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 | ||
a9ee6cf5 | 93 | #ifndef CONFIG_NUMA |
1da177e4 | 94 | unsigned long max_mapnr; |
1da177e4 | 95 | EXPORT_SYMBOL(max_mapnr); |
166f61b9 TH |
96 | |
97 | struct page *mem_map; | |
1da177e4 LT |
98 | EXPORT_SYMBOL(mem_map); |
99 | #endif | |
100 | ||
1da177e4 LT |
101 | /* |
102 | * A number of key systems in x86 including ioremap() rely on the assumption | |
103 | * that high_memory defines the upper bound on direct map memory, then end | |
104 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
105 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
106 | * and ZONE_HIGHMEM. | |
107 | */ | |
166f61b9 | 108 | void *high_memory; |
1da177e4 | 109 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 110 | |
32a93233 IM |
111 | /* |
112 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
113 | * | |
114 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
115 | * as ancient (libc5 based) binaries can segfault. ) | |
116 | */ | |
117 | int randomize_va_space __read_mostly = | |
118 | #ifdef CONFIG_COMPAT_BRK | |
119 | 1; | |
120 | #else | |
121 | 2; | |
122 | #endif | |
a62eaf15 | 123 | |
83d116c5 JH |
124 | #ifndef arch_faults_on_old_pte |
125 | static inline bool arch_faults_on_old_pte(void) | |
126 | { | |
127 | /* | |
128 | * Those arches which don't have hw access flag feature need to | |
129 | * implement their own helper. By default, "true" means pagefault | |
130 | * will be hit on old pte. | |
131 | */ | |
132 | return true; | |
133 | } | |
134 | #endif | |
135 | ||
46bdb427 WD |
136 | #ifndef arch_wants_old_prefaulted_pte |
137 | static inline bool arch_wants_old_prefaulted_pte(void) | |
138 | { | |
139 | /* | |
140 | * Transitioning a PTE from 'old' to 'young' can be expensive on | |
141 | * some architectures, even if it's performed in hardware. By | |
142 | * default, "false" means prefaulted entries will be 'young'. | |
143 | */ | |
144 | return false; | |
145 | } | |
146 | #endif | |
147 | ||
a62eaf15 AK |
148 | static int __init disable_randmaps(char *s) |
149 | { | |
150 | randomize_va_space = 0; | |
9b41046c | 151 | return 1; |
a62eaf15 AK |
152 | } |
153 | __setup("norandmaps", disable_randmaps); | |
154 | ||
62eede62 | 155 | unsigned long zero_pfn __read_mostly; |
0b70068e AB |
156 | EXPORT_SYMBOL(zero_pfn); |
157 | ||
166f61b9 TH |
158 | unsigned long highest_memmap_pfn __read_mostly; |
159 | ||
a13ea5b7 HD |
160 | /* |
161 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
162 | */ | |
163 | static int __init init_zero_pfn(void) | |
164 | { | |
165 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
166 | return 0; | |
167 | } | |
e720e7d0 | 168 | early_initcall(init_zero_pfn); |
a62eaf15 | 169 | |
e4dcad20 | 170 | void mm_trace_rss_stat(struct mm_struct *mm, int member, long count) |
b3d1411b | 171 | { |
e4dcad20 | 172 | trace_rss_stat(mm, member, count); |
b3d1411b | 173 | } |
d559db08 | 174 | |
34e55232 KH |
175 | #if defined(SPLIT_RSS_COUNTING) |
176 | ||
ea48cf78 | 177 | void sync_mm_rss(struct mm_struct *mm) |
34e55232 KH |
178 | { |
179 | int i; | |
180 | ||
181 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
05af2e10 DR |
182 | if (current->rss_stat.count[i]) { |
183 | add_mm_counter(mm, i, current->rss_stat.count[i]); | |
184 | current->rss_stat.count[i] = 0; | |
34e55232 KH |
185 | } |
186 | } | |
05af2e10 | 187 | current->rss_stat.events = 0; |
34e55232 KH |
188 | } |
189 | ||
190 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
191 | { | |
192 | struct task_struct *task = current; | |
193 | ||
194 | if (likely(task->mm == mm)) | |
195 | task->rss_stat.count[member] += val; | |
196 | else | |
197 | add_mm_counter(mm, member, val); | |
198 | } | |
199 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
200 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
201 | ||
202 | /* sync counter once per 64 page faults */ | |
203 | #define TASK_RSS_EVENTS_THRESH (64) | |
204 | static void check_sync_rss_stat(struct task_struct *task) | |
205 | { | |
206 | if (unlikely(task != current)) | |
207 | return; | |
208 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
ea48cf78 | 209 | sync_mm_rss(task->mm); |
34e55232 | 210 | } |
9547d01b | 211 | #else /* SPLIT_RSS_COUNTING */ |
34e55232 KH |
212 | |
213 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
214 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
215 | ||
216 | static void check_sync_rss_stat(struct task_struct *task) | |
217 | { | |
218 | } | |
219 | ||
9547d01b PZ |
220 | #endif /* SPLIT_RSS_COUNTING */ |
221 | ||
1da177e4 LT |
222 | /* |
223 | * Note: this doesn't free the actual pages themselves. That | |
224 | * has been handled earlier when unmapping all the memory regions. | |
225 | */ | |
9e1b32ca BH |
226 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
227 | unsigned long addr) | |
1da177e4 | 228 | { |
2f569afd | 229 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 230 | pmd_clear(pmd); |
9e1b32ca | 231 | pte_free_tlb(tlb, token, addr); |
c4812909 | 232 | mm_dec_nr_ptes(tlb->mm); |
1da177e4 LT |
233 | } |
234 | ||
e0da382c HD |
235 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
236 | unsigned long addr, unsigned long end, | |
237 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
238 | { |
239 | pmd_t *pmd; | |
240 | unsigned long next; | |
e0da382c | 241 | unsigned long start; |
1da177e4 | 242 | |
e0da382c | 243 | start = addr; |
1da177e4 | 244 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
245 | do { |
246 | next = pmd_addr_end(addr, end); | |
247 | if (pmd_none_or_clear_bad(pmd)) | |
248 | continue; | |
9e1b32ca | 249 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
250 | } while (pmd++, addr = next, addr != end); |
251 | ||
e0da382c HD |
252 | start &= PUD_MASK; |
253 | if (start < floor) | |
254 | return; | |
255 | if (ceiling) { | |
256 | ceiling &= PUD_MASK; | |
257 | if (!ceiling) | |
258 | return; | |
1da177e4 | 259 | } |
e0da382c HD |
260 | if (end - 1 > ceiling - 1) |
261 | return; | |
262 | ||
263 | pmd = pmd_offset(pud, start); | |
264 | pud_clear(pud); | |
9e1b32ca | 265 | pmd_free_tlb(tlb, pmd, start); |
dc6c9a35 | 266 | mm_dec_nr_pmds(tlb->mm); |
1da177e4 LT |
267 | } |
268 | ||
c2febafc | 269 | static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, |
e0da382c HD |
270 | unsigned long addr, unsigned long end, |
271 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
272 | { |
273 | pud_t *pud; | |
274 | unsigned long next; | |
e0da382c | 275 | unsigned long start; |
1da177e4 | 276 | |
e0da382c | 277 | start = addr; |
c2febafc | 278 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
279 | do { |
280 | next = pud_addr_end(addr, end); | |
281 | if (pud_none_or_clear_bad(pud)) | |
282 | continue; | |
e0da382c | 283 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
284 | } while (pud++, addr = next, addr != end); |
285 | ||
c2febafc KS |
286 | start &= P4D_MASK; |
287 | if (start < floor) | |
288 | return; | |
289 | if (ceiling) { | |
290 | ceiling &= P4D_MASK; | |
291 | if (!ceiling) | |
292 | return; | |
293 | } | |
294 | if (end - 1 > ceiling - 1) | |
295 | return; | |
296 | ||
297 | pud = pud_offset(p4d, start); | |
298 | p4d_clear(p4d); | |
299 | pud_free_tlb(tlb, pud, start); | |
b4e98d9a | 300 | mm_dec_nr_puds(tlb->mm); |
c2febafc KS |
301 | } |
302 | ||
303 | static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, | |
304 | unsigned long addr, unsigned long end, | |
305 | unsigned long floor, unsigned long ceiling) | |
306 | { | |
307 | p4d_t *p4d; | |
308 | unsigned long next; | |
309 | unsigned long start; | |
310 | ||
311 | start = addr; | |
312 | p4d = p4d_offset(pgd, addr); | |
313 | do { | |
314 | next = p4d_addr_end(addr, end); | |
315 | if (p4d_none_or_clear_bad(p4d)) | |
316 | continue; | |
317 | free_pud_range(tlb, p4d, addr, next, floor, ceiling); | |
318 | } while (p4d++, addr = next, addr != end); | |
319 | ||
e0da382c HD |
320 | start &= PGDIR_MASK; |
321 | if (start < floor) | |
322 | return; | |
323 | if (ceiling) { | |
324 | ceiling &= PGDIR_MASK; | |
325 | if (!ceiling) | |
326 | return; | |
1da177e4 | 327 | } |
e0da382c HD |
328 | if (end - 1 > ceiling - 1) |
329 | return; | |
330 | ||
c2febafc | 331 | p4d = p4d_offset(pgd, start); |
e0da382c | 332 | pgd_clear(pgd); |
c2febafc | 333 | p4d_free_tlb(tlb, p4d, start); |
1da177e4 LT |
334 | } |
335 | ||
336 | /* | |
e0da382c | 337 | * This function frees user-level page tables of a process. |
1da177e4 | 338 | */ |
42b77728 | 339 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
340 | unsigned long addr, unsigned long end, |
341 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
342 | { |
343 | pgd_t *pgd; | |
344 | unsigned long next; | |
e0da382c HD |
345 | |
346 | /* | |
347 | * The next few lines have given us lots of grief... | |
348 | * | |
349 | * Why are we testing PMD* at this top level? Because often | |
350 | * there will be no work to do at all, and we'd prefer not to | |
351 | * go all the way down to the bottom just to discover that. | |
352 | * | |
353 | * Why all these "- 1"s? Because 0 represents both the bottom | |
354 | * of the address space and the top of it (using -1 for the | |
355 | * top wouldn't help much: the masks would do the wrong thing). | |
356 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
357 | * the address space, but end 0 and ceiling 0 refer to the top | |
358 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
359 | * that end 0 case should be mythical). | |
360 | * | |
361 | * Wherever addr is brought up or ceiling brought down, we must | |
362 | * be careful to reject "the opposite 0" before it confuses the | |
363 | * subsequent tests. But what about where end is brought down | |
364 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
365 | * | |
366 | * Whereas we round start (addr) and ceiling down, by different | |
367 | * masks at different levels, in order to test whether a table | |
368 | * now has no other vmas using it, so can be freed, we don't | |
369 | * bother to round floor or end up - the tests don't need that. | |
370 | */ | |
1da177e4 | 371 | |
e0da382c HD |
372 | addr &= PMD_MASK; |
373 | if (addr < floor) { | |
374 | addr += PMD_SIZE; | |
375 | if (!addr) | |
376 | return; | |
377 | } | |
378 | if (ceiling) { | |
379 | ceiling &= PMD_MASK; | |
380 | if (!ceiling) | |
381 | return; | |
382 | } | |
383 | if (end - 1 > ceiling - 1) | |
384 | end -= PMD_SIZE; | |
385 | if (addr > end - 1) | |
386 | return; | |
07e32661 AK |
387 | /* |
388 | * We add page table cache pages with PAGE_SIZE, | |
389 | * (see pte_free_tlb()), flush the tlb if we need | |
390 | */ | |
ed6a7935 | 391 | tlb_change_page_size(tlb, PAGE_SIZE); |
42b77728 | 392 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
393 | do { |
394 | next = pgd_addr_end(addr, end); | |
395 | if (pgd_none_or_clear_bad(pgd)) | |
396 | continue; | |
c2febafc | 397 | free_p4d_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 398 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
399 | } |
400 | ||
42b77728 | 401 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 402 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
403 | { |
404 | while (vma) { | |
405 | struct vm_area_struct *next = vma->vm_next; | |
406 | unsigned long addr = vma->vm_start; | |
407 | ||
8f4f8c16 | 408 | /* |
25d9e2d1 | 409 | * Hide vma from rmap and truncate_pagecache before freeing |
410 | * pgtables | |
8f4f8c16 | 411 | */ |
5beb4930 | 412 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
413 | unlink_file_vma(vma); |
414 | ||
9da61aef | 415 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 416 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
166f61b9 | 417 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 HD |
418 | } else { |
419 | /* | |
420 | * Optimization: gather nearby vmas into one call down | |
421 | */ | |
422 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 423 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
424 | vma = next; |
425 | next = vma->vm_next; | |
5beb4930 | 426 | unlink_anon_vmas(vma); |
8f4f8c16 | 427 | unlink_file_vma(vma); |
3bf5ee95 HD |
428 | } |
429 | free_pgd_range(tlb, addr, vma->vm_end, | |
166f61b9 | 430 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 | 431 | } |
e0da382c HD |
432 | vma = next; |
433 | } | |
1da177e4 LT |
434 | } |
435 | ||
4cf58924 | 436 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd) |
1da177e4 | 437 | { |
c4088ebd | 438 | spinlock_t *ptl; |
4cf58924 | 439 | pgtable_t new = pte_alloc_one(mm); |
1bb3630e HD |
440 | if (!new) |
441 | return -ENOMEM; | |
442 | ||
362a61ad NP |
443 | /* |
444 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
445 | * visible before the pte is made visible to other CPUs by being | |
446 | * put into page tables. | |
447 | * | |
448 | * The other side of the story is the pointer chasing in the page | |
449 | * table walking code (when walking the page table without locking; | |
450 | * ie. most of the time). Fortunately, these data accesses consist | |
451 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
452 | * being the notable exception) will already guarantee loads are | |
453 | * seen in-order. See the alpha page table accessors for the | |
bb7cdd38 | 454 | * smp_rmb() barriers in page table walking code. |
362a61ad NP |
455 | */ |
456 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
457 | ||
c4088ebd | 458 | ptl = pmd_lock(mm, pmd); |
8ac1f832 | 459 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
c4812909 | 460 | mm_inc_nr_ptes(mm); |
1da177e4 | 461 | pmd_populate(mm, pmd, new); |
2f569afd | 462 | new = NULL; |
4b471e88 | 463 | } |
c4088ebd | 464 | spin_unlock(ptl); |
2f569afd MS |
465 | if (new) |
466 | pte_free(mm, new); | |
1bb3630e | 467 | return 0; |
1da177e4 LT |
468 | } |
469 | ||
4cf58924 | 470 | int __pte_alloc_kernel(pmd_t *pmd) |
1da177e4 | 471 | { |
4cf58924 | 472 | pte_t *new = pte_alloc_one_kernel(&init_mm); |
1bb3630e HD |
473 | if (!new) |
474 | return -ENOMEM; | |
475 | ||
362a61ad NP |
476 | smp_wmb(); /* See comment in __pte_alloc */ |
477 | ||
1bb3630e | 478 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 479 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 480 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 481 | new = NULL; |
4b471e88 | 482 | } |
1bb3630e | 483 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
484 | if (new) |
485 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 486 | return 0; |
1da177e4 LT |
487 | } |
488 | ||
d559db08 KH |
489 | static inline void init_rss_vec(int *rss) |
490 | { | |
491 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
492 | } | |
493 | ||
494 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 495 | { |
d559db08 KH |
496 | int i; |
497 | ||
34e55232 | 498 | if (current->mm == mm) |
05af2e10 | 499 | sync_mm_rss(mm); |
d559db08 KH |
500 | for (i = 0; i < NR_MM_COUNTERS; i++) |
501 | if (rss[i]) | |
502 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
503 | } |
504 | ||
b5810039 | 505 | /* |
6aab341e LT |
506 | * This function is called to print an error when a bad pte |
507 | * is found. For example, we might have a PFN-mapped pte in | |
508 | * a region that doesn't allow it. | |
b5810039 NP |
509 | * |
510 | * The calling function must still handle the error. | |
511 | */ | |
3dc14741 HD |
512 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
513 | pte_t pte, struct page *page) | |
b5810039 | 514 | { |
3dc14741 | 515 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
c2febafc KS |
516 | p4d_t *p4d = p4d_offset(pgd, addr); |
517 | pud_t *pud = pud_offset(p4d, addr); | |
3dc14741 HD |
518 | pmd_t *pmd = pmd_offset(pud, addr); |
519 | struct address_space *mapping; | |
520 | pgoff_t index; | |
d936cf9b HD |
521 | static unsigned long resume; |
522 | static unsigned long nr_shown; | |
523 | static unsigned long nr_unshown; | |
524 | ||
525 | /* | |
526 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
527 | * or allow a steady drip of one report per second. | |
528 | */ | |
529 | if (nr_shown == 60) { | |
530 | if (time_before(jiffies, resume)) { | |
531 | nr_unshown++; | |
532 | return; | |
533 | } | |
534 | if (nr_unshown) { | |
1170532b JP |
535 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
536 | nr_unshown); | |
d936cf9b HD |
537 | nr_unshown = 0; |
538 | } | |
539 | nr_shown = 0; | |
540 | } | |
541 | if (nr_shown++ == 0) | |
542 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
543 | |
544 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
545 | index = linear_page_index(vma, addr); | |
546 | ||
1170532b JP |
547 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
548 | current->comm, | |
549 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 550 | if (page) |
f0b791a3 | 551 | dump_page(page, "bad pte"); |
6aa9b8b2 | 552 | pr_alert("addr:%px vm_flags:%08lx anon_vma:%px mapping:%px index:%lx\n", |
1170532b | 553 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); |
d75f773c | 554 | pr_alert("file:%pD fault:%ps mmap:%ps readpage:%ps\n", |
2682582a KK |
555 | vma->vm_file, |
556 | vma->vm_ops ? vma->vm_ops->fault : NULL, | |
557 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, | |
558 | mapping ? mapping->a_ops->readpage : NULL); | |
b5810039 | 559 | dump_stack(); |
373d4d09 | 560 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
561 | } |
562 | ||
ee498ed7 | 563 | /* |
7e675137 | 564 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 565 | * |
7e675137 NP |
566 | * "Special" mappings do not wish to be associated with a "struct page" (either |
567 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
568 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 569 | * |
7e675137 NP |
570 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
571 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
572 | * may not have a spare pte bit, which requires a more complicated scheme, | |
573 | * described below. | |
574 | * | |
575 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
576 | * special mapping (even if there are underlying and valid "struct pages"). | |
577 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 578 | * |
b379d790 JH |
579 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
580 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
581 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
582 | * mapping will always honor the rule | |
6aab341e LT |
583 | * |
584 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
585 | * | |
7e675137 NP |
586 | * And for normal mappings this is false. |
587 | * | |
588 | * This restricts such mappings to be a linear translation from virtual address | |
589 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
590 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
591 | * special (because none can have been COWed). | |
b379d790 | 592 | * |
b379d790 | 593 | * |
7e675137 | 594 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
595 | * |
596 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
597 | * page" backing, however the difference is that _all_ pages with a struct | |
598 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
599 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
600 | * (which can be slower and simply not an option for some PFNMAP users). The | |
601 | * advantage is that we don't have to follow the strict linearity rule of | |
602 | * PFNMAP mappings in order to support COWable mappings. | |
603 | * | |
ee498ed7 | 604 | */ |
25b2995a CH |
605 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
606 | pte_t pte) | |
ee498ed7 | 607 | { |
22b31eec | 608 | unsigned long pfn = pte_pfn(pte); |
7e675137 | 609 | |
00b3a331 | 610 | if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) { |
b38af472 | 611 | if (likely(!pte_special(pte))) |
22b31eec | 612 | goto check_pfn; |
667a0a06 DV |
613 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
614 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
615 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
616 | return NULL; | |
df6ad698 JG |
617 | if (is_zero_pfn(pfn)) |
618 | return NULL; | |
e1fb4a08 DJ |
619 | if (pte_devmap(pte)) |
620 | return NULL; | |
621 | ||
df6ad698 | 622 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
623 | return NULL; |
624 | } | |
625 | ||
00b3a331 | 626 | /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */ |
7e675137 | 627 | |
b379d790 JH |
628 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
629 | if (vma->vm_flags & VM_MIXEDMAP) { | |
630 | if (!pfn_valid(pfn)) | |
631 | return NULL; | |
632 | goto out; | |
633 | } else { | |
7e675137 NP |
634 | unsigned long off; |
635 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
636 | if (pfn == vma->vm_pgoff + off) |
637 | return NULL; | |
638 | if (!is_cow_mapping(vma->vm_flags)) | |
639 | return NULL; | |
640 | } | |
6aab341e LT |
641 | } |
642 | ||
b38af472 HD |
643 | if (is_zero_pfn(pfn)) |
644 | return NULL; | |
00b3a331 | 645 | |
22b31eec HD |
646 | check_pfn: |
647 | if (unlikely(pfn > highest_memmap_pfn)) { | |
648 | print_bad_pte(vma, addr, pte, NULL); | |
649 | return NULL; | |
650 | } | |
6aab341e LT |
651 | |
652 | /* | |
7e675137 | 653 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 654 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 655 | */ |
b379d790 | 656 | out: |
6aab341e | 657 | return pfn_to_page(pfn); |
ee498ed7 HD |
658 | } |
659 | ||
28093f9f GS |
660 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
661 | struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, | |
662 | pmd_t pmd) | |
663 | { | |
664 | unsigned long pfn = pmd_pfn(pmd); | |
665 | ||
666 | /* | |
667 | * There is no pmd_special() but there may be special pmds, e.g. | |
668 | * in a direct-access (dax) mapping, so let's just replicate the | |
00b3a331 | 669 | * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here. |
28093f9f GS |
670 | */ |
671 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { | |
672 | if (vma->vm_flags & VM_MIXEDMAP) { | |
673 | if (!pfn_valid(pfn)) | |
674 | return NULL; | |
675 | goto out; | |
676 | } else { | |
677 | unsigned long off; | |
678 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
679 | if (pfn == vma->vm_pgoff + off) | |
680 | return NULL; | |
681 | if (!is_cow_mapping(vma->vm_flags)) | |
682 | return NULL; | |
683 | } | |
684 | } | |
685 | ||
e1fb4a08 DJ |
686 | if (pmd_devmap(pmd)) |
687 | return NULL; | |
3cde287b | 688 | if (is_huge_zero_pmd(pmd)) |
28093f9f GS |
689 | return NULL; |
690 | if (unlikely(pfn > highest_memmap_pfn)) | |
691 | return NULL; | |
692 | ||
693 | /* | |
694 | * NOTE! We still have PageReserved() pages in the page tables. | |
695 | * eg. VDSO mappings can cause them to exist. | |
696 | */ | |
697 | out: | |
698 | return pfn_to_page(pfn); | |
699 | } | |
700 | #endif | |
701 | ||
1da177e4 LT |
702 | /* |
703 | * copy one vm_area from one task to the other. Assumes the page tables | |
704 | * already present in the new task to be cleared in the whole range | |
705 | * covered by this vma. | |
1da177e4 LT |
706 | */ |
707 | ||
df3a57d1 LT |
708 | static unsigned long |
709 | copy_nonpresent_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
8f34f1ea PX |
710 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *dst_vma, |
711 | struct vm_area_struct *src_vma, unsigned long addr, int *rss) | |
1da177e4 | 712 | { |
8f34f1ea | 713 | unsigned long vm_flags = dst_vma->vm_flags; |
1da177e4 LT |
714 | pte_t pte = *src_pte; |
715 | struct page *page; | |
df3a57d1 LT |
716 | swp_entry_t entry = pte_to_swp_entry(pte); |
717 | ||
718 | if (likely(!non_swap_entry(entry))) { | |
719 | if (swap_duplicate(entry) < 0) | |
9a5cc85c | 720 | return -EIO; |
df3a57d1 LT |
721 | |
722 | /* make sure dst_mm is on swapoff's mmlist. */ | |
723 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
724 | spin_lock(&mmlist_lock); | |
725 | if (list_empty(&dst_mm->mmlist)) | |
726 | list_add(&dst_mm->mmlist, | |
727 | &src_mm->mmlist); | |
728 | spin_unlock(&mmlist_lock); | |
729 | } | |
730 | rss[MM_SWAPENTS]++; | |
731 | } else if (is_migration_entry(entry)) { | |
af5cdaf8 | 732 | page = pfn_swap_entry_to_page(entry); |
1da177e4 | 733 | |
df3a57d1 | 734 | rss[mm_counter(page)]++; |
5042db43 | 735 | |
4dd845b5 | 736 | if (is_writable_migration_entry(entry) && |
df3a57d1 | 737 | is_cow_mapping(vm_flags)) { |
5042db43 | 738 | /* |
df3a57d1 LT |
739 | * COW mappings require pages in both |
740 | * parent and child to be set to read. | |
5042db43 | 741 | */ |
4dd845b5 AP |
742 | entry = make_readable_migration_entry( |
743 | swp_offset(entry)); | |
df3a57d1 LT |
744 | pte = swp_entry_to_pte(entry); |
745 | if (pte_swp_soft_dirty(*src_pte)) | |
746 | pte = pte_swp_mksoft_dirty(pte); | |
747 | if (pte_swp_uffd_wp(*src_pte)) | |
748 | pte = pte_swp_mkuffd_wp(pte); | |
749 | set_pte_at(src_mm, addr, src_pte, pte); | |
750 | } | |
751 | } else if (is_device_private_entry(entry)) { | |
af5cdaf8 | 752 | page = pfn_swap_entry_to_page(entry); |
5042db43 | 753 | |
df3a57d1 LT |
754 | /* |
755 | * Update rss count even for unaddressable pages, as | |
756 | * they should treated just like normal pages in this | |
757 | * respect. | |
758 | * | |
759 | * We will likely want to have some new rss counters | |
760 | * for unaddressable pages, at some point. But for now | |
761 | * keep things as they are. | |
762 | */ | |
763 | get_page(page); | |
764 | rss[mm_counter(page)]++; | |
765 | page_dup_rmap(page, false); | |
766 | ||
767 | /* | |
768 | * We do not preserve soft-dirty information, because so | |
769 | * far, checkpoint/restore is the only feature that | |
770 | * requires that. And checkpoint/restore does not work | |
771 | * when a device driver is involved (you cannot easily | |
772 | * save and restore device driver state). | |
773 | */ | |
4dd845b5 | 774 | if (is_writable_device_private_entry(entry) && |
df3a57d1 | 775 | is_cow_mapping(vm_flags)) { |
4dd845b5 AP |
776 | entry = make_readable_device_private_entry( |
777 | swp_offset(entry)); | |
df3a57d1 LT |
778 | pte = swp_entry_to_pte(entry); |
779 | if (pte_swp_uffd_wp(*src_pte)) | |
780 | pte = pte_swp_mkuffd_wp(pte); | |
781 | set_pte_at(src_mm, addr, src_pte, pte); | |
1da177e4 | 782 | } |
1da177e4 | 783 | } |
8f34f1ea PX |
784 | if (!userfaultfd_wp(dst_vma)) |
785 | pte = pte_swp_clear_uffd_wp(pte); | |
df3a57d1 LT |
786 | set_pte_at(dst_mm, addr, dst_pte, pte); |
787 | return 0; | |
788 | } | |
789 | ||
70e806e4 PX |
790 | /* |
791 | * Copy a present and normal page if necessary. | |
792 | * | |
793 | * NOTE! The usual case is that this doesn't need to do | |
794 | * anything, and can just return a positive value. That | |
795 | * will let the caller know that it can just increase | |
796 | * the page refcount and re-use the pte the traditional | |
797 | * way. | |
798 | * | |
799 | * But _if_ we need to copy it because it needs to be | |
800 | * pinned in the parent (and the child should get its own | |
801 | * copy rather than just a reference to the same page), | |
802 | * we'll do that here and return zero to let the caller | |
803 | * know we're done. | |
804 | * | |
805 | * And if we need a pre-allocated page but don't yet have | |
806 | * one, return a negative error to let the preallocation | |
807 | * code know so that it can do so outside the page table | |
808 | * lock. | |
809 | */ | |
810 | static inline int | |
c78f4636 PX |
811 | copy_present_page(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, |
812 | pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, | |
813 | struct page **prealloc, pte_t pte, struct page *page) | |
70e806e4 PX |
814 | { |
815 | struct page *new_page; | |
816 | ||
70e806e4 | 817 | /* |
70e806e4 PX |
818 | * What we want to do is to check whether this page may |
819 | * have been pinned by the parent process. If so, | |
820 | * instead of wrprotect the pte on both sides, we copy | |
821 | * the page immediately so that we'll always guarantee | |
822 | * the pinned page won't be randomly replaced in the | |
823 | * future. | |
824 | * | |
f3c64eda LT |
825 | * The page pinning checks are just "has this mm ever |
826 | * seen pinning", along with the (inexact) check of | |
827 | * the page count. That might give false positives for | |
828 | * for pinning, but it will work correctly. | |
70e806e4 | 829 | */ |
97a7e473 | 830 | if (likely(!page_needs_cow_for_dma(src_vma, page))) |
70e806e4 PX |
831 | return 1; |
832 | ||
70e806e4 PX |
833 | new_page = *prealloc; |
834 | if (!new_page) | |
835 | return -EAGAIN; | |
836 | ||
837 | /* | |
838 | * We have a prealloc page, all good! Take it | |
839 | * over and copy the page & arm it. | |
840 | */ | |
841 | *prealloc = NULL; | |
c78f4636 | 842 | copy_user_highpage(new_page, page, addr, src_vma); |
70e806e4 | 843 | __SetPageUptodate(new_page); |
c78f4636 PX |
844 | page_add_new_anon_rmap(new_page, dst_vma, addr, false); |
845 | lru_cache_add_inactive_or_unevictable(new_page, dst_vma); | |
70e806e4 PX |
846 | rss[mm_counter(new_page)]++; |
847 | ||
848 | /* All done, just insert the new page copy in the child */ | |
c78f4636 PX |
849 | pte = mk_pte(new_page, dst_vma->vm_page_prot); |
850 | pte = maybe_mkwrite(pte_mkdirty(pte), dst_vma); | |
8f34f1ea PX |
851 | if (userfaultfd_pte_wp(dst_vma, *src_pte)) |
852 | /* Uffd-wp needs to be delivered to dest pte as well */ | |
853 | pte = pte_wrprotect(pte_mkuffd_wp(pte)); | |
c78f4636 | 854 | set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); |
70e806e4 PX |
855 | return 0; |
856 | } | |
857 | ||
858 | /* | |
859 | * Copy one pte. Returns 0 if succeeded, or -EAGAIN if one preallocated page | |
860 | * is required to copy this pte. | |
861 | */ | |
862 | static inline int | |
c78f4636 PX |
863 | copy_present_pte(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, |
864 | pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, | |
865 | struct page **prealloc) | |
df3a57d1 | 866 | { |
c78f4636 PX |
867 | struct mm_struct *src_mm = src_vma->vm_mm; |
868 | unsigned long vm_flags = src_vma->vm_flags; | |
df3a57d1 LT |
869 | pte_t pte = *src_pte; |
870 | struct page *page; | |
871 | ||
c78f4636 | 872 | page = vm_normal_page(src_vma, addr, pte); |
70e806e4 PX |
873 | if (page) { |
874 | int retval; | |
875 | ||
c78f4636 PX |
876 | retval = copy_present_page(dst_vma, src_vma, dst_pte, src_pte, |
877 | addr, rss, prealloc, pte, page); | |
70e806e4 PX |
878 | if (retval <= 0) |
879 | return retval; | |
880 | ||
881 | get_page(page); | |
882 | page_dup_rmap(page, false); | |
883 | rss[mm_counter(page)]++; | |
884 | } | |
885 | ||
1da177e4 LT |
886 | /* |
887 | * If it's a COW mapping, write protect it both | |
888 | * in the parent and the child | |
889 | */ | |
1b2de5d0 | 890 | if (is_cow_mapping(vm_flags) && pte_write(pte)) { |
1da177e4 | 891 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 892 | pte = pte_wrprotect(pte); |
1da177e4 LT |
893 | } |
894 | ||
895 | /* | |
896 | * If it's a shared mapping, mark it clean in | |
897 | * the child | |
898 | */ | |
899 | if (vm_flags & VM_SHARED) | |
900 | pte = pte_mkclean(pte); | |
901 | pte = pte_mkold(pte); | |
6aab341e | 902 | |
8f34f1ea | 903 | if (!userfaultfd_wp(dst_vma)) |
b569a176 PX |
904 | pte = pte_clear_uffd_wp(pte); |
905 | ||
c78f4636 | 906 | set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); |
70e806e4 PX |
907 | return 0; |
908 | } | |
909 | ||
910 | static inline struct page * | |
911 | page_copy_prealloc(struct mm_struct *src_mm, struct vm_area_struct *vma, | |
912 | unsigned long addr) | |
913 | { | |
914 | struct page *new_page; | |
915 | ||
916 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, addr); | |
917 | if (!new_page) | |
918 | return NULL; | |
919 | ||
920 | if (mem_cgroup_charge(new_page, src_mm, GFP_KERNEL)) { | |
921 | put_page(new_page); | |
922 | return NULL; | |
6aab341e | 923 | } |
70e806e4 | 924 | cgroup_throttle_swaprate(new_page, GFP_KERNEL); |
ae859762 | 925 | |
70e806e4 | 926 | return new_page; |
1da177e4 LT |
927 | } |
928 | ||
c78f4636 PX |
929 | static int |
930 | copy_pte_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
931 | pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr, | |
932 | unsigned long end) | |
1da177e4 | 933 | { |
c78f4636 PX |
934 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
935 | struct mm_struct *src_mm = src_vma->vm_mm; | |
c36987e2 | 936 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 937 | pte_t *src_pte, *dst_pte; |
c74df32c | 938 | spinlock_t *src_ptl, *dst_ptl; |
70e806e4 | 939 | int progress, ret = 0; |
d559db08 | 940 | int rss[NR_MM_COUNTERS]; |
570a335b | 941 | swp_entry_t entry = (swp_entry_t){0}; |
70e806e4 | 942 | struct page *prealloc = NULL; |
1da177e4 LT |
943 | |
944 | again: | |
70e806e4 | 945 | progress = 0; |
d559db08 KH |
946 | init_rss_vec(rss); |
947 | ||
c74df32c | 948 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
70e806e4 PX |
949 | if (!dst_pte) { |
950 | ret = -ENOMEM; | |
951 | goto out; | |
952 | } | |
ece0e2b6 | 953 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 954 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 955 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
956 | orig_src_pte = src_pte; |
957 | orig_dst_pte = dst_pte; | |
6606c3e0 | 958 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 959 | |
1da177e4 LT |
960 | do { |
961 | /* | |
962 | * We are holding two locks at this point - either of them | |
963 | * could generate latencies in another task on another CPU. | |
964 | */ | |
e040f218 HD |
965 | if (progress >= 32) { |
966 | progress = 0; | |
967 | if (need_resched() || | |
95c354fe | 968 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
969 | break; |
970 | } | |
1da177e4 LT |
971 | if (pte_none(*src_pte)) { |
972 | progress++; | |
973 | continue; | |
974 | } | |
79a1971c | 975 | if (unlikely(!pte_present(*src_pte))) { |
9a5cc85c AP |
976 | ret = copy_nonpresent_pte(dst_mm, src_mm, |
977 | dst_pte, src_pte, | |
978 | dst_vma, src_vma, | |
979 | addr, rss); | |
980 | if (ret == -EIO) { | |
981 | entry = pte_to_swp_entry(*src_pte); | |
79a1971c | 982 | break; |
9a5cc85c | 983 | } |
79a1971c LT |
984 | progress += 8; |
985 | continue; | |
986 | } | |
70e806e4 | 987 | /* copy_present_pte() will clear `*prealloc' if consumed */ |
c78f4636 PX |
988 | ret = copy_present_pte(dst_vma, src_vma, dst_pte, src_pte, |
989 | addr, rss, &prealloc); | |
70e806e4 PX |
990 | /* |
991 | * If we need a pre-allocated page for this pte, drop the | |
992 | * locks, allocate, and try again. | |
993 | */ | |
994 | if (unlikely(ret == -EAGAIN)) | |
995 | break; | |
996 | if (unlikely(prealloc)) { | |
997 | /* | |
998 | * pre-alloc page cannot be reused by next time so as | |
999 | * to strictly follow mempolicy (e.g., alloc_page_vma() | |
1000 | * will allocate page according to address). This | |
1001 | * could only happen if one pinned pte changed. | |
1002 | */ | |
1003 | put_page(prealloc); | |
1004 | prealloc = NULL; | |
1005 | } | |
1da177e4 LT |
1006 | progress += 8; |
1007 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 1008 | |
6606c3e0 | 1009 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1010 | spin_unlock(src_ptl); |
ece0e2b6 | 1011 | pte_unmap(orig_src_pte); |
d559db08 | 1012 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 1013 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 1014 | cond_resched(); |
570a335b | 1015 | |
9a5cc85c AP |
1016 | if (ret == -EIO) { |
1017 | VM_WARN_ON_ONCE(!entry.val); | |
70e806e4 PX |
1018 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) { |
1019 | ret = -ENOMEM; | |
1020 | goto out; | |
1021 | } | |
1022 | entry.val = 0; | |
9a5cc85c | 1023 | } else if (ret == -EAGAIN) { |
c78f4636 | 1024 | prealloc = page_copy_prealloc(src_mm, src_vma, addr); |
70e806e4 | 1025 | if (!prealloc) |
570a335b | 1026 | return -ENOMEM; |
9a5cc85c AP |
1027 | } else if (ret) { |
1028 | VM_WARN_ON_ONCE(1); | |
570a335b | 1029 | } |
9a5cc85c AP |
1030 | |
1031 | /* We've captured and resolved the error. Reset, try again. */ | |
1032 | ret = 0; | |
1033 | ||
1da177e4 LT |
1034 | if (addr != end) |
1035 | goto again; | |
70e806e4 PX |
1036 | out: |
1037 | if (unlikely(prealloc)) | |
1038 | put_page(prealloc); | |
1039 | return ret; | |
1da177e4 LT |
1040 | } |
1041 | ||
c78f4636 PX |
1042 | static inline int |
1043 | copy_pmd_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1044 | pud_t *dst_pud, pud_t *src_pud, unsigned long addr, | |
1045 | unsigned long end) | |
1da177e4 | 1046 | { |
c78f4636 PX |
1047 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
1048 | struct mm_struct *src_mm = src_vma->vm_mm; | |
1da177e4 LT |
1049 | pmd_t *src_pmd, *dst_pmd; |
1050 | unsigned long next; | |
1051 | ||
1052 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
1053 | if (!dst_pmd) | |
1054 | return -ENOMEM; | |
1055 | src_pmd = pmd_offset(src_pud, addr); | |
1056 | do { | |
1057 | next = pmd_addr_end(addr, end); | |
84c3fc4e ZY |
1058 | if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) |
1059 | || pmd_devmap(*src_pmd)) { | |
71e3aac0 | 1060 | int err; |
c78f4636 | 1061 | VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, src_vma); |
8f34f1ea PX |
1062 | err = copy_huge_pmd(dst_mm, src_mm, dst_pmd, src_pmd, |
1063 | addr, dst_vma, src_vma); | |
71e3aac0 AA |
1064 | if (err == -ENOMEM) |
1065 | return -ENOMEM; | |
1066 | if (!err) | |
1067 | continue; | |
1068 | /* fall through */ | |
1069 | } | |
1da177e4 LT |
1070 | if (pmd_none_or_clear_bad(src_pmd)) |
1071 | continue; | |
c78f4636 PX |
1072 | if (copy_pte_range(dst_vma, src_vma, dst_pmd, src_pmd, |
1073 | addr, next)) | |
1da177e4 LT |
1074 | return -ENOMEM; |
1075 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
1076 | return 0; | |
1077 | } | |
1078 | ||
c78f4636 PX |
1079 | static inline int |
1080 | copy_pud_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1081 | p4d_t *dst_p4d, p4d_t *src_p4d, unsigned long addr, | |
1082 | unsigned long end) | |
1da177e4 | 1083 | { |
c78f4636 PX |
1084 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
1085 | struct mm_struct *src_mm = src_vma->vm_mm; | |
1da177e4 LT |
1086 | pud_t *src_pud, *dst_pud; |
1087 | unsigned long next; | |
1088 | ||
c2febafc | 1089 | dst_pud = pud_alloc(dst_mm, dst_p4d, addr); |
1da177e4 LT |
1090 | if (!dst_pud) |
1091 | return -ENOMEM; | |
c2febafc | 1092 | src_pud = pud_offset(src_p4d, addr); |
1da177e4 LT |
1093 | do { |
1094 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1095 | if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { |
1096 | int err; | |
1097 | ||
c78f4636 | 1098 | VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, src_vma); |
a00cc7d9 | 1099 | err = copy_huge_pud(dst_mm, src_mm, |
c78f4636 | 1100 | dst_pud, src_pud, addr, src_vma); |
a00cc7d9 MW |
1101 | if (err == -ENOMEM) |
1102 | return -ENOMEM; | |
1103 | if (!err) | |
1104 | continue; | |
1105 | /* fall through */ | |
1106 | } | |
1da177e4 LT |
1107 | if (pud_none_or_clear_bad(src_pud)) |
1108 | continue; | |
c78f4636 PX |
1109 | if (copy_pmd_range(dst_vma, src_vma, dst_pud, src_pud, |
1110 | addr, next)) | |
1da177e4 LT |
1111 | return -ENOMEM; |
1112 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
1113 | return 0; | |
1114 | } | |
1115 | ||
c78f4636 PX |
1116 | static inline int |
1117 | copy_p4d_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1118 | pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long addr, | |
1119 | unsigned long end) | |
c2febafc | 1120 | { |
c78f4636 | 1121 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
c2febafc KS |
1122 | p4d_t *src_p4d, *dst_p4d; |
1123 | unsigned long next; | |
1124 | ||
1125 | dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); | |
1126 | if (!dst_p4d) | |
1127 | return -ENOMEM; | |
1128 | src_p4d = p4d_offset(src_pgd, addr); | |
1129 | do { | |
1130 | next = p4d_addr_end(addr, end); | |
1131 | if (p4d_none_or_clear_bad(src_p4d)) | |
1132 | continue; | |
c78f4636 PX |
1133 | if (copy_pud_range(dst_vma, src_vma, dst_p4d, src_p4d, |
1134 | addr, next)) | |
c2febafc KS |
1135 | return -ENOMEM; |
1136 | } while (dst_p4d++, src_p4d++, addr = next, addr != end); | |
1137 | return 0; | |
1138 | } | |
1139 | ||
c78f4636 PX |
1140 | int |
1141 | copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) | |
1da177e4 LT |
1142 | { |
1143 | pgd_t *src_pgd, *dst_pgd; | |
1144 | unsigned long next; | |
c78f4636 PX |
1145 | unsigned long addr = src_vma->vm_start; |
1146 | unsigned long end = src_vma->vm_end; | |
1147 | struct mm_struct *dst_mm = dst_vma->vm_mm; | |
1148 | struct mm_struct *src_mm = src_vma->vm_mm; | |
ac46d4f3 | 1149 | struct mmu_notifier_range range; |
2ec74c3e | 1150 | bool is_cow; |
cddb8a5c | 1151 | int ret; |
1da177e4 | 1152 | |
d992895b NP |
1153 | /* |
1154 | * Don't copy ptes where a page fault will fill them correctly. | |
1155 | * Fork becomes much lighter when there are big shared or private | |
1156 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1157 | * efficient than faulting. | |
1158 | */ | |
c78f4636 PX |
1159 | if (!(src_vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
1160 | !src_vma->anon_vma) | |
0661a336 | 1161 | return 0; |
d992895b | 1162 | |
c78f4636 PX |
1163 | if (is_vm_hugetlb_page(src_vma)) |
1164 | return copy_hugetlb_page_range(dst_mm, src_mm, src_vma); | |
1da177e4 | 1165 | |
c78f4636 | 1166 | if (unlikely(src_vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1167 | /* |
1168 | * We do not free on error cases below as remove_vma | |
1169 | * gets called on error from higher level routine | |
1170 | */ | |
c78f4636 | 1171 | ret = track_pfn_copy(src_vma); |
2ab64037 | 1172 | if (ret) |
1173 | return ret; | |
1174 | } | |
1175 | ||
cddb8a5c AA |
1176 | /* |
1177 | * We need to invalidate the secondary MMU mappings only when | |
1178 | * there could be a permission downgrade on the ptes of the | |
1179 | * parent mm. And a permission downgrade will only happen if | |
1180 | * is_cow_mapping() returns true. | |
1181 | */ | |
c78f4636 | 1182 | is_cow = is_cow_mapping(src_vma->vm_flags); |
ac46d4f3 JG |
1183 | |
1184 | if (is_cow) { | |
7269f999 | 1185 | mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_PAGE, |
c78f4636 | 1186 | 0, src_vma, src_mm, addr, end); |
ac46d4f3 | 1187 | mmu_notifier_invalidate_range_start(&range); |
57efa1fe JG |
1188 | /* |
1189 | * Disabling preemption is not needed for the write side, as | |
1190 | * the read side doesn't spin, but goes to the mmap_lock. | |
1191 | * | |
1192 | * Use the raw variant of the seqcount_t write API to avoid | |
1193 | * lockdep complaining about preemptibility. | |
1194 | */ | |
1195 | mmap_assert_write_locked(src_mm); | |
1196 | raw_write_seqcount_begin(&src_mm->write_protect_seq); | |
ac46d4f3 | 1197 | } |
cddb8a5c AA |
1198 | |
1199 | ret = 0; | |
1da177e4 LT |
1200 | dst_pgd = pgd_offset(dst_mm, addr); |
1201 | src_pgd = pgd_offset(src_mm, addr); | |
1202 | do { | |
1203 | next = pgd_addr_end(addr, end); | |
1204 | if (pgd_none_or_clear_bad(src_pgd)) | |
1205 | continue; | |
c78f4636 PX |
1206 | if (unlikely(copy_p4d_range(dst_vma, src_vma, dst_pgd, src_pgd, |
1207 | addr, next))) { | |
cddb8a5c AA |
1208 | ret = -ENOMEM; |
1209 | break; | |
1210 | } | |
1da177e4 | 1211 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1212 | |
57efa1fe JG |
1213 | if (is_cow) { |
1214 | raw_write_seqcount_end(&src_mm->write_protect_seq); | |
ac46d4f3 | 1215 | mmu_notifier_invalidate_range_end(&range); |
57efa1fe | 1216 | } |
cddb8a5c | 1217 | return ret; |
1da177e4 LT |
1218 | } |
1219 | ||
51c6f666 | 1220 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1221 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1222 | unsigned long addr, unsigned long end, |
97a89413 | 1223 | struct zap_details *details) |
1da177e4 | 1224 | { |
b5810039 | 1225 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1226 | int force_flush = 0; |
d559db08 | 1227 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1228 | spinlock_t *ptl; |
5f1a1907 | 1229 | pte_t *start_pte; |
97a89413 | 1230 | pte_t *pte; |
8a5f14a2 | 1231 | swp_entry_t entry; |
d559db08 | 1232 | |
ed6a7935 | 1233 | tlb_change_page_size(tlb, PAGE_SIZE); |
d16dfc55 | 1234 | again: |
e303297e | 1235 | init_rss_vec(rss); |
5f1a1907 SR |
1236 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1237 | pte = start_pte; | |
3ea27719 | 1238 | flush_tlb_batched_pending(mm); |
6606c3e0 | 1239 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1240 | do { |
1241 | pte_t ptent = *pte; | |
166f61b9 | 1242 | if (pte_none(ptent)) |
1da177e4 | 1243 | continue; |
6f5e6b9e | 1244 | |
7b167b68 MK |
1245 | if (need_resched()) |
1246 | break; | |
1247 | ||
1da177e4 | 1248 | if (pte_present(ptent)) { |
ee498ed7 | 1249 | struct page *page; |
51c6f666 | 1250 | |
25b2995a | 1251 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
1252 | if (unlikely(details) && page) { |
1253 | /* | |
1254 | * unmap_shared_mapping_pages() wants to | |
1255 | * invalidate cache without truncating: | |
1256 | * unmap shared but keep private pages. | |
1257 | */ | |
1258 | if (details->check_mapping && | |
800d8c63 | 1259 | details->check_mapping != page_rmapping(page)) |
1da177e4 | 1260 | continue; |
1da177e4 | 1261 | } |
b5810039 | 1262 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1263 | tlb->fullmm); |
1da177e4 LT |
1264 | tlb_remove_tlb_entry(tlb, pte, addr); |
1265 | if (unlikely(!page)) | |
1266 | continue; | |
eca56ff9 JM |
1267 | |
1268 | if (!PageAnon(page)) { | |
1cf35d47 LT |
1269 | if (pte_dirty(ptent)) { |
1270 | force_flush = 1; | |
6237bcd9 | 1271 | set_page_dirty(page); |
1cf35d47 | 1272 | } |
4917e5d0 | 1273 | if (pte_young(ptent) && |
64363aad | 1274 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1275 | mark_page_accessed(page); |
6237bcd9 | 1276 | } |
eca56ff9 | 1277 | rss[mm_counter(page)]--; |
d281ee61 | 1278 | page_remove_rmap(page, false); |
3dc14741 HD |
1279 | if (unlikely(page_mapcount(page) < 0)) |
1280 | print_bad_pte(vma, addr, ptent, page); | |
e9d55e15 | 1281 | if (unlikely(__tlb_remove_page(tlb, page))) { |
1cf35d47 | 1282 | force_flush = 1; |
ce9ec37b | 1283 | addr += PAGE_SIZE; |
d16dfc55 | 1284 | break; |
1cf35d47 | 1285 | } |
1da177e4 LT |
1286 | continue; |
1287 | } | |
5042db43 JG |
1288 | |
1289 | entry = pte_to_swp_entry(ptent); | |
463b7a17 | 1290 | if (is_device_private_entry(entry)) { |
af5cdaf8 | 1291 | struct page *page = pfn_swap_entry_to_page(entry); |
5042db43 JG |
1292 | |
1293 | if (unlikely(details && details->check_mapping)) { | |
1294 | /* | |
1295 | * unmap_shared_mapping_pages() wants to | |
1296 | * invalidate cache without truncating: | |
1297 | * unmap shared but keep private pages. | |
1298 | */ | |
1299 | if (details->check_mapping != | |
1300 | page_rmapping(page)) | |
1301 | continue; | |
1302 | } | |
1303 | ||
1304 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
1305 | rss[mm_counter(page)]--; | |
1306 | page_remove_rmap(page, false); | |
1307 | put_page(page); | |
1308 | continue; | |
1309 | } | |
1310 | ||
3e8715fd KS |
1311 | /* If details->check_mapping, we leave swap entries. */ |
1312 | if (unlikely(details)) | |
1da177e4 | 1313 | continue; |
b084d435 | 1314 | |
8a5f14a2 KS |
1315 | if (!non_swap_entry(entry)) |
1316 | rss[MM_SWAPENTS]--; | |
1317 | else if (is_migration_entry(entry)) { | |
1318 | struct page *page; | |
9f9f1acd | 1319 | |
af5cdaf8 | 1320 | page = pfn_swap_entry_to_page(entry); |
eca56ff9 | 1321 | rss[mm_counter(page)]--; |
b084d435 | 1322 | } |
8a5f14a2 KS |
1323 | if (unlikely(!free_swap_and_cache(entry))) |
1324 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 1325 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1326 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1327 | |
d559db08 | 1328 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1329 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1330 | |
1cf35d47 | 1331 | /* Do the actual TLB flush before dropping ptl */ |
fb7332a9 | 1332 | if (force_flush) |
1cf35d47 | 1333 | tlb_flush_mmu_tlbonly(tlb); |
1cf35d47 LT |
1334 | pte_unmap_unlock(start_pte, ptl); |
1335 | ||
1336 | /* | |
1337 | * If we forced a TLB flush (either due to running out of | |
1338 | * batch buffers or because we needed to flush dirty TLB | |
1339 | * entries before releasing the ptl), free the batched | |
1340 | * memory too. Restart if we didn't do everything. | |
1341 | */ | |
1342 | if (force_flush) { | |
1343 | force_flush = 0; | |
fa0aafb8 | 1344 | tlb_flush_mmu(tlb); |
7b167b68 MK |
1345 | } |
1346 | ||
1347 | if (addr != end) { | |
1348 | cond_resched(); | |
1349 | goto again; | |
d16dfc55 PZ |
1350 | } |
1351 | ||
51c6f666 | 1352 | return addr; |
1da177e4 LT |
1353 | } |
1354 | ||
51c6f666 | 1355 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1356 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1357 | unsigned long addr, unsigned long end, |
97a89413 | 1358 | struct zap_details *details) |
1da177e4 LT |
1359 | { |
1360 | pmd_t *pmd; | |
1361 | unsigned long next; | |
1362 | ||
1363 | pmd = pmd_offset(pud, addr); | |
1364 | do { | |
1365 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 1366 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
53406ed1 | 1367 | if (next - addr != HPAGE_PMD_SIZE) |
fd60775a | 1368 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
53406ed1 | 1369 | else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1370 | goto next; |
71e3aac0 | 1371 | /* fall through */ |
22061a1f HD |
1372 | } else if (details && details->single_page && |
1373 | PageTransCompound(details->single_page) && | |
1374 | next - addr == HPAGE_PMD_SIZE && pmd_none(*pmd)) { | |
1375 | spinlock_t *ptl = pmd_lock(tlb->mm, pmd); | |
1376 | /* | |
1377 | * Take and drop THP pmd lock so that we cannot return | |
1378 | * prematurely, while zap_huge_pmd() has cleared *pmd, | |
1379 | * but not yet decremented compound_mapcount(). | |
1380 | */ | |
1381 | spin_unlock(ptl); | |
71e3aac0 | 1382 | } |
22061a1f | 1383 | |
1a5a9906 AA |
1384 | /* |
1385 | * Here there can be other concurrent MADV_DONTNEED or | |
1386 | * trans huge page faults running, and if the pmd is | |
1387 | * none or trans huge it can change under us. This is | |
c1e8d7c6 | 1388 | * because MADV_DONTNEED holds the mmap_lock in read |
1a5a9906 AA |
1389 | * mode. |
1390 | */ | |
1391 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1392 | goto next; | |
97a89413 | 1393 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1394 | next: |
97a89413 PZ |
1395 | cond_resched(); |
1396 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1397 | |
1398 | return addr; | |
1da177e4 LT |
1399 | } |
1400 | ||
51c6f666 | 1401 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
c2febafc | 1402 | struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1403 | unsigned long addr, unsigned long end, |
97a89413 | 1404 | struct zap_details *details) |
1da177e4 LT |
1405 | { |
1406 | pud_t *pud; | |
1407 | unsigned long next; | |
1408 | ||
c2febafc | 1409 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1410 | do { |
1411 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1412 | if (pud_trans_huge(*pud) || pud_devmap(*pud)) { |
1413 | if (next - addr != HPAGE_PUD_SIZE) { | |
42fc5414 | 1414 | mmap_assert_locked(tlb->mm); |
a00cc7d9 MW |
1415 | split_huge_pud(vma, pud, addr); |
1416 | } else if (zap_huge_pud(tlb, vma, pud, addr)) | |
1417 | goto next; | |
1418 | /* fall through */ | |
1419 | } | |
97a89413 | 1420 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1421 | continue; |
97a89413 | 1422 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
a00cc7d9 MW |
1423 | next: |
1424 | cond_resched(); | |
97a89413 | 1425 | } while (pud++, addr = next, addr != end); |
51c6f666 RH |
1426 | |
1427 | return addr; | |
1da177e4 LT |
1428 | } |
1429 | ||
c2febafc KS |
1430 | static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, |
1431 | struct vm_area_struct *vma, pgd_t *pgd, | |
1432 | unsigned long addr, unsigned long end, | |
1433 | struct zap_details *details) | |
1434 | { | |
1435 | p4d_t *p4d; | |
1436 | unsigned long next; | |
1437 | ||
1438 | p4d = p4d_offset(pgd, addr); | |
1439 | do { | |
1440 | next = p4d_addr_end(addr, end); | |
1441 | if (p4d_none_or_clear_bad(p4d)) | |
1442 | continue; | |
1443 | next = zap_pud_range(tlb, vma, p4d, addr, next, details); | |
1444 | } while (p4d++, addr = next, addr != end); | |
1445 | ||
1446 | return addr; | |
1447 | } | |
1448 | ||
aac45363 | 1449 | void unmap_page_range(struct mmu_gather *tlb, |
038c7aa1 AV |
1450 | struct vm_area_struct *vma, |
1451 | unsigned long addr, unsigned long end, | |
1452 | struct zap_details *details) | |
1da177e4 LT |
1453 | { |
1454 | pgd_t *pgd; | |
1455 | unsigned long next; | |
1456 | ||
1da177e4 LT |
1457 | BUG_ON(addr >= end); |
1458 | tlb_start_vma(tlb, vma); | |
1459 | pgd = pgd_offset(vma->vm_mm, addr); | |
1460 | do { | |
1461 | next = pgd_addr_end(addr, end); | |
97a89413 | 1462 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1463 | continue; |
c2febafc | 1464 | next = zap_p4d_range(tlb, vma, pgd, addr, next, details); |
97a89413 | 1465 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
1466 | tlb_end_vma(tlb, vma); |
1467 | } | |
51c6f666 | 1468 | |
f5cc4eef AV |
1469 | |
1470 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1471 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1472 | unsigned long end_addr, |
f5cc4eef AV |
1473 | struct zap_details *details) |
1474 | { | |
1475 | unsigned long start = max(vma->vm_start, start_addr); | |
1476 | unsigned long end; | |
1477 | ||
1478 | if (start >= vma->vm_end) | |
1479 | return; | |
1480 | end = min(vma->vm_end, end_addr); | |
1481 | if (end <= vma->vm_start) | |
1482 | return; | |
1483 | ||
cbc91f71 SD |
1484 | if (vma->vm_file) |
1485 | uprobe_munmap(vma, start, end); | |
1486 | ||
b3b9c293 | 1487 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1488 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1489 | |
1490 | if (start != end) { | |
1491 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1492 | /* | |
1493 | * It is undesirable to test vma->vm_file as it | |
1494 | * should be non-null for valid hugetlb area. | |
1495 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1496 | * cleanup path of mmap_region. When |
f5cc4eef | 1497 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1498 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1499 | * before calling this function to clean up. |
1500 | * Since no pte has actually been setup, it is | |
1501 | * safe to do nothing in this case. | |
1502 | */ | |
24669e58 | 1503 | if (vma->vm_file) { |
83cde9e8 | 1504 | i_mmap_lock_write(vma->vm_file->f_mapping); |
d833352a | 1505 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
83cde9e8 | 1506 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
24669e58 | 1507 | } |
f5cc4eef AV |
1508 | } else |
1509 | unmap_page_range(tlb, vma, start, end, details); | |
1510 | } | |
1da177e4 LT |
1511 | } |
1512 | ||
1da177e4 LT |
1513 | /** |
1514 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1515 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1516 | * @vma: the starting vma |
1517 | * @start_addr: virtual address at which to start unmapping | |
1518 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1519 | * |
508034a3 | 1520 | * Unmap all pages in the vma list. |
1da177e4 | 1521 | * |
1da177e4 LT |
1522 | * Only addresses between `start' and `end' will be unmapped. |
1523 | * | |
1524 | * The VMA list must be sorted in ascending virtual address order. | |
1525 | * | |
1526 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1527 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1528 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1529 | * drops the lock and schedules. | |
1530 | */ | |
6e8bb019 | 1531 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1532 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1533 | unsigned long end_addr) |
1da177e4 | 1534 | { |
ac46d4f3 | 1535 | struct mmu_notifier_range range; |
1da177e4 | 1536 | |
6f4f13e8 JG |
1537 | mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, vma->vm_mm, |
1538 | start_addr, end_addr); | |
ac46d4f3 | 1539 | mmu_notifier_invalidate_range_start(&range); |
f5cc4eef | 1540 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1541 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
ac46d4f3 | 1542 | mmu_notifier_invalidate_range_end(&range); |
1da177e4 LT |
1543 | } |
1544 | ||
1545 | /** | |
1546 | * zap_page_range - remove user pages in a given range | |
1547 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1548 | * @start: starting address of pages to zap |
1da177e4 | 1549 | * @size: number of bytes to zap |
f5cc4eef AV |
1550 | * |
1551 | * Caller must protect the VMA list | |
1da177e4 | 1552 | */ |
7e027b14 | 1553 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
ecf1385d | 1554 | unsigned long size) |
1da177e4 | 1555 | { |
ac46d4f3 | 1556 | struct mmu_notifier_range range; |
d16dfc55 | 1557 | struct mmu_gather tlb; |
1da177e4 | 1558 | |
1da177e4 | 1559 | lru_add_drain(); |
7269f999 | 1560 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, vma->vm_mm, |
6f4f13e8 | 1561 | start, start + size); |
a72afd87 | 1562 | tlb_gather_mmu(&tlb, vma->vm_mm); |
ac46d4f3 JG |
1563 | update_hiwater_rss(vma->vm_mm); |
1564 | mmu_notifier_invalidate_range_start(&range); | |
1565 | for ( ; vma && vma->vm_start < range.end; vma = vma->vm_next) | |
1566 | unmap_single_vma(&tlb, vma, start, range.end, NULL); | |
1567 | mmu_notifier_invalidate_range_end(&range); | |
ae8eba8b | 1568 | tlb_finish_mmu(&tlb); |
1da177e4 LT |
1569 | } |
1570 | ||
f5cc4eef AV |
1571 | /** |
1572 | * zap_page_range_single - remove user pages in a given range | |
1573 | * @vma: vm_area_struct holding the applicable pages | |
1574 | * @address: starting address of pages to zap | |
1575 | * @size: number of bytes to zap | |
8a5f14a2 | 1576 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1577 | * |
1578 | * The range must fit into one VMA. | |
1da177e4 | 1579 | */ |
f5cc4eef | 1580 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1581 | unsigned long size, struct zap_details *details) |
1582 | { | |
ac46d4f3 | 1583 | struct mmu_notifier_range range; |
d16dfc55 | 1584 | struct mmu_gather tlb; |
1da177e4 | 1585 | |
1da177e4 | 1586 | lru_add_drain(); |
7269f999 | 1587 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, vma->vm_mm, |
6f4f13e8 | 1588 | address, address + size); |
a72afd87 | 1589 | tlb_gather_mmu(&tlb, vma->vm_mm); |
ac46d4f3 JG |
1590 | update_hiwater_rss(vma->vm_mm); |
1591 | mmu_notifier_invalidate_range_start(&range); | |
1592 | unmap_single_vma(&tlb, vma, address, range.end, details); | |
1593 | mmu_notifier_invalidate_range_end(&range); | |
ae8eba8b | 1594 | tlb_finish_mmu(&tlb); |
1da177e4 LT |
1595 | } |
1596 | ||
c627f9cc JS |
1597 | /** |
1598 | * zap_vma_ptes - remove ptes mapping the vma | |
1599 | * @vma: vm_area_struct holding ptes to be zapped | |
1600 | * @address: starting address of pages to zap | |
1601 | * @size: number of bytes to zap | |
1602 | * | |
1603 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1604 | * | |
1605 | * The entire address range must be fully contained within the vma. | |
1606 | * | |
c627f9cc | 1607 | */ |
27d036e3 | 1608 | void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, |
c627f9cc JS |
1609 | unsigned long size) |
1610 | { | |
1611 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1612 | !(vma->vm_flags & VM_PFNMAP)) | |
27d036e3 LR |
1613 | return; |
1614 | ||
f5cc4eef | 1615 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1616 | } |
1617 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1618 | ||
8cd3984d | 1619 | static pmd_t *walk_to_pmd(struct mm_struct *mm, unsigned long addr) |
c9cfcddf | 1620 | { |
c2febafc KS |
1621 | pgd_t *pgd; |
1622 | p4d_t *p4d; | |
1623 | pud_t *pud; | |
1624 | pmd_t *pmd; | |
1625 | ||
1626 | pgd = pgd_offset(mm, addr); | |
1627 | p4d = p4d_alloc(mm, pgd, addr); | |
1628 | if (!p4d) | |
1629 | return NULL; | |
1630 | pud = pud_alloc(mm, p4d, addr); | |
1631 | if (!pud) | |
1632 | return NULL; | |
1633 | pmd = pmd_alloc(mm, pud, addr); | |
1634 | if (!pmd) | |
1635 | return NULL; | |
1636 | ||
1637 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
8cd3984d AR |
1638 | return pmd; |
1639 | } | |
1640 | ||
1641 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, | |
1642 | spinlock_t **ptl) | |
1643 | { | |
1644 | pmd_t *pmd = walk_to_pmd(mm, addr); | |
1645 | ||
1646 | if (!pmd) | |
1647 | return NULL; | |
c2febafc | 1648 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
c9cfcddf LT |
1649 | } |
1650 | ||
8efd6f5b AR |
1651 | static int validate_page_before_insert(struct page *page) |
1652 | { | |
1653 | if (PageAnon(page) || PageSlab(page) || page_has_type(page)) | |
1654 | return -EINVAL; | |
1655 | flush_dcache_page(page); | |
1656 | return 0; | |
1657 | } | |
1658 | ||
1659 | static int insert_page_into_pte_locked(struct mm_struct *mm, pte_t *pte, | |
1660 | unsigned long addr, struct page *page, pgprot_t prot) | |
1661 | { | |
1662 | if (!pte_none(*pte)) | |
1663 | return -EBUSY; | |
1664 | /* Ok, finally just insert the thing.. */ | |
1665 | get_page(page); | |
1666 | inc_mm_counter_fast(mm, mm_counter_file(page)); | |
1667 | page_add_file_rmap(page, false); | |
1668 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); | |
1669 | return 0; | |
1670 | } | |
1671 | ||
238f58d8 LT |
1672 | /* |
1673 | * This is the old fallback for page remapping. | |
1674 | * | |
1675 | * For historical reasons, it only allows reserved pages. Only | |
1676 | * old drivers should use this, and they needed to mark their | |
1677 | * pages reserved for the old functions anyway. | |
1678 | */ | |
423bad60 NP |
1679 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1680 | struct page *page, pgprot_t prot) | |
238f58d8 | 1681 | { |
423bad60 | 1682 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1683 | int retval; |
c9cfcddf | 1684 | pte_t *pte; |
8a9f3ccd BS |
1685 | spinlock_t *ptl; |
1686 | ||
8efd6f5b AR |
1687 | retval = validate_page_before_insert(page); |
1688 | if (retval) | |
5b4e655e | 1689 | goto out; |
238f58d8 | 1690 | retval = -ENOMEM; |
c9cfcddf | 1691 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1692 | if (!pte) |
5b4e655e | 1693 | goto out; |
8efd6f5b | 1694 | retval = insert_page_into_pte_locked(mm, pte, addr, page, prot); |
238f58d8 LT |
1695 | pte_unmap_unlock(pte, ptl); |
1696 | out: | |
1697 | return retval; | |
1698 | } | |
1699 | ||
8cd3984d | 1700 | #ifdef pte_index |
7f70c2a6 | 1701 | static int insert_page_in_batch_locked(struct mm_struct *mm, pte_t *pte, |
8cd3984d AR |
1702 | unsigned long addr, struct page *page, pgprot_t prot) |
1703 | { | |
1704 | int err; | |
1705 | ||
1706 | if (!page_count(page)) | |
1707 | return -EINVAL; | |
1708 | err = validate_page_before_insert(page); | |
7f70c2a6 AR |
1709 | if (err) |
1710 | return err; | |
1711 | return insert_page_into_pte_locked(mm, pte, addr, page, prot); | |
8cd3984d AR |
1712 | } |
1713 | ||
1714 | /* insert_pages() amortizes the cost of spinlock operations | |
1715 | * when inserting pages in a loop. Arch *must* define pte_index. | |
1716 | */ | |
1717 | static int insert_pages(struct vm_area_struct *vma, unsigned long addr, | |
1718 | struct page **pages, unsigned long *num, pgprot_t prot) | |
1719 | { | |
1720 | pmd_t *pmd = NULL; | |
7f70c2a6 AR |
1721 | pte_t *start_pte, *pte; |
1722 | spinlock_t *pte_lock; | |
8cd3984d AR |
1723 | struct mm_struct *const mm = vma->vm_mm; |
1724 | unsigned long curr_page_idx = 0; | |
1725 | unsigned long remaining_pages_total = *num; | |
1726 | unsigned long pages_to_write_in_pmd; | |
1727 | int ret; | |
1728 | more: | |
1729 | ret = -EFAULT; | |
1730 | pmd = walk_to_pmd(mm, addr); | |
1731 | if (!pmd) | |
1732 | goto out; | |
1733 | ||
1734 | pages_to_write_in_pmd = min_t(unsigned long, | |
1735 | remaining_pages_total, PTRS_PER_PTE - pte_index(addr)); | |
1736 | ||
1737 | /* Allocate the PTE if necessary; takes PMD lock once only. */ | |
1738 | ret = -ENOMEM; | |
1739 | if (pte_alloc(mm, pmd)) | |
1740 | goto out; | |
8cd3984d AR |
1741 | |
1742 | while (pages_to_write_in_pmd) { | |
1743 | int pte_idx = 0; | |
1744 | const int batch_size = min_t(int, pages_to_write_in_pmd, 8); | |
1745 | ||
7f70c2a6 AR |
1746 | start_pte = pte_offset_map_lock(mm, pmd, addr, &pte_lock); |
1747 | for (pte = start_pte; pte_idx < batch_size; ++pte, ++pte_idx) { | |
1748 | int err = insert_page_in_batch_locked(mm, pte, | |
8cd3984d AR |
1749 | addr, pages[curr_page_idx], prot); |
1750 | if (unlikely(err)) { | |
7f70c2a6 | 1751 | pte_unmap_unlock(start_pte, pte_lock); |
8cd3984d AR |
1752 | ret = err; |
1753 | remaining_pages_total -= pte_idx; | |
1754 | goto out; | |
1755 | } | |
1756 | addr += PAGE_SIZE; | |
1757 | ++curr_page_idx; | |
1758 | } | |
7f70c2a6 | 1759 | pte_unmap_unlock(start_pte, pte_lock); |
8cd3984d AR |
1760 | pages_to_write_in_pmd -= batch_size; |
1761 | remaining_pages_total -= batch_size; | |
1762 | } | |
1763 | if (remaining_pages_total) | |
1764 | goto more; | |
1765 | ret = 0; | |
1766 | out: | |
1767 | *num = remaining_pages_total; | |
1768 | return ret; | |
1769 | } | |
1770 | #endif /* ifdef pte_index */ | |
1771 | ||
1772 | /** | |
1773 | * vm_insert_pages - insert multiple pages into user vma, batching the pmd lock. | |
1774 | * @vma: user vma to map to | |
1775 | * @addr: target start user address of these pages | |
1776 | * @pages: source kernel pages | |
1777 | * @num: in: number of pages to map. out: number of pages that were *not* | |
1778 | * mapped. (0 means all pages were successfully mapped). | |
1779 | * | |
1780 | * Preferred over vm_insert_page() when inserting multiple pages. | |
1781 | * | |
1782 | * In case of error, we may have mapped a subset of the provided | |
1783 | * pages. It is the caller's responsibility to account for this case. | |
1784 | * | |
1785 | * The same restrictions apply as in vm_insert_page(). | |
1786 | */ | |
1787 | int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr, | |
1788 | struct page **pages, unsigned long *num) | |
1789 | { | |
1790 | #ifdef pte_index | |
1791 | const unsigned long end_addr = addr + (*num * PAGE_SIZE) - 1; | |
1792 | ||
1793 | if (addr < vma->vm_start || end_addr >= vma->vm_end) | |
1794 | return -EFAULT; | |
1795 | if (!(vma->vm_flags & VM_MIXEDMAP)) { | |
d8ed45c5 | 1796 | BUG_ON(mmap_read_trylock(vma->vm_mm)); |
8cd3984d AR |
1797 | BUG_ON(vma->vm_flags & VM_PFNMAP); |
1798 | vma->vm_flags |= VM_MIXEDMAP; | |
1799 | } | |
1800 | /* Defer page refcount checking till we're about to map that page. */ | |
1801 | return insert_pages(vma, addr, pages, num, vma->vm_page_prot); | |
1802 | #else | |
1803 | unsigned long idx = 0, pgcount = *num; | |
45779b03 | 1804 | int err = -EINVAL; |
8cd3984d AR |
1805 | |
1806 | for (; idx < pgcount; ++idx) { | |
1807 | err = vm_insert_page(vma, addr + (PAGE_SIZE * idx), pages[idx]); | |
1808 | if (err) | |
1809 | break; | |
1810 | } | |
1811 | *num = pgcount - idx; | |
1812 | return err; | |
1813 | #endif /* ifdef pte_index */ | |
1814 | } | |
1815 | EXPORT_SYMBOL(vm_insert_pages); | |
1816 | ||
bfa5bf6d REB |
1817 | /** |
1818 | * vm_insert_page - insert single page into user vma | |
1819 | * @vma: user vma to map to | |
1820 | * @addr: target user address of this page | |
1821 | * @page: source kernel page | |
1822 | * | |
a145dd41 LT |
1823 | * This allows drivers to insert individual pages they've allocated |
1824 | * into a user vma. | |
1825 | * | |
1826 | * The page has to be a nice clean _individual_ kernel allocation. | |
1827 | * If you allocate a compound page, you need to have marked it as | |
1828 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1829 | * (see split_page()). |
a145dd41 LT |
1830 | * |
1831 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1832 | * took an arbitrary page protection parameter. This doesn't allow | |
1833 | * that. Your vma protection will have to be set up correctly, which | |
1834 | * means that if you want a shared writable mapping, you'd better | |
1835 | * ask for a shared writable mapping! | |
1836 | * | |
1837 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1838 | * |
1839 | * Usually this function is called from f_op->mmap() handler | |
c1e8d7c6 | 1840 | * under mm->mmap_lock write-lock, so it can change vma->vm_flags. |
4b6e1e37 KK |
1841 | * Caller must set VM_MIXEDMAP on vma if it wants to call this |
1842 | * function from other places, for example from page-fault handler. | |
a862f68a MR |
1843 | * |
1844 | * Return: %0 on success, negative error code otherwise. | |
a145dd41 | 1845 | */ |
423bad60 NP |
1846 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1847 | struct page *page) | |
a145dd41 LT |
1848 | { |
1849 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1850 | return -EFAULT; | |
1851 | if (!page_count(page)) | |
1852 | return -EINVAL; | |
4b6e1e37 | 1853 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
d8ed45c5 | 1854 | BUG_ON(mmap_read_trylock(vma->vm_mm)); |
4b6e1e37 KK |
1855 | BUG_ON(vma->vm_flags & VM_PFNMAP); |
1856 | vma->vm_flags |= VM_MIXEDMAP; | |
1857 | } | |
423bad60 | 1858 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1859 | } |
e3c3374f | 1860 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1861 | |
a667d745 SJ |
1862 | /* |
1863 | * __vm_map_pages - maps range of kernel pages into user vma | |
1864 | * @vma: user vma to map to | |
1865 | * @pages: pointer to array of source kernel pages | |
1866 | * @num: number of pages in page array | |
1867 | * @offset: user's requested vm_pgoff | |
1868 | * | |
1869 | * This allows drivers to map range of kernel pages into a user vma. | |
1870 | * | |
1871 | * Return: 0 on success and error code otherwise. | |
1872 | */ | |
1873 | static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages, | |
1874 | unsigned long num, unsigned long offset) | |
1875 | { | |
1876 | unsigned long count = vma_pages(vma); | |
1877 | unsigned long uaddr = vma->vm_start; | |
1878 | int ret, i; | |
1879 | ||
1880 | /* Fail if the user requested offset is beyond the end of the object */ | |
96756fcb | 1881 | if (offset >= num) |
a667d745 SJ |
1882 | return -ENXIO; |
1883 | ||
1884 | /* Fail if the user requested size exceeds available object size */ | |
1885 | if (count > num - offset) | |
1886 | return -ENXIO; | |
1887 | ||
1888 | for (i = 0; i < count; i++) { | |
1889 | ret = vm_insert_page(vma, uaddr, pages[offset + i]); | |
1890 | if (ret < 0) | |
1891 | return ret; | |
1892 | uaddr += PAGE_SIZE; | |
1893 | } | |
1894 | ||
1895 | return 0; | |
1896 | } | |
1897 | ||
1898 | /** | |
1899 | * vm_map_pages - maps range of kernel pages starts with non zero offset | |
1900 | * @vma: user vma to map to | |
1901 | * @pages: pointer to array of source kernel pages | |
1902 | * @num: number of pages in page array | |
1903 | * | |
1904 | * Maps an object consisting of @num pages, catering for the user's | |
1905 | * requested vm_pgoff | |
1906 | * | |
1907 | * If we fail to insert any page into the vma, the function will return | |
1908 | * immediately leaving any previously inserted pages present. Callers | |
1909 | * from the mmap handler may immediately return the error as their caller | |
1910 | * will destroy the vma, removing any successfully inserted pages. Other | |
1911 | * callers should make their own arrangements for calling unmap_region(). | |
1912 | * | |
1913 | * Context: Process context. Called by mmap handlers. | |
1914 | * Return: 0 on success and error code otherwise. | |
1915 | */ | |
1916 | int vm_map_pages(struct vm_area_struct *vma, struct page **pages, | |
1917 | unsigned long num) | |
1918 | { | |
1919 | return __vm_map_pages(vma, pages, num, vma->vm_pgoff); | |
1920 | } | |
1921 | EXPORT_SYMBOL(vm_map_pages); | |
1922 | ||
1923 | /** | |
1924 | * vm_map_pages_zero - map range of kernel pages starts with zero offset | |
1925 | * @vma: user vma to map to | |
1926 | * @pages: pointer to array of source kernel pages | |
1927 | * @num: number of pages in page array | |
1928 | * | |
1929 | * Similar to vm_map_pages(), except that it explicitly sets the offset | |
1930 | * to 0. This function is intended for the drivers that did not consider | |
1931 | * vm_pgoff. | |
1932 | * | |
1933 | * Context: Process context. Called by mmap handlers. | |
1934 | * Return: 0 on success and error code otherwise. | |
1935 | */ | |
1936 | int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, | |
1937 | unsigned long num) | |
1938 | { | |
1939 | return __vm_map_pages(vma, pages, num, 0); | |
1940 | } | |
1941 | EXPORT_SYMBOL(vm_map_pages_zero); | |
1942 | ||
9b5a8e00 | 1943 | static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
b2770da6 | 1944 | pfn_t pfn, pgprot_t prot, bool mkwrite) |
423bad60 NP |
1945 | { |
1946 | struct mm_struct *mm = vma->vm_mm; | |
423bad60 NP |
1947 | pte_t *pte, entry; |
1948 | spinlock_t *ptl; | |
1949 | ||
423bad60 NP |
1950 | pte = get_locked_pte(mm, addr, &ptl); |
1951 | if (!pte) | |
9b5a8e00 | 1952 | return VM_FAULT_OOM; |
b2770da6 RZ |
1953 | if (!pte_none(*pte)) { |
1954 | if (mkwrite) { | |
1955 | /* | |
1956 | * For read faults on private mappings the PFN passed | |
1957 | * in may not match the PFN we have mapped if the | |
1958 | * mapped PFN is a writeable COW page. In the mkwrite | |
1959 | * case we are creating a writable PTE for a shared | |
f2c57d91 JK |
1960 | * mapping and we expect the PFNs to match. If they |
1961 | * don't match, we are likely racing with block | |
1962 | * allocation and mapping invalidation so just skip the | |
1963 | * update. | |
b2770da6 | 1964 | */ |
f2c57d91 JK |
1965 | if (pte_pfn(*pte) != pfn_t_to_pfn(pfn)) { |
1966 | WARN_ON_ONCE(!is_zero_pfn(pte_pfn(*pte))); | |
b2770da6 | 1967 | goto out_unlock; |
f2c57d91 | 1968 | } |
cae85cb8 JK |
1969 | entry = pte_mkyoung(*pte); |
1970 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1971 | if (ptep_set_access_flags(vma, addr, pte, entry, 1)) | |
1972 | update_mmu_cache(vma, addr, pte); | |
1973 | } | |
1974 | goto out_unlock; | |
b2770da6 | 1975 | } |
423bad60 NP |
1976 | |
1977 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
1978 | if (pfn_t_devmap(pfn)) |
1979 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
1980 | else | |
1981 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
b2770da6 | 1982 | |
b2770da6 RZ |
1983 | if (mkwrite) { |
1984 | entry = pte_mkyoung(entry); | |
1985 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1986 | } | |
1987 | ||
423bad60 | 1988 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 1989 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 | 1990 | |
423bad60 NP |
1991 | out_unlock: |
1992 | pte_unmap_unlock(pte, ptl); | |
9b5a8e00 | 1993 | return VM_FAULT_NOPAGE; |
423bad60 NP |
1994 | } |
1995 | ||
f5e6d1d5 MW |
1996 | /** |
1997 | * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
1998 | * @vma: user vma to map to | |
1999 | * @addr: target user address of this page | |
2000 | * @pfn: source kernel pfn | |
2001 | * @pgprot: pgprot flags for the inserted page | |
2002 | * | |
a1a0aea5 | 2003 | * This is exactly like vmf_insert_pfn(), except that it allows drivers |
f5e6d1d5 MW |
2004 | * to override pgprot on a per-page basis. |
2005 | * | |
2006 | * This only makes sense for IO mappings, and it makes no sense for | |
2007 | * COW mappings. In general, using multiple vmas is preferable; | |
ae2b01f3 | 2008 | * vmf_insert_pfn_prot should only be used if using multiple VMAs is |
f5e6d1d5 MW |
2009 | * impractical. |
2010 | * | |
574c5b3d TH |
2011 | * See vmf_insert_mixed_prot() for a discussion of the implication of using |
2012 | * a value of @pgprot different from that of @vma->vm_page_prot. | |
2013 | * | |
ae2b01f3 | 2014 | * Context: Process context. May allocate using %GFP_KERNEL. |
f5e6d1d5 MW |
2015 | * Return: vm_fault_t value. |
2016 | */ | |
2017 | vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
2018 | unsigned long pfn, pgprot_t pgprot) | |
2019 | { | |
6d958546 MW |
2020 | /* |
2021 | * Technically, architectures with pte_special can avoid all these | |
2022 | * restrictions (same for remap_pfn_range). However we would like | |
2023 | * consistency in testing and feature parity among all, so we should | |
2024 | * try to keep these invariants in place for everybody. | |
2025 | */ | |
2026 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); | |
2027 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
2028 | (VM_PFNMAP|VM_MIXEDMAP)); | |
2029 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
2030 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
2031 | ||
2032 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
2033 | return VM_FAULT_SIGBUS; | |
2034 | ||
2035 | if (!pfn_modify_allowed(pfn, pgprot)) | |
2036 | return VM_FAULT_SIGBUS; | |
2037 | ||
2038 | track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); | |
2039 | ||
9b5a8e00 | 2040 | return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, |
6d958546 | 2041 | false); |
f5e6d1d5 MW |
2042 | } |
2043 | EXPORT_SYMBOL(vmf_insert_pfn_prot); | |
e0dc0d8f | 2044 | |
ae2b01f3 MW |
2045 | /** |
2046 | * vmf_insert_pfn - insert single pfn into user vma | |
2047 | * @vma: user vma to map to | |
2048 | * @addr: target user address of this page | |
2049 | * @pfn: source kernel pfn | |
2050 | * | |
2051 | * Similar to vm_insert_page, this allows drivers to insert individual pages | |
2052 | * they've allocated into a user vma. Same comments apply. | |
2053 | * | |
2054 | * This function should only be called from a vm_ops->fault handler, and | |
2055 | * in that case the handler should return the result of this function. | |
2056 | * | |
2057 | * vma cannot be a COW mapping. | |
2058 | * | |
2059 | * As this is called only for pages that do not currently exist, we | |
2060 | * do not need to flush old virtual caches or the TLB. | |
2061 | * | |
2062 | * Context: Process context. May allocate using %GFP_KERNEL. | |
2063 | * Return: vm_fault_t value. | |
2064 | */ | |
2065 | vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
2066 | unsigned long pfn) | |
2067 | { | |
2068 | return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
2069 | } | |
2070 | EXPORT_SYMBOL(vmf_insert_pfn); | |
2071 | ||
785a3fab DW |
2072 | static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) |
2073 | { | |
2074 | /* these checks mirror the abort conditions in vm_normal_page */ | |
2075 | if (vma->vm_flags & VM_MIXEDMAP) | |
2076 | return true; | |
2077 | if (pfn_t_devmap(pfn)) | |
2078 | return true; | |
2079 | if (pfn_t_special(pfn)) | |
2080 | return true; | |
2081 | if (is_zero_pfn(pfn_t_to_pfn(pfn))) | |
2082 | return true; | |
2083 | return false; | |
2084 | } | |
2085 | ||
79f3aa5b | 2086 | static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma, |
574c5b3d TH |
2087 | unsigned long addr, pfn_t pfn, pgprot_t pgprot, |
2088 | bool mkwrite) | |
423bad60 | 2089 | { |
79f3aa5b | 2090 | int err; |
87744ab3 | 2091 | |
785a3fab | 2092 | BUG_ON(!vm_mixed_ok(vma, pfn)); |
e0dc0d8f | 2093 | |
423bad60 | 2094 | if (addr < vma->vm_start || addr >= vma->vm_end) |
79f3aa5b | 2095 | return VM_FAULT_SIGBUS; |
308a047c BP |
2096 | |
2097 | track_pfn_insert(vma, &pgprot, pfn); | |
e0dc0d8f | 2098 | |
42e4089c | 2099 | if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) |
79f3aa5b | 2100 | return VM_FAULT_SIGBUS; |
42e4089c | 2101 | |
423bad60 NP |
2102 | /* |
2103 | * If we don't have pte special, then we have to use the pfn_valid() | |
2104 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
2105 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
2106 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
2107 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 2108 | */ |
00b3a331 LD |
2109 | if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && |
2110 | !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { | |
423bad60 NP |
2111 | struct page *page; |
2112 | ||
03fc2da6 DW |
2113 | /* |
2114 | * At this point we are committed to insert_page() | |
2115 | * regardless of whether the caller specified flags that | |
2116 | * result in pfn_t_has_page() == false. | |
2117 | */ | |
2118 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
79f3aa5b MW |
2119 | err = insert_page(vma, addr, page, pgprot); |
2120 | } else { | |
9b5a8e00 | 2121 | return insert_pfn(vma, addr, pfn, pgprot, mkwrite); |
423bad60 | 2122 | } |
b2770da6 | 2123 | |
5d747637 MW |
2124 | if (err == -ENOMEM) |
2125 | return VM_FAULT_OOM; | |
2126 | if (err < 0 && err != -EBUSY) | |
2127 | return VM_FAULT_SIGBUS; | |
2128 | ||
2129 | return VM_FAULT_NOPAGE; | |
e0dc0d8f | 2130 | } |
79f3aa5b | 2131 | |
574c5b3d TH |
2132 | /** |
2133 | * vmf_insert_mixed_prot - insert single pfn into user vma with specified pgprot | |
2134 | * @vma: user vma to map to | |
2135 | * @addr: target user address of this page | |
2136 | * @pfn: source kernel pfn | |
2137 | * @pgprot: pgprot flags for the inserted page | |
2138 | * | |
a1a0aea5 | 2139 | * This is exactly like vmf_insert_mixed(), except that it allows drivers |
574c5b3d TH |
2140 | * to override pgprot on a per-page basis. |
2141 | * | |
2142 | * Typically this function should be used by drivers to set caching- and | |
2143 | * encryption bits different than those of @vma->vm_page_prot, because | |
2144 | * the caching- or encryption mode may not be known at mmap() time. | |
2145 | * This is ok as long as @vma->vm_page_prot is not used by the core vm | |
2146 | * to set caching and encryption bits for those vmas (except for COW pages). | |
2147 | * This is ensured by core vm only modifying these page table entries using | |
2148 | * functions that don't touch caching- or encryption bits, using pte_modify() | |
2149 | * if needed. (See for example mprotect()). | |
2150 | * Also when new page-table entries are created, this is only done using the | |
2151 | * fault() callback, and never using the value of vma->vm_page_prot, | |
2152 | * except for page-table entries that point to anonymous pages as the result | |
2153 | * of COW. | |
2154 | * | |
2155 | * Context: Process context. May allocate using %GFP_KERNEL. | |
2156 | * Return: vm_fault_t value. | |
2157 | */ | |
2158 | vm_fault_t vmf_insert_mixed_prot(struct vm_area_struct *vma, unsigned long addr, | |
2159 | pfn_t pfn, pgprot_t pgprot) | |
2160 | { | |
2161 | return __vm_insert_mixed(vma, addr, pfn, pgprot, false); | |
2162 | } | |
5379e4dd | 2163 | EXPORT_SYMBOL(vmf_insert_mixed_prot); |
574c5b3d | 2164 | |
79f3aa5b MW |
2165 | vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
2166 | pfn_t pfn) | |
2167 | { | |
574c5b3d | 2168 | return __vm_insert_mixed(vma, addr, pfn, vma->vm_page_prot, false); |
79f3aa5b | 2169 | } |
5d747637 | 2170 | EXPORT_SYMBOL(vmf_insert_mixed); |
e0dc0d8f | 2171 | |
ab77dab4 SJ |
2172 | /* |
2173 | * If the insertion of PTE failed because someone else already added a | |
2174 | * different entry in the mean time, we treat that as success as we assume | |
2175 | * the same entry was actually inserted. | |
2176 | */ | |
ab77dab4 SJ |
2177 | vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, |
2178 | unsigned long addr, pfn_t pfn) | |
b2770da6 | 2179 | { |
574c5b3d | 2180 | return __vm_insert_mixed(vma, addr, pfn, vma->vm_page_prot, true); |
b2770da6 | 2181 | } |
ab77dab4 | 2182 | EXPORT_SYMBOL(vmf_insert_mixed_mkwrite); |
b2770da6 | 2183 | |
1da177e4 LT |
2184 | /* |
2185 | * maps a range of physical memory into the requested pages. the old | |
2186 | * mappings are removed. any references to nonexistent pages results | |
2187 | * in null mappings (currently treated as "copy-on-access") | |
2188 | */ | |
2189 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
2190 | unsigned long addr, unsigned long end, | |
2191 | unsigned long pfn, pgprot_t prot) | |
2192 | { | |
90a3e375 | 2193 | pte_t *pte, *mapped_pte; |
c74df32c | 2194 | spinlock_t *ptl; |
42e4089c | 2195 | int err = 0; |
1da177e4 | 2196 | |
90a3e375 | 2197 | mapped_pte = pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
2198 | if (!pte) |
2199 | return -ENOMEM; | |
6606c3e0 | 2200 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
2201 | do { |
2202 | BUG_ON(!pte_none(*pte)); | |
42e4089c AK |
2203 | if (!pfn_modify_allowed(pfn, prot)) { |
2204 | err = -EACCES; | |
2205 | break; | |
2206 | } | |
7e675137 | 2207 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
2208 | pfn++; |
2209 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 2210 | arch_leave_lazy_mmu_mode(); |
90a3e375 | 2211 | pte_unmap_unlock(mapped_pte, ptl); |
42e4089c | 2212 | return err; |
1da177e4 LT |
2213 | } |
2214 | ||
2215 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2216 | unsigned long addr, unsigned long end, | |
2217 | unsigned long pfn, pgprot_t prot) | |
2218 | { | |
2219 | pmd_t *pmd; | |
2220 | unsigned long next; | |
42e4089c | 2221 | int err; |
1da177e4 LT |
2222 | |
2223 | pfn -= addr >> PAGE_SHIFT; | |
2224 | pmd = pmd_alloc(mm, pud, addr); | |
2225 | if (!pmd) | |
2226 | return -ENOMEM; | |
f66055ab | 2227 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
2228 | do { |
2229 | next = pmd_addr_end(addr, end); | |
42e4089c AK |
2230 | err = remap_pte_range(mm, pmd, addr, next, |
2231 | pfn + (addr >> PAGE_SHIFT), prot); | |
2232 | if (err) | |
2233 | return err; | |
1da177e4 LT |
2234 | } while (pmd++, addr = next, addr != end); |
2235 | return 0; | |
2236 | } | |
2237 | ||
c2febafc | 2238 | static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, |
1da177e4 LT |
2239 | unsigned long addr, unsigned long end, |
2240 | unsigned long pfn, pgprot_t prot) | |
2241 | { | |
2242 | pud_t *pud; | |
2243 | unsigned long next; | |
42e4089c | 2244 | int err; |
1da177e4 LT |
2245 | |
2246 | pfn -= addr >> PAGE_SHIFT; | |
c2febafc | 2247 | pud = pud_alloc(mm, p4d, addr); |
1da177e4 LT |
2248 | if (!pud) |
2249 | return -ENOMEM; | |
2250 | do { | |
2251 | next = pud_addr_end(addr, end); | |
42e4089c AK |
2252 | err = remap_pmd_range(mm, pud, addr, next, |
2253 | pfn + (addr >> PAGE_SHIFT), prot); | |
2254 | if (err) | |
2255 | return err; | |
1da177e4 LT |
2256 | } while (pud++, addr = next, addr != end); |
2257 | return 0; | |
2258 | } | |
2259 | ||
c2febafc KS |
2260 | static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2261 | unsigned long addr, unsigned long end, | |
2262 | unsigned long pfn, pgprot_t prot) | |
2263 | { | |
2264 | p4d_t *p4d; | |
2265 | unsigned long next; | |
42e4089c | 2266 | int err; |
c2febafc KS |
2267 | |
2268 | pfn -= addr >> PAGE_SHIFT; | |
2269 | p4d = p4d_alloc(mm, pgd, addr); | |
2270 | if (!p4d) | |
2271 | return -ENOMEM; | |
2272 | do { | |
2273 | next = p4d_addr_end(addr, end); | |
42e4089c AK |
2274 | err = remap_pud_range(mm, p4d, addr, next, |
2275 | pfn + (addr >> PAGE_SHIFT), prot); | |
2276 | if (err) | |
2277 | return err; | |
c2febafc KS |
2278 | } while (p4d++, addr = next, addr != end); |
2279 | return 0; | |
2280 | } | |
2281 | ||
74ffa5a3 CH |
2282 | /* |
2283 | * Variant of remap_pfn_range that does not call track_pfn_remap. The caller | |
2284 | * must have pre-validated the caching bits of the pgprot_t. | |
bfa5bf6d | 2285 | */ |
74ffa5a3 CH |
2286 | int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr, |
2287 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1da177e4 LT |
2288 | { |
2289 | pgd_t *pgd; | |
2290 | unsigned long next; | |
2d15cab8 | 2291 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 LT |
2292 | struct mm_struct *mm = vma->vm_mm; |
2293 | int err; | |
2294 | ||
0c4123e3 AZ |
2295 | if (WARN_ON_ONCE(!PAGE_ALIGNED(addr))) |
2296 | return -EINVAL; | |
2297 | ||
1da177e4 LT |
2298 | /* |
2299 | * Physically remapped pages are special. Tell the | |
2300 | * rest of the world about it: | |
2301 | * VM_IO tells people not to look at these pages | |
2302 | * (accesses can have side effects). | |
6aab341e LT |
2303 | * VM_PFNMAP tells the core MM that the base pages are just |
2304 | * raw PFN mappings, and do not have a "struct page" associated | |
2305 | * with them. | |
314e51b9 KK |
2306 | * VM_DONTEXPAND |
2307 | * Disable vma merging and expanding with mremap(). | |
2308 | * VM_DONTDUMP | |
2309 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
2310 | * |
2311 | * There's a horrible special case to handle copy-on-write | |
2312 | * behaviour that some programs depend on. We mark the "original" | |
2313 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 2314 | * See vm_normal_page() for details. |
1da177e4 | 2315 | */ |
b3b9c293 KK |
2316 | if (is_cow_mapping(vma->vm_flags)) { |
2317 | if (addr != vma->vm_start || end != vma->vm_end) | |
2318 | return -EINVAL; | |
fb155c16 | 2319 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
2320 | } |
2321 | ||
314e51b9 | 2322 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
2323 | |
2324 | BUG_ON(addr >= end); | |
2325 | pfn -= addr >> PAGE_SHIFT; | |
2326 | pgd = pgd_offset(mm, addr); | |
2327 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
2328 | do { |
2329 | next = pgd_addr_end(addr, end); | |
c2febafc | 2330 | err = remap_p4d_range(mm, pgd, addr, next, |
1da177e4 LT |
2331 | pfn + (addr >> PAGE_SHIFT), prot); |
2332 | if (err) | |
74ffa5a3 | 2333 | return err; |
1da177e4 | 2334 | } while (pgd++, addr = next, addr != end); |
2ab64037 | 2335 | |
74ffa5a3 CH |
2336 | return 0; |
2337 | } | |
2338 | ||
2339 | /** | |
2340 | * remap_pfn_range - remap kernel memory to userspace | |
2341 | * @vma: user vma to map to | |
2342 | * @addr: target page aligned user address to start at | |
2343 | * @pfn: page frame number of kernel physical memory address | |
2344 | * @size: size of mapping area | |
2345 | * @prot: page protection flags for this mapping | |
2346 | * | |
2347 | * Note: this is only safe if the mm semaphore is held when called. | |
2348 | * | |
2349 | * Return: %0 on success, negative error code otherwise. | |
2350 | */ | |
2351 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, | |
2352 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
2353 | { | |
2354 | int err; | |
2355 | ||
2356 | err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); | |
2ab64037 | 2357 | if (err) |
74ffa5a3 | 2358 | return -EINVAL; |
2ab64037 | 2359 | |
74ffa5a3 CH |
2360 | err = remap_pfn_range_notrack(vma, addr, pfn, size, prot); |
2361 | if (err) | |
2362 | untrack_pfn(vma, pfn, PAGE_ALIGN(size)); | |
1da177e4 LT |
2363 | return err; |
2364 | } | |
2365 | EXPORT_SYMBOL(remap_pfn_range); | |
2366 | ||
b4cbb197 LT |
2367 | /** |
2368 | * vm_iomap_memory - remap memory to userspace | |
2369 | * @vma: user vma to map to | |
abd69b9e | 2370 | * @start: start of the physical memory to be mapped |
b4cbb197 LT |
2371 | * @len: size of area |
2372 | * | |
2373 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
2374 | * driver just needs to give us the physical memory range to be mapped, | |
2375 | * we'll figure out the rest from the vma information. | |
2376 | * | |
2377 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
2378 | * whatever write-combining details or similar. | |
a862f68a MR |
2379 | * |
2380 | * Return: %0 on success, negative error code otherwise. | |
b4cbb197 LT |
2381 | */ |
2382 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
2383 | { | |
2384 | unsigned long vm_len, pfn, pages; | |
2385 | ||
2386 | /* Check that the physical memory area passed in looks valid */ | |
2387 | if (start + len < start) | |
2388 | return -EINVAL; | |
2389 | /* | |
2390 | * You *really* shouldn't map things that aren't page-aligned, | |
2391 | * but we've historically allowed it because IO memory might | |
2392 | * just have smaller alignment. | |
2393 | */ | |
2394 | len += start & ~PAGE_MASK; | |
2395 | pfn = start >> PAGE_SHIFT; | |
2396 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
2397 | if (pfn + pages < pfn) | |
2398 | return -EINVAL; | |
2399 | ||
2400 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
2401 | if (vma->vm_pgoff > pages) | |
2402 | return -EINVAL; | |
2403 | pfn += vma->vm_pgoff; | |
2404 | pages -= vma->vm_pgoff; | |
2405 | ||
2406 | /* Can we fit all of the mapping? */ | |
2407 | vm_len = vma->vm_end - vma->vm_start; | |
2408 | if (vm_len >> PAGE_SHIFT > pages) | |
2409 | return -EINVAL; | |
2410 | ||
2411 | /* Ok, let it rip */ | |
2412 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
2413 | } | |
2414 | EXPORT_SYMBOL(vm_iomap_memory); | |
2415 | ||
aee16b3c JF |
2416 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
2417 | unsigned long addr, unsigned long end, | |
e80d3909 JR |
2418 | pte_fn_t fn, void *data, bool create, |
2419 | pgtbl_mod_mask *mask) | |
aee16b3c | 2420 | { |
8abb50c7 | 2421 | pte_t *pte, *mapped_pte; |
be1db475 | 2422 | int err = 0; |
3f649ab7 | 2423 | spinlock_t *ptl; |
aee16b3c | 2424 | |
be1db475 | 2425 | if (create) { |
8abb50c7 | 2426 | mapped_pte = pte = (mm == &init_mm) ? |
e80d3909 | 2427 | pte_alloc_kernel_track(pmd, addr, mask) : |
be1db475 DA |
2428 | pte_alloc_map_lock(mm, pmd, addr, &ptl); |
2429 | if (!pte) | |
2430 | return -ENOMEM; | |
2431 | } else { | |
8abb50c7 | 2432 | mapped_pte = pte = (mm == &init_mm) ? |
be1db475 DA |
2433 | pte_offset_kernel(pmd, addr) : |
2434 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
2435 | } | |
aee16b3c JF |
2436 | |
2437 | BUG_ON(pmd_huge(*pmd)); | |
2438 | ||
38e0edb1 JF |
2439 | arch_enter_lazy_mmu_mode(); |
2440 | ||
eeb4a05f CH |
2441 | if (fn) { |
2442 | do { | |
2443 | if (create || !pte_none(*pte)) { | |
2444 | err = fn(pte++, addr, data); | |
2445 | if (err) | |
2446 | break; | |
2447 | } | |
2448 | } while (addr += PAGE_SIZE, addr != end); | |
2449 | } | |
e80d3909 | 2450 | *mask |= PGTBL_PTE_MODIFIED; |
aee16b3c | 2451 | |
38e0edb1 JF |
2452 | arch_leave_lazy_mmu_mode(); |
2453 | ||
aee16b3c | 2454 | if (mm != &init_mm) |
8abb50c7 | 2455 | pte_unmap_unlock(mapped_pte, ptl); |
aee16b3c JF |
2456 | return err; |
2457 | } | |
2458 | ||
2459 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2460 | unsigned long addr, unsigned long end, | |
e80d3909 JR |
2461 | pte_fn_t fn, void *data, bool create, |
2462 | pgtbl_mod_mask *mask) | |
aee16b3c JF |
2463 | { |
2464 | pmd_t *pmd; | |
2465 | unsigned long next; | |
be1db475 | 2466 | int err = 0; |
aee16b3c | 2467 | |
ceb86879 AK |
2468 | BUG_ON(pud_huge(*pud)); |
2469 | ||
be1db475 | 2470 | if (create) { |
e80d3909 | 2471 | pmd = pmd_alloc_track(mm, pud, addr, mask); |
be1db475 DA |
2472 | if (!pmd) |
2473 | return -ENOMEM; | |
2474 | } else { | |
2475 | pmd = pmd_offset(pud, addr); | |
2476 | } | |
aee16b3c JF |
2477 | do { |
2478 | next = pmd_addr_end(addr, end); | |
0c95cba4 NP |
2479 | if (pmd_none(*pmd) && !create) |
2480 | continue; | |
2481 | if (WARN_ON_ONCE(pmd_leaf(*pmd))) | |
2482 | return -EINVAL; | |
2483 | if (!pmd_none(*pmd) && WARN_ON_ONCE(pmd_bad(*pmd))) { | |
2484 | if (!create) | |
2485 | continue; | |
2486 | pmd_clear_bad(pmd); | |
be1db475 | 2487 | } |
0c95cba4 NP |
2488 | err = apply_to_pte_range(mm, pmd, addr, next, |
2489 | fn, data, create, mask); | |
2490 | if (err) | |
2491 | break; | |
aee16b3c | 2492 | } while (pmd++, addr = next, addr != end); |
0c95cba4 | 2493 | |
aee16b3c JF |
2494 | return err; |
2495 | } | |
2496 | ||
c2febafc | 2497 | static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, |
aee16b3c | 2498 | unsigned long addr, unsigned long end, |
e80d3909 JR |
2499 | pte_fn_t fn, void *data, bool create, |
2500 | pgtbl_mod_mask *mask) | |
aee16b3c JF |
2501 | { |
2502 | pud_t *pud; | |
2503 | unsigned long next; | |
be1db475 | 2504 | int err = 0; |
aee16b3c | 2505 | |
be1db475 | 2506 | if (create) { |
e80d3909 | 2507 | pud = pud_alloc_track(mm, p4d, addr, mask); |
be1db475 DA |
2508 | if (!pud) |
2509 | return -ENOMEM; | |
2510 | } else { | |
2511 | pud = pud_offset(p4d, addr); | |
2512 | } | |
aee16b3c JF |
2513 | do { |
2514 | next = pud_addr_end(addr, end); | |
0c95cba4 NP |
2515 | if (pud_none(*pud) && !create) |
2516 | continue; | |
2517 | if (WARN_ON_ONCE(pud_leaf(*pud))) | |
2518 | return -EINVAL; | |
2519 | if (!pud_none(*pud) && WARN_ON_ONCE(pud_bad(*pud))) { | |
2520 | if (!create) | |
2521 | continue; | |
2522 | pud_clear_bad(pud); | |
be1db475 | 2523 | } |
0c95cba4 NP |
2524 | err = apply_to_pmd_range(mm, pud, addr, next, |
2525 | fn, data, create, mask); | |
2526 | if (err) | |
2527 | break; | |
aee16b3c | 2528 | } while (pud++, addr = next, addr != end); |
0c95cba4 | 2529 | |
aee16b3c JF |
2530 | return err; |
2531 | } | |
2532 | ||
c2febafc KS |
2533 | static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2534 | unsigned long addr, unsigned long end, | |
e80d3909 JR |
2535 | pte_fn_t fn, void *data, bool create, |
2536 | pgtbl_mod_mask *mask) | |
c2febafc KS |
2537 | { |
2538 | p4d_t *p4d; | |
2539 | unsigned long next; | |
be1db475 | 2540 | int err = 0; |
c2febafc | 2541 | |
be1db475 | 2542 | if (create) { |
e80d3909 | 2543 | p4d = p4d_alloc_track(mm, pgd, addr, mask); |
be1db475 DA |
2544 | if (!p4d) |
2545 | return -ENOMEM; | |
2546 | } else { | |
2547 | p4d = p4d_offset(pgd, addr); | |
2548 | } | |
c2febafc KS |
2549 | do { |
2550 | next = p4d_addr_end(addr, end); | |
0c95cba4 NP |
2551 | if (p4d_none(*p4d) && !create) |
2552 | continue; | |
2553 | if (WARN_ON_ONCE(p4d_leaf(*p4d))) | |
2554 | return -EINVAL; | |
2555 | if (!p4d_none(*p4d) && WARN_ON_ONCE(p4d_bad(*p4d))) { | |
2556 | if (!create) | |
2557 | continue; | |
2558 | p4d_clear_bad(p4d); | |
be1db475 | 2559 | } |
0c95cba4 NP |
2560 | err = apply_to_pud_range(mm, p4d, addr, next, |
2561 | fn, data, create, mask); | |
2562 | if (err) | |
2563 | break; | |
c2febafc | 2564 | } while (p4d++, addr = next, addr != end); |
0c95cba4 | 2565 | |
c2febafc KS |
2566 | return err; |
2567 | } | |
2568 | ||
be1db475 DA |
2569 | static int __apply_to_page_range(struct mm_struct *mm, unsigned long addr, |
2570 | unsigned long size, pte_fn_t fn, | |
2571 | void *data, bool create) | |
aee16b3c JF |
2572 | { |
2573 | pgd_t *pgd; | |
e80d3909 | 2574 | unsigned long start = addr, next; |
57250a5b | 2575 | unsigned long end = addr + size; |
e80d3909 | 2576 | pgtbl_mod_mask mask = 0; |
be1db475 | 2577 | int err = 0; |
aee16b3c | 2578 | |
9cb65bc3 MP |
2579 | if (WARN_ON(addr >= end)) |
2580 | return -EINVAL; | |
2581 | ||
aee16b3c JF |
2582 | pgd = pgd_offset(mm, addr); |
2583 | do { | |
2584 | next = pgd_addr_end(addr, end); | |
0c95cba4 | 2585 | if (pgd_none(*pgd) && !create) |
be1db475 | 2586 | continue; |
0c95cba4 NP |
2587 | if (WARN_ON_ONCE(pgd_leaf(*pgd))) |
2588 | return -EINVAL; | |
2589 | if (!pgd_none(*pgd) && WARN_ON_ONCE(pgd_bad(*pgd))) { | |
2590 | if (!create) | |
2591 | continue; | |
2592 | pgd_clear_bad(pgd); | |
2593 | } | |
2594 | err = apply_to_p4d_range(mm, pgd, addr, next, | |
2595 | fn, data, create, &mask); | |
aee16b3c JF |
2596 | if (err) |
2597 | break; | |
2598 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2599 | |
e80d3909 JR |
2600 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
2601 | arch_sync_kernel_mappings(start, start + size); | |
2602 | ||
aee16b3c JF |
2603 | return err; |
2604 | } | |
be1db475 DA |
2605 | |
2606 | /* | |
2607 | * Scan a region of virtual memory, filling in page tables as necessary | |
2608 | * and calling a provided function on each leaf page table. | |
2609 | */ | |
2610 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2611 | unsigned long size, pte_fn_t fn, void *data) | |
2612 | { | |
2613 | return __apply_to_page_range(mm, addr, size, fn, data, true); | |
2614 | } | |
aee16b3c JF |
2615 | EXPORT_SYMBOL_GPL(apply_to_page_range); |
2616 | ||
be1db475 DA |
2617 | /* |
2618 | * Scan a region of virtual memory, calling a provided function on | |
2619 | * each leaf page table where it exists. | |
2620 | * | |
2621 | * Unlike apply_to_page_range, this does _not_ fill in page tables | |
2622 | * where they are absent. | |
2623 | */ | |
2624 | int apply_to_existing_page_range(struct mm_struct *mm, unsigned long addr, | |
2625 | unsigned long size, pte_fn_t fn, void *data) | |
2626 | { | |
2627 | return __apply_to_page_range(mm, addr, size, fn, data, false); | |
2628 | } | |
2629 | EXPORT_SYMBOL_GPL(apply_to_existing_page_range); | |
2630 | ||
8f4e2101 | 2631 | /* |
9b4bdd2f KS |
2632 | * handle_pte_fault chooses page fault handler according to an entry which was |
2633 | * read non-atomically. Before making any commitment, on those architectures | |
2634 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
2635 | * parts, do_swap_page must check under lock before unmapping the pte and | |
2636 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2637 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2638 | */ |
4c21e2f2 | 2639 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2640 | pte_t *page_table, pte_t orig_pte) |
2641 | { | |
2642 | int same = 1; | |
923717cb | 2643 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPTION) |
8f4e2101 | 2644 | if (sizeof(pte_t) > sizeof(unsigned long)) { |
4c21e2f2 HD |
2645 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2646 | spin_lock(ptl); | |
8f4e2101 | 2647 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2648 | spin_unlock(ptl); |
8f4e2101 HD |
2649 | } |
2650 | #endif | |
2651 | pte_unmap(page_table); | |
2652 | return same; | |
2653 | } | |
2654 | ||
83d116c5 JH |
2655 | static inline bool cow_user_page(struct page *dst, struct page *src, |
2656 | struct vm_fault *vmf) | |
6aab341e | 2657 | { |
83d116c5 JH |
2658 | bool ret; |
2659 | void *kaddr; | |
2660 | void __user *uaddr; | |
c3e5ea6e | 2661 | bool locked = false; |
83d116c5 JH |
2662 | struct vm_area_struct *vma = vmf->vma; |
2663 | struct mm_struct *mm = vma->vm_mm; | |
2664 | unsigned long addr = vmf->address; | |
2665 | ||
83d116c5 JH |
2666 | if (likely(src)) { |
2667 | copy_user_highpage(dst, src, addr, vma); | |
2668 | return true; | |
2669 | } | |
2670 | ||
6aab341e LT |
2671 | /* |
2672 | * If the source page was a PFN mapping, we don't have | |
2673 | * a "struct page" for it. We do a best-effort copy by | |
2674 | * just copying from the original user address. If that | |
2675 | * fails, we just zero-fill it. Live with it. | |
2676 | */ | |
83d116c5 JH |
2677 | kaddr = kmap_atomic(dst); |
2678 | uaddr = (void __user *)(addr & PAGE_MASK); | |
2679 | ||
2680 | /* | |
2681 | * On architectures with software "accessed" bits, we would | |
2682 | * take a double page fault, so mark it accessed here. | |
2683 | */ | |
c3e5ea6e | 2684 | if (arch_faults_on_old_pte() && !pte_young(vmf->orig_pte)) { |
83d116c5 | 2685 | pte_t entry; |
5d2a2dbb | 2686 | |
83d116c5 | 2687 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); |
c3e5ea6e | 2688 | locked = true; |
83d116c5 JH |
2689 | if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2690 | /* | |
2691 | * Other thread has already handled the fault | |
7df67697 | 2692 | * and update local tlb only |
83d116c5 | 2693 | */ |
7df67697 | 2694 | update_mmu_tlb(vma, addr, vmf->pte); |
83d116c5 JH |
2695 | ret = false; |
2696 | goto pte_unlock; | |
2697 | } | |
2698 | ||
2699 | entry = pte_mkyoung(vmf->orig_pte); | |
2700 | if (ptep_set_access_flags(vma, addr, vmf->pte, entry, 0)) | |
2701 | update_mmu_cache(vma, addr, vmf->pte); | |
2702 | } | |
2703 | ||
2704 | /* | |
2705 | * This really shouldn't fail, because the page is there | |
2706 | * in the page tables. But it might just be unreadable, | |
2707 | * in which case we just give up and fill the result with | |
2708 | * zeroes. | |
2709 | */ | |
2710 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { | |
c3e5ea6e KS |
2711 | if (locked) |
2712 | goto warn; | |
2713 | ||
2714 | /* Re-validate under PTL if the page is still mapped */ | |
2715 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); | |
2716 | locked = true; | |
2717 | if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) { | |
7df67697 BM |
2718 | /* The PTE changed under us, update local tlb */ |
2719 | update_mmu_tlb(vma, addr, vmf->pte); | |
c3e5ea6e KS |
2720 | ret = false; |
2721 | goto pte_unlock; | |
2722 | } | |
2723 | ||
5d2a2dbb | 2724 | /* |
985ba004 | 2725 | * The same page can be mapped back since last copy attempt. |
c3e5ea6e | 2726 | * Try to copy again under PTL. |
5d2a2dbb | 2727 | */ |
c3e5ea6e KS |
2728 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { |
2729 | /* | |
2730 | * Give a warn in case there can be some obscure | |
2731 | * use-case | |
2732 | */ | |
2733 | warn: | |
2734 | WARN_ON_ONCE(1); | |
2735 | clear_page(kaddr); | |
2736 | } | |
83d116c5 JH |
2737 | } |
2738 | ||
2739 | ret = true; | |
2740 | ||
2741 | pte_unlock: | |
c3e5ea6e | 2742 | if (locked) |
83d116c5 JH |
2743 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
2744 | kunmap_atomic(kaddr); | |
2745 | flush_dcache_page(dst); | |
2746 | ||
2747 | return ret; | |
6aab341e LT |
2748 | } |
2749 | ||
c20cd45e MH |
2750 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
2751 | { | |
2752 | struct file *vm_file = vma->vm_file; | |
2753 | ||
2754 | if (vm_file) | |
2755 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
2756 | ||
2757 | /* | |
2758 | * Special mappings (e.g. VDSO) do not have any file so fake | |
2759 | * a default GFP_KERNEL for them. | |
2760 | */ | |
2761 | return GFP_KERNEL; | |
2762 | } | |
2763 | ||
fb09a464 KS |
2764 | /* |
2765 | * Notify the address space that the page is about to become writable so that | |
2766 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2767 | * | |
2768 | * We do this without the lock held, so that it can sleep if it needs to. | |
2769 | */ | |
2b740303 | 2770 | static vm_fault_t do_page_mkwrite(struct vm_fault *vmf) |
fb09a464 | 2771 | { |
2b740303 | 2772 | vm_fault_t ret; |
38b8cb7f JK |
2773 | struct page *page = vmf->page; |
2774 | unsigned int old_flags = vmf->flags; | |
fb09a464 | 2775 | |
38b8cb7f | 2776 | vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
fb09a464 | 2777 | |
dc617f29 DW |
2778 | if (vmf->vma->vm_file && |
2779 | IS_SWAPFILE(vmf->vma->vm_file->f_mapping->host)) | |
2780 | return VM_FAULT_SIGBUS; | |
2781 | ||
11bac800 | 2782 | ret = vmf->vma->vm_ops->page_mkwrite(vmf); |
38b8cb7f JK |
2783 | /* Restore original flags so that caller is not surprised */ |
2784 | vmf->flags = old_flags; | |
fb09a464 KS |
2785 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2786 | return ret; | |
2787 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2788 | lock_page(page); | |
2789 | if (!page->mapping) { | |
2790 | unlock_page(page); | |
2791 | return 0; /* retry */ | |
2792 | } | |
2793 | ret |= VM_FAULT_LOCKED; | |
2794 | } else | |
2795 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2796 | return ret; | |
2797 | } | |
2798 | ||
97ba0c2b JK |
2799 | /* |
2800 | * Handle dirtying of a page in shared file mapping on a write fault. | |
2801 | * | |
2802 | * The function expects the page to be locked and unlocks it. | |
2803 | */ | |
89b15332 | 2804 | static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf) |
97ba0c2b | 2805 | { |
89b15332 | 2806 | struct vm_area_struct *vma = vmf->vma; |
97ba0c2b | 2807 | struct address_space *mapping; |
89b15332 | 2808 | struct page *page = vmf->page; |
97ba0c2b JK |
2809 | bool dirtied; |
2810 | bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; | |
2811 | ||
2812 | dirtied = set_page_dirty(page); | |
2813 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2814 | /* | |
2815 | * Take a local copy of the address_space - page.mapping may be zeroed | |
2816 | * by truncate after unlock_page(). The address_space itself remains | |
2817 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
2818 | * release semantics to prevent the compiler from undoing this copying. | |
2819 | */ | |
2820 | mapping = page_rmapping(page); | |
2821 | unlock_page(page); | |
2822 | ||
89b15332 JW |
2823 | if (!page_mkwrite) |
2824 | file_update_time(vma->vm_file); | |
2825 | ||
2826 | /* | |
2827 | * Throttle page dirtying rate down to writeback speed. | |
2828 | * | |
2829 | * mapping may be NULL here because some device drivers do not | |
2830 | * set page.mapping but still dirty their pages | |
2831 | * | |
c1e8d7c6 | 2832 | * Drop the mmap_lock before waiting on IO, if we can. The file |
89b15332 JW |
2833 | * is pinning the mapping, as per above. |
2834 | */ | |
97ba0c2b | 2835 | if ((dirtied || page_mkwrite) && mapping) { |
89b15332 JW |
2836 | struct file *fpin; |
2837 | ||
2838 | fpin = maybe_unlock_mmap_for_io(vmf, NULL); | |
97ba0c2b | 2839 | balance_dirty_pages_ratelimited(mapping); |
89b15332 JW |
2840 | if (fpin) { |
2841 | fput(fpin); | |
2842 | return VM_FAULT_RETRY; | |
2843 | } | |
97ba0c2b JK |
2844 | } |
2845 | ||
89b15332 | 2846 | return 0; |
97ba0c2b JK |
2847 | } |
2848 | ||
4e047f89 SR |
2849 | /* |
2850 | * Handle write page faults for pages that can be reused in the current vma | |
2851 | * | |
2852 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
2853 | * or due to us being the last reference standing to the page. In either | |
2854 | * case, all we need to do here is to mark the page as writable and update | |
2855 | * any related book-keeping. | |
2856 | */ | |
997dd98d | 2857 | static inline void wp_page_reuse(struct vm_fault *vmf) |
82b0f8c3 | 2858 | __releases(vmf->ptl) |
4e047f89 | 2859 | { |
82b0f8c3 | 2860 | struct vm_area_struct *vma = vmf->vma; |
a41b70d6 | 2861 | struct page *page = vmf->page; |
4e047f89 SR |
2862 | pte_t entry; |
2863 | /* | |
2864 | * Clear the pages cpupid information as the existing | |
2865 | * information potentially belongs to a now completely | |
2866 | * unrelated process. | |
2867 | */ | |
2868 | if (page) | |
2869 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2870 | ||
2994302b JK |
2871 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2872 | entry = pte_mkyoung(vmf->orig_pte); | |
4e047f89 | 2873 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
82b0f8c3 JK |
2874 | if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) |
2875 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2876 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
798a6b87 | 2877 | count_vm_event(PGREUSE); |
4e047f89 SR |
2878 | } |
2879 | ||
2f38ab2c SR |
2880 | /* |
2881 | * Handle the case of a page which we actually need to copy to a new page. | |
2882 | * | |
c1e8d7c6 | 2883 | * Called with mmap_lock locked and the old page referenced, but |
2f38ab2c SR |
2884 | * without the ptl held. |
2885 | * | |
2886 | * High level logic flow: | |
2887 | * | |
2888 | * - Allocate a page, copy the content of the old page to the new one. | |
2889 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2890 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2891 | * - If the pte is still the way we remember it, update the page table and all | |
2892 | * relevant references. This includes dropping the reference the page-table | |
2893 | * held to the old page, as well as updating the rmap. | |
2894 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2895 | */ | |
2b740303 | 2896 | static vm_fault_t wp_page_copy(struct vm_fault *vmf) |
2f38ab2c | 2897 | { |
82b0f8c3 | 2898 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2899 | struct mm_struct *mm = vma->vm_mm; |
a41b70d6 | 2900 | struct page *old_page = vmf->page; |
2f38ab2c | 2901 | struct page *new_page = NULL; |
2f38ab2c SR |
2902 | pte_t entry; |
2903 | int page_copied = 0; | |
ac46d4f3 | 2904 | struct mmu_notifier_range range; |
2f38ab2c SR |
2905 | |
2906 | if (unlikely(anon_vma_prepare(vma))) | |
2907 | goto oom; | |
2908 | ||
2994302b | 2909 | if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { |
82b0f8c3 JK |
2910 | new_page = alloc_zeroed_user_highpage_movable(vma, |
2911 | vmf->address); | |
2f38ab2c SR |
2912 | if (!new_page) |
2913 | goto oom; | |
2914 | } else { | |
bae473a4 | 2915 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
82b0f8c3 | 2916 | vmf->address); |
2f38ab2c SR |
2917 | if (!new_page) |
2918 | goto oom; | |
83d116c5 JH |
2919 | |
2920 | if (!cow_user_page(new_page, old_page, vmf)) { | |
2921 | /* | |
2922 | * COW failed, if the fault was solved by other, | |
2923 | * it's fine. If not, userspace would re-fault on | |
2924 | * the same address and we will handle the fault | |
2925 | * from the second attempt. | |
2926 | */ | |
2927 | put_page(new_page); | |
2928 | if (old_page) | |
2929 | put_page(old_page); | |
2930 | return 0; | |
2931 | } | |
2f38ab2c | 2932 | } |
2f38ab2c | 2933 | |
d9eb1ea2 | 2934 | if (mem_cgroup_charge(new_page, mm, GFP_KERNEL)) |
2f38ab2c | 2935 | goto oom_free_new; |
9d82c694 | 2936 | cgroup_throttle_swaprate(new_page, GFP_KERNEL); |
2f38ab2c | 2937 | |
eb3c24f3 MG |
2938 | __SetPageUptodate(new_page); |
2939 | ||
7269f999 | 2940 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, |
6f4f13e8 | 2941 | vmf->address & PAGE_MASK, |
ac46d4f3 JG |
2942 | (vmf->address & PAGE_MASK) + PAGE_SIZE); |
2943 | mmu_notifier_invalidate_range_start(&range); | |
2f38ab2c SR |
2944 | |
2945 | /* | |
2946 | * Re-check the pte - we dropped the lock | |
2947 | */ | |
82b0f8c3 | 2948 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); |
2994302b | 2949 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2f38ab2c SR |
2950 | if (old_page) { |
2951 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
2952 | dec_mm_counter_fast(mm, |
2953 | mm_counter_file(old_page)); | |
2f38ab2c SR |
2954 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
2955 | } | |
2956 | } else { | |
2957 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
2958 | } | |
2994302b | 2959 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2f38ab2c | 2960 | entry = mk_pte(new_page, vma->vm_page_prot); |
50c25ee9 | 2961 | entry = pte_sw_mkyoung(entry); |
2f38ab2c | 2962 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
111fe718 | 2963 | |
2f38ab2c SR |
2964 | /* |
2965 | * Clear the pte entry and flush it first, before updating the | |
111fe718 NP |
2966 | * pte with the new entry, to keep TLBs on different CPUs in |
2967 | * sync. This code used to set the new PTE then flush TLBs, but | |
2968 | * that left a window where the new PTE could be loaded into | |
2969 | * some TLBs while the old PTE remains in others. | |
2f38ab2c | 2970 | */ |
82b0f8c3 JK |
2971 | ptep_clear_flush_notify(vma, vmf->address, vmf->pte); |
2972 | page_add_new_anon_rmap(new_page, vma, vmf->address, false); | |
b518154e | 2973 | lru_cache_add_inactive_or_unevictable(new_page, vma); |
2f38ab2c SR |
2974 | /* |
2975 | * We call the notify macro here because, when using secondary | |
2976 | * mmu page tables (such as kvm shadow page tables), we want the | |
2977 | * new page to be mapped directly into the secondary page table. | |
2978 | */ | |
82b0f8c3 JK |
2979 | set_pte_at_notify(mm, vmf->address, vmf->pte, entry); |
2980 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2f38ab2c SR |
2981 | if (old_page) { |
2982 | /* | |
2983 | * Only after switching the pte to the new page may | |
2984 | * we remove the mapcount here. Otherwise another | |
2985 | * process may come and find the rmap count decremented | |
2986 | * before the pte is switched to the new page, and | |
2987 | * "reuse" the old page writing into it while our pte | |
2988 | * here still points into it and can be read by other | |
2989 | * threads. | |
2990 | * | |
2991 | * The critical issue is to order this | |
2992 | * page_remove_rmap with the ptp_clear_flush above. | |
2993 | * Those stores are ordered by (if nothing else,) | |
2994 | * the barrier present in the atomic_add_negative | |
2995 | * in page_remove_rmap. | |
2996 | * | |
2997 | * Then the TLB flush in ptep_clear_flush ensures that | |
2998 | * no process can access the old page before the | |
2999 | * decremented mapcount is visible. And the old page | |
3000 | * cannot be reused until after the decremented | |
3001 | * mapcount is visible. So transitively, TLBs to | |
3002 | * old page will be flushed before it can be reused. | |
3003 | */ | |
d281ee61 | 3004 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
3005 | } |
3006 | ||
3007 | /* Free the old page.. */ | |
3008 | new_page = old_page; | |
3009 | page_copied = 1; | |
3010 | } else { | |
7df67697 | 3011 | update_mmu_tlb(vma, vmf->address, vmf->pte); |
2f38ab2c SR |
3012 | } |
3013 | ||
3014 | if (new_page) | |
09cbfeaf | 3015 | put_page(new_page); |
2f38ab2c | 3016 | |
82b0f8c3 | 3017 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4645b9fe JG |
3018 | /* |
3019 | * No need to double call mmu_notifier->invalidate_range() callback as | |
3020 | * the above ptep_clear_flush_notify() did already call it. | |
3021 | */ | |
ac46d4f3 | 3022 | mmu_notifier_invalidate_range_only_end(&range); |
2f38ab2c SR |
3023 | if (old_page) { |
3024 | /* | |
3025 | * Don't let another task, with possibly unlocked vma, | |
3026 | * keep the mlocked page. | |
3027 | */ | |
3028 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
3029 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
3030 | if (PageMlocked(old_page)) |
3031 | munlock_vma_page(old_page); | |
2f38ab2c SR |
3032 | unlock_page(old_page); |
3033 | } | |
f4c4a3f4 HY |
3034 | if (page_copied) |
3035 | free_swap_cache(old_page); | |
09cbfeaf | 3036 | put_page(old_page); |
2f38ab2c SR |
3037 | } |
3038 | return page_copied ? VM_FAULT_WRITE : 0; | |
3039 | oom_free_new: | |
09cbfeaf | 3040 | put_page(new_page); |
2f38ab2c SR |
3041 | oom: |
3042 | if (old_page) | |
09cbfeaf | 3043 | put_page(old_page); |
2f38ab2c SR |
3044 | return VM_FAULT_OOM; |
3045 | } | |
3046 | ||
66a6197c JK |
3047 | /** |
3048 | * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE | |
3049 | * writeable once the page is prepared | |
3050 | * | |
3051 | * @vmf: structure describing the fault | |
3052 | * | |
3053 | * This function handles all that is needed to finish a write page fault in a | |
3054 | * shared mapping due to PTE being read-only once the mapped page is prepared. | |
a862f68a | 3055 | * It handles locking of PTE and modifying it. |
66a6197c JK |
3056 | * |
3057 | * The function expects the page to be locked or other protection against | |
3058 | * concurrent faults / writeback (such as DAX radix tree locks). | |
a862f68a | 3059 | * |
2797e79f | 3060 | * Return: %0 on success, %VM_FAULT_NOPAGE when PTE got changed before |
a862f68a | 3061 | * we acquired PTE lock. |
66a6197c | 3062 | */ |
2b740303 | 3063 | vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf) |
66a6197c JK |
3064 | { |
3065 | WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); | |
3066 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, | |
3067 | &vmf->ptl); | |
3068 | /* | |
3069 | * We might have raced with another page fault while we released the | |
3070 | * pte_offset_map_lock. | |
3071 | */ | |
3072 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { | |
7df67697 | 3073 | update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); |
66a6197c | 3074 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a19e2553 | 3075 | return VM_FAULT_NOPAGE; |
66a6197c JK |
3076 | } |
3077 | wp_page_reuse(vmf); | |
a19e2553 | 3078 | return 0; |
66a6197c JK |
3079 | } |
3080 | ||
dd906184 BH |
3081 | /* |
3082 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
3083 | * mapping | |
3084 | */ | |
2b740303 | 3085 | static vm_fault_t wp_pfn_shared(struct vm_fault *vmf) |
dd906184 | 3086 | { |
82b0f8c3 | 3087 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 3088 | |
dd906184 | 3089 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
2b740303 | 3090 | vm_fault_t ret; |
dd906184 | 3091 | |
82b0f8c3 | 3092 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
fe82221f | 3093 | vmf->flags |= FAULT_FLAG_MKWRITE; |
11bac800 | 3094 | ret = vma->vm_ops->pfn_mkwrite(vmf); |
2f89dc12 | 3095 | if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) |
dd906184 | 3096 | return ret; |
66a6197c | 3097 | return finish_mkwrite_fault(vmf); |
dd906184 | 3098 | } |
997dd98d JK |
3099 | wp_page_reuse(vmf); |
3100 | return VM_FAULT_WRITE; | |
dd906184 BH |
3101 | } |
3102 | ||
2b740303 | 3103 | static vm_fault_t wp_page_shared(struct vm_fault *vmf) |
82b0f8c3 | 3104 | __releases(vmf->ptl) |
93e478d4 | 3105 | { |
82b0f8c3 | 3106 | struct vm_area_struct *vma = vmf->vma; |
89b15332 | 3107 | vm_fault_t ret = VM_FAULT_WRITE; |
93e478d4 | 3108 | |
a41b70d6 | 3109 | get_page(vmf->page); |
93e478d4 | 3110 | |
93e478d4 | 3111 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2b740303 | 3112 | vm_fault_t tmp; |
93e478d4 | 3113 | |
82b0f8c3 | 3114 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
38b8cb7f | 3115 | tmp = do_page_mkwrite(vmf); |
93e478d4 SR |
3116 | if (unlikely(!tmp || (tmp & |
3117 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
a41b70d6 | 3118 | put_page(vmf->page); |
93e478d4 SR |
3119 | return tmp; |
3120 | } | |
66a6197c | 3121 | tmp = finish_mkwrite_fault(vmf); |
a19e2553 | 3122 | if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
a41b70d6 | 3123 | unlock_page(vmf->page); |
a41b70d6 | 3124 | put_page(vmf->page); |
66a6197c | 3125 | return tmp; |
93e478d4 | 3126 | } |
66a6197c JK |
3127 | } else { |
3128 | wp_page_reuse(vmf); | |
997dd98d | 3129 | lock_page(vmf->page); |
93e478d4 | 3130 | } |
89b15332 | 3131 | ret |= fault_dirty_shared_page(vmf); |
997dd98d | 3132 | put_page(vmf->page); |
93e478d4 | 3133 | |
89b15332 | 3134 | return ret; |
93e478d4 SR |
3135 | } |
3136 | ||
1da177e4 LT |
3137 | /* |
3138 | * This routine handles present pages, when users try to write | |
3139 | * to a shared page. It is done by copying the page to a new address | |
3140 | * and decrementing the shared-page counter for the old page. | |
3141 | * | |
1da177e4 LT |
3142 | * Note that this routine assumes that the protection checks have been |
3143 | * done by the caller (the low-level page fault routine in most cases). | |
3144 | * Thus we can safely just mark it writable once we've done any necessary | |
3145 | * COW. | |
3146 | * | |
3147 | * We also mark the page dirty at this point even though the page will | |
3148 | * change only once the write actually happens. This avoids a few races, | |
3149 | * and potentially makes it more efficient. | |
3150 | * | |
c1e8d7c6 | 3151 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
8f4e2101 | 3152 | * but allow concurrent faults), with pte both mapped and locked. |
c1e8d7c6 | 3153 | * We return with mmap_lock still held, but pte unmapped and unlocked. |
1da177e4 | 3154 | */ |
2b740303 | 3155 | static vm_fault_t do_wp_page(struct vm_fault *vmf) |
82b0f8c3 | 3156 | __releases(vmf->ptl) |
1da177e4 | 3157 | { |
82b0f8c3 | 3158 | struct vm_area_struct *vma = vmf->vma; |
1da177e4 | 3159 | |
292924b2 | 3160 | if (userfaultfd_pte_wp(vma, *vmf->pte)) { |
529b930b AA |
3161 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3162 | return handle_userfault(vmf, VM_UFFD_WP); | |
3163 | } | |
3164 | ||
6ce64428 NA |
3165 | /* |
3166 | * Userfaultfd write-protect can defer flushes. Ensure the TLB | |
3167 | * is flushed in this case before copying. | |
3168 | */ | |
3169 | if (unlikely(userfaultfd_wp(vmf->vma) && | |
3170 | mm_tlb_flush_pending(vmf->vma->vm_mm))) | |
3171 | flush_tlb_page(vmf->vma, vmf->address); | |
3172 | ||
a41b70d6 JK |
3173 | vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); |
3174 | if (!vmf->page) { | |
251b97f5 | 3175 | /* |
64e45507 PF |
3176 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
3177 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
3178 | * |
3179 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 3180 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
3181 | */ |
3182 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
3183 | (VM_WRITE|VM_SHARED)) | |
2994302b | 3184 | return wp_pfn_shared(vmf); |
2f38ab2c | 3185 | |
82b0f8c3 | 3186 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 3187 | return wp_page_copy(vmf); |
251b97f5 | 3188 | } |
1da177e4 | 3189 | |
d08b3851 | 3190 | /* |
ee6a6457 PZ |
3191 | * Take out anonymous pages first, anonymous shared vmas are |
3192 | * not dirty accountable. | |
d08b3851 | 3193 | */ |
52d1e606 | 3194 | if (PageAnon(vmf->page)) { |
09854ba9 LT |
3195 | struct page *page = vmf->page; |
3196 | ||
3197 | /* PageKsm() doesn't necessarily raise the page refcount */ | |
3198 | if (PageKsm(page) || page_count(page) != 1) | |
3199 | goto copy; | |
3200 | if (!trylock_page(page)) | |
3201 | goto copy; | |
3202 | if (PageKsm(page) || page_mapcount(page) != 1 || page_count(page) != 1) { | |
3203 | unlock_page(page); | |
52d1e606 | 3204 | goto copy; |
b009c024 | 3205 | } |
09854ba9 LT |
3206 | /* |
3207 | * Ok, we've got the only map reference, and the only | |
3208 | * page count reference, and the page is locked, | |
3209 | * it's dark out, and we're wearing sunglasses. Hit it. | |
3210 | */ | |
09854ba9 | 3211 | unlock_page(page); |
be068f29 | 3212 | wp_page_reuse(vmf); |
09854ba9 | 3213 | return VM_FAULT_WRITE; |
ee6a6457 | 3214 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 3215 | (VM_WRITE|VM_SHARED))) { |
a41b70d6 | 3216 | return wp_page_shared(vmf); |
1da177e4 | 3217 | } |
52d1e606 | 3218 | copy: |
1da177e4 LT |
3219 | /* |
3220 | * Ok, we need to copy. Oh, well.. | |
3221 | */ | |
a41b70d6 | 3222 | get_page(vmf->page); |
28766805 | 3223 | |
82b0f8c3 | 3224 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 3225 | return wp_page_copy(vmf); |
1da177e4 LT |
3226 | } |
3227 | ||
97a89413 | 3228 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
3229 | unsigned long start_addr, unsigned long end_addr, |
3230 | struct zap_details *details) | |
3231 | { | |
f5cc4eef | 3232 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
3233 | } |
3234 | ||
f808c13f | 3235 | static inline void unmap_mapping_range_tree(struct rb_root_cached *root, |
1da177e4 LT |
3236 | struct zap_details *details) |
3237 | { | |
3238 | struct vm_area_struct *vma; | |
1da177e4 LT |
3239 | pgoff_t vba, vea, zba, zea; |
3240 | ||
6b2dbba8 | 3241 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 3242 | details->first_index, details->last_index) { |
1da177e4 LT |
3243 | |
3244 | vba = vma->vm_pgoff; | |
d6e93217 | 3245 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
3246 | zba = details->first_index; |
3247 | if (zba < vba) | |
3248 | zba = vba; | |
3249 | zea = details->last_index; | |
3250 | if (zea > vea) | |
3251 | zea = vea; | |
3252 | ||
97a89413 | 3253 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
3254 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
3255 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 3256 | details); |
1da177e4 LT |
3257 | } |
3258 | } | |
3259 | ||
22061a1f HD |
3260 | /** |
3261 | * unmap_mapping_page() - Unmap single page from processes. | |
3262 | * @page: The locked page to be unmapped. | |
3263 | * | |
3264 | * Unmap this page from any userspace process which still has it mmaped. | |
3265 | * Typically, for efficiency, the range of nearby pages has already been | |
3266 | * unmapped by unmap_mapping_pages() or unmap_mapping_range(). But once | |
3267 | * truncation or invalidation holds the lock on a page, it may find that | |
3268 | * the page has been remapped again: and then uses unmap_mapping_page() | |
3269 | * to unmap it finally. | |
3270 | */ | |
3271 | void unmap_mapping_page(struct page *page) | |
3272 | { | |
3273 | struct address_space *mapping = page->mapping; | |
3274 | struct zap_details details = { }; | |
3275 | ||
3276 | VM_BUG_ON(!PageLocked(page)); | |
3277 | VM_BUG_ON(PageTail(page)); | |
3278 | ||
3279 | details.check_mapping = mapping; | |
3280 | details.first_index = page->index; | |
3281 | details.last_index = page->index + thp_nr_pages(page) - 1; | |
3282 | details.single_page = page; | |
3283 | ||
3284 | i_mmap_lock_write(mapping); | |
3285 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) | |
3286 | unmap_mapping_range_tree(&mapping->i_mmap, &details); | |
3287 | i_mmap_unlock_write(mapping); | |
3288 | } | |
3289 | ||
977fbdcd MW |
3290 | /** |
3291 | * unmap_mapping_pages() - Unmap pages from processes. | |
3292 | * @mapping: The address space containing pages to be unmapped. | |
3293 | * @start: Index of first page to be unmapped. | |
3294 | * @nr: Number of pages to be unmapped. 0 to unmap to end of file. | |
3295 | * @even_cows: Whether to unmap even private COWed pages. | |
3296 | * | |
3297 | * Unmap the pages in this address space from any userspace process which | |
3298 | * has them mmaped. Generally, you want to remove COWed pages as well when | |
3299 | * a file is being truncated, but not when invalidating pages from the page | |
3300 | * cache. | |
3301 | */ | |
3302 | void unmap_mapping_pages(struct address_space *mapping, pgoff_t start, | |
3303 | pgoff_t nr, bool even_cows) | |
3304 | { | |
3305 | struct zap_details details = { }; | |
3306 | ||
3307 | details.check_mapping = even_cows ? NULL : mapping; | |
3308 | details.first_index = start; | |
3309 | details.last_index = start + nr - 1; | |
3310 | if (details.last_index < details.first_index) | |
3311 | details.last_index = ULONG_MAX; | |
3312 | ||
3313 | i_mmap_lock_write(mapping); | |
3314 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) | |
3315 | unmap_mapping_range_tree(&mapping->i_mmap, &details); | |
3316 | i_mmap_unlock_write(mapping); | |
3317 | } | |
3318 | ||
1da177e4 | 3319 | /** |
8a5f14a2 | 3320 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
977fbdcd | 3321 | * address_space corresponding to the specified byte range in the underlying |
8a5f14a2 KS |
3322 | * file. |
3323 | * | |
3d41088f | 3324 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
3325 | * @holebegin: byte in first page to unmap, relative to the start of |
3326 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 3327 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
3328 | * must keep the partial page. In contrast, we must get rid of |
3329 | * partial pages. | |
3330 | * @holelen: size of prospective hole in bytes. This will be rounded | |
3331 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
3332 | * end of the file. | |
3333 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
3334 | * but 0 when invalidating pagecache, don't throw away private data. | |
3335 | */ | |
3336 | void unmap_mapping_range(struct address_space *mapping, | |
3337 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
3338 | { | |
1da177e4 LT |
3339 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
3340 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
3341 | ||
3342 | /* Check for overflow. */ | |
3343 | if (sizeof(holelen) > sizeof(hlen)) { | |
3344 | long long holeend = | |
3345 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
3346 | if (holeend & ~(long long)ULONG_MAX) | |
3347 | hlen = ULONG_MAX - hba + 1; | |
3348 | } | |
3349 | ||
977fbdcd | 3350 | unmap_mapping_pages(mapping, hba, hlen, even_cows); |
1da177e4 LT |
3351 | } |
3352 | EXPORT_SYMBOL(unmap_mapping_range); | |
3353 | ||
1da177e4 | 3354 | /* |
c1e8d7c6 | 3355 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
8f4e2101 | 3356 | * but allow concurrent faults), and pte mapped but not yet locked. |
9a95f3cf PC |
3357 | * We return with pte unmapped and unlocked. |
3358 | * | |
c1e8d7c6 | 3359 | * We return with the mmap_lock locked or unlocked in the same cases |
9a95f3cf | 3360 | * as does filemap_fault(). |
1da177e4 | 3361 | */ |
2b740303 | 3362 | vm_fault_t do_swap_page(struct vm_fault *vmf) |
1da177e4 | 3363 | { |
82b0f8c3 | 3364 | struct vm_area_struct *vma = vmf->vma; |
eaf649eb | 3365 | struct page *page = NULL, *swapcache; |
2799e775 | 3366 | struct swap_info_struct *si = NULL; |
65500d23 | 3367 | swp_entry_t entry; |
1da177e4 | 3368 | pte_t pte; |
d065bd81 | 3369 | int locked; |
ad8c2ee8 | 3370 | int exclusive = 0; |
2b740303 | 3371 | vm_fault_t ret = 0; |
aae466b0 | 3372 | void *shadow = NULL; |
1da177e4 | 3373 | |
eaf649eb | 3374 | if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte)) |
8f4e2101 | 3375 | goto out; |
65500d23 | 3376 | |
2994302b | 3377 | entry = pte_to_swp_entry(vmf->orig_pte); |
d1737fdb AK |
3378 | if (unlikely(non_swap_entry(entry))) { |
3379 | if (is_migration_entry(entry)) { | |
82b0f8c3 JK |
3380 | migration_entry_wait(vma->vm_mm, vmf->pmd, |
3381 | vmf->address); | |
5042db43 | 3382 | } else if (is_device_private_entry(entry)) { |
af5cdaf8 | 3383 | vmf->page = pfn_swap_entry_to_page(entry); |
897e6365 | 3384 | ret = vmf->page->pgmap->ops->migrate_to_ram(vmf); |
d1737fdb AK |
3385 | } else if (is_hwpoison_entry(entry)) { |
3386 | ret = VM_FAULT_HWPOISON; | |
3387 | } else { | |
2994302b | 3388 | print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); |
d99be1a8 | 3389 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 3390 | } |
0697212a CL |
3391 | goto out; |
3392 | } | |
0bcac06f | 3393 | |
2799e775 ML |
3394 | /* Prevent swapoff from happening to us. */ |
3395 | si = get_swap_device(entry); | |
3396 | if (unlikely(!si)) | |
3397 | goto out; | |
0bcac06f | 3398 | |
3d1c7fd9 | 3399 | delayacct_set_flag(current, DELAYACCT_PF_SWAPIN); |
eaf649eb MK |
3400 | page = lookup_swap_cache(entry, vma, vmf->address); |
3401 | swapcache = page; | |
f8020772 | 3402 | |
1da177e4 | 3403 | if (!page) { |
a449bf58 QC |
3404 | if (data_race(si->flags & SWP_SYNCHRONOUS_IO) && |
3405 | __swap_count(entry) == 1) { | |
0bcac06f | 3406 | /* skip swapcache */ |
e9e9b7ec MK |
3407 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
3408 | vmf->address); | |
0bcac06f MK |
3409 | if (page) { |
3410 | __SetPageLocked(page); | |
3411 | __SetPageSwapBacked(page); | |
4c6355b2 | 3412 | |
0add0c77 SB |
3413 | if (mem_cgroup_swapin_charge_page(page, |
3414 | vma->vm_mm, GFP_KERNEL, entry)) { | |
545b1b07 | 3415 | ret = VM_FAULT_OOM; |
4c6355b2 | 3416 | goto out_page; |
545b1b07 | 3417 | } |
0add0c77 | 3418 | mem_cgroup_swapin_uncharge_swap(entry); |
4c6355b2 | 3419 | |
aae466b0 JK |
3420 | shadow = get_shadow_from_swap_cache(entry); |
3421 | if (shadow) | |
3422 | workingset_refault(page, shadow); | |
0076f029 | 3423 | |
6058eaec | 3424 | lru_cache_add(page); |
0add0c77 SB |
3425 | |
3426 | /* To provide entry to swap_readpage() */ | |
3427 | set_page_private(page, entry.val); | |
0bcac06f | 3428 | swap_readpage(page, true); |
0add0c77 | 3429 | set_page_private(page, 0); |
0bcac06f | 3430 | } |
aa8d22a1 | 3431 | } else { |
e9e9b7ec MK |
3432 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
3433 | vmf); | |
aa8d22a1 | 3434 | swapcache = page; |
0bcac06f MK |
3435 | } |
3436 | ||
1da177e4 LT |
3437 | if (!page) { |
3438 | /* | |
8f4e2101 HD |
3439 | * Back out if somebody else faulted in this pte |
3440 | * while we released the pte lock. | |
1da177e4 | 3441 | */ |
82b0f8c3 JK |
3442 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
3443 | vmf->address, &vmf->ptl); | |
2994302b | 3444 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) |
1da177e4 | 3445 | ret = VM_FAULT_OOM; |
3d1c7fd9 | 3446 | delayacct_clear_flag(current, DELAYACCT_PF_SWAPIN); |
65500d23 | 3447 | goto unlock; |
1da177e4 LT |
3448 | } |
3449 | ||
3450 | /* Had to read the page from swap area: Major fault */ | |
3451 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 3452 | count_vm_event(PGMAJFAULT); |
2262185c | 3453 | count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); |
d1737fdb | 3454 | } else if (PageHWPoison(page)) { |
71f72525 WF |
3455 | /* |
3456 | * hwpoisoned dirty swapcache pages are kept for killing | |
3457 | * owner processes (which may be unknown at hwpoison time) | |
3458 | */ | |
d1737fdb | 3459 | ret = VM_FAULT_HWPOISON; |
3d1c7fd9 | 3460 | delayacct_clear_flag(current, DELAYACCT_PF_SWAPIN); |
4779cb31 | 3461 | goto out_release; |
1da177e4 LT |
3462 | } |
3463 | ||
82b0f8c3 | 3464 | locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags); |
e709ffd6 | 3465 | |
3d1c7fd9 | 3466 | delayacct_clear_flag(current, DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
3467 | if (!locked) { |
3468 | ret |= VM_FAULT_RETRY; | |
3469 | goto out_release; | |
3470 | } | |
073e587e | 3471 | |
4969c119 | 3472 | /* |
31c4a3d3 HD |
3473 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
3474 | * release the swapcache from under us. The page pin, and pte_same | |
3475 | * test below, are not enough to exclude that. Even if it is still | |
3476 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 3477 | */ |
0bcac06f MK |
3478 | if (unlikely((!PageSwapCache(page) || |
3479 | page_private(page) != entry.val)) && swapcache) | |
4969c119 AA |
3480 | goto out_page; |
3481 | ||
82b0f8c3 | 3482 | page = ksm_might_need_to_copy(page, vma, vmf->address); |
cbf86cfe HD |
3483 | if (unlikely(!page)) { |
3484 | ret = VM_FAULT_OOM; | |
3485 | page = swapcache; | |
cbf86cfe | 3486 | goto out_page; |
5ad64688 HD |
3487 | } |
3488 | ||
9d82c694 | 3489 | cgroup_throttle_swaprate(page, GFP_KERNEL); |
8a9f3ccd | 3490 | |
1da177e4 | 3491 | /* |
8f4e2101 | 3492 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 3493 | */ |
82b0f8c3 JK |
3494 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3495 | &vmf->ptl); | |
2994302b | 3496 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) |
b8107480 | 3497 | goto out_nomap; |
b8107480 KK |
3498 | |
3499 | if (unlikely(!PageUptodate(page))) { | |
3500 | ret = VM_FAULT_SIGBUS; | |
3501 | goto out_nomap; | |
1da177e4 LT |
3502 | } |
3503 | ||
8c7c6e34 KH |
3504 | /* |
3505 | * The page isn't present yet, go ahead with the fault. | |
3506 | * | |
3507 | * Be careful about the sequence of operations here. | |
3508 | * To get its accounting right, reuse_swap_page() must be called | |
3509 | * while the page is counted on swap but not yet in mapcount i.e. | |
3510 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
3511 | * must be called after the swap_free(), or it will never succeed. | |
8c7c6e34 | 3512 | */ |
1da177e4 | 3513 | |
bae473a4 KS |
3514 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
3515 | dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS); | |
1da177e4 | 3516 | pte = mk_pte(page, vma->vm_page_prot); |
82b0f8c3 | 3517 | if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { |
1da177e4 | 3518 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
82b0f8c3 | 3519 | vmf->flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 3520 | ret |= VM_FAULT_WRITE; |
d281ee61 | 3521 | exclusive = RMAP_EXCLUSIVE; |
1da177e4 | 3522 | } |
1da177e4 | 3523 | flush_icache_page(vma, page); |
2994302b | 3524 | if (pte_swp_soft_dirty(vmf->orig_pte)) |
179ef71c | 3525 | pte = pte_mksoft_dirty(pte); |
f45ec5ff PX |
3526 | if (pte_swp_uffd_wp(vmf->orig_pte)) { |
3527 | pte = pte_mkuffd_wp(pte); | |
3528 | pte = pte_wrprotect(pte); | |
3529 | } | |
82b0f8c3 | 3530 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); |
ca827d55 | 3531 | arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte); |
2994302b | 3532 | vmf->orig_pte = pte; |
0bcac06f MK |
3533 | |
3534 | /* ksm created a completely new copy */ | |
3535 | if (unlikely(page != swapcache && swapcache)) { | |
82b0f8c3 | 3536 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
b518154e | 3537 | lru_cache_add_inactive_or_unevictable(page, vma); |
0bcac06f MK |
3538 | } else { |
3539 | do_page_add_anon_rmap(page, vma, vmf->address, exclusive); | |
00501b53 | 3540 | } |
1da177e4 | 3541 | |
c475a8ab | 3542 | swap_free(entry); |
5ccc5aba VD |
3543 | if (mem_cgroup_swap_full(page) || |
3544 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 3545 | try_to_free_swap(page); |
c475a8ab | 3546 | unlock_page(page); |
0bcac06f | 3547 | if (page != swapcache && swapcache) { |
4969c119 AA |
3548 | /* |
3549 | * Hold the lock to avoid the swap entry to be reused | |
3550 | * until we take the PT lock for the pte_same() check | |
3551 | * (to avoid false positives from pte_same). For | |
3552 | * further safety release the lock after the swap_free | |
3553 | * so that the swap count won't change under a | |
3554 | * parallel locked swapcache. | |
3555 | */ | |
3556 | unlock_page(swapcache); | |
09cbfeaf | 3557 | put_page(swapcache); |
4969c119 | 3558 | } |
c475a8ab | 3559 | |
82b0f8c3 | 3560 | if (vmf->flags & FAULT_FLAG_WRITE) { |
2994302b | 3561 | ret |= do_wp_page(vmf); |
61469f1d HD |
3562 | if (ret & VM_FAULT_ERROR) |
3563 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
3564 | goto out; |
3565 | } | |
3566 | ||
3567 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3568 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3569 | unlock: |
82b0f8c3 | 3570 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
1da177e4 | 3571 | out: |
2799e775 ML |
3572 | if (si) |
3573 | put_swap_device(si); | |
1da177e4 | 3574 | return ret; |
b8107480 | 3575 | out_nomap: |
82b0f8c3 | 3576 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bc43f75c | 3577 | out_page: |
b8107480 | 3578 | unlock_page(page); |
4779cb31 | 3579 | out_release: |
09cbfeaf | 3580 | put_page(page); |
0bcac06f | 3581 | if (page != swapcache && swapcache) { |
4969c119 | 3582 | unlock_page(swapcache); |
09cbfeaf | 3583 | put_page(swapcache); |
4969c119 | 3584 | } |
2799e775 ML |
3585 | if (si) |
3586 | put_swap_device(si); | |
65500d23 | 3587 | return ret; |
1da177e4 LT |
3588 | } |
3589 | ||
3590 | /* | |
c1e8d7c6 | 3591 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
8f4e2101 | 3592 | * but allow concurrent faults), and pte mapped but not yet locked. |
c1e8d7c6 | 3593 | * We return with mmap_lock still held, but pte unmapped and unlocked. |
1da177e4 | 3594 | */ |
2b740303 | 3595 | static vm_fault_t do_anonymous_page(struct vm_fault *vmf) |
1da177e4 | 3596 | { |
82b0f8c3 | 3597 | struct vm_area_struct *vma = vmf->vma; |
8f4e2101 | 3598 | struct page *page; |
2b740303 | 3599 | vm_fault_t ret = 0; |
1da177e4 | 3600 | pte_t entry; |
1da177e4 | 3601 | |
6b7339f4 KS |
3602 | /* File mapping without ->vm_ops ? */ |
3603 | if (vma->vm_flags & VM_SHARED) | |
3604 | return VM_FAULT_SIGBUS; | |
3605 | ||
7267ec00 KS |
3606 | /* |
3607 | * Use pte_alloc() instead of pte_alloc_map(). We can't run | |
3608 | * pte_offset_map() on pmds where a huge pmd might be created | |
3609 | * from a different thread. | |
3610 | * | |
3e4e28c5 | 3611 | * pte_alloc_map() is safe to use under mmap_write_lock(mm) or when |
7267ec00 KS |
3612 | * parallel threads are excluded by other means. |
3613 | * | |
3e4e28c5 | 3614 | * Here we only have mmap_read_lock(mm). |
7267ec00 | 3615 | */ |
4cf58924 | 3616 | if (pte_alloc(vma->vm_mm, vmf->pmd)) |
7267ec00 KS |
3617 | return VM_FAULT_OOM; |
3618 | ||
f9ce0be7 | 3619 | /* See comment in handle_pte_fault() */ |
82b0f8c3 | 3620 | if (unlikely(pmd_trans_unstable(vmf->pmd))) |
7267ec00 KS |
3621 | return 0; |
3622 | ||
11ac5524 | 3623 | /* Use the zero-page for reads */ |
82b0f8c3 | 3624 | if (!(vmf->flags & FAULT_FLAG_WRITE) && |
bae473a4 | 3625 | !mm_forbids_zeropage(vma->vm_mm)) { |
82b0f8c3 | 3626 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), |
62eede62 | 3627 | vma->vm_page_prot)); |
82b0f8c3 JK |
3628 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
3629 | vmf->address, &vmf->ptl); | |
7df67697 BM |
3630 | if (!pte_none(*vmf->pte)) { |
3631 | update_mmu_tlb(vma, vmf->address, vmf->pte); | |
a13ea5b7 | 3632 | goto unlock; |
7df67697 | 3633 | } |
6b31d595 MH |
3634 | ret = check_stable_address_space(vma->vm_mm); |
3635 | if (ret) | |
3636 | goto unlock; | |
6b251fc9 AA |
3637 | /* Deliver the page fault to userland, check inside PT lock */ |
3638 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 JK |
3639 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3640 | return handle_userfault(vmf, VM_UFFD_MISSING); | |
6b251fc9 | 3641 | } |
a13ea5b7 HD |
3642 | goto setpte; |
3643 | } | |
3644 | ||
557ed1fa | 3645 | /* Allocate our own private page. */ |
557ed1fa NP |
3646 | if (unlikely(anon_vma_prepare(vma))) |
3647 | goto oom; | |
82b0f8c3 | 3648 | page = alloc_zeroed_user_highpage_movable(vma, vmf->address); |
557ed1fa NP |
3649 | if (!page) |
3650 | goto oom; | |
eb3c24f3 | 3651 | |
d9eb1ea2 | 3652 | if (mem_cgroup_charge(page, vma->vm_mm, GFP_KERNEL)) |
eb3c24f3 | 3653 | goto oom_free_page; |
9d82c694 | 3654 | cgroup_throttle_swaprate(page, GFP_KERNEL); |
eb3c24f3 | 3655 | |
52f37629 MK |
3656 | /* |
3657 | * The memory barrier inside __SetPageUptodate makes sure that | |
f4f5329d | 3658 | * preceding stores to the page contents become visible before |
52f37629 MK |
3659 | * the set_pte_at() write. |
3660 | */ | |
0ed361de | 3661 | __SetPageUptodate(page); |
8f4e2101 | 3662 | |
557ed1fa | 3663 | entry = mk_pte(page, vma->vm_page_prot); |
50c25ee9 | 3664 | entry = pte_sw_mkyoung(entry); |
1ac0cb5d HD |
3665 | if (vma->vm_flags & VM_WRITE) |
3666 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 3667 | |
82b0f8c3 JK |
3668 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3669 | &vmf->ptl); | |
7df67697 BM |
3670 | if (!pte_none(*vmf->pte)) { |
3671 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
557ed1fa | 3672 | goto release; |
7df67697 | 3673 | } |
9ba69294 | 3674 | |
6b31d595 MH |
3675 | ret = check_stable_address_space(vma->vm_mm); |
3676 | if (ret) | |
3677 | goto release; | |
3678 | ||
6b251fc9 AA |
3679 | /* Deliver the page fault to userland, check inside PT lock */ |
3680 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 | 3681 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
09cbfeaf | 3682 | put_page(page); |
82b0f8c3 | 3683 | return handle_userfault(vmf, VM_UFFD_MISSING); |
6b251fc9 AA |
3684 | } |
3685 | ||
bae473a4 | 3686 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3687 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
b518154e | 3688 | lru_cache_add_inactive_or_unevictable(page, vma); |
a13ea5b7 | 3689 | setpte: |
82b0f8c3 | 3690 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
1da177e4 LT |
3691 | |
3692 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3693 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3694 | unlock: |
82b0f8c3 | 3695 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
6b31d595 | 3696 | return ret; |
8f4e2101 | 3697 | release: |
09cbfeaf | 3698 | put_page(page); |
8f4e2101 | 3699 | goto unlock; |
8a9f3ccd | 3700 | oom_free_page: |
09cbfeaf | 3701 | put_page(page); |
65500d23 | 3702 | oom: |
1da177e4 LT |
3703 | return VM_FAULT_OOM; |
3704 | } | |
3705 | ||
9a95f3cf | 3706 | /* |
c1e8d7c6 | 3707 | * The mmap_lock must have been held on entry, and may have been |
9a95f3cf PC |
3708 | * released depending on flags and vma->vm_ops->fault() return value. |
3709 | * See filemap_fault() and __lock_page_retry(). | |
3710 | */ | |
2b740303 | 3711 | static vm_fault_t __do_fault(struct vm_fault *vmf) |
7eae74af | 3712 | { |
82b0f8c3 | 3713 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 3714 | vm_fault_t ret; |
7eae74af | 3715 | |
63f3655f MH |
3716 | /* |
3717 | * Preallocate pte before we take page_lock because this might lead to | |
3718 | * deadlocks for memcg reclaim which waits for pages under writeback: | |
3719 | * lock_page(A) | |
3720 | * SetPageWriteback(A) | |
3721 | * unlock_page(A) | |
3722 | * lock_page(B) | |
3723 | * lock_page(B) | |
d383807a | 3724 | * pte_alloc_one |
63f3655f MH |
3725 | * shrink_page_list |
3726 | * wait_on_page_writeback(A) | |
3727 | * SetPageWriteback(B) | |
3728 | * unlock_page(B) | |
3729 | * # flush A, B to clear the writeback | |
3730 | */ | |
3731 | if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) { | |
a7069ee3 | 3732 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); |
63f3655f MH |
3733 | if (!vmf->prealloc_pte) |
3734 | return VM_FAULT_OOM; | |
3735 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3736 | } | |
3737 | ||
11bac800 | 3738 | ret = vma->vm_ops->fault(vmf); |
3917048d | 3739 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | |
b1aa812b | 3740 | VM_FAULT_DONE_COW))) |
bc2466e4 | 3741 | return ret; |
7eae74af | 3742 | |
667240e0 | 3743 | if (unlikely(PageHWPoison(vmf->page))) { |
7eae74af | 3744 | if (ret & VM_FAULT_LOCKED) |
667240e0 JK |
3745 | unlock_page(vmf->page); |
3746 | put_page(vmf->page); | |
936ca80d | 3747 | vmf->page = NULL; |
7eae74af KS |
3748 | return VM_FAULT_HWPOISON; |
3749 | } | |
3750 | ||
3751 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
667240e0 | 3752 | lock_page(vmf->page); |
7eae74af | 3753 | else |
667240e0 | 3754 | VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); |
7eae74af | 3755 | |
7eae74af KS |
3756 | return ret; |
3757 | } | |
3758 | ||
396bcc52 | 3759 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
82b0f8c3 | 3760 | static void deposit_prealloc_pte(struct vm_fault *vmf) |
953c66c2 | 3761 | { |
82b0f8c3 | 3762 | struct vm_area_struct *vma = vmf->vma; |
953c66c2 | 3763 | |
82b0f8c3 | 3764 | pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
953c66c2 AK |
3765 | /* |
3766 | * We are going to consume the prealloc table, | |
3767 | * count that as nr_ptes. | |
3768 | */ | |
c4812909 | 3769 | mm_inc_nr_ptes(vma->vm_mm); |
7f2b6ce8 | 3770 | vmf->prealloc_pte = NULL; |
953c66c2 AK |
3771 | } |
3772 | ||
f9ce0be7 | 3773 | vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3774 | { |
82b0f8c3 JK |
3775 | struct vm_area_struct *vma = vmf->vma; |
3776 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3777 | unsigned long haddr = vmf->address & HPAGE_PMD_MASK; | |
10102459 | 3778 | pmd_t entry; |
2b740303 | 3779 | int i; |
d01ac3c3 | 3780 | vm_fault_t ret = VM_FAULT_FALLBACK; |
10102459 KS |
3781 | |
3782 | if (!transhuge_vma_suitable(vma, haddr)) | |
d01ac3c3 | 3783 | return ret; |
10102459 | 3784 | |
10102459 | 3785 | page = compound_head(page); |
d01ac3c3 MWO |
3786 | if (compound_order(page) != HPAGE_PMD_ORDER) |
3787 | return ret; | |
10102459 | 3788 | |
953c66c2 | 3789 | /* |
f0953a1b | 3790 | * Archs like ppc64 need additional space to store information |
953c66c2 AK |
3791 | * related to pte entry. Use the preallocated table for that. |
3792 | */ | |
82b0f8c3 | 3793 | if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { |
4cf58924 | 3794 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); |
82b0f8c3 | 3795 | if (!vmf->prealloc_pte) |
953c66c2 AK |
3796 | return VM_FAULT_OOM; |
3797 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3798 | } | |
3799 | ||
82b0f8c3 JK |
3800 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); |
3801 | if (unlikely(!pmd_none(*vmf->pmd))) | |
10102459 KS |
3802 | goto out; |
3803 | ||
3804 | for (i = 0; i < HPAGE_PMD_NR; i++) | |
3805 | flush_icache_page(vma, page + i); | |
3806 | ||
3807 | entry = mk_huge_pmd(page, vma->vm_page_prot); | |
3808 | if (write) | |
f55e1014 | 3809 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); |
10102459 | 3810 | |
fadae295 | 3811 | add_mm_counter(vma->vm_mm, mm_counter_file(page), HPAGE_PMD_NR); |
10102459 | 3812 | page_add_file_rmap(page, true); |
953c66c2 AK |
3813 | /* |
3814 | * deposit and withdraw with pmd lock held | |
3815 | */ | |
3816 | if (arch_needs_pgtable_deposit()) | |
82b0f8c3 | 3817 | deposit_prealloc_pte(vmf); |
10102459 | 3818 | |
82b0f8c3 | 3819 | set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); |
10102459 | 3820 | |
82b0f8c3 | 3821 | update_mmu_cache_pmd(vma, haddr, vmf->pmd); |
10102459 KS |
3822 | |
3823 | /* fault is handled */ | |
3824 | ret = 0; | |
95ecedcd | 3825 | count_vm_event(THP_FILE_MAPPED); |
10102459 | 3826 | out: |
82b0f8c3 | 3827 | spin_unlock(vmf->ptl); |
10102459 KS |
3828 | return ret; |
3829 | } | |
3830 | #else | |
f9ce0be7 | 3831 | vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3832 | { |
f9ce0be7 | 3833 | return VM_FAULT_FALLBACK; |
10102459 KS |
3834 | } |
3835 | #endif | |
3836 | ||
9d3af4b4 | 3837 | void do_set_pte(struct vm_fault *vmf, struct page *page, unsigned long addr) |
3bb97794 | 3838 | { |
82b0f8c3 JK |
3839 | struct vm_area_struct *vma = vmf->vma; |
3840 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
9d3af4b4 | 3841 | bool prefault = vmf->address != addr; |
3bb97794 | 3842 | pte_t entry; |
7267ec00 | 3843 | |
3bb97794 KS |
3844 | flush_icache_page(vma, page); |
3845 | entry = mk_pte(page, vma->vm_page_prot); | |
46bdb427 WD |
3846 | |
3847 | if (prefault && arch_wants_old_prefaulted_pte()) | |
3848 | entry = pte_mkold(entry); | |
50c25ee9 TB |
3849 | else |
3850 | entry = pte_sw_mkyoung(entry); | |
46bdb427 | 3851 | |
3bb97794 KS |
3852 | if (write) |
3853 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
bae473a4 KS |
3854 | /* copy-on-write page */ |
3855 | if (write && !(vma->vm_flags & VM_SHARED)) { | |
3bb97794 | 3856 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
9d3af4b4 | 3857 | page_add_new_anon_rmap(page, vma, addr, false); |
b518154e | 3858 | lru_cache_add_inactive_or_unevictable(page, vma); |
3bb97794 | 3859 | } else { |
eca56ff9 | 3860 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
dd78fedd | 3861 | page_add_file_rmap(page, false); |
3bb97794 | 3862 | } |
9d3af4b4 | 3863 | set_pte_at(vma->vm_mm, addr, vmf->pte, entry); |
3bb97794 KS |
3864 | } |
3865 | ||
9118c0cb JK |
3866 | /** |
3867 | * finish_fault - finish page fault once we have prepared the page to fault | |
3868 | * | |
3869 | * @vmf: structure describing the fault | |
3870 | * | |
3871 | * This function handles all that is needed to finish a page fault once the | |
3872 | * page to fault in is prepared. It handles locking of PTEs, inserts PTE for | |
3873 | * given page, adds reverse page mapping, handles memcg charges and LRU | |
a862f68a | 3874 | * addition. |
9118c0cb JK |
3875 | * |
3876 | * The function expects the page to be locked and on success it consumes a | |
3877 | * reference of a page being mapped (for the PTE which maps it). | |
a862f68a MR |
3878 | * |
3879 | * Return: %0 on success, %VM_FAULT_ code in case of error. | |
9118c0cb | 3880 | */ |
2b740303 | 3881 | vm_fault_t finish_fault(struct vm_fault *vmf) |
9118c0cb | 3882 | { |
f9ce0be7 | 3883 | struct vm_area_struct *vma = vmf->vma; |
9118c0cb | 3884 | struct page *page; |
f9ce0be7 | 3885 | vm_fault_t ret; |
9118c0cb JK |
3886 | |
3887 | /* Did we COW the page? */ | |
f9ce0be7 | 3888 | if ((vmf->flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) |
9118c0cb JK |
3889 | page = vmf->cow_page; |
3890 | else | |
3891 | page = vmf->page; | |
6b31d595 MH |
3892 | |
3893 | /* | |
3894 | * check even for read faults because we might have lost our CoWed | |
3895 | * page | |
3896 | */ | |
f9ce0be7 KS |
3897 | if (!(vma->vm_flags & VM_SHARED)) { |
3898 | ret = check_stable_address_space(vma->vm_mm); | |
3899 | if (ret) | |
3900 | return ret; | |
3901 | } | |
3902 | ||
3903 | if (pmd_none(*vmf->pmd)) { | |
3904 | if (PageTransCompound(page)) { | |
3905 | ret = do_set_pmd(vmf, page); | |
3906 | if (ret != VM_FAULT_FALLBACK) | |
3907 | return ret; | |
3908 | } | |
3909 | ||
3910 | if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd))) | |
3911 | return VM_FAULT_OOM; | |
3912 | } | |
3913 | ||
3914 | /* See comment in handle_pte_fault() */ | |
3915 | if (pmd_devmap_trans_unstable(vmf->pmd)) | |
3916 | return 0; | |
3917 | ||
3918 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, | |
3919 | vmf->address, &vmf->ptl); | |
3920 | ret = 0; | |
3921 | /* Re-check under ptl */ | |
3922 | if (likely(pte_none(*vmf->pte))) | |
9d3af4b4 | 3923 | do_set_pte(vmf, page, vmf->address); |
f9ce0be7 KS |
3924 | else |
3925 | ret = VM_FAULT_NOPAGE; | |
3926 | ||
3927 | update_mmu_tlb(vma, vmf->address, vmf->pte); | |
3928 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
9118c0cb JK |
3929 | return ret; |
3930 | } | |
3931 | ||
3a91053a KS |
3932 | static unsigned long fault_around_bytes __read_mostly = |
3933 | rounddown_pow_of_two(65536); | |
a9b0f861 | 3934 | |
a9b0f861 KS |
3935 | #ifdef CONFIG_DEBUG_FS |
3936 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 3937 | { |
a9b0f861 | 3938 | *val = fault_around_bytes; |
1592eef0 KS |
3939 | return 0; |
3940 | } | |
3941 | ||
b4903d6e | 3942 | /* |
da391d64 WK |
3943 | * fault_around_bytes must be rounded down to the nearest page order as it's |
3944 | * what do_fault_around() expects to see. | |
b4903d6e | 3945 | */ |
a9b0f861 | 3946 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 3947 | { |
a9b0f861 | 3948 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 3949 | return -EINVAL; |
b4903d6e AR |
3950 | if (val > PAGE_SIZE) |
3951 | fault_around_bytes = rounddown_pow_of_two(val); | |
3952 | else | |
3953 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
3954 | return 0; |
3955 | } | |
0a1345f8 | 3956 | DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, |
a9b0f861 | 3957 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
1592eef0 KS |
3958 | |
3959 | static int __init fault_around_debugfs(void) | |
3960 | { | |
d9f7979c GKH |
3961 | debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, |
3962 | &fault_around_bytes_fops); | |
1592eef0 KS |
3963 | return 0; |
3964 | } | |
3965 | late_initcall(fault_around_debugfs); | |
1592eef0 | 3966 | #endif |
8c6e50b0 | 3967 | |
1fdb412b KS |
3968 | /* |
3969 | * do_fault_around() tries to map few pages around the fault address. The hope | |
3970 | * is that the pages will be needed soon and this will lower the number of | |
3971 | * faults to handle. | |
3972 | * | |
3973 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
3974 | * not ready to be mapped: not up-to-date, locked, etc. | |
3975 | * | |
3976 | * This function is called with the page table lock taken. In the split ptlock | |
3977 | * case the page table lock only protects only those entries which belong to | |
3978 | * the page table corresponding to the fault address. | |
3979 | * | |
3980 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
3981 | * only once. | |
3982 | * | |
da391d64 WK |
3983 | * fault_around_bytes defines how many bytes we'll try to map. |
3984 | * do_fault_around() expects it to be set to a power of two less than or equal | |
3985 | * to PTRS_PER_PTE. | |
1fdb412b | 3986 | * |
da391d64 WK |
3987 | * The virtual address of the area that we map is naturally aligned to |
3988 | * fault_around_bytes rounded down to the machine page size | |
3989 | * (and therefore to page order). This way it's easier to guarantee | |
3990 | * that we don't cross page table boundaries. | |
1fdb412b | 3991 | */ |
2b740303 | 3992 | static vm_fault_t do_fault_around(struct vm_fault *vmf) |
8c6e50b0 | 3993 | { |
82b0f8c3 | 3994 | unsigned long address = vmf->address, nr_pages, mask; |
0721ec8b | 3995 | pgoff_t start_pgoff = vmf->pgoff; |
bae473a4 | 3996 | pgoff_t end_pgoff; |
2b740303 | 3997 | int off; |
8c6e50b0 | 3998 | |
4db0c3c2 | 3999 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
4000 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
4001 | ||
f9ce0be7 KS |
4002 | address = max(address & mask, vmf->vma->vm_start); |
4003 | off = ((vmf->address - address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); | |
bae473a4 | 4004 | start_pgoff -= off; |
8c6e50b0 KS |
4005 | |
4006 | /* | |
da391d64 WK |
4007 | * end_pgoff is either the end of the page table, the end of |
4008 | * the vma or nr_pages from start_pgoff, depending what is nearest. | |
8c6e50b0 | 4009 | */ |
bae473a4 | 4010 | end_pgoff = start_pgoff - |
f9ce0be7 | 4011 | ((address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + |
8c6e50b0 | 4012 | PTRS_PER_PTE - 1; |
82b0f8c3 | 4013 | end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1, |
bae473a4 | 4014 | start_pgoff + nr_pages - 1); |
8c6e50b0 | 4015 | |
82b0f8c3 | 4016 | if (pmd_none(*vmf->pmd)) { |
4cf58924 | 4017 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm); |
82b0f8c3 | 4018 | if (!vmf->prealloc_pte) |
f9ce0be7 | 4019 | return VM_FAULT_OOM; |
7267ec00 | 4020 | smp_wmb(); /* See comment in __pte_alloc() */ |
8c6e50b0 KS |
4021 | } |
4022 | ||
f9ce0be7 | 4023 | return vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff); |
8c6e50b0 KS |
4024 | } |
4025 | ||
2b740303 | 4026 | static vm_fault_t do_read_fault(struct vm_fault *vmf) |
e655fb29 | 4027 | { |
82b0f8c3 | 4028 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 4029 | vm_fault_t ret = 0; |
8c6e50b0 KS |
4030 | |
4031 | /* | |
4032 | * Let's call ->map_pages() first and use ->fault() as fallback | |
4033 | * if page by the offset is not ready to be mapped (cold cache or | |
4034 | * something). | |
4035 | */ | |
9b4bdd2f | 4036 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
c949b097 AR |
4037 | if (likely(!userfaultfd_minor(vmf->vma))) { |
4038 | ret = do_fault_around(vmf); | |
4039 | if (ret) | |
4040 | return ret; | |
4041 | } | |
8c6e50b0 | 4042 | } |
e655fb29 | 4043 | |
936ca80d | 4044 | ret = __do_fault(vmf); |
e655fb29 KS |
4045 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
4046 | return ret; | |
4047 | ||
9118c0cb | 4048 | ret |= finish_fault(vmf); |
936ca80d | 4049 | unlock_page(vmf->page); |
7267ec00 | 4050 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
936ca80d | 4051 | put_page(vmf->page); |
e655fb29 KS |
4052 | return ret; |
4053 | } | |
4054 | ||
2b740303 | 4055 | static vm_fault_t do_cow_fault(struct vm_fault *vmf) |
ec47c3b9 | 4056 | { |
82b0f8c3 | 4057 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 4058 | vm_fault_t ret; |
ec47c3b9 KS |
4059 | |
4060 | if (unlikely(anon_vma_prepare(vma))) | |
4061 | return VM_FAULT_OOM; | |
4062 | ||
936ca80d JK |
4063 | vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); |
4064 | if (!vmf->cow_page) | |
ec47c3b9 KS |
4065 | return VM_FAULT_OOM; |
4066 | ||
d9eb1ea2 | 4067 | if (mem_cgroup_charge(vmf->cow_page, vma->vm_mm, GFP_KERNEL)) { |
936ca80d | 4068 | put_page(vmf->cow_page); |
ec47c3b9 KS |
4069 | return VM_FAULT_OOM; |
4070 | } | |
9d82c694 | 4071 | cgroup_throttle_swaprate(vmf->cow_page, GFP_KERNEL); |
ec47c3b9 | 4072 | |
936ca80d | 4073 | ret = __do_fault(vmf); |
ec47c3b9 KS |
4074 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
4075 | goto uncharge_out; | |
3917048d JK |
4076 | if (ret & VM_FAULT_DONE_COW) |
4077 | return ret; | |
ec47c3b9 | 4078 | |
b1aa812b | 4079 | copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); |
936ca80d | 4080 | __SetPageUptodate(vmf->cow_page); |
ec47c3b9 | 4081 | |
9118c0cb | 4082 | ret |= finish_fault(vmf); |
b1aa812b JK |
4083 | unlock_page(vmf->page); |
4084 | put_page(vmf->page); | |
7267ec00 KS |
4085 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
4086 | goto uncharge_out; | |
ec47c3b9 KS |
4087 | return ret; |
4088 | uncharge_out: | |
936ca80d | 4089 | put_page(vmf->cow_page); |
ec47c3b9 KS |
4090 | return ret; |
4091 | } | |
4092 | ||
2b740303 | 4093 | static vm_fault_t do_shared_fault(struct vm_fault *vmf) |
1da177e4 | 4094 | { |
82b0f8c3 | 4095 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 4096 | vm_fault_t ret, tmp; |
1d65f86d | 4097 | |
936ca80d | 4098 | ret = __do_fault(vmf); |
7eae74af | 4099 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 4100 | return ret; |
1da177e4 LT |
4101 | |
4102 | /* | |
f0c6d4d2 KS |
4103 | * Check if the backing address space wants to know that the page is |
4104 | * about to become writable | |
1da177e4 | 4105 | */ |
fb09a464 | 4106 | if (vma->vm_ops->page_mkwrite) { |
936ca80d | 4107 | unlock_page(vmf->page); |
38b8cb7f | 4108 | tmp = do_page_mkwrite(vmf); |
fb09a464 KS |
4109 | if (unlikely(!tmp || |
4110 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
936ca80d | 4111 | put_page(vmf->page); |
fb09a464 | 4112 | return tmp; |
4294621f | 4113 | } |
fb09a464 KS |
4114 | } |
4115 | ||
9118c0cb | 4116 | ret |= finish_fault(vmf); |
7267ec00 KS |
4117 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
4118 | VM_FAULT_RETRY))) { | |
936ca80d JK |
4119 | unlock_page(vmf->page); |
4120 | put_page(vmf->page); | |
f0c6d4d2 | 4121 | return ret; |
1da177e4 | 4122 | } |
b827e496 | 4123 | |
89b15332 | 4124 | ret |= fault_dirty_shared_page(vmf); |
1d65f86d | 4125 | return ret; |
54cb8821 | 4126 | } |
d00806b1 | 4127 | |
9a95f3cf | 4128 | /* |
c1e8d7c6 | 4129 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
9a95f3cf | 4130 | * but allow concurrent faults). |
c1e8d7c6 | 4131 | * The mmap_lock may have been released depending on flags and our |
9a95f3cf | 4132 | * return value. See filemap_fault() and __lock_page_or_retry(). |
c1e8d7c6 | 4133 | * If mmap_lock is released, vma may become invalid (for example |
fc8efd2d | 4134 | * by other thread calling munmap()). |
9a95f3cf | 4135 | */ |
2b740303 | 4136 | static vm_fault_t do_fault(struct vm_fault *vmf) |
54cb8821 | 4137 | { |
82b0f8c3 | 4138 | struct vm_area_struct *vma = vmf->vma; |
fc8efd2d | 4139 | struct mm_struct *vm_mm = vma->vm_mm; |
2b740303 | 4140 | vm_fault_t ret; |
54cb8821 | 4141 | |
ff09d7ec AK |
4142 | /* |
4143 | * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND | |
4144 | */ | |
4145 | if (!vma->vm_ops->fault) { | |
4146 | /* | |
4147 | * If we find a migration pmd entry or a none pmd entry, which | |
4148 | * should never happen, return SIGBUS | |
4149 | */ | |
4150 | if (unlikely(!pmd_present(*vmf->pmd))) | |
4151 | ret = VM_FAULT_SIGBUS; | |
4152 | else { | |
4153 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, | |
4154 | vmf->pmd, | |
4155 | vmf->address, | |
4156 | &vmf->ptl); | |
4157 | /* | |
4158 | * Make sure this is not a temporary clearing of pte | |
4159 | * by holding ptl and checking again. A R/M/W update | |
4160 | * of pte involves: take ptl, clearing the pte so that | |
4161 | * we don't have concurrent modification by hardware | |
4162 | * followed by an update. | |
4163 | */ | |
4164 | if (unlikely(pte_none(*vmf->pte))) | |
4165 | ret = VM_FAULT_SIGBUS; | |
4166 | else | |
4167 | ret = VM_FAULT_NOPAGE; | |
4168 | ||
4169 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4170 | } | |
4171 | } else if (!(vmf->flags & FAULT_FLAG_WRITE)) | |
b0b9b3df HD |
4172 | ret = do_read_fault(vmf); |
4173 | else if (!(vma->vm_flags & VM_SHARED)) | |
4174 | ret = do_cow_fault(vmf); | |
4175 | else | |
4176 | ret = do_shared_fault(vmf); | |
4177 | ||
4178 | /* preallocated pagetable is unused: free it */ | |
4179 | if (vmf->prealloc_pte) { | |
fc8efd2d | 4180 | pte_free(vm_mm, vmf->prealloc_pte); |
7f2b6ce8 | 4181 | vmf->prealloc_pte = NULL; |
b0b9b3df HD |
4182 | } |
4183 | return ret; | |
54cb8821 NP |
4184 | } |
4185 | ||
f4c0d836 YS |
4186 | int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
4187 | unsigned long addr, int page_nid, int *flags) | |
9532fec1 MG |
4188 | { |
4189 | get_page(page); | |
4190 | ||
4191 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 4192 | if (page_nid == numa_node_id()) { |
9532fec1 | 4193 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
4194 | *flags |= TNF_FAULT_LOCAL; |
4195 | } | |
9532fec1 MG |
4196 | |
4197 | return mpol_misplaced(page, vma, addr); | |
4198 | } | |
4199 | ||
2b740303 | 4200 | static vm_fault_t do_numa_page(struct vm_fault *vmf) |
d10e63f2 | 4201 | { |
82b0f8c3 | 4202 | struct vm_area_struct *vma = vmf->vma; |
4daae3b4 | 4203 | struct page *page = NULL; |
98fa15f3 | 4204 | int page_nid = NUMA_NO_NODE; |
90572890 | 4205 | int last_cpupid; |
cbee9f88 | 4206 | int target_nid; |
04a86453 | 4207 | pte_t pte, old_pte; |
288bc549 | 4208 | bool was_writable = pte_savedwrite(vmf->orig_pte); |
6688cc05 | 4209 | int flags = 0; |
d10e63f2 MG |
4210 | |
4211 | /* | |
166f61b9 TH |
4212 | * The "pte" at this point cannot be used safely without |
4213 | * validation through pte_unmap_same(). It's of NUMA type but | |
4214 | * the pfn may be screwed if the read is non atomic. | |
166f61b9 | 4215 | */ |
82b0f8c3 JK |
4216 | vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); |
4217 | spin_lock(vmf->ptl); | |
cee216a6 | 4218 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { |
82b0f8c3 | 4219 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
4220 | goto out; |
4221 | } | |
4222 | ||
b99a342d HY |
4223 | /* Get the normal PTE */ |
4224 | old_pte = ptep_get(vmf->pte); | |
04a86453 | 4225 | pte = pte_modify(old_pte, vma->vm_page_prot); |
d10e63f2 | 4226 | |
82b0f8c3 | 4227 | page = vm_normal_page(vma, vmf->address, pte); |
b99a342d HY |
4228 | if (!page) |
4229 | goto out_map; | |
d10e63f2 | 4230 | |
e81c4802 | 4231 | /* TODO: handle PTE-mapped THP */ |
b99a342d HY |
4232 | if (PageCompound(page)) |
4233 | goto out_map; | |
e81c4802 | 4234 | |
6688cc05 | 4235 | /* |
bea66fbd MG |
4236 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
4237 | * much anyway since they can be in shared cache state. This misses | |
4238 | * the case where a mapping is writable but the process never writes | |
4239 | * to it but pte_write gets cleared during protection updates and | |
4240 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
4241 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 4242 | */ |
b99a342d | 4243 | if (!was_writable) |
6688cc05 PZ |
4244 | flags |= TNF_NO_GROUP; |
4245 | ||
dabe1d99 RR |
4246 | /* |
4247 | * Flag if the page is shared between multiple address spaces. This | |
4248 | * is later used when determining whether to group tasks together | |
4249 | */ | |
4250 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
4251 | flags |= TNF_SHARED; | |
4252 | ||
90572890 | 4253 | last_cpupid = page_cpupid_last(page); |
8191acbd | 4254 | page_nid = page_to_nid(page); |
82b0f8c3 | 4255 | target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, |
bae473a4 | 4256 | &flags); |
98fa15f3 | 4257 | if (target_nid == NUMA_NO_NODE) { |
4daae3b4 | 4258 | put_page(page); |
b99a342d | 4259 | goto out_map; |
4daae3b4 | 4260 | } |
b99a342d | 4261 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
4262 | |
4263 | /* Migrate to the requested node */ | |
bf90ac19 | 4264 | if (migrate_misplaced_page(page, vma, target_nid)) { |
8191acbd | 4265 | page_nid = target_nid; |
6688cc05 | 4266 | flags |= TNF_MIGRATED; |
b99a342d | 4267 | } else { |
074c2381 | 4268 | flags |= TNF_MIGRATE_FAIL; |
b99a342d HY |
4269 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
4270 | spin_lock(vmf->ptl); | |
4271 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { | |
4272 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4273 | goto out; | |
4274 | } | |
4275 | goto out_map; | |
4276 | } | |
4daae3b4 MG |
4277 | |
4278 | out: | |
98fa15f3 | 4279 | if (page_nid != NUMA_NO_NODE) |
6688cc05 | 4280 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 | 4281 | return 0; |
b99a342d HY |
4282 | out_map: |
4283 | /* | |
4284 | * Make it present again, depending on how arch implements | |
4285 | * non-accessible ptes, some can allow access by kernel mode. | |
4286 | */ | |
4287 | old_pte = ptep_modify_prot_start(vma, vmf->address, vmf->pte); | |
4288 | pte = pte_modify(old_pte, vma->vm_page_prot); | |
4289 | pte = pte_mkyoung(pte); | |
4290 | if (was_writable) | |
4291 | pte = pte_mkwrite(pte); | |
4292 | ptep_modify_prot_commit(vma, vmf->address, vmf->pte, old_pte, pte); | |
4293 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
4294 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4295 | goto out; | |
d10e63f2 MG |
4296 | } |
4297 | ||
2b740303 | 4298 | static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 4299 | { |
f4200391 | 4300 | if (vma_is_anonymous(vmf->vma)) |
82b0f8c3 | 4301 | return do_huge_pmd_anonymous_page(vmf); |
a2d58167 | 4302 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 4303 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
b96375f7 MW |
4304 | return VM_FAULT_FALLBACK; |
4305 | } | |
4306 | ||
183f24aa | 4307 | /* `inline' is required to avoid gcc 4.1.2 build error */ |
5db4f15c | 4308 | static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 4309 | { |
529b930b | 4310 | if (vma_is_anonymous(vmf->vma)) { |
5db4f15c | 4311 | if (userfaultfd_huge_pmd_wp(vmf->vma, vmf->orig_pmd)) |
529b930b | 4312 | return handle_userfault(vmf, VM_UFFD_WP); |
5db4f15c | 4313 | return do_huge_pmd_wp_page(vmf); |
529b930b | 4314 | } |
327e9fd4 THV |
4315 | if (vmf->vma->vm_ops->huge_fault) { |
4316 | vm_fault_t ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); | |
4317 | ||
4318 | if (!(ret & VM_FAULT_FALLBACK)) | |
4319 | return ret; | |
4320 | } | |
af9e4d5f | 4321 | |
327e9fd4 | 4322 | /* COW or write-notify handled on pte level: split pmd. */ |
82b0f8c3 | 4323 | __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); |
af9e4d5f | 4324 | |
b96375f7 MW |
4325 | return VM_FAULT_FALLBACK; |
4326 | } | |
4327 | ||
2b740303 | 4328 | static vm_fault_t create_huge_pud(struct vm_fault *vmf) |
a00cc7d9 | 4329 | { |
327e9fd4 THV |
4330 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ |
4331 | defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) | |
a00cc7d9 MW |
4332 | /* No support for anonymous transparent PUD pages yet */ |
4333 | if (vma_is_anonymous(vmf->vma)) | |
327e9fd4 THV |
4334 | goto split; |
4335 | if (vmf->vma->vm_ops->huge_fault) { | |
4336 | vm_fault_t ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); | |
4337 | ||
4338 | if (!(ret & VM_FAULT_FALLBACK)) | |
4339 | return ret; | |
4340 | } | |
4341 | split: | |
4342 | /* COW or write-notify not handled on PUD level: split pud.*/ | |
4343 | __split_huge_pud(vmf->vma, vmf->pud, vmf->address); | |
a00cc7d9 MW |
4344 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
4345 | return VM_FAULT_FALLBACK; | |
4346 | } | |
4347 | ||
2b740303 | 4348 | static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) |
a00cc7d9 MW |
4349 | { |
4350 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4351 | /* No support for anonymous transparent PUD pages yet */ | |
4352 | if (vma_is_anonymous(vmf->vma)) | |
4353 | return VM_FAULT_FALLBACK; | |
4354 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 4355 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
4356 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
4357 | return VM_FAULT_FALLBACK; | |
4358 | } | |
4359 | ||
1da177e4 LT |
4360 | /* |
4361 | * These routines also need to handle stuff like marking pages dirty | |
4362 | * and/or accessed for architectures that don't do it in hardware (most | |
4363 | * RISC architectures). The early dirtying is also good on the i386. | |
4364 | * | |
4365 | * There is also a hook called "update_mmu_cache()" that architectures | |
4366 | * with external mmu caches can use to update those (ie the Sparc or | |
4367 | * PowerPC hashed page tables that act as extended TLBs). | |
4368 | * | |
c1e8d7c6 | 4369 | * We enter with non-exclusive mmap_lock (to exclude vma changes, but allow |
7267ec00 | 4370 | * concurrent faults). |
9a95f3cf | 4371 | * |
c1e8d7c6 | 4372 | * The mmap_lock may have been released depending on flags and our return value. |
7267ec00 | 4373 | * See filemap_fault() and __lock_page_or_retry(). |
1da177e4 | 4374 | */ |
2b740303 | 4375 | static vm_fault_t handle_pte_fault(struct vm_fault *vmf) |
1da177e4 LT |
4376 | { |
4377 | pte_t entry; | |
4378 | ||
82b0f8c3 | 4379 | if (unlikely(pmd_none(*vmf->pmd))) { |
7267ec00 KS |
4380 | /* |
4381 | * Leave __pte_alloc() until later: because vm_ops->fault may | |
4382 | * want to allocate huge page, and if we expose page table | |
4383 | * for an instant, it will be difficult to retract from | |
4384 | * concurrent faults and from rmap lookups. | |
4385 | */ | |
82b0f8c3 | 4386 | vmf->pte = NULL; |
7267ec00 | 4387 | } else { |
f9ce0be7 KS |
4388 | /* |
4389 | * If a huge pmd materialized under us just retry later. Use | |
4390 | * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead | |
4391 | * of pmd_trans_huge() to ensure the pmd didn't become | |
4392 | * pmd_trans_huge under us and then back to pmd_none, as a | |
4393 | * result of MADV_DONTNEED running immediately after a huge pmd | |
4394 | * fault in a different thread of this mm, in turn leading to a | |
4395 | * misleading pmd_trans_huge() retval. All we have to ensure is | |
4396 | * that it is a regular pmd that we can walk with | |
4397 | * pte_offset_map() and we can do that through an atomic read | |
4398 | * in C, which is what pmd_trans_unstable() provides. | |
4399 | */ | |
d0f0931d | 4400 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
4401 | return 0; |
4402 | /* | |
4403 | * A regular pmd is established and it can't morph into a huge | |
4404 | * pmd from under us anymore at this point because we hold the | |
c1e8d7c6 | 4405 | * mmap_lock read mode and khugepaged takes it in write mode. |
7267ec00 KS |
4406 | * So now it's safe to run pte_offset_map(). |
4407 | */ | |
82b0f8c3 | 4408 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
2994302b | 4409 | vmf->orig_pte = *vmf->pte; |
7267ec00 KS |
4410 | |
4411 | /* | |
4412 | * some architectures can have larger ptes than wordsize, | |
4413 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and | |
b03a0fe0 PM |
4414 | * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic |
4415 | * accesses. The code below just needs a consistent view | |
4416 | * for the ifs and we later double check anyway with the | |
7267ec00 KS |
4417 | * ptl lock held. So here a barrier will do. |
4418 | */ | |
4419 | barrier(); | |
2994302b | 4420 | if (pte_none(vmf->orig_pte)) { |
82b0f8c3 JK |
4421 | pte_unmap(vmf->pte); |
4422 | vmf->pte = NULL; | |
65500d23 | 4423 | } |
1da177e4 LT |
4424 | } |
4425 | ||
82b0f8c3 JK |
4426 | if (!vmf->pte) { |
4427 | if (vma_is_anonymous(vmf->vma)) | |
4428 | return do_anonymous_page(vmf); | |
7267ec00 | 4429 | else |
82b0f8c3 | 4430 | return do_fault(vmf); |
7267ec00 KS |
4431 | } |
4432 | ||
2994302b JK |
4433 | if (!pte_present(vmf->orig_pte)) |
4434 | return do_swap_page(vmf); | |
7267ec00 | 4435 | |
2994302b JK |
4436 | if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) |
4437 | return do_numa_page(vmf); | |
d10e63f2 | 4438 | |
82b0f8c3 JK |
4439 | vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); |
4440 | spin_lock(vmf->ptl); | |
2994302b | 4441 | entry = vmf->orig_pte; |
7df67697 BM |
4442 | if (unlikely(!pte_same(*vmf->pte, entry))) { |
4443 | update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); | |
8f4e2101 | 4444 | goto unlock; |
7df67697 | 4445 | } |
82b0f8c3 | 4446 | if (vmf->flags & FAULT_FLAG_WRITE) { |
f6f37321 | 4447 | if (!pte_write(entry)) |
2994302b | 4448 | return do_wp_page(vmf); |
1da177e4 LT |
4449 | entry = pte_mkdirty(entry); |
4450 | } | |
4451 | entry = pte_mkyoung(entry); | |
82b0f8c3 JK |
4452 | if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, |
4453 | vmf->flags & FAULT_FLAG_WRITE)) { | |
4454 | update_mmu_cache(vmf->vma, vmf->address, vmf->pte); | |
1a44e149 | 4455 | } else { |
b7333b58 YS |
4456 | /* Skip spurious TLB flush for retried page fault */ |
4457 | if (vmf->flags & FAULT_FLAG_TRIED) | |
4458 | goto unlock; | |
1a44e149 AA |
4459 | /* |
4460 | * This is needed only for protection faults but the arch code | |
4461 | * is not yet telling us if this is a protection fault or not. | |
4462 | * This still avoids useless tlb flushes for .text page faults | |
4463 | * with threads. | |
4464 | */ | |
82b0f8c3 JK |
4465 | if (vmf->flags & FAULT_FLAG_WRITE) |
4466 | flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); | |
1a44e149 | 4467 | } |
8f4e2101 | 4468 | unlock: |
82b0f8c3 | 4469 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 4470 | return 0; |
1da177e4 LT |
4471 | } |
4472 | ||
4473 | /* | |
4474 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf | 4475 | * |
c1e8d7c6 | 4476 | * The mmap_lock may have been released depending on flags and our |
9a95f3cf | 4477 | * return value. See filemap_fault() and __lock_page_or_retry(). |
1da177e4 | 4478 | */ |
2b740303 SJ |
4479 | static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma, |
4480 | unsigned long address, unsigned int flags) | |
1da177e4 | 4481 | { |
82b0f8c3 | 4482 | struct vm_fault vmf = { |
bae473a4 | 4483 | .vma = vma, |
1a29d85e | 4484 | .address = address & PAGE_MASK, |
bae473a4 | 4485 | .flags = flags, |
0721ec8b | 4486 | .pgoff = linear_page_index(vma, address), |
667240e0 | 4487 | .gfp_mask = __get_fault_gfp_mask(vma), |
bae473a4 | 4488 | }; |
fde26bed | 4489 | unsigned int dirty = flags & FAULT_FLAG_WRITE; |
dcddffd4 | 4490 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 4491 | pgd_t *pgd; |
c2febafc | 4492 | p4d_t *p4d; |
2b740303 | 4493 | vm_fault_t ret; |
1da177e4 | 4494 | |
1da177e4 | 4495 | pgd = pgd_offset(mm, address); |
c2febafc KS |
4496 | p4d = p4d_alloc(mm, pgd, address); |
4497 | if (!p4d) | |
4498 | return VM_FAULT_OOM; | |
a00cc7d9 | 4499 | |
c2febafc | 4500 | vmf.pud = pud_alloc(mm, p4d, address); |
a00cc7d9 | 4501 | if (!vmf.pud) |
c74df32c | 4502 | return VM_FAULT_OOM; |
625110b5 | 4503 | retry_pud: |
7635d9cb | 4504 | if (pud_none(*vmf.pud) && __transparent_hugepage_enabled(vma)) { |
a00cc7d9 MW |
4505 | ret = create_huge_pud(&vmf); |
4506 | if (!(ret & VM_FAULT_FALLBACK)) | |
4507 | return ret; | |
4508 | } else { | |
4509 | pud_t orig_pud = *vmf.pud; | |
4510 | ||
4511 | barrier(); | |
4512 | if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { | |
a00cc7d9 | 4513 | |
a00cc7d9 MW |
4514 | /* NUMA case for anonymous PUDs would go here */ |
4515 | ||
f6f37321 | 4516 | if (dirty && !pud_write(orig_pud)) { |
a00cc7d9 MW |
4517 | ret = wp_huge_pud(&vmf, orig_pud); |
4518 | if (!(ret & VM_FAULT_FALLBACK)) | |
4519 | return ret; | |
4520 | } else { | |
4521 | huge_pud_set_accessed(&vmf, orig_pud); | |
4522 | return 0; | |
4523 | } | |
4524 | } | |
4525 | } | |
4526 | ||
4527 | vmf.pmd = pmd_alloc(mm, vmf.pud, address); | |
82b0f8c3 | 4528 | if (!vmf.pmd) |
c74df32c | 4529 | return VM_FAULT_OOM; |
625110b5 TH |
4530 | |
4531 | /* Huge pud page fault raced with pmd_alloc? */ | |
4532 | if (pud_trans_unstable(vmf.pud)) | |
4533 | goto retry_pud; | |
4534 | ||
7635d9cb | 4535 | if (pmd_none(*vmf.pmd) && __transparent_hugepage_enabled(vma)) { |
a2d58167 | 4536 | ret = create_huge_pmd(&vmf); |
c0292554 KS |
4537 | if (!(ret & VM_FAULT_FALLBACK)) |
4538 | return ret; | |
71e3aac0 | 4539 | } else { |
5db4f15c | 4540 | vmf.orig_pmd = *vmf.pmd; |
1f1d06c3 | 4541 | |
71e3aac0 | 4542 | barrier(); |
5db4f15c | 4543 | if (unlikely(is_swap_pmd(vmf.orig_pmd))) { |
84c3fc4e | 4544 | VM_BUG_ON(thp_migration_supported() && |
5db4f15c YS |
4545 | !is_pmd_migration_entry(vmf.orig_pmd)); |
4546 | if (is_pmd_migration_entry(vmf.orig_pmd)) | |
84c3fc4e ZY |
4547 | pmd_migration_entry_wait(mm, vmf.pmd); |
4548 | return 0; | |
4549 | } | |
5db4f15c YS |
4550 | if (pmd_trans_huge(vmf.orig_pmd) || pmd_devmap(vmf.orig_pmd)) { |
4551 | if (pmd_protnone(vmf.orig_pmd) && vma_is_accessible(vma)) | |
4552 | return do_huge_pmd_numa_page(&vmf); | |
d10e63f2 | 4553 | |
5db4f15c YS |
4554 | if (dirty && !pmd_write(vmf.orig_pmd)) { |
4555 | ret = wp_huge_pmd(&vmf); | |
9845cbbd KS |
4556 | if (!(ret & VM_FAULT_FALLBACK)) |
4557 | return ret; | |
a1dd450b | 4558 | } else { |
5db4f15c | 4559 | huge_pmd_set_accessed(&vmf); |
9845cbbd | 4560 | return 0; |
1f1d06c3 | 4561 | } |
71e3aac0 AA |
4562 | } |
4563 | } | |
4564 | ||
82b0f8c3 | 4565 | return handle_pte_fault(&vmf); |
1da177e4 LT |
4566 | } |
4567 | ||
bce617ed | 4568 | /** |
f0953a1b | 4569 | * mm_account_fault - Do page fault accounting |
bce617ed PX |
4570 | * |
4571 | * @regs: the pt_regs struct pointer. When set to NULL, will skip accounting | |
4572 | * of perf event counters, but we'll still do the per-task accounting to | |
4573 | * the task who triggered this page fault. | |
4574 | * @address: the faulted address. | |
4575 | * @flags: the fault flags. | |
4576 | * @ret: the fault retcode. | |
4577 | * | |
f0953a1b | 4578 | * This will take care of most of the page fault accounting. Meanwhile, it |
bce617ed | 4579 | * will also include the PERF_COUNT_SW_PAGE_FAULTS_[MAJ|MIN] perf counter |
f0953a1b | 4580 | * updates. However, note that the handling of PERF_COUNT_SW_PAGE_FAULTS should |
bce617ed PX |
4581 | * still be in per-arch page fault handlers at the entry of page fault. |
4582 | */ | |
4583 | static inline void mm_account_fault(struct pt_regs *regs, | |
4584 | unsigned long address, unsigned int flags, | |
4585 | vm_fault_t ret) | |
4586 | { | |
4587 | bool major; | |
4588 | ||
4589 | /* | |
4590 | * We don't do accounting for some specific faults: | |
4591 | * | |
4592 | * - Unsuccessful faults (e.g. when the address wasn't valid). That | |
4593 | * includes arch_vma_access_permitted() failing before reaching here. | |
4594 | * So this is not a "this many hardware page faults" counter. We | |
4595 | * should use the hw profiling for that. | |
4596 | * | |
4597 | * - Incomplete faults (VM_FAULT_RETRY). They will only be counted | |
4598 | * once they're completed. | |
4599 | */ | |
4600 | if (ret & (VM_FAULT_ERROR | VM_FAULT_RETRY)) | |
4601 | return; | |
4602 | ||
4603 | /* | |
4604 | * We define the fault as a major fault when the final successful fault | |
4605 | * is VM_FAULT_MAJOR, or if it retried (which implies that we couldn't | |
4606 | * handle it immediately previously). | |
4607 | */ | |
4608 | major = (ret & VM_FAULT_MAJOR) || (flags & FAULT_FLAG_TRIED); | |
4609 | ||
a2beb5f1 PX |
4610 | if (major) |
4611 | current->maj_flt++; | |
4612 | else | |
4613 | current->min_flt++; | |
4614 | ||
bce617ed | 4615 | /* |
a2beb5f1 PX |
4616 | * If the fault is done for GUP, regs will be NULL. We only do the |
4617 | * accounting for the per thread fault counters who triggered the | |
4618 | * fault, and we skip the perf event updates. | |
bce617ed PX |
4619 | */ |
4620 | if (!regs) | |
4621 | return; | |
4622 | ||
a2beb5f1 | 4623 | if (major) |
bce617ed | 4624 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); |
a2beb5f1 | 4625 | else |
bce617ed | 4626 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); |
bce617ed PX |
4627 | } |
4628 | ||
9a95f3cf PC |
4629 | /* |
4630 | * By the time we get here, we already hold the mm semaphore | |
4631 | * | |
c1e8d7c6 | 4632 | * The mmap_lock may have been released depending on flags and our |
9a95f3cf PC |
4633 | * return value. See filemap_fault() and __lock_page_or_retry(). |
4634 | */ | |
2b740303 | 4635 | vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
bce617ed | 4636 | unsigned int flags, struct pt_regs *regs) |
519e5247 | 4637 | { |
2b740303 | 4638 | vm_fault_t ret; |
519e5247 JW |
4639 | |
4640 | __set_current_state(TASK_RUNNING); | |
4641 | ||
4642 | count_vm_event(PGFAULT); | |
2262185c | 4643 | count_memcg_event_mm(vma->vm_mm, PGFAULT); |
519e5247 JW |
4644 | |
4645 | /* do counter updates before entering really critical section. */ | |
4646 | check_sync_rss_stat(current); | |
4647 | ||
de0c799b LD |
4648 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
4649 | flags & FAULT_FLAG_INSTRUCTION, | |
4650 | flags & FAULT_FLAG_REMOTE)) | |
4651 | return VM_FAULT_SIGSEGV; | |
4652 | ||
519e5247 JW |
4653 | /* |
4654 | * Enable the memcg OOM handling for faults triggered in user | |
4655 | * space. Kernel faults are handled more gracefully. | |
4656 | */ | |
4657 | if (flags & FAULT_FLAG_USER) | |
29ef680a | 4658 | mem_cgroup_enter_user_fault(); |
519e5247 | 4659 | |
bae473a4 KS |
4660 | if (unlikely(is_vm_hugetlb_page(vma))) |
4661 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); | |
4662 | else | |
4663 | ret = __handle_mm_fault(vma, address, flags); | |
519e5247 | 4664 | |
49426420 | 4665 | if (flags & FAULT_FLAG_USER) { |
29ef680a | 4666 | mem_cgroup_exit_user_fault(); |
166f61b9 TH |
4667 | /* |
4668 | * The task may have entered a memcg OOM situation but | |
4669 | * if the allocation error was handled gracefully (no | |
4670 | * VM_FAULT_OOM), there is no need to kill anything. | |
4671 | * Just clean up the OOM state peacefully. | |
4672 | */ | |
4673 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
4674 | mem_cgroup_oom_synchronize(false); | |
49426420 | 4675 | } |
3812c8c8 | 4676 | |
bce617ed PX |
4677 | mm_account_fault(regs, address, flags, ret); |
4678 | ||
519e5247 JW |
4679 | return ret; |
4680 | } | |
e1d6d01a | 4681 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 4682 | |
90eceff1 KS |
4683 | #ifndef __PAGETABLE_P4D_FOLDED |
4684 | /* | |
4685 | * Allocate p4d page table. | |
4686 | * We've already handled the fast-path in-line. | |
4687 | */ | |
4688 | int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
4689 | { | |
4690 | p4d_t *new = p4d_alloc_one(mm, address); | |
4691 | if (!new) | |
4692 | return -ENOMEM; | |
4693 | ||
4694 | smp_wmb(); /* See comment in __pte_alloc */ | |
4695 | ||
4696 | spin_lock(&mm->page_table_lock); | |
4697 | if (pgd_present(*pgd)) /* Another has populated it */ | |
4698 | p4d_free(mm, new); | |
4699 | else | |
4700 | pgd_populate(mm, pgd, new); | |
4701 | spin_unlock(&mm->page_table_lock); | |
4702 | return 0; | |
4703 | } | |
4704 | #endif /* __PAGETABLE_P4D_FOLDED */ | |
4705 | ||
1da177e4 LT |
4706 | #ifndef __PAGETABLE_PUD_FOLDED |
4707 | /* | |
4708 | * Allocate page upper directory. | |
872fec16 | 4709 | * We've already handled the fast-path in-line. |
1da177e4 | 4710 | */ |
c2febafc | 4711 | int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) |
1da177e4 | 4712 | { |
c74df32c HD |
4713 | pud_t *new = pud_alloc_one(mm, address); |
4714 | if (!new) | |
1bb3630e | 4715 | return -ENOMEM; |
1da177e4 | 4716 | |
362a61ad NP |
4717 | smp_wmb(); /* See comment in __pte_alloc */ |
4718 | ||
872fec16 | 4719 | spin_lock(&mm->page_table_lock); |
b4e98d9a KS |
4720 | if (!p4d_present(*p4d)) { |
4721 | mm_inc_nr_puds(mm); | |
c2febafc | 4722 | p4d_populate(mm, p4d, new); |
b4e98d9a | 4723 | } else /* Another has populated it */ |
5e541973 | 4724 | pud_free(mm, new); |
c74df32c | 4725 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 4726 | return 0; |
1da177e4 LT |
4727 | } |
4728 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
4729 | ||
4730 | #ifndef __PAGETABLE_PMD_FOLDED | |
4731 | /* | |
4732 | * Allocate page middle directory. | |
872fec16 | 4733 | * We've already handled the fast-path in-line. |
1da177e4 | 4734 | */ |
1bb3630e | 4735 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 4736 | { |
a00cc7d9 | 4737 | spinlock_t *ptl; |
c74df32c HD |
4738 | pmd_t *new = pmd_alloc_one(mm, address); |
4739 | if (!new) | |
1bb3630e | 4740 | return -ENOMEM; |
1da177e4 | 4741 | |
362a61ad NP |
4742 | smp_wmb(); /* See comment in __pte_alloc */ |
4743 | ||
a00cc7d9 | 4744 | ptl = pud_lock(mm, pud); |
dc6c9a35 KS |
4745 | if (!pud_present(*pud)) { |
4746 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4747 | pud_populate(mm, pud, new); |
dc6c9a35 | 4748 | } else /* Another has populated it */ |
5e541973 | 4749 | pmd_free(mm, new); |
a00cc7d9 | 4750 | spin_unlock(ptl); |
1bb3630e | 4751 | return 0; |
e0f39591 | 4752 | } |
1da177e4 LT |
4753 | #endif /* __PAGETABLE_PMD_FOLDED */ |
4754 | ||
9fd6dad1 PB |
4755 | int follow_invalidate_pte(struct mm_struct *mm, unsigned long address, |
4756 | struct mmu_notifier_range *range, pte_t **ptepp, | |
4757 | pmd_t **pmdpp, spinlock_t **ptlp) | |
f8ad0f49 JW |
4758 | { |
4759 | pgd_t *pgd; | |
c2febafc | 4760 | p4d_t *p4d; |
f8ad0f49 JW |
4761 | pud_t *pud; |
4762 | pmd_t *pmd; | |
4763 | pte_t *ptep; | |
4764 | ||
4765 | pgd = pgd_offset(mm, address); | |
4766 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
4767 | goto out; | |
4768 | ||
c2febafc KS |
4769 | p4d = p4d_offset(pgd, address); |
4770 | if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) | |
4771 | goto out; | |
4772 | ||
4773 | pud = pud_offset(p4d, address); | |
f8ad0f49 JW |
4774 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
4775 | goto out; | |
4776 | ||
4777 | pmd = pmd_offset(pud, address); | |
f66055ab | 4778 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 | 4779 | |
09796395 RZ |
4780 | if (pmd_huge(*pmd)) { |
4781 | if (!pmdpp) | |
4782 | goto out; | |
4783 | ||
ac46d4f3 | 4784 | if (range) { |
7269f999 | 4785 | mmu_notifier_range_init(range, MMU_NOTIFY_CLEAR, 0, |
6f4f13e8 JG |
4786 | NULL, mm, address & PMD_MASK, |
4787 | (address & PMD_MASK) + PMD_SIZE); | |
ac46d4f3 | 4788 | mmu_notifier_invalidate_range_start(range); |
a4d1a885 | 4789 | } |
09796395 RZ |
4790 | *ptlp = pmd_lock(mm, pmd); |
4791 | if (pmd_huge(*pmd)) { | |
4792 | *pmdpp = pmd; | |
4793 | return 0; | |
4794 | } | |
4795 | spin_unlock(*ptlp); | |
ac46d4f3 JG |
4796 | if (range) |
4797 | mmu_notifier_invalidate_range_end(range); | |
09796395 RZ |
4798 | } |
4799 | ||
4800 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
f8ad0f49 JW |
4801 | goto out; |
4802 | ||
ac46d4f3 | 4803 | if (range) { |
7269f999 | 4804 | mmu_notifier_range_init(range, MMU_NOTIFY_CLEAR, 0, NULL, mm, |
6f4f13e8 JG |
4805 | address & PAGE_MASK, |
4806 | (address & PAGE_MASK) + PAGE_SIZE); | |
ac46d4f3 | 4807 | mmu_notifier_invalidate_range_start(range); |
a4d1a885 | 4808 | } |
f8ad0f49 | 4809 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
f8ad0f49 JW |
4810 | if (!pte_present(*ptep)) |
4811 | goto unlock; | |
4812 | *ptepp = ptep; | |
4813 | return 0; | |
4814 | unlock: | |
4815 | pte_unmap_unlock(ptep, *ptlp); | |
ac46d4f3 JG |
4816 | if (range) |
4817 | mmu_notifier_invalidate_range_end(range); | |
f8ad0f49 JW |
4818 | out: |
4819 | return -EINVAL; | |
4820 | } | |
4821 | ||
9fd6dad1 PB |
4822 | /** |
4823 | * follow_pte - look up PTE at a user virtual address | |
4824 | * @mm: the mm_struct of the target address space | |
4825 | * @address: user virtual address | |
4826 | * @ptepp: location to store found PTE | |
4827 | * @ptlp: location to store the lock for the PTE | |
4828 | * | |
4829 | * On a successful return, the pointer to the PTE is stored in @ptepp; | |
4830 | * the corresponding lock is taken and its location is stored in @ptlp. | |
4831 | * The contents of the PTE are only stable until @ptlp is released; | |
4832 | * any further use, if any, must be protected against invalidation | |
4833 | * with MMU notifiers. | |
4834 | * | |
4835 | * Only IO mappings and raw PFN mappings are allowed. The mmap semaphore | |
4836 | * should be taken for read. | |
4837 | * | |
4838 | * KVM uses this function. While it is arguably less bad than ``follow_pfn``, | |
4839 | * it is not a good general-purpose API. | |
4840 | * | |
4841 | * Return: zero on success, -ve otherwise. | |
4842 | */ | |
4843 | int follow_pte(struct mm_struct *mm, unsigned long address, | |
4844 | pte_t **ptepp, spinlock_t **ptlp) | |
4845 | { | |
4846 | return follow_invalidate_pte(mm, address, NULL, ptepp, NULL, ptlp); | |
4847 | } | |
4848 | EXPORT_SYMBOL_GPL(follow_pte); | |
4849 | ||
3b6748e2 JW |
4850 | /** |
4851 | * follow_pfn - look up PFN at a user virtual address | |
4852 | * @vma: memory mapping | |
4853 | * @address: user virtual address | |
4854 | * @pfn: location to store found PFN | |
4855 | * | |
4856 | * Only IO mappings and raw PFN mappings are allowed. | |
4857 | * | |
9fd6dad1 PB |
4858 | * This function does not allow the caller to read the permissions |
4859 | * of the PTE. Do not use it. | |
4860 | * | |
a862f68a | 4861 | * Return: zero and the pfn at @pfn on success, -ve otherwise. |
3b6748e2 JW |
4862 | */ |
4863 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
4864 | unsigned long *pfn) | |
4865 | { | |
4866 | int ret = -EINVAL; | |
4867 | spinlock_t *ptl; | |
4868 | pte_t *ptep; | |
4869 | ||
4870 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
4871 | return ret; | |
4872 | ||
9fd6dad1 | 4873 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); |
3b6748e2 JW |
4874 | if (ret) |
4875 | return ret; | |
4876 | *pfn = pte_pfn(*ptep); | |
4877 | pte_unmap_unlock(ptep, ptl); | |
4878 | return 0; | |
4879 | } | |
4880 | EXPORT_SYMBOL(follow_pfn); | |
4881 | ||
28b2ee20 | 4882 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 4883 | int follow_phys(struct vm_area_struct *vma, |
4884 | unsigned long address, unsigned int flags, | |
4885 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 4886 | { |
03668a4d | 4887 | int ret = -EINVAL; |
28b2ee20 RR |
4888 | pte_t *ptep, pte; |
4889 | spinlock_t *ptl; | |
28b2ee20 | 4890 | |
d87fe660 | 4891 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
4892 | goto out; | |
28b2ee20 | 4893 | |
9fd6dad1 | 4894 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 4895 | goto out; |
28b2ee20 | 4896 | pte = *ptep; |
03668a4d | 4897 | |
f6f37321 | 4898 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
28b2ee20 | 4899 | goto unlock; |
28b2ee20 RR |
4900 | |
4901 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 4902 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 4903 | |
03668a4d | 4904 | ret = 0; |
28b2ee20 RR |
4905 | unlock: |
4906 | pte_unmap_unlock(ptep, ptl); | |
4907 | out: | |
d87fe660 | 4908 | return ret; |
28b2ee20 RR |
4909 | } |
4910 | ||
96667f8a DV |
4911 | /** |
4912 | * generic_access_phys - generic implementation for iomem mmap access | |
4913 | * @vma: the vma to access | |
f0953a1b | 4914 | * @addr: userspace address, not relative offset within @vma |
96667f8a DV |
4915 | * @buf: buffer to read/write |
4916 | * @len: length of transfer | |
4917 | * @write: set to FOLL_WRITE when writing, otherwise reading | |
4918 | * | |
4919 | * This is a generic implementation for &vm_operations_struct.access for an | |
4920 | * iomem mapping. This callback is used by access_process_vm() when the @vma is | |
4921 | * not page based. | |
4922 | */ | |
28b2ee20 RR |
4923 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, |
4924 | void *buf, int len, int write) | |
4925 | { | |
4926 | resource_size_t phys_addr; | |
4927 | unsigned long prot = 0; | |
2bc7273b | 4928 | void __iomem *maddr; |
96667f8a DV |
4929 | pte_t *ptep, pte; |
4930 | spinlock_t *ptl; | |
4931 | int offset = offset_in_page(addr); | |
4932 | int ret = -EINVAL; | |
4933 | ||
4934 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
4935 | return -EINVAL; | |
4936 | ||
4937 | retry: | |
e913a8cd | 4938 | if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) |
96667f8a DV |
4939 | return -EINVAL; |
4940 | pte = *ptep; | |
4941 | pte_unmap_unlock(ptep, ptl); | |
28b2ee20 | 4942 | |
96667f8a DV |
4943 | prot = pgprot_val(pte_pgprot(pte)); |
4944 | phys_addr = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; | |
4945 | ||
4946 | if ((write & FOLL_WRITE) && !pte_write(pte)) | |
28b2ee20 RR |
4947 | return -EINVAL; |
4948 | ||
9cb12d7b | 4949 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
24eee1e4 | 4950 | if (!maddr) |
4951 | return -ENOMEM; | |
4952 | ||
e913a8cd | 4953 | if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) |
96667f8a DV |
4954 | goto out_unmap; |
4955 | ||
4956 | if (!pte_same(pte, *ptep)) { | |
4957 | pte_unmap_unlock(ptep, ptl); | |
4958 | iounmap(maddr); | |
4959 | ||
4960 | goto retry; | |
4961 | } | |
4962 | ||
28b2ee20 RR |
4963 | if (write) |
4964 | memcpy_toio(maddr + offset, buf, len); | |
4965 | else | |
4966 | memcpy_fromio(buf, maddr + offset, len); | |
96667f8a DV |
4967 | ret = len; |
4968 | pte_unmap_unlock(ptep, ptl); | |
4969 | out_unmap: | |
28b2ee20 RR |
4970 | iounmap(maddr); |
4971 | ||
96667f8a | 4972 | return ret; |
28b2ee20 | 4973 | } |
5a73633e | 4974 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
4975 | #endif |
4976 | ||
0ec76a11 | 4977 | /* |
d3f5ffca | 4978 | * Access another process' address space as given in mm. |
0ec76a11 | 4979 | */ |
d3f5ffca JH |
4980 | int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf, |
4981 | int len, unsigned int gup_flags) | |
0ec76a11 | 4982 | { |
0ec76a11 | 4983 | struct vm_area_struct *vma; |
0ec76a11 | 4984 | void *old_buf = buf; |
442486ec | 4985 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 4986 | |
d8ed45c5 | 4987 | if (mmap_read_lock_killable(mm)) |
1e426fe2 KK |
4988 | return 0; |
4989 | ||
183ff22b | 4990 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
4991 | while (len) { |
4992 | int bytes, ret, offset; | |
4993 | void *maddr; | |
28b2ee20 | 4994 | struct page *page = NULL; |
0ec76a11 | 4995 | |
64019a2e | 4996 | ret = get_user_pages_remote(mm, addr, 1, |
5b56d49f | 4997 | gup_flags, &page, &vma, NULL); |
28b2ee20 | 4998 | if (ret <= 0) { |
dbffcd03 RR |
4999 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
5000 | break; | |
5001 | #else | |
28b2ee20 RR |
5002 | /* |
5003 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
5004 | * we can access using slightly different code. | |
5005 | */ | |
3e418f98 LH |
5006 | vma = vma_lookup(mm, addr); |
5007 | if (!vma) | |
28b2ee20 RR |
5008 | break; |
5009 | if (vma->vm_ops && vma->vm_ops->access) | |
5010 | ret = vma->vm_ops->access(vma, addr, buf, | |
5011 | len, write); | |
5012 | if (ret <= 0) | |
28b2ee20 RR |
5013 | break; |
5014 | bytes = ret; | |
dbffcd03 | 5015 | #endif |
0ec76a11 | 5016 | } else { |
28b2ee20 RR |
5017 | bytes = len; |
5018 | offset = addr & (PAGE_SIZE-1); | |
5019 | if (bytes > PAGE_SIZE-offset) | |
5020 | bytes = PAGE_SIZE-offset; | |
5021 | ||
5022 | maddr = kmap(page); | |
5023 | if (write) { | |
5024 | copy_to_user_page(vma, page, addr, | |
5025 | maddr + offset, buf, bytes); | |
5026 | set_page_dirty_lock(page); | |
5027 | } else { | |
5028 | copy_from_user_page(vma, page, addr, | |
5029 | buf, maddr + offset, bytes); | |
5030 | } | |
5031 | kunmap(page); | |
09cbfeaf | 5032 | put_page(page); |
0ec76a11 | 5033 | } |
0ec76a11 DH |
5034 | len -= bytes; |
5035 | buf += bytes; | |
5036 | addr += bytes; | |
5037 | } | |
d8ed45c5 | 5038 | mmap_read_unlock(mm); |
0ec76a11 DH |
5039 | |
5040 | return buf - old_buf; | |
5041 | } | |
03252919 | 5042 | |
5ddd36b9 | 5043 | /** |
ae91dbfc | 5044 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
5045 | * @mm: the mm_struct of the target address space |
5046 | * @addr: start address to access | |
5047 | * @buf: source or destination buffer | |
5048 | * @len: number of bytes to transfer | |
6347e8d5 | 5049 | * @gup_flags: flags modifying lookup behaviour |
5ddd36b9 SW |
5050 | * |
5051 | * The caller must hold a reference on @mm. | |
a862f68a MR |
5052 | * |
5053 | * Return: number of bytes copied from source to destination. | |
5ddd36b9 SW |
5054 | */ |
5055 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 5056 | void *buf, int len, unsigned int gup_flags) |
5ddd36b9 | 5057 | { |
d3f5ffca | 5058 | return __access_remote_vm(mm, addr, buf, len, gup_flags); |
5ddd36b9 SW |
5059 | } |
5060 | ||
206cb636 SW |
5061 | /* |
5062 | * Access another process' address space. | |
5063 | * Source/target buffer must be kernel space, | |
5064 | * Do not walk the page table directly, use get_user_pages | |
5065 | */ | |
5066 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
f307ab6d | 5067 | void *buf, int len, unsigned int gup_flags) |
206cb636 SW |
5068 | { |
5069 | struct mm_struct *mm; | |
5070 | int ret; | |
5071 | ||
5072 | mm = get_task_mm(tsk); | |
5073 | if (!mm) | |
5074 | return 0; | |
5075 | ||
d3f5ffca | 5076 | ret = __access_remote_vm(mm, addr, buf, len, gup_flags); |
442486ec | 5077 | |
206cb636 SW |
5078 | mmput(mm); |
5079 | ||
5080 | return ret; | |
5081 | } | |
fcd35857 | 5082 | EXPORT_SYMBOL_GPL(access_process_vm); |
206cb636 | 5083 | |
03252919 AK |
5084 | /* |
5085 | * Print the name of a VMA. | |
5086 | */ | |
5087 | void print_vma_addr(char *prefix, unsigned long ip) | |
5088 | { | |
5089 | struct mm_struct *mm = current->mm; | |
5090 | struct vm_area_struct *vma; | |
5091 | ||
e8bff74a | 5092 | /* |
0a7f682d | 5093 | * we might be running from an atomic context so we cannot sleep |
e8bff74a | 5094 | */ |
d8ed45c5 | 5095 | if (!mmap_read_trylock(mm)) |
e8bff74a IM |
5096 | return; |
5097 | ||
03252919 AK |
5098 | vma = find_vma(mm, ip); |
5099 | if (vma && vma->vm_file) { | |
5100 | struct file *f = vma->vm_file; | |
0a7f682d | 5101 | char *buf = (char *)__get_free_page(GFP_NOWAIT); |
03252919 | 5102 | if (buf) { |
2fbc57c5 | 5103 | char *p; |
03252919 | 5104 | |
9bf39ab2 | 5105 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
5106 | if (IS_ERR(p)) |
5107 | p = "?"; | |
2fbc57c5 | 5108 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
5109 | vma->vm_start, |
5110 | vma->vm_end - vma->vm_start); | |
5111 | free_page((unsigned long)buf); | |
5112 | } | |
5113 | } | |
d8ed45c5 | 5114 | mmap_read_unlock(mm); |
03252919 | 5115 | } |
3ee1afa3 | 5116 | |
662bbcb2 | 5117 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 5118 | void __might_fault(const char *file, int line) |
3ee1afa3 | 5119 | { |
95156f00 PZ |
5120 | /* |
5121 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
c1e8d7c6 | 5122 | * holding the mmap_lock, this is safe because kernel memory doesn't |
95156f00 PZ |
5123 | * get paged out, therefore we'll never actually fault, and the |
5124 | * below annotations will generate false positives. | |
5125 | */ | |
db68ce10 | 5126 | if (uaccess_kernel()) |
95156f00 | 5127 | return; |
9ec23531 | 5128 | if (pagefault_disabled()) |
662bbcb2 | 5129 | return; |
9ec23531 DH |
5130 | __might_sleep(file, line, 0); |
5131 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) | |
662bbcb2 | 5132 | if (current->mm) |
da1c55f1 | 5133 | might_lock_read(¤t->mm->mmap_lock); |
9ec23531 | 5134 | #endif |
3ee1afa3 | 5135 | } |
9ec23531 | 5136 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 5137 | #endif |
47ad8475 AA |
5138 | |
5139 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
c6ddfb6c HY |
5140 | /* |
5141 | * Process all subpages of the specified huge page with the specified | |
5142 | * operation. The target subpage will be processed last to keep its | |
5143 | * cache lines hot. | |
5144 | */ | |
5145 | static inline void process_huge_page( | |
5146 | unsigned long addr_hint, unsigned int pages_per_huge_page, | |
5147 | void (*process_subpage)(unsigned long addr, int idx, void *arg), | |
5148 | void *arg) | |
47ad8475 | 5149 | { |
c79b57e4 HY |
5150 | int i, n, base, l; |
5151 | unsigned long addr = addr_hint & | |
5152 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
47ad8475 | 5153 | |
c6ddfb6c | 5154 | /* Process target subpage last to keep its cache lines hot */ |
47ad8475 | 5155 | might_sleep(); |
c79b57e4 HY |
5156 | n = (addr_hint - addr) / PAGE_SIZE; |
5157 | if (2 * n <= pages_per_huge_page) { | |
c6ddfb6c | 5158 | /* If target subpage in first half of huge page */ |
c79b57e4 HY |
5159 | base = 0; |
5160 | l = n; | |
c6ddfb6c | 5161 | /* Process subpages at the end of huge page */ |
c79b57e4 HY |
5162 | for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { |
5163 | cond_resched(); | |
c6ddfb6c | 5164 | process_subpage(addr + i * PAGE_SIZE, i, arg); |
c79b57e4 HY |
5165 | } |
5166 | } else { | |
c6ddfb6c | 5167 | /* If target subpage in second half of huge page */ |
c79b57e4 HY |
5168 | base = pages_per_huge_page - 2 * (pages_per_huge_page - n); |
5169 | l = pages_per_huge_page - n; | |
c6ddfb6c | 5170 | /* Process subpages at the begin of huge page */ |
c79b57e4 HY |
5171 | for (i = 0; i < base; i++) { |
5172 | cond_resched(); | |
c6ddfb6c | 5173 | process_subpage(addr + i * PAGE_SIZE, i, arg); |
c79b57e4 HY |
5174 | } |
5175 | } | |
5176 | /* | |
c6ddfb6c HY |
5177 | * Process remaining subpages in left-right-left-right pattern |
5178 | * towards the target subpage | |
c79b57e4 HY |
5179 | */ |
5180 | for (i = 0; i < l; i++) { | |
5181 | int left_idx = base + i; | |
5182 | int right_idx = base + 2 * l - 1 - i; | |
5183 | ||
5184 | cond_resched(); | |
c6ddfb6c | 5185 | process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg); |
47ad8475 | 5186 | cond_resched(); |
c6ddfb6c | 5187 | process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg); |
47ad8475 AA |
5188 | } |
5189 | } | |
5190 | ||
c6ddfb6c HY |
5191 | static void clear_gigantic_page(struct page *page, |
5192 | unsigned long addr, | |
5193 | unsigned int pages_per_huge_page) | |
5194 | { | |
5195 | int i; | |
5196 | struct page *p = page; | |
5197 | ||
5198 | might_sleep(); | |
5199 | for (i = 0; i < pages_per_huge_page; | |
5200 | i++, p = mem_map_next(p, page, i)) { | |
5201 | cond_resched(); | |
5202 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
5203 | } | |
5204 | } | |
5205 | ||
5206 | static void clear_subpage(unsigned long addr, int idx, void *arg) | |
5207 | { | |
5208 | struct page *page = arg; | |
5209 | ||
5210 | clear_user_highpage(page + idx, addr); | |
5211 | } | |
5212 | ||
5213 | void clear_huge_page(struct page *page, | |
5214 | unsigned long addr_hint, unsigned int pages_per_huge_page) | |
5215 | { | |
5216 | unsigned long addr = addr_hint & | |
5217 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
5218 | ||
5219 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
5220 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
5221 | return; | |
5222 | } | |
5223 | ||
5224 | process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page); | |
5225 | } | |
5226 | ||
47ad8475 AA |
5227 | static void copy_user_gigantic_page(struct page *dst, struct page *src, |
5228 | unsigned long addr, | |
5229 | struct vm_area_struct *vma, | |
5230 | unsigned int pages_per_huge_page) | |
5231 | { | |
5232 | int i; | |
5233 | struct page *dst_base = dst; | |
5234 | struct page *src_base = src; | |
5235 | ||
5236 | for (i = 0; i < pages_per_huge_page; ) { | |
5237 | cond_resched(); | |
5238 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
5239 | ||
5240 | i++; | |
5241 | dst = mem_map_next(dst, dst_base, i); | |
5242 | src = mem_map_next(src, src_base, i); | |
5243 | } | |
5244 | } | |
5245 | ||
c9f4cd71 HY |
5246 | struct copy_subpage_arg { |
5247 | struct page *dst; | |
5248 | struct page *src; | |
5249 | struct vm_area_struct *vma; | |
5250 | }; | |
5251 | ||
5252 | static void copy_subpage(unsigned long addr, int idx, void *arg) | |
5253 | { | |
5254 | struct copy_subpage_arg *copy_arg = arg; | |
5255 | ||
5256 | copy_user_highpage(copy_arg->dst + idx, copy_arg->src + idx, | |
5257 | addr, copy_arg->vma); | |
5258 | } | |
5259 | ||
47ad8475 | 5260 | void copy_user_huge_page(struct page *dst, struct page *src, |
c9f4cd71 | 5261 | unsigned long addr_hint, struct vm_area_struct *vma, |
47ad8475 AA |
5262 | unsigned int pages_per_huge_page) |
5263 | { | |
c9f4cd71 HY |
5264 | unsigned long addr = addr_hint & |
5265 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
5266 | struct copy_subpage_arg arg = { | |
5267 | .dst = dst, | |
5268 | .src = src, | |
5269 | .vma = vma, | |
5270 | }; | |
47ad8475 AA |
5271 | |
5272 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
5273 | copy_user_gigantic_page(dst, src, addr, vma, | |
5274 | pages_per_huge_page); | |
5275 | return; | |
5276 | } | |
5277 | ||
c9f4cd71 | 5278 | process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg); |
47ad8475 | 5279 | } |
fa4d75c1 MK |
5280 | |
5281 | long copy_huge_page_from_user(struct page *dst_page, | |
5282 | const void __user *usr_src, | |
810a56b9 MK |
5283 | unsigned int pages_per_huge_page, |
5284 | bool allow_pagefault) | |
fa4d75c1 MK |
5285 | { |
5286 | void *src = (void *)usr_src; | |
5287 | void *page_kaddr; | |
5288 | unsigned long i, rc = 0; | |
5289 | unsigned long ret_val = pages_per_huge_page * PAGE_SIZE; | |
3272cfc2 | 5290 | struct page *subpage = dst_page; |
fa4d75c1 | 5291 | |
3272cfc2 MK |
5292 | for (i = 0; i < pages_per_huge_page; |
5293 | i++, subpage = mem_map_next(subpage, dst_page, i)) { | |
810a56b9 | 5294 | if (allow_pagefault) |
3272cfc2 | 5295 | page_kaddr = kmap(subpage); |
810a56b9 | 5296 | else |
3272cfc2 | 5297 | page_kaddr = kmap_atomic(subpage); |
fa4d75c1 MK |
5298 | rc = copy_from_user(page_kaddr, |
5299 | (const void __user *)(src + i * PAGE_SIZE), | |
5300 | PAGE_SIZE); | |
810a56b9 | 5301 | if (allow_pagefault) |
3272cfc2 | 5302 | kunmap(subpage); |
810a56b9 MK |
5303 | else |
5304 | kunmap_atomic(page_kaddr); | |
fa4d75c1 MK |
5305 | |
5306 | ret_val -= (PAGE_SIZE - rc); | |
5307 | if (rc) | |
5308 | break; | |
5309 | ||
5310 | cond_resched(); | |
5311 | } | |
5312 | return ret_val; | |
5313 | } | |
47ad8475 | 5314 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
49076ec2 | 5315 | |
40b64acd | 5316 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
5317 | |
5318 | static struct kmem_cache *page_ptl_cachep; | |
5319 | ||
5320 | void __init ptlock_cache_init(void) | |
5321 | { | |
5322 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
5323 | SLAB_PANIC, NULL); | |
5324 | } | |
5325 | ||
539edb58 | 5326 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
5327 | { |
5328 | spinlock_t *ptl; | |
5329 | ||
b35f1819 | 5330 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
5331 | if (!ptl) |
5332 | return false; | |
539edb58 | 5333 | page->ptl = ptl; |
49076ec2 KS |
5334 | return true; |
5335 | } | |
5336 | ||
539edb58 | 5337 | void ptlock_free(struct page *page) |
49076ec2 | 5338 | { |
b35f1819 | 5339 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
5340 | } |
5341 | #endif |