]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/memory.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
6 | */ | |
7 | ||
8 | /* | |
9 | * demand-loading started 01.12.91 - seems it is high on the list of | |
10 | * things wanted, and it should be easy to implement. - Linus | |
11 | */ | |
12 | ||
13 | /* | |
14 | * Ok, demand-loading was easy, shared pages a little bit tricker. Shared | |
15 | * pages started 02.12.91, seems to work. - Linus. | |
16 | * | |
17 | * Tested sharing by executing about 30 /bin/sh: under the old kernel it | |
18 | * would have taken more than the 6M I have free, but it worked well as | |
19 | * far as I could see. | |
20 | * | |
21 | * Also corrected some "invalidate()"s - I wasn't doing enough of them. | |
22 | */ | |
23 | ||
24 | /* | |
25 | * Real VM (paging to/from disk) started 18.12.91. Much more work and | |
26 | * thought has to go into this. Oh, well.. | |
27 | * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. | |
28 | * Found it. Everything seems to work now. | |
29 | * 20.12.91 - Ok, making the swap-device changeable like the root. | |
30 | */ | |
31 | ||
32 | /* | |
33 | * 05.04.94 - Multi-page memory management added for v1.1. | |
166f61b9 | 34 | * Idea by Alex Bligh (alex@cconcepts.co.uk) |
1da177e4 LT |
35 | * |
36 | * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG | |
37 | * (Gerhard.Wichert@pdb.siemens.de) | |
38 | * | |
39 | * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) | |
40 | */ | |
41 | ||
42 | #include <linux/kernel_stat.h> | |
43 | #include <linux/mm.h> | |
6e84f315 | 44 | #include <linux/sched/mm.h> |
f7ccbae4 | 45 | #include <linux/sched/coredump.h> |
6a3827d7 | 46 | #include <linux/sched/numa_balancing.h> |
29930025 | 47 | #include <linux/sched/task.h> |
1da177e4 LT |
48 | #include <linux/hugetlb.h> |
49 | #include <linux/mman.h> | |
50 | #include <linux/swap.h> | |
51 | #include <linux/highmem.h> | |
52 | #include <linux/pagemap.h> | |
5042db43 | 53 | #include <linux/memremap.h> |
9a840895 | 54 | #include <linux/ksm.h> |
1da177e4 | 55 | #include <linux/rmap.h> |
b95f1b31 | 56 | #include <linux/export.h> |
0ff92245 | 57 | #include <linux/delayacct.h> |
1da177e4 | 58 | #include <linux/init.h> |
01c8f1c4 | 59 | #include <linux/pfn_t.h> |
edc79b2a | 60 | #include <linux/writeback.h> |
8a9f3ccd | 61 | #include <linux/memcontrol.h> |
cddb8a5c | 62 | #include <linux/mmu_notifier.h> |
3dc14741 HD |
63 | #include <linux/swapops.h> |
64 | #include <linux/elf.h> | |
5a0e3ad6 | 65 | #include <linux/gfp.h> |
4daae3b4 | 66 | #include <linux/migrate.h> |
2fbc57c5 | 67 | #include <linux/string.h> |
1592eef0 | 68 | #include <linux/debugfs.h> |
6b251fc9 | 69 | #include <linux/userfaultfd_k.h> |
bc2466e4 | 70 | #include <linux/dax.h> |
6b31d595 | 71 | #include <linux/oom.h> |
98fa15f3 | 72 | #include <linux/numa.h> |
bce617ed PX |
73 | #include <linux/perf_event.h> |
74 | #include <linux/ptrace.h> | |
e80d3909 | 75 | #include <linux/vmalloc.h> |
1da177e4 | 76 | |
b3d1411b JFG |
77 | #include <trace/events/kmem.h> |
78 | ||
6952b61d | 79 | #include <asm/io.h> |
33a709b2 | 80 | #include <asm/mmu_context.h> |
1da177e4 | 81 | #include <asm/pgalloc.h> |
7c0f6ba6 | 82 | #include <linux/uaccess.h> |
1da177e4 LT |
83 | #include <asm/tlb.h> |
84 | #include <asm/tlbflush.h> | |
1da177e4 | 85 | |
e80d3909 | 86 | #include "pgalloc-track.h" |
42b77728 JB |
87 | #include "internal.h" |
88 | ||
af27d940 | 89 | #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST) |
90572890 | 90 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. |
75980e97 PZ |
91 | #endif |
92 | ||
d41dee36 | 93 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
94 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
95 | unsigned long max_mapnr; | |
1da177e4 | 96 | EXPORT_SYMBOL(max_mapnr); |
166f61b9 TH |
97 | |
98 | struct page *mem_map; | |
1da177e4 LT |
99 | EXPORT_SYMBOL(mem_map); |
100 | #endif | |
101 | ||
1da177e4 LT |
102 | /* |
103 | * A number of key systems in x86 including ioremap() rely on the assumption | |
104 | * that high_memory defines the upper bound on direct map memory, then end | |
105 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
106 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
107 | * and ZONE_HIGHMEM. | |
108 | */ | |
166f61b9 | 109 | void *high_memory; |
1da177e4 | 110 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 111 | |
32a93233 IM |
112 | /* |
113 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
114 | * | |
115 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
116 | * as ancient (libc5 based) binaries can segfault. ) | |
117 | */ | |
118 | int randomize_va_space __read_mostly = | |
119 | #ifdef CONFIG_COMPAT_BRK | |
120 | 1; | |
121 | #else | |
122 | 2; | |
123 | #endif | |
a62eaf15 | 124 | |
83d116c5 JH |
125 | #ifndef arch_faults_on_old_pte |
126 | static inline bool arch_faults_on_old_pte(void) | |
127 | { | |
128 | /* | |
129 | * Those arches which don't have hw access flag feature need to | |
130 | * implement their own helper. By default, "true" means pagefault | |
131 | * will be hit on old pte. | |
132 | */ | |
133 | return true; | |
134 | } | |
135 | #endif | |
136 | ||
a62eaf15 AK |
137 | static int __init disable_randmaps(char *s) |
138 | { | |
139 | randomize_va_space = 0; | |
9b41046c | 140 | return 1; |
a62eaf15 AK |
141 | } |
142 | __setup("norandmaps", disable_randmaps); | |
143 | ||
62eede62 | 144 | unsigned long zero_pfn __read_mostly; |
0b70068e AB |
145 | EXPORT_SYMBOL(zero_pfn); |
146 | ||
166f61b9 TH |
147 | unsigned long highest_memmap_pfn __read_mostly; |
148 | ||
a13ea5b7 HD |
149 | /* |
150 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
151 | */ | |
152 | static int __init init_zero_pfn(void) | |
153 | { | |
154 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
155 | return 0; | |
156 | } | |
75f7c053 | 157 | early_initcall(init_zero_pfn); |
a62eaf15 | 158 | |
e4dcad20 | 159 | void mm_trace_rss_stat(struct mm_struct *mm, int member, long count) |
b3d1411b | 160 | { |
e4dcad20 | 161 | trace_rss_stat(mm, member, count); |
b3d1411b | 162 | } |
d559db08 | 163 | |
34e55232 KH |
164 | #if defined(SPLIT_RSS_COUNTING) |
165 | ||
ea48cf78 | 166 | void sync_mm_rss(struct mm_struct *mm) |
34e55232 KH |
167 | { |
168 | int i; | |
169 | ||
170 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
05af2e10 DR |
171 | if (current->rss_stat.count[i]) { |
172 | add_mm_counter(mm, i, current->rss_stat.count[i]); | |
173 | current->rss_stat.count[i] = 0; | |
34e55232 KH |
174 | } |
175 | } | |
05af2e10 | 176 | current->rss_stat.events = 0; |
34e55232 KH |
177 | } |
178 | ||
179 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
180 | { | |
181 | struct task_struct *task = current; | |
182 | ||
183 | if (likely(task->mm == mm)) | |
184 | task->rss_stat.count[member] += val; | |
185 | else | |
186 | add_mm_counter(mm, member, val); | |
187 | } | |
188 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
189 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
190 | ||
191 | /* sync counter once per 64 page faults */ | |
192 | #define TASK_RSS_EVENTS_THRESH (64) | |
193 | static void check_sync_rss_stat(struct task_struct *task) | |
194 | { | |
195 | if (unlikely(task != current)) | |
196 | return; | |
197 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
ea48cf78 | 198 | sync_mm_rss(task->mm); |
34e55232 | 199 | } |
9547d01b | 200 | #else /* SPLIT_RSS_COUNTING */ |
34e55232 KH |
201 | |
202 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
203 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
204 | ||
205 | static void check_sync_rss_stat(struct task_struct *task) | |
206 | { | |
207 | } | |
208 | ||
9547d01b PZ |
209 | #endif /* SPLIT_RSS_COUNTING */ |
210 | ||
1da177e4 LT |
211 | /* |
212 | * Note: this doesn't free the actual pages themselves. That | |
213 | * has been handled earlier when unmapping all the memory regions. | |
214 | */ | |
9e1b32ca BH |
215 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
216 | unsigned long addr) | |
1da177e4 | 217 | { |
2f569afd | 218 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 219 | pmd_clear(pmd); |
9e1b32ca | 220 | pte_free_tlb(tlb, token, addr); |
c4812909 | 221 | mm_dec_nr_ptes(tlb->mm); |
1da177e4 LT |
222 | } |
223 | ||
e0da382c HD |
224 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
225 | unsigned long addr, unsigned long end, | |
226 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
227 | { |
228 | pmd_t *pmd; | |
229 | unsigned long next; | |
e0da382c | 230 | unsigned long start; |
1da177e4 | 231 | |
e0da382c | 232 | start = addr; |
1da177e4 | 233 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
234 | do { |
235 | next = pmd_addr_end(addr, end); | |
236 | if (pmd_none_or_clear_bad(pmd)) | |
237 | continue; | |
9e1b32ca | 238 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
239 | } while (pmd++, addr = next, addr != end); |
240 | ||
e0da382c HD |
241 | start &= PUD_MASK; |
242 | if (start < floor) | |
243 | return; | |
244 | if (ceiling) { | |
245 | ceiling &= PUD_MASK; | |
246 | if (!ceiling) | |
247 | return; | |
1da177e4 | 248 | } |
e0da382c HD |
249 | if (end - 1 > ceiling - 1) |
250 | return; | |
251 | ||
252 | pmd = pmd_offset(pud, start); | |
253 | pud_clear(pud); | |
9e1b32ca | 254 | pmd_free_tlb(tlb, pmd, start); |
dc6c9a35 | 255 | mm_dec_nr_pmds(tlb->mm); |
1da177e4 LT |
256 | } |
257 | ||
c2febafc | 258 | static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, |
e0da382c HD |
259 | unsigned long addr, unsigned long end, |
260 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
261 | { |
262 | pud_t *pud; | |
263 | unsigned long next; | |
e0da382c | 264 | unsigned long start; |
1da177e4 | 265 | |
e0da382c | 266 | start = addr; |
c2febafc | 267 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
268 | do { |
269 | next = pud_addr_end(addr, end); | |
270 | if (pud_none_or_clear_bad(pud)) | |
271 | continue; | |
e0da382c | 272 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
273 | } while (pud++, addr = next, addr != end); |
274 | ||
c2febafc KS |
275 | start &= P4D_MASK; |
276 | if (start < floor) | |
277 | return; | |
278 | if (ceiling) { | |
279 | ceiling &= P4D_MASK; | |
280 | if (!ceiling) | |
281 | return; | |
282 | } | |
283 | if (end - 1 > ceiling - 1) | |
284 | return; | |
285 | ||
286 | pud = pud_offset(p4d, start); | |
287 | p4d_clear(p4d); | |
288 | pud_free_tlb(tlb, pud, start); | |
b4e98d9a | 289 | mm_dec_nr_puds(tlb->mm); |
c2febafc KS |
290 | } |
291 | ||
292 | static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, | |
293 | unsigned long addr, unsigned long end, | |
294 | unsigned long floor, unsigned long ceiling) | |
295 | { | |
296 | p4d_t *p4d; | |
297 | unsigned long next; | |
298 | unsigned long start; | |
299 | ||
300 | start = addr; | |
301 | p4d = p4d_offset(pgd, addr); | |
302 | do { | |
303 | next = p4d_addr_end(addr, end); | |
304 | if (p4d_none_or_clear_bad(p4d)) | |
305 | continue; | |
306 | free_pud_range(tlb, p4d, addr, next, floor, ceiling); | |
307 | } while (p4d++, addr = next, addr != end); | |
308 | ||
e0da382c HD |
309 | start &= PGDIR_MASK; |
310 | if (start < floor) | |
311 | return; | |
312 | if (ceiling) { | |
313 | ceiling &= PGDIR_MASK; | |
314 | if (!ceiling) | |
315 | return; | |
1da177e4 | 316 | } |
e0da382c HD |
317 | if (end - 1 > ceiling - 1) |
318 | return; | |
319 | ||
c2febafc | 320 | p4d = p4d_offset(pgd, start); |
e0da382c | 321 | pgd_clear(pgd); |
c2febafc | 322 | p4d_free_tlb(tlb, p4d, start); |
1da177e4 LT |
323 | } |
324 | ||
325 | /* | |
e0da382c | 326 | * This function frees user-level page tables of a process. |
1da177e4 | 327 | */ |
42b77728 | 328 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
329 | unsigned long addr, unsigned long end, |
330 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
331 | { |
332 | pgd_t *pgd; | |
333 | unsigned long next; | |
e0da382c HD |
334 | |
335 | /* | |
336 | * The next few lines have given us lots of grief... | |
337 | * | |
338 | * Why are we testing PMD* at this top level? Because often | |
339 | * there will be no work to do at all, and we'd prefer not to | |
340 | * go all the way down to the bottom just to discover that. | |
341 | * | |
342 | * Why all these "- 1"s? Because 0 represents both the bottom | |
343 | * of the address space and the top of it (using -1 for the | |
344 | * top wouldn't help much: the masks would do the wrong thing). | |
345 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
346 | * the address space, but end 0 and ceiling 0 refer to the top | |
347 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
348 | * that end 0 case should be mythical). | |
349 | * | |
350 | * Wherever addr is brought up or ceiling brought down, we must | |
351 | * be careful to reject "the opposite 0" before it confuses the | |
352 | * subsequent tests. But what about where end is brought down | |
353 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
354 | * | |
355 | * Whereas we round start (addr) and ceiling down, by different | |
356 | * masks at different levels, in order to test whether a table | |
357 | * now has no other vmas using it, so can be freed, we don't | |
358 | * bother to round floor or end up - the tests don't need that. | |
359 | */ | |
1da177e4 | 360 | |
e0da382c HD |
361 | addr &= PMD_MASK; |
362 | if (addr < floor) { | |
363 | addr += PMD_SIZE; | |
364 | if (!addr) | |
365 | return; | |
366 | } | |
367 | if (ceiling) { | |
368 | ceiling &= PMD_MASK; | |
369 | if (!ceiling) | |
370 | return; | |
371 | } | |
372 | if (end - 1 > ceiling - 1) | |
373 | end -= PMD_SIZE; | |
374 | if (addr > end - 1) | |
375 | return; | |
07e32661 AK |
376 | /* |
377 | * We add page table cache pages with PAGE_SIZE, | |
378 | * (see pte_free_tlb()), flush the tlb if we need | |
379 | */ | |
ed6a7935 | 380 | tlb_change_page_size(tlb, PAGE_SIZE); |
42b77728 | 381 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
382 | do { |
383 | next = pgd_addr_end(addr, end); | |
384 | if (pgd_none_or_clear_bad(pgd)) | |
385 | continue; | |
c2febafc | 386 | free_p4d_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 387 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
388 | } |
389 | ||
42b77728 | 390 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 391 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
392 | { |
393 | while (vma) { | |
394 | struct vm_area_struct *next = vma->vm_next; | |
395 | unsigned long addr = vma->vm_start; | |
396 | ||
8f4f8c16 | 397 | /* |
25d9e2d1 | 398 | * Hide vma from rmap and truncate_pagecache before freeing |
399 | * pgtables | |
8f4f8c16 | 400 | */ |
5beb4930 | 401 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
402 | unlink_file_vma(vma); |
403 | ||
9da61aef | 404 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 405 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
166f61b9 | 406 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 HD |
407 | } else { |
408 | /* | |
409 | * Optimization: gather nearby vmas into one call down | |
410 | */ | |
411 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 412 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
413 | vma = next; |
414 | next = vma->vm_next; | |
5beb4930 | 415 | unlink_anon_vmas(vma); |
8f4f8c16 | 416 | unlink_file_vma(vma); |
3bf5ee95 HD |
417 | } |
418 | free_pgd_range(tlb, addr, vma->vm_end, | |
166f61b9 | 419 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 | 420 | } |
e0da382c HD |
421 | vma = next; |
422 | } | |
1da177e4 LT |
423 | } |
424 | ||
4cf58924 | 425 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd) |
1da177e4 | 426 | { |
c4088ebd | 427 | spinlock_t *ptl; |
4cf58924 | 428 | pgtable_t new = pte_alloc_one(mm); |
1bb3630e HD |
429 | if (!new) |
430 | return -ENOMEM; | |
431 | ||
362a61ad NP |
432 | /* |
433 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
434 | * visible before the pte is made visible to other CPUs by being | |
435 | * put into page tables. | |
436 | * | |
437 | * The other side of the story is the pointer chasing in the page | |
438 | * table walking code (when walking the page table without locking; | |
439 | * ie. most of the time). Fortunately, these data accesses consist | |
440 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
441 | * being the notable exception) will already guarantee loads are | |
442 | * seen in-order. See the alpha page table accessors for the | |
bb7cdd38 | 443 | * smp_rmb() barriers in page table walking code. |
362a61ad NP |
444 | */ |
445 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
446 | ||
c4088ebd | 447 | ptl = pmd_lock(mm, pmd); |
8ac1f832 | 448 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
c4812909 | 449 | mm_inc_nr_ptes(mm); |
1da177e4 | 450 | pmd_populate(mm, pmd, new); |
2f569afd | 451 | new = NULL; |
4b471e88 | 452 | } |
c4088ebd | 453 | spin_unlock(ptl); |
2f569afd MS |
454 | if (new) |
455 | pte_free(mm, new); | |
1bb3630e | 456 | return 0; |
1da177e4 LT |
457 | } |
458 | ||
4cf58924 | 459 | int __pte_alloc_kernel(pmd_t *pmd) |
1da177e4 | 460 | { |
4cf58924 | 461 | pte_t *new = pte_alloc_one_kernel(&init_mm); |
1bb3630e HD |
462 | if (!new) |
463 | return -ENOMEM; | |
464 | ||
362a61ad NP |
465 | smp_wmb(); /* See comment in __pte_alloc */ |
466 | ||
1bb3630e | 467 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 468 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 469 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 470 | new = NULL; |
4b471e88 | 471 | } |
1bb3630e | 472 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
473 | if (new) |
474 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 475 | return 0; |
1da177e4 LT |
476 | } |
477 | ||
d559db08 KH |
478 | static inline void init_rss_vec(int *rss) |
479 | { | |
480 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
481 | } | |
482 | ||
483 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 484 | { |
d559db08 KH |
485 | int i; |
486 | ||
34e55232 | 487 | if (current->mm == mm) |
05af2e10 | 488 | sync_mm_rss(mm); |
d559db08 KH |
489 | for (i = 0; i < NR_MM_COUNTERS; i++) |
490 | if (rss[i]) | |
491 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
492 | } |
493 | ||
b5810039 | 494 | /* |
6aab341e LT |
495 | * This function is called to print an error when a bad pte |
496 | * is found. For example, we might have a PFN-mapped pte in | |
497 | * a region that doesn't allow it. | |
b5810039 NP |
498 | * |
499 | * The calling function must still handle the error. | |
500 | */ | |
3dc14741 HD |
501 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
502 | pte_t pte, struct page *page) | |
b5810039 | 503 | { |
3dc14741 | 504 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
c2febafc KS |
505 | p4d_t *p4d = p4d_offset(pgd, addr); |
506 | pud_t *pud = pud_offset(p4d, addr); | |
3dc14741 HD |
507 | pmd_t *pmd = pmd_offset(pud, addr); |
508 | struct address_space *mapping; | |
509 | pgoff_t index; | |
d936cf9b HD |
510 | static unsigned long resume; |
511 | static unsigned long nr_shown; | |
512 | static unsigned long nr_unshown; | |
513 | ||
514 | /* | |
515 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
516 | * or allow a steady drip of one report per second. | |
517 | */ | |
518 | if (nr_shown == 60) { | |
519 | if (time_before(jiffies, resume)) { | |
520 | nr_unshown++; | |
521 | return; | |
522 | } | |
523 | if (nr_unshown) { | |
1170532b JP |
524 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
525 | nr_unshown); | |
d936cf9b HD |
526 | nr_unshown = 0; |
527 | } | |
528 | nr_shown = 0; | |
529 | } | |
530 | if (nr_shown++ == 0) | |
531 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
532 | |
533 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
534 | index = linear_page_index(vma, addr); | |
535 | ||
1170532b JP |
536 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
537 | current->comm, | |
538 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 539 | if (page) |
f0b791a3 | 540 | dump_page(page, "bad pte"); |
6aa9b8b2 | 541 | pr_alert("addr:%px vm_flags:%08lx anon_vma:%px mapping:%px index:%lx\n", |
1170532b | 542 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); |
d75f773c | 543 | pr_alert("file:%pD fault:%ps mmap:%ps readpage:%ps\n", |
2682582a KK |
544 | vma->vm_file, |
545 | vma->vm_ops ? vma->vm_ops->fault : NULL, | |
546 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, | |
547 | mapping ? mapping->a_ops->readpage : NULL); | |
b5810039 | 548 | dump_stack(); |
373d4d09 | 549 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
550 | } |
551 | ||
ee498ed7 | 552 | /* |
7e675137 | 553 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 554 | * |
7e675137 NP |
555 | * "Special" mappings do not wish to be associated with a "struct page" (either |
556 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
557 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 558 | * |
7e675137 NP |
559 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
560 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
561 | * may not have a spare pte bit, which requires a more complicated scheme, | |
562 | * described below. | |
563 | * | |
564 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
565 | * special mapping (even if there are underlying and valid "struct pages"). | |
566 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 567 | * |
b379d790 JH |
568 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
569 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
570 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
571 | * mapping will always honor the rule | |
6aab341e LT |
572 | * |
573 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
574 | * | |
7e675137 NP |
575 | * And for normal mappings this is false. |
576 | * | |
577 | * This restricts such mappings to be a linear translation from virtual address | |
578 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
579 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
580 | * special (because none can have been COWed). | |
b379d790 | 581 | * |
b379d790 | 582 | * |
7e675137 | 583 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
584 | * |
585 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
586 | * page" backing, however the difference is that _all_ pages with a struct | |
587 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
588 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
589 | * (which can be slower and simply not an option for some PFNMAP users). The | |
590 | * advantage is that we don't have to follow the strict linearity rule of | |
591 | * PFNMAP mappings in order to support COWable mappings. | |
592 | * | |
ee498ed7 | 593 | */ |
25b2995a CH |
594 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
595 | pte_t pte) | |
ee498ed7 | 596 | { |
22b31eec | 597 | unsigned long pfn = pte_pfn(pte); |
7e675137 | 598 | |
00b3a331 | 599 | if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) { |
b38af472 | 600 | if (likely(!pte_special(pte))) |
22b31eec | 601 | goto check_pfn; |
667a0a06 DV |
602 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
603 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
604 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
605 | return NULL; | |
df6ad698 JG |
606 | if (is_zero_pfn(pfn)) |
607 | return NULL; | |
e1fb4a08 DJ |
608 | if (pte_devmap(pte)) |
609 | return NULL; | |
610 | ||
df6ad698 | 611 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
612 | return NULL; |
613 | } | |
614 | ||
00b3a331 | 615 | /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */ |
7e675137 | 616 | |
b379d790 JH |
617 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
618 | if (vma->vm_flags & VM_MIXEDMAP) { | |
619 | if (!pfn_valid(pfn)) | |
620 | return NULL; | |
621 | goto out; | |
622 | } else { | |
7e675137 NP |
623 | unsigned long off; |
624 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
625 | if (pfn == vma->vm_pgoff + off) |
626 | return NULL; | |
627 | if (!is_cow_mapping(vma->vm_flags)) | |
628 | return NULL; | |
629 | } | |
6aab341e LT |
630 | } |
631 | ||
b38af472 HD |
632 | if (is_zero_pfn(pfn)) |
633 | return NULL; | |
00b3a331 | 634 | |
22b31eec HD |
635 | check_pfn: |
636 | if (unlikely(pfn > highest_memmap_pfn)) { | |
637 | print_bad_pte(vma, addr, pte, NULL); | |
638 | return NULL; | |
639 | } | |
6aab341e LT |
640 | |
641 | /* | |
7e675137 | 642 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 643 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 644 | */ |
b379d790 | 645 | out: |
6aab341e | 646 | return pfn_to_page(pfn); |
ee498ed7 HD |
647 | } |
648 | ||
28093f9f GS |
649 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
650 | struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, | |
651 | pmd_t pmd) | |
652 | { | |
653 | unsigned long pfn = pmd_pfn(pmd); | |
654 | ||
655 | /* | |
656 | * There is no pmd_special() but there may be special pmds, e.g. | |
657 | * in a direct-access (dax) mapping, so let's just replicate the | |
00b3a331 | 658 | * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here. |
28093f9f GS |
659 | */ |
660 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { | |
661 | if (vma->vm_flags & VM_MIXEDMAP) { | |
662 | if (!pfn_valid(pfn)) | |
663 | return NULL; | |
664 | goto out; | |
665 | } else { | |
666 | unsigned long off; | |
667 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
668 | if (pfn == vma->vm_pgoff + off) | |
669 | return NULL; | |
670 | if (!is_cow_mapping(vma->vm_flags)) | |
671 | return NULL; | |
672 | } | |
673 | } | |
674 | ||
e1fb4a08 DJ |
675 | if (pmd_devmap(pmd)) |
676 | return NULL; | |
3cde287b | 677 | if (is_huge_zero_pmd(pmd)) |
28093f9f GS |
678 | return NULL; |
679 | if (unlikely(pfn > highest_memmap_pfn)) | |
680 | return NULL; | |
681 | ||
682 | /* | |
683 | * NOTE! We still have PageReserved() pages in the page tables. | |
684 | * eg. VDSO mappings can cause them to exist. | |
685 | */ | |
686 | out: | |
687 | return pfn_to_page(pfn); | |
688 | } | |
689 | #endif | |
690 | ||
1da177e4 LT |
691 | /* |
692 | * copy one vm_area from one task to the other. Assumes the page tables | |
693 | * already present in the new task to be cleared in the whole range | |
694 | * covered by this vma. | |
1da177e4 LT |
695 | */ |
696 | ||
df3a57d1 LT |
697 | static unsigned long |
698 | copy_nonpresent_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
b5810039 | 699 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 700 | unsigned long addr, int *rss) |
1da177e4 | 701 | { |
b5810039 | 702 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
703 | pte_t pte = *src_pte; |
704 | struct page *page; | |
df3a57d1 LT |
705 | swp_entry_t entry = pte_to_swp_entry(pte); |
706 | ||
707 | if (likely(!non_swap_entry(entry))) { | |
708 | if (swap_duplicate(entry) < 0) | |
709 | return entry.val; | |
710 | ||
711 | /* make sure dst_mm is on swapoff's mmlist. */ | |
712 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
713 | spin_lock(&mmlist_lock); | |
714 | if (list_empty(&dst_mm->mmlist)) | |
715 | list_add(&dst_mm->mmlist, | |
716 | &src_mm->mmlist); | |
717 | spin_unlock(&mmlist_lock); | |
718 | } | |
719 | rss[MM_SWAPENTS]++; | |
720 | } else if (is_migration_entry(entry)) { | |
721 | page = migration_entry_to_page(entry); | |
1da177e4 | 722 | |
df3a57d1 | 723 | rss[mm_counter(page)]++; |
5042db43 | 724 | |
df3a57d1 LT |
725 | if (is_write_migration_entry(entry) && |
726 | is_cow_mapping(vm_flags)) { | |
5042db43 | 727 | /* |
df3a57d1 LT |
728 | * COW mappings require pages in both |
729 | * parent and child to be set to read. | |
5042db43 | 730 | */ |
df3a57d1 LT |
731 | make_migration_entry_read(&entry); |
732 | pte = swp_entry_to_pte(entry); | |
733 | if (pte_swp_soft_dirty(*src_pte)) | |
734 | pte = pte_swp_mksoft_dirty(pte); | |
735 | if (pte_swp_uffd_wp(*src_pte)) | |
736 | pte = pte_swp_mkuffd_wp(pte); | |
737 | set_pte_at(src_mm, addr, src_pte, pte); | |
738 | } | |
739 | } else if (is_device_private_entry(entry)) { | |
740 | page = device_private_entry_to_page(entry); | |
5042db43 | 741 | |
df3a57d1 LT |
742 | /* |
743 | * Update rss count even for unaddressable pages, as | |
744 | * they should treated just like normal pages in this | |
745 | * respect. | |
746 | * | |
747 | * We will likely want to have some new rss counters | |
748 | * for unaddressable pages, at some point. But for now | |
749 | * keep things as they are. | |
750 | */ | |
751 | get_page(page); | |
752 | rss[mm_counter(page)]++; | |
753 | page_dup_rmap(page, false); | |
754 | ||
755 | /* | |
756 | * We do not preserve soft-dirty information, because so | |
757 | * far, checkpoint/restore is the only feature that | |
758 | * requires that. And checkpoint/restore does not work | |
759 | * when a device driver is involved (you cannot easily | |
760 | * save and restore device driver state). | |
761 | */ | |
762 | if (is_write_device_private_entry(entry) && | |
763 | is_cow_mapping(vm_flags)) { | |
764 | make_device_private_entry_read(&entry); | |
765 | pte = swp_entry_to_pte(entry); | |
766 | if (pte_swp_uffd_wp(*src_pte)) | |
767 | pte = pte_swp_mkuffd_wp(pte); | |
768 | set_pte_at(src_mm, addr, src_pte, pte); | |
1da177e4 | 769 | } |
1da177e4 | 770 | } |
df3a57d1 LT |
771 | set_pte_at(dst_mm, addr, dst_pte, pte); |
772 | return 0; | |
773 | } | |
774 | ||
70e806e4 PX |
775 | /* |
776 | * Copy a present and normal page if necessary. | |
777 | * | |
778 | * NOTE! The usual case is that this doesn't need to do | |
779 | * anything, and can just return a positive value. That | |
780 | * will let the caller know that it can just increase | |
781 | * the page refcount and re-use the pte the traditional | |
782 | * way. | |
783 | * | |
784 | * But _if_ we need to copy it because it needs to be | |
785 | * pinned in the parent (and the child should get its own | |
786 | * copy rather than just a reference to the same page), | |
787 | * we'll do that here and return zero to let the caller | |
788 | * know we're done. | |
789 | * | |
790 | * And if we need a pre-allocated page but don't yet have | |
791 | * one, return a negative error to let the preallocation | |
792 | * code know so that it can do so outside the page table | |
793 | * lock. | |
794 | */ | |
795 | static inline int | |
c78f4636 PX |
796 | copy_present_page(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, |
797 | pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, | |
798 | struct page **prealloc, pte_t pte, struct page *page) | |
70e806e4 | 799 | { |
c78f4636 | 800 | struct mm_struct *src_mm = src_vma->vm_mm; |
70e806e4 PX |
801 | struct page *new_page; |
802 | ||
c78f4636 | 803 | if (!is_cow_mapping(src_vma->vm_flags)) |
70e806e4 PX |
804 | return 1; |
805 | ||
806 | /* | |
70e806e4 PX |
807 | * What we want to do is to check whether this page may |
808 | * have been pinned by the parent process. If so, | |
809 | * instead of wrprotect the pte on both sides, we copy | |
810 | * the page immediately so that we'll always guarantee | |
811 | * the pinned page won't be randomly replaced in the | |
812 | * future. | |
813 | * | |
f3c64eda LT |
814 | * The page pinning checks are just "has this mm ever |
815 | * seen pinning", along with the (inexact) check of | |
816 | * the page count. That might give false positives for | |
817 | * for pinning, but it will work correctly. | |
70e806e4 PX |
818 | */ |
819 | if (likely(!atomic_read(&src_mm->has_pinned))) | |
820 | return 1; | |
821 | if (likely(!page_maybe_dma_pinned(page))) | |
822 | return 1; | |
823 | ||
70e806e4 PX |
824 | new_page = *prealloc; |
825 | if (!new_page) | |
826 | return -EAGAIN; | |
827 | ||
828 | /* | |
829 | * We have a prealloc page, all good! Take it | |
830 | * over and copy the page & arm it. | |
831 | */ | |
832 | *prealloc = NULL; | |
c78f4636 | 833 | copy_user_highpage(new_page, page, addr, src_vma); |
70e806e4 | 834 | __SetPageUptodate(new_page); |
c78f4636 PX |
835 | page_add_new_anon_rmap(new_page, dst_vma, addr, false); |
836 | lru_cache_add_inactive_or_unevictable(new_page, dst_vma); | |
70e806e4 PX |
837 | rss[mm_counter(new_page)]++; |
838 | ||
839 | /* All done, just insert the new page copy in the child */ | |
c78f4636 PX |
840 | pte = mk_pte(new_page, dst_vma->vm_page_prot); |
841 | pte = maybe_mkwrite(pte_mkdirty(pte), dst_vma); | |
842 | set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); | |
70e806e4 PX |
843 | return 0; |
844 | } | |
845 | ||
846 | /* | |
847 | * Copy one pte. Returns 0 if succeeded, or -EAGAIN if one preallocated page | |
848 | * is required to copy this pte. | |
849 | */ | |
850 | static inline int | |
c78f4636 PX |
851 | copy_present_pte(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, |
852 | pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, | |
853 | struct page **prealloc) | |
df3a57d1 | 854 | { |
c78f4636 PX |
855 | struct mm_struct *src_mm = src_vma->vm_mm; |
856 | unsigned long vm_flags = src_vma->vm_flags; | |
df3a57d1 LT |
857 | pte_t pte = *src_pte; |
858 | struct page *page; | |
859 | ||
c78f4636 | 860 | page = vm_normal_page(src_vma, addr, pte); |
70e806e4 PX |
861 | if (page) { |
862 | int retval; | |
863 | ||
c78f4636 PX |
864 | retval = copy_present_page(dst_vma, src_vma, dst_pte, src_pte, |
865 | addr, rss, prealloc, pte, page); | |
70e806e4 PX |
866 | if (retval <= 0) |
867 | return retval; | |
868 | ||
869 | get_page(page); | |
870 | page_dup_rmap(page, false); | |
871 | rss[mm_counter(page)]++; | |
872 | } | |
873 | ||
1da177e4 LT |
874 | /* |
875 | * If it's a COW mapping, write protect it both | |
876 | * in the parent and the child | |
877 | */ | |
1b2de5d0 | 878 | if (is_cow_mapping(vm_flags) && pte_write(pte)) { |
1da177e4 | 879 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 880 | pte = pte_wrprotect(pte); |
1da177e4 LT |
881 | } |
882 | ||
883 | /* | |
884 | * If it's a shared mapping, mark it clean in | |
885 | * the child | |
886 | */ | |
887 | if (vm_flags & VM_SHARED) | |
888 | pte = pte_mkclean(pte); | |
889 | pte = pte_mkold(pte); | |
6aab341e | 890 | |
b569a176 PX |
891 | /* |
892 | * Make sure the _PAGE_UFFD_WP bit is cleared if the new VMA | |
893 | * does not have the VM_UFFD_WP, which means that the uffd | |
894 | * fork event is not enabled. | |
895 | */ | |
896 | if (!(vm_flags & VM_UFFD_WP)) | |
897 | pte = pte_clear_uffd_wp(pte); | |
898 | ||
c78f4636 | 899 | set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); |
70e806e4 PX |
900 | return 0; |
901 | } | |
902 | ||
903 | static inline struct page * | |
904 | page_copy_prealloc(struct mm_struct *src_mm, struct vm_area_struct *vma, | |
905 | unsigned long addr) | |
906 | { | |
907 | struct page *new_page; | |
908 | ||
909 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, addr); | |
910 | if (!new_page) | |
911 | return NULL; | |
912 | ||
913 | if (mem_cgroup_charge(new_page, src_mm, GFP_KERNEL)) { | |
914 | put_page(new_page); | |
915 | return NULL; | |
6aab341e | 916 | } |
70e806e4 | 917 | cgroup_throttle_swaprate(new_page, GFP_KERNEL); |
ae859762 | 918 | |
70e806e4 | 919 | return new_page; |
1da177e4 LT |
920 | } |
921 | ||
c78f4636 PX |
922 | static int |
923 | copy_pte_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
924 | pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr, | |
925 | unsigned long end) | |
1da177e4 | 926 | { |
c78f4636 PX |
927 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
928 | struct mm_struct *src_mm = src_vma->vm_mm; | |
c36987e2 | 929 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 930 | pte_t *src_pte, *dst_pte; |
c74df32c | 931 | spinlock_t *src_ptl, *dst_ptl; |
70e806e4 | 932 | int progress, ret = 0; |
d559db08 | 933 | int rss[NR_MM_COUNTERS]; |
570a335b | 934 | swp_entry_t entry = (swp_entry_t){0}; |
70e806e4 | 935 | struct page *prealloc = NULL; |
1da177e4 LT |
936 | |
937 | again: | |
70e806e4 | 938 | progress = 0; |
d559db08 KH |
939 | init_rss_vec(rss); |
940 | ||
c74df32c | 941 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
70e806e4 PX |
942 | if (!dst_pte) { |
943 | ret = -ENOMEM; | |
944 | goto out; | |
945 | } | |
ece0e2b6 | 946 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 947 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 948 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
949 | orig_src_pte = src_pte; |
950 | orig_dst_pte = dst_pte; | |
6606c3e0 | 951 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 952 | |
1da177e4 LT |
953 | do { |
954 | /* | |
955 | * We are holding two locks at this point - either of them | |
956 | * could generate latencies in another task on another CPU. | |
957 | */ | |
e040f218 HD |
958 | if (progress >= 32) { |
959 | progress = 0; | |
960 | if (need_resched() || | |
95c354fe | 961 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
962 | break; |
963 | } | |
1da177e4 LT |
964 | if (pte_none(*src_pte)) { |
965 | progress++; | |
966 | continue; | |
967 | } | |
79a1971c LT |
968 | if (unlikely(!pte_present(*src_pte))) { |
969 | entry.val = copy_nonpresent_pte(dst_mm, src_mm, | |
970 | dst_pte, src_pte, | |
c78f4636 | 971 | src_vma, addr, rss); |
79a1971c LT |
972 | if (entry.val) |
973 | break; | |
974 | progress += 8; | |
975 | continue; | |
976 | } | |
70e806e4 | 977 | /* copy_present_pte() will clear `*prealloc' if consumed */ |
c78f4636 PX |
978 | ret = copy_present_pte(dst_vma, src_vma, dst_pte, src_pte, |
979 | addr, rss, &prealloc); | |
70e806e4 PX |
980 | /* |
981 | * If we need a pre-allocated page for this pte, drop the | |
982 | * locks, allocate, and try again. | |
983 | */ | |
984 | if (unlikely(ret == -EAGAIN)) | |
985 | break; | |
986 | if (unlikely(prealloc)) { | |
987 | /* | |
988 | * pre-alloc page cannot be reused by next time so as | |
989 | * to strictly follow mempolicy (e.g., alloc_page_vma() | |
990 | * will allocate page according to address). This | |
991 | * could only happen if one pinned pte changed. | |
992 | */ | |
993 | put_page(prealloc); | |
994 | prealloc = NULL; | |
995 | } | |
1da177e4 LT |
996 | progress += 8; |
997 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 998 | |
6606c3e0 | 999 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1000 | spin_unlock(src_ptl); |
ece0e2b6 | 1001 | pte_unmap(orig_src_pte); |
d559db08 | 1002 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 1003 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 1004 | cond_resched(); |
570a335b HD |
1005 | |
1006 | if (entry.val) { | |
70e806e4 PX |
1007 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) { |
1008 | ret = -ENOMEM; | |
1009 | goto out; | |
1010 | } | |
1011 | entry.val = 0; | |
1012 | } else if (ret) { | |
1013 | WARN_ON_ONCE(ret != -EAGAIN); | |
c78f4636 | 1014 | prealloc = page_copy_prealloc(src_mm, src_vma, addr); |
70e806e4 | 1015 | if (!prealloc) |
570a335b | 1016 | return -ENOMEM; |
70e806e4 PX |
1017 | /* We've captured and resolved the error. Reset, try again. */ |
1018 | ret = 0; | |
570a335b | 1019 | } |
1da177e4 LT |
1020 | if (addr != end) |
1021 | goto again; | |
70e806e4 PX |
1022 | out: |
1023 | if (unlikely(prealloc)) | |
1024 | put_page(prealloc); | |
1025 | return ret; | |
1da177e4 LT |
1026 | } |
1027 | ||
c78f4636 PX |
1028 | static inline int |
1029 | copy_pmd_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1030 | pud_t *dst_pud, pud_t *src_pud, unsigned long addr, | |
1031 | unsigned long end) | |
1da177e4 | 1032 | { |
c78f4636 PX |
1033 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
1034 | struct mm_struct *src_mm = src_vma->vm_mm; | |
1da177e4 LT |
1035 | pmd_t *src_pmd, *dst_pmd; |
1036 | unsigned long next; | |
1037 | ||
1038 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
1039 | if (!dst_pmd) | |
1040 | return -ENOMEM; | |
1041 | src_pmd = pmd_offset(src_pud, addr); | |
1042 | do { | |
1043 | next = pmd_addr_end(addr, end); | |
84c3fc4e ZY |
1044 | if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) |
1045 | || pmd_devmap(*src_pmd)) { | |
71e3aac0 | 1046 | int err; |
c78f4636 | 1047 | VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, src_vma); |
71e3aac0 | 1048 | err = copy_huge_pmd(dst_mm, src_mm, |
c78f4636 | 1049 | dst_pmd, src_pmd, addr, src_vma); |
71e3aac0 AA |
1050 | if (err == -ENOMEM) |
1051 | return -ENOMEM; | |
1052 | if (!err) | |
1053 | continue; | |
1054 | /* fall through */ | |
1055 | } | |
1da177e4 LT |
1056 | if (pmd_none_or_clear_bad(src_pmd)) |
1057 | continue; | |
c78f4636 PX |
1058 | if (copy_pte_range(dst_vma, src_vma, dst_pmd, src_pmd, |
1059 | addr, next)) | |
1da177e4 LT |
1060 | return -ENOMEM; |
1061 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
1062 | return 0; | |
1063 | } | |
1064 | ||
c78f4636 PX |
1065 | static inline int |
1066 | copy_pud_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1067 | p4d_t *dst_p4d, p4d_t *src_p4d, unsigned long addr, | |
1068 | unsigned long end) | |
1da177e4 | 1069 | { |
c78f4636 PX |
1070 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
1071 | struct mm_struct *src_mm = src_vma->vm_mm; | |
1da177e4 LT |
1072 | pud_t *src_pud, *dst_pud; |
1073 | unsigned long next; | |
1074 | ||
c2febafc | 1075 | dst_pud = pud_alloc(dst_mm, dst_p4d, addr); |
1da177e4 LT |
1076 | if (!dst_pud) |
1077 | return -ENOMEM; | |
c2febafc | 1078 | src_pud = pud_offset(src_p4d, addr); |
1da177e4 LT |
1079 | do { |
1080 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1081 | if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { |
1082 | int err; | |
1083 | ||
c78f4636 | 1084 | VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, src_vma); |
a00cc7d9 | 1085 | err = copy_huge_pud(dst_mm, src_mm, |
c78f4636 | 1086 | dst_pud, src_pud, addr, src_vma); |
a00cc7d9 MW |
1087 | if (err == -ENOMEM) |
1088 | return -ENOMEM; | |
1089 | if (!err) | |
1090 | continue; | |
1091 | /* fall through */ | |
1092 | } | |
1da177e4 LT |
1093 | if (pud_none_or_clear_bad(src_pud)) |
1094 | continue; | |
c78f4636 PX |
1095 | if (copy_pmd_range(dst_vma, src_vma, dst_pud, src_pud, |
1096 | addr, next)) | |
1da177e4 LT |
1097 | return -ENOMEM; |
1098 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
1099 | return 0; | |
1100 | } | |
1101 | ||
c78f4636 PX |
1102 | static inline int |
1103 | copy_p4d_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1104 | pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long addr, | |
1105 | unsigned long end) | |
c2febafc | 1106 | { |
c78f4636 | 1107 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
c2febafc KS |
1108 | p4d_t *src_p4d, *dst_p4d; |
1109 | unsigned long next; | |
1110 | ||
1111 | dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); | |
1112 | if (!dst_p4d) | |
1113 | return -ENOMEM; | |
1114 | src_p4d = p4d_offset(src_pgd, addr); | |
1115 | do { | |
1116 | next = p4d_addr_end(addr, end); | |
1117 | if (p4d_none_or_clear_bad(src_p4d)) | |
1118 | continue; | |
c78f4636 PX |
1119 | if (copy_pud_range(dst_vma, src_vma, dst_p4d, src_p4d, |
1120 | addr, next)) | |
c2febafc KS |
1121 | return -ENOMEM; |
1122 | } while (dst_p4d++, src_p4d++, addr = next, addr != end); | |
1123 | return 0; | |
1124 | } | |
1125 | ||
c78f4636 PX |
1126 | int |
1127 | copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) | |
1da177e4 LT |
1128 | { |
1129 | pgd_t *src_pgd, *dst_pgd; | |
1130 | unsigned long next; | |
c78f4636 PX |
1131 | unsigned long addr = src_vma->vm_start; |
1132 | unsigned long end = src_vma->vm_end; | |
1133 | struct mm_struct *dst_mm = dst_vma->vm_mm; | |
1134 | struct mm_struct *src_mm = src_vma->vm_mm; | |
ac46d4f3 | 1135 | struct mmu_notifier_range range; |
2ec74c3e | 1136 | bool is_cow; |
cddb8a5c | 1137 | int ret; |
1da177e4 | 1138 | |
d992895b NP |
1139 | /* |
1140 | * Don't copy ptes where a page fault will fill them correctly. | |
1141 | * Fork becomes much lighter when there are big shared or private | |
1142 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1143 | * efficient than faulting. | |
1144 | */ | |
c78f4636 PX |
1145 | if (!(src_vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
1146 | !src_vma->anon_vma) | |
0661a336 | 1147 | return 0; |
d992895b | 1148 | |
c78f4636 PX |
1149 | if (is_vm_hugetlb_page(src_vma)) |
1150 | return copy_hugetlb_page_range(dst_mm, src_mm, src_vma); | |
1da177e4 | 1151 | |
c78f4636 | 1152 | if (unlikely(src_vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1153 | /* |
1154 | * We do not free on error cases below as remove_vma | |
1155 | * gets called on error from higher level routine | |
1156 | */ | |
c78f4636 | 1157 | ret = track_pfn_copy(src_vma); |
2ab64037 | 1158 | if (ret) |
1159 | return ret; | |
1160 | } | |
1161 | ||
cddb8a5c AA |
1162 | /* |
1163 | * We need to invalidate the secondary MMU mappings only when | |
1164 | * there could be a permission downgrade on the ptes of the | |
1165 | * parent mm. And a permission downgrade will only happen if | |
1166 | * is_cow_mapping() returns true. | |
1167 | */ | |
c78f4636 | 1168 | is_cow = is_cow_mapping(src_vma->vm_flags); |
ac46d4f3 JG |
1169 | |
1170 | if (is_cow) { | |
7269f999 | 1171 | mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_PAGE, |
c78f4636 | 1172 | 0, src_vma, src_mm, addr, end); |
ac46d4f3 | 1173 | mmu_notifier_invalidate_range_start(&range); |
57efa1fe JG |
1174 | /* |
1175 | * Disabling preemption is not needed for the write side, as | |
1176 | * the read side doesn't spin, but goes to the mmap_lock. | |
1177 | * | |
1178 | * Use the raw variant of the seqcount_t write API to avoid | |
1179 | * lockdep complaining about preemptibility. | |
1180 | */ | |
1181 | mmap_assert_write_locked(src_mm); | |
1182 | raw_write_seqcount_begin(&src_mm->write_protect_seq); | |
ac46d4f3 | 1183 | } |
cddb8a5c AA |
1184 | |
1185 | ret = 0; | |
1da177e4 LT |
1186 | dst_pgd = pgd_offset(dst_mm, addr); |
1187 | src_pgd = pgd_offset(src_mm, addr); | |
1188 | do { | |
1189 | next = pgd_addr_end(addr, end); | |
1190 | if (pgd_none_or_clear_bad(src_pgd)) | |
1191 | continue; | |
c78f4636 PX |
1192 | if (unlikely(copy_p4d_range(dst_vma, src_vma, dst_pgd, src_pgd, |
1193 | addr, next))) { | |
cddb8a5c AA |
1194 | ret = -ENOMEM; |
1195 | break; | |
1196 | } | |
1da177e4 | 1197 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1198 | |
57efa1fe JG |
1199 | if (is_cow) { |
1200 | raw_write_seqcount_end(&src_mm->write_protect_seq); | |
ac46d4f3 | 1201 | mmu_notifier_invalidate_range_end(&range); |
57efa1fe | 1202 | } |
cddb8a5c | 1203 | return ret; |
1da177e4 LT |
1204 | } |
1205 | ||
51c6f666 | 1206 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1207 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1208 | unsigned long addr, unsigned long end, |
97a89413 | 1209 | struct zap_details *details) |
1da177e4 | 1210 | { |
b5810039 | 1211 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1212 | int force_flush = 0; |
d559db08 | 1213 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1214 | spinlock_t *ptl; |
5f1a1907 | 1215 | pte_t *start_pte; |
97a89413 | 1216 | pte_t *pte; |
8a5f14a2 | 1217 | swp_entry_t entry; |
d559db08 | 1218 | |
ed6a7935 | 1219 | tlb_change_page_size(tlb, PAGE_SIZE); |
d16dfc55 | 1220 | again: |
e303297e | 1221 | init_rss_vec(rss); |
5f1a1907 SR |
1222 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1223 | pte = start_pte; | |
3ea27719 | 1224 | flush_tlb_batched_pending(mm); |
6606c3e0 | 1225 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1226 | do { |
1227 | pte_t ptent = *pte; | |
166f61b9 | 1228 | if (pte_none(ptent)) |
1da177e4 | 1229 | continue; |
6f5e6b9e | 1230 | |
7b167b68 MK |
1231 | if (need_resched()) |
1232 | break; | |
1233 | ||
1da177e4 | 1234 | if (pte_present(ptent)) { |
ee498ed7 | 1235 | struct page *page; |
51c6f666 | 1236 | |
25b2995a | 1237 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
1238 | if (unlikely(details) && page) { |
1239 | /* | |
1240 | * unmap_shared_mapping_pages() wants to | |
1241 | * invalidate cache without truncating: | |
1242 | * unmap shared but keep private pages. | |
1243 | */ | |
1244 | if (details->check_mapping && | |
800d8c63 | 1245 | details->check_mapping != page_rmapping(page)) |
1da177e4 | 1246 | continue; |
1da177e4 | 1247 | } |
b5810039 | 1248 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1249 | tlb->fullmm); |
1da177e4 LT |
1250 | tlb_remove_tlb_entry(tlb, pte, addr); |
1251 | if (unlikely(!page)) | |
1252 | continue; | |
eca56ff9 JM |
1253 | |
1254 | if (!PageAnon(page)) { | |
1cf35d47 LT |
1255 | if (pte_dirty(ptent)) { |
1256 | force_flush = 1; | |
6237bcd9 | 1257 | set_page_dirty(page); |
1cf35d47 | 1258 | } |
4917e5d0 | 1259 | if (pte_young(ptent) && |
64363aad | 1260 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1261 | mark_page_accessed(page); |
6237bcd9 | 1262 | } |
eca56ff9 | 1263 | rss[mm_counter(page)]--; |
d281ee61 | 1264 | page_remove_rmap(page, false); |
3dc14741 HD |
1265 | if (unlikely(page_mapcount(page) < 0)) |
1266 | print_bad_pte(vma, addr, ptent, page); | |
e9d55e15 | 1267 | if (unlikely(__tlb_remove_page(tlb, page))) { |
1cf35d47 | 1268 | force_flush = 1; |
ce9ec37b | 1269 | addr += PAGE_SIZE; |
d16dfc55 | 1270 | break; |
1cf35d47 | 1271 | } |
1da177e4 LT |
1272 | continue; |
1273 | } | |
5042db43 JG |
1274 | |
1275 | entry = pte_to_swp_entry(ptent); | |
463b7a17 | 1276 | if (is_device_private_entry(entry)) { |
5042db43 JG |
1277 | struct page *page = device_private_entry_to_page(entry); |
1278 | ||
1279 | if (unlikely(details && details->check_mapping)) { | |
1280 | /* | |
1281 | * unmap_shared_mapping_pages() wants to | |
1282 | * invalidate cache without truncating: | |
1283 | * unmap shared but keep private pages. | |
1284 | */ | |
1285 | if (details->check_mapping != | |
1286 | page_rmapping(page)) | |
1287 | continue; | |
1288 | } | |
1289 | ||
1290 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
1291 | rss[mm_counter(page)]--; | |
1292 | page_remove_rmap(page, false); | |
1293 | put_page(page); | |
1294 | continue; | |
1295 | } | |
1296 | ||
3e8715fd KS |
1297 | /* If details->check_mapping, we leave swap entries. */ |
1298 | if (unlikely(details)) | |
1da177e4 | 1299 | continue; |
b084d435 | 1300 | |
8a5f14a2 KS |
1301 | if (!non_swap_entry(entry)) |
1302 | rss[MM_SWAPENTS]--; | |
1303 | else if (is_migration_entry(entry)) { | |
1304 | struct page *page; | |
9f9f1acd | 1305 | |
8a5f14a2 | 1306 | page = migration_entry_to_page(entry); |
eca56ff9 | 1307 | rss[mm_counter(page)]--; |
b084d435 | 1308 | } |
8a5f14a2 KS |
1309 | if (unlikely(!free_swap_and_cache(entry))) |
1310 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 1311 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1312 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1313 | |
d559db08 | 1314 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1315 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1316 | |
1cf35d47 | 1317 | /* Do the actual TLB flush before dropping ptl */ |
fb7332a9 | 1318 | if (force_flush) |
1cf35d47 | 1319 | tlb_flush_mmu_tlbonly(tlb); |
1cf35d47 LT |
1320 | pte_unmap_unlock(start_pte, ptl); |
1321 | ||
1322 | /* | |
1323 | * If we forced a TLB flush (either due to running out of | |
1324 | * batch buffers or because we needed to flush dirty TLB | |
1325 | * entries before releasing the ptl), free the batched | |
1326 | * memory too. Restart if we didn't do everything. | |
1327 | */ | |
1328 | if (force_flush) { | |
1329 | force_flush = 0; | |
fa0aafb8 | 1330 | tlb_flush_mmu(tlb); |
7b167b68 MK |
1331 | } |
1332 | ||
1333 | if (addr != end) { | |
1334 | cond_resched(); | |
1335 | goto again; | |
d16dfc55 PZ |
1336 | } |
1337 | ||
51c6f666 | 1338 | return addr; |
1da177e4 LT |
1339 | } |
1340 | ||
51c6f666 | 1341 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1342 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1343 | unsigned long addr, unsigned long end, |
97a89413 | 1344 | struct zap_details *details) |
1da177e4 LT |
1345 | { |
1346 | pmd_t *pmd; | |
1347 | unsigned long next; | |
1348 | ||
1349 | pmd = pmd_offset(pud, addr); | |
1350 | do { | |
1351 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 1352 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
53406ed1 | 1353 | if (next - addr != HPAGE_PMD_SIZE) |
fd60775a | 1354 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
53406ed1 | 1355 | else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1356 | goto next; |
71e3aac0 AA |
1357 | /* fall through */ |
1358 | } | |
1a5a9906 AA |
1359 | /* |
1360 | * Here there can be other concurrent MADV_DONTNEED or | |
1361 | * trans huge page faults running, and if the pmd is | |
1362 | * none or trans huge it can change under us. This is | |
c1e8d7c6 | 1363 | * because MADV_DONTNEED holds the mmap_lock in read |
1a5a9906 AA |
1364 | * mode. |
1365 | */ | |
1366 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1367 | goto next; | |
97a89413 | 1368 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1369 | next: |
97a89413 PZ |
1370 | cond_resched(); |
1371 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1372 | |
1373 | return addr; | |
1da177e4 LT |
1374 | } |
1375 | ||
51c6f666 | 1376 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
c2febafc | 1377 | struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1378 | unsigned long addr, unsigned long end, |
97a89413 | 1379 | struct zap_details *details) |
1da177e4 LT |
1380 | { |
1381 | pud_t *pud; | |
1382 | unsigned long next; | |
1383 | ||
c2febafc | 1384 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1385 | do { |
1386 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1387 | if (pud_trans_huge(*pud) || pud_devmap(*pud)) { |
1388 | if (next - addr != HPAGE_PUD_SIZE) { | |
42fc5414 | 1389 | mmap_assert_locked(tlb->mm); |
a00cc7d9 MW |
1390 | split_huge_pud(vma, pud, addr); |
1391 | } else if (zap_huge_pud(tlb, vma, pud, addr)) | |
1392 | goto next; | |
1393 | /* fall through */ | |
1394 | } | |
97a89413 | 1395 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1396 | continue; |
97a89413 | 1397 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
a00cc7d9 MW |
1398 | next: |
1399 | cond_resched(); | |
97a89413 | 1400 | } while (pud++, addr = next, addr != end); |
51c6f666 RH |
1401 | |
1402 | return addr; | |
1da177e4 LT |
1403 | } |
1404 | ||
c2febafc KS |
1405 | static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, |
1406 | struct vm_area_struct *vma, pgd_t *pgd, | |
1407 | unsigned long addr, unsigned long end, | |
1408 | struct zap_details *details) | |
1409 | { | |
1410 | p4d_t *p4d; | |
1411 | unsigned long next; | |
1412 | ||
1413 | p4d = p4d_offset(pgd, addr); | |
1414 | do { | |
1415 | next = p4d_addr_end(addr, end); | |
1416 | if (p4d_none_or_clear_bad(p4d)) | |
1417 | continue; | |
1418 | next = zap_pud_range(tlb, vma, p4d, addr, next, details); | |
1419 | } while (p4d++, addr = next, addr != end); | |
1420 | ||
1421 | return addr; | |
1422 | } | |
1423 | ||
aac45363 | 1424 | void unmap_page_range(struct mmu_gather *tlb, |
038c7aa1 AV |
1425 | struct vm_area_struct *vma, |
1426 | unsigned long addr, unsigned long end, | |
1427 | struct zap_details *details) | |
1da177e4 LT |
1428 | { |
1429 | pgd_t *pgd; | |
1430 | unsigned long next; | |
1431 | ||
1da177e4 LT |
1432 | BUG_ON(addr >= end); |
1433 | tlb_start_vma(tlb, vma); | |
1434 | pgd = pgd_offset(vma->vm_mm, addr); | |
1435 | do { | |
1436 | next = pgd_addr_end(addr, end); | |
97a89413 | 1437 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1438 | continue; |
c2febafc | 1439 | next = zap_p4d_range(tlb, vma, pgd, addr, next, details); |
97a89413 | 1440 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
1441 | tlb_end_vma(tlb, vma); |
1442 | } | |
51c6f666 | 1443 | |
f5cc4eef AV |
1444 | |
1445 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1446 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1447 | unsigned long end_addr, |
f5cc4eef AV |
1448 | struct zap_details *details) |
1449 | { | |
1450 | unsigned long start = max(vma->vm_start, start_addr); | |
1451 | unsigned long end; | |
1452 | ||
1453 | if (start >= vma->vm_end) | |
1454 | return; | |
1455 | end = min(vma->vm_end, end_addr); | |
1456 | if (end <= vma->vm_start) | |
1457 | return; | |
1458 | ||
cbc91f71 SD |
1459 | if (vma->vm_file) |
1460 | uprobe_munmap(vma, start, end); | |
1461 | ||
b3b9c293 | 1462 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1463 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1464 | |
1465 | if (start != end) { | |
1466 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1467 | /* | |
1468 | * It is undesirable to test vma->vm_file as it | |
1469 | * should be non-null for valid hugetlb area. | |
1470 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1471 | * cleanup path of mmap_region. When |
f5cc4eef | 1472 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1473 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1474 | * before calling this function to clean up. |
1475 | * Since no pte has actually been setup, it is | |
1476 | * safe to do nothing in this case. | |
1477 | */ | |
24669e58 | 1478 | if (vma->vm_file) { |
83cde9e8 | 1479 | i_mmap_lock_write(vma->vm_file->f_mapping); |
d833352a | 1480 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
83cde9e8 | 1481 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
24669e58 | 1482 | } |
f5cc4eef AV |
1483 | } else |
1484 | unmap_page_range(tlb, vma, start, end, details); | |
1485 | } | |
1da177e4 LT |
1486 | } |
1487 | ||
1da177e4 LT |
1488 | /** |
1489 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1490 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1491 | * @vma: the starting vma |
1492 | * @start_addr: virtual address at which to start unmapping | |
1493 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1494 | * |
508034a3 | 1495 | * Unmap all pages in the vma list. |
1da177e4 | 1496 | * |
1da177e4 LT |
1497 | * Only addresses between `start' and `end' will be unmapped. |
1498 | * | |
1499 | * The VMA list must be sorted in ascending virtual address order. | |
1500 | * | |
1501 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1502 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1503 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1504 | * drops the lock and schedules. | |
1505 | */ | |
6e8bb019 | 1506 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1507 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1508 | unsigned long end_addr) |
1da177e4 | 1509 | { |
ac46d4f3 | 1510 | struct mmu_notifier_range range; |
1da177e4 | 1511 | |
6f4f13e8 JG |
1512 | mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, vma->vm_mm, |
1513 | start_addr, end_addr); | |
ac46d4f3 | 1514 | mmu_notifier_invalidate_range_start(&range); |
f5cc4eef | 1515 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1516 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
ac46d4f3 | 1517 | mmu_notifier_invalidate_range_end(&range); |
1da177e4 LT |
1518 | } |
1519 | ||
1520 | /** | |
1521 | * zap_page_range - remove user pages in a given range | |
1522 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1523 | * @start: starting address of pages to zap |
1da177e4 | 1524 | * @size: number of bytes to zap |
f5cc4eef AV |
1525 | * |
1526 | * Caller must protect the VMA list | |
1da177e4 | 1527 | */ |
7e027b14 | 1528 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
ecf1385d | 1529 | unsigned long size) |
1da177e4 | 1530 | { |
ac46d4f3 | 1531 | struct mmu_notifier_range range; |
d16dfc55 | 1532 | struct mmu_gather tlb; |
1da177e4 | 1533 | |
1da177e4 | 1534 | lru_add_drain(); |
7269f999 | 1535 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, vma->vm_mm, |
6f4f13e8 | 1536 | start, start + size); |
ac46d4f3 JG |
1537 | tlb_gather_mmu(&tlb, vma->vm_mm, start, range.end); |
1538 | update_hiwater_rss(vma->vm_mm); | |
1539 | mmu_notifier_invalidate_range_start(&range); | |
1540 | for ( ; vma && vma->vm_start < range.end; vma = vma->vm_next) | |
1541 | unmap_single_vma(&tlb, vma, start, range.end, NULL); | |
1542 | mmu_notifier_invalidate_range_end(&range); | |
1543 | tlb_finish_mmu(&tlb, start, range.end); | |
1da177e4 | 1544 | } |
2e0647d3 | 1545 | EXPORT_SYMBOL(zap_page_range); |
1da177e4 | 1546 | |
f5cc4eef AV |
1547 | /** |
1548 | * zap_page_range_single - remove user pages in a given range | |
1549 | * @vma: vm_area_struct holding the applicable pages | |
1550 | * @address: starting address of pages to zap | |
1551 | * @size: number of bytes to zap | |
8a5f14a2 | 1552 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1553 | * |
1554 | * The range must fit into one VMA. | |
1da177e4 | 1555 | */ |
f5cc4eef | 1556 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1557 | unsigned long size, struct zap_details *details) |
1558 | { | |
ac46d4f3 | 1559 | struct mmu_notifier_range range; |
d16dfc55 | 1560 | struct mmu_gather tlb; |
1da177e4 | 1561 | |
1da177e4 | 1562 | lru_add_drain(); |
7269f999 | 1563 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, vma->vm_mm, |
6f4f13e8 | 1564 | address, address + size); |
ac46d4f3 JG |
1565 | tlb_gather_mmu(&tlb, vma->vm_mm, address, range.end); |
1566 | update_hiwater_rss(vma->vm_mm); | |
1567 | mmu_notifier_invalidate_range_start(&range); | |
1568 | unmap_single_vma(&tlb, vma, address, range.end, details); | |
1569 | mmu_notifier_invalidate_range_end(&range); | |
1570 | tlb_finish_mmu(&tlb, address, range.end); | |
1da177e4 LT |
1571 | } |
1572 | ||
c627f9cc JS |
1573 | /** |
1574 | * zap_vma_ptes - remove ptes mapping the vma | |
1575 | * @vma: vm_area_struct holding ptes to be zapped | |
1576 | * @address: starting address of pages to zap | |
1577 | * @size: number of bytes to zap | |
1578 | * | |
1579 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1580 | * | |
1581 | * The entire address range must be fully contained within the vma. | |
1582 | * | |
c627f9cc | 1583 | */ |
27d036e3 | 1584 | void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, |
c627f9cc JS |
1585 | unsigned long size) |
1586 | { | |
1587 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1588 | !(vma->vm_flags & VM_PFNMAP)) | |
27d036e3 LR |
1589 | return; |
1590 | ||
f5cc4eef | 1591 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1592 | } |
1593 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1594 | ||
8cd3984d | 1595 | static pmd_t *walk_to_pmd(struct mm_struct *mm, unsigned long addr) |
c9cfcddf | 1596 | { |
c2febafc KS |
1597 | pgd_t *pgd; |
1598 | p4d_t *p4d; | |
1599 | pud_t *pud; | |
1600 | pmd_t *pmd; | |
1601 | ||
1602 | pgd = pgd_offset(mm, addr); | |
1603 | p4d = p4d_alloc(mm, pgd, addr); | |
1604 | if (!p4d) | |
1605 | return NULL; | |
1606 | pud = pud_alloc(mm, p4d, addr); | |
1607 | if (!pud) | |
1608 | return NULL; | |
1609 | pmd = pmd_alloc(mm, pud, addr); | |
1610 | if (!pmd) | |
1611 | return NULL; | |
1612 | ||
1613 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
8cd3984d AR |
1614 | return pmd; |
1615 | } | |
1616 | ||
1617 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, | |
1618 | spinlock_t **ptl) | |
1619 | { | |
1620 | pmd_t *pmd = walk_to_pmd(mm, addr); | |
1621 | ||
1622 | if (!pmd) | |
1623 | return NULL; | |
c2febafc | 1624 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
c9cfcddf LT |
1625 | } |
1626 | ||
8efd6f5b AR |
1627 | static int validate_page_before_insert(struct page *page) |
1628 | { | |
1629 | if (PageAnon(page) || PageSlab(page) || page_has_type(page)) | |
1630 | return -EINVAL; | |
1631 | flush_dcache_page(page); | |
1632 | return 0; | |
1633 | } | |
1634 | ||
1635 | static int insert_page_into_pte_locked(struct mm_struct *mm, pte_t *pte, | |
1636 | unsigned long addr, struct page *page, pgprot_t prot) | |
1637 | { | |
1638 | if (!pte_none(*pte)) | |
1639 | return -EBUSY; | |
1640 | /* Ok, finally just insert the thing.. */ | |
1641 | get_page(page); | |
1642 | inc_mm_counter_fast(mm, mm_counter_file(page)); | |
1643 | page_add_file_rmap(page, false); | |
1644 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); | |
1645 | return 0; | |
1646 | } | |
1647 | ||
238f58d8 LT |
1648 | /* |
1649 | * This is the old fallback for page remapping. | |
1650 | * | |
1651 | * For historical reasons, it only allows reserved pages. Only | |
1652 | * old drivers should use this, and they needed to mark their | |
1653 | * pages reserved for the old functions anyway. | |
1654 | */ | |
423bad60 NP |
1655 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1656 | struct page *page, pgprot_t prot) | |
238f58d8 | 1657 | { |
423bad60 | 1658 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1659 | int retval; |
c9cfcddf | 1660 | pte_t *pte; |
8a9f3ccd BS |
1661 | spinlock_t *ptl; |
1662 | ||
8efd6f5b AR |
1663 | retval = validate_page_before_insert(page); |
1664 | if (retval) | |
5b4e655e | 1665 | goto out; |
238f58d8 | 1666 | retval = -ENOMEM; |
c9cfcddf | 1667 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1668 | if (!pte) |
5b4e655e | 1669 | goto out; |
8efd6f5b | 1670 | retval = insert_page_into_pte_locked(mm, pte, addr, page, prot); |
238f58d8 LT |
1671 | pte_unmap_unlock(pte, ptl); |
1672 | out: | |
1673 | return retval; | |
1674 | } | |
1675 | ||
8cd3984d | 1676 | #ifdef pte_index |
7f70c2a6 | 1677 | static int insert_page_in_batch_locked(struct mm_struct *mm, pte_t *pte, |
8cd3984d AR |
1678 | unsigned long addr, struct page *page, pgprot_t prot) |
1679 | { | |
1680 | int err; | |
1681 | ||
1682 | if (!page_count(page)) | |
1683 | return -EINVAL; | |
1684 | err = validate_page_before_insert(page); | |
7f70c2a6 AR |
1685 | if (err) |
1686 | return err; | |
1687 | return insert_page_into_pte_locked(mm, pte, addr, page, prot); | |
8cd3984d AR |
1688 | } |
1689 | ||
1690 | /* insert_pages() amortizes the cost of spinlock operations | |
1691 | * when inserting pages in a loop. Arch *must* define pte_index. | |
1692 | */ | |
1693 | static int insert_pages(struct vm_area_struct *vma, unsigned long addr, | |
1694 | struct page **pages, unsigned long *num, pgprot_t prot) | |
1695 | { | |
1696 | pmd_t *pmd = NULL; | |
7f70c2a6 AR |
1697 | pte_t *start_pte, *pte; |
1698 | spinlock_t *pte_lock; | |
8cd3984d AR |
1699 | struct mm_struct *const mm = vma->vm_mm; |
1700 | unsigned long curr_page_idx = 0; | |
1701 | unsigned long remaining_pages_total = *num; | |
1702 | unsigned long pages_to_write_in_pmd; | |
1703 | int ret; | |
1704 | more: | |
1705 | ret = -EFAULT; | |
1706 | pmd = walk_to_pmd(mm, addr); | |
1707 | if (!pmd) | |
1708 | goto out; | |
1709 | ||
1710 | pages_to_write_in_pmd = min_t(unsigned long, | |
1711 | remaining_pages_total, PTRS_PER_PTE - pte_index(addr)); | |
1712 | ||
1713 | /* Allocate the PTE if necessary; takes PMD lock once only. */ | |
1714 | ret = -ENOMEM; | |
1715 | if (pte_alloc(mm, pmd)) | |
1716 | goto out; | |
8cd3984d AR |
1717 | |
1718 | while (pages_to_write_in_pmd) { | |
1719 | int pte_idx = 0; | |
1720 | const int batch_size = min_t(int, pages_to_write_in_pmd, 8); | |
1721 | ||
7f70c2a6 AR |
1722 | start_pte = pte_offset_map_lock(mm, pmd, addr, &pte_lock); |
1723 | for (pte = start_pte; pte_idx < batch_size; ++pte, ++pte_idx) { | |
1724 | int err = insert_page_in_batch_locked(mm, pte, | |
8cd3984d AR |
1725 | addr, pages[curr_page_idx], prot); |
1726 | if (unlikely(err)) { | |
7f70c2a6 | 1727 | pte_unmap_unlock(start_pte, pte_lock); |
8cd3984d AR |
1728 | ret = err; |
1729 | remaining_pages_total -= pte_idx; | |
1730 | goto out; | |
1731 | } | |
1732 | addr += PAGE_SIZE; | |
1733 | ++curr_page_idx; | |
1734 | } | |
7f70c2a6 | 1735 | pte_unmap_unlock(start_pte, pte_lock); |
8cd3984d AR |
1736 | pages_to_write_in_pmd -= batch_size; |
1737 | remaining_pages_total -= batch_size; | |
1738 | } | |
1739 | if (remaining_pages_total) | |
1740 | goto more; | |
1741 | ret = 0; | |
1742 | out: | |
1743 | *num = remaining_pages_total; | |
1744 | return ret; | |
1745 | } | |
1746 | #endif /* ifdef pte_index */ | |
1747 | ||
1748 | /** | |
1749 | * vm_insert_pages - insert multiple pages into user vma, batching the pmd lock. | |
1750 | * @vma: user vma to map to | |
1751 | * @addr: target start user address of these pages | |
1752 | * @pages: source kernel pages | |
1753 | * @num: in: number of pages to map. out: number of pages that were *not* | |
1754 | * mapped. (0 means all pages were successfully mapped). | |
1755 | * | |
1756 | * Preferred over vm_insert_page() when inserting multiple pages. | |
1757 | * | |
1758 | * In case of error, we may have mapped a subset of the provided | |
1759 | * pages. It is the caller's responsibility to account for this case. | |
1760 | * | |
1761 | * The same restrictions apply as in vm_insert_page(). | |
1762 | */ | |
1763 | int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr, | |
1764 | struct page **pages, unsigned long *num) | |
1765 | { | |
1766 | #ifdef pte_index | |
1767 | const unsigned long end_addr = addr + (*num * PAGE_SIZE) - 1; | |
1768 | ||
1769 | if (addr < vma->vm_start || end_addr >= vma->vm_end) | |
1770 | return -EFAULT; | |
1771 | if (!(vma->vm_flags & VM_MIXEDMAP)) { | |
d8ed45c5 | 1772 | BUG_ON(mmap_read_trylock(vma->vm_mm)); |
8cd3984d AR |
1773 | BUG_ON(vma->vm_flags & VM_PFNMAP); |
1774 | vma->vm_flags |= VM_MIXEDMAP; | |
1775 | } | |
1776 | /* Defer page refcount checking till we're about to map that page. */ | |
1777 | return insert_pages(vma, addr, pages, num, vma->vm_page_prot); | |
1778 | #else | |
1779 | unsigned long idx = 0, pgcount = *num; | |
45779b03 | 1780 | int err = -EINVAL; |
8cd3984d AR |
1781 | |
1782 | for (; idx < pgcount; ++idx) { | |
1783 | err = vm_insert_page(vma, addr + (PAGE_SIZE * idx), pages[idx]); | |
1784 | if (err) | |
1785 | break; | |
1786 | } | |
1787 | *num = pgcount - idx; | |
1788 | return err; | |
1789 | #endif /* ifdef pte_index */ | |
1790 | } | |
1791 | EXPORT_SYMBOL(vm_insert_pages); | |
1792 | ||
bfa5bf6d REB |
1793 | /** |
1794 | * vm_insert_page - insert single page into user vma | |
1795 | * @vma: user vma to map to | |
1796 | * @addr: target user address of this page | |
1797 | * @page: source kernel page | |
1798 | * | |
a145dd41 LT |
1799 | * This allows drivers to insert individual pages they've allocated |
1800 | * into a user vma. | |
1801 | * | |
1802 | * The page has to be a nice clean _individual_ kernel allocation. | |
1803 | * If you allocate a compound page, you need to have marked it as | |
1804 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1805 | * (see split_page()). |
a145dd41 LT |
1806 | * |
1807 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1808 | * took an arbitrary page protection parameter. This doesn't allow | |
1809 | * that. Your vma protection will have to be set up correctly, which | |
1810 | * means that if you want a shared writable mapping, you'd better | |
1811 | * ask for a shared writable mapping! | |
1812 | * | |
1813 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1814 | * |
1815 | * Usually this function is called from f_op->mmap() handler | |
c1e8d7c6 | 1816 | * under mm->mmap_lock write-lock, so it can change vma->vm_flags. |
4b6e1e37 KK |
1817 | * Caller must set VM_MIXEDMAP on vma if it wants to call this |
1818 | * function from other places, for example from page-fault handler. | |
a862f68a MR |
1819 | * |
1820 | * Return: %0 on success, negative error code otherwise. | |
a145dd41 | 1821 | */ |
423bad60 NP |
1822 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1823 | struct page *page) | |
a145dd41 LT |
1824 | { |
1825 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1826 | return -EFAULT; | |
1827 | if (!page_count(page)) | |
1828 | return -EINVAL; | |
4b6e1e37 | 1829 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
d8ed45c5 | 1830 | BUG_ON(mmap_read_trylock(vma->vm_mm)); |
4b6e1e37 KK |
1831 | BUG_ON(vma->vm_flags & VM_PFNMAP); |
1832 | vma->vm_flags |= VM_MIXEDMAP; | |
1833 | } | |
423bad60 | 1834 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1835 | } |
e3c3374f | 1836 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1837 | |
a667d745 SJ |
1838 | /* |
1839 | * __vm_map_pages - maps range of kernel pages into user vma | |
1840 | * @vma: user vma to map to | |
1841 | * @pages: pointer to array of source kernel pages | |
1842 | * @num: number of pages in page array | |
1843 | * @offset: user's requested vm_pgoff | |
1844 | * | |
1845 | * This allows drivers to map range of kernel pages into a user vma. | |
1846 | * | |
1847 | * Return: 0 on success and error code otherwise. | |
1848 | */ | |
1849 | static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages, | |
1850 | unsigned long num, unsigned long offset) | |
1851 | { | |
1852 | unsigned long count = vma_pages(vma); | |
1853 | unsigned long uaddr = vma->vm_start; | |
1854 | int ret, i; | |
1855 | ||
1856 | /* Fail if the user requested offset is beyond the end of the object */ | |
96756fcb | 1857 | if (offset >= num) |
a667d745 SJ |
1858 | return -ENXIO; |
1859 | ||
1860 | /* Fail if the user requested size exceeds available object size */ | |
1861 | if (count > num - offset) | |
1862 | return -ENXIO; | |
1863 | ||
1864 | for (i = 0; i < count; i++) { | |
1865 | ret = vm_insert_page(vma, uaddr, pages[offset + i]); | |
1866 | if (ret < 0) | |
1867 | return ret; | |
1868 | uaddr += PAGE_SIZE; | |
1869 | } | |
1870 | ||
1871 | return 0; | |
1872 | } | |
1873 | ||
1874 | /** | |
1875 | * vm_map_pages - maps range of kernel pages starts with non zero offset | |
1876 | * @vma: user vma to map to | |
1877 | * @pages: pointer to array of source kernel pages | |
1878 | * @num: number of pages in page array | |
1879 | * | |
1880 | * Maps an object consisting of @num pages, catering for the user's | |
1881 | * requested vm_pgoff | |
1882 | * | |
1883 | * If we fail to insert any page into the vma, the function will return | |
1884 | * immediately leaving any previously inserted pages present. Callers | |
1885 | * from the mmap handler may immediately return the error as their caller | |
1886 | * will destroy the vma, removing any successfully inserted pages. Other | |
1887 | * callers should make their own arrangements for calling unmap_region(). | |
1888 | * | |
1889 | * Context: Process context. Called by mmap handlers. | |
1890 | * Return: 0 on success and error code otherwise. | |
1891 | */ | |
1892 | int vm_map_pages(struct vm_area_struct *vma, struct page **pages, | |
1893 | unsigned long num) | |
1894 | { | |
1895 | return __vm_map_pages(vma, pages, num, vma->vm_pgoff); | |
1896 | } | |
1897 | EXPORT_SYMBOL(vm_map_pages); | |
1898 | ||
1899 | /** | |
1900 | * vm_map_pages_zero - map range of kernel pages starts with zero offset | |
1901 | * @vma: user vma to map to | |
1902 | * @pages: pointer to array of source kernel pages | |
1903 | * @num: number of pages in page array | |
1904 | * | |
1905 | * Similar to vm_map_pages(), except that it explicitly sets the offset | |
1906 | * to 0. This function is intended for the drivers that did not consider | |
1907 | * vm_pgoff. | |
1908 | * | |
1909 | * Context: Process context. Called by mmap handlers. | |
1910 | * Return: 0 on success and error code otherwise. | |
1911 | */ | |
1912 | int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, | |
1913 | unsigned long num) | |
1914 | { | |
1915 | return __vm_map_pages(vma, pages, num, 0); | |
1916 | } | |
1917 | EXPORT_SYMBOL(vm_map_pages_zero); | |
1918 | ||
9b5a8e00 | 1919 | static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
b2770da6 | 1920 | pfn_t pfn, pgprot_t prot, bool mkwrite) |
423bad60 NP |
1921 | { |
1922 | struct mm_struct *mm = vma->vm_mm; | |
423bad60 NP |
1923 | pte_t *pte, entry; |
1924 | spinlock_t *ptl; | |
1925 | ||
423bad60 NP |
1926 | pte = get_locked_pte(mm, addr, &ptl); |
1927 | if (!pte) | |
9b5a8e00 | 1928 | return VM_FAULT_OOM; |
b2770da6 RZ |
1929 | if (!pte_none(*pte)) { |
1930 | if (mkwrite) { | |
1931 | /* | |
1932 | * For read faults on private mappings the PFN passed | |
1933 | * in may not match the PFN we have mapped if the | |
1934 | * mapped PFN is a writeable COW page. In the mkwrite | |
1935 | * case we are creating a writable PTE for a shared | |
f2c57d91 JK |
1936 | * mapping and we expect the PFNs to match. If they |
1937 | * don't match, we are likely racing with block | |
1938 | * allocation and mapping invalidation so just skip the | |
1939 | * update. | |
b2770da6 | 1940 | */ |
f2c57d91 JK |
1941 | if (pte_pfn(*pte) != pfn_t_to_pfn(pfn)) { |
1942 | WARN_ON_ONCE(!is_zero_pfn(pte_pfn(*pte))); | |
b2770da6 | 1943 | goto out_unlock; |
f2c57d91 | 1944 | } |
cae85cb8 JK |
1945 | entry = pte_mkyoung(*pte); |
1946 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1947 | if (ptep_set_access_flags(vma, addr, pte, entry, 1)) | |
1948 | update_mmu_cache(vma, addr, pte); | |
1949 | } | |
1950 | goto out_unlock; | |
b2770da6 | 1951 | } |
423bad60 NP |
1952 | |
1953 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
1954 | if (pfn_t_devmap(pfn)) |
1955 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
1956 | else | |
1957 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
b2770da6 | 1958 | |
b2770da6 RZ |
1959 | if (mkwrite) { |
1960 | entry = pte_mkyoung(entry); | |
1961 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1962 | } | |
1963 | ||
423bad60 | 1964 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 1965 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 | 1966 | |
423bad60 NP |
1967 | out_unlock: |
1968 | pte_unmap_unlock(pte, ptl); | |
9b5a8e00 | 1969 | return VM_FAULT_NOPAGE; |
423bad60 NP |
1970 | } |
1971 | ||
f5e6d1d5 MW |
1972 | /** |
1973 | * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
1974 | * @vma: user vma to map to | |
1975 | * @addr: target user address of this page | |
1976 | * @pfn: source kernel pfn | |
1977 | * @pgprot: pgprot flags for the inserted page | |
1978 | * | |
a1a0aea5 | 1979 | * This is exactly like vmf_insert_pfn(), except that it allows drivers |
f5e6d1d5 MW |
1980 | * to override pgprot on a per-page basis. |
1981 | * | |
1982 | * This only makes sense for IO mappings, and it makes no sense for | |
1983 | * COW mappings. In general, using multiple vmas is preferable; | |
ae2b01f3 | 1984 | * vmf_insert_pfn_prot should only be used if using multiple VMAs is |
f5e6d1d5 MW |
1985 | * impractical. |
1986 | * | |
574c5b3d TH |
1987 | * See vmf_insert_mixed_prot() for a discussion of the implication of using |
1988 | * a value of @pgprot different from that of @vma->vm_page_prot. | |
1989 | * | |
ae2b01f3 | 1990 | * Context: Process context. May allocate using %GFP_KERNEL. |
f5e6d1d5 MW |
1991 | * Return: vm_fault_t value. |
1992 | */ | |
1993 | vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
1994 | unsigned long pfn, pgprot_t pgprot) | |
1995 | { | |
6d958546 MW |
1996 | /* |
1997 | * Technically, architectures with pte_special can avoid all these | |
1998 | * restrictions (same for remap_pfn_range). However we would like | |
1999 | * consistency in testing and feature parity among all, so we should | |
2000 | * try to keep these invariants in place for everybody. | |
2001 | */ | |
2002 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); | |
2003 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
2004 | (VM_PFNMAP|VM_MIXEDMAP)); | |
2005 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
2006 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
2007 | ||
2008 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
2009 | return VM_FAULT_SIGBUS; | |
2010 | ||
2011 | if (!pfn_modify_allowed(pfn, pgprot)) | |
2012 | return VM_FAULT_SIGBUS; | |
2013 | ||
2014 | track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); | |
2015 | ||
9b5a8e00 | 2016 | return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, |
6d958546 | 2017 | false); |
f5e6d1d5 MW |
2018 | } |
2019 | EXPORT_SYMBOL(vmf_insert_pfn_prot); | |
e0dc0d8f | 2020 | |
ae2b01f3 MW |
2021 | /** |
2022 | * vmf_insert_pfn - insert single pfn into user vma | |
2023 | * @vma: user vma to map to | |
2024 | * @addr: target user address of this page | |
2025 | * @pfn: source kernel pfn | |
2026 | * | |
2027 | * Similar to vm_insert_page, this allows drivers to insert individual pages | |
2028 | * they've allocated into a user vma. Same comments apply. | |
2029 | * | |
2030 | * This function should only be called from a vm_ops->fault handler, and | |
2031 | * in that case the handler should return the result of this function. | |
2032 | * | |
2033 | * vma cannot be a COW mapping. | |
2034 | * | |
2035 | * As this is called only for pages that do not currently exist, we | |
2036 | * do not need to flush old virtual caches or the TLB. | |
2037 | * | |
2038 | * Context: Process context. May allocate using %GFP_KERNEL. | |
2039 | * Return: vm_fault_t value. | |
2040 | */ | |
2041 | vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
2042 | unsigned long pfn) | |
2043 | { | |
2044 | return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
2045 | } | |
2046 | EXPORT_SYMBOL(vmf_insert_pfn); | |
2047 | ||
785a3fab DW |
2048 | static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) |
2049 | { | |
2050 | /* these checks mirror the abort conditions in vm_normal_page */ | |
2051 | if (vma->vm_flags & VM_MIXEDMAP) | |
2052 | return true; | |
2053 | if (pfn_t_devmap(pfn)) | |
2054 | return true; | |
2055 | if (pfn_t_special(pfn)) | |
2056 | return true; | |
2057 | if (is_zero_pfn(pfn_t_to_pfn(pfn))) | |
2058 | return true; | |
2059 | return false; | |
2060 | } | |
2061 | ||
79f3aa5b | 2062 | static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma, |
574c5b3d TH |
2063 | unsigned long addr, pfn_t pfn, pgprot_t pgprot, |
2064 | bool mkwrite) | |
423bad60 | 2065 | { |
79f3aa5b | 2066 | int err; |
87744ab3 | 2067 | |
785a3fab | 2068 | BUG_ON(!vm_mixed_ok(vma, pfn)); |
e0dc0d8f | 2069 | |
423bad60 | 2070 | if (addr < vma->vm_start || addr >= vma->vm_end) |
79f3aa5b | 2071 | return VM_FAULT_SIGBUS; |
308a047c BP |
2072 | |
2073 | track_pfn_insert(vma, &pgprot, pfn); | |
e0dc0d8f | 2074 | |
42e4089c | 2075 | if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) |
79f3aa5b | 2076 | return VM_FAULT_SIGBUS; |
42e4089c | 2077 | |
423bad60 NP |
2078 | /* |
2079 | * If we don't have pte special, then we have to use the pfn_valid() | |
2080 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
2081 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
2082 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
2083 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 2084 | */ |
00b3a331 LD |
2085 | if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && |
2086 | !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { | |
423bad60 NP |
2087 | struct page *page; |
2088 | ||
03fc2da6 DW |
2089 | /* |
2090 | * At this point we are committed to insert_page() | |
2091 | * regardless of whether the caller specified flags that | |
2092 | * result in pfn_t_has_page() == false. | |
2093 | */ | |
2094 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
79f3aa5b MW |
2095 | err = insert_page(vma, addr, page, pgprot); |
2096 | } else { | |
9b5a8e00 | 2097 | return insert_pfn(vma, addr, pfn, pgprot, mkwrite); |
423bad60 | 2098 | } |
b2770da6 | 2099 | |
5d747637 MW |
2100 | if (err == -ENOMEM) |
2101 | return VM_FAULT_OOM; | |
2102 | if (err < 0 && err != -EBUSY) | |
2103 | return VM_FAULT_SIGBUS; | |
2104 | ||
2105 | return VM_FAULT_NOPAGE; | |
e0dc0d8f | 2106 | } |
79f3aa5b | 2107 | |
574c5b3d TH |
2108 | /** |
2109 | * vmf_insert_mixed_prot - insert single pfn into user vma with specified pgprot | |
2110 | * @vma: user vma to map to | |
2111 | * @addr: target user address of this page | |
2112 | * @pfn: source kernel pfn | |
2113 | * @pgprot: pgprot flags for the inserted page | |
2114 | * | |
a1a0aea5 | 2115 | * This is exactly like vmf_insert_mixed(), except that it allows drivers |
574c5b3d TH |
2116 | * to override pgprot on a per-page basis. |
2117 | * | |
2118 | * Typically this function should be used by drivers to set caching- and | |
2119 | * encryption bits different than those of @vma->vm_page_prot, because | |
2120 | * the caching- or encryption mode may not be known at mmap() time. | |
2121 | * This is ok as long as @vma->vm_page_prot is not used by the core vm | |
2122 | * to set caching and encryption bits for those vmas (except for COW pages). | |
2123 | * This is ensured by core vm only modifying these page table entries using | |
2124 | * functions that don't touch caching- or encryption bits, using pte_modify() | |
2125 | * if needed. (See for example mprotect()). | |
2126 | * Also when new page-table entries are created, this is only done using the | |
2127 | * fault() callback, and never using the value of vma->vm_page_prot, | |
2128 | * except for page-table entries that point to anonymous pages as the result | |
2129 | * of COW. | |
2130 | * | |
2131 | * Context: Process context. May allocate using %GFP_KERNEL. | |
2132 | * Return: vm_fault_t value. | |
2133 | */ | |
2134 | vm_fault_t vmf_insert_mixed_prot(struct vm_area_struct *vma, unsigned long addr, | |
2135 | pfn_t pfn, pgprot_t pgprot) | |
2136 | { | |
2137 | return __vm_insert_mixed(vma, addr, pfn, pgprot, false); | |
2138 | } | |
5379e4dd | 2139 | EXPORT_SYMBOL(vmf_insert_mixed_prot); |
574c5b3d | 2140 | |
79f3aa5b MW |
2141 | vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
2142 | pfn_t pfn) | |
2143 | { | |
574c5b3d | 2144 | return __vm_insert_mixed(vma, addr, pfn, vma->vm_page_prot, false); |
79f3aa5b | 2145 | } |
5d747637 | 2146 | EXPORT_SYMBOL(vmf_insert_mixed); |
e0dc0d8f | 2147 | |
ab77dab4 SJ |
2148 | /* |
2149 | * If the insertion of PTE failed because someone else already added a | |
2150 | * different entry in the mean time, we treat that as success as we assume | |
2151 | * the same entry was actually inserted. | |
2152 | */ | |
ab77dab4 SJ |
2153 | vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, |
2154 | unsigned long addr, pfn_t pfn) | |
b2770da6 | 2155 | { |
574c5b3d | 2156 | return __vm_insert_mixed(vma, addr, pfn, vma->vm_page_prot, true); |
b2770da6 | 2157 | } |
ab77dab4 | 2158 | EXPORT_SYMBOL(vmf_insert_mixed_mkwrite); |
b2770da6 | 2159 | |
1da177e4 LT |
2160 | /* |
2161 | * maps a range of physical memory into the requested pages. the old | |
2162 | * mappings are removed. any references to nonexistent pages results | |
2163 | * in null mappings (currently treated as "copy-on-access") | |
2164 | */ | |
2165 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
2166 | unsigned long addr, unsigned long end, | |
2167 | unsigned long pfn, pgprot_t prot) | |
2168 | { | |
fbc6e026 | 2169 | pte_t *pte, *mapped_pte; |
c74df32c | 2170 | spinlock_t *ptl; |
42e4089c | 2171 | int err = 0; |
1da177e4 | 2172 | |
fbc6e026 | 2173 | mapped_pte = pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
2174 | if (!pte) |
2175 | return -ENOMEM; | |
6606c3e0 | 2176 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
2177 | do { |
2178 | BUG_ON(!pte_none(*pte)); | |
42e4089c AK |
2179 | if (!pfn_modify_allowed(pfn, prot)) { |
2180 | err = -EACCES; | |
2181 | break; | |
2182 | } | |
7e675137 | 2183 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
2184 | pfn++; |
2185 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 2186 | arch_leave_lazy_mmu_mode(); |
fbc6e026 | 2187 | pte_unmap_unlock(mapped_pte, ptl); |
42e4089c | 2188 | return err; |
1da177e4 LT |
2189 | } |
2190 | ||
2191 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2192 | unsigned long addr, unsigned long end, | |
2193 | unsigned long pfn, pgprot_t prot) | |
2194 | { | |
2195 | pmd_t *pmd; | |
2196 | unsigned long next; | |
42e4089c | 2197 | int err; |
1da177e4 LT |
2198 | |
2199 | pfn -= addr >> PAGE_SHIFT; | |
2200 | pmd = pmd_alloc(mm, pud, addr); | |
2201 | if (!pmd) | |
2202 | return -ENOMEM; | |
f66055ab | 2203 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
2204 | do { |
2205 | next = pmd_addr_end(addr, end); | |
42e4089c AK |
2206 | err = remap_pte_range(mm, pmd, addr, next, |
2207 | pfn + (addr >> PAGE_SHIFT), prot); | |
2208 | if (err) | |
2209 | return err; | |
1da177e4 LT |
2210 | } while (pmd++, addr = next, addr != end); |
2211 | return 0; | |
2212 | } | |
2213 | ||
c2febafc | 2214 | static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, |
1da177e4 LT |
2215 | unsigned long addr, unsigned long end, |
2216 | unsigned long pfn, pgprot_t prot) | |
2217 | { | |
2218 | pud_t *pud; | |
2219 | unsigned long next; | |
42e4089c | 2220 | int err; |
1da177e4 LT |
2221 | |
2222 | pfn -= addr >> PAGE_SHIFT; | |
c2febafc | 2223 | pud = pud_alloc(mm, p4d, addr); |
1da177e4 LT |
2224 | if (!pud) |
2225 | return -ENOMEM; | |
2226 | do { | |
2227 | next = pud_addr_end(addr, end); | |
42e4089c AK |
2228 | err = remap_pmd_range(mm, pud, addr, next, |
2229 | pfn + (addr >> PAGE_SHIFT), prot); | |
2230 | if (err) | |
2231 | return err; | |
1da177e4 LT |
2232 | } while (pud++, addr = next, addr != end); |
2233 | return 0; | |
2234 | } | |
2235 | ||
c2febafc KS |
2236 | static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2237 | unsigned long addr, unsigned long end, | |
2238 | unsigned long pfn, pgprot_t prot) | |
2239 | { | |
2240 | p4d_t *p4d; | |
2241 | unsigned long next; | |
42e4089c | 2242 | int err; |
c2febafc KS |
2243 | |
2244 | pfn -= addr >> PAGE_SHIFT; | |
2245 | p4d = p4d_alloc(mm, pgd, addr); | |
2246 | if (!p4d) | |
2247 | return -ENOMEM; | |
2248 | do { | |
2249 | next = p4d_addr_end(addr, end); | |
42e4089c AK |
2250 | err = remap_pud_range(mm, p4d, addr, next, |
2251 | pfn + (addr >> PAGE_SHIFT), prot); | |
2252 | if (err) | |
2253 | return err; | |
c2febafc KS |
2254 | } while (p4d++, addr = next, addr != end); |
2255 | return 0; | |
2256 | } | |
2257 | ||
bfa5bf6d REB |
2258 | /** |
2259 | * remap_pfn_range - remap kernel memory to userspace | |
2260 | * @vma: user vma to map to | |
0c4123e3 | 2261 | * @addr: target page aligned user address to start at |
86a76331 | 2262 | * @pfn: page frame number of kernel physical memory address |
552657b7 | 2263 | * @size: size of mapping area |
bfa5bf6d REB |
2264 | * @prot: page protection flags for this mapping |
2265 | * | |
a862f68a MR |
2266 | * Note: this is only safe if the mm semaphore is held when called. |
2267 | * | |
2268 | * Return: %0 on success, negative error code otherwise. | |
bfa5bf6d | 2269 | */ |
1da177e4 LT |
2270 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
2271 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
2272 | { | |
2273 | pgd_t *pgd; | |
2274 | unsigned long next; | |
2d15cab8 | 2275 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 | 2276 | struct mm_struct *mm = vma->vm_mm; |
d5957d2f | 2277 | unsigned long remap_pfn = pfn; |
1da177e4 LT |
2278 | int err; |
2279 | ||
0c4123e3 AZ |
2280 | if (WARN_ON_ONCE(!PAGE_ALIGNED(addr))) |
2281 | return -EINVAL; | |
2282 | ||
1da177e4 LT |
2283 | /* |
2284 | * Physically remapped pages are special. Tell the | |
2285 | * rest of the world about it: | |
2286 | * VM_IO tells people not to look at these pages | |
2287 | * (accesses can have side effects). | |
6aab341e LT |
2288 | * VM_PFNMAP tells the core MM that the base pages are just |
2289 | * raw PFN mappings, and do not have a "struct page" associated | |
2290 | * with them. | |
314e51b9 KK |
2291 | * VM_DONTEXPAND |
2292 | * Disable vma merging and expanding with mremap(). | |
2293 | * VM_DONTDUMP | |
2294 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
2295 | * |
2296 | * There's a horrible special case to handle copy-on-write | |
2297 | * behaviour that some programs depend on. We mark the "original" | |
2298 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 2299 | * See vm_normal_page() for details. |
1da177e4 | 2300 | */ |
b3b9c293 KK |
2301 | if (is_cow_mapping(vma->vm_flags)) { |
2302 | if (addr != vma->vm_start || end != vma->vm_end) | |
2303 | return -EINVAL; | |
fb155c16 | 2304 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
2305 | } |
2306 | ||
d5957d2f | 2307 | err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size)); |
b3b9c293 | 2308 | if (err) |
3c8bb73a | 2309 | return -EINVAL; |
fb155c16 | 2310 | |
314e51b9 | 2311 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
2312 | |
2313 | BUG_ON(addr >= end); | |
2314 | pfn -= addr >> PAGE_SHIFT; | |
2315 | pgd = pgd_offset(mm, addr); | |
2316 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
2317 | do { |
2318 | next = pgd_addr_end(addr, end); | |
c2febafc | 2319 | err = remap_p4d_range(mm, pgd, addr, next, |
1da177e4 LT |
2320 | pfn + (addr >> PAGE_SHIFT), prot); |
2321 | if (err) | |
2322 | break; | |
2323 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 2324 | |
2325 | if (err) | |
d5957d2f | 2326 | untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size)); |
2ab64037 | 2327 | |
1da177e4 LT |
2328 | return err; |
2329 | } | |
2330 | EXPORT_SYMBOL(remap_pfn_range); | |
2331 | ||
b4cbb197 LT |
2332 | /** |
2333 | * vm_iomap_memory - remap memory to userspace | |
2334 | * @vma: user vma to map to | |
abd69b9e | 2335 | * @start: start of the physical memory to be mapped |
b4cbb197 LT |
2336 | * @len: size of area |
2337 | * | |
2338 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
2339 | * driver just needs to give us the physical memory range to be mapped, | |
2340 | * we'll figure out the rest from the vma information. | |
2341 | * | |
2342 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
2343 | * whatever write-combining details or similar. | |
a862f68a MR |
2344 | * |
2345 | * Return: %0 on success, negative error code otherwise. | |
b4cbb197 LT |
2346 | */ |
2347 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
2348 | { | |
2349 | unsigned long vm_len, pfn, pages; | |
2350 | ||
2351 | /* Check that the physical memory area passed in looks valid */ | |
2352 | if (start + len < start) | |
2353 | return -EINVAL; | |
2354 | /* | |
2355 | * You *really* shouldn't map things that aren't page-aligned, | |
2356 | * but we've historically allowed it because IO memory might | |
2357 | * just have smaller alignment. | |
2358 | */ | |
2359 | len += start & ~PAGE_MASK; | |
2360 | pfn = start >> PAGE_SHIFT; | |
2361 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
2362 | if (pfn + pages < pfn) | |
2363 | return -EINVAL; | |
2364 | ||
2365 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
2366 | if (vma->vm_pgoff > pages) | |
2367 | return -EINVAL; | |
2368 | pfn += vma->vm_pgoff; | |
2369 | pages -= vma->vm_pgoff; | |
2370 | ||
2371 | /* Can we fit all of the mapping? */ | |
2372 | vm_len = vma->vm_end - vma->vm_start; | |
2373 | if (vm_len >> PAGE_SHIFT > pages) | |
2374 | return -EINVAL; | |
2375 | ||
2376 | /* Ok, let it rip */ | |
2377 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
2378 | } | |
2379 | EXPORT_SYMBOL(vm_iomap_memory); | |
2380 | ||
aee16b3c JF |
2381 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
2382 | unsigned long addr, unsigned long end, | |
e80d3909 JR |
2383 | pte_fn_t fn, void *data, bool create, |
2384 | pgtbl_mod_mask *mask) | |
aee16b3c JF |
2385 | { |
2386 | pte_t *pte; | |
be1db475 | 2387 | int err = 0; |
3f649ab7 | 2388 | spinlock_t *ptl; |
aee16b3c | 2389 | |
be1db475 DA |
2390 | if (create) { |
2391 | pte = (mm == &init_mm) ? | |
e80d3909 | 2392 | pte_alloc_kernel_track(pmd, addr, mask) : |
be1db475 DA |
2393 | pte_alloc_map_lock(mm, pmd, addr, &ptl); |
2394 | if (!pte) | |
2395 | return -ENOMEM; | |
2396 | } else { | |
2397 | pte = (mm == &init_mm) ? | |
2398 | pte_offset_kernel(pmd, addr) : | |
2399 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
2400 | } | |
aee16b3c JF |
2401 | |
2402 | BUG_ON(pmd_huge(*pmd)); | |
2403 | ||
38e0edb1 JF |
2404 | arch_enter_lazy_mmu_mode(); |
2405 | ||
eeb4a05f CH |
2406 | if (fn) { |
2407 | do { | |
2408 | if (create || !pte_none(*pte)) { | |
2409 | err = fn(pte++, addr, data); | |
2410 | if (err) | |
2411 | break; | |
2412 | } | |
2413 | } while (addr += PAGE_SIZE, addr != end); | |
2414 | } | |
e80d3909 | 2415 | *mask |= PGTBL_PTE_MODIFIED; |
aee16b3c | 2416 | |
38e0edb1 JF |
2417 | arch_leave_lazy_mmu_mode(); |
2418 | ||
aee16b3c JF |
2419 | if (mm != &init_mm) |
2420 | pte_unmap_unlock(pte-1, ptl); | |
2421 | return err; | |
2422 | } | |
2423 | ||
2424 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2425 | unsigned long addr, unsigned long end, | |
e80d3909 JR |
2426 | pte_fn_t fn, void *data, bool create, |
2427 | pgtbl_mod_mask *mask) | |
aee16b3c JF |
2428 | { |
2429 | pmd_t *pmd; | |
2430 | unsigned long next; | |
be1db475 | 2431 | int err = 0; |
aee16b3c | 2432 | |
ceb86879 AK |
2433 | BUG_ON(pud_huge(*pud)); |
2434 | ||
be1db475 | 2435 | if (create) { |
e80d3909 | 2436 | pmd = pmd_alloc_track(mm, pud, addr, mask); |
be1db475 DA |
2437 | if (!pmd) |
2438 | return -ENOMEM; | |
2439 | } else { | |
2440 | pmd = pmd_offset(pud, addr); | |
2441 | } | |
aee16b3c JF |
2442 | do { |
2443 | next = pmd_addr_end(addr, end); | |
be1db475 DA |
2444 | if (create || !pmd_none_or_clear_bad(pmd)) { |
2445 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data, | |
e80d3909 | 2446 | create, mask); |
be1db475 DA |
2447 | if (err) |
2448 | break; | |
2449 | } | |
aee16b3c JF |
2450 | } while (pmd++, addr = next, addr != end); |
2451 | return err; | |
2452 | } | |
2453 | ||
c2febafc | 2454 | static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, |
aee16b3c | 2455 | unsigned long addr, unsigned long end, |
e80d3909 JR |
2456 | pte_fn_t fn, void *data, bool create, |
2457 | pgtbl_mod_mask *mask) | |
aee16b3c JF |
2458 | { |
2459 | pud_t *pud; | |
2460 | unsigned long next; | |
be1db475 | 2461 | int err = 0; |
aee16b3c | 2462 | |
be1db475 | 2463 | if (create) { |
e80d3909 | 2464 | pud = pud_alloc_track(mm, p4d, addr, mask); |
be1db475 DA |
2465 | if (!pud) |
2466 | return -ENOMEM; | |
2467 | } else { | |
2468 | pud = pud_offset(p4d, addr); | |
2469 | } | |
aee16b3c JF |
2470 | do { |
2471 | next = pud_addr_end(addr, end); | |
be1db475 DA |
2472 | if (create || !pud_none_or_clear_bad(pud)) { |
2473 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data, | |
e80d3909 | 2474 | create, mask); |
be1db475 DA |
2475 | if (err) |
2476 | break; | |
2477 | } | |
aee16b3c JF |
2478 | } while (pud++, addr = next, addr != end); |
2479 | return err; | |
2480 | } | |
2481 | ||
c2febafc KS |
2482 | static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2483 | unsigned long addr, unsigned long end, | |
e80d3909 JR |
2484 | pte_fn_t fn, void *data, bool create, |
2485 | pgtbl_mod_mask *mask) | |
c2febafc KS |
2486 | { |
2487 | p4d_t *p4d; | |
2488 | unsigned long next; | |
be1db475 | 2489 | int err = 0; |
c2febafc | 2490 | |
be1db475 | 2491 | if (create) { |
e80d3909 | 2492 | p4d = p4d_alloc_track(mm, pgd, addr, mask); |
be1db475 DA |
2493 | if (!p4d) |
2494 | return -ENOMEM; | |
2495 | } else { | |
2496 | p4d = p4d_offset(pgd, addr); | |
2497 | } | |
c2febafc KS |
2498 | do { |
2499 | next = p4d_addr_end(addr, end); | |
be1db475 DA |
2500 | if (create || !p4d_none_or_clear_bad(p4d)) { |
2501 | err = apply_to_pud_range(mm, p4d, addr, next, fn, data, | |
e80d3909 | 2502 | create, mask); |
be1db475 DA |
2503 | if (err) |
2504 | break; | |
2505 | } | |
c2febafc KS |
2506 | } while (p4d++, addr = next, addr != end); |
2507 | return err; | |
2508 | } | |
2509 | ||
be1db475 DA |
2510 | static int __apply_to_page_range(struct mm_struct *mm, unsigned long addr, |
2511 | unsigned long size, pte_fn_t fn, | |
2512 | void *data, bool create) | |
aee16b3c JF |
2513 | { |
2514 | pgd_t *pgd; | |
e80d3909 | 2515 | unsigned long start = addr, next; |
57250a5b | 2516 | unsigned long end = addr + size; |
e80d3909 | 2517 | pgtbl_mod_mask mask = 0; |
be1db475 | 2518 | int err = 0; |
aee16b3c | 2519 | |
9cb65bc3 MP |
2520 | if (WARN_ON(addr >= end)) |
2521 | return -EINVAL; | |
2522 | ||
aee16b3c JF |
2523 | pgd = pgd_offset(mm, addr); |
2524 | do { | |
2525 | next = pgd_addr_end(addr, end); | |
be1db475 DA |
2526 | if (!create && pgd_none_or_clear_bad(pgd)) |
2527 | continue; | |
e80d3909 | 2528 | err = apply_to_p4d_range(mm, pgd, addr, next, fn, data, create, &mask); |
aee16b3c JF |
2529 | if (err) |
2530 | break; | |
2531 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2532 | |
e80d3909 JR |
2533 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
2534 | arch_sync_kernel_mappings(start, start + size); | |
2535 | ||
aee16b3c JF |
2536 | return err; |
2537 | } | |
be1db475 DA |
2538 | |
2539 | /* | |
2540 | * Scan a region of virtual memory, filling in page tables as necessary | |
2541 | * and calling a provided function on each leaf page table. | |
2542 | */ | |
2543 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2544 | unsigned long size, pte_fn_t fn, void *data) | |
2545 | { | |
2546 | return __apply_to_page_range(mm, addr, size, fn, data, true); | |
2547 | } | |
aee16b3c JF |
2548 | EXPORT_SYMBOL_GPL(apply_to_page_range); |
2549 | ||
be1db475 DA |
2550 | /* |
2551 | * Scan a region of virtual memory, calling a provided function on | |
2552 | * each leaf page table where it exists. | |
2553 | * | |
2554 | * Unlike apply_to_page_range, this does _not_ fill in page tables | |
2555 | * where they are absent. | |
2556 | */ | |
2557 | int apply_to_existing_page_range(struct mm_struct *mm, unsigned long addr, | |
2558 | unsigned long size, pte_fn_t fn, void *data) | |
2559 | { | |
2560 | return __apply_to_page_range(mm, addr, size, fn, data, false); | |
2561 | } | |
2562 | EXPORT_SYMBOL_GPL(apply_to_existing_page_range); | |
2563 | ||
8f4e2101 | 2564 | /* |
9b4bdd2f KS |
2565 | * handle_pte_fault chooses page fault handler according to an entry which was |
2566 | * read non-atomically. Before making any commitment, on those architectures | |
2567 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
2568 | * parts, do_swap_page must check under lock before unmapping the pte and | |
2569 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2570 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2571 | */ |
4c21e2f2 | 2572 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2573 | pte_t *page_table, pte_t orig_pte) |
2574 | { | |
2575 | int same = 1; | |
923717cb | 2576 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPTION) |
8f4e2101 | 2577 | if (sizeof(pte_t) > sizeof(unsigned long)) { |
4c21e2f2 HD |
2578 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2579 | spin_lock(ptl); | |
8f4e2101 | 2580 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2581 | spin_unlock(ptl); |
8f4e2101 HD |
2582 | } |
2583 | #endif | |
2584 | pte_unmap(page_table); | |
2585 | return same; | |
2586 | } | |
2587 | ||
83d116c5 JH |
2588 | static inline bool cow_user_page(struct page *dst, struct page *src, |
2589 | struct vm_fault *vmf) | |
6aab341e | 2590 | { |
83d116c5 JH |
2591 | bool ret; |
2592 | void *kaddr; | |
2593 | void __user *uaddr; | |
c3e5ea6e | 2594 | bool locked = false; |
83d116c5 JH |
2595 | struct vm_area_struct *vma = vmf->vma; |
2596 | struct mm_struct *mm = vma->vm_mm; | |
2597 | unsigned long addr = vmf->address; | |
2598 | ||
83d116c5 JH |
2599 | if (likely(src)) { |
2600 | copy_user_highpage(dst, src, addr, vma); | |
2601 | return true; | |
2602 | } | |
2603 | ||
6aab341e LT |
2604 | /* |
2605 | * If the source page was a PFN mapping, we don't have | |
2606 | * a "struct page" for it. We do a best-effort copy by | |
2607 | * just copying from the original user address. If that | |
2608 | * fails, we just zero-fill it. Live with it. | |
2609 | */ | |
83d116c5 JH |
2610 | kaddr = kmap_atomic(dst); |
2611 | uaddr = (void __user *)(addr & PAGE_MASK); | |
2612 | ||
2613 | /* | |
2614 | * On architectures with software "accessed" bits, we would | |
2615 | * take a double page fault, so mark it accessed here. | |
2616 | */ | |
c3e5ea6e | 2617 | if (arch_faults_on_old_pte() && !pte_young(vmf->orig_pte)) { |
83d116c5 | 2618 | pte_t entry; |
5d2a2dbb | 2619 | |
83d116c5 | 2620 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); |
c3e5ea6e | 2621 | locked = true; |
83d116c5 JH |
2622 | if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2623 | /* | |
2624 | * Other thread has already handled the fault | |
7df67697 | 2625 | * and update local tlb only |
83d116c5 | 2626 | */ |
7df67697 | 2627 | update_mmu_tlb(vma, addr, vmf->pte); |
83d116c5 JH |
2628 | ret = false; |
2629 | goto pte_unlock; | |
2630 | } | |
2631 | ||
2632 | entry = pte_mkyoung(vmf->orig_pte); | |
2633 | if (ptep_set_access_flags(vma, addr, vmf->pte, entry, 0)) | |
2634 | update_mmu_cache(vma, addr, vmf->pte); | |
2635 | } | |
2636 | ||
2637 | /* | |
2638 | * This really shouldn't fail, because the page is there | |
2639 | * in the page tables. But it might just be unreadable, | |
2640 | * in which case we just give up and fill the result with | |
2641 | * zeroes. | |
2642 | */ | |
2643 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { | |
c3e5ea6e KS |
2644 | if (locked) |
2645 | goto warn; | |
2646 | ||
2647 | /* Re-validate under PTL if the page is still mapped */ | |
2648 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); | |
2649 | locked = true; | |
2650 | if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) { | |
7df67697 BM |
2651 | /* The PTE changed under us, update local tlb */ |
2652 | update_mmu_tlb(vma, addr, vmf->pte); | |
c3e5ea6e KS |
2653 | ret = false; |
2654 | goto pte_unlock; | |
2655 | } | |
2656 | ||
5d2a2dbb | 2657 | /* |
985ba004 | 2658 | * The same page can be mapped back since last copy attempt. |
c3e5ea6e | 2659 | * Try to copy again under PTL. |
5d2a2dbb | 2660 | */ |
c3e5ea6e KS |
2661 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { |
2662 | /* | |
2663 | * Give a warn in case there can be some obscure | |
2664 | * use-case | |
2665 | */ | |
2666 | warn: | |
2667 | WARN_ON_ONCE(1); | |
2668 | clear_page(kaddr); | |
2669 | } | |
83d116c5 JH |
2670 | } |
2671 | ||
2672 | ret = true; | |
2673 | ||
2674 | pte_unlock: | |
c3e5ea6e | 2675 | if (locked) |
83d116c5 JH |
2676 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
2677 | kunmap_atomic(kaddr); | |
2678 | flush_dcache_page(dst); | |
2679 | ||
2680 | return ret; | |
6aab341e LT |
2681 | } |
2682 | ||
c20cd45e MH |
2683 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
2684 | { | |
2685 | struct file *vm_file = vma->vm_file; | |
2686 | ||
2687 | if (vm_file) | |
2688 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
2689 | ||
2690 | /* | |
2691 | * Special mappings (e.g. VDSO) do not have any file so fake | |
2692 | * a default GFP_KERNEL for them. | |
2693 | */ | |
2694 | return GFP_KERNEL; | |
2695 | } | |
2696 | ||
fb09a464 KS |
2697 | /* |
2698 | * Notify the address space that the page is about to become writable so that | |
2699 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2700 | * | |
2701 | * We do this without the lock held, so that it can sleep if it needs to. | |
2702 | */ | |
2b740303 | 2703 | static vm_fault_t do_page_mkwrite(struct vm_fault *vmf) |
fb09a464 | 2704 | { |
2b740303 | 2705 | vm_fault_t ret; |
38b8cb7f JK |
2706 | struct page *page = vmf->page; |
2707 | unsigned int old_flags = vmf->flags; | |
fb09a464 | 2708 | |
38b8cb7f | 2709 | vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
fb09a464 | 2710 | |
dc617f29 DW |
2711 | if (vmf->vma->vm_file && |
2712 | IS_SWAPFILE(vmf->vma->vm_file->f_mapping->host)) | |
2713 | return VM_FAULT_SIGBUS; | |
2714 | ||
11bac800 | 2715 | ret = vmf->vma->vm_ops->page_mkwrite(vmf); |
38b8cb7f JK |
2716 | /* Restore original flags so that caller is not surprised */ |
2717 | vmf->flags = old_flags; | |
fb09a464 KS |
2718 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2719 | return ret; | |
2720 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2721 | lock_page(page); | |
2722 | if (!page->mapping) { | |
2723 | unlock_page(page); | |
2724 | return 0; /* retry */ | |
2725 | } | |
2726 | ret |= VM_FAULT_LOCKED; | |
2727 | } else | |
2728 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2729 | return ret; | |
2730 | } | |
2731 | ||
97ba0c2b JK |
2732 | /* |
2733 | * Handle dirtying of a page in shared file mapping on a write fault. | |
2734 | * | |
2735 | * The function expects the page to be locked and unlocks it. | |
2736 | */ | |
89b15332 | 2737 | static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf) |
97ba0c2b | 2738 | { |
89b15332 | 2739 | struct vm_area_struct *vma = vmf->vma; |
97ba0c2b | 2740 | struct address_space *mapping; |
89b15332 | 2741 | struct page *page = vmf->page; |
97ba0c2b JK |
2742 | bool dirtied; |
2743 | bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; | |
2744 | ||
2745 | dirtied = set_page_dirty(page); | |
2746 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2747 | /* | |
2748 | * Take a local copy of the address_space - page.mapping may be zeroed | |
2749 | * by truncate after unlock_page(). The address_space itself remains | |
2750 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
2751 | * release semantics to prevent the compiler from undoing this copying. | |
2752 | */ | |
2753 | mapping = page_rmapping(page); | |
2754 | unlock_page(page); | |
2755 | ||
89b15332 JW |
2756 | if (!page_mkwrite) |
2757 | file_update_time(vma->vm_file); | |
2758 | ||
2759 | /* | |
2760 | * Throttle page dirtying rate down to writeback speed. | |
2761 | * | |
2762 | * mapping may be NULL here because some device drivers do not | |
2763 | * set page.mapping but still dirty their pages | |
2764 | * | |
c1e8d7c6 | 2765 | * Drop the mmap_lock before waiting on IO, if we can. The file |
89b15332 JW |
2766 | * is pinning the mapping, as per above. |
2767 | */ | |
97ba0c2b | 2768 | if ((dirtied || page_mkwrite) && mapping) { |
89b15332 JW |
2769 | struct file *fpin; |
2770 | ||
2771 | fpin = maybe_unlock_mmap_for_io(vmf, NULL); | |
97ba0c2b | 2772 | balance_dirty_pages_ratelimited(mapping); |
89b15332 JW |
2773 | if (fpin) { |
2774 | fput(fpin); | |
2775 | return VM_FAULT_RETRY; | |
2776 | } | |
97ba0c2b JK |
2777 | } |
2778 | ||
89b15332 | 2779 | return 0; |
97ba0c2b JK |
2780 | } |
2781 | ||
4e047f89 SR |
2782 | /* |
2783 | * Handle write page faults for pages that can be reused in the current vma | |
2784 | * | |
2785 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
2786 | * or due to us being the last reference standing to the page. In either | |
2787 | * case, all we need to do here is to mark the page as writable and update | |
2788 | * any related book-keeping. | |
2789 | */ | |
997dd98d | 2790 | static inline void wp_page_reuse(struct vm_fault *vmf) |
82b0f8c3 | 2791 | __releases(vmf->ptl) |
4e047f89 | 2792 | { |
82b0f8c3 | 2793 | struct vm_area_struct *vma = vmf->vma; |
a41b70d6 | 2794 | struct page *page = vmf->page; |
4e047f89 SR |
2795 | pte_t entry; |
2796 | /* | |
2797 | * Clear the pages cpupid information as the existing | |
2798 | * information potentially belongs to a now completely | |
2799 | * unrelated process. | |
2800 | */ | |
2801 | if (page) | |
2802 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2803 | ||
2994302b JK |
2804 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2805 | entry = pte_mkyoung(vmf->orig_pte); | |
4e047f89 | 2806 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
82b0f8c3 JK |
2807 | if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) |
2808 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2809 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
798a6b87 | 2810 | count_vm_event(PGREUSE); |
4e047f89 SR |
2811 | } |
2812 | ||
2f38ab2c SR |
2813 | /* |
2814 | * Handle the case of a page which we actually need to copy to a new page. | |
2815 | * | |
c1e8d7c6 | 2816 | * Called with mmap_lock locked and the old page referenced, but |
2f38ab2c SR |
2817 | * without the ptl held. |
2818 | * | |
2819 | * High level logic flow: | |
2820 | * | |
2821 | * - Allocate a page, copy the content of the old page to the new one. | |
2822 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2823 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2824 | * - If the pte is still the way we remember it, update the page table and all | |
2825 | * relevant references. This includes dropping the reference the page-table | |
2826 | * held to the old page, as well as updating the rmap. | |
2827 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2828 | */ | |
2b740303 | 2829 | static vm_fault_t wp_page_copy(struct vm_fault *vmf) |
2f38ab2c | 2830 | { |
82b0f8c3 | 2831 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2832 | struct mm_struct *mm = vma->vm_mm; |
a41b70d6 | 2833 | struct page *old_page = vmf->page; |
2f38ab2c | 2834 | struct page *new_page = NULL; |
2f38ab2c SR |
2835 | pte_t entry; |
2836 | int page_copied = 0; | |
ac46d4f3 | 2837 | struct mmu_notifier_range range; |
2f38ab2c SR |
2838 | |
2839 | if (unlikely(anon_vma_prepare(vma))) | |
2840 | goto oom; | |
2841 | ||
2994302b | 2842 | if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { |
82b0f8c3 JK |
2843 | new_page = alloc_zeroed_user_highpage_movable(vma, |
2844 | vmf->address); | |
2f38ab2c SR |
2845 | if (!new_page) |
2846 | goto oom; | |
2847 | } else { | |
bae473a4 | 2848 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
82b0f8c3 | 2849 | vmf->address); |
2f38ab2c SR |
2850 | if (!new_page) |
2851 | goto oom; | |
83d116c5 JH |
2852 | |
2853 | if (!cow_user_page(new_page, old_page, vmf)) { | |
2854 | /* | |
2855 | * COW failed, if the fault was solved by other, | |
2856 | * it's fine. If not, userspace would re-fault on | |
2857 | * the same address and we will handle the fault | |
2858 | * from the second attempt. | |
2859 | */ | |
2860 | put_page(new_page); | |
2861 | if (old_page) | |
2862 | put_page(old_page); | |
2863 | return 0; | |
2864 | } | |
2f38ab2c | 2865 | } |
2f38ab2c | 2866 | |
d9eb1ea2 | 2867 | if (mem_cgroup_charge(new_page, mm, GFP_KERNEL)) |
2f38ab2c | 2868 | goto oom_free_new; |
9d82c694 | 2869 | cgroup_throttle_swaprate(new_page, GFP_KERNEL); |
2f38ab2c | 2870 | |
eb3c24f3 MG |
2871 | __SetPageUptodate(new_page); |
2872 | ||
7269f999 | 2873 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, |
6f4f13e8 | 2874 | vmf->address & PAGE_MASK, |
ac46d4f3 JG |
2875 | (vmf->address & PAGE_MASK) + PAGE_SIZE); |
2876 | mmu_notifier_invalidate_range_start(&range); | |
2f38ab2c SR |
2877 | |
2878 | /* | |
2879 | * Re-check the pte - we dropped the lock | |
2880 | */ | |
82b0f8c3 | 2881 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); |
2994302b | 2882 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2f38ab2c SR |
2883 | if (old_page) { |
2884 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
2885 | dec_mm_counter_fast(mm, |
2886 | mm_counter_file(old_page)); | |
2f38ab2c SR |
2887 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
2888 | } | |
2889 | } else { | |
2890 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
2891 | } | |
2994302b | 2892 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2f38ab2c | 2893 | entry = mk_pte(new_page, vma->vm_page_prot); |
44bf431b | 2894 | entry = pte_sw_mkyoung(entry); |
2f38ab2c | 2895 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
111fe718 | 2896 | |
2f38ab2c SR |
2897 | /* |
2898 | * Clear the pte entry and flush it first, before updating the | |
111fe718 NP |
2899 | * pte with the new entry, to keep TLBs on different CPUs in |
2900 | * sync. This code used to set the new PTE then flush TLBs, but | |
2901 | * that left a window where the new PTE could be loaded into | |
2902 | * some TLBs while the old PTE remains in others. | |
2f38ab2c | 2903 | */ |
82b0f8c3 JK |
2904 | ptep_clear_flush_notify(vma, vmf->address, vmf->pte); |
2905 | page_add_new_anon_rmap(new_page, vma, vmf->address, false); | |
b518154e | 2906 | lru_cache_add_inactive_or_unevictable(new_page, vma); |
2f38ab2c SR |
2907 | /* |
2908 | * We call the notify macro here because, when using secondary | |
2909 | * mmu page tables (such as kvm shadow page tables), we want the | |
2910 | * new page to be mapped directly into the secondary page table. | |
2911 | */ | |
82b0f8c3 JK |
2912 | set_pte_at_notify(mm, vmf->address, vmf->pte, entry); |
2913 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2f38ab2c SR |
2914 | if (old_page) { |
2915 | /* | |
2916 | * Only after switching the pte to the new page may | |
2917 | * we remove the mapcount here. Otherwise another | |
2918 | * process may come and find the rmap count decremented | |
2919 | * before the pte is switched to the new page, and | |
2920 | * "reuse" the old page writing into it while our pte | |
2921 | * here still points into it and can be read by other | |
2922 | * threads. | |
2923 | * | |
2924 | * The critical issue is to order this | |
2925 | * page_remove_rmap with the ptp_clear_flush above. | |
2926 | * Those stores are ordered by (if nothing else,) | |
2927 | * the barrier present in the atomic_add_negative | |
2928 | * in page_remove_rmap. | |
2929 | * | |
2930 | * Then the TLB flush in ptep_clear_flush ensures that | |
2931 | * no process can access the old page before the | |
2932 | * decremented mapcount is visible. And the old page | |
2933 | * cannot be reused until after the decremented | |
2934 | * mapcount is visible. So transitively, TLBs to | |
2935 | * old page will be flushed before it can be reused. | |
2936 | */ | |
d281ee61 | 2937 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
2938 | } |
2939 | ||
2940 | /* Free the old page.. */ | |
2941 | new_page = old_page; | |
2942 | page_copied = 1; | |
2943 | } else { | |
7df67697 | 2944 | update_mmu_tlb(vma, vmf->address, vmf->pte); |
2f38ab2c SR |
2945 | } |
2946 | ||
2947 | if (new_page) | |
09cbfeaf | 2948 | put_page(new_page); |
2f38ab2c | 2949 | |
82b0f8c3 | 2950 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4645b9fe JG |
2951 | /* |
2952 | * No need to double call mmu_notifier->invalidate_range() callback as | |
2953 | * the above ptep_clear_flush_notify() did already call it. | |
2954 | */ | |
ac46d4f3 | 2955 | mmu_notifier_invalidate_range_only_end(&range); |
2f38ab2c SR |
2956 | if (old_page) { |
2957 | /* | |
2958 | * Don't let another task, with possibly unlocked vma, | |
2959 | * keep the mlocked page. | |
2960 | */ | |
2961 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
2962 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
2963 | if (PageMlocked(old_page)) |
2964 | munlock_vma_page(old_page); | |
2f38ab2c SR |
2965 | unlock_page(old_page); |
2966 | } | |
09cbfeaf | 2967 | put_page(old_page); |
2f38ab2c SR |
2968 | } |
2969 | return page_copied ? VM_FAULT_WRITE : 0; | |
2970 | oom_free_new: | |
09cbfeaf | 2971 | put_page(new_page); |
2f38ab2c SR |
2972 | oom: |
2973 | if (old_page) | |
09cbfeaf | 2974 | put_page(old_page); |
2f38ab2c SR |
2975 | return VM_FAULT_OOM; |
2976 | } | |
2977 | ||
66a6197c JK |
2978 | /** |
2979 | * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE | |
2980 | * writeable once the page is prepared | |
2981 | * | |
2982 | * @vmf: structure describing the fault | |
2983 | * | |
2984 | * This function handles all that is needed to finish a write page fault in a | |
2985 | * shared mapping due to PTE being read-only once the mapped page is prepared. | |
a862f68a | 2986 | * It handles locking of PTE and modifying it. |
66a6197c JK |
2987 | * |
2988 | * The function expects the page to be locked or other protection against | |
2989 | * concurrent faults / writeback (such as DAX radix tree locks). | |
a862f68a MR |
2990 | * |
2991 | * Return: %VM_FAULT_WRITE on success, %0 when PTE got changed before | |
2992 | * we acquired PTE lock. | |
66a6197c | 2993 | */ |
2b740303 | 2994 | vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf) |
66a6197c JK |
2995 | { |
2996 | WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); | |
2997 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, | |
2998 | &vmf->ptl); | |
2999 | /* | |
3000 | * We might have raced with another page fault while we released the | |
3001 | * pte_offset_map_lock. | |
3002 | */ | |
3003 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { | |
7df67697 | 3004 | update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); |
66a6197c | 3005 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a19e2553 | 3006 | return VM_FAULT_NOPAGE; |
66a6197c JK |
3007 | } |
3008 | wp_page_reuse(vmf); | |
a19e2553 | 3009 | return 0; |
66a6197c JK |
3010 | } |
3011 | ||
dd906184 BH |
3012 | /* |
3013 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
3014 | * mapping | |
3015 | */ | |
2b740303 | 3016 | static vm_fault_t wp_pfn_shared(struct vm_fault *vmf) |
dd906184 | 3017 | { |
82b0f8c3 | 3018 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 3019 | |
dd906184 | 3020 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
2b740303 | 3021 | vm_fault_t ret; |
dd906184 | 3022 | |
82b0f8c3 | 3023 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
fe82221f | 3024 | vmf->flags |= FAULT_FLAG_MKWRITE; |
11bac800 | 3025 | ret = vma->vm_ops->pfn_mkwrite(vmf); |
2f89dc12 | 3026 | if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) |
dd906184 | 3027 | return ret; |
66a6197c | 3028 | return finish_mkwrite_fault(vmf); |
dd906184 | 3029 | } |
997dd98d JK |
3030 | wp_page_reuse(vmf); |
3031 | return VM_FAULT_WRITE; | |
dd906184 BH |
3032 | } |
3033 | ||
2b740303 | 3034 | static vm_fault_t wp_page_shared(struct vm_fault *vmf) |
82b0f8c3 | 3035 | __releases(vmf->ptl) |
93e478d4 | 3036 | { |
82b0f8c3 | 3037 | struct vm_area_struct *vma = vmf->vma; |
89b15332 | 3038 | vm_fault_t ret = VM_FAULT_WRITE; |
93e478d4 | 3039 | |
a41b70d6 | 3040 | get_page(vmf->page); |
93e478d4 | 3041 | |
93e478d4 | 3042 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2b740303 | 3043 | vm_fault_t tmp; |
93e478d4 | 3044 | |
82b0f8c3 | 3045 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
38b8cb7f | 3046 | tmp = do_page_mkwrite(vmf); |
93e478d4 SR |
3047 | if (unlikely(!tmp || (tmp & |
3048 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
a41b70d6 | 3049 | put_page(vmf->page); |
93e478d4 SR |
3050 | return tmp; |
3051 | } | |
66a6197c | 3052 | tmp = finish_mkwrite_fault(vmf); |
a19e2553 | 3053 | if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
a41b70d6 | 3054 | unlock_page(vmf->page); |
a41b70d6 | 3055 | put_page(vmf->page); |
66a6197c | 3056 | return tmp; |
93e478d4 | 3057 | } |
66a6197c JK |
3058 | } else { |
3059 | wp_page_reuse(vmf); | |
997dd98d | 3060 | lock_page(vmf->page); |
93e478d4 | 3061 | } |
89b15332 | 3062 | ret |= fault_dirty_shared_page(vmf); |
997dd98d | 3063 | put_page(vmf->page); |
93e478d4 | 3064 | |
89b15332 | 3065 | return ret; |
93e478d4 SR |
3066 | } |
3067 | ||
1da177e4 LT |
3068 | /* |
3069 | * This routine handles present pages, when users try to write | |
3070 | * to a shared page. It is done by copying the page to a new address | |
3071 | * and decrementing the shared-page counter for the old page. | |
3072 | * | |
1da177e4 LT |
3073 | * Note that this routine assumes that the protection checks have been |
3074 | * done by the caller (the low-level page fault routine in most cases). | |
3075 | * Thus we can safely just mark it writable once we've done any necessary | |
3076 | * COW. | |
3077 | * | |
3078 | * We also mark the page dirty at this point even though the page will | |
3079 | * change only once the write actually happens. This avoids a few races, | |
3080 | * and potentially makes it more efficient. | |
3081 | * | |
c1e8d7c6 | 3082 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
8f4e2101 | 3083 | * but allow concurrent faults), with pte both mapped and locked. |
c1e8d7c6 | 3084 | * We return with mmap_lock still held, but pte unmapped and unlocked. |
1da177e4 | 3085 | */ |
2b740303 | 3086 | static vm_fault_t do_wp_page(struct vm_fault *vmf) |
82b0f8c3 | 3087 | __releases(vmf->ptl) |
1da177e4 | 3088 | { |
82b0f8c3 | 3089 | struct vm_area_struct *vma = vmf->vma; |
1da177e4 | 3090 | |
292924b2 | 3091 | if (userfaultfd_pte_wp(vma, *vmf->pte)) { |
529b930b AA |
3092 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3093 | return handle_userfault(vmf, VM_UFFD_WP); | |
3094 | } | |
3095 | ||
bef4cea8 NA |
3096 | /* |
3097 | * Userfaultfd write-protect can defer flushes. Ensure the TLB | |
3098 | * is flushed in this case before copying. | |
3099 | */ | |
3100 | if (unlikely(userfaultfd_wp(vmf->vma) && | |
3101 | mm_tlb_flush_pending(vmf->vma->vm_mm))) | |
3102 | flush_tlb_page(vmf->vma, vmf->address); | |
3103 | ||
a41b70d6 JK |
3104 | vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); |
3105 | if (!vmf->page) { | |
251b97f5 | 3106 | /* |
64e45507 PF |
3107 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
3108 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
3109 | * |
3110 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 3111 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
3112 | */ |
3113 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
3114 | (VM_WRITE|VM_SHARED)) | |
2994302b | 3115 | return wp_pfn_shared(vmf); |
2f38ab2c | 3116 | |
82b0f8c3 | 3117 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 3118 | return wp_page_copy(vmf); |
251b97f5 | 3119 | } |
1da177e4 | 3120 | |
d08b3851 | 3121 | /* |
ee6a6457 PZ |
3122 | * Take out anonymous pages first, anonymous shared vmas are |
3123 | * not dirty accountable. | |
d08b3851 | 3124 | */ |
52d1e606 | 3125 | if (PageAnon(vmf->page)) { |
09854ba9 LT |
3126 | struct page *page = vmf->page; |
3127 | ||
3128 | /* PageKsm() doesn't necessarily raise the page refcount */ | |
3129 | if (PageKsm(page) || page_count(page) != 1) | |
3130 | goto copy; | |
3131 | if (!trylock_page(page)) | |
3132 | goto copy; | |
3133 | if (PageKsm(page) || page_mapcount(page) != 1 || page_count(page) != 1) { | |
3134 | unlock_page(page); | |
52d1e606 | 3135 | goto copy; |
b009c024 | 3136 | } |
09854ba9 LT |
3137 | /* |
3138 | * Ok, we've got the only map reference, and the only | |
3139 | * page count reference, and the page is locked, | |
3140 | * it's dark out, and we're wearing sunglasses. Hit it. | |
3141 | */ | |
09854ba9 | 3142 | unlock_page(page); |
be068f29 | 3143 | wp_page_reuse(vmf); |
09854ba9 | 3144 | return VM_FAULT_WRITE; |
ee6a6457 | 3145 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 3146 | (VM_WRITE|VM_SHARED))) { |
a41b70d6 | 3147 | return wp_page_shared(vmf); |
1da177e4 | 3148 | } |
52d1e606 | 3149 | copy: |
1da177e4 LT |
3150 | /* |
3151 | * Ok, we need to copy. Oh, well.. | |
3152 | */ | |
a41b70d6 | 3153 | get_page(vmf->page); |
28766805 | 3154 | |
82b0f8c3 | 3155 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 3156 | return wp_page_copy(vmf); |
1da177e4 LT |
3157 | } |
3158 | ||
97a89413 | 3159 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
3160 | unsigned long start_addr, unsigned long end_addr, |
3161 | struct zap_details *details) | |
3162 | { | |
f5cc4eef | 3163 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
3164 | } |
3165 | ||
f808c13f | 3166 | static inline void unmap_mapping_range_tree(struct rb_root_cached *root, |
1da177e4 LT |
3167 | struct zap_details *details) |
3168 | { | |
3169 | struct vm_area_struct *vma; | |
1da177e4 LT |
3170 | pgoff_t vba, vea, zba, zea; |
3171 | ||
6b2dbba8 | 3172 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 3173 | details->first_index, details->last_index) { |
1da177e4 LT |
3174 | |
3175 | vba = vma->vm_pgoff; | |
d6e93217 | 3176 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
3177 | zba = details->first_index; |
3178 | if (zba < vba) | |
3179 | zba = vba; | |
3180 | zea = details->last_index; | |
3181 | if (zea > vea) | |
3182 | zea = vea; | |
3183 | ||
97a89413 | 3184 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
3185 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
3186 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 3187 | details); |
1da177e4 LT |
3188 | } |
3189 | } | |
3190 | ||
977fbdcd MW |
3191 | /** |
3192 | * unmap_mapping_pages() - Unmap pages from processes. | |
3193 | * @mapping: The address space containing pages to be unmapped. | |
3194 | * @start: Index of first page to be unmapped. | |
3195 | * @nr: Number of pages to be unmapped. 0 to unmap to end of file. | |
3196 | * @even_cows: Whether to unmap even private COWed pages. | |
3197 | * | |
3198 | * Unmap the pages in this address space from any userspace process which | |
3199 | * has them mmaped. Generally, you want to remove COWed pages as well when | |
3200 | * a file is being truncated, but not when invalidating pages from the page | |
3201 | * cache. | |
3202 | */ | |
3203 | void unmap_mapping_pages(struct address_space *mapping, pgoff_t start, | |
3204 | pgoff_t nr, bool even_cows) | |
3205 | { | |
3206 | struct zap_details details = { }; | |
3207 | ||
3208 | details.check_mapping = even_cows ? NULL : mapping; | |
3209 | details.first_index = start; | |
3210 | details.last_index = start + nr - 1; | |
3211 | if (details.last_index < details.first_index) | |
3212 | details.last_index = ULONG_MAX; | |
3213 | ||
3214 | i_mmap_lock_write(mapping); | |
3215 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) | |
3216 | unmap_mapping_range_tree(&mapping->i_mmap, &details); | |
3217 | i_mmap_unlock_write(mapping); | |
3218 | } | |
3219 | ||
1da177e4 | 3220 | /** |
8a5f14a2 | 3221 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
977fbdcd | 3222 | * address_space corresponding to the specified byte range in the underlying |
8a5f14a2 KS |
3223 | * file. |
3224 | * | |
3d41088f | 3225 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
3226 | * @holebegin: byte in first page to unmap, relative to the start of |
3227 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 3228 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
3229 | * must keep the partial page. In contrast, we must get rid of |
3230 | * partial pages. | |
3231 | * @holelen: size of prospective hole in bytes. This will be rounded | |
3232 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
3233 | * end of the file. | |
3234 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
3235 | * but 0 when invalidating pagecache, don't throw away private data. | |
3236 | */ | |
3237 | void unmap_mapping_range(struct address_space *mapping, | |
3238 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
3239 | { | |
1da177e4 LT |
3240 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
3241 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
3242 | ||
3243 | /* Check for overflow. */ | |
3244 | if (sizeof(holelen) > sizeof(hlen)) { | |
3245 | long long holeend = | |
3246 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
3247 | if (holeend & ~(long long)ULONG_MAX) | |
3248 | hlen = ULONG_MAX - hba + 1; | |
3249 | } | |
3250 | ||
977fbdcd | 3251 | unmap_mapping_pages(mapping, hba, hlen, even_cows); |
1da177e4 LT |
3252 | } |
3253 | EXPORT_SYMBOL(unmap_mapping_range); | |
3254 | ||
1da177e4 | 3255 | /* |
c1e8d7c6 | 3256 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
8f4e2101 | 3257 | * but allow concurrent faults), and pte mapped but not yet locked. |
9a95f3cf PC |
3258 | * We return with pte unmapped and unlocked. |
3259 | * | |
c1e8d7c6 | 3260 | * We return with the mmap_lock locked or unlocked in the same cases |
9a95f3cf | 3261 | * as does filemap_fault(). |
1da177e4 | 3262 | */ |
2b740303 | 3263 | vm_fault_t do_swap_page(struct vm_fault *vmf) |
1da177e4 | 3264 | { |
82b0f8c3 | 3265 | struct vm_area_struct *vma = vmf->vma; |
eaf649eb | 3266 | struct page *page = NULL, *swapcache; |
65500d23 | 3267 | swp_entry_t entry; |
1da177e4 | 3268 | pte_t pte; |
d065bd81 | 3269 | int locked; |
ad8c2ee8 | 3270 | int exclusive = 0; |
2b740303 | 3271 | vm_fault_t ret = 0; |
aae466b0 | 3272 | void *shadow = NULL; |
1da177e4 | 3273 | |
eaf649eb | 3274 | if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte)) |
8f4e2101 | 3275 | goto out; |
65500d23 | 3276 | |
2994302b | 3277 | entry = pte_to_swp_entry(vmf->orig_pte); |
d1737fdb AK |
3278 | if (unlikely(non_swap_entry(entry))) { |
3279 | if (is_migration_entry(entry)) { | |
82b0f8c3 JK |
3280 | migration_entry_wait(vma->vm_mm, vmf->pmd, |
3281 | vmf->address); | |
5042db43 | 3282 | } else if (is_device_private_entry(entry)) { |
897e6365 CH |
3283 | vmf->page = device_private_entry_to_page(entry); |
3284 | ret = vmf->page->pgmap->ops->migrate_to_ram(vmf); | |
d1737fdb AK |
3285 | } else if (is_hwpoison_entry(entry)) { |
3286 | ret = VM_FAULT_HWPOISON; | |
3287 | } else { | |
2994302b | 3288 | print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); |
d99be1a8 | 3289 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 3290 | } |
0697212a CL |
3291 | goto out; |
3292 | } | |
0bcac06f MK |
3293 | |
3294 | ||
0ff92245 | 3295 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
eaf649eb MK |
3296 | page = lookup_swap_cache(entry, vma, vmf->address); |
3297 | swapcache = page; | |
f8020772 | 3298 | |
1da177e4 | 3299 | if (!page) { |
0bcac06f MK |
3300 | struct swap_info_struct *si = swp_swap_info(entry); |
3301 | ||
a449bf58 QC |
3302 | if (data_race(si->flags & SWP_SYNCHRONOUS_IO) && |
3303 | __swap_count(entry) == 1) { | |
0bcac06f | 3304 | /* skip swapcache */ |
e9e9b7ec MK |
3305 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
3306 | vmf->address); | |
0bcac06f | 3307 | if (page) { |
4c6355b2 JW |
3308 | int err; |
3309 | ||
0bcac06f MK |
3310 | __SetPageLocked(page); |
3311 | __SetPageSwapBacked(page); | |
3312 | set_page_private(page, entry.val); | |
4c6355b2 JW |
3313 | |
3314 | /* Tell memcg to use swap ownership records */ | |
3315 | SetPageSwapCache(page); | |
3316 | err = mem_cgroup_charge(page, vma->vm_mm, | |
d9eb1ea2 | 3317 | GFP_KERNEL); |
4c6355b2 | 3318 | ClearPageSwapCache(page); |
545b1b07 MH |
3319 | if (err) { |
3320 | ret = VM_FAULT_OOM; | |
4c6355b2 | 3321 | goto out_page; |
545b1b07 | 3322 | } |
4c6355b2 | 3323 | |
aae466b0 JK |
3324 | shadow = get_shadow_from_swap_cache(entry); |
3325 | if (shadow) | |
3326 | workingset_refault(page, shadow); | |
0076f029 | 3327 | |
6058eaec | 3328 | lru_cache_add(page); |
0bcac06f MK |
3329 | swap_readpage(page, true); |
3330 | } | |
aa8d22a1 | 3331 | } else { |
e9e9b7ec MK |
3332 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
3333 | vmf); | |
aa8d22a1 | 3334 | swapcache = page; |
0bcac06f MK |
3335 | } |
3336 | ||
1da177e4 LT |
3337 | if (!page) { |
3338 | /* | |
8f4e2101 HD |
3339 | * Back out if somebody else faulted in this pte |
3340 | * while we released the pte lock. | |
1da177e4 | 3341 | */ |
82b0f8c3 JK |
3342 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
3343 | vmf->address, &vmf->ptl); | |
2994302b | 3344 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) |
1da177e4 | 3345 | ret = VM_FAULT_OOM; |
0ff92245 | 3346 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 3347 | goto unlock; |
1da177e4 LT |
3348 | } |
3349 | ||
3350 | /* Had to read the page from swap area: Major fault */ | |
3351 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 3352 | count_vm_event(PGMAJFAULT); |
2262185c | 3353 | count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); |
d1737fdb | 3354 | } else if (PageHWPoison(page)) { |
71f72525 WF |
3355 | /* |
3356 | * hwpoisoned dirty swapcache pages are kept for killing | |
3357 | * owner processes (which may be unknown at hwpoison time) | |
3358 | */ | |
d1737fdb AK |
3359 | ret = VM_FAULT_HWPOISON; |
3360 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
4779cb31 | 3361 | goto out_release; |
1da177e4 LT |
3362 | } |
3363 | ||
82b0f8c3 | 3364 | locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags); |
e709ffd6 | 3365 | |
073e587e | 3366 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
3367 | if (!locked) { |
3368 | ret |= VM_FAULT_RETRY; | |
3369 | goto out_release; | |
3370 | } | |
073e587e | 3371 | |
4969c119 | 3372 | /* |
31c4a3d3 HD |
3373 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
3374 | * release the swapcache from under us. The page pin, and pte_same | |
3375 | * test below, are not enough to exclude that. Even if it is still | |
3376 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 3377 | */ |
0bcac06f MK |
3378 | if (unlikely((!PageSwapCache(page) || |
3379 | page_private(page) != entry.val)) && swapcache) | |
4969c119 AA |
3380 | goto out_page; |
3381 | ||
82b0f8c3 | 3382 | page = ksm_might_need_to_copy(page, vma, vmf->address); |
cbf86cfe HD |
3383 | if (unlikely(!page)) { |
3384 | ret = VM_FAULT_OOM; | |
3385 | page = swapcache; | |
cbf86cfe | 3386 | goto out_page; |
5ad64688 HD |
3387 | } |
3388 | ||
9d82c694 | 3389 | cgroup_throttle_swaprate(page, GFP_KERNEL); |
8a9f3ccd | 3390 | |
1da177e4 | 3391 | /* |
8f4e2101 | 3392 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 3393 | */ |
82b0f8c3 JK |
3394 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3395 | &vmf->ptl); | |
2994302b | 3396 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) |
b8107480 | 3397 | goto out_nomap; |
b8107480 KK |
3398 | |
3399 | if (unlikely(!PageUptodate(page))) { | |
3400 | ret = VM_FAULT_SIGBUS; | |
3401 | goto out_nomap; | |
1da177e4 LT |
3402 | } |
3403 | ||
8c7c6e34 KH |
3404 | /* |
3405 | * The page isn't present yet, go ahead with the fault. | |
3406 | * | |
3407 | * Be careful about the sequence of operations here. | |
3408 | * To get its accounting right, reuse_swap_page() must be called | |
3409 | * while the page is counted on swap but not yet in mapcount i.e. | |
3410 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
3411 | * must be called after the swap_free(), or it will never succeed. | |
8c7c6e34 | 3412 | */ |
1da177e4 | 3413 | |
bae473a4 KS |
3414 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
3415 | dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS); | |
1da177e4 | 3416 | pte = mk_pte(page, vma->vm_page_prot); |
82b0f8c3 | 3417 | if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { |
1da177e4 | 3418 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
82b0f8c3 | 3419 | vmf->flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 3420 | ret |= VM_FAULT_WRITE; |
d281ee61 | 3421 | exclusive = RMAP_EXCLUSIVE; |
1da177e4 | 3422 | } |
1da177e4 | 3423 | flush_icache_page(vma, page); |
2994302b | 3424 | if (pte_swp_soft_dirty(vmf->orig_pte)) |
179ef71c | 3425 | pte = pte_mksoft_dirty(pte); |
f45ec5ff PX |
3426 | if (pte_swp_uffd_wp(vmf->orig_pte)) { |
3427 | pte = pte_mkuffd_wp(pte); | |
3428 | pte = pte_wrprotect(pte); | |
3429 | } | |
82b0f8c3 | 3430 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); |
ca827d55 | 3431 | arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte); |
2994302b | 3432 | vmf->orig_pte = pte; |
0bcac06f MK |
3433 | |
3434 | /* ksm created a completely new copy */ | |
3435 | if (unlikely(page != swapcache && swapcache)) { | |
82b0f8c3 | 3436 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
b518154e | 3437 | lru_cache_add_inactive_or_unevictable(page, vma); |
0bcac06f MK |
3438 | } else { |
3439 | do_page_add_anon_rmap(page, vma, vmf->address, exclusive); | |
00501b53 | 3440 | } |
1da177e4 | 3441 | |
c475a8ab | 3442 | swap_free(entry); |
5ccc5aba VD |
3443 | if (mem_cgroup_swap_full(page) || |
3444 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 3445 | try_to_free_swap(page); |
c475a8ab | 3446 | unlock_page(page); |
0bcac06f | 3447 | if (page != swapcache && swapcache) { |
4969c119 AA |
3448 | /* |
3449 | * Hold the lock to avoid the swap entry to be reused | |
3450 | * until we take the PT lock for the pte_same() check | |
3451 | * (to avoid false positives from pte_same). For | |
3452 | * further safety release the lock after the swap_free | |
3453 | * so that the swap count won't change under a | |
3454 | * parallel locked swapcache. | |
3455 | */ | |
3456 | unlock_page(swapcache); | |
09cbfeaf | 3457 | put_page(swapcache); |
4969c119 | 3458 | } |
c475a8ab | 3459 | |
82b0f8c3 | 3460 | if (vmf->flags & FAULT_FLAG_WRITE) { |
2994302b | 3461 | ret |= do_wp_page(vmf); |
61469f1d HD |
3462 | if (ret & VM_FAULT_ERROR) |
3463 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
3464 | goto out; |
3465 | } | |
3466 | ||
3467 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3468 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3469 | unlock: |
82b0f8c3 | 3470 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
1da177e4 LT |
3471 | out: |
3472 | return ret; | |
b8107480 | 3473 | out_nomap: |
82b0f8c3 | 3474 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bc43f75c | 3475 | out_page: |
b8107480 | 3476 | unlock_page(page); |
4779cb31 | 3477 | out_release: |
09cbfeaf | 3478 | put_page(page); |
0bcac06f | 3479 | if (page != swapcache && swapcache) { |
4969c119 | 3480 | unlock_page(swapcache); |
09cbfeaf | 3481 | put_page(swapcache); |
4969c119 | 3482 | } |
65500d23 | 3483 | return ret; |
1da177e4 LT |
3484 | } |
3485 | ||
3486 | /* | |
c1e8d7c6 | 3487 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
8f4e2101 | 3488 | * but allow concurrent faults), and pte mapped but not yet locked. |
c1e8d7c6 | 3489 | * We return with mmap_lock still held, but pte unmapped and unlocked. |
1da177e4 | 3490 | */ |
2b740303 | 3491 | static vm_fault_t do_anonymous_page(struct vm_fault *vmf) |
1da177e4 | 3492 | { |
82b0f8c3 | 3493 | struct vm_area_struct *vma = vmf->vma; |
8f4e2101 | 3494 | struct page *page; |
2b740303 | 3495 | vm_fault_t ret = 0; |
1da177e4 | 3496 | pte_t entry; |
1da177e4 | 3497 | |
6b7339f4 KS |
3498 | /* File mapping without ->vm_ops ? */ |
3499 | if (vma->vm_flags & VM_SHARED) | |
3500 | return VM_FAULT_SIGBUS; | |
3501 | ||
7267ec00 KS |
3502 | /* |
3503 | * Use pte_alloc() instead of pte_alloc_map(). We can't run | |
3504 | * pte_offset_map() on pmds where a huge pmd might be created | |
3505 | * from a different thread. | |
3506 | * | |
3e4e28c5 | 3507 | * pte_alloc_map() is safe to use under mmap_write_lock(mm) or when |
7267ec00 KS |
3508 | * parallel threads are excluded by other means. |
3509 | * | |
3e4e28c5 | 3510 | * Here we only have mmap_read_lock(mm). |
7267ec00 | 3511 | */ |
4cf58924 | 3512 | if (pte_alloc(vma->vm_mm, vmf->pmd)) |
7267ec00 KS |
3513 | return VM_FAULT_OOM; |
3514 | ||
3515 | /* See the comment in pte_alloc_one_map() */ | |
82b0f8c3 | 3516 | if (unlikely(pmd_trans_unstable(vmf->pmd))) |
7267ec00 KS |
3517 | return 0; |
3518 | ||
11ac5524 | 3519 | /* Use the zero-page for reads */ |
82b0f8c3 | 3520 | if (!(vmf->flags & FAULT_FLAG_WRITE) && |
bae473a4 | 3521 | !mm_forbids_zeropage(vma->vm_mm)) { |
82b0f8c3 | 3522 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), |
62eede62 | 3523 | vma->vm_page_prot)); |
82b0f8c3 JK |
3524 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
3525 | vmf->address, &vmf->ptl); | |
7df67697 BM |
3526 | if (!pte_none(*vmf->pte)) { |
3527 | update_mmu_tlb(vma, vmf->address, vmf->pte); | |
a13ea5b7 | 3528 | goto unlock; |
7df67697 | 3529 | } |
6b31d595 MH |
3530 | ret = check_stable_address_space(vma->vm_mm); |
3531 | if (ret) | |
3532 | goto unlock; | |
6b251fc9 AA |
3533 | /* Deliver the page fault to userland, check inside PT lock */ |
3534 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 JK |
3535 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3536 | return handle_userfault(vmf, VM_UFFD_MISSING); | |
6b251fc9 | 3537 | } |
a13ea5b7 HD |
3538 | goto setpte; |
3539 | } | |
3540 | ||
557ed1fa | 3541 | /* Allocate our own private page. */ |
557ed1fa NP |
3542 | if (unlikely(anon_vma_prepare(vma))) |
3543 | goto oom; | |
82b0f8c3 | 3544 | page = alloc_zeroed_user_highpage_movable(vma, vmf->address); |
557ed1fa NP |
3545 | if (!page) |
3546 | goto oom; | |
eb3c24f3 | 3547 | |
d9eb1ea2 | 3548 | if (mem_cgroup_charge(page, vma->vm_mm, GFP_KERNEL)) |
eb3c24f3 | 3549 | goto oom_free_page; |
9d82c694 | 3550 | cgroup_throttle_swaprate(page, GFP_KERNEL); |
eb3c24f3 | 3551 | |
52f37629 MK |
3552 | /* |
3553 | * The memory barrier inside __SetPageUptodate makes sure that | |
f4f5329d | 3554 | * preceding stores to the page contents become visible before |
52f37629 MK |
3555 | * the set_pte_at() write. |
3556 | */ | |
0ed361de | 3557 | __SetPageUptodate(page); |
8f4e2101 | 3558 | |
557ed1fa | 3559 | entry = mk_pte(page, vma->vm_page_prot); |
44bf431b | 3560 | entry = pte_sw_mkyoung(entry); |
1ac0cb5d HD |
3561 | if (vma->vm_flags & VM_WRITE) |
3562 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 3563 | |
82b0f8c3 JK |
3564 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3565 | &vmf->ptl); | |
7df67697 BM |
3566 | if (!pte_none(*vmf->pte)) { |
3567 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
557ed1fa | 3568 | goto release; |
7df67697 | 3569 | } |
9ba69294 | 3570 | |
6b31d595 MH |
3571 | ret = check_stable_address_space(vma->vm_mm); |
3572 | if (ret) | |
3573 | goto release; | |
3574 | ||
6b251fc9 AA |
3575 | /* Deliver the page fault to userland, check inside PT lock */ |
3576 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 | 3577 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
09cbfeaf | 3578 | put_page(page); |
82b0f8c3 | 3579 | return handle_userfault(vmf, VM_UFFD_MISSING); |
6b251fc9 AA |
3580 | } |
3581 | ||
bae473a4 | 3582 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3583 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
b518154e | 3584 | lru_cache_add_inactive_or_unevictable(page, vma); |
a13ea5b7 | 3585 | setpte: |
82b0f8c3 | 3586 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
1da177e4 LT |
3587 | |
3588 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3589 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3590 | unlock: |
82b0f8c3 | 3591 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
6b31d595 | 3592 | return ret; |
8f4e2101 | 3593 | release: |
09cbfeaf | 3594 | put_page(page); |
8f4e2101 | 3595 | goto unlock; |
8a9f3ccd | 3596 | oom_free_page: |
09cbfeaf | 3597 | put_page(page); |
65500d23 | 3598 | oom: |
1da177e4 LT |
3599 | return VM_FAULT_OOM; |
3600 | } | |
3601 | ||
9a95f3cf | 3602 | /* |
c1e8d7c6 | 3603 | * The mmap_lock must have been held on entry, and may have been |
9a95f3cf PC |
3604 | * released depending on flags and vma->vm_ops->fault() return value. |
3605 | * See filemap_fault() and __lock_page_retry(). | |
3606 | */ | |
2b740303 | 3607 | static vm_fault_t __do_fault(struct vm_fault *vmf) |
7eae74af | 3608 | { |
82b0f8c3 | 3609 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 3610 | vm_fault_t ret; |
7eae74af | 3611 | |
63f3655f MH |
3612 | /* |
3613 | * Preallocate pte before we take page_lock because this might lead to | |
3614 | * deadlocks for memcg reclaim which waits for pages under writeback: | |
3615 | * lock_page(A) | |
3616 | * SetPageWriteback(A) | |
3617 | * unlock_page(A) | |
3618 | * lock_page(B) | |
3619 | * lock_page(B) | |
d383807a | 3620 | * pte_alloc_one |
63f3655f MH |
3621 | * shrink_page_list |
3622 | * wait_on_page_writeback(A) | |
3623 | * SetPageWriteback(B) | |
3624 | * unlock_page(B) | |
3625 | * # flush A, B to clear the writeback | |
3626 | */ | |
3627 | if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) { | |
a7069ee3 | 3628 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); |
63f3655f MH |
3629 | if (!vmf->prealloc_pte) |
3630 | return VM_FAULT_OOM; | |
3631 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3632 | } | |
3633 | ||
11bac800 | 3634 | ret = vma->vm_ops->fault(vmf); |
3917048d | 3635 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | |
b1aa812b | 3636 | VM_FAULT_DONE_COW))) |
bc2466e4 | 3637 | return ret; |
7eae74af | 3638 | |
667240e0 | 3639 | if (unlikely(PageHWPoison(vmf->page))) { |
7eae74af | 3640 | if (ret & VM_FAULT_LOCKED) |
667240e0 JK |
3641 | unlock_page(vmf->page); |
3642 | put_page(vmf->page); | |
936ca80d | 3643 | vmf->page = NULL; |
7eae74af KS |
3644 | return VM_FAULT_HWPOISON; |
3645 | } | |
3646 | ||
3647 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
667240e0 | 3648 | lock_page(vmf->page); |
7eae74af | 3649 | else |
667240e0 | 3650 | VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); |
7eae74af | 3651 | |
7eae74af KS |
3652 | return ret; |
3653 | } | |
3654 | ||
d0f0931d RZ |
3655 | /* |
3656 | * The ordering of these checks is important for pmds with _PAGE_DEVMAP set. | |
3657 | * If we check pmd_trans_unstable() first we will trip the bad_pmd() check | |
3658 | * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly | |
3659 | * returning 1 but not before it spams dmesg with the pmd_clear_bad() output. | |
3660 | */ | |
3661 | static int pmd_devmap_trans_unstable(pmd_t *pmd) | |
3662 | { | |
3663 | return pmd_devmap(*pmd) || pmd_trans_unstable(pmd); | |
3664 | } | |
3665 | ||
2b740303 | 3666 | static vm_fault_t pte_alloc_one_map(struct vm_fault *vmf) |
7267ec00 | 3667 | { |
82b0f8c3 | 3668 | struct vm_area_struct *vma = vmf->vma; |
7267ec00 | 3669 | |
82b0f8c3 | 3670 | if (!pmd_none(*vmf->pmd)) |
7267ec00 | 3671 | goto map_pte; |
82b0f8c3 JK |
3672 | if (vmf->prealloc_pte) { |
3673 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); | |
3674 | if (unlikely(!pmd_none(*vmf->pmd))) { | |
3675 | spin_unlock(vmf->ptl); | |
7267ec00 KS |
3676 | goto map_pte; |
3677 | } | |
3678 | ||
c4812909 | 3679 | mm_inc_nr_ptes(vma->vm_mm); |
82b0f8c3 JK |
3680 | pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
3681 | spin_unlock(vmf->ptl); | |
7f2b6ce8 | 3682 | vmf->prealloc_pte = NULL; |
4cf58924 | 3683 | } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd))) { |
7267ec00 KS |
3684 | return VM_FAULT_OOM; |
3685 | } | |
3686 | map_pte: | |
3687 | /* | |
3688 | * If a huge pmd materialized under us just retry later. Use | |
d0f0931d RZ |
3689 | * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of |
3690 | * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge | |
3691 | * under us and then back to pmd_none, as a result of MADV_DONTNEED | |
3692 | * running immediately after a huge pmd fault in a different thread of | |
3693 | * this mm, in turn leading to a misleading pmd_trans_huge() retval. | |
3694 | * All we have to ensure is that it is a regular pmd that we can walk | |
3695 | * with pte_offset_map() and we can do that through an atomic read in | |
3696 | * C, which is what pmd_trans_unstable() provides. | |
7267ec00 | 3697 | */ |
d0f0931d | 3698 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3699 | return VM_FAULT_NOPAGE; |
3700 | ||
d0f0931d RZ |
3701 | /* |
3702 | * At this point we know that our vmf->pmd points to a page of ptes | |
3703 | * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge() | |
3704 | * for the duration of the fault. If a racing MADV_DONTNEED runs and | |
3705 | * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still | |
3706 | * be valid and we will re-check to make sure the vmf->pte isn't | |
3707 | * pte_none() under vmf->ptl protection when we return to | |
3708 | * alloc_set_pte(). | |
3709 | */ | |
82b0f8c3 JK |
3710 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3711 | &vmf->ptl); | |
7267ec00 KS |
3712 | return 0; |
3713 | } | |
3714 | ||
396bcc52 | 3715 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
82b0f8c3 | 3716 | static void deposit_prealloc_pte(struct vm_fault *vmf) |
953c66c2 | 3717 | { |
82b0f8c3 | 3718 | struct vm_area_struct *vma = vmf->vma; |
953c66c2 | 3719 | |
82b0f8c3 | 3720 | pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
953c66c2 AK |
3721 | /* |
3722 | * We are going to consume the prealloc table, | |
3723 | * count that as nr_ptes. | |
3724 | */ | |
c4812909 | 3725 | mm_inc_nr_ptes(vma->vm_mm); |
7f2b6ce8 | 3726 | vmf->prealloc_pte = NULL; |
953c66c2 AK |
3727 | } |
3728 | ||
2b740303 | 3729 | static vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3730 | { |
82b0f8c3 JK |
3731 | struct vm_area_struct *vma = vmf->vma; |
3732 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3733 | unsigned long haddr = vmf->address & HPAGE_PMD_MASK; | |
10102459 | 3734 | pmd_t entry; |
2b740303 | 3735 | int i; |
d01ac3c3 | 3736 | vm_fault_t ret = VM_FAULT_FALLBACK; |
10102459 KS |
3737 | |
3738 | if (!transhuge_vma_suitable(vma, haddr)) | |
d01ac3c3 | 3739 | return ret; |
10102459 | 3740 | |
10102459 | 3741 | page = compound_head(page); |
d01ac3c3 MWO |
3742 | if (compound_order(page) != HPAGE_PMD_ORDER) |
3743 | return ret; | |
10102459 | 3744 | |
953c66c2 AK |
3745 | /* |
3746 | * Archs like ppc64 need additonal space to store information | |
3747 | * related to pte entry. Use the preallocated table for that. | |
3748 | */ | |
82b0f8c3 | 3749 | if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { |
4cf58924 | 3750 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); |
82b0f8c3 | 3751 | if (!vmf->prealloc_pte) |
953c66c2 AK |
3752 | return VM_FAULT_OOM; |
3753 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3754 | } | |
3755 | ||
82b0f8c3 JK |
3756 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); |
3757 | if (unlikely(!pmd_none(*vmf->pmd))) | |
10102459 KS |
3758 | goto out; |
3759 | ||
3760 | for (i = 0; i < HPAGE_PMD_NR; i++) | |
3761 | flush_icache_page(vma, page + i); | |
3762 | ||
3763 | entry = mk_huge_pmd(page, vma->vm_page_prot); | |
3764 | if (write) | |
f55e1014 | 3765 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); |
10102459 | 3766 | |
fadae295 | 3767 | add_mm_counter(vma->vm_mm, mm_counter_file(page), HPAGE_PMD_NR); |
10102459 | 3768 | page_add_file_rmap(page, true); |
953c66c2 AK |
3769 | /* |
3770 | * deposit and withdraw with pmd lock held | |
3771 | */ | |
3772 | if (arch_needs_pgtable_deposit()) | |
82b0f8c3 | 3773 | deposit_prealloc_pte(vmf); |
10102459 | 3774 | |
82b0f8c3 | 3775 | set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); |
10102459 | 3776 | |
82b0f8c3 | 3777 | update_mmu_cache_pmd(vma, haddr, vmf->pmd); |
10102459 KS |
3778 | |
3779 | /* fault is handled */ | |
3780 | ret = 0; | |
95ecedcd | 3781 | count_vm_event(THP_FILE_MAPPED); |
10102459 | 3782 | out: |
82b0f8c3 | 3783 | spin_unlock(vmf->ptl); |
10102459 KS |
3784 | return ret; |
3785 | } | |
3786 | #else | |
2b740303 | 3787 | static vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 KS |
3788 | { |
3789 | BUILD_BUG(); | |
3790 | return 0; | |
3791 | } | |
3792 | #endif | |
3793 | ||
8c6e50b0 | 3794 | /** |
7267ec00 | 3795 | * alloc_set_pte - setup new PTE entry for given page and add reverse page |
f1dc1685 | 3796 | * mapping. If needed, the function allocates page table or use pre-allocated. |
8c6e50b0 | 3797 | * |
82b0f8c3 | 3798 | * @vmf: fault environment |
8c6e50b0 | 3799 | * @page: page to map |
8c6e50b0 | 3800 | * |
82b0f8c3 JK |
3801 | * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on |
3802 | * return. | |
8c6e50b0 KS |
3803 | * |
3804 | * Target users are page handler itself and implementations of | |
3805 | * vm_ops->map_pages. | |
a862f68a MR |
3806 | * |
3807 | * Return: %0 on success, %VM_FAULT_ code in case of error. | |
8c6e50b0 | 3808 | */ |
9d82c694 | 3809 | vm_fault_t alloc_set_pte(struct vm_fault *vmf, struct page *page) |
3bb97794 | 3810 | { |
82b0f8c3 JK |
3811 | struct vm_area_struct *vma = vmf->vma; |
3812 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3bb97794 | 3813 | pte_t entry; |
2b740303 | 3814 | vm_fault_t ret; |
10102459 | 3815 | |
396bcc52 | 3816 | if (pmd_none(*vmf->pmd) && PageTransCompound(page)) { |
82b0f8c3 | 3817 | ret = do_set_pmd(vmf, page); |
10102459 | 3818 | if (ret != VM_FAULT_FALLBACK) |
b0b9b3df | 3819 | return ret; |
10102459 | 3820 | } |
3bb97794 | 3821 | |
82b0f8c3 JK |
3822 | if (!vmf->pte) { |
3823 | ret = pte_alloc_one_map(vmf); | |
7267ec00 | 3824 | if (ret) |
b0b9b3df | 3825 | return ret; |
7267ec00 KS |
3826 | } |
3827 | ||
3828 | /* Re-check under ptl */ | |
7df67697 BM |
3829 | if (unlikely(!pte_none(*vmf->pte))) { |
3830 | update_mmu_tlb(vma, vmf->address, vmf->pte); | |
b0b9b3df | 3831 | return VM_FAULT_NOPAGE; |
7df67697 | 3832 | } |
7267ec00 | 3833 | |
3bb97794 KS |
3834 | flush_icache_page(vma, page); |
3835 | entry = mk_pte(page, vma->vm_page_prot); | |
44bf431b | 3836 | entry = pte_sw_mkyoung(entry); |
3bb97794 KS |
3837 | if (write) |
3838 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
bae473a4 KS |
3839 | /* copy-on-write page */ |
3840 | if (write && !(vma->vm_flags & VM_SHARED)) { | |
3bb97794 | 3841 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3842 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
b518154e | 3843 | lru_cache_add_inactive_or_unevictable(page, vma); |
3bb97794 | 3844 | } else { |
eca56ff9 | 3845 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
dd78fedd | 3846 | page_add_file_rmap(page, false); |
3bb97794 | 3847 | } |
82b0f8c3 | 3848 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
3bb97794 KS |
3849 | |
3850 | /* no need to invalidate: a not-present page won't be cached */ | |
82b0f8c3 | 3851 | update_mmu_cache(vma, vmf->address, vmf->pte); |
7267ec00 | 3852 | |
b0b9b3df | 3853 | return 0; |
3bb97794 KS |
3854 | } |
3855 | ||
9118c0cb JK |
3856 | |
3857 | /** | |
3858 | * finish_fault - finish page fault once we have prepared the page to fault | |
3859 | * | |
3860 | * @vmf: structure describing the fault | |
3861 | * | |
3862 | * This function handles all that is needed to finish a page fault once the | |
3863 | * page to fault in is prepared. It handles locking of PTEs, inserts PTE for | |
3864 | * given page, adds reverse page mapping, handles memcg charges and LRU | |
a862f68a | 3865 | * addition. |
9118c0cb JK |
3866 | * |
3867 | * The function expects the page to be locked and on success it consumes a | |
3868 | * reference of a page being mapped (for the PTE which maps it). | |
a862f68a MR |
3869 | * |
3870 | * Return: %0 on success, %VM_FAULT_ code in case of error. | |
9118c0cb | 3871 | */ |
2b740303 | 3872 | vm_fault_t finish_fault(struct vm_fault *vmf) |
9118c0cb JK |
3873 | { |
3874 | struct page *page; | |
2b740303 | 3875 | vm_fault_t ret = 0; |
9118c0cb JK |
3876 | |
3877 | /* Did we COW the page? */ | |
3878 | if ((vmf->flags & FAULT_FLAG_WRITE) && | |
3879 | !(vmf->vma->vm_flags & VM_SHARED)) | |
3880 | page = vmf->cow_page; | |
3881 | else | |
3882 | page = vmf->page; | |
6b31d595 MH |
3883 | |
3884 | /* | |
3885 | * check even for read faults because we might have lost our CoWed | |
3886 | * page | |
3887 | */ | |
3888 | if (!(vmf->vma->vm_flags & VM_SHARED)) | |
3889 | ret = check_stable_address_space(vmf->vma->vm_mm); | |
3890 | if (!ret) | |
9d82c694 | 3891 | ret = alloc_set_pte(vmf, page); |
9118c0cb JK |
3892 | if (vmf->pte) |
3893 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3894 | return ret; | |
3895 | } | |
3896 | ||
3a91053a KS |
3897 | static unsigned long fault_around_bytes __read_mostly = |
3898 | rounddown_pow_of_two(65536); | |
a9b0f861 | 3899 | |
a9b0f861 KS |
3900 | #ifdef CONFIG_DEBUG_FS |
3901 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 3902 | { |
a9b0f861 | 3903 | *val = fault_around_bytes; |
1592eef0 KS |
3904 | return 0; |
3905 | } | |
3906 | ||
b4903d6e | 3907 | /* |
da391d64 WK |
3908 | * fault_around_bytes must be rounded down to the nearest page order as it's |
3909 | * what do_fault_around() expects to see. | |
b4903d6e | 3910 | */ |
a9b0f861 | 3911 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 3912 | { |
a9b0f861 | 3913 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 3914 | return -EINVAL; |
b4903d6e AR |
3915 | if (val > PAGE_SIZE) |
3916 | fault_around_bytes = rounddown_pow_of_two(val); | |
3917 | else | |
3918 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
3919 | return 0; |
3920 | } | |
0a1345f8 | 3921 | DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, |
a9b0f861 | 3922 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
1592eef0 KS |
3923 | |
3924 | static int __init fault_around_debugfs(void) | |
3925 | { | |
d9f7979c GKH |
3926 | debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, |
3927 | &fault_around_bytes_fops); | |
1592eef0 KS |
3928 | return 0; |
3929 | } | |
3930 | late_initcall(fault_around_debugfs); | |
1592eef0 | 3931 | #endif |
8c6e50b0 | 3932 | |
1fdb412b KS |
3933 | /* |
3934 | * do_fault_around() tries to map few pages around the fault address. The hope | |
3935 | * is that the pages will be needed soon and this will lower the number of | |
3936 | * faults to handle. | |
3937 | * | |
3938 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
3939 | * not ready to be mapped: not up-to-date, locked, etc. | |
3940 | * | |
3941 | * This function is called with the page table lock taken. In the split ptlock | |
3942 | * case the page table lock only protects only those entries which belong to | |
3943 | * the page table corresponding to the fault address. | |
3944 | * | |
3945 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
3946 | * only once. | |
3947 | * | |
da391d64 WK |
3948 | * fault_around_bytes defines how many bytes we'll try to map. |
3949 | * do_fault_around() expects it to be set to a power of two less than or equal | |
3950 | * to PTRS_PER_PTE. | |
1fdb412b | 3951 | * |
da391d64 WK |
3952 | * The virtual address of the area that we map is naturally aligned to |
3953 | * fault_around_bytes rounded down to the machine page size | |
3954 | * (and therefore to page order). This way it's easier to guarantee | |
3955 | * that we don't cross page table boundaries. | |
1fdb412b | 3956 | */ |
2b740303 | 3957 | static vm_fault_t do_fault_around(struct vm_fault *vmf) |
8c6e50b0 | 3958 | { |
82b0f8c3 | 3959 | unsigned long address = vmf->address, nr_pages, mask; |
0721ec8b | 3960 | pgoff_t start_pgoff = vmf->pgoff; |
bae473a4 | 3961 | pgoff_t end_pgoff; |
2b740303 SJ |
3962 | int off; |
3963 | vm_fault_t ret = 0; | |
8c6e50b0 | 3964 | |
4db0c3c2 | 3965 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
3966 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
3967 | ||
82b0f8c3 JK |
3968 | vmf->address = max(address & mask, vmf->vma->vm_start); |
3969 | off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); | |
bae473a4 | 3970 | start_pgoff -= off; |
8c6e50b0 KS |
3971 | |
3972 | /* | |
da391d64 WK |
3973 | * end_pgoff is either the end of the page table, the end of |
3974 | * the vma or nr_pages from start_pgoff, depending what is nearest. | |
8c6e50b0 | 3975 | */ |
bae473a4 | 3976 | end_pgoff = start_pgoff - |
82b0f8c3 | 3977 | ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + |
8c6e50b0 | 3978 | PTRS_PER_PTE - 1; |
82b0f8c3 | 3979 | end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1, |
bae473a4 | 3980 | start_pgoff + nr_pages - 1); |
8c6e50b0 | 3981 | |
82b0f8c3 | 3982 | if (pmd_none(*vmf->pmd)) { |
4cf58924 | 3983 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm); |
82b0f8c3 | 3984 | if (!vmf->prealloc_pte) |
c5f88bd2 | 3985 | goto out; |
7267ec00 | 3986 | smp_wmb(); /* See comment in __pte_alloc() */ |
8c6e50b0 KS |
3987 | } |
3988 | ||
82b0f8c3 | 3989 | vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff); |
7267ec00 | 3990 | |
7267ec00 | 3991 | /* Huge page is mapped? Page fault is solved */ |
82b0f8c3 | 3992 | if (pmd_trans_huge(*vmf->pmd)) { |
7267ec00 KS |
3993 | ret = VM_FAULT_NOPAGE; |
3994 | goto out; | |
3995 | } | |
3996 | ||
3997 | /* ->map_pages() haven't done anything useful. Cold page cache? */ | |
82b0f8c3 | 3998 | if (!vmf->pte) |
7267ec00 KS |
3999 | goto out; |
4000 | ||
4001 | /* check if the page fault is solved */ | |
82b0f8c3 JK |
4002 | vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT); |
4003 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 4004 | ret = VM_FAULT_NOPAGE; |
82b0f8c3 | 4005 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bae473a4 | 4006 | out: |
82b0f8c3 JK |
4007 | vmf->address = address; |
4008 | vmf->pte = NULL; | |
7267ec00 | 4009 | return ret; |
8c6e50b0 KS |
4010 | } |
4011 | ||
2b740303 | 4012 | static vm_fault_t do_read_fault(struct vm_fault *vmf) |
e655fb29 | 4013 | { |
82b0f8c3 | 4014 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 4015 | vm_fault_t ret = 0; |
8c6e50b0 KS |
4016 | |
4017 | /* | |
4018 | * Let's call ->map_pages() first and use ->fault() as fallback | |
4019 | * if page by the offset is not ready to be mapped (cold cache or | |
4020 | * something). | |
4021 | */ | |
9b4bdd2f | 4022 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
0721ec8b | 4023 | ret = do_fault_around(vmf); |
7267ec00 KS |
4024 | if (ret) |
4025 | return ret; | |
8c6e50b0 | 4026 | } |
e655fb29 | 4027 | |
936ca80d | 4028 | ret = __do_fault(vmf); |
e655fb29 KS |
4029 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
4030 | return ret; | |
4031 | ||
9118c0cb | 4032 | ret |= finish_fault(vmf); |
936ca80d | 4033 | unlock_page(vmf->page); |
7267ec00 | 4034 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
936ca80d | 4035 | put_page(vmf->page); |
e655fb29 KS |
4036 | return ret; |
4037 | } | |
4038 | ||
2b740303 | 4039 | static vm_fault_t do_cow_fault(struct vm_fault *vmf) |
ec47c3b9 | 4040 | { |
82b0f8c3 | 4041 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 4042 | vm_fault_t ret; |
ec47c3b9 KS |
4043 | |
4044 | if (unlikely(anon_vma_prepare(vma))) | |
4045 | return VM_FAULT_OOM; | |
4046 | ||
936ca80d JK |
4047 | vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); |
4048 | if (!vmf->cow_page) | |
ec47c3b9 KS |
4049 | return VM_FAULT_OOM; |
4050 | ||
d9eb1ea2 | 4051 | if (mem_cgroup_charge(vmf->cow_page, vma->vm_mm, GFP_KERNEL)) { |
936ca80d | 4052 | put_page(vmf->cow_page); |
ec47c3b9 KS |
4053 | return VM_FAULT_OOM; |
4054 | } | |
9d82c694 | 4055 | cgroup_throttle_swaprate(vmf->cow_page, GFP_KERNEL); |
ec47c3b9 | 4056 | |
936ca80d | 4057 | ret = __do_fault(vmf); |
ec47c3b9 KS |
4058 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
4059 | goto uncharge_out; | |
3917048d JK |
4060 | if (ret & VM_FAULT_DONE_COW) |
4061 | return ret; | |
ec47c3b9 | 4062 | |
b1aa812b | 4063 | copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); |
936ca80d | 4064 | __SetPageUptodate(vmf->cow_page); |
ec47c3b9 | 4065 | |
9118c0cb | 4066 | ret |= finish_fault(vmf); |
b1aa812b JK |
4067 | unlock_page(vmf->page); |
4068 | put_page(vmf->page); | |
7267ec00 KS |
4069 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
4070 | goto uncharge_out; | |
ec47c3b9 KS |
4071 | return ret; |
4072 | uncharge_out: | |
936ca80d | 4073 | put_page(vmf->cow_page); |
ec47c3b9 KS |
4074 | return ret; |
4075 | } | |
4076 | ||
2b740303 | 4077 | static vm_fault_t do_shared_fault(struct vm_fault *vmf) |
1da177e4 | 4078 | { |
82b0f8c3 | 4079 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 4080 | vm_fault_t ret, tmp; |
1d65f86d | 4081 | |
936ca80d | 4082 | ret = __do_fault(vmf); |
7eae74af | 4083 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 4084 | return ret; |
1da177e4 LT |
4085 | |
4086 | /* | |
f0c6d4d2 KS |
4087 | * Check if the backing address space wants to know that the page is |
4088 | * about to become writable | |
1da177e4 | 4089 | */ |
fb09a464 | 4090 | if (vma->vm_ops->page_mkwrite) { |
936ca80d | 4091 | unlock_page(vmf->page); |
38b8cb7f | 4092 | tmp = do_page_mkwrite(vmf); |
fb09a464 KS |
4093 | if (unlikely(!tmp || |
4094 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
936ca80d | 4095 | put_page(vmf->page); |
fb09a464 | 4096 | return tmp; |
4294621f | 4097 | } |
fb09a464 KS |
4098 | } |
4099 | ||
9118c0cb | 4100 | ret |= finish_fault(vmf); |
7267ec00 KS |
4101 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
4102 | VM_FAULT_RETRY))) { | |
936ca80d JK |
4103 | unlock_page(vmf->page); |
4104 | put_page(vmf->page); | |
f0c6d4d2 | 4105 | return ret; |
1da177e4 | 4106 | } |
b827e496 | 4107 | |
89b15332 | 4108 | ret |= fault_dirty_shared_page(vmf); |
1d65f86d | 4109 | return ret; |
54cb8821 | 4110 | } |
d00806b1 | 4111 | |
9a95f3cf | 4112 | /* |
c1e8d7c6 | 4113 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
9a95f3cf | 4114 | * but allow concurrent faults). |
c1e8d7c6 | 4115 | * The mmap_lock may have been released depending on flags and our |
9a95f3cf | 4116 | * return value. See filemap_fault() and __lock_page_or_retry(). |
c1e8d7c6 | 4117 | * If mmap_lock is released, vma may become invalid (for example |
fc8efd2d | 4118 | * by other thread calling munmap()). |
9a95f3cf | 4119 | */ |
2b740303 | 4120 | static vm_fault_t do_fault(struct vm_fault *vmf) |
54cb8821 | 4121 | { |
82b0f8c3 | 4122 | struct vm_area_struct *vma = vmf->vma; |
fc8efd2d | 4123 | struct mm_struct *vm_mm = vma->vm_mm; |
2b740303 | 4124 | vm_fault_t ret; |
54cb8821 | 4125 | |
ff09d7ec AK |
4126 | /* |
4127 | * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND | |
4128 | */ | |
4129 | if (!vma->vm_ops->fault) { | |
4130 | /* | |
4131 | * If we find a migration pmd entry or a none pmd entry, which | |
4132 | * should never happen, return SIGBUS | |
4133 | */ | |
4134 | if (unlikely(!pmd_present(*vmf->pmd))) | |
4135 | ret = VM_FAULT_SIGBUS; | |
4136 | else { | |
4137 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, | |
4138 | vmf->pmd, | |
4139 | vmf->address, | |
4140 | &vmf->ptl); | |
4141 | /* | |
4142 | * Make sure this is not a temporary clearing of pte | |
4143 | * by holding ptl and checking again. A R/M/W update | |
4144 | * of pte involves: take ptl, clearing the pte so that | |
4145 | * we don't have concurrent modification by hardware | |
4146 | * followed by an update. | |
4147 | */ | |
4148 | if (unlikely(pte_none(*vmf->pte))) | |
4149 | ret = VM_FAULT_SIGBUS; | |
4150 | else | |
4151 | ret = VM_FAULT_NOPAGE; | |
4152 | ||
4153 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4154 | } | |
4155 | } else if (!(vmf->flags & FAULT_FLAG_WRITE)) | |
b0b9b3df HD |
4156 | ret = do_read_fault(vmf); |
4157 | else if (!(vma->vm_flags & VM_SHARED)) | |
4158 | ret = do_cow_fault(vmf); | |
4159 | else | |
4160 | ret = do_shared_fault(vmf); | |
4161 | ||
4162 | /* preallocated pagetable is unused: free it */ | |
4163 | if (vmf->prealloc_pte) { | |
fc8efd2d | 4164 | pte_free(vm_mm, vmf->prealloc_pte); |
7f2b6ce8 | 4165 | vmf->prealloc_pte = NULL; |
b0b9b3df HD |
4166 | } |
4167 | return ret; | |
54cb8821 NP |
4168 | } |
4169 | ||
b19a9939 | 4170 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
4171 | unsigned long addr, int page_nid, |
4172 | int *flags) | |
9532fec1 MG |
4173 | { |
4174 | get_page(page); | |
4175 | ||
4176 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 4177 | if (page_nid == numa_node_id()) { |
9532fec1 | 4178 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
4179 | *flags |= TNF_FAULT_LOCAL; |
4180 | } | |
9532fec1 MG |
4181 | |
4182 | return mpol_misplaced(page, vma, addr); | |
4183 | } | |
4184 | ||
2b740303 | 4185 | static vm_fault_t do_numa_page(struct vm_fault *vmf) |
d10e63f2 | 4186 | { |
82b0f8c3 | 4187 | struct vm_area_struct *vma = vmf->vma; |
4daae3b4 | 4188 | struct page *page = NULL; |
98fa15f3 | 4189 | int page_nid = NUMA_NO_NODE; |
90572890 | 4190 | int last_cpupid; |
cbee9f88 | 4191 | int target_nid; |
b8593bfd | 4192 | bool migrated = false; |
04a86453 | 4193 | pte_t pte, old_pte; |
288bc549 | 4194 | bool was_writable = pte_savedwrite(vmf->orig_pte); |
6688cc05 | 4195 | int flags = 0; |
d10e63f2 MG |
4196 | |
4197 | /* | |
166f61b9 TH |
4198 | * The "pte" at this point cannot be used safely without |
4199 | * validation through pte_unmap_same(). It's of NUMA type but | |
4200 | * the pfn may be screwed if the read is non atomic. | |
166f61b9 | 4201 | */ |
82b0f8c3 JK |
4202 | vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); |
4203 | spin_lock(vmf->ptl); | |
cee216a6 | 4204 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { |
82b0f8c3 | 4205 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
4206 | goto out; |
4207 | } | |
4208 | ||
cee216a6 AK |
4209 | /* |
4210 | * Make it present again, Depending on how arch implementes non | |
4211 | * accessible ptes, some can allow access by kernel mode. | |
4212 | */ | |
04a86453 AK |
4213 | old_pte = ptep_modify_prot_start(vma, vmf->address, vmf->pte); |
4214 | pte = pte_modify(old_pte, vma->vm_page_prot); | |
4d942466 | 4215 | pte = pte_mkyoung(pte); |
b191f9b1 MG |
4216 | if (was_writable) |
4217 | pte = pte_mkwrite(pte); | |
04a86453 | 4218 | ptep_modify_prot_commit(vma, vmf->address, vmf->pte, old_pte, pte); |
82b0f8c3 | 4219 | update_mmu_cache(vma, vmf->address, vmf->pte); |
d10e63f2 | 4220 | |
82b0f8c3 | 4221 | page = vm_normal_page(vma, vmf->address, pte); |
d10e63f2 | 4222 | if (!page) { |
82b0f8c3 | 4223 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
d10e63f2 MG |
4224 | return 0; |
4225 | } | |
4226 | ||
e81c4802 KS |
4227 | /* TODO: handle PTE-mapped THP */ |
4228 | if (PageCompound(page)) { | |
82b0f8c3 | 4229 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
e81c4802 KS |
4230 | return 0; |
4231 | } | |
4232 | ||
6688cc05 | 4233 | /* |
bea66fbd MG |
4234 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
4235 | * much anyway since they can be in shared cache state. This misses | |
4236 | * the case where a mapping is writable but the process never writes | |
4237 | * to it but pte_write gets cleared during protection updates and | |
4238 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
4239 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 4240 | */ |
d59dc7bc | 4241 | if (!pte_write(pte)) |
6688cc05 PZ |
4242 | flags |= TNF_NO_GROUP; |
4243 | ||
dabe1d99 RR |
4244 | /* |
4245 | * Flag if the page is shared between multiple address spaces. This | |
4246 | * is later used when determining whether to group tasks together | |
4247 | */ | |
4248 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
4249 | flags |= TNF_SHARED; | |
4250 | ||
90572890 | 4251 | last_cpupid = page_cpupid_last(page); |
8191acbd | 4252 | page_nid = page_to_nid(page); |
82b0f8c3 | 4253 | target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, |
bae473a4 | 4254 | &flags); |
82b0f8c3 | 4255 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
98fa15f3 | 4256 | if (target_nid == NUMA_NO_NODE) { |
4daae3b4 MG |
4257 | put_page(page); |
4258 | goto out; | |
4259 | } | |
4260 | ||
4261 | /* Migrate to the requested node */ | |
1bc115d8 | 4262 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 4263 | if (migrated) { |
8191acbd | 4264 | page_nid = target_nid; |
6688cc05 | 4265 | flags |= TNF_MIGRATED; |
074c2381 MG |
4266 | } else |
4267 | flags |= TNF_MIGRATE_FAIL; | |
4daae3b4 MG |
4268 | |
4269 | out: | |
98fa15f3 | 4270 | if (page_nid != NUMA_NO_NODE) |
6688cc05 | 4271 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
4272 | return 0; |
4273 | } | |
4274 | ||
2b740303 | 4275 | static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 4276 | { |
f4200391 | 4277 | if (vma_is_anonymous(vmf->vma)) |
82b0f8c3 | 4278 | return do_huge_pmd_anonymous_page(vmf); |
a2d58167 | 4279 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 4280 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
b96375f7 MW |
4281 | return VM_FAULT_FALLBACK; |
4282 | } | |
4283 | ||
183f24aa | 4284 | /* `inline' is required to avoid gcc 4.1.2 build error */ |
2b740303 | 4285 | static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd) |
b96375f7 | 4286 | { |
529b930b | 4287 | if (vma_is_anonymous(vmf->vma)) { |
292924b2 | 4288 | if (userfaultfd_huge_pmd_wp(vmf->vma, orig_pmd)) |
529b930b | 4289 | return handle_userfault(vmf, VM_UFFD_WP); |
82b0f8c3 | 4290 | return do_huge_pmd_wp_page(vmf, orig_pmd); |
529b930b | 4291 | } |
327e9fd4 THV |
4292 | if (vmf->vma->vm_ops->huge_fault) { |
4293 | vm_fault_t ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); | |
4294 | ||
4295 | if (!(ret & VM_FAULT_FALLBACK)) | |
4296 | return ret; | |
4297 | } | |
af9e4d5f | 4298 | |
327e9fd4 | 4299 | /* COW or write-notify handled on pte level: split pmd. */ |
82b0f8c3 | 4300 | __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); |
af9e4d5f | 4301 | |
b96375f7 MW |
4302 | return VM_FAULT_FALLBACK; |
4303 | } | |
4304 | ||
2b740303 | 4305 | static vm_fault_t create_huge_pud(struct vm_fault *vmf) |
a00cc7d9 | 4306 | { |
327e9fd4 THV |
4307 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ |
4308 | defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) | |
a00cc7d9 MW |
4309 | /* No support for anonymous transparent PUD pages yet */ |
4310 | if (vma_is_anonymous(vmf->vma)) | |
327e9fd4 THV |
4311 | goto split; |
4312 | if (vmf->vma->vm_ops->huge_fault) { | |
4313 | vm_fault_t ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); | |
4314 | ||
4315 | if (!(ret & VM_FAULT_FALLBACK)) | |
4316 | return ret; | |
4317 | } | |
4318 | split: | |
4319 | /* COW or write-notify not handled on PUD level: split pud.*/ | |
4320 | __split_huge_pud(vmf->vma, vmf->pud, vmf->address); | |
a00cc7d9 MW |
4321 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
4322 | return VM_FAULT_FALLBACK; | |
4323 | } | |
4324 | ||
2b740303 | 4325 | static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) |
a00cc7d9 MW |
4326 | { |
4327 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4328 | /* No support for anonymous transparent PUD pages yet */ | |
4329 | if (vma_is_anonymous(vmf->vma)) | |
4330 | return VM_FAULT_FALLBACK; | |
4331 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 4332 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
4333 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
4334 | return VM_FAULT_FALLBACK; | |
4335 | } | |
4336 | ||
1da177e4 LT |
4337 | /* |
4338 | * These routines also need to handle stuff like marking pages dirty | |
4339 | * and/or accessed for architectures that don't do it in hardware (most | |
4340 | * RISC architectures). The early dirtying is also good on the i386. | |
4341 | * | |
4342 | * There is also a hook called "update_mmu_cache()" that architectures | |
4343 | * with external mmu caches can use to update those (ie the Sparc or | |
4344 | * PowerPC hashed page tables that act as extended TLBs). | |
4345 | * | |
c1e8d7c6 | 4346 | * We enter with non-exclusive mmap_lock (to exclude vma changes, but allow |
7267ec00 | 4347 | * concurrent faults). |
9a95f3cf | 4348 | * |
c1e8d7c6 | 4349 | * The mmap_lock may have been released depending on flags and our return value. |
7267ec00 | 4350 | * See filemap_fault() and __lock_page_or_retry(). |
1da177e4 | 4351 | */ |
2b740303 | 4352 | static vm_fault_t handle_pte_fault(struct vm_fault *vmf) |
1da177e4 LT |
4353 | { |
4354 | pte_t entry; | |
4355 | ||
82b0f8c3 | 4356 | if (unlikely(pmd_none(*vmf->pmd))) { |
7267ec00 KS |
4357 | /* |
4358 | * Leave __pte_alloc() until later: because vm_ops->fault may | |
4359 | * want to allocate huge page, and if we expose page table | |
4360 | * for an instant, it will be difficult to retract from | |
4361 | * concurrent faults and from rmap lookups. | |
4362 | */ | |
82b0f8c3 | 4363 | vmf->pte = NULL; |
7267ec00 KS |
4364 | } else { |
4365 | /* See comment in pte_alloc_one_map() */ | |
d0f0931d | 4366 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
4367 | return 0; |
4368 | /* | |
4369 | * A regular pmd is established and it can't morph into a huge | |
4370 | * pmd from under us anymore at this point because we hold the | |
c1e8d7c6 | 4371 | * mmap_lock read mode and khugepaged takes it in write mode. |
7267ec00 KS |
4372 | * So now it's safe to run pte_offset_map(). |
4373 | */ | |
82b0f8c3 | 4374 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
2994302b | 4375 | vmf->orig_pte = *vmf->pte; |
7267ec00 KS |
4376 | |
4377 | /* | |
4378 | * some architectures can have larger ptes than wordsize, | |
4379 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and | |
b03a0fe0 PM |
4380 | * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic |
4381 | * accesses. The code below just needs a consistent view | |
4382 | * for the ifs and we later double check anyway with the | |
7267ec00 KS |
4383 | * ptl lock held. So here a barrier will do. |
4384 | */ | |
4385 | barrier(); | |
2994302b | 4386 | if (pte_none(vmf->orig_pte)) { |
82b0f8c3 JK |
4387 | pte_unmap(vmf->pte); |
4388 | vmf->pte = NULL; | |
65500d23 | 4389 | } |
1da177e4 LT |
4390 | } |
4391 | ||
82b0f8c3 JK |
4392 | if (!vmf->pte) { |
4393 | if (vma_is_anonymous(vmf->vma)) | |
4394 | return do_anonymous_page(vmf); | |
7267ec00 | 4395 | else |
82b0f8c3 | 4396 | return do_fault(vmf); |
7267ec00 KS |
4397 | } |
4398 | ||
2994302b JK |
4399 | if (!pte_present(vmf->orig_pte)) |
4400 | return do_swap_page(vmf); | |
7267ec00 | 4401 | |
2994302b JK |
4402 | if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) |
4403 | return do_numa_page(vmf); | |
d10e63f2 | 4404 | |
82b0f8c3 JK |
4405 | vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); |
4406 | spin_lock(vmf->ptl); | |
2994302b | 4407 | entry = vmf->orig_pte; |
7df67697 BM |
4408 | if (unlikely(!pte_same(*vmf->pte, entry))) { |
4409 | update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); | |
8f4e2101 | 4410 | goto unlock; |
7df67697 | 4411 | } |
82b0f8c3 | 4412 | if (vmf->flags & FAULT_FLAG_WRITE) { |
f6f37321 | 4413 | if (!pte_write(entry)) |
2994302b | 4414 | return do_wp_page(vmf); |
1da177e4 LT |
4415 | entry = pte_mkdirty(entry); |
4416 | } | |
4417 | entry = pte_mkyoung(entry); | |
82b0f8c3 JK |
4418 | if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, |
4419 | vmf->flags & FAULT_FLAG_WRITE)) { | |
4420 | update_mmu_cache(vmf->vma, vmf->address, vmf->pte); | |
1a44e149 | 4421 | } else { |
b7333b58 YS |
4422 | /* Skip spurious TLB flush for retried page fault */ |
4423 | if (vmf->flags & FAULT_FLAG_TRIED) | |
4424 | goto unlock; | |
1a44e149 AA |
4425 | /* |
4426 | * This is needed only for protection faults but the arch code | |
4427 | * is not yet telling us if this is a protection fault or not. | |
4428 | * This still avoids useless tlb flushes for .text page faults | |
4429 | * with threads. | |
4430 | */ | |
82b0f8c3 JK |
4431 | if (vmf->flags & FAULT_FLAG_WRITE) |
4432 | flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); | |
1a44e149 | 4433 | } |
8f4e2101 | 4434 | unlock: |
82b0f8c3 | 4435 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 4436 | return 0; |
1da177e4 LT |
4437 | } |
4438 | ||
4439 | /* | |
4440 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf | 4441 | * |
c1e8d7c6 | 4442 | * The mmap_lock may have been released depending on flags and our |
9a95f3cf | 4443 | * return value. See filemap_fault() and __lock_page_or_retry(). |
1da177e4 | 4444 | */ |
2b740303 SJ |
4445 | static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma, |
4446 | unsigned long address, unsigned int flags) | |
1da177e4 | 4447 | { |
82b0f8c3 | 4448 | struct vm_fault vmf = { |
bae473a4 | 4449 | .vma = vma, |
1a29d85e | 4450 | .address = address & PAGE_MASK, |
bae473a4 | 4451 | .flags = flags, |
0721ec8b | 4452 | .pgoff = linear_page_index(vma, address), |
667240e0 | 4453 | .gfp_mask = __get_fault_gfp_mask(vma), |
bae473a4 | 4454 | }; |
fde26bed | 4455 | unsigned int dirty = flags & FAULT_FLAG_WRITE; |
dcddffd4 | 4456 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 4457 | pgd_t *pgd; |
c2febafc | 4458 | p4d_t *p4d; |
2b740303 | 4459 | vm_fault_t ret; |
1da177e4 | 4460 | |
1da177e4 | 4461 | pgd = pgd_offset(mm, address); |
c2febafc KS |
4462 | p4d = p4d_alloc(mm, pgd, address); |
4463 | if (!p4d) | |
4464 | return VM_FAULT_OOM; | |
a00cc7d9 | 4465 | |
c2febafc | 4466 | vmf.pud = pud_alloc(mm, p4d, address); |
a00cc7d9 | 4467 | if (!vmf.pud) |
c74df32c | 4468 | return VM_FAULT_OOM; |
625110b5 | 4469 | retry_pud: |
7635d9cb | 4470 | if (pud_none(*vmf.pud) && __transparent_hugepage_enabled(vma)) { |
a00cc7d9 MW |
4471 | ret = create_huge_pud(&vmf); |
4472 | if (!(ret & VM_FAULT_FALLBACK)) | |
4473 | return ret; | |
4474 | } else { | |
4475 | pud_t orig_pud = *vmf.pud; | |
4476 | ||
4477 | barrier(); | |
4478 | if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { | |
a00cc7d9 | 4479 | |
a00cc7d9 MW |
4480 | /* NUMA case for anonymous PUDs would go here */ |
4481 | ||
f6f37321 | 4482 | if (dirty && !pud_write(orig_pud)) { |
a00cc7d9 MW |
4483 | ret = wp_huge_pud(&vmf, orig_pud); |
4484 | if (!(ret & VM_FAULT_FALLBACK)) | |
4485 | return ret; | |
4486 | } else { | |
4487 | huge_pud_set_accessed(&vmf, orig_pud); | |
4488 | return 0; | |
4489 | } | |
4490 | } | |
4491 | } | |
4492 | ||
4493 | vmf.pmd = pmd_alloc(mm, vmf.pud, address); | |
82b0f8c3 | 4494 | if (!vmf.pmd) |
c74df32c | 4495 | return VM_FAULT_OOM; |
625110b5 TH |
4496 | |
4497 | /* Huge pud page fault raced with pmd_alloc? */ | |
4498 | if (pud_trans_unstable(vmf.pud)) | |
4499 | goto retry_pud; | |
4500 | ||
7635d9cb | 4501 | if (pmd_none(*vmf.pmd) && __transparent_hugepage_enabled(vma)) { |
a2d58167 | 4502 | ret = create_huge_pmd(&vmf); |
c0292554 KS |
4503 | if (!(ret & VM_FAULT_FALLBACK)) |
4504 | return ret; | |
71e3aac0 | 4505 | } else { |
82b0f8c3 | 4506 | pmd_t orig_pmd = *vmf.pmd; |
1f1d06c3 | 4507 | |
71e3aac0 | 4508 | barrier(); |
84c3fc4e ZY |
4509 | if (unlikely(is_swap_pmd(orig_pmd))) { |
4510 | VM_BUG_ON(thp_migration_supported() && | |
4511 | !is_pmd_migration_entry(orig_pmd)); | |
4512 | if (is_pmd_migration_entry(orig_pmd)) | |
4513 | pmd_migration_entry_wait(mm, vmf.pmd); | |
4514 | return 0; | |
4515 | } | |
5c7fb56e | 4516 | if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { |
38e08854 | 4517 | if (pmd_protnone(orig_pmd) && vma_is_accessible(vma)) |
82b0f8c3 | 4518 | return do_huge_pmd_numa_page(&vmf, orig_pmd); |
d10e63f2 | 4519 | |
f6f37321 | 4520 | if (dirty && !pmd_write(orig_pmd)) { |
82b0f8c3 | 4521 | ret = wp_huge_pmd(&vmf, orig_pmd); |
9845cbbd KS |
4522 | if (!(ret & VM_FAULT_FALLBACK)) |
4523 | return ret; | |
a1dd450b | 4524 | } else { |
82b0f8c3 | 4525 | huge_pmd_set_accessed(&vmf, orig_pmd); |
9845cbbd | 4526 | return 0; |
1f1d06c3 | 4527 | } |
71e3aac0 AA |
4528 | } |
4529 | } | |
4530 | ||
82b0f8c3 | 4531 | return handle_pte_fault(&vmf); |
1da177e4 LT |
4532 | } |
4533 | ||
bce617ed PX |
4534 | /** |
4535 | * mm_account_fault - Do page fault accountings | |
4536 | * | |
4537 | * @regs: the pt_regs struct pointer. When set to NULL, will skip accounting | |
4538 | * of perf event counters, but we'll still do the per-task accounting to | |
4539 | * the task who triggered this page fault. | |
4540 | * @address: the faulted address. | |
4541 | * @flags: the fault flags. | |
4542 | * @ret: the fault retcode. | |
4543 | * | |
4544 | * This will take care of most of the page fault accountings. Meanwhile, it | |
4545 | * will also include the PERF_COUNT_SW_PAGE_FAULTS_[MAJ|MIN] perf counter | |
4546 | * updates. However note that the handling of PERF_COUNT_SW_PAGE_FAULTS should | |
4547 | * still be in per-arch page fault handlers at the entry of page fault. | |
4548 | */ | |
4549 | static inline void mm_account_fault(struct pt_regs *regs, | |
4550 | unsigned long address, unsigned int flags, | |
4551 | vm_fault_t ret) | |
4552 | { | |
4553 | bool major; | |
4554 | ||
4555 | /* | |
4556 | * We don't do accounting for some specific faults: | |
4557 | * | |
4558 | * - Unsuccessful faults (e.g. when the address wasn't valid). That | |
4559 | * includes arch_vma_access_permitted() failing before reaching here. | |
4560 | * So this is not a "this many hardware page faults" counter. We | |
4561 | * should use the hw profiling for that. | |
4562 | * | |
4563 | * - Incomplete faults (VM_FAULT_RETRY). They will only be counted | |
4564 | * once they're completed. | |
4565 | */ | |
4566 | if (ret & (VM_FAULT_ERROR | VM_FAULT_RETRY)) | |
4567 | return; | |
4568 | ||
4569 | /* | |
4570 | * We define the fault as a major fault when the final successful fault | |
4571 | * is VM_FAULT_MAJOR, or if it retried (which implies that we couldn't | |
4572 | * handle it immediately previously). | |
4573 | */ | |
4574 | major = (ret & VM_FAULT_MAJOR) || (flags & FAULT_FLAG_TRIED); | |
4575 | ||
a2beb5f1 PX |
4576 | if (major) |
4577 | current->maj_flt++; | |
4578 | else | |
4579 | current->min_flt++; | |
4580 | ||
bce617ed | 4581 | /* |
a2beb5f1 PX |
4582 | * If the fault is done for GUP, regs will be NULL. We only do the |
4583 | * accounting for the per thread fault counters who triggered the | |
4584 | * fault, and we skip the perf event updates. | |
bce617ed PX |
4585 | */ |
4586 | if (!regs) | |
4587 | return; | |
4588 | ||
a2beb5f1 | 4589 | if (major) |
bce617ed | 4590 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); |
a2beb5f1 | 4591 | else |
bce617ed | 4592 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); |
bce617ed PX |
4593 | } |
4594 | ||
9a95f3cf PC |
4595 | /* |
4596 | * By the time we get here, we already hold the mm semaphore | |
4597 | * | |
c1e8d7c6 | 4598 | * The mmap_lock may have been released depending on flags and our |
9a95f3cf PC |
4599 | * return value. See filemap_fault() and __lock_page_or_retry(). |
4600 | */ | |
2b740303 | 4601 | vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
bce617ed | 4602 | unsigned int flags, struct pt_regs *regs) |
519e5247 | 4603 | { |
2b740303 | 4604 | vm_fault_t ret; |
519e5247 JW |
4605 | |
4606 | __set_current_state(TASK_RUNNING); | |
4607 | ||
4608 | count_vm_event(PGFAULT); | |
2262185c | 4609 | count_memcg_event_mm(vma->vm_mm, PGFAULT); |
519e5247 JW |
4610 | |
4611 | /* do counter updates before entering really critical section. */ | |
4612 | check_sync_rss_stat(current); | |
4613 | ||
de0c799b LD |
4614 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
4615 | flags & FAULT_FLAG_INSTRUCTION, | |
4616 | flags & FAULT_FLAG_REMOTE)) | |
4617 | return VM_FAULT_SIGSEGV; | |
4618 | ||
519e5247 JW |
4619 | /* |
4620 | * Enable the memcg OOM handling for faults triggered in user | |
4621 | * space. Kernel faults are handled more gracefully. | |
4622 | */ | |
4623 | if (flags & FAULT_FLAG_USER) | |
29ef680a | 4624 | mem_cgroup_enter_user_fault(); |
519e5247 | 4625 | |
bae473a4 KS |
4626 | if (unlikely(is_vm_hugetlb_page(vma))) |
4627 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); | |
4628 | else | |
4629 | ret = __handle_mm_fault(vma, address, flags); | |
519e5247 | 4630 | |
49426420 | 4631 | if (flags & FAULT_FLAG_USER) { |
29ef680a | 4632 | mem_cgroup_exit_user_fault(); |
166f61b9 TH |
4633 | /* |
4634 | * The task may have entered a memcg OOM situation but | |
4635 | * if the allocation error was handled gracefully (no | |
4636 | * VM_FAULT_OOM), there is no need to kill anything. | |
4637 | * Just clean up the OOM state peacefully. | |
4638 | */ | |
4639 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
4640 | mem_cgroup_oom_synchronize(false); | |
49426420 | 4641 | } |
3812c8c8 | 4642 | |
bce617ed PX |
4643 | mm_account_fault(regs, address, flags, ret); |
4644 | ||
519e5247 JW |
4645 | return ret; |
4646 | } | |
e1d6d01a | 4647 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 4648 | |
90eceff1 KS |
4649 | #ifndef __PAGETABLE_P4D_FOLDED |
4650 | /* | |
4651 | * Allocate p4d page table. | |
4652 | * We've already handled the fast-path in-line. | |
4653 | */ | |
4654 | int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
4655 | { | |
4656 | p4d_t *new = p4d_alloc_one(mm, address); | |
4657 | if (!new) | |
4658 | return -ENOMEM; | |
4659 | ||
4660 | smp_wmb(); /* See comment in __pte_alloc */ | |
4661 | ||
4662 | spin_lock(&mm->page_table_lock); | |
4663 | if (pgd_present(*pgd)) /* Another has populated it */ | |
4664 | p4d_free(mm, new); | |
4665 | else | |
4666 | pgd_populate(mm, pgd, new); | |
4667 | spin_unlock(&mm->page_table_lock); | |
4668 | return 0; | |
4669 | } | |
4670 | #endif /* __PAGETABLE_P4D_FOLDED */ | |
4671 | ||
1da177e4 LT |
4672 | #ifndef __PAGETABLE_PUD_FOLDED |
4673 | /* | |
4674 | * Allocate page upper directory. | |
872fec16 | 4675 | * We've already handled the fast-path in-line. |
1da177e4 | 4676 | */ |
c2febafc | 4677 | int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) |
1da177e4 | 4678 | { |
c74df32c HD |
4679 | pud_t *new = pud_alloc_one(mm, address); |
4680 | if (!new) | |
1bb3630e | 4681 | return -ENOMEM; |
1da177e4 | 4682 | |
362a61ad NP |
4683 | smp_wmb(); /* See comment in __pte_alloc */ |
4684 | ||
872fec16 | 4685 | spin_lock(&mm->page_table_lock); |
b4e98d9a KS |
4686 | if (!p4d_present(*p4d)) { |
4687 | mm_inc_nr_puds(mm); | |
c2febafc | 4688 | p4d_populate(mm, p4d, new); |
b4e98d9a | 4689 | } else /* Another has populated it */ |
5e541973 | 4690 | pud_free(mm, new); |
c74df32c | 4691 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 4692 | return 0; |
1da177e4 LT |
4693 | } |
4694 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
4695 | ||
4696 | #ifndef __PAGETABLE_PMD_FOLDED | |
4697 | /* | |
4698 | * Allocate page middle directory. | |
872fec16 | 4699 | * We've already handled the fast-path in-line. |
1da177e4 | 4700 | */ |
1bb3630e | 4701 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 4702 | { |
a00cc7d9 | 4703 | spinlock_t *ptl; |
c74df32c HD |
4704 | pmd_t *new = pmd_alloc_one(mm, address); |
4705 | if (!new) | |
1bb3630e | 4706 | return -ENOMEM; |
1da177e4 | 4707 | |
362a61ad NP |
4708 | smp_wmb(); /* See comment in __pte_alloc */ |
4709 | ||
a00cc7d9 | 4710 | ptl = pud_lock(mm, pud); |
dc6c9a35 KS |
4711 | if (!pud_present(*pud)) { |
4712 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4713 | pud_populate(mm, pud, new); |
dc6c9a35 | 4714 | } else /* Another has populated it */ |
5e541973 | 4715 | pmd_free(mm, new); |
a00cc7d9 | 4716 | spin_unlock(ptl); |
1bb3630e | 4717 | return 0; |
e0f39591 | 4718 | } |
1da177e4 LT |
4719 | #endif /* __PAGETABLE_PMD_FOLDED */ |
4720 | ||
992e4554 PB |
4721 | int follow_invalidate_pte(struct mm_struct *mm, unsigned long address, |
4722 | struct mmu_notifier_range *range, pte_t **ptepp, | |
4723 | pmd_t **pmdpp, spinlock_t **ptlp) | |
f8ad0f49 JW |
4724 | { |
4725 | pgd_t *pgd; | |
c2febafc | 4726 | p4d_t *p4d; |
f8ad0f49 JW |
4727 | pud_t *pud; |
4728 | pmd_t *pmd; | |
4729 | pte_t *ptep; | |
4730 | ||
4731 | pgd = pgd_offset(mm, address); | |
4732 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
4733 | goto out; | |
4734 | ||
c2febafc KS |
4735 | p4d = p4d_offset(pgd, address); |
4736 | if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) | |
4737 | goto out; | |
4738 | ||
4739 | pud = pud_offset(p4d, address); | |
f8ad0f49 JW |
4740 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
4741 | goto out; | |
4742 | ||
4743 | pmd = pmd_offset(pud, address); | |
f66055ab | 4744 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 | 4745 | |
09796395 RZ |
4746 | if (pmd_huge(*pmd)) { |
4747 | if (!pmdpp) | |
4748 | goto out; | |
4749 | ||
ac46d4f3 | 4750 | if (range) { |
7269f999 | 4751 | mmu_notifier_range_init(range, MMU_NOTIFY_CLEAR, 0, |
6f4f13e8 JG |
4752 | NULL, mm, address & PMD_MASK, |
4753 | (address & PMD_MASK) + PMD_SIZE); | |
ac46d4f3 | 4754 | mmu_notifier_invalidate_range_start(range); |
a4d1a885 | 4755 | } |
09796395 RZ |
4756 | *ptlp = pmd_lock(mm, pmd); |
4757 | if (pmd_huge(*pmd)) { | |
4758 | *pmdpp = pmd; | |
4759 | return 0; | |
4760 | } | |
4761 | spin_unlock(*ptlp); | |
ac46d4f3 JG |
4762 | if (range) |
4763 | mmu_notifier_invalidate_range_end(range); | |
09796395 RZ |
4764 | } |
4765 | ||
4766 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
f8ad0f49 JW |
4767 | goto out; |
4768 | ||
ac46d4f3 | 4769 | if (range) { |
7269f999 | 4770 | mmu_notifier_range_init(range, MMU_NOTIFY_CLEAR, 0, NULL, mm, |
6f4f13e8 JG |
4771 | address & PAGE_MASK, |
4772 | (address & PAGE_MASK) + PAGE_SIZE); | |
ac46d4f3 | 4773 | mmu_notifier_invalidate_range_start(range); |
a4d1a885 | 4774 | } |
f8ad0f49 | 4775 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
f8ad0f49 JW |
4776 | if (!pte_present(*ptep)) |
4777 | goto unlock; | |
4778 | *ptepp = ptep; | |
4779 | return 0; | |
4780 | unlock: | |
4781 | pte_unmap_unlock(ptep, *ptlp); | |
ac46d4f3 JG |
4782 | if (range) |
4783 | mmu_notifier_invalidate_range_end(range); | |
f8ad0f49 JW |
4784 | out: |
4785 | return -EINVAL; | |
4786 | } | |
4787 | ||
992e4554 PB |
4788 | /** |
4789 | * follow_pte - look up PTE at a user virtual address | |
4790 | * @mm: the mm_struct of the target address space | |
4791 | * @address: user virtual address | |
4792 | * @ptepp: location to store found PTE | |
4793 | * @ptlp: location to store the lock for the PTE | |
4794 | * | |
4795 | * On a successful return, the pointer to the PTE is stored in @ptepp; | |
4796 | * the corresponding lock is taken and its location is stored in @ptlp. | |
4797 | * The contents of the PTE are only stable until @ptlp is released; | |
4798 | * any further use, if any, must be protected against invalidation | |
4799 | * with MMU notifiers. | |
4800 | * | |
4801 | * Only IO mappings and raw PFN mappings are allowed. The mmap semaphore | |
4802 | * should be taken for read. | |
4803 | * | |
4804 | * KVM uses this function. While it is arguably less bad than ``follow_pfn``, | |
4805 | * it is not a good general-purpose API. | |
4806 | * | |
4807 | * Return: zero on success, -ve otherwise. | |
4808 | */ | |
4809 | int follow_pte(struct mm_struct *mm, unsigned long address, | |
4810 | pte_t **ptepp, spinlock_t **ptlp) | |
4811 | { | |
4812 | return follow_invalidate_pte(mm, address, NULL, ptepp, NULL, ptlp); | |
4813 | } | |
4814 | EXPORT_SYMBOL_GPL(follow_pte); | |
4815 | ||
3b6748e2 JW |
4816 | /** |
4817 | * follow_pfn - look up PFN at a user virtual address | |
4818 | * @vma: memory mapping | |
4819 | * @address: user virtual address | |
4820 | * @pfn: location to store found PFN | |
4821 | * | |
4822 | * Only IO mappings and raw PFN mappings are allowed. | |
4823 | * | |
992e4554 PB |
4824 | * This function does not allow the caller to read the permissions |
4825 | * of the PTE. Do not use it. | |
4826 | * | |
a862f68a | 4827 | * Return: zero and the pfn at @pfn on success, -ve otherwise. |
3b6748e2 JW |
4828 | */ |
4829 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
4830 | unsigned long *pfn) | |
4831 | { | |
4832 | int ret = -EINVAL; | |
4833 | spinlock_t *ptl; | |
4834 | pte_t *ptep; | |
4835 | ||
4836 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
4837 | return ret; | |
4838 | ||
992e4554 | 4839 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); |
3b6748e2 JW |
4840 | if (ret) |
4841 | return ret; | |
4842 | *pfn = pte_pfn(*ptep); | |
4843 | pte_unmap_unlock(ptep, ptl); | |
4844 | return 0; | |
4845 | } | |
4846 | EXPORT_SYMBOL(follow_pfn); | |
4847 | ||
28b2ee20 | 4848 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 4849 | int follow_phys(struct vm_area_struct *vma, |
4850 | unsigned long address, unsigned int flags, | |
4851 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 4852 | { |
03668a4d | 4853 | int ret = -EINVAL; |
28b2ee20 RR |
4854 | pte_t *ptep, pte; |
4855 | spinlock_t *ptl; | |
28b2ee20 | 4856 | |
d87fe660 | 4857 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
4858 | goto out; | |
28b2ee20 | 4859 | |
992e4554 | 4860 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 4861 | goto out; |
28b2ee20 | 4862 | pte = *ptep; |
03668a4d | 4863 | |
f6f37321 | 4864 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
28b2ee20 | 4865 | goto unlock; |
28b2ee20 RR |
4866 | |
4867 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 4868 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 4869 | |
03668a4d | 4870 | ret = 0; |
28b2ee20 RR |
4871 | unlock: |
4872 | pte_unmap_unlock(ptep, ptl); | |
4873 | out: | |
d87fe660 | 4874 | return ret; |
28b2ee20 RR |
4875 | } |
4876 | ||
4877 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
4878 | void *buf, int len, int write) | |
4879 | { | |
4880 | resource_size_t phys_addr; | |
4881 | unsigned long prot = 0; | |
2bc7273b | 4882 | void __iomem *maddr; |
28b2ee20 RR |
4883 | int offset = addr & (PAGE_SIZE-1); |
4884 | ||
d87fe660 | 4885 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
4886 | return -EINVAL; |
4887 | ||
9cb12d7b | 4888 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
24eee1e4 | 4889 | if (!maddr) |
4890 | return -ENOMEM; | |
4891 | ||
28b2ee20 RR |
4892 | if (write) |
4893 | memcpy_toio(maddr + offset, buf, len); | |
4894 | else | |
4895 | memcpy_fromio(buf, maddr + offset, len); | |
4896 | iounmap(maddr); | |
4897 | ||
4898 | return len; | |
4899 | } | |
5a73633e | 4900 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
4901 | #endif |
4902 | ||
0ec76a11 | 4903 | /* |
d3f5ffca | 4904 | * Access another process' address space as given in mm. |
0ec76a11 | 4905 | */ |
d3f5ffca JH |
4906 | int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf, |
4907 | int len, unsigned int gup_flags) | |
0ec76a11 | 4908 | { |
0ec76a11 | 4909 | struct vm_area_struct *vma; |
0ec76a11 | 4910 | void *old_buf = buf; |
442486ec | 4911 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 4912 | |
d8ed45c5 | 4913 | if (mmap_read_lock_killable(mm)) |
1e426fe2 KK |
4914 | return 0; |
4915 | ||
183ff22b | 4916 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
4917 | while (len) { |
4918 | int bytes, ret, offset; | |
4919 | void *maddr; | |
28b2ee20 | 4920 | struct page *page = NULL; |
0ec76a11 | 4921 | |
64019a2e | 4922 | ret = get_user_pages_remote(mm, addr, 1, |
5b56d49f | 4923 | gup_flags, &page, &vma, NULL); |
28b2ee20 | 4924 | if (ret <= 0) { |
dbffcd03 RR |
4925 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
4926 | break; | |
4927 | #else | |
28b2ee20 RR |
4928 | /* |
4929 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
4930 | * we can access using slightly different code. | |
4931 | */ | |
28b2ee20 | 4932 | vma = find_vma(mm, addr); |
fe936dfc | 4933 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
4934 | break; |
4935 | if (vma->vm_ops && vma->vm_ops->access) | |
4936 | ret = vma->vm_ops->access(vma, addr, buf, | |
4937 | len, write); | |
4938 | if (ret <= 0) | |
28b2ee20 RR |
4939 | break; |
4940 | bytes = ret; | |
dbffcd03 | 4941 | #endif |
0ec76a11 | 4942 | } else { |
28b2ee20 RR |
4943 | bytes = len; |
4944 | offset = addr & (PAGE_SIZE-1); | |
4945 | if (bytes > PAGE_SIZE-offset) | |
4946 | bytes = PAGE_SIZE-offset; | |
4947 | ||
4948 | maddr = kmap(page); | |
4949 | if (write) { | |
4950 | copy_to_user_page(vma, page, addr, | |
4951 | maddr + offset, buf, bytes); | |
4952 | set_page_dirty_lock(page); | |
4953 | } else { | |
4954 | copy_from_user_page(vma, page, addr, | |
4955 | buf, maddr + offset, bytes); | |
4956 | } | |
4957 | kunmap(page); | |
09cbfeaf | 4958 | put_page(page); |
0ec76a11 | 4959 | } |
0ec76a11 DH |
4960 | len -= bytes; |
4961 | buf += bytes; | |
4962 | addr += bytes; | |
4963 | } | |
d8ed45c5 | 4964 | mmap_read_unlock(mm); |
0ec76a11 DH |
4965 | |
4966 | return buf - old_buf; | |
4967 | } | |
03252919 | 4968 | |
5ddd36b9 | 4969 | /** |
ae91dbfc | 4970 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
4971 | * @mm: the mm_struct of the target address space |
4972 | * @addr: start address to access | |
4973 | * @buf: source or destination buffer | |
4974 | * @len: number of bytes to transfer | |
6347e8d5 | 4975 | * @gup_flags: flags modifying lookup behaviour |
5ddd36b9 SW |
4976 | * |
4977 | * The caller must hold a reference on @mm. | |
a862f68a MR |
4978 | * |
4979 | * Return: number of bytes copied from source to destination. | |
5ddd36b9 SW |
4980 | */ |
4981 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 4982 | void *buf, int len, unsigned int gup_flags) |
5ddd36b9 | 4983 | { |
d3f5ffca | 4984 | return __access_remote_vm(mm, addr, buf, len, gup_flags); |
5ddd36b9 SW |
4985 | } |
4986 | ||
206cb636 SW |
4987 | /* |
4988 | * Access another process' address space. | |
4989 | * Source/target buffer must be kernel space, | |
4990 | * Do not walk the page table directly, use get_user_pages | |
4991 | */ | |
4992 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
f307ab6d | 4993 | void *buf, int len, unsigned int gup_flags) |
206cb636 SW |
4994 | { |
4995 | struct mm_struct *mm; | |
4996 | int ret; | |
4997 | ||
4998 | mm = get_task_mm(tsk); | |
4999 | if (!mm) | |
5000 | return 0; | |
5001 | ||
d3f5ffca | 5002 | ret = __access_remote_vm(mm, addr, buf, len, gup_flags); |
442486ec | 5003 | |
206cb636 SW |
5004 | mmput(mm); |
5005 | ||
5006 | return ret; | |
5007 | } | |
fcd35857 | 5008 | EXPORT_SYMBOL_GPL(access_process_vm); |
206cb636 | 5009 | |
03252919 AK |
5010 | /* |
5011 | * Print the name of a VMA. | |
5012 | */ | |
5013 | void print_vma_addr(char *prefix, unsigned long ip) | |
5014 | { | |
5015 | struct mm_struct *mm = current->mm; | |
5016 | struct vm_area_struct *vma; | |
5017 | ||
e8bff74a | 5018 | /* |
0a7f682d | 5019 | * we might be running from an atomic context so we cannot sleep |
e8bff74a | 5020 | */ |
d8ed45c5 | 5021 | if (!mmap_read_trylock(mm)) |
e8bff74a IM |
5022 | return; |
5023 | ||
03252919 AK |
5024 | vma = find_vma(mm, ip); |
5025 | if (vma && vma->vm_file) { | |
5026 | struct file *f = vma->vm_file; | |
0a7f682d | 5027 | char *buf = (char *)__get_free_page(GFP_NOWAIT); |
03252919 | 5028 | if (buf) { |
2fbc57c5 | 5029 | char *p; |
03252919 | 5030 | |
9bf39ab2 | 5031 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
5032 | if (IS_ERR(p)) |
5033 | p = "?"; | |
2fbc57c5 | 5034 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
5035 | vma->vm_start, |
5036 | vma->vm_end - vma->vm_start); | |
5037 | free_page((unsigned long)buf); | |
5038 | } | |
5039 | } | |
d8ed45c5 | 5040 | mmap_read_unlock(mm); |
03252919 | 5041 | } |
3ee1afa3 | 5042 | |
662bbcb2 | 5043 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 5044 | void __might_fault(const char *file, int line) |
3ee1afa3 | 5045 | { |
95156f00 PZ |
5046 | /* |
5047 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
c1e8d7c6 | 5048 | * holding the mmap_lock, this is safe because kernel memory doesn't |
95156f00 PZ |
5049 | * get paged out, therefore we'll never actually fault, and the |
5050 | * below annotations will generate false positives. | |
5051 | */ | |
db68ce10 | 5052 | if (uaccess_kernel()) |
95156f00 | 5053 | return; |
9ec23531 | 5054 | if (pagefault_disabled()) |
662bbcb2 | 5055 | return; |
9ec23531 DH |
5056 | __might_sleep(file, line, 0); |
5057 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) | |
662bbcb2 | 5058 | if (current->mm) |
da1c55f1 | 5059 | might_lock_read(¤t->mm->mmap_lock); |
9ec23531 | 5060 | #endif |
3ee1afa3 | 5061 | } |
9ec23531 | 5062 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 5063 | #endif |
47ad8475 AA |
5064 | |
5065 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
c6ddfb6c HY |
5066 | /* |
5067 | * Process all subpages of the specified huge page with the specified | |
5068 | * operation. The target subpage will be processed last to keep its | |
5069 | * cache lines hot. | |
5070 | */ | |
5071 | static inline void process_huge_page( | |
5072 | unsigned long addr_hint, unsigned int pages_per_huge_page, | |
5073 | void (*process_subpage)(unsigned long addr, int idx, void *arg), | |
5074 | void *arg) | |
47ad8475 | 5075 | { |
c79b57e4 HY |
5076 | int i, n, base, l; |
5077 | unsigned long addr = addr_hint & | |
5078 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
47ad8475 | 5079 | |
c6ddfb6c | 5080 | /* Process target subpage last to keep its cache lines hot */ |
47ad8475 | 5081 | might_sleep(); |
c79b57e4 HY |
5082 | n = (addr_hint - addr) / PAGE_SIZE; |
5083 | if (2 * n <= pages_per_huge_page) { | |
c6ddfb6c | 5084 | /* If target subpage in first half of huge page */ |
c79b57e4 HY |
5085 | base = 0; |
5086 | l = n; | |
c6ddfb6c | 5087 | /* Process subpages at the end of huge page */ |
c79b57e4 HY |
5088 | for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { |
5089 | cond_resched(); | |
c6ddfb6c | 5090 | process_subpage(addr + i * PAGE_SIZE, i, arg); |
c79b57e4 HY |
5091 | } |
5092 | } else { | |
c6ddfb6c | 5093 | /* If target subpage in second half of huge page */ |
c79b57e4 HY |
5094 | base = pages_per_huge_page - 2 * (pages_per_huge_page - n); |
5095 | l = pages_per_huge_page - n; | |
c6ddfb6c | 5096 | /* Process subpages at the begin of huge page */ |
c79b57e4 HY |
5097 | for (i = 0; i < base; i++) { |
5098 | cond_resched(); | |
c6ddfb6c | 5099 | process_subpage(addr + i * PAGE_SIZE, i, arg); |
c79b57e4 HY |
5100 | } |
5101 | } | |
5102 | /* | |
c6ddfb6c HY |
5103 | * Process remaining subpages in left-right-left-right pattern |
5104 | * towards the target subpage | |
c79b57e4 HY |
5105 | */ |
5106 | for (i = 0; i < l; i++) { | |
5107 | int left_idx = base + i; | |
5108 | int right_idx = base + 2 * l - 1 - i; | |
5109 | ||
5110 | cond_resched(); | |
c6ddfb6c | 5111 | process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg); |
47ad8475 | 5112 | cond_resched(); |
c6ddfb6c | 5113 | process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg); |
47ad8475 AA |
5114 | } |
5115 | } | |
5116 | ||
c6ddfb6c HY |
5117 | static void clear_gigantic_page(struct page *page, |
5118 | unsigned long addr, | |
5119 | unsigned int pages_per_huge_page) | |
5120 | { | |
5121 | int i; | |
5122 | struct page *p = page; | |
5123 | ||
5124 | might_sleep(); | |
5125 | for (i = 0; i < pages_per_huge_page; | |
5126 | i++, p = mem_map_next(p, page, i)) { | |
5127 | cond_resched(); | |
5128 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
5129 | } | |
5130 | } | |
5131 | ||
5132 | static void clear_subpage(unsigned long addr, int idx, void *arg) | |
5133 | { | |
5134 | struct page *page = arg; | |
5135 | ||
5136 | clear_user_highpage(page + idx, addr); | |
5137 | } | |
5138 | ||
5139 | void clear_huge_page(struct page *page, | |
5140 | unsigned long addr_hint, unsigned int pages_per_huge_page) | |
5141 | { | |
5142 | unsigned long addr = addr_hint & | |
5143 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
5144 | ||
5145 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
5146 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
5147 | return; | |
5148 | } | |
5149 | ||
5150 | process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page); | |
5151 | } | |
5152 | ||
47ad8475 AA |
5153 | static void copy_user_gigantic_page(struct page *dst, struct page *src, |
5154 | unsigned long addr, | |
5155 | struct vm_area_struct *vma, | |
5156 | unsigned int pages_per_huge_page) | |
5157 | { | |
5158 | int i; | |
5159 | struct page *dst_base = dst; | |
5160 | struct page *src_base = src; | |
5161 | ||
5162 | for (i = 0; i < pages_per_huge_page; ) { | |
5163 | cond_resched(); | |
5164 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
5165 | ||
5166 | i++; | |
5167 | dst = mem_map_next(dst, dst_base, i); | |
5168 | src = mem_map_next(src, src_base, i); | |
5169 | } | |
5170 | } | |
5171 | ||
c9f4cd71 HY |
5172 | struct copy_subpage_arg { |
5173 | struct page *dst; | |
5174 | struct page *src; | |
5175 | struct vm_area_struct *vma; | |
5176 | }; | |
5177 | ||
5178 | static void copy_subpage(unsigned long addr, int idx, void *arg) | |
5179 | { | |
5180 | struct copy_subpage_arg *copy_arg = arg; | |
5181 | ||
5182 | copy_user_highpage(copy_arg->dst + idx, copy_arg->src + idx, | |
5183 | addr, copy_arg->vma); | |
5184 | } | |
5185 | ||
47ad8475 | 5186 | void copy_user_huge_page(struct page *dst, struct page *src, |
c9f4cd71 | 5187 | unsigned long addr_hint, struct vm_area_struct *vma, |
47ad8475 AA |
5188 | unsigned int pages_per_huge_page) |
5189 | { | |
c9f4cd71 HY |
5190 | unsigned long addr = addr_hint & |
5191 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
5192 | struct copy_subpage_arg arg = { | |
5193 | .dst = dst, | |
5194 | .src = src, | |
5195 | .vma = vma, | |
5196 | }; | |
47ad8475 AA |
5197 | |
5198 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
5199 | copy_user_gigantic_page(dst, src, addr, vma, | |
5200 | pages_per_huge_page); | |
5201 | return; | |
5202 | } | |
5203 | ||
c9f4cd71 | 5204 | process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg); |
47ad8475 | 5205 | } |
fa4d75c1 MK |
5206 | |
5207 | long copy_huge_page_from_user(struct page *dst_page, | |
5208 | const void __user *usr_src, | |
810a56b9 MK |
5209 | unsigned int pages_per_huge_page, |
5210 | bool allow_pagefault) | |
fa4d75c1 MK |
5211 | { |
5212 | void *src = (void *)usr_src; | |
5213 | void *page_kaddr; | |
5214 | unsigned long i, rc = 0; | |
5215 | unsigned long ret_val = pages_per_huge_page * PAGE_SIZE; | |
ae24fdea | 5216 | struct page *subpage = dst_page; |
fa4d75c1 | 5217 | |
ae24fdea MK |
5218 | for (i = 0; i < pages_per_huge_page; |
5219 | i++, subpage = mem_map_next(subpage, dst_page, i)) { | |
810a56b9 | 5220 | if (allow_pagefault) |
ae24fdea | 5221 | page_kaddr = kmap(subpage); |
810a56b9 | 5222 | else |
ae24fdea | 5223 | page_kaddr = kmap_atomic(subpage); |
fa4d75c1 MK |
5224 | rc = copy_from_user(page_kaddr, |
5225 | (const void __user *)(src + i * PAGE_SIZE), | |
5226 | PAGE_SIZE); | |
810a56b9 | 5227 | if (allow_pagefault) |
ae24fdea | 5228 | kunmap(subpage); |
810a56b9 MK |
5229 | else |
5230 | kunmap_atomic(page_kaddr); | |
fa4d75c1 MK |
5231 | |
5232 | ret_val -= (PAGE_SIZE - rc); | |
5233 | if (rc) | |
5234 | break; | |
5235 | ||
5236 | cond_resched(); | |
5237 | } | |
5238 | return ret_val; | |
5239 | } | |
47ad8475 | 5240 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
49076ec2 | 5241 | |
40b64acd | 5242 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
5243 | |
5244 | static struct kmem_cache *page_ptl_cachep; | |
5245 | ||
5246 | void __init ptlock_cache_init(void) | |
5247 | { | |
5248 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
5249 | SLAB_PANIC, NULL); | |
5250 | } | |
5251 | ||
539edb58 | 5252 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
5253 | { |
5254 | spinlock_t *ptl; | |
5255 | ||
b35f1819 | 5256 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
5257 | if (!ptl) |
5258 | return false; | |
539edb58 | 5259 | page->ptl = ptl; |
49076ec2 KS |
5260 | return true; |
5261 | } | |
5262 | ||
539edb58 | 5263 | void ptlock_free(struct page *page) |
49076ec2 | 5264 | { |
b35f1819 | 5265 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
5266 | } |
5267 | #endif |