA |
880 | if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, |
881 | vma, addr, next))) { | |
882 | ret = -ENOMEM; | |
883 | break; | |
884 | } | |
1da177e4 | 885 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c AA |
886 | |
887 | if (is_cow_mapping(vma->vm_flags)) | |
888 | mmu_notifier_invalidate_range_end(src_mm, | |
889 | vma->vm_start, end); | |
890 | return ret; | |
1da177e4 LT |
891 | } |
892 | ||
51c6f666 | 893 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 894 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 895 | unsigned long addr, unsigned long end, |
51c6f666 | 896 | long *zap_work, struct zap_details *details) |
1da177e4 | 897 | { |
b5810039 | 898 | struct mm_struct *mm = tlb->mm; |
1da177e4 | 899 | pte_t *pte; |
508034a3 | 900 | spinlock_t *ptl; |
d559db08 KH |
901 | int rss[NR_MM_COUNTERS]; |
902 | ||
903 | init_rss_vec(rss); | |
1da177e4 | 904 | |
508034a3 | 905 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
6606c3e0 | 906 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
907 | do { |
908 | pte_t ptent = *pte; | |
51c6f666 RH |
909 | if (pte_none(ptent)) { |
910 | (*zap_work)--; | |
1da177e4 | 911 | continue; |
51c6f666 | 912 | } |
6f5e6b9e HD |
913 | |
914 | (*zap_work) -= PAGE_SIZE; | |
915 | ||
1da177e4 | 916 | if (pte_present(ptent)) { |
ee498ed7 | 917 | struct page *page; |
51c6f666 | 918 | |
6aab341e | 919 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
920 | if (unlikely(details) && page) { |
921 | /* | |
922 | * unmap_shared_mapping_pages() wants to | |
923 | * invalidate cache without truncating: | |
924 | * unmap shared but keep private pages. | |
925 | */ | |
926 | if (details->check_mapping && | |
927 | details->check_mapping != page->mapping) | |
928 | continue; | |
929 | /* | |
930 | * Each page->index must be checked when | |
931 | * invalidating or truncating nonlinear. | |
932 | */ | |
933 | if (details->nonlinear_vma && | |
934 | (page->index < details->first_index || | |
935 | page->index > details->last_index)) | |
936 | continue; | |
937 | } | |
b5810039 | 938 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 939 | tlb->fullmm); |
1da177e4 LT |
940 | tlb_remove_tlb_entry(tlb, pte, addr); |
941 | if (unlikely(!page)) | |
942 | continue; | |
943 | if (unlikely(details) && details->nonlinear_vma | |
944 | && linear_page_index(details->nonlinear_vma, | |
945 | addr) != page->index) | |
b5810039 | 946 | set_pte_at(mm, addr, pte, |
1da177e4 | 947 | pgoff_to_pte(page->index)); |
1da177e4 | 948 | if (PageAnon(page)) |
d559db08 | 949 | rss[MM_ANONPAGES]--; |
6237bcd9 HD |
950 | else { |
951 | if (pte_dirty(ptent)) | |
952 | set_page_dirty(page); | |
4917e5d0 JW |
953 | if (pte_young(ptent) && |
954 | likely(!VM_SequentialReadHint(vma))) | |
bf3f3bc5 | 955 | mark_page_accessed(page); |
d559db08 | 956 | rss[MM_FILEPAGES]--; |
6237bcd9 | 957 | } |
edc315fd | 958 | page_remove_rmap(page); |
3dc14741 HD |
959 | if (unlikely(page_mapcount(page) < 0)) |
960 | print_bad_pte(vma, addr, ptent, page); | |
1da177e4 LT |
961 | tlb_remove_page(tlb, page); |
962 | continue; | |
963 | } | |
964 | /* | |
965 | * If details->check_mapping, we leave swap entries; | |
966 | * if details->nonlinear_vma, we leave file entries. | |
967 | */ | |
968 | if (unlikely(details)) | |
969 | continue; | |
2509ef26 HD |
970 | if (pte_file(ptent)) { |
971 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) | |
972 | print_bad_pte(vma, addr, ptent, NULL); | |
b084d435 KH |
973 | } else { |
974 | swp_entry_t entry = pte_to_swp_entry(ptent); | |
975 | ||
976 | if (!non_swap_entry(entry)) | |
977 | rss[MM_SWAPENTS]--; | |
978 | if (unlikely(!free_swap_and_cache(entry))) | |
979 | print_bad_pte(vma, addr, ptent, NULL); | |
980 | } | |
9888a1ca | 981 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
51c6f666 | 982 | } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0)); |
ae859762 | 983 | |
d559db08 | 984 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 985 | arch_leave_lazy_mmu_mode(); |
508034a3 | 986 | pte_unmap_unlock(pte - 1, ptl); |
51c6f666 RH |
987 | |
988 | return addr; | |
1da177e4 LT |
989 | } |
990 | ||
51c6f666 | 991 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 992 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 993 | unsigned long addr, unsigned long end, |
51c6f666 | 994 | long *zap_work, struct zap_details *details) |
1da177e4 LT |
995 | { |
996 | pmd_t *pmd; | |
997 | unsigned long next; | |
998 | ||
999 | pmd = pmd_offset(pud, addr); | |
1000 | do { | |
1001 | next = pmd_addr_end(addr, end); | |
51c6f666 RH |
1002 | if (pmd_none_or_clear_bad(pmd)) { |
1003 | (*zap_work)--; | |
1da177e4 | 1004 | continue; |
51c6f666 RH |
1005 | } |
1006 | next = zap_pte_range(tlb, vma, pmd, addr, next, | |
1007 | zap_work, details); | |
1008 | } while (pmd++, addr = next, (addr != end && *zap_work > 0)); | |
1009 | ||
1010 | return addr; | |
1da177e4 LT |
1011 | } |
1012 | ||
51c6f666 | 1013 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
b5810039 | 1014 | struct vm_area_struct *vma, pgd_t *pgd, |
1da177e4 | 1015 | unsigned long addr, unsigned long end, |
51c6f666 | 1016 | long *zap_work, struct zap_details *details) |
1da177e4 LT |
1017 | { |
1018 | pud_t *pud; | |
1019 | unsigned long next; | |
1020 | ||
1021 | pud = pud_offset(pgd, addr); | |
1022 | do { | |
1023 | next = pud_addr_end(addr, end); | |
51c6f666 RH |
1024 | if (pud_none_or_clear_bad(pud)) { |
1025 | (*zap_work)--; | |
1da177e4 | 1026 | continue; |
51c6f666 RH |
1027 | } |
1028 | next = zap_pmd_range(tlb, vma, pud, addr, next, | |
1029 | zap_work, details); | |
1030 | } while (pud++, addr = next, (addr != end && *zap_work > 0)); | |
1031 | ||
1032 | return addr; | |
1da177e4 LT |
1033 | } |
1034 | ||
51c6f666 RH |
1035 | static unsigned long unmap_page_range(struct mmu_gather *tlb, |
1036 | struct vm_area_struct *vma, | |
1da177e4 | 1037 | unsigned long addr, unsigned long end, |
51c6f666 | 1038 | long *zap_work, struct zap_details *details) |
1da177e4 LT |
1039 | { |
1040 | pgd_t *pgd; | |
1041 | unsigned long next; | |
1042 | ||
1043 | if (details && !details->check_mapping && !details->nonlinear_vma) | |
1044 | details = NULL; | |
1045 | ||
1046 | BUG_ON(addr >= end); | |
569b846d | 1047 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
1048 | tlb_start_vma(tlb, vma); |
1049 | pgd = pgd_offset(vma->vm_mm, addr); | |
1050 | do { | |
1051 | next = pgd_addr_end(addr, end); | |
51c6f666 RH |
1052 | if (pgd_none_or_clear_bad(pgd)) { |
1053 | (*zap_work)--; | |
1da177e4 | 1054 | continue; |
51c6f666 RH |
1055 | } |
1056 | next = zap_pud_range(tlb, vma, pgd, addr, next, | |
1057 | zap_work, details); | |
1058 | } while (pgd++, addr = next, (addr != end && *zap_work > 0)); | |
1da177e4 | 1059 | tlb_end_vma(tlb, vma); |
569b846d | 1060 | mem_cgroup_uncharge_end(); |
51c6f666 RH |
1061 | |
1062 | return addr; | |
1da177e4 LT |
1063 | } |
1064 | ||
1065 | #ifdef CONFIG_PREEMPT | |
1066 | # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE) | |
1067 | #else | |
1068 | /* No preempt: go for improved straight-line efficiency */ | |
1069 | # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE) | |
1070 | #endif | |
1071 | ||
1072 | /** | |
1073 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
1074 | * @tlbp: address of the caller's struct mmu_gather | |
1da177e4 LT |
1075 | * @vma: the starting vma |
1076 | * @start_addr: virtual address at which to start unmapping | |
1077 | * @end_addr: virtual address at which to end unmapping | |
1078 | * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here | |
1079 | * @details: details of nonlinear truncation or shared cache invalidation | |
1080 | * | |
ee39b37b | 1081 | * Returns the end address of the unmapping (restart addr if interrupted). |
1da177e4 | 1082 | * |
508034a3 | 1083 | * Unmap all pages in the vma list. |
1da177e4 | 1084 | * |
508034a3 HD |
1085 | * We aim to not hold locks for too long (for scheduling latency reasons). |
1086 | * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to | |
1da177e4 LT |
1087 | * return the ending mmu_gather to the caller. |
1088 | * | |
1089 | * Only addresses between `start' and `end' will be unmapped. | |
1090 | * | |
1091 | * The VMA list must be sorted in ascending virtual address order. | |
1092 | * | |
1093 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1094 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1095 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1096 | * drops the lock and schedules. | |
1097 | */ | |
508034a3 | 1098 | unsigned long unmap_vmas(struct mmu_gather **tlbp, |
1da177e4 LT |
1099 | struct vm_area_struct *vma, unsigned long start_addr, |
1100 | unsigned long end_addr, unsigned long *nr_accounted, | |
1101 | struct zap_details *details) | |
1102 | { | |
51c6f666 | 1103 | long zap_work = ZAP_BLOCK_SIZE; |
1da177e4 LT |
1104 | unsigned long tlb_start = 0; /* For tlb_finish_mmu */ |
1105 | int tlb_start_valid = 0; | |
ee39b37b | 1106 | unsigned long start = start_addr; |
1da177e4 | 1107 | spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL; |
4d6ddfa9 | 1108 | int fullmm = (*tlbp)->fullmm; |
cddb8a5c | 1109 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1110 | |
cddb8a5c | 1111 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
1da177e4 | 1112 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) { |
1da177e4 LT |
1113 | unsigned long end; |
1114 | ||
1115 | start = max(vma->vm_start, start_addr); | |
1116 | if (start >= vma->vm_end) | |
1117 | continue; | |
1118 | end = min(vma->vm_end, end_addr); | |
1119 | if (end <= vma->vm_start) | |
1120 | continue; | |
1121 | ||
1122 | if (vma->vm_flags & VM_ACCOUNT) | |
1123 | *nr_accounted += (end - start) >> PAGE_SHIFT; | |
1124 | ||
34801ba9 | 1125 | if (unlikely(is_pfn_mapping(vma))) |
2ab64037 | 1126 | untrack_pfn_vma(vma, 0, 0); |
1127 | ||
1da177e4 | 1128 | while (start != end) { |
1da177e4 LT |
1129 | if (!tlb_start_valid) { |
1130 | tlb_start = start; | |
1131 | tlb_start_valid = 1; | |
1132 | } | |
1133 | ||
51c6f666 | 1134 | if (unlikely(is_vm_hugetlb_page(vma))) { |
a137e1cc AK |
1135 | /* |
1136 | * It is undesirable to test vma->vm_file as it | |
1137 | * should be non-null for valid hugetlb area. | |
1138 | * However, vm_file will be NULL in the error | |
1139 | * cleanup path of do_mmap_pgoff. When | |
1140 | * hugetlbfs ->mmap method fails, | |
1141 | * do_mmap_pgoff() nullifies vma->vm_file | |
1142 | * before calling this function to clean up. | |
1143 | * Since no pte has actually been setup, it is | |
1144 | * safe to do nothing in this case. | |
1145 | */ | |
1146 | if (vma->vm_file) { | |
1147 | unmap_hugepage_range(vma, start, end, NULL); | |
1148 | zap_work -= (end - start) / | |
a5516438 | 1149 | pages_per_huge_page(hstate_vma(vma)); |
a137e1cc AK |
1150 | } |
1151 | ||
51c6f666 RH |
1152 | start = end; |
1153 | } else | |
1154 | start = unmap_page_range(*tlbp, vma, | |
1155 | start, end, &zap_work, details); | |
1156 | ||
1157 | if (zap_work > 0) { | |
1158 | BUG_ON(start != end); | |
1159 | break; | |
1da177e4 LT |
1160 | } |
1161 | ||
1da177e4 LT |
1162 | tlb_finish_mmu(*tlbp, tlb_start, start); |
1163 | ||
1164 | if (need_resched() || | |
95c354fe | 1165 | (i_mmap_lock && spin_needbreak(i_mmap_lock))) { |
1da177e4 | 1166 | if (i_mmap_lock) { |
508034a3 | 1167 | *tlbp = NULL; |
1da177e4 LT |
1168 | goto out; |
1169 | } | |
1da177e4 | 1170 | cond_resched(); |
1da177e4 LT |
1171 | } |
1172 | ||
508034a3 | 1173 | *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm); |
1da177e4 | 1174 | tlb_start_valid = 0; |
51c6f666 | 1175 | zap_work = ZAP_BLOCK_SIZE; |
1da177e4 LT |
1176 | } |
1177 | } | |
1178 | out: | |
cddb8a5c | 1179 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
ee39b37b | 1180 | return start; /* which is now the end (or restart) address */ |
1da177e4 LT |
1181 | } |
1182 | ||
1183 | /** | |
1184 | * zap_page_range - remove user pages in a given range | |
1185 | * @vma: vm_area_struct holding the applicable pages | |
1186 | * @address: starting address of pages to zap | |
1187 | * @size: number of bytes to zap | |
1188 | * @details: details of nonlinear truncation or shared cache invalidation | |
1189 | */ | |
ee39b37b | 1190 | unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1191 | unsigned long size, struct zap_details *details) |
1192 | { | |
1193 | struct mm_struct *mm = vma->vm_mm; | |
1194 | struct mmu_gather *tlb; | |
1195 | unsigned long end = address + size; | |
1196 | unsigned long nr_accounted = 0; | |
1197 | ||
1da177e4 | 1198 | lru_add_drain(); |
1da177e4 | 1199 | tlb = tlb_gather_mmu(mm, 0); |
365e9c87 | 1200 | update_hiwater_rss(mm); |
508034a3 HD |
1201 | end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details); |
1202 | if (tlb) | |
1203 | tlb_finish_mmu(tlb, address, end); | |
ee39b37b | 1204 | return end; |
1da177e4 LT |
1205 | } |
1206 | ||
c627f9cc JS |
1207 | /** |
1208 | * zap_vma_ptes - remove ptes mapping the vma | |
1209 | * @vma: vm_area_struct holding ptes to be zapped | |
1210 | * @address: starting address of pages to zap | |
1211 | * @size: number of bytes to zap | |
1212 | * | |
1213 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1214 | * | |
1215 | * The entire address range must be fully contained within the vma. | |
1216 | * | |
1217 | * Returns 0 if successful. | |
1218 | */ | |
1219 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1220 | unsigned long size) | |
1221 | { | |
1222 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1223 | !(vma->vm_flags & VM_PFNMAP)) | |
1224 | return -1; | |
1225 | zap_page_range(vma, address, size, NULL); | |
1226 | return 0; | |
1227 | } | |
1228 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1229 | ||
1da177e4 LT |
1230 | /* |
1231 | * Do a quick page-table lookup for a single page. | |
1da177e4 | 1232 | */ |
6aab341e | 1233 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, |
deceb6cd | 1234 | unsigned int flags) |
1da177e4 LT |
1235 | { |
1236 | pgd_t *pgd; | |
1237 | pud_t *pud; | |
1238 | pmd_t *pmd; | |
1239 | pte_t *ptep, pte; | |
deceb6cd | 1240 | spinlock_t *ptl; |
1da177e4 | 1241 | struct page *page; |
6aab341e | 1242 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1243 | |
deceb6cd HD |
1244 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); |
1245 | if (!IS_ERR(page)) { | |
1246 | BUG_ON(flags & FOLL_GET); | |
1247 | goto out; | |
1248 | } | |
1da177e4 | 1249 | |
deceb6cd | 1250 | page = NULL; |
1da177e4 LT |
1251 | pgd = pgd_offset(mm, address); |
1252 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
deceb6cd | 1253 | goto no_page_table; |
1da177e4 LT |
1254 | |
1255 | pud = pud_offset(pgd, address); | |
ceb86879 | 1256 | if (pud_none(*pud)) |
deceb6cd | 1257 | goto no_page_table; |
ceb86879 AK |
1258 | if (pud_huge(*pud)) { |
1259 | BUG_ON(flags & FOLL_GET); | |
1260 | page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE); | |
1261 | goto out; | |
1262 | } | |
1263 | if (unlikely(pud_bad(*pud))) | |
1264 | goto no_page_table; | |
1265 | ||
1da177e4 | 1266 | pmd = pmd_offset(pud, address); |
aeed5fce | 1267 | if (pmd_none(*pmd)) |
deceb6cd | 1268 | goto no_page_table; |
deceb6cd HD |
1269 | if (pmd_huge(*pmd)) { |
1270 | BUG_ON(flags & FOLL_GET); | |
1271 | page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); | |
1da177e4 | 1272 | goto out; |
deceb6cd | 1273 | } |
aeed5fce HD |
1274 | if (unlikely(pmd_bad(*pmd))) |
1275 | goto no_page_table; | |
1276 | ||
deceb6cd | 1277 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
1278 | |
1279 | pte = *ptep; | |
deceb6cd | 1280 | if (!pte_present(pte)) |
89f5b7da | 1281 | goto no_page; |
deceb6cd HD |
1282 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
1283 | goto unlock; | |
a13ea5b7 | 1284 | |
6aab341e | 1285 | page = vm_normal_page(vma, address, pte); |
a13ea5b7 HD |
1286 | if (unlikely(!page)) { |
1287 | if ((flags & FOLL_DUMP) || | |
62eede62 | 1288 | !is_zero_pfn(pte_pfn(pte))) |
a13ea5b7 HD |
1289 | goto bad_page; |
1290 | page = pte_page(pte); | |
1291 | } | |
1da177e4 | 1292 | |
deceb6cd HD |
1293 | if (flags & FOLL_GET) |
1294 | get_page(page); | |
1295 | if (flags & FOLL_TOUCH) { | |
1296 | if ((flags & FOLL_WRITE) && | |
1297 | !pte_dirty(pte) && !PageDirty(page)) | |
1298 | set_page_dirty(page); | |
bd775c42 KM |
1299 | /* |
1300 | * pte_mkyoung() would be more correct here, but atomic care | |
1301 | * is needed to avoid losing the dirty bit: it is easier to use | |
1302 | * mark_page_accessed(). | |
1303 | */ | |
deceb6cd HD |
1304 | mark_page_accessed(page); |
1305 | } | |
1306 | unlock: | |
1307 | pte_unmap_unlock(ptep, ptl); | |
1da177e4 | 1308 | out: |
deceb6cd | 1309 | return page; |
1da177e4 | 1310 | |
89f5b7da LT |
1311 | bad_page: |
1312 | pte_unmap_unlock(ptep, ptl); | |
1313 | return ERR_PTR(-EFAULT); | |
1314 | ||
1315 | no_page: | |
1316 | pte_unmap_unlock(ptep, ptl); | |
1317 | if (!pte_none(pte)) | |
1318 | return page; | |
8e4b9a60 | 1319 | |
deceb6cd HD |
1320 | no_page_table: |
1321 | /* | |
1322 | * When core dumping an enormous anonymous area that nobody | |
8e4b9a60 HD |
1323 | * has touched so far, we don't want to allocate unnecessary pages or |
1324 | * page tables. Return error instead of NULL to skip handle_mm_fault, | |
1325 | * then get_dump_page() will return NULL to leave a hole in the dump. | |
1326 | * But we can only make this optimization where a hole would surely | |
1327 | * be zero-filled if handle_mm_fault() actually did handle it. | |
deceb6cd | 1328 | */ |
8e4b9a60 HD |
1329 | if ((flags & FOLL_DUMP) && |
1330 | (!vma->vm_ops || !vma->vm_ops->fault)) | |
1331 | return ERR_PTR(-EFAULT); | |
deceb6cd | 1332 | return page; |
1da177e4 LT |
1333 | } |
1334 | ||
b291f000 | 1335 | int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
58fa879e | 1336 | unsigned long start, int nr_pages, unsigned int gup_flags, |
9d73777e | 1337 | struct page **pages, struct vm_area_struct **vmas) |
1da177e4 LT |
1338 | { |
1339 | int i; | |
58fa879e | 1340 | unsigned long vm_flags; |
1da177e4 | 1341 | |
9d73777e | 1342 | if (nr_pages <= 0) |
900cf086 | 1343 | return 0; |
58fa879e HD |
1344 | |
1345 | VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET)); | |
1346 | ||
1da177e4 LT |
1347 | /* |
1348 | * Require read or write permissions. | |
58fa879e | 1349 | * If FOLL_FORCE is set, we only require the "MAY" flags. |
1da177e4 | 1350 | */ |
58fa879e HD |
1351 | vm_flags = (gup_flags & FOLL_WRITE) ? |
1352 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
1353 | vm_flags &= (gup_flags & FOLL_FORCE) ? | |
1354 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1da177e4 LT |
1355 | i = 0; |
1356 | ||
1357 | do { | |
deceb6cd | 1358 | struct vm_area_struct *vma; |
1da177e4 LT |
1359 | |
1360 | vma = find_extend_vma(mm, start); | |
1361 | if (!vma && in_gate_area(tsk, start)) { | |
1362 | unsigned long pg = start & PAGE_MASK; | |
1363 | struct vm_area_struct *gate_vma = get_gate_vma(tsk); | |
1364 | pgd_t *pgd; | |
1365 | pud_t *pud; | |
1366 | pmd_t *pmd; | |
1367 | pte_t *pte; | |
b291f000 NP |
1368 | |
1369 | /* user gate pages are read-only */ | |
58fa879e | 1370 | if (gup_flags & FOLL_WRITE) |
1da177e4 LT |
1371 | return i ? : -EFAULT; |
1372 | if (pg > TASK_SIZE) | |
1373 | pgd = pgd_offset_k(pg); | |
1374 | else | |
1375 | pgd = pgd_offset_gate(mm, pg); | |
1376 | BUG_ON(pgd_none(*pgd)); | |
1377 | pud = pud_offset(pgd, pg); | |
1378 | BUG_ON(pud_none(*pud)); | |
1379 | pmd = pmd_offset(pud, pg); | |
690dbe1c HD |
1380 | if (pmd_none(*pmd)) |
1381 | return i ? : -EFAULT; | |
1da177e4 | 1382 | pte = pte_offset_map(pmd, pg); |
690dbe1c HD |
1383 | if (pte_none(*pte)) { |
1384 | pte_unmap(pte); | |
1385 | return i ? : -EFAULT; | |
1386 | } | |
1da177e4 | 1387 | if (pages) { |
fa2a455b | 1388 | struct page *page = vm_normal_page(gate_vma, start, *pte); |
6aab341e LT |
1389 | pages[i] = page; |
1390 | if (page) | |
1391 | get_page(page); | |
1da177e4 LT |
1392 | } |
1393 | pte_unmap(pte); | |
1394 | if (vmas) | |
1395 | vmas[i] = gate_vma; | |
1396 | i++; | |
1397 | start += PAGE_SIZE; | |
9d73777e | 1398 | nr_pages--; |
1da177e4 LT |
1399 | continue; |
1400 | } | |
1401 | ||
b291f000 NP |
1402 | if (!vma || |
1403 | (vma->vm_flags & (VM_IO | VM_PFNMAP)) || | |
1c3aff1c | 1404 | !(vm_flags & vma->vm_flags)) |
1da177e4 LT |
1405 | return i ? : -EFAULT; |
1406 | ||
2a15efc9 HD |
1407 | if (is_vm_hugetlb_page(vma)) { |
1408 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
58fa879e | 1409 | &start, &nr_pages, i, gup_flags); |
2a15efc9 HD |
1410 | continue; |
1411 | } | |
deceb6cd | 1412 | |
1da177e4 | 1413 | do { |
08ef4729 | 1414 | struct page *page; |
58fa879e | 1415 | unsigned int foll_flags = gup_flags; |
1da177e4 | 1416 | |
462e00cc | 1417 | /* |
4779280d | 1418 | * If we have a pending SIGKILL, don't keep faulting |
1c3aff1c | 1419 | * pages and potentially allocating memory. |
462e00cc | 1420 | */ |
1c3aff1c | 1421 | if (unlikely(fatal_signal_pending(current))) |
4779280d | 1422 | return i ? i : -ERESTARTSYS; |
462e00cc | 1423 | |
deceb6cd | 1424 | cond_resched(); |
6aab341e | 1425 | while (!(page = follow_page(vma, start, foll_flags))) { |
deceb6cd | 1426 | int ret; |
d06063cc | 1427 | |
d26ed650 HD |
1428 | ret = handle_mm_fault(mm, vma, start, |
1429 | (foll_flags & FOLL_WRITE) ? | |
1430 | FAULT_FLAG_WRITE : 0); | |
1431 | ||
83c54070 NP |
1432 | if (ret & VM_FAULT_ERROR) { |
1433 | if (ret & VM_FAULT_OOM) | |
1434 | return i ? i : -ENOMEM; | |
d1737fdb AK |
1435 | if (ret & |
1436 | (VM_FAULT_HWPOISON|VM_FAULT_SIGBUS)) | |
83c54070 NP |
1437 | return i ? i : -EFAULT; |
1438 | BUG(); | |
1439 | } | |
1440 | if (ret & VM_FAULT_MAJOR) | |
1441 | tsk->maj_flt++; | |
1442 | else | |
1443 | tsk->min_flt++; | |
1444 | ||
a68d2ebc | 1445 | /* |
83c54070 NP |
1446 | * The VM_FAULT_WRITE bit tells us that |
1447 | * do_wp_page has broken COW when necessary, | |
1448 | * even if maybe_mkwrite decided not to set | |
1449 | * pte_write. We can thus safely do subsequent | |
878b63ac HD |
1450 | * page lookups as if they were reads. But only |
1451 | * do so when looping for pte_write is futile: | |
1452 | * in some cases userspace may also be wanting | |
1453 | * to write to the gotten user page, which a | |
1454 | * read fault here might prevent (a readonly | |
1455 | * page might get reCOWed by userspace write). | |
a68d2ebc | 1456 | */ |
878b63ac HD |
1457 | if ((ret & VM_FAULT_WRITE) && |
1458 | !(vma->vm_flags & VM_WRITE)) | |
deceb6cd | 1459 | foll_flags &= ~FOLL_WRITE; |
83c54070 | 1460 | |
7f7bbbe5 | 1461 | cond_resched(); |
1da177e4 | 1462 | } |
89f5b7da LT |
1463 | if (IS_ERR(page)) |
1464 | return i ? i : PTR_ERR(page); | |
1da177e4 | 1465 | if (pages) { |
08ef4729 | 1466 | pages[i] = page; |
03beb076 | 1467 | |
a6f36be3 | 1468 | flush_anon_page(vma, page, start); |
08ef4729 | 1469 | flush_dcache_page(page); |
1da177e4 LT |
1470 | } |
1471 | if (vmas) | |
1472 | vmas[i] = vma; | |
1473 | i++; | |
1474 | start += PAGE_SIZE; | |
9d73777e PZ |
1475 | nr_pages--; |
1476 | } while (nr_pages && start < vma->vm_end); | |
1477 | } while (nr_pages); | |
1da177e4 LT |
1478 | return i; |
1479 | } | |
b291f000 | 1480 | |
d2bf6be8 NP |
1481 | /** |
1482 | * get_user_pages() - pin user pages in memory | |
1483 | * @tsk: task_struct of target task | |
1484 | * @mm: mm_struct of target mm | |
1485 | * @start: starting user address | |
9d73777e | 1486 | * @nr_pages: number of pages from start to pin |
d2bf6be8 NP |
1487 | * @write: whether pages will be written to by the caller |
1488 | * @force: whether to force write access even if user mapping is | |
1489 | * readonly. This will result in the page being COWed even | |
1490 | * in MAP_SHARED mappings. You do not want this. | |
1491 | * @pages: array that receives pointers to the pages pinned. | |
1492 | * Should be at least nr_pages long. Or NULL, if caller | |
1493 | * only intends to ensure the pages are faulted in. | |
1494 | * @vmas: array of pointers to vmas corresponding to each page. | |
1495 | * Or NULL if the caller does not require them. | |
1496 | * | |
1497 | * Returns number of pages pinned. This may be fewer than the number | |
9d73777e | 1498 | * requested. If nr_pages is 0 or negative, returns 0. If no pages |
d2bf6be8 NP |
1499 | * were pinned, returns -errno. Each page returned must be released |
1500 | * with a put_page() call when it is finished with. vmas will only | |
1501 | * remain valid while mmap_sem is held. | |
1502 | * | |
1503 | * Must be called with mmap_sem held for read or write. | |
1504 | * | |
1505 | * get_user_pages walks a process's page tables and takes a reference to | |
1506 | * each struct page that each user address corresponds to at a given | |
1507 | * instant. That is, it takes the page that would be accessed if a user | |
1508 | * thread accesses the given user virtual address at that instant. | |
1509 | * | |
1510 | * This does not guarantee that the page exists in the user mappings when | |
1511 | * get_user_pages returns, and there may even be a completely different | |
1512 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
1513 | * and subsequently re faulted). However it does guarantee that the page | |
1514 | * won't be freed completely. And mostly callers simply care that the page | |
1515 | * contains data that was valid *at some point in time*. Typically, an IO | |
1516 | * or similar operation cannot guarantee anything stronger anyway because | |
1517 | * locks can't be held over the syscall boundary. | |
1518 | * | |
1519 | * If write=0, the page must not be written to. If the page is written to, | |
1520 | * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called | |
1521 | * after the page is finished with, and before put_page is called. | |
1522 | * | |
1523 | * get_user_pages is typically used for fewer-copy IO operations, to get a | |
1524 | * handle on the memory by some means other than accesses via the user virtual | |
1525 | * addresses. The pages may be submitted for DMA to devices or accessed via | |
1526 | * their kernel linear mapping (via the kmap APIs). Care should be taken to | |
1527 | * use the correct cache flushing APIs. | |
1528 | * | |
1529 | * See also get_user_pages_fast, for performance critical applications. | |
1530 | */ | |
b291f000 | 1531 | int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
9d73777e | 1532 | unsigned long start, int nr_pages, int write, int force, |
b291f000 NP |
1533 | struct page **pages, struct vm_area_struct **vmas) |
1534 | { | |
58fa879e | 1535 | int flags = FOLL_TOUCH; |
b291f000 | 1536 | |
58fa879e HD |
1537 | if (pages) |
1538 | flags |= FOLL_GET; | |
b291f000 | 1539 | if (write) |
58fa879e | 1540 | flags |= FOLL_WRITE; |
b291f000 | 1541 | if (force) |
58fa879e | 1542 | flags |= FOLL_FORCE; |
b291f000 | 1543 | |
9d73777e | 1544 | return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas); |
b291f000 | 1545 | } |
1da177e4 LT |
1546 | EXPORT_SYMBOL(get_user_pages); |
1547 | ||
f3e8fccd HD |
1548 | /** |
1549 | * get_dump_page() - pin user page in memory while writing it to core dump | |
1550 | * @addr: user address | |
1551 | * | |
1552 | * Returns struct page pointer of user page pinned for dump, | |
1553 | * to be freed afterwards by page_cache_release() or put_page(). | |
1554 | * | |
1555 | * Returns NULL on any kind of failure - a hole must then be inserted into | |
1556 | * the corefile, to preserve alignment with its headers; and also returns | |
1557 | * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - | |
1558 | * allowing a hole to be left in the corefile to save diskspace. | |
1559 | * | |
1560 | * Called without mmap_sem, but after all other threads have been killed. | |
1561 | */ | |
1562 | #ifdef CONFIG_ELF_CORE | |
1563 | struct page *get_dump_page(unsigned long addr) | |
1564 | { | |
1565 | struct vm_area_struct *vma; | |
1566 | struct page *page; | |
1567 | ||
1568 | if (__get_user_pages(current, current->mm, addr, 1, | |
58fa879e | 1569 | FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma) < 1) |
f3e8fccd | 1570 | return NULL; |
f3e8fccd HD |
1571 | flush_cache_page(vma, addr, page_to_pfn(page)); |
1572 | return page; | |
1573 | } | |
1574 | #endif /* CONFIG_ELF_CORE */ | |
1575 | ||
920c7a5d HH |
1576 | pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr, |
1577 | spinlock_t **ptl) | |
c9cfcddf LT |
1578 | { |
1579 | pgd_t * pgd = pgd_offset(mm, addr); | |
1580 | pud_t * pud = pud_alloc(mm, pgd, addr); | |
1581 | if (pud) { | |
49c91fb0 | 1582 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
c9cfcddf LT |
1583 | if (pmd) |
1584 | return pte_alloc_map_lock(mm, pmd, addr, ptl); | |
1585 | } | |
1586 | return NULL; | |
1587 | } | |
1588 | ||
238f58d8 LT |
1589 | /* |
1590 | * This is the old fallback for page remapping. | |
1591 | * | |
1592 | * For historical reasons, it only allows reserved pages. Only | |
1593 | * old drivers should use this, and they needed to mark their | |
1594 | * pages reserved for the old functions anyway. | |
1595 | */ | |
423bad60 NP |
1596 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1597 | struct page *page, pgprot_t prot) | |
238f58d8 | 1598 | { |
423bad60 | 1599 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1600 | int retval; |
c9cfcddf | 1601 | pte_t *pte; |
8a9f3ccd BS |
1602 | spinlock_t *ptl; |
1603 | ||
238f58d8 | 1604 | retval = -EINVAL; |
a145dd41 | 1605 | if (PageAnon(page)) |
5b4e655e | 1606 | goto out; |
238f58d8 LT |
1607 | retval = -ENOMEM; |
1608 | flush_dcache_page(page); | |
c9cfcddf | 1609 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1610 | if (!pte) |
5b4e655e | 1611 | goto out; |
238f58d8 LT |
1612 | retval = -EBUSY; |
1613 | if (!pte_none(*pte)) | |
1614 | goto out_unlock; | |
1615 | ||
1616 | /* Ok, finally just insert the thing.. */ | |
1617 | get_page(page); | |
34e55232 | 1618 | inc_mm_counter_fast(mm, MM_FILEPAGES); |
238f58d8 LT |
1619 | page_add_file_rmap(page); |
1620 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); | |
1621 | ||
1622 | retval = 0; | |
8a9f3ccd BS |
1623 | pte_unmap_unlock(pte, ptl); |
1624 | return retval; | |
238f58d8 LT |
1625 | out_unlock: |
1626 | pte_unmap_unlock(pte, ptl); | |
1627 | out: | |
1628 | return retval; | |
1629 | } | |
1630 | ||
bfa5bf6d REB |
1631 | /** |
1632 | * vm_insert_page - insert single page into user vma | |
1633 | * @vma: user vma to map to | |
1634 | * @addr: target user address of this page | |
1635 | * @page: source kernel page | |
1636 | * | |
a145dd41 LT |
1637 | * This allows drivers to insert individual pages they've allocated |
1638 | * into a user vma. | |
1639 | * | |
1640 | * The page has to be a nice clean _individual_ kernel allocation. | |
1641 | * If you allocate a compound page, you need to have marked it as | |
1642 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1643 | * (see split_page()). |
a145dd41 LT |
1644 | * |
1645 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1646 | * took an arbitrary page protection parameter. This doesn't allow | |
1647 | * that. Your vma protection will have to be set up correctly, which | |
1648 | * means that if you want a shared writable mapping, you'd better | |
1649 | * ask for a shared writable mapping! | |
1650 | * | |
1651 | * The page does not need to be reserved. | |
1652 | */ | |
423bad60 NP |
1653 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1654 | struct page *page) | |
a145dd41 LT |
1655 | { |
1656 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1657 | return -EFAULT; | |
1658 | if (!page_count(page)) | |
1659 | return -EINVAL; | |
4d7672b4 | 1660 | vma->vm_flags |= VM_INSERTPAGE; |
423bad60 | 1661 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1662 | } |
e3c3374f | 1663 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1664 | |
423bad60 NP |
1665 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
1666 | unsigned long pfn, pgprot_t prot) | |
1667 | { | |
1668 | struct mm_struct *mm = vma->vm_mm; | |
1669 | int retval; | |
1670 | pte_t *pte, entry; | |
1671 | spinlock_t *ptl; | |
1672 | ||
1673 | retval = -ENOMEM; | |
1674 | pte = get_locked_pte(mm, addr, &ptl); | |
1675 | if (!pte) | |
1676 | goto out; | |
1677 | retval = -EBUSY; | |
1678 | if (!pte_none(*pte)) | |
1679 | goto out_unlock; | |
1680 | ||
1681 | /* Ok, finally just insert the thing.. */ | |
1682 | entry = pte_mkspecial(pfn_pte(pfn, prot)); | |
1683 | set_pte_at(mm, addr, pte, entry); | |
4b3073e1 | 1684 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
1685 | |
1686 | retval = 0; | |
1687 | out_unlock: | |
1688 | pte_unmap_unlock(pte, ptl); | |
1689 | out: | |
1690 | return retval; | |
1691 | } | |
1692 | ||
e0dc0d8f NP |
1693 | /** |
1694 | * vm_insert_pfn - insert single pfn into user vma | |
1695 | * @vma: user vma to map to | |
1696 | * @addr: target user address of this page | |
1697 | * @pfn: source kernel pfn | |
1698 | * | |
1699 | * Similar to vm_inert_page, this allows drivers to insert individual pages | |
1700 | * they've allocated into a user vma. Same comments apply. | |
1701 | * | |
1702 | * This function should only be called from a vm_ops->fault handler, and | |
1703 | * in that case the handler should return NULL. | |
0d71d10a NP |
1704 | * |
1705 | * vma cannot be a COW mapping. | |
1706 | * | |
1707 | * As this is called only for pages that do not currently exist, we | |
1708 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1709 | */ |
1710 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1711 | unsigned long pfn) |
e0dc0d8f | 1712 | { |
2ab64037 | 1713 | int ret; |
e4b866ed | 1714 | pgprot_t pgprot = vma->vm_page_prot; |
7e675137 NP |
1715 | /* |
1716 | * Technically, architectures with pte_special can avoid all these | |
1717 | * restrictions (same for remap_pfn_range). However we would like | |
1718 | * consistency in testing and feature parity among all, so we should | |
1719 | * try to keep these invariants in place for everybody. | |
1720 | */ | |
b379d790 JH |
1721 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1722 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1723 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1724 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1725 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1726 | |
423bad60 NP |
1727 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1728 | return -EFAULT; | |
e4b866ed | 1729 | if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE)) |
2ab64037 | 1730 | return -EINVAL; |
1731 | ||
e4b866ed | 1732 | ret = insert_pfn(vma, addr, pfn, pgprot); |
2ab64037 | 1733 | |
1734 | if (ret) | |
1735 | untrack_pfn_vma(vma, pfn, PAGE_SIZE); | |
1736 | ||
1737 | return ret; | |
423bad60 NP |
1738 | } |
1739 | EXPORT_SYMBOL(vm_insert_pfn); | |
e0dc0d8f | 1740 | |
423bad60 NP |
1741 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
1742 | unsigned long pfn) | |
1743 | { | |
1744 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); | |
e0dc0d8f | 1745 | |
423bad60 NP |
1746 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1747 | return -EFAULT; | |
e0dc0d8f | 1748 | |
423bad60 NP |
1749 | /* |
1750 | * If we don't have pte special, then we have to use the pfn_valid() | |
1751 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1752 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1753 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1754 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 NP |
1755 | */ |
1756 | if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { | |
1757 | struct page *page; | |
1758 | ||
1759 | page = pfn_to_page(pfn); | |
1760 | return insert_page(vma, addr, page, vma->vm_page_prot); | |
1761 | } | |
1762 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); | |
e0dc0d8f | 1763 | } |
423bad60 | 1764 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1765 | |
1da177e4 LT |
1766 | /* |
1767 | * maps a range of physical memory into the requested pages. the old | |
1768 | * mappings are removed. any references to nonexistent pages results | |
1769 | * in null mappings (currently treated as "copy-on-access") | |
1770 | */ | |
1771 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1772 | unsigned long addr, unsigned long end, | |
1773 | unsigned long pfn, pgprot_t prot) | |
1774 | { | |
1775 | pte_t *pte; | |
c74df32c | 1776 | spinlock_t *ptl; |
1da177e4 | 1777 | |
c74df32c | 1778 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1779 | if (!pte) |
1780 | return -ENOMEM; | |
6606c3e0 | 1781 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1782 | do { |
1783 | BUG_ON(!pte_none(*pte)); | |
7e675137 | 1784 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1785 | pfn++; |
1786 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1787 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1788 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
1789 | return 0; |
1790 | } | |
1791 | ||
1792 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1793 | unsigned long addr, unsigned long end, | |
1794 | unsigned long pfn, pgprot_t prot) | |
1795 | { | |
1796 | pmd_t *pmd; | |
1797 | unsigned long next; | |
1798 | ||
1799 | pfn -= addr >> PAGE_SHIFT; | |
1800 | pmd = pmd_alloc(mm, pud, addr); | |
1801 | if (!pmd) | |
1802 | return -ENOMEM; | |
1803 | do { | |
1804 | next = pmd_addr_end(addr, end); | |
1805 | if (remap_pte_range(mm, pmd, addr, next, | |
1806 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1807 | return -ENOMEM; | |
1808 | } while (pmd++, addr = next, addr != end); | |
1809 | return 0; | |
1810 | } | |
1811 | ||
1812 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1813 | unsigned long addr, unsigned long end, | |
1814 | unsigned long pfn, pgprot_t prot) | |
1815 | { | |
1816 | pud_t *pud; | |
1817 | unsigned long next; | |
1818 | ||
1819 | pfn -= addr >> PAGE_SHIFT; | |
1820 | pud = pud_alloc(mm, pgd, addr); | |
1821 | if (!pud) | |
1822 | return -ENOMEM; | |
1823 | do { | |
1824 | next = pud_addr_end(addr, end); | |
1825 | if (remap_pmd_range(mm, pud, addr, next, | |
1826 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1827 | return -ENOMEM; | |
1828 | } while (pud++, addr = next, addr != end); | |
1829 | return 0; | |
1830 | } | |
1831 | ||
bfa5bf6d REB |
1832 | /** |
1833 | * remap_pfn_range - remap kernel memory to userspace | |
1834 | * @vma: user vma to map to | |
1835 | * @addr: target user address to start at | |
1836 | * @pfn: physical address of kernel memory | |
1837 | * @size: size of map area | |
1838 | * @prot: page protection flags for this mapping | |
1839 | * | |
1840 | * Note: this is only safe if the mm semaphore is held when called. | |
1841 | */ | |
1da177e4 LT |
1842 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1843 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1844 | { | |
1845 | pgd_t *pgd; | |
1846 | unsigned long next; | |
2d15cab8 | 1847 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 LT |
1848 | struct mm_struct *mm = vma->vm_mm; |
1849 | int err; | |
1850 | ||
1851 | /* | |
1852 | * Physically remapped pages are special. Tell the | |
1853 | * rest of the world about it: | |
1854 | * VM_IO tells people not to look at these pages | |
1855 | * (accesses can have side effects). | |
0b14c179 HD |
1856 | * VM_RESERVED is specified all over the place, because |
1857 | * in 2.4 it kept swapout's vma scan off this vma; but | |
1858 | * in 2.6 the LRU scan won't even find its pages, so this | |
1859 | * flag means no more than count its pages in reserved_vm, | |
1860 | * and omit it from core dump, even when VM_IO turned off. | |
6aab341e LT |
1861 | * VM_PFNMAP tells the core MM that the base pages are just |
1862 | * raw PFN mappings, and do not have a "struct page" associated | |
1863 | * with them. | |
fb155c16 LT |
1864 | * |
1865 | * There's a horrible special case to handle copy-on-write | |
1866 | * behaviour that some programs depend on. We mark the "original" | |
1867 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
1da177e4 | 1868 | */ |
4bb9c5c0 | 1869 | if (addr == vma->vm_start && end == vma->vm_end) { |
fb155c16 | 1870 | vma->vm_pgoff = pfn; |
895791da | 1871 | vma->vm_flags |= VM_PFN_AT_MMAP; |
4bb9c5c0 | 1872 | } else if (is_cow_mapping(vma->vm_flags)) |
3c8bb73a | 1873 | return -EINVAL; |
fb155c16 | 1874 | |
6aab341e | 1875 | vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; |
1da177e4 | 1876 | |
e4b866ed | 1877 | err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size)); |
a3670613 | 1878 | if (err) { |
1879 | /* | |
1880 | * To indicate that track_pfn related cleanup is not | |
1881 | * needed from higher level routine calling unmap_vmas | |
1882 | */ | |
1883 | vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP); | |
895791da | 1884 | vma->vm_flags &= ~VM_PFN_AT_MMAP; |
2ab64037 | 1885 | return -EINVAL; |
a3670613 | 1886 | } |
2ab64037 | 1887 | |
1da177e4 LT |
1888 | BUG_ON(addr >= end); |
1889 | pfn -= addr >> PAGE_SHIFT; | |
1890 | pgd = pgd_offset(mm, addr); | |
1891 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
1892 | do { |
1893 | next = pgd_addr_end(addr, end); | |
1894 | err = remap_pud_range(mm, pgd, addr, next, | |
1895 | pfn + (addr >> PAGE_SHIFT), prot); | |
1896 | if (err) | |
1897 | break; | |
1898 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 1899 | |
1900 | if (err) | |
1901 | untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size)); | |
1902 | ||
1da177e4 LT |
1903 | return err; |
1904 | } | |
1905 | EXPORT_SYMBOL(remap_pfn_range); | |
1906 | ||
aee16b3c JF |
1907 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
1908 | unsigned long addr, unsigned long end, | |
1909 | pte_fn_t fn, void *data) | |
1910 | { | |
1911 | pte_t *pte; | |
1912 | int err; | |
2f569afd | 1913 | pgtable_t token; |
94909914 | 1914 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
1915 | |
1916 | pte = (mm == &init_mm) ? | |
1917 | pte_alloc_kernel(pmd, addr) : | |
1918 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
1919 | if (!pte) | |
1920 | return -ENOMEM; | |
1921 | ||
1922 | BUG_ON(pmd_huge(*pmd)); | |
1923 | ||
38e0edb1 JF |
1924 | arch_enter_lazy_mmu_mode(); |
1925 | ||
2f569afd | 1926 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
1927 | |
1928 | do { | |
c36987e2 | 1929 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
1930 | if (err) |
1931 | break; | |
c36987e2 | 1932 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 1933 | |
38e0edb1 JF |
1934 | arch_leave_lazy_mmu_mode(); |
1935 | ||
aee16b3c JF |
1936 | if (mm != &init_mm) |
1937 | pte_unmap_unlock(pte-1, ptl); | |
1938 | return err; | |
1939 | } | |
1940 | ||
1941 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1942 | unsigned long addr, unsigned long end, | |
1943 | pte_fn_t fn, void *data) | |
1944 | { | |
1945 | pmd_t *pmd; | |
1946 | unsigned long next; | |
1947 | int err; | |
1948 | ||
ceb86879 AK |
1949 | BUG_ON(pud_huge(*pud)); |
1950 | ||
aee16b3c JF |
1951 | pmd = pmd_alloc(mm, pud, addr); |
1952 | if (!pmd) | |
1953 | return -ENOMEM; | |
1954 | do { | |
1955 | next = pmd_addr_end(addr, end); | |
1956 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
1957 | if (err) | |
1958 | break; | |
1959 | } while (pmd++, addr = next, addr != end); | |
1960 | return err; | |
1961 | } | |
1962 | ||
1963 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1964 | unsigned long addr, unsigned long end, | |
1965 | pte_fn_t fn, void *data) | |
1966 | { | |
1967 | pud_t *pud; | |
1968 | unsigned long next; | |
1969 | int err; | |
1970 | ||
1971 | pud = pud_alloc(mm, pgd, addr); | |
1972 | if (!pud) | |
1973 | return -ENOMEM; | |
1974 | do { | |
1975 | next = pud_addr_end(addr, end); | |
1976 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
1977 | if (err) | |
1978 | break; | |
1979 | } while (pud++, addr = next, addr != end); | |
1980 | return err; | |
1981 | } | |
1982 | ||
1983 | /* | |
1984 | * Scan a region of virtual memory, filling in page tables as necessary | |
1985 | * and calling a provided function on each leaf page table. | |
1986 | */ | |
1987 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
1988 | unsigned long size, pte_fn_t fn, void *data) | |
1989 | { | |
1990 | pgd_t *pgd; | |
1991 | unsigned long next; | |
cddb8a5c | 1992 | unsigned long start = addr, end = addr + size; |
aee16b3c JF |
1993 | int err; |
1994 | ||
1995 | BUG_ON(addr >= end); | |
cddb8a5c | 1996 | mmu_notifier_invalidate_range_start(mm, start, end); |
aee16b3c JF |
1997 | pgd = pgd_offset(mm, addr); |
1998 | do { | |
1999 | next = pgd_addr_end(addr, end); | |
2000 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); | |
2001 | if (err) | |
2002 | break; | |
2003 | } while (pgd++, addr = next, addr != end); | |
cddb8a5c | 2004 | mmu_notifier_invalidate_range_end(mm, start, end); |
aee16b3c JF |
2005 | return err; |
2006 | } | |
2007 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
2008 | ||
8f4e2101 HD |
2009 | /* |
2010 | * handle_pte_fault chooses page fault handler according to an entry | |
2011 | * which was read non-atomically. Before making any commitment, on | |
2012 | * those architectures or configurations (e.g. i386 with PAE) which | |
2013 | * might give a mix of unmatched parts, do_swap_page and do_file_page | |
2014 | * must check under lock before unmapping the pte and proceeding | |
2015 | * (but do_wp_page is only called after already making such a check; | |
2016 | * and do_anonymous_page and do_no_page can safely check later on). | |
2017 | */ | |
4c21e2f2 | 2018 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2019 | pte_t *page_table, pte_t orig_pte) |
2020 | { | |
2021 | int same = 1; | |
2022 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
2023 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
2024 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2025 | spin_lock(ptl); | |
8f4e2101 | 2026 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2027 | spin_unlock(ptl); |
8f4e2101 HD |
2028 | } |
2029 | #endif | |
2030 | pte_unmap(page_table); | |
2031 | return same; | |
2032 | } | |
2033 | ||
1da177e4 LT |
2034 | /* |
2035 | * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when | |
2036 | * servicing faults for write access. In the normal case, do always want | |
2037 | * pte_mkwrite. But get_user_pages can cause write faults for mappings | |
2038 | * that do not have writing enabled, when used by access_process_vm. | |
2039 | */ | |
2040 | static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma) | |
2041 | { | |
2042 | if (likely(vma->vm_flags & VM_WRITE)) | |
2043 | pte = pte_mkwrite(pte); | |
2044 | return pte; | |
2045 | } | |
2046 | ||
9de455b2 | 2047 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e LT |
2048 | { |
2049 | /* | |
2050 | * If the source page was a PFN mapping, we don't have | |
2051 | * a "struct page" for it. We do a best-effort copy by | |
2052 | * just copying from the original user address. If that | |
2053 | * fails, we just zero-fill it. Live with it. | |
2054 | */ | |
2055 | if (unlikely(!src)) { | |
2056 | void *kaddr = kmap_atomic(dst, KM_USER0); | |
5d2a2dbb LT |
2057 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
2058 | ||
2059 | /* | |
2060 | * This really shouldn't fail, because the page is there | |
2061 | * in the page tables. But it might just be unreadable, | |
2062 | * in which case we just give up and fill the result with | |
2063 | * zeroes. | |
2064 | */ | |
2065 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
6aab341e LT |
2066 | memset(kaddr, 0, PAGE_SIZE); |
2067 | kunmap_atomic(kaddr, KM_USER0); | |
c4ec7b0d | 2068 | flush_dcache_page(dst); |
0ed361de NP |
2069 | } else |
2070 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
2071 | } |
2072 | ||
1da177e4 LT |
2073 | /* |
2074 | * This routine handles present pages, when users try to write | |
2075 | * to a shared page. It is done by copying the page to a new address | |
2076 | * and decrementing the shared-page counter for the old page. | |
2077 | * | |
1da177e4 LT |
2078 | * Note that this routine assumes that the protection checks have been |
2079 | * done by the caller (the low-level page fault routine in most cases). | |
2080 | * Thus we can safely just mark it writable once we've done any necessary | |
2081 | * COW. | |
2082 | * | |
2083 | * We also mark the page dirty at this point even though the page will | |
2084 | * change only once the write actually happens. This avoids a few races, | |
2085 | * and potentially makes it more efficient. | |
2086 | * | |
8f4e2101 HD |
2087 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2088 | * but allow concurrent faults), with pte both mapped and locked. | |
2089 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2090 | */ |
65500d23 HD |
2091 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2092 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
8f4e2101 | 2093 | spinlock_t *ptl, pte_t orig_pte) |
1da177e4 | 2094 | { |
e5bbe4df | 2095 | struct page *old_page, *new_page; |
1da177e4 | 2096 | pte_t entry; |
83c54070 | 2097 | int reuse = 0, ret = 0; |
a200ee18 | 2098 | int page_mkwrite = 0; |
d08b3851 | 2099 | struct page *dirty_page = NULL; |
1da177e4 | 2100 | |
6aab341e | 2101 | old_page = vm_normal_page(vma, address, orig_pte); |
251b97f5 PZ |
2102 | if (!old_page) { |
2103 | /* | |
2104 | * VM_MIXEDMAP !pfn_valid() case | |
2105 | * | |
2106 | * We should not cow pages in a shared writeable mapping. | |
2107 | * Just mark the pages writable as we can't do any dirty | |
2108 | * accounting on raw pfn maps. | |
2109 | */ | |
2110 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2111 | (VM_WRITE|VM_SHARED)) | |
2112 | goto reuse; | |
6aab341e | 2113 | goto gotten; |
251b97f5 | 2114 | } |
1da177e4 | 2115 | |
d08b3851 | 2116 | /* |
ee6a6457 PZ |
2117 | * Take out anonymous pages first, anonymous shared vmas are |
2118 | * not dirty accountable. | |
d08b3851 | 2119 | */ |
9a840895 | 2120 | if (PageAnon(old_page) && !PageKsm(old_page)) { |
ab967d86 HD |
2121 | if (!trylock_page(old_page)) { |
2122 | page_cache_get(old_page); | |
2123 | pte_unmap_unlock(page_table, ptl); | |
2124 | lock_page(old_page); | |
2125 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2126 | &ptl); | |
2127 | if (!pte_same(*page_table, orig_pte)) { | |
2128 | unlock_page(old_page); | |
2129 | page_cache_release(old_page); | |
2130 | goto unlock; | |
2131 | } | |
2132 | page_cache_release(old_page); | |
ee6a6457 | 2133 | } |
7b1fe597 | 2134 | reuse = reuse_swap_page(old_page); |
c44b6743 RR |
2135 | if (reuse) |
2136 | /* | |
2137 | * The page is all ours. Move it to our anon_vma so | |
2138 | * the rmap code will not search our parent or siblings. | |
2139 | * Protected against the rmap code by the page lock. | |
2140 | */ | |
2141 | page_move_anon_rmap(old_page, vma, address); | |
ab967d86 | 2142 | unlock_page(old_page); |
ee6a6457 | 2143 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2144 | (VM_WRITE|VM_SHARED))) { |
ee6a6457 PZ |
2145 | /* |
2146 | * Only catch write-faults on shared writable pages, | |
2147 | * read-only shared pages can get COWed by | |
2148 | * get_user_pages(.write=1, .force=1). | |
2149 | */ | |
9637a5ef | 2150 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
c2ec175c NP |
2151 | struct vm_fault vmf; |
2152 | int tmp; | |
2153 | ||
2154 | vmf.virtual_address = (void __user *)(address & | |
2155 | PAGE_MASK); | |
2156 | vmf.pgoff = old_page->index; | |
2157 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; | |
2158 | vmf.page = old_page; | |
2159 | ||
9637a5ef DH |
2160 | /* |
2161 | * Notify the address space that the page is about to | |
2162 | * become writable so that it can prohibit this or wait | |
2163 | * for the page to get into an appropriate state. | |
2164 | * | |
2165 | * We do this without the lock held, so that it can | |
2166 | * sleep if it needs to. | |
2167 | */ | |
2168 | page_cache_get(old_page); | |
2169 | pte_unmap_unlock(page_table, ptl); | |
2170 | ||
c2ec175c NP |
2171 | tmp = vma->vm_ops->page_mkwrite(vma, &vmf); |
2172 | if (unlikely(tmp & | |
2173 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { | |
2174 | ret = tmp; | |
9637a5ef | 2175 | goto unwritable_page; |
c2ec175c | 2176 | } |
b827e496 NP |
2177 | if (unlikely(!(tmp & VM_FAULT_LOCKED))) { |
2178 | lock_page(old_page); | |
2179 | if (!old_page->mapping) { | |
2180 | ret = 0; /* retry the fault */ | |
2181 | unlock_page(old_page); | |
2182 | goto unwritable_page; | |
2183 | } | |
2184 | } else | |
2185 | VM_BUG_ON(!PageLocked(old_page)); | |
9637a5ef | 2186 | |
9637a5ef DH |
2187 | /* |
2188 | * Since we dropped the lock we need to revalidate | |
2189 | * the PTE as someone else may have changed it. If | |
2190 | * they did, we just return, as we can count on the | |
2191 | * MMU to tell us if they didn't also make it writable. | |
2192 | */ | |
2193 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2194 | &ptl); | |
b827e496 NP |
2195 | if (!pte_same(*page_table, orig_pte)) { |
2196 | unlock_page(old_page); | |
2197 | page_cache_release(old_page); | |
9637a5ef | 2198 | goto unlock; |
b827e496 | 2199 | } |
a200ee18 PZ |
2200 | |
2201 | page_mkwrite = 1; | |
1da177e4 | 2202 | } |
d08b3851 PZ |
2203 | dirty_page = old_page; |
2204 | get_page(dirty_page); | |
9637a5ef | 2205 | reuse = 1; |
9637a5ef DH |
2206 | } |
2207 | ||
2208 | if (reuse) { | |
251b97f5 | 2209 | reuse: |
9637a5ef DH |
2210 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
2211 | entry = pte_mkyoung(orig_pte); | |
2212 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
954ffcb3 | 2213 | if (ptep_set_access_flags(vma, address, page_table, entry,1)) |
4b3073e1 | 2214 | update_mmu_cache(vma, address, page_table); |
9637a5ef DH |
2215 | ret |= VM_FAULT_WRITE; |
2216 | goto unlock; | |
1da177e4 | 2217 | } |
1da177e4 LT |
2218 | |
2219 | /* | |
2220 | * Ok, we need to copy. Oh, well.. | |
2221 | */ | |
b5810039 | 2222 | page_cache_get(old_page); |
920fc356 | 2223 | gotten: |
8f4e2101 | 2224 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2225 | |
2226 | if (unlikely(anon_vma_prepare(vma))) | |
65500d23 | 2227 | goto oom; |
a13ea5b7 | 2228 | |
62eede62 | 2229 | if (is_zero_pfn(pte_pfn(orig_pte))) { |
a13ea5b7 HD |
2230 | new_page = alloc_zeroed_user_highpage_movable(vma, address); |
2231 | if (!new_page) | |
2232 | goto oom; | |
2233 | } else { | |
2234 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
2235 | if (!new_page) | |
2236 | goto oom; | |
2237 | cow_user_page(new_page, old_page, address, vma); | |
2238 | } | |
2239 | __SetPageUptodate(new_page); | |
2240 | ||
b291f000 NP |
2241 | /* |
2242 | * Don't let another task, with possibly unlocked vma, | |
2243 | * keep the mlocked page. | |
2244 | */ | |
ab92661d | 2245 | if ((vma->vm_flags & VM_LOCKED) && old_page) { |
b291f000 NP |
2246 | lock_page(old_page); /* for LRU manipulation */ |
2247 | clear_page_mlock(old_page); | |
2248 | unlock_page(old_page); | |
2249 | } | |
65500d23 | 2250 | |
2c26fdd7 | 2251 | if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL)) |
8a9f3ccd BS |
2252 | goto oom_free_new; |
2253 | ||
1da177e4 LT |
2254 | /* |
2255 | * Re-check the pte - we dropped the lock | |
2256 | */ | |
8f4e2101 | 2257 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
65500d23 | 2258 | if (likely(pte_same(*page_table, orig_pte))) { |
920fc356 | 2259 | if (old_page) { |
920fc356 | 2260 | if (!PageAnon(old_page)) { |
34e55232 KH |
2261 | dec_mm_counter_fast(mm, MM_FILEPAGES); |
2262 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
920fc356 HD |
2263 | } |
2264 | } else | |
34e55232 | 2265 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
eca35133 | 2266 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
65500d23 HD |
2267 | entry = mk_pte(new_page, vma->vm_page_prot); |
2268 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
4ce072f1 SS |
2269 | /* |
2270 | * Clear the pte entry and flush it first, before updating the | |
2271 | * pte with the new entry. This will avoid a race condition | |
2272 | * seen in the presence of one thread doing SMC and another | |
2273 | * thread doing COW. | |
2274 | */ | |
828502d3 | 2275 | ptep_clear_flush(vma, address, page_table); |
9617d95e | 2276 | page_add_new_anon_rmap(new_page, vma, address); |
828502d3 IE |
2277 | /* |
2278 | * We call the notify macro here because, when using secondary | |
2279 | * mmu page tables (such as kvm shadow page tables), we want the | |
2280 | * new page to be mapped directly into the secondary page table. | |
2281 | */ | |
2282 | set_pte_at_notify(mm, address, page_table, entry); | |
4b3073e1 | 2283 | update_mmu_cache(vma, address, page_table); |
945754a1 NP |
2284 | if (old_page) { |
2285 | /* | |
2286 | * Only after switching the pte to the new page may | |
2287 | * we remove the mapcount here. Otherwise another | |
2288 | * process may come and find the rmap count decremented | |
2289 | * before the pte is switched to the new page, and | |
2290 | * "reuse" the old page writing into it while our pte | |
2291 | * here still points into it and can be read by other | |
2292 | * threads. | |
2293 | * | |
2294 | * The critical issue is to order this | |
2295 | * page_remove_rmap with the ptp_clear_flush above. | |
2296 | * Those stores are ordered by (if nothing else,) | |
2297 | * the barrier present in the atomic_add_negative | |
2298 | * in page_remove_rmap. | |
2299 | * | |
2300 | * Then the TLB flush in ptep_clear_flush ensures that | |
2301 | * no process can access the old page before the | |
2302 | * decremented mapcount is visible. And the old page | |
2303 | * cannot be reused until after the decremented | |
2304 | * mapcount is visible. So transitively, TLBs to | |
2305 | * old page will be flushed before it can be reused. | |
2306 | */ | |
edc315fd | 2307 | page_remove_rmap(old_page); |
945754a1 NP |
2308 | } |
2309 | ||
1da177e4 LT |
2310 | /* Free the old page.. */ |
2311 | new_page = old_page; | |
f33ea7f4 | 2312 | ret |= VM_FAULT_WRITE; |
8a9f3ccd BS |
2313 | } else |
2314 | mem_cgroup_uncharge_page(new_page); | |
2315 | ||
920fc356 HD |
2316 | if (new_page) |
2317 | page_cache_release(new_page); | |
2318 | if (old_page) | |
2319 | page_cache_release(old_page); | |
65500d23 | 2320 | unlock: |
8f4e2101 | 2321 | pte_unmap_unlock(page_table, ptl); |
d08b3851 | 2322 | if (dirty_page) { |
79352894 NP |
2323 | /* |
2324 | * Yes, Virginia, this is actually required to prevent a race | |
2325 | * with clear_page_dirty_for_io() from clearing the page dirty | |
2326 | * bit after it clear all dirty ptes, but before a racing | |
2327 | * do_wp_page installs a dirty pte. | |
2328 | * | |
2329 | * do_no_page is protected similarly. | |
2330 | */ | |
b827e496 NP |
2331 | if (!page_mkwrite) { |
2332 | wait_on_page_locked(dirty_page); | |
2333 | set_page_dirty_balance(dirty_page, page_mkwrite); | |
2334 | } | |
d08b3851 | 2335 | put_page(dirty_page); |
b827e496 NP |
2336 | if (page_mkwrite) { |
2337 | struct address_space *mapping = dirty_page->mapping; | |
2338 | ||
2339 | set_page_dirty(dirty_page); | |
2340 | unlock_page(dirty_page); | |
2341 | page_cache_release(dirty_page); | |
2342 | if (mapping) { | |
2343 | /* | |
2344 | * Some device drivers do not set page.mapping | |
2345 | * but still dirty their pages | |
2346 | */ | |
2347 | balance_dirty_pages_ratelimited(mapping); | |
2348 | } | |
2349 | } | |
2350 | ||
2351 | /* file_update_time outside page_lock */ | |
2352 | if (vma->vm_file) | |
2353 | file_update_time(vma->vm_file); | |
d08b3851 | 2354 | } |
f33ea7f4 | 2355 | return ret; |
8a9f3ccd | 2356 | oom_free_new: |
6dbf6d3b | 2357 | page_cache_release(new_page); |
65500d23 | 2358 | oom: |
b827e496 NP |
2359 | if (old_page) { |
2360 | if (page_mkwrite) { | |
2361 | unlock_page(old_page); | |
2362 | page_cache_release(old_page); | |
2363 | } | |
920fc356 | 2364 | page_cache_release(old_page); |
b827e496 | 2365 | } |
1da177e4 | 2366 | return VM_FAULT_OOM; |
9637a5ef DH |
2367 | |
2368 | unwritable_page: | |
2369 | page_cache_release(old_page); | |
c2ec175c | 2370 | return ret; |
1da177e4 LT |
2371 | } |
2372 | ||
2373 | /* | |
2374 | * Helper functions for unmap_mapping_range(). | |
2375 | * | |
2376 | * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __ | |
2377 | * | |
2378 | * We have to restart searching the prio_tree whenever we drop the lock, | |
2379 | * since the iterator is only valid while the lock is held, and anyway | |
2380 | * a later vma might be split and reinserted earlier while lock dropped. | |
2381 | * | |
2382 | * The list of nonlinear vmas could be handled more efficiently, using | |
2383 | * a placeholder, but handle it in the same way until a need is shown. | |
2384 | * It is important to search the prio_tree before nonlinear list: a vma | |
2385 | * may become nonlinear and be shifted from prio_tree to nonlinear list | |
2386 | * while the lock is dropped; but never shifted from list to prio_tree. | |
2387 | * | |
2388 | * In order to make forward progress despite restarting the search, | |
2389 | * vm_truncate_count is used to mark a vma as now dealt with, so we can | |
2390 | * quickly skip it next time around. Since the prio_tree search only | |
2391 | * shows us those vmas affected by unmapping the range in question, we | |
2392 | * can't efficiently keep all vmas in step with mapping->truncate_count: | |
2393 | * so instead reset them all whenever it wraps back to 0 (then go to 1). | |
2394 | * mapping->truncate_count and vma->vm_truncate_count are protected by | |
2395 | * i_mmap_lock. | |
2396 | * | |
2397 | * In order to make forward progress despite repeatedly restarting some | |
ee39b37b | 2398 | * large vma, note the restart_addr from unmap_vmas when it breaks out: |
1da177e4 LT |
2399 | * and restart from that address when we reach that vma again. It might |
2400 | * have been split or merged, shrunk or extended, but never shifted: so | |
2401 | * restart_addr remains valid so long as it remains in the vma's range. | |
2402 | * unmap_mapping_range forces truncate_count to leap over page-aligned | |
2403 | * values so we can save vma's restart_addr in its truncate_count field. | |
2404 | */ | |
2405 | #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK)) | |
2406 | ||
2407 | static void reset_vma_truncate_counts(struct address_space *mapping) | |
2408 | { | |
2409 | struct vm_area_struct *vma; | |
2410 | struct prio_tree_iter iter; | |
2411 | ||
2412 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX) | |
2413 | vma->vm_truncate_count = 0; | |
2414 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) | |
2415 | vma->vm_truncate_count = 0; | |
2416 | } | |
2417 | ||
2418 | static int unmap_mapping_range_vma(struct vm_area_struct *vma, | |
2419 | unsigned long start_addr, unsigned long end_addr, | |
2420 | struct zap_details *details) | |
2421 | { | |
2422 | unsigned long restart_addr; | |
2423 | int need_break; | |
2424 | ||
d00806b1 NP |
2425 | /* |
2426 | * files that support invalidating or truncating portions of the | |
d0217ac0 | 2427 | * file from under mmaped areas must have their ->fault function |
83c54070 NP |
2428 | * return a locked page (and set VM_FAULT_LOCKED in the return). |
2429 | * This provides synchronisation against concurrent unmapping here. | |
d00806b1 | 2430 | */ |
d00806b1 | 2431 | |
1da177e4 LT |
2432 | again: |
2433 | restart_addr = vma->vm_truncate_count; | |
2434 | if (is_restart_addr(restart_addr) && start_addr < restart_addr) { | |
2435 | start_addr = restart_addr; | |
2436 | if (start_addr >= end_addr) { | |
2437 | /* Top of vma has been split off since last time */ | |
2438 | vma->vm_truncate_count = details->truncate_count; | |
2439 | return 0; | |
2440 | } | |
2441 | } | |
2442 | ||
ee39b37b HD |
2443 | restart_addr = zap_page_range(vma, start_addr, |
2444 | end_addr - start_addr, details); | |
95c354fe | 2445 | need_break = need_resched() || spin_needbreak(details->i_mmap_lock); |
1da177e4 | 2446 | |
ee39b37b | 2447 | if (restart_addr >= end_addr) { |
1da177e4 LT |
2448 | /* We have now completed this vma: mark it so */ |
2449 | vma->vm_truncate_count = details->truncate_count; | |
2450 | if (!need_break) | |
2451 | return 0; | |
2452 | } else { | |
2453 | /* Note restart_addr in vma's truncate_count field */ | |
ee39b37b | 2454 | vma->vm_truncate_count = restart_addr; |
1da177e4 LT |
2455 | if (!need_break) |
2456 | goto again; | |
2457 | } | |
2458 | ||
2459 | spin_unlock(details->i_mmap_lock); | |
2460 | cond_resched(); | |
2461 | spin_lock(details->i_mmap_lock); | |
2462 | return -EINTR; | |
2463 | } | |
2464 | ||
2465 | static inline void unmap_mapping_range_tree(struct prio_tree_root *root, | |
2466 | struct zap_details *details) | |
2467 | { | |
2468 | struct vm_area_struct *vma; | |
2469 | struct prio_tree_iter iter; | |
2470 | pgoff_t vba, vea, zba, zea; | |
2471 | ||
2472 | restart: | |
2473 | vma_prio_tree_foreach(vma, &iter, root, | |
2474 | details->first_index, details->last_index) { | |
2475 | /* Skip quickly over those we have already dealt with */ | |
2476 | if (vma->vm_truncate_count == details->truncate_count) | |
2477 | continue; | |
2478 | ||
2479 | vba = vma->vm_pgoff; | |
2480 | vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; | |
2481 | /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ | |
2482 | zba = details->first_index; | |
2483 | if (zba < vba) | |
2484 | zba = vba; | |
2485 | zea = details->last_index; | |
2486 | if (zea > vea) | |
2487 | zea = vea; | |
2488 | ||
2489 | if (unmap_mapping_range_vma(vma, | |
2490 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, | |
2491 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
2492 | details) < 0) | |
2493 | goto restart; | |
2494 | } | |
2495 | } | |
2496 | ||
2497 | static inline void unmap_mapping_range_list(struct list_head *head, | |
2498 | struct zap_details *details) | |
2499 | { | |
2500 | struct vm_area_struct *vma; | |
2501 | ||
2502 | /* | |
2503 | * In nonlinear VMAs there is no correspondence between virtual address | |
2504 | * offset and file offset. So we must perform an exhaustive search | |
2505 | * across *all* the pages in each nonlinear VMA, not just the pages | |
2506 | * whose virtual address lies outside the file truncation point. | |
2507 | */ | |
2508 | restart: | |
2509 | list_for_each_entry(vma, head, shared.vm_set.list) { | |
2510 | /* Skip quickly over those we have already dealt with */ | |
2511 | if (vma->vm_truncate_count == details->truncate_count) | |
2512 | continue; | |
2513 | details->nonlinear_vma = vma; | |
2514 | if (unmap_mapping_range_vma(vma, vma->vm_start, | |
2515 | vma->vm_end, details) < 0) | |
2516 | goto restart; | |
2517 | } | |
2518 | } | |
2519 | ||
2520 | /** | |
72fd4a35 | 2521 | * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file. |
3d41088f | 2522 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2523 | * @holebegin: byte in first page to unmap, relative to the start of |
2524 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2525 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2526 | * must keep the partial page. In contrast, we must get rid of |
2527 | * partial pages. | |
2528 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2529 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2530 | * end of the file. | |
2531 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2532 | * but 0 when invalidating pagecache, don't throw away private data. | |
2533 | */ | |
2534 | void unmap_mapping_range(struct address_space *mapping, | |
2535 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2536 | { | |
2537 | struct zap_details details; | |
2538 | pgoff_t hba = holebegin >> PAGE_SHIFT; | |
2539 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2540 | ||
2541 | /* Check for overflow. */ | |
2542 | if (sizeof(holelen) > sizeof(hlen)) { | |
2543 | long long holeend = | |
2544 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2545 | if (holeend & ~(long long)ULONG_MAX) | |
2546 | hlen = ULONG_MAX - hba + 1; | |
2547 | } | |
2548 | ||
2549 | details.check_mapping = even_cows? NULL: mapping; | |
2550 | details.nonlinear_vma = NULL; | |
2551 | details.first_index = hba; | |
2552 | details.last_index = hba + hlen - 1; | |
2553 | if (details.last_index < details.first_index) | |
2554 | details.last_index = ULONG_MAX; | |
2555 | details.i_mmap_lock = &mapping->i_mmap_lock; | |
2556 | ||
2557 | spin_lock(&mapping->i_mmap_lock); | |
2558 | ||
d00806b1 | 2559 | /* Protect against endless unmapping loops */ |
1da177e4 | 2560 | mapping->truncate_count++; |
1da177e4 LT |
2561 | if (unlikely(is_restart_addr(mapping->truncate_count))) { |
2562 | if (mapping->truncate_count == 0) | |
2563 | reset_vma_truncate_counts(mapping); | |
2564 | mapping->truncate_count++; | |
2565 | } | |
2566 | details.truncate_count = mapping->truncate_count; | |
2567 | ||
2568 | if (unlikely(!prio_tree_empty(&mapping->i_mmap))) | |
2569 | unmap_mapping_range_tree(&mapping->i_mmap, &details); | |
2570 | if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) | |
2571 | unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); | |
2572 | spin_unlock(&mapping->i_mmap_lock); | |
2573 | } | |
2574 | EXPORT_SYMBOL(unmap_mapping_range); | |
2575 | ||
f6b3ec23 BP |
2576 | int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end) |
2577 | { | |
2578 | struct address_space *mapping = inode->i_mapping; | |
2579 | ||
2580 | /* | |
2581 | * If the underlying filesystem is not going to provide | |
2582 | * a way to truncate a range of blocks (punch a hole) - | |
2583 | * we should return failure right now. | |
2584 | */ | |
acfa4380 | 2585 | if (!inode->i_op->truncate_range) |
f6b3ec23 BP |
2586 | return -ENOSYS; |
2587 | ||
1b1dcc1b | 2588 | mutex_lock(&inode->i_mutex); |
f6b3ec23 BP |
2589 | down_write(&inode->i_alloc_sem); |
2590 | unmap_mapping_range(mapping, offset, (end - offset), 1); | |
2591 | truncate_inode_pages_range(mapping, offset, end); | |
d00806b1 | 2592 | unmap_mapping_range(mapping, offset, (end - offset), 1); |
f6b3ec23 BP |
2593 | inode->i_op->truncate_range(inode, offset, end); |
2594 | up_write(&inode->i_alloc_sem); | |
1b1dcc1b | 2595 | mutex_unlock(&inode->i_mutex); |
f6b3ec23 BP |
2596 | |
2597 | return 0; | |
2598 | } | |
f6b3ec23 | 2599 | |
1da177e4 | 2600 | /* |
8f4e2101 HD |
2601 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2602 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2603 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2604 | */ |
65500d23 HD |
2605 | static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2606 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2607 | unsigned int flags, pte_t orig_pte) |
1da177e4 | 2608 | { |
8f4e2101 | 2609 | spinlock_t *ptl; |
1da177e4 | 2610 | struct page *page; |
65500d23 | 2611 | swp_entry_t entry; |
1da177e4 | 2612 | pte_t pte; |
7a81b88c | 2613 | struct mem_cgroup *ptr = NULL; |
83c54070 | 2614 | int ret = 0; |
1da177e4 | 2615 | |
4c21e2f2 | 2616 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
8f4e2101 | 2617 | goto out; |
65500d23 HD |
2618 | |
2619 | entry = pte_to_swp_entry(orig_pte); | |
d1737fdb AK |
2620 | if (unlikely(non_swap_entry(entry))) { |
2621 | if (is_migration_entry(entry)) { | |
2622 | migration_entry_wait(mm, pmd, address); | |
2623 | } else if (is_hwpoison_entry(entry)) { | |
2624 | ret = VM_FAULT_HWPOISON; | |
2625 | } else { | |
2626 | print_bad_pte(vma, address, orig_pte, NULL); | |
d99be1a8 | 2627 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2628 | } |
0697212a CL |
2629 | goto out; |
2630 | } | |
0ff92245 | 2631 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
2632 | page = lookup_swap_cache(entry); |
2633 | if (!page) { | |
a5c9b696 | 2634 | grab_swap_token(mm); /* Contend for token _before_ read-in */ |
02098fea HD |
2635 | page = swapin_readahead(entry, |
2636 | GFP_HIGHUSER_MOVABLE, vma, address); | |
1da177e4 LT |
2637 | if (!page) { |
2638 | /* | |
8f4e2101 HD |
2639 | * Back out if somebody else faulted in this pte |
2640 | * while we released the pte lock. | |
1da177e4 | 2641 | */ |
8f4e2101 | 2642 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
2643 | if (likely(pte_same(*page_table, orig_pte))) |
2644 | ret = VM_FAULT_OOM; | |
0ff92245 | 2645 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2646 | goto unlock; |
1da177e4 LT |
2647 | } |
2648 | ||
2649 | /* Had to read the page from swap area: Major fault */ | |
2650 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2651 | count_vm_event(PGMAJFAULT); |
d1737fdb | 2652 | } else if (PageHWPoison(page)) { |
71f72525 WF |
2653 | /* |
2654 | * hwpoisoned dirty swapcache pages are kept for killing | |
2655 | * owner processes (which may be unknown at hwpoison time) | |
2656 | */ | |
d1737fdb AK |
2657 | ret = VM_FAULT_HWPOISON; |
2658 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
4779cb31 | 2659 | goto out_release; |
1da177e4 LT |
2660 | } |
2661 | ||
073e587e KH |
2662 | lock_page(page); |
2663 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
2664 | ||
5ad64688 HD |
2665 | page = ksm_might_need_to_copy(page, vma, address); |
2666 | if (!page) { | |
2667 | ret = VM_FAULT_OOM; | |
2668 | goto out; | |
2669 | } | |
2670 | ||
2c26fdd7 | 2671 | if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) { |
8a9f3ccd | 2672 | ret = VM_FAULT_OOM; |
bc43f75c | 2673 | goto out_page; |
8a9f3ccd BS |
2674 | } |
2675 | ||
1da177e4 | 2676 | /* |
8f4e2101 | 2677 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2678 | */ |
8f4e2101 | 2679 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
9e9bef07 | 2680 | if (unlikely(!pte_same(*page_table, orig_pte))) |
b8107480 | 2681 | goto out_nomap; |
b8107480 KK |
2682 | |
2683 | if (unlikely(!PageUptodate(page))) { | |
2684 | ret = VM_FAULT_SIGBUS; | |
2685 | goto out_nomap; | |
1da177e4 LT |
2686 | } |
2687 | ||
8c7c6e34 KH |
2688 | /* |
2689 | * The page isn't present yet, go ahead with the fault. | |
2690 | * | |
2691 | * Be careful about the sequence of operations here. | |
2692 | * To get its accounting right, reuse_swap_page() must be called | |
2693 | * while the page is counted on swap but not yet in mapcount i.e. | |
2694 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
2695 | * must be called after the swap_free(), or it will never succeed. | |
03f3c433 KH |
2696 | * Because delete_from_swap_page() may be called by reuse_swap_page(), |
2697 | * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry | |
2698 | * in page->private. In this case, a record in swap_cgroup is silently | |
2699 | * discarded at swap_free(). | |
8c7c6e34 | 2700 | */ |
1da177e4 | 2701 | |
34e55232 | 2702 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
b084d435 | 2703 | dec_mm_counter_fast(mm, MM_SWAPENTS); |
1da177e4 | 2704 | pte = mk_pte(page, vma->vm_page_prot); |
30c9f3a9 | 2705 | if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { |
1da177e4 | 2706 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
30c9f3a9 | 2707 | flags &= ~FAULT_FLAG_WRITE; |
1da177e4 | 2708 | } |
1da177e4 LT |
2709 | flush_icache_page(vma, page); |
2710 | set_pte_at(mm, address, page_table, pte); | |
2711 | page_add_anon_rmap(page, vma, address); | |
03f3c433 KH |
2712 | /* It's better to call commit-charge after rmap is established */ |
2713 | mem_cgroup_commit_charge_swapin(page, ptr); | |
1da177e4 | 2714 | |
c475a8ab | 2715 | swap_free(entry); |
b291f000 | 2716 | if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) |
a2c43eed | 2717 | try_to_free_swap(page); |
c475a8ab HD |
2718 | unlock_page(page); |
2719 | ||
30c9f3a9 | 2720 | if (flags & FAULT_FLAG_WRITE) { |
61469f1d HD |
2721 | ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); |
2722 | if (ret & VM_FAULT_ERROR) | |
2723 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
2724 | goto out; |
2725 | } | |
2726 | ||
2727 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 2728 | update_mmu_cache(vma, address, page_table); |
65500d23 | 2729 | unlock: |
8f4e2101 | 2730 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2731 | out: |
2732 | return ret; | |
b8107480 | 2733 | out_nomap: |
7a81b88c | 2734 | mem_cgroup_cancel_charge_swapin(ptr); |
8f4e2101 | 2735 | pte_unmap_unlock(page_table, ptl); |
bc43f75c | 2736 | out_page: |
b8107480 | 2737 | unlock_page(page); |
4779cb31 | 2738 | out_release: |
b8107480 | 2739 | page_cache_release(page); |
65500d23 | 2740 | return ret; |
1da177e4 LT |
2741 | } |
2742 | ||
2743 | /* | |
8f4e2101 HD |
2744 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2745 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2746 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2747 | */ |
65500d23 HD |
2748 | static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2749 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2750 | unsigned int flags) |
1da177e4 | 2751 | { |
8f4e2101 HD |
2752 | struct page *page; |
2753 | spinlock_t *ptl; | |
1da177e4 | 2754 | pte_t entry; |
1da177e4 | 2755 | |
62eede62 HD |
2756 | if (!(flags & FAULT_FLAG_WRITE)) { |
2757 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), | |
2758 | vma->vm_page_prot)); | |
a13ea5b7 HD |
2759 | ptl = pte_lockptr(mm, pmd); |
2760 | spin_lock(ptl); | |
2761 | if (!pte_none(*page_table)) | |
2762 | goto unlock; | |
2763 | goto setpte; | |
2764 | } | |
2765 | ||
557ed1fa NP |
2766 | /* Allocate our own private page. */ |
2767 | pte_unmap(page_table); | |
8f4e2101 | 2768 | |
557ed1fa NP |
2769 | if (unlikely(anon_vma_prepare(vma))) |
2770 | goto oom; | |
2771 | page = alloc_zeroed_user_highpage_movable(vma, address); | |
2772 | if (!page) | |
2773 | goto oom; | |
0ed361de | 2774 | __SetPageUptodate(page); |
8f4e2101 | 2775 | |
2c26fdd7 | 2776 | if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) |
8a9f3ccd BS |
2777 | goto oom_free_page; |
2778 | ||
557ed1fa | 2779 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
2780 | if (vma->vm_flags & VM_WRITE) |
2781 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 2782 | |
557ed1fa | 2783 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1c2fb7a4 | 2784 | if (!pte_none(*page_table)) |
557ed1fa | 2785 | goto release; |
9ba69294 | 2786 | |
34e55232 | 2787 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
557ed1fa | 2788 | page_add_new_anon_rmap(page, vma, address); |
a13ea5b7 | 2789 | setpte: |
65500d23 | 2790 | set_pte_at(mm, address, page_table, entry); |
1da177e4 LT |
2791 | |
2792 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 2793 | update_mmu_cache(vma, address, page_table); |
65500d23 | 2794 | unlock: |
8f4e2101 | 2795 | pte_unmap_unlock(page_table, ptl); |
83c54070 | 2796 | return 0; |
8f4e2101 | 2797 | release: |
8a9f3ccd | 2798 | mem_cgroup_uncharge_page(page); |
8f4e2101 HD |
2799 | page_cache_release(page); |
2800 | goto unlock; | |
8a9f3ccd | 2801 | oom_free_page: |
6dbf6d3b | 2802 | page_cache_release(page); |
65500d23 | 2803 | oom: |
1da177e4 LT |
2804 | return VM_FAULT_OOM; |
2805 | } | |
2806 | ||
2807 | /* | |
54cb8821 | 2808 | * __do_fault() tries to create a new page mapping. It aggressively |
1da177e4 | 2809 | * tries to share with existing pages, but makes a separate copy if |
54cb8821 NP |
2810 | * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid |
2811 | * the next page fault. | |
1da177e4 LT |
2812 | * |
2813 | * As this is called only for pages that do not currently exist, we | |
2814 | * do not need to flush old virtual caches or the TLB. | |
2815 | * | |
8f4e2101 | 2816 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
16abfa08 | 2817 | * but allow concurrent faults), and pte neither mapped nor locked. |
8f4e2101 | 2818 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
1da177e4 | 2819 | */ |
54cb8821 | 2820 | static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
16abfa08 | 2821 | unsigned long address, pmd_t *pmd, |
54cb8821 | 2822 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) |
1da177e4 | 2823 | { |
16abfa08 | 2824 | pte_t *page_table; |
8f4e2101 | 2825 | spinlock_t *ptl; |
d0217ac0 | 2826 | struct page *page; |
1da177e4 | 2827 | pte_t entry; |
1da177e4 | 2828 | int anon = 0; |
5b4e655e | 2829 | int charged = 0; |
d08b3851 | 2830 | struct page *dirty_page = NULL; |
d0217ac0 NP |
2831 | struct vm_fault vmf; |
2832 | int ret; | |
a200ee18 | 2833 | int page_mkwrite = 0; |
54cb8821 | 2834 | |
d0217ac0 NP |
2835 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); |
2836 | vmf.pgoff = pgoff; | |
2837 | vmf.flags = flags; | |
2838 | vmf.page = NULL; | |
1da177e4 | 2839 | |
3c18ddd1 NP |
2840 | ret = vma->vm_ops->fault(vma, &vmf); |
2841 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) | |
2842 | return ret; | |
1da177e4 | 2843 | |
a3b947ea AK |
2844 | if (unlikely(PageHWPoison(vmf.page))) { |
2845 | if (ret & VM_FAULT_LOCKED) | |
2846 | unlock_page(vmf.page); | |
2847 | return VM_FAULT_HWPOISON; | |
2848 | } | |
2849 | ||
d00806b1 | 2850 | /* |
d0217ac0 | 2851 | * For consistency in subsequent calls, make the faulted page always |
d00806b1 NP |
2852 | * locked. |
2853 | */ | |
83c54070 | 2854 | if (unlikely(!(ret & VM_FAULT_LOCKED))) |
d0217ac0 | 2855 | lock_page(vmf.page); |
54cb8821 | 2856 | else |
d0217ac0 | 2857 | VM_BUG_ON(!PageLocked(vmf.page)); |
d00806b1 | 2858 | |
1da177e4 LT |
2859 | /* |
2860 | * Should we do an early C-O-W break? | |
2861 | */ | |
d0217ac0 | 2862 | page = vmf.page; |
54cb8821 | 2863 | if (flags & FAULT_FLAG_WRITE) { |
9637a5ef | 2864 | if (!(vma->vm_flags & VM_SHARED)) { |
54cb8821 | 2865 | anon = 1; |
d00806b1 | 2866 | if (unlikely(anon_vma_prepare(vma))) { |
d0217ac0 | 2867 | ret = VM_FAULT_OOM; |
54cb8821 | 2868 | goto out; |
d00806b1 | 2869 | } |
83c54070 NP |
2870 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, |
2871 | vma, address); | |
d00806b1 | 2872 | if (!page) { |
d0217ac0 | 2873 | ret = VM_FAULT_OOM; |
54cb8821 | 2874 | goto out; |
d00806b1 | 2875 | } |
2c26fdd7 | 2876 | if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) { |
5b4e655e KH |
2877 | ret = VM_FAULT_OOM; |
2878 | page_cache_release(page); | |
2879 | goto out; | |
2880 | } | |
2881 | charged = 1; | |
b291f000 NP |
2882 | /* |
2883 | * Don't let another task, with possibly unlocked vma, | |
2884 | * keep the mlocked page. | |
2885 | */ | |
2886 | if (vma->vm_flags & VM_LOCKED) | |
2887 | clear_page_mlock(vmf.page); | |
d0217ac0 | 2888 | copy_user_highpage(page, vmf.page, address, vma); |
0ed361de | 2889 | __SetPageUptodate(page); |
9637a5ef | 2890 | } else { |
54cb8821 NP |
2891 | /* |
2892 | * If the page will be shareable, see if the backing | |
9637a5ef | 2893 | * address space wants to know that the page is about |
54cb8821 NP |
2894 | * to become writable |
2895 | */ | |
69676147 | 2896 | if (vma->vm_ops->page_mkwrite) { |
c2ec175c NP |
2897 | int tmp; |
2898 | ||
69676147 | 2899 | unlock_page(page); |
b827e496 | 2900 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
c2ec175c NP |
2901 | tmp = vma->vm_ops->page_mkwrite(vma, &vmf); |
2902 | if (unlikely(tmp & | |
2903 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { | |
2904 | ret = tmp; | |
b827e496 | 2905 | goto unwritable_page; |
d0217ac0 | 2906 | } |
b827e496 NP |
2907 | if (unlikely(!(tmp & VM_FAULT_LOCKED))) { |
2908 | lock_page(page); | |
2909 | if (!page->mapping) { | |
2910 | ret = 0; /* retry the fault */ | |
2911 | unlock_page(page); | |
2912 | goto unwritable_page; | |
2913 | } | |
2914 | } else | |
2915 | VM_BUG_ON(!PageLocked(page)); | |
a200ee18 | 2916 | page_mkwrite = 1; |
9637a5ef DH |
2917 | } |
2918 | } | |
54cb8821 | 2919 | |
1da177e4 LT |
2920 | } |
2921 | ||
8f4e2101 | 2922 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
2923 | |
2924 | /* | |
2925 | * This silly early PAGE_DIRTY setting removes a race | |
2926 | * due to the bad i386 page protection. But it's valid | |
2927 | * for other architectures too. | |
2928 | * | |
30c9f3a9 | 2929 | * Note that if FAULT_FLAG_WRITE is set, we either now have |
1da177e4 LT |
2930 | * an exclusive copy of the page, or this is a shared mapping, |
2931 | * so we can make it writable and dirty to avoid having to | |
2932 | * handle that later. | |
2933 | */ | |
2934 | /* Only go through if we didn't race with anybody else... */ | |
1c2fb7a4 | 2935 | if (likely(pte_same(*page_table, orig_pte))) { |
d00806b1 NP |
2936 | flush_icache_page(vma, page); |
2937 | entry = mk_pte(page, vma->vm_page_prot); | |
54cb8821 | 2938 | if (flags & FAULT_FLAG_WRITE) |
1da177e4 | 2939 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
1da177e4 | 2940 | if (anon) { |
34e55232 | 2941 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
64d6519d | 2942 | page_add_new_anon_rmap(page, vma, address); |
f57e88a8 | 2943 | } else { |
34e55232 | 2944 | inc_mm_counter_fast(mm, MM_FILEPAGES); |
d00806b1 | 2945 | page_add_file_rmap(page); |
54cb8821 | 2946 | if (flags & FAULT_FLAG_WRITE) { |
d00806b1 | 2947 | dirty_page = page; |
d08b3851 PZ |
2948 | get_page(dirty_page); |
2949 | } | |
4294621f | 2950 | } |
64d6519d | 2951 | set_pte_at(mm, address, page_table, entry); |
d00806b1 NP |
2952 | |
2953 | /* no need to invalidate: a not-present page won't be cached */ | |
4b3073e1 | 2954 | update_mmu_cache(vma, address, page_table); |
1da177e4 | 2955 | } else { |
5b4e655e KH |
2956 | if (charged) |
2957 | mem_cgroup_uncharge_page(page); | |
d00806b1 NP |
2958 | if (anon) |
2959 | page_cache_release(page); | |
2960 | else | |
54cb8821 | 2961 | anon = 1; /* no anon but release faulted_page */ |
1da177e4 LT |
2962 | } |
2963 | ||
8f4e2101 | 2964 | pte_unmap_unlock(page_table, ptl); |
d00806b1 NP |
2965 | |
2966 | out: | |
b827e496 NP |
2967 | if (dirty_page) { |
2968 | struct address_space *mapping = page->mapping; | |
8f7b3d15 | 2969 | |
b827e496 NP |
2970 | if (set_page_dirty(dirty_page)) |
2971 | page_mkwrite = 1; | |
2972 | unlock_page(dirty_page); | |
d08b3851 | 2973 | put_page(dirty_page); |
b827e496 NP |
2974 | if (page_mkwrite && mapping) { |
2975 | /* | |
2976 | * Some device drivers do not set page.mapping but still | |
2977 | * dirty their pages | |
2978 | */ | |
2979 | balance_dirty_pages_ratelimited(mapping); | |
2980 | } | |
2981 | ||
2982 | /* file_update_time outside page_lock */ | |
2983 | if (vma->vm_file) | |
2984 | file_update_time(vma->vm_file); | |
2985 | } else { | |
2986 | unlock_page(vmf.page); | |
2987 | if (anon) | |
2988 | page_cache_release(vmf.page); | |
d08b3851 | 2989 | } |
d00806b1 | 2990 | |
83c54070 | 2991 | return ret; |
b827e496 NP |
2992 | |
2993 | unwritable_page: | |
2994 | page_cache_release(page); | |
2995 | return ret; | |
54cb8821 | 2996 | } |
d00806b1 | 2997 | |
54cb8821 NP |
2998 | static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
2999 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 3000 | unsigned int flags, pte_t orig_pte) |
54cb8821 NP |
3001 | { |
3002 | pgoff_t pgoff = (((address & PAGE_MASK) | |
0da7e01f | 3003 | - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
54cb8821 | 3004 | |
16abfa08 HD |
3005 | pte_unmap(page_table); |
3006 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); | |
54cb8821 NP |
3007 | } |
3008 | ||
1da177e4 LT |
3009 | /* |
3010 | * Fault of a previously existing named mapping. Repopulate the pte | |
3011 | * from the encoded file_pte if possible. This enables swappable | |
3012 | * nonlinear vmas. | |
8f4e2101 HD |
3013 | * |
3014 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3015 | * but allow concurrent faults), and pte mapped but not yet locked. | |
3016 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 3017 | */ |
d0217ac0 | 3018 | static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
65500d23 | 3019 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
30c9f3a9 | 3020 | unsigned int flags, pte_t orig_pte) |
1da177e4 | 3021 | { |
65500d23 | 3022 | pgoff_t pgoff; |
1da177e4 | 3023 | |
30c9f3a9 LT |
3024 | flags |= FAULT_FLAG_NONLINEAR; |
3025 | ||
4c21e2f2 | 3026 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
83c54070 | 3027 | return 0; |
1da177e4 | 3028 | |
2509ef26 | 3029 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) { |
65500d23 HD |
3030 | /* |
3031 | * Page table corrupted: show pte and kill process. | |
3032 | */ | |
3dc14741 | 3033 | print_bad_pte(vma, address, orig_pte, NULL); |
d99be1a8 | 3034 | return VM_FAULT_SIGBUS; |
65500d23 | 3035 | } |
65500d23 HD |
3036 | |
3037 | pgoff = pte_to_pgoff(orig_pte); | |
16abfa08 | 3038 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
1da177e4 LT |
3039 | } |
3040 | ||
3041 | /* | |
3042 | * These routines also need to handle stuff like marking pages dirty | |
3043 | * and/or accessed for architectures that don't do it in hardware (most | |
3044 | * RISC architectures). The early dirtying is also good on the i386. | |
3045 | * | |
3046 | * There is also a hook called "update_mmu_cache()" that architectures | |
3047 | * with external mmu caches can use to update those (ie the Sparc or | |
3048 | * PowerPC hashed page tables that act as extended TLBs). | |
3049 | * | |
c74df32c HD |
3050 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
3051 | * but allow concurrent faults), and pte mapped but not yet locked. | |
3052 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 LT |
3053 | */ |
3054 | static inline int handle_pte_fault(struct mm_struct *mm, | |
65500d23 | 3055 | struct vm_area_struct *vma, unsigned long address, |
30c9f3a9 | 3056 | pte_t *pte, pmd_t *pmd, unsigned int flags) |
1da177e4 LT |
3057 | { |
3058 | pte_t entry; | |
8f4e2101 | 3059 | spinlock_t *ptl; |
1da177e4 | 3060 | |
8dab5241 | 3061 | entry = *pte; |
1da177e4 | 3062 | if (!pte_present(entry)) { |
65500d23 | 3063 | if (pte_none(entry)) { |
f4b81804 | 3064 | if (vma->vm_ops) { |
3c18ddd1 | 3065 | if (likely(vma->vm_ops->fault)) |
54cb8821 | 3066 | return do_linear_fault(mm, vma, address, |
30c9f3a9 | 3067 | pte, pmd, flags, entry); |
f4b81804 JS |
3068 | } |
3069 | return do_anonymous_page(mm, vma, address, | |
30c9f3a9 | 3070 | pte, pmd, flags); |
65500d23 | 3071 | } |
1da177e4 | 3072 | if (pte_file(entry)) |
d0217ac0 | 3073 | return do_nonlinear_fault(mm, vma, address, |
30c9f3a9 | 3074 | pte, pmd, flags, entry); |
65500d23 | 3075 | return do_swap_page(mm, vma, address, |
30c9f3a9 | 3076 | pte, pmd, flags, entry); |
1da177e4 LT |
3077 | } |
3078 | ||
4c21e2f2 | 3079 | ptl = pte_lockptr(mm, pmd); |
8f4e2101 HD |
3080 | spin_lock(ptl); |
3081 | if (unlikely(!pte_same(*pte, entry))) | |
3082 | goto unlock; | |
30c9f3a9 | 3083 | if (flags & FAULT_FLAG_WRITE) { |
1da177e4 | 3084 | if (!pte_write(entry)) |
8f4e2101 HD |
3085 | return do_wp_page(mm, vma, address, |
3086 | pte, pmd, ptl, entry); | |
1da177e4 LT |
3087 | entry = pte_mkdirty(entry); |
3088 | } | |
3089 | entry = pte_mkyoung(entry); | |
30c9f3a9 | 3090 | if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { |
4b3073e1 | 3091 | update_mmu_cache(vma, address, pte); |
1a44e149 AA |
3092 | } else { |
3093 | /* | |
3094 | * This is needed only for protection faults but the arch code | |
3095 | * is not yet telling us if this is a protection fault or not. | |
3096 | * This still avoids useless tlb flushes for .text page faults | |
3097 | * with threads. | |
3098 | */ | |
30c9f3a9 | 3099 | if (flags & FAULT_FLAG_WRITE) |
1a44e149 AA |
3100 | flush_tlb_page(vma, address); |
3101 | } | |
8f4e2101 HD |
3102 | unlock: |
3103 | pte_unmap_unlock(pte, ptl); | |
83c54070 | 3104 | return 0; |
1da177e4 LT |
3105 | } |
3106 | ||
3107 | /* | |
3108 | * By the time we get here, we already hold the mm semaphore | |
3109 | */ | |
83c54070 | 3110 | int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
d06063cc | 3111 | unsigned long address, unsigned int flags) |
1da177e4 LT |
3112 | { |
3113 | pgd_t *pgd; | |
3114 | pud_t *pud; | |
3115 | pmd_t *pmd; | |
3116 | pte_t *pte; | |
3117 | ||
3118 | __set_current_state(TASK_RUNNING); | |
3119 | ||
f8891e5e | 3120 | count_vm_event(PGFAULT); |
1da177e4 | 3121 | |
34e55232 KH |
3122 | /* do counter updates before entering really critical section. */ |
3123 | check_sync_rss_stat(current); | |
3124 | ||
ac9b9c66 | 3125 | if (unlikely(is_vm_hugetlb_page(vma))) |
30c9f3a9 | 3126 | return hugetlb_fault(mm, vma, address, flags); |
1da177e4 | 3127 | |
1da177e4 | 3128 | pgd = pgd_offset(mm, address); |
1da177e4 LT |
3129 | pud = pud_alloc(mm, pgd, address); |
3130 | if (!pud) | |
c74df32c | 3131 | return VM_FAULT_OOM; |
1da177e4 LT |
3132 | pmd = pmd_alloc(mm, pud, address); |
3133 | if (!pmd) | |
c74df32c | 3134 | return VM_FAULT_OOM; |
1da177e4 LT |
3135 | pte = pte_alloc_map(mm, pmd, address); |
3136 | if (!pte) | |
c74df32c | 3137 | return VM_FAULT_OOM; |
1da177e4 | 3138 | |
30c9f3a9 | 3139 | return handle_pte_fault(mm, vma, address, pte, pmd, flags); |
1da177e4 LT |
3140 | } |
3141 | ||
3142 | #ifndef __PAGETABLE_PUD_FOLDED | |
3143 | /* | |
3144 | * Allocate page upper directory. | |
872fec16 | 3145 | * We've already handled the fast-path in-line. |
1da177e4 | 3146 | */ |
1bb3630e | 3147 | int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) |
1da177e4 | 3148 | { |
c74df32c HD |
3149 | pud_t *new = pud_alloc_one(mm, address); |
3150 | if (!new) | |
1bb3630e | 3151 | return -ENOMEM; |
1da177e4 | 3152 | |
362a61ad NP |
3153 | smp_wmb(); /* See comment in __pte_alloc */ |
3154 | ||
872fec16 | 3155 | spin_lock(&mm->page_table_lock); |
1bb3630e | 3156 | if (pgd_present(*pgd)) /* Another has populated it */ |
5e541973 | 3157 | pud_free(mm, new); |
1bb3630e HD |
3158 | else |
3159 | pgd_populate(mm, pgd, new); | |
c74df32c | 3160 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3161 | return 0; |
1da177e4 LT |
3162 | } |
3163 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
3164 | ||
3165 | #ifndef __PAGETABLE_PMD_FOLDED | |
3166 | /* | |
3167 | * Allocate page middle directory. | |
872fec16 | 3168 | * We've already handled the fast-path in-line. |
1da177e4 | 3169 | */ |
1bb3630e | 3170 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 3171 | { |
c74df32c HD |
3172 | pmd_t *new = pmd_alloc_one(mm, address); |
3173 | if (!new) | |
1bb3630e | 3174 | return -ENOMEM; |
1da177e4 | 3175 | |
362a61ad NP |
3176 | smp_wmb(); /* See comment in __pte_alloc */ |
3177 | ||
872fec16 | 3178 | spin_lock(&mm->page_table_lock); |
1da177e4 | 3179 | #ifndef __ARCH_HAS_4LEVEL_HACK |
1bb3630e | 3180 | if (pud_present(*pud)) /* Another has populated it */ |
5e541973 | 3181 | pmd_free(mm, new); |
1bb3630e HD |
3182 | else |
3183 | pud_populate(mm, pud, new); | |
1da177e4 | 3184 | #else |
1bb3630e | 3185 | if (pgd_present(*pud)) /* Another has populated it */ |
5e541973 | 3186 | pmd_free(mm, new); |
1bb3630e HD |
3187 | else |
3188 | pgd_populate(mm, pud, new); | |
1da177e4 | 3189 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
c74df32c | 3190 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3191 | return 0; |
e0f39591 | 3192 | } |
1da177e4 LT |
3193 | #endif /* __PAGETABLE_PMD_FOLDED */ |
3194 | ||
3195 | int make_pages_present(unsigned long addr, unsigned long end) | |
3196 | { | |
3197 | int ret, len, write; | |
3198 | struct vm_area_struct * vma; | |
3199 | ||
3200 | vma = find_vma(current->mm, addr); | |
3201 | if (!vma) | |
a477097d | 3202 | return -ENOMEM; |
1da177e4 | 3203 | write = (vma->vm_flags & VM_WRITE) != 0; |
5bcb28b1 ES |
3204 | BUG_ON(addr >= end); |
3205 | BUG_ON(end > vma->vm_end); | |
68e116a3 | 3206 | len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE; |
1da177e4 LT |
3207 | ret = get_user_pages(current, current->mm, addr, |
3208 | len, write, 0, NULL, NULL); | |
c11d69d8 | 3209 | if (ret < 0) |
1da177e4 | 3210 | return ret; |
9978ad58 | 3211 | return ret == len ? 0 : -EFAULT; |
1da177e4 LT |
3212 | } |
3213 | ||
1da177e4 LT |
3214 | #if !defined(__HAVE_ARCH_GATE_AREA) |
3215 | ||
3216 | #if defined(AT_SYSINFO_EHDR) | |
5ce7852c | 3217 | static struct vm_area_struct gate_vma; |
1da177e4 LT |
3218 | |
3219 | static int __init gate_vma_init(void) | |
3220 | { | |
3221 | gate_vma.vm_mm = NULL; | |
3222 | gate_vma.vm_start = FIXADDR_USER_START; | |
3223 | gate_vma.vm_end = FIXADDR_USER_END; | |
b6558c4a RM |
3224 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; |
3225 | gate_vma.vm_page_prot = __P101; | |
f47aef55 RM |
3226 | /* |
3227 | * Make sure the vDSO gets into every core dump. | |
3228 | * Dumping its contents makes post-mortem fully interpretable later | |
3229 | * without matching up the same kernel and hardware config to see | |
3230 | * what PC values meant. | |
3231 | */ | |
3232 | gate_vma.vm_flags |= VM_ALWAYSDUMP; | |
1da177e4 LT |
3233 | return 0; |
3234 | } | |
3235 | __initcall(gate_vma_init); | |
3236 | #endif | |
3237 | ||
3238 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) | |
3239 | { | |
3240 | #ifdef AT_SYSINFO_EHDR | |
3241 | return &gate_vma; | |
3242 | #else | |
3243 | return NULL; | |
3244 | #endif | |
3245 | } | |
3246 | ||
3247 | int in_gate_area_no_task(unsigned long addr) | |
3248 | { | |
3249 | #ifdef AT_SYSINFO_EHDR | |
3250 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) | |
3251 | return 1; | |
3252 | #endif | |
3253 | return 0; | |
3254 | } | |
3255 | ||
3256 | #endif /* __HAVE_ARCH_GATE_AREA */ | |
0ec76a11 | 3257 | |
f8ad0f49 JW |
3258 | static int follow_pte(struct mm_struct *mm, unsigned long address, |
3259 | pte_t **ptepp, spinlock_t **ptlp) | |
3260 | { | |
3261 | pgd_t *pgd; | |
3262 | pud_t *pud; | |
3263 | pmd_t *pmd; | |
3264 | pte_t *ptep; | |
3265 | ||
3266 | pgd = pgd_offset(mm, address); | |
3267 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
3268 | goto out; | |
3269 | ||
3270 | pud = pud_offset(pgd, address); | |
3271 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) | |
3272 | goto out; | |
3273 | ||
3274 | pmd = pmd_offset(pud, address); | |
3275 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
3276 | goto out; | |
3277 | ||
3278 | /* We cannot handle huge page PFN maps. Luckily they don't exist. */ | |
3279 | if (pmd_huge(*pmd)) | |
3280 | goto out; | |
3281 | ||
3282 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); | |
3283 | if (!ptep) | |
3284 | goto out; | |
3285 | if (!pte_present(*ptep)) | |
3286 | goto unlock; | |
3287 | *ptepp = ptep; | |
3288 | return 0; | |
3289 | unlock: | |
3290 | pte_unmap_unlock(ptep, *ptlp); | |
3291 | out: | |
3292 | return -EINVAL; | |
3293 | } | |
3294 | ||
3b6748e2 JW |
3295 | /** |
3296 | * follow_pfn - look up PFN at a user virtual address | |
3297 | * @vma: memory mapping | |
3298 | * @address: user virtual address | |
3299 | * @pfn: location to store found PFN | |
3300 | * | |
3301 | * Only IO mappings and raw PFN mappings are allowed. | |
3302 | * | |
3303 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
3304 | */ | |
3305 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
3306 | unsigned long *pfn) | |
3307 | { | |
3308 | int ret = -EINVAL; | |
3309 | spinlock_t *ptl; | |
3310 | pte_t *ptep; | |
3311 | ||
3312 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
3313 | return ret; | |
3314 | ||
3315 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
3316 | if (ret) | |
3317 | return ret; | |
3318 | *pfn = pte_pfn(*ptep); | |
3319 | pte_unmap_unlock(ptep, ptl); | |
3320 | return 0; | |
3321 | } | |
3322 | EXPORT_SYMBOL(follow_pfn); | |
3323 | ||
28b2ee20 | 3324 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 3325 | int follow_phys(struct vm_area_struct *vma, |
3326 | unsigned long address, unsigned int flags, | |
3327 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 3328 | { |
03668a4d | 3329 | int ret = -EINVAL; |
28b2ee20 RR |
3330 | pte_t *ptep, pte; |
3331 | spinlock_t *ptl; | |
28b2ee20 | 3332 | |
d87fe660 | 3333 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
3334 | goto out; | |
28b2ee20 | 3335 | |
03668a4d | 3336 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 3337 | goto out; |
28b2ee20 | 3338 | pte = *ptep; |
03668a4d | 3339 | |
28b2ee20 RR |
3340 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
3341 | goto unlock; | |
28b2ee20 RR |
3342 | |
3343 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 3344 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 3345 | |
03668a4d | 3346 | ret = 0; |
28b2ee20 RR |
3347 | unlock: |
3348 | pte_unmap_unlock(ptep, ptl); | |
3349 | out: | |
d87fe660 | 3350 | return ret; |
28b2ee20 RR |
3351 | } |
3352 | ||
3353 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
3354 | void *buf, int len, int write) | |
3355 | { | |
3356 | resource_size_t phys_addr; | |
3357 | unsigned long prot = 0; | |
2bc7273b | 3358 | void __iomem *maddr; |
28b2ee20 RR |
3359 | int offset = addr & (PAGE_SIZE-1); |
3360 | ||
d87fe660 | 3361 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
3362 | return -EINVAL; |
3363 | ||
3364 | maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot); | |
3365 | if (write) | |
3366 | memcpy_toio(maddr + offset, buf, len); | |
3367 | else | |
3368 | memcpy_fromio(buf, maddr + offset, len); | |
3369 | iounmap(maddr); | |
3370 | ||
3371 | return len; | |
3372 | } | |
3373 | #endif | |
3374 | ||
0ec76a11 DH |
3375 | /* |
3376 | * Access another process' address space. | |
3377 | * Source/target buffer must be kernel space, | |
3378 | * Do not walk the page table directly, use get_user_pages | |
3379 | */ | |
3380 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) | |
3381 | { | |
3382 | struct mm_struct *mm; | |
3383 | struct vm_area_struct *vma; | |
0ec76a11 DH |
3384 | void *old_buf = buf; |
3385 | ||
3386 | mm = get_task_mm(tsk); | |
3387 | if (!mm) | |
3388 | return 0; | |
3389 | ||
3390 | down_read(&mm->mmap_sem); | |
183ff22b | 3391 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
3392 | while (len) { |
3393 | int bytes, ret, offset; | |
3394 | void *maddr; | |
28b2ee20 | 3395 | struct page *page = NULL; |
0ec76a11 DH |
3396 | |
3397 | ret = get_user_pages(tsk, mm, addr, 1, | |
3398 | write, 1, &page, &vma); | |
28b2ee20 RR |
3399 | if (ret <= 0) { |
3400 | /* | |
3401 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
3402 | * we can access using slightly different code. | |
3403 | */ | |
3404 | #ifdef CONFIG_HAVE_IOREMAP_PROT | |
3405 | vma = find_vma(mm, addr); | |
3406 | if (!vma) | |
3407 | break; | |
3408 | if (vma->vm_ops && vma->vm_ops->access) | |
3409 | ret = vma->vm_ops->access(vma, addr, buf, | |
3410 | len, write); | |
3411 | if (ret <= 0) | |
3412 | #endif | |
3413 | break; | |
3414 | bytes = ret; | |
0ec76a11 | 3415 | } else { |
28b2ee20 RR |
3416 | bytes = len; |
3417 | offset = addr & (PAGE_SIZE-1); | |
3418 | if (bytes > PAGE_SIZE-offset) | |
3419 | bytes = PAGE_SIZE-offset; | |
3420 | ||
3421 | maddr = kmap(page); | |
3422 | if (write) { | |
3423 | copy_to_user_page(vma, page, addr, | |
3424 | maddr + offset, buf, bytes); | |
3425 | set_page_dirty_lock(page); | |
3426 | } else { | |
3427 | copy_from_user_page(vma, page, addr, | |
3428 | buf, maddr + offset, bytes); | |
3429 | } | |
3430 | kunmap(page); | |
3431 | page_cache_release(page); | |
0ec76a11 | 3432 | } |
0ec76a11 DH |
3433 | len -= bytes; |
3434 | buf += bytes; | |
3435 | addr += bytes; | |
3436 | } | |
3437 | up_read(&mm->mmap_sem); | |
3438 | mmput(mm); | |
3439 | ||
3440 | return buf - old_buf; | |
3441 | } | |
03252919 AK |
3442 | |
3443 | /* | |
3444 | * Print the name of a VMA. | |
3445 | */ | |
3446 | void print_vma_addr(char *prefix, unsigned long ip) | |
3447 | { | |
3448 | struct mm_struct *mm = current->mm; | |
3449 | struct vm_area_struct *vma; | |
3450 | ||
e8bff74a IM |
3451 | /* |
3452 | * Do not print if we are in atomic | |
3453 | * contexts (in exception stacks, etc.): | |
3454 | */ | |
3455 | if (preempt_count()) | |
3456 | return; | |
3457 | ||
03252919 AK |
3458 | down_read(&mm->mmap_sem); |
3459 | vma = find_vma(mm, ip); | |
3460 | if (vma && vma->vm_file) { | |
3461 | struct file *f = vma->vm_file; | |
3462 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
3463 | if (buf) { | |
3464 | char *p, *s; | |
3465 | ||
cf28b486 | 3466 | p = d_path(&f->f_path, buf, PAGE_SIZE); |
03252919 AK |
3467 | if (IS_ERR(p)) |
3468 | p = "?"; | |
3469 | s = strrchr(p, '/'); | |
3470 | if (s) | |
3471 | p = s+1; | |
3472 | printk("%s%s[%lx+%lx]", prefix, p, | |
3473 | vma->vm_start, | |
3474 | vma->vm_end - vma->vm_start); | |
3475 | free_page((unsigned long)buf); | |
3476 | } | |
3477 | } | |
3478 | up_read(¤t->mm->mmap_sem); | |
3479 | } | |
3ee1afa3 NP |
3480 | |
3481 | #ifdef CONFIG_PROVE_LOCKING | |
3482 | void might_fault(void) | |
3483 | { | |
95156f00 PZ |
3484 | /* |
3485 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
3486 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
3487 | * get paged out, therefore we'll never actually fault, and the | |
3488 | * below annotations will generate false positives. | |
3489 | */ | |
3490 | if (segment_eq(get_fs(), KERNEL_DS)) | |
3491 | return; | |
3492 | ||
3ee1afa3 NP |
3493 | might_sleep(); |
3494 | /* | |
3495 | * it would be nicer only to annotate paths which are not under | |
3496 | * pagefault_disable, however that requires a larger audit and | |
3497 | * providing helpers like get_user_atomic. | |
3498 | */ | |
3499 | if (!in_atomic() && current->mm) | |
3500 | might_lock_read(¤t->mm->mmap_sem); | |
3501 | } | |
3502 | EXPORT_SYMBOL(might_fault); | |
3503 | #endif |