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 | ||
03c4f204 | 436 | void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte) |
1da177e4 | 437 | { |
03c4f204 | 438 | spinlock_t *ptl = pmd_lock(mm, pmd); |
1bb3630e | 439 | |
8ac1f832 | 440 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
c4812909 | 441 | mm_inc_nr_ptes(mm); |
ed33b5a6 QZ |
442 | /* |
443 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
444 | * visible before the pte is made visible to other CPUs by being | |
445 | * put into page tables. | |
446 | * | |
447 | * The other side of the story is the pointer chasing in the page | |
448 | * table walking code (when walking the page table without locking; | |
449 | * ie. most of the time). Fortunately, these data accesses consist | |
450 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
451 | * being the notable exception) will already guarantee loads are | |
452 | * seen in-order. See the alpha page table accessors for the | |
453 | * smp_rmb() barriers in page table walking code. | |
454 | */ | |
455 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
03c4f204 QZ |
456 | pmd_populate(mm, pmd, *pte); |
457 | *pte = NULL; | |
4b471e88 | 458 | } |
c4088ebd | 459 | spin_unlock(ptl); |
03c4f204 QZ |
460 | } |
461 | ||
4cf58924 | 462 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd) |
1da177e4 | 463 | { |
4cf58924 | 464 | pgtable_t new = pte_alloc_one(mm); |
1bb3630e HD |
465 | if (!new) |
466 | return -ENOMEM; | |
467 | ||
03c4f204 | 468 | pmd_install(mm, pmd, &new); |
2f569afd MS |
469 | if (new) |
470 | pte_free(mm, new); | |
1bb3630e | 471 | return 0; |
1da177e4 LT |
472 | } |
473 | ||
4cf58924 | 474 | int __pte_alloc_kernel(pmd_t *pmd) |
1da177e4 | 475 | { |
4cf58924 | 476 | pte_t *new = pte_alloc_one_kernel(&init_mm); |
1bb3630e HD |
477 | if (!new) |
478 | return -ENOMEM; | |
479 | ||
480 | spin_lock(&init_mm.page_table_lock); | |
8ac1f832 | 481 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
ed33b5a6 | 482 | smp_wmb(); /* See comment in pmd_install() */ |
1bb3630e | 483 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 484 | new = NULL; |
4b471e88 | 485 | } |
1bb3630e | 486 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
487 | if (new) |
488 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 489 | return 0; |
1da177e4 LT |
490 | } |
491 | ||
d559db08 KH |
492 | static inline void init_rss_vec(int *rss) |
493 | { | |
494 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
495 | } | |
496 | ||
497 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 498 | { |
d559db08 KH |
499 | int i; |
500 | ||
34e55232 | 501 | if (current->mm == mm) |
05af2e10 | 502 | sync_mm_rss(mm); |
d559db08 KH |
503 | for (i = 0; i < NR_MM_COUNTERS; i++) |
504 | if (rss[i]) | |
505 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
506 | } |
507 | ||
b5810039 | 508 | /* |
6aab341e LT |
509 | * This function is called to print an error when a bad pte |
510 | * is found. For example, we might have a PFN-mapped pte in | |
511 | * a region that doesn't allow it. | |
b5810039 NP |
512 | * |
513 | * The calling function must still handle the error. | |
514 | */ | |
3dc14741 HD |
515 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
516 | pte_t pte, struct page *page) | |
b5810039 | 517 | { |
3dc14741 | 518 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
c2febafc KS |
519 | p4d_t *p4d = p4d_offset(pgd, addr); |
520 | pud_t *pud = pud_offset(p4d, addr); | |
3dc14741 HD |
521 | pmd_t *pmd = pmd_offset(pud, addr); |
522 | struct address_space *mapping; | |
523 | pgoff_t index; | |
d936cf9b HD |
524 | static unsigned long resume; |
525 | static unsigned long nr_shown; | |
526 | static unsigned long nr_unshown; | |
527 | ||
528 | /* | |
529 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
530 | * or allow a steady drip of one report per second. | |
531 | */ | |
532 | if (nr_shown == 60) { | |
533 | if (time_before(jiffies, resume)) { | |
534 | nr_unshown++; | |
535 | return; | |
536 | } | |
537 | if (nr_unshown) { | |
1170532b JP |
538 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
539 | nr_unshown); | |
d936cf9b HD |
540 | nr_unshown = 0; |
541 | } | |
542 | nr_shown = 0; | |
543 | } | |
544 | if (nr_shown++ == 0) | |
545 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
546 | |
547 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
548 | index = linear_page_index(vma, addr); | |
549 | ||
1170532b JP |
550 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
551 | current->comm, | |
552 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 553 | if (page) |
f0b791a3 | 554 | dump_page(page, "bad pte"); |
6aa9b8b2 | 555 | pr_alert("addr:%px vm_flags:%08lx anon_vma:%px mapping:%px index:%lx\n", |
1170532b | 556 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); |
d75f773c | 557 | pr_alert("file:%pD fault:%ps mmap:%ps readpage:%ps\n", |
2682582a KK |
558 | vma->vm_file, |
559 | vma->vm_ops ? vma->vm_ops->fault : NULL, | |
560 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, | |
561 | mapping ? mapping->a_ops->readpage : NULL); | |
b5810039 | 562 | dump_stack(); |
373d4d09 | 563 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
564 | } |
565 | ||
ee498ed7 | 566 | /* |
7e675137 | 567 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 568 | * |
7e675137 NP |
569 | * "Special" mappings do not wish to be associated with a "struct page" (either |
570 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
571 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 572 | * |
7e675137 NP |
573 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
574 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
575 | * may not have a spare pte bit, which requires a more complicated scheme, | |
576 | * described below. | |
577 | * | |
578 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
579 | * special mapping (even if there are underlying and valid "struct pages"). | |
580 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 581 | * |
b379d790 JH |
582 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
583 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
584 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
585 | * mapping will always honor the rule | |
6aab341e LT |
586 | * |
587 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
588 | * | |
7e675137 NP |
589 | * And for normal mappings this is false. |
590 | * | |
591 | * This restricts such mappings to be a linear translation from virtual address | |
592 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
593 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
594 | * special (because none can have been COWed). | |
b379d790 | 595 | * |
b379d790 | 596 | * |
7e675137 | 597 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
598 | * |
599 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
600 | * page" backing, however the difference is that _all_ pages with a struct | |
601 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
602 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
603 | * (which can be slower and simply not an option for some PFNMAP users). The | |
604 | * advantage is that we don't have to follow the strict linearity rule of | |
605 | * PFNMAP mappings in order to support COWable mappings. | |
606 | * | |
ee498ed7 | 607 | */ |
25b2995a CH |
608 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
609 | pte_t pte) | |
ee498ed7 | 610 | { |
22b31eec | 611 | unsigned long pfn = pte_pfn(pte); |
7e675137 | 612 | |
00b3a331 | 613 | if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) { |
b38af472 | 614 | if (likely(!pte_special(pte))) |
22b31eec | 615 | goto check_pfn; |
667a0a06 DV |
616 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
617 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
618 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
619 | return NULL; | |
df6ad698 JG |
620 | if (is_zero_pfn(pfn)) |
621 | return NULL; | |
e1fb4a08 DJ |
622 | if (pte_devmap(pte)) |
623 | return NULL; | |
624 | ||
df6ad698 | 625 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
626 | return NULL; |
627 | } | |
628 | ||
00b3a331 | 629 | /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */ |
7e675137 | 630 | |
b379d790 JH |
631 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
632 | if (vma->vm_flags & VM_MIXEDMAP) { | |
633 | if (!pfn_valid(pfn)) | |
634 | return NULL; | |
635 | goto out; | |
636 | } else { | |
7e675137 NP |
637 | unsigned long off; |
638 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
639 | if (pfn == vma->vm_pgoff + off) |
640 | return NULL; | |
641 | if (!is_cow_mapping(vma->vm_flags)) | |
642 | return NULL; | |
643 | } | |
6aab341e LT |
644 | } |
645 | ||
b38af472 HD |
646 | if (is_zero_pfn(pfn)) |
647 | return NULL; | |
00b3a331 | 648 | |
22b31eec HD |
649 | check_pfn: |
650 | if (unlikely(pfn > highest_memmap_pfn)) { | |
651 | print_bad_pte(vma, addr, pte, NULL); | |
652 | return NULL; | |
653 | } | |
6aab341e LT |
654 | |
655 | /* | |
7e675137 | 656 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 657 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 658 | */ |
b379d790 | 659 | out: |
6aab341e | 660 | return pfn_to_page(pfn); |
ee498ed7 HD |
661 | } |
662 | ||
28093f9f GS |
663 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
664 | struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, | |
665 | pmd_t pmd) | |
666 | { | |
667 | unsigned long pfn = pmd_pfn(pmd); | |
668 | ||
669 | /* | |
670 | * There is no pmd_special() but there may be special pmds, e.g. | |
671 | * in a direct-access (dax) mapping, so let's just replicate the | |
00b3a331 | 672 | * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here. |
28093f9f GS |
673 | */ |
674 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { | |
675 | if (vma->vm_flags & VM_MIXEDMAP) { | |
676 | if (!pfn_valid(pfn)) | |
677 | return NULL; | |
678 | goto out; | |
679 | } else { | |
680 | unsigned long off; | |
681 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
682 | if (pfn == vma->vm_pgoff + off) | |
683 | return NULL; | |
684 | if (!is_cow_mapping(vma->vm_flags)) | |
685 | return NULL; | |
686 | } | |
687 | } | |
688 | ||
e1fb4a08 DJ |
689 | if (pmd_devmap(pmd)) |
690 | return NULL; | |
3cde287b | 691 | if (is_huge_zero_pmd(pmd)) |
28093f9f GS |
692 | return NULL; |
693 | if (unlikely(pfn > highest_memmap_pfn)) | |
694 | return NULL; | |
695 | ||
696 | /* | |
697 | * NOTE! We still have PageReserved() pages in the page tables. | |
698 | * eg. VDSO mappings can cause them to exist. | |
699 | */ | |
700 | out: | |
701 | return pfn_to_page(pfn); | |
702 | } | |
703 | #endif | |
704 | ||
b756a3b5 AP |
705 | static void restore_exclusive_pte(struct vm_area_struct *vma, |
706 | struct page *page, unsigned long address, | |
707 | pte_t *ptep) | |
708 | { | |
709 | pte_t pte; | |
710 | swp_entry_t entry; | |
711 | ||
712 | pte = pte_mkold(mk_pte(page, READ_ONCE(vma->vm_page_prot))); | |
713 | if (pte_swp_soft_dirty(*ptep)) | |
714 | pte = pte_mksoft_dirty(pte); | |
715 | ||
716 | entry = pte_to_swp_entry(*ptep); | |
717 | if (pte_swp_uffd_wp(*ptep)) | |
718 | pte = pte_mkuffd_wp(pte); | |
719 | else if (is_writable_device_exclusive_entry(entry)) | |
720 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); | |
721 | ||
722 | set_pte_at(vma->vm_mm, address, ptep, pte); | |
723 | ||
724 | /* | |
725 | * No need to take a page reference as one was already | |
726 | * created when the swap entry was made. | |
727 | */ | |
728 | if (PageAnon(page)) | |
729 | page_add_anon_rmap(page, vma, address, false); | |
730 | else | |
731 | /* | |
732 | * Currently device exclusive access only supports anonymous | |
733 | * memory so the entry shouldn't point to a filebacked page. | |
734 | */ | |
735 | WARN_ON_ONCE(!PageAnon(page)); | |
736 | ||
737 | if (vma->vm_flags & VM_LOCKED) | |
738 | mlock_vma_page(page); | |
739 | ||
740 | /* | |
741 | * No need to invalidate - it was non-present before. However | |
742 | * secondary CPUs may have mappings that need invalidating. | |
743 | */ | |
744 | update_mmu_cache(vma, address, ptep); | |
745 | } | |
746 | ||
747 | /* | |
748 | * Tries to restore an exclusive pte if the page lock can be acquired without | |
749 | * sleeping. | |
750 | */ | |
751 | static int | |
752 | try_restore_exclusive_pte(pte_t *src_pte, struct vm_area_struct *vma, | |
753 | unsigned long addr) | |
754 | { | |
755 | swp_entry_t entry = pte_to_swp_entry(*src_pte); | |
756 | struct page *page = pfn_swap_entry_to_page(entry); | |
757 | ||
758 | if (trylock_page(page)) { | |
759 | restore_exclusive_pte(vma, page, addr, src_pte); | |
760 | unlock_page(page); | |
761 | return 0; | |
762 | } | |
763 | ||
764 | return -EBUSY; | |
765 | } | |
766 | ||
1da177e4 LT |
767 | /* |
768 | * copy one vm_area from one task to the other. Assumes the page tables | |
769 | * already present in the new task to be cleared in the whole range | |
770 | * covered by this vma. | |
1da177e4 LT |
771 | */ |
772 | ||
df3a57d1 LT |
773 | static unsigned long |
774 | copy_nonpresent_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
8f34f1ea PX |
775 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *dst_vma, |
776 | struct vm_area_struct *src_vma, unsigned long addr, int *rss) | |
1da177e4 | 777 | { |
8f34f1ea | 778 | unsigned long vm_flags = dst_vma->vm_flags; |
1da177e4 LT |
779 | pte_t pte = *src_pte; |
780 | struct page *page; | |
df3a57d1 LT |
781 | swp_entry_t entry = pte_to_swp_entry(pte); |
782 | ||
783 | if (likely(!non_swap_entry(entry))) { | |
784 | if (swap_duplicate(entry) < 0) | |
9a5cc85c | 785 | return -EIO; |
df3a57d1 LT |
786 | |
787 | /* make sure dst_mm is on swapoff's mmlist. */ | |
788 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
789 | spin_lock(&mmlist_lock); | |
790 | if (list_empty(&dst_mm->mmlist)) | |
791 | list_add(&dst_mm->mmlist, | |
792 | &src_mm->mmlist); | |
793 | spin_unlock(&mmlist_lock); | |
794 | } | |
795 | rss[MM_SWAPENTS]++; | |
796 | } else if (is_migration_entry(entry)) { | |
af5cdaf8 | 797 | page = pfn_swap_entry_to_page(entry); |
1da177e4 | 798 | |
df3a57d1 | 799 | rss[mm_counter(page)]++; |
5042db43 | 800 | |
4dd845b5 | 801 | if (is_writable_migration_entry(entry) && |
df3a57d1 | 802 | is_cow_mapping(vm_flags)) { |
5042db43 | 803 | /* |
df3a57d1 LT |
804 | * COW mappings require pages in both |
805 | * parent and child to be set to read. | |
5042db43 | 806 | */ |
4dd845b5 AP |
807 | entry = make_readable_migration_entry( |
808 | swp_offset(entry)); | |
df3a57d1 LT |
809 | pte = swp_entry_to_pte(entry); |
810 | if (pte_swp_soft_dirty(*src_pte)) | |
811 | pte = pte_swp_mksoft_dirty(pte); | |
812 | if (pte_swp_uffd_wp(*src_pte)) | |
813 | pte = pte_swp_mkuffd_wp(pte); | |
814 | set_pte_at(src_mm, addr, src_pte, pte); | |
815 | } | |
816 | } else if (is_device_private_entry(entry)) { | |
af5cdaf8 | 817 | page = pfn_swap_entry_to_page(entry); |
5042db43 | 818 | |
df3a57d1 LT |
819 | /* |
820 | * Update rss count even for unaddressable pages, as | |
821 | * they should treated just like normal pages in this | |
822 | * respect. | |
823 | * | |
824 | * We will likely want to have some new rss counters | |
825 | * for unaddressable pages, at some point. But for now | |
826 | * keep things as they are. | |
827 | */ | |
828 | get_page(page); | |
829 | rss[mm_counter(page)]++; | |
830 | page_dup_rmap(page, false); | |
831 | ||
832 | /* | |
833 | * We do not preserve soft-dirty information, because so | |
834 | * far, checkpoint/restore is the only feature that | |
835 | * requires that. And checkpoint/restore does not work | |
836 | * when a device driver is involved (you cannot easily | |
837 | * save and restore device driver state). | |
838 | */ | |
4dd845b5 | 839 | if (is_writable_device_private_entry(entry) && |
df3a57d1 | 840 | is_cow_mapping(vm_flags)) { |
4dd845b5 AP |
841 | entry = make_readable_device_private_entry( |
842 | swp_offset(entry)); | |
df3a57d1 LT |
843 | pte = swp_entry_to_pte(entry); |
844 | if (pte_swp_uffd_wp(*src_pte)) | |
845 | pte = pte_swp_mkuffd_wp(pte); | |
846 | set_pte_at(src_mm, addr, src_pte, pte); | |
1da177e4 | 847 | } |
b756a3b5 AP |
848 | } else if (is_device_exclusive_entry(entry)) { |
849 | /* | |
850 | * Make device exclusive entries present by restoring the | |
851 | * original entry then copying as for a present pte. Device | |
852 | * exclusive entries currently only support private writable | |
853 | * (ie. COW) mappings. | |
854 | */ | |
855 | VM_BUG_ON(!is_cow_mapping(src_vma->vm_flags)); | |
856 | if (try_restore_exclusive_pte(src_pte, src_vma, addr)) | |
857 | return -EBUSY; | |
858 | return -ENOENT; | |
1da177e4 | 859 | } |
8f34f1ea PX |
860 | if (!userfaultfd_wp(dst_vma)) |
861 | pte = pte_swp_clear_uffd_wp(pte); | |
df3a57d1 LT |
862 | set_pte_at(dst_mm, addr, dst_pte, pte); |
863 | return 0; | |
864 | } | |
865 | ||
70e806e4 PX |
866 | /* |
867 | * Copy a present and normal page if necessary. | |
868 | * | |
869 | * NOTE! The usual case is that this doesn't need to do | |
870 | * anything, and can just return a positive value. That | |
871 | * will let the caller know that it can just increase | |
872 | * the page refcount and re-use the pte the traditional | |
873 | * way. | |
874 | * | |
875 | * But _if_ we need to copy it because it needs to be | |
876 | * pinned in the parent (and the child should get its own | |
877 | * copy rather than just a reference to the same page), | |
878 | * we'll do that here and return zero to let the caller | |
879 | * know we're done. | |
880 | * | |
881 | * And if we need a pre-allocated page but don't yet have | |
882 | * one, return a negative error to let the preallocation | |
883 | * code know so that it can do so outside the page table | |
884 | * lock. | |
885 | */ | |
886 | static inline int | |
c78f4636 PX |
887 | copy_present_page(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, |
888 | pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, | |
889 | struct page **prealloc, pte_t pte, struct page *page) | |
70e806e4 PX |
890 | { |
891 | struct page *new_page; | |
892 | ||
70e806e4 | 893 | /* |
70e806e4 PX |
894 | * What we want to do is to check whether this page may |
895 | * have been pinned by the parent process. If so, | |
896 | * instead of wrprotect the pte on both sides, we copy | |
897 | * the page immediately so that we'll always guarantee | |
898 | * the pinned page won't be randomly replaced in the | |
899 | * future. | |
900 | * | |
f3c64eda LT |
901 | * The page pinning checks are just "has this mm ever |
902 | * seen pinning", along with the (inexact) check of | |
903 | * the page count. That might give false positives for | |
904 | * for pinning, but it will work correctly. | |
70e806e4 | 905 | */ |
97a7e473 | 906 | if (likely(!page_needs_cow_for_dma(src_vma, page))) |
70e806e4 PX |
907 | return 1; |
908 | ||
70e806e4 PX |
909 | new_page = *prealloc; |
910 | if (!new_page) | |
911 | return -EAGAIN; | |
912 | ||
913 | /* | |
914 | * We have a prealloc page, all good! Take it | |
915 | * over and copy the page & arm it. | |
916 | */ | |
917 | *prealloc = NULL; | |
c78f4636 | 918 | copy_user_highpage(new_page, page, addr, src_vma); |
70e806e4 | 919 | __SetPageUptodate(new_page); |
c78f4636 PX |
920 | page_add_new_anon_rmap(new_page, dst_vma, addr, false); |
921 | lru_cache_add_inactive_or_unevictable(new_page, dst_vma); | |
70e806e4 PX |
922 | rss[mm_counter(new_page)]++; |
923 | ||
924 | /* All done, just insert the new page copy in the child */ | |
c78f4636 PX |
925 | pte = mk_pte(new_page, dst_vma->vm_page_prot); |
926 | pte = maybe_mkwrite(pte_mkdirty(pte), dst_vma); | |
8f34f1ea PX |
927 | if (userfaultfd_pte_wp(dst_vma, *src_pte)) |
928 | /* Uffd-wp needs to be delivered to dest pte as well */ | |
929 | pte = pte_wrprotect(pte_mkuffd_wp(pte)); | |
c78f4636 | 930 | set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); |
70e806e4 PX |
931 | return 0; |
932 | } | |
933 | ||
934 | /* | |
935 | * Copy one pte. Returns 0 if succeeded, or -EAGAIN if one preallocated page | |
936 | * is required to copy this pte. | |
937 | */ | |
938 | static inline int | |
c78f4636 PX |
939 | copy_present_pte(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, |
940 | pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, | |
941 | struct page **prealloc) | |
df3a57d1 | 942 | { |
c78f4636 PX |
943 | struct mm_struct *src_mm = src_vma->vm_mm; |
944 | unsigned long vm_flags = src_vma->vm_flags; | |
df3a57d1 LT |
945 | pte_t pte = *src_pte; |
946 | struct page *page; | |
947 | ||
c78f4636 | 948 | page = vm_normal_page(src_vma, addr, pte); |
70e806e4 PX |
949 | if (page) { |
950 | int retval; | |
951 | ||
c78f4636 PX |
952 | retval = copy_present_page(dst_vma, src_vma, dst_pte, src_pte, |
953 | addr, rss, prealloc, pte, page); | |
70e806e4 PX |
954 | if (retval <= 0) |
955 | return retval; | |
956 | ||
957 | get_page(page); | |
958 | page_dup_rmap(page, false); | |
959 | rss[mm_counter(page)]++; | |
960 | } | |
961 | ||
1da177e4 LT |
962 | /* |
963 | * If it's a COW mapping, write protect it both | |
964 | * in the parent and the child | |
965 | */ | |
1b2de5d0 | 966 | if (is_cow_mapping(vm_flags) && pte_write(pte)) { |
1da177e4 | 967 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 968 | pte = pte_wrprotect(pte); |
1da177e4 LT |
969 | } |
970 | ||
971 | /* | |
972 | * If it's a shared mapping, mark it clean in | |
973 | * the child | |
974 | */ | |
975 | if (vm_flags & VM_SHARED) | |
976 | pte = pte_mkclean(pte); | |
977 | pte = pte_mkold(pte); | |
6aab341e | 978 | |
8f34f1ea | 979 | if (!userfaultfd_wp(dst_vma)) |
b569a176 PX |
980 | pte = pte_clear_uffd_wp(pte); |
981 | ||
c78f4636 | 982 | set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); |
70e806e4 PX |
983 | return 0; |
984 | } | |
985 | ||
986 | static inline struct page * | |
987 | page_copy_prealloc(struct mm_struct *src_mm, struct vm_area_struct *vma, | |
988 | unsigned long addr) | |
989 | { | |
990 | struct page *new_page; | |
991 | ||
992 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, addr); | |
993 | if (!new_page) | |
994 | return NULL; | |
995 | ||
8f425e4e | 996 | if (mem_cgroup_charge(page_folio(new_page), src_mm, GFP_KERNEL)) { |
70e806e4 PX |
997 | put_page(new_page); |
998 | return NULL; | |
6aab341e | 999 | } |
70e806e4 | 1000 | cgroup_throttle_swaprate(new_page, GFP_KERNEL); |
ae859762 | 1001 | |
70e806e4 | 1002 | return new_page; |
1da177e4 LT |
1003 | } |
1004 | ||
c78f4636 PX |
1005 | static int |
1006 | copy_pte_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1007 | pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr, | |
1008 | unsigned long end) | |
1da177e4 | 1009 | { |
c78f4636 PX |
1010 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
1011 | struct mm_struct *src_mm = src_vma->vm_mm; | |
c36987e2 | 1012 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 1013 | pte_t *src_pte, *dst_pte; |
c74df32c | 1014 | spinlock_t *src_ptl, *dst_ptl; |
70e806e4 | 1015 | int progress, ret = 0; |
d559db08 | 1016 | int rss[NR_MM_COUNTERS]; |
570a335b | 1017 | swp_entry_t entry = (swp_entry_t){0}; |
70e806e4 | 1018 | struct page *prealloc = NULL; |
1da177e4 LT |
1019 | |
1020 | again: | |
70e806e4 | 1021 | progress = 0; |
d559db08 KH |
1022 | init_rss_vec(rss); |
1023 | ||
c74df32c | 1024 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
70e806e4 PX |
1025 | if (!dst_pte) { |
1026 | ret = -ENOMEM; | |
1027 | goto out; | |
1028 | } | |
ece0e2b6 | 1029 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 1030 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 1031 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
1032 | orig_src_pte = src_pte; |
1033 | orig_dst_pte = dst_pte; | |
6606c3e0 | 1034 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 1035 | |
1da177e4 LT |
1036 | do { |
1037 | /* | |
1038 | * We are holding two locks at this point - either of them | |
1039 | * could generate latencies in another task on another CPU. | |
1040 | */ | |
e040f218 HD |
1041 | if (progress >= 32) { |
1042 | progress = 0; | |
1043 | if (need_resched() || | |
95c354fe | 1044 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
1045 | break; |
1046 | } | |
1da177e4 LT |
1047 | if (pte_none(*src_pte)) { |
1048 | progress++; | |
1049 | continue; | |
1050 | } | |
79a1971c | 1051 | if (unlikely(!pte_present(*src_pte))) { |
9a5cc85c AP |
1052 | ret = copy_nonpresent_pte(dst_mm, src_mm, |
1053 | dst_pte, src_pte, | |
1054 | dst_vma, src_vma, | |
1055 | addr, rss); | |
1056 | if (ret == -EIO) { | |
1057 | entry = pte_to_swp_entry(*src_pte); | |
79a1971c | 1058 | break; |
b756a3b5 AP |
1059 | } else if (ret == -EBUSY) { |
1060 | break; | |
1061 | } else if (!ret) { | |
1062 | progress += 8; | |
1063 | continue; | |
9a5cc85c | 1064 | } |
b756a3b5 AP |
1065 | |
1066 | /* | |
1067 | * Device exclusive entry restored, continue by copying | |
1068 | * the now present pte. | |
1069 | */ | |
1070 | WARN_ON_ONCE(ret != -ENOENT); | |
79a1971c | 1071 | } |
70e806e4 | 1072 | /* copy_present_pte() will clear `*prealloc' if consumed */ |
c78f4636 PX |
1073 | ret = copy_present_pte(dst_vma, src_vma, dst_pte, src_pte, |
1074 | addr, rss, &prealloc); | |
70e806e4 PX |
1075 | /* |
1076 | * If we need a pre-allocated page for this pte, drop the | |
1077 | * locks, allocate, and try again. | |
1078 | */ | |
1079 | if (unlikely(ret == -EAGAIN)) | |
1080 | break; | |
1081 | if (unlikely(prealloc)) { | |
1082 | /* | |
1083 | * pre-alloc page cannot be reused by next time so as | |
1084 | * to strictly follow mempolicy (e.g., alloc_page_vma() | |
1085 | * will allocate page according to address). This | |
1086 | * could only happen if one pinned pte changed. | |
1087 | */ | |
1088 | put_page(prealloc); | |
1089 | prealloc = NULL; | |
1090 | } | |
1da177e4 LT |
1091 | progress += 8; |
1092 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 1093 | |
6606c3e0 | 1094 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1095 | spin_unlock(src_ptl); |
ece0e2b6 | 1096 | pte_unmap(orig_src_pte); |
d559db08 | 1097 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 1098 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 1099 | cond_resched(); |
570a335b | 1100 | |
9a5cc85c AP |
1101 | if (ret == -EIO) { |
1102 | VM_WARN_ON_ONCE(!entry.val); | |
70e806e4 PX |
1103 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) { |
1104 | ret = -ENOMEM; | |
1105 | goto out; | |
1106 | } | |
1107 | entry.val = 0; | |
b756a3b5 AP |
1108 | } else if (ret == -EBUSY) { |
1109 | goto out; | |
9a5cc85c | 1110 | } else if (ret == -EAGAIN) { |
c78f4636 | 1111 | prealloc = page_copy_prealloc(src_mm, src_vma, addr); |
70e806e4 | 1112 | if (!prealloc) |
570a335b | 1113 | return -ENOMEM; |
9a5cc85c AP |
1114 | } else if (ret) { |
1115 | VM_WARN_ON_ONCE(1); | |
570a335b | 1116 | } |
9a5cc85c AP |
1117 | |
1118 | /* We've captured and resolved the error. Reset, try again. */ | |
1119 | ret = 0; | |
1120 | ||
1da177e4 LT |
1121 | if (addr != end) |
1122 | goto again; | |
70e806e4 PX |
1123 | out: |
1124 | if (unlikely(prealloc)) | |
1125 | put_page(prealloc); | |
1126 | return ret; | |
1da177e4 LT |
1127 | } |
1128 | ||
c78f4636 PX |
1129 | static inline int |
1130 | copy_pmd_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1131 | pud_t *dst_pud, pud_t *src_pud, unsigned long addr, | |
1132 | unsigned long end) | |
1da177e4 | 1133 | { |
c78f4636 PX |
1134 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
1135 | struct mm_struct *src_mm = src_vma->vm_mm; | |
1da177e4 LT |
1136 | pmd_t *src_pmd, *dst_pmd; |
1137 | unsigned long next; | |
1138 | ||
1139 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
1140 | if (!dst_pmd) | |
1141 | return -ENOMEM; | |
1142 | src_pmd = pmd_offset(src_pud, addr); | |
1143 | do { | |
1144 | next = pmd_addr_end(addr, end); | |
84c3fc4e ZY |
1145 | if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) |
1146 | || pmd_devmap(*src_pmd)) { | |
71e3aac0 | 1147 | int err; |
c78f4636 | 1148 | VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, src_vma); |
8f34f1ea PX |
1149 | err = copy_huge_pmd(dst_mm, src_mm, dst_pmd, src_pmd, |
1150 | addr, dst_vma, src_vma); | |
71e3aac0 AA |
1151 | if (err == -ENOMEM) |
1152 | return -ENOMEM; | |
1153 | if (!err) | |
1154 | continue; | |
1155 | /* fall through */ | |
1156 | } | |
1da177e4 LT |
1157 | if (pmd_none_or_clear_bad(src_pmd)) |
1158 | continue; | |
c78f4636 PX |
1159 | if (copy_pte_range(dst_vma, src_vma, dst_pmd, src_pmd, |
1160 | addr, next)) | |
1da177e4 LT |
1161 | return -ENOMEM; |
1162 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
1163 | return 0; | |
1164 | } | |
1165 | ||
c78f4636 PX |
1166 | static inline int |
1167 | copy_pud_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1168 | p4d_t *dst_p4d, p4d_t *src_p4d, unsigned long addr, | |
1169 | unsigned long end) | |
1da177e4 | 1170 | { |
c78f4636 PX |
1171 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
1172 | struct mm_struct *src_mm = src_vma->vm_mm; | |
1da177e4 LT |
1173 | pud_t *src_pud, *dst_pud; |
1174 | unsigned long next; | |
1175 | ||
c2febafc | 1176 | dst_pud = pud_alloc(dst_mm, dst_p4d, addr); |
1da177e4 LT |
1177 | if (!dst_pud) |
1178 | return -ENOMEM; | |
c2febafc | 1179 | src_pud = pud_offset(src_p4d, addr); |
1da177e4 LT |
1180 | do { |
1181 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1182 | if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { |
1183 | int err; | |
1184 | ||
c78f4636 | 1185 | VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, src_vma); |
a00cc7d9 | 1186 | err = copy_huge_pud(dst_mm, src_mm, |
c78f4636 | 1187 | dst_pud, src_pud, addr, src_vma); |
a00cc7d9 MW |
1188 | if (err == -ENOMEM) |
1189 | return -ENOMEM; | |
1190 | if (!err) | |
1191 | continue; | |
1192 | /* fall through */ | |
1193 | } | |
1da177e4 LT |
1194 | if (pud_none_or_clear_bad(src_pud)) |
1195 | continue; | |
c78f4636 PX |
1196 | if (copy_pmd_range(dst_vma, src_vma, dst_pud, src_pud, |
1197 | addr, next)) | |
1da177e4 LT |
1198 | return -ENOMEM; |
1199 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
1200 | return 0; | |
1201 | } | |
1202 | ||
c78f4636 PX |
1203 | static inline int |
1204 | copy_p4d_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1205 | pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long addr, | |
1206 | unsigned long end) | |
c2febafc | 1207 | { |
c78f4636 | 1208 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
c2febafc KS |
1209 | p4d_t *src_p4d, *dst_p4d; |
1210 | unsigned long next; | |
1211 | ||
1212 | dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); | |
1213 | if (!dst_p4d) | |
1214 | return -ENOMEM; | |
1215 | src_p4d = p4d_offset(src_pgd, addr); | |
1216 | do { | |
1217 | next = p4d_addr_end(addr, end); | |
1218 | if (p4d_none_or_clear_bad(src_p4d)) | |
1219 | continue; | |
c78f4636 PX |
1220 | if (copy_pud_range(dst_vma, src_vma, dst_p4d, src_p4d, |
1221 | addr, next)) | |
c2febafc KS |
1222 | return -ENOMEM; |
1223 | } while (dst_p4d++, src_p4d++, addr = next, addr != end); | |
1224 | return 0; | |
1225 | } | |
1226 | ||
c78f4636 PX |
1227 | int |
1228 | copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) | |
1da177e4 LT |
1229 | { |
1230 | pgd_t *src_pgd, *dst_pgd; | |
1231 | unsigned long next; | |
c78f4636 PX |
1232 | unsigned long addr = src_vma->vm_start; |
1233 | unsigned long end = src_vma->vm_end; | |
1234 | struct mm_struct *dst_mm = dst_vma->vm_mm; | |
1235 | struct mm_struct *src_mm = src_vma->vm_mm; | |
ac46d4f3 | 1236 | struct mmu_notifier_range range; |
2ec74c3e | 1237 | bool is_cow; |
cddb8a5c | 1238 | int ret; |
1da177e4 | 1239 | |
d992895b NP |
1240 | /* |
1241 | * Don't copy ptes where a page fault will fill them correctly. | |
1242 | * Fork becomes much lighter when there are big shared or private | |
1243 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1244 | * efficient than faulting. | |
1245 | */ | |
c78f4636 PX |
1246 | if (!(src_vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
1247 | !src_vma->anon_vma) | |
0661a336 | 1248 | return 0; |
d992895b | 1249 | |
c78f4636 PX |
1250 | if (is_vm_hugetlb_page(src_vma)) |
1251 | return copy_hugetlb_page_range(dst_mm, src_mm, src_vma); | |
1da177e4 | 1252 | |
c78f4636 | 1253 | if (unlikely(src_vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1254 | /* |
1255 | * We do not free on error cases below as remove_vma | |
1256 | * gets called on error from higher level routine | |
1257 | */ | |
c78f4636 | 1258 | ret = track_pfn_copy(src_vma); |
2ab64037 | 1259 | if (ret) |
1260 | return ret; | |
1261 | } | |
1262 | ||
cddb8a5c AA |
1263 | /* |
1264 | * We need to invalidate the secondary MMU mappings only when | |
1265 | * there could be a permission downgrade on the ptes of the | |
1266 | * parent mm. And a permission downgrade will only happen if | |
1267 | * is_cow_mapping() returns true. | |
1268 | */ | |
c78f4636 | 1269 | is_cow = is_cow_mapping(src_vma->vm_flags); |
ac46d4f3 JG |
1270 | |
1271 | if (is_cow) { | |
7269f999 | 1272 | mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_PAGE, |
c78f4636 | 1273 | 0, src_vma, src_mm, addr, end); |
ac46d4f3 | 1274 | mmu_notifier_invalidate_range_start(&range); |
57efa1fe JG |
1275 | /* |
1276 | * Disabling preemption is not needed for the write side, as | |
1277 | * the read side doesn't spin, but goes to the mmap_lock. | |
1278 | * | |
1279 | * Use the raw variant of the seqcount_t write API to avoid | |
1280 | * lockdep complaining about preemptibility. | |
1281 | */ | |
1282 | mmap_assert_write_locked(src_mm); | |
1283 | raw_write_seqcount_begin(&src_mm->write_protect_seq); | |
ac46d4f3 | 1284 | } |
cddb8a5c AA |
1285 | |
1286 | ret = 0; | |
1da177e4 LT |
1287 | dst_pgd = pgd_offset(dst_mm, addr); |
1288 | src_pgd = pgd_offset(src_mm, addr); | |
1289 | do { | |
1290 | next = pgd_addr_end(addr, end); | |
1291 | if (pgd_none_or_clear_bad(src_pgd)) | |
1292 | continue; | |
c78f4636 PX |
1293 | if (unlikely(copy_p4d_range(dst_vma, src_vma, dst_pgd, src_pgd, |
1294 | addr, next))) { | |
cddb8a5c AA |
1295 | ret = -ENOMEM; |
1296 | break; | |
1297 | } | |
1da177e4 | 1298 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1299 | |
57efa1fe JG |
1300 | if (is_cow) { |
1301 | raw_write_seqcount_end(&src_mm->write_protect_seq); | |
ac46d4f3 | 1302 | mmu_notifier_invalidate_range_end(&range); |
57efa1fe | 1303 | } |
cddb8a5c | 1304 | return ret; |
1da177e4 LT |
1305 | } |
1306 | ||
51c6f666 | 1307 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1308 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1309 | unsigned long addr, unsigned long end, |
97a89413 | 1310 | struct zap_details *details) |
1da177e4 | 1311 | { |
b5810039 | 1312 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1313 | int force_flush = 0; |
d559db08 | 1314 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1315 | spinlock_t *ptl; |
5f1a1907 | 1316 | pte_t *start_pte; |
97a89413 | 1317 | pte_t *pte; |
8a5f14a2 | 1318 | swp_entry_t entry; |
d559db08 | 1319 | |
ed6a7935 | 1320 | tlb_change_page_size(tlb, PAGE_SIZE); |
d16dfc55 | 1321 | again: |
e303297e | 1322 | init_rss_vec(rss); |
5f1a1907 SR |
1323 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1324 | pte = start_pte; | |
3ea27719 | 1325 | flush_tlb_batched_pending(mm); |
6606c3e0 | 1326 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1327 | do { |
1328 | pte_t ptent = *pte; | |
166f61b9 | 1329 | if (pte_none(ptent)) |
1da177e4 | 1330 | continue; |
6f5e6b9e | 1331 | |
7b167b68 MK |
1332 | if (need_resched()) |
1333 | break; | |
1334 | ||
1da177e4 | 1335 | if (pte_present(ptent)) { |
ee498ed7 | 1336 | struct page *page; |
51c6f666 | 1337 | |
25b2995a | 1338 | page = vm_normal_page(vma, addr, ptent); |
91b61ef3 PX |
1339 | if (unlikely(zap_skip_check_mapping(details, page))) |
1340 | continue; | |
b5810039 | 1341 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1342 | tlb->fullmm); |
1da177e4 LT |
1343 | tlb_remove_tlb_entry(tlb, pte, addr); |
1344 | if (unlikely(!page)) | |
1345 | continue; | |
eca56ff9 JM |
1346 | |
1347 | if (!PageAnon(page)) { | |
1cf35d47 LT |
1348 | if (pte_dirty(ptent)) { |
1349 | force_flush = 1; | |
6237bcd9 | 1350 | set_page_dirty(page); |
1cf35d47 | 1351 | } |
4917e5d0 | 1352 | if (pte_young(ptent) && |
64363aad | 1353 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1354 | mark_page_accessed(page); |
6237bcd9 | 1355 | } |
eca56ff9 | 1356 | rss[mm_counter(page)]--; |
d281ee61 | 1357 | page_remove_rmap(page, false); |
3dc14741 HD |
1358 | if (unlikely(page_mapcount(page) < 0)) |
1359 | print_bad_pte(vma, addr, ptent, page); | |
e9d55e15 | 1360 | if (unlikely(__tlb_remove_page(tlb, page))) { |
1cf35d47 | 1361 | force_flush = 1; |
ce9ec37b | 1362 | addr += PAGE_SIZE; |
d16dfc55 | 1363 | break; |
1cf35d47 | 1364 | } |
1da177e4 LT |
1365 | continue; |
1366 | } | |
5042db43 JG |
1367 | |
1368 | entry = pte_to_swp_entry(ptent); | |
b756a3b5 AP |
1369 | if (is_device_private_entry(entry) || |
1370 | is_device_exclusive_entry(entry)) { | |
af5cdaf8 | 1371 | struct page *page = pfn_swap_entry_to_page(entry); |
5042db43 | 1372 | |
91b61ef3 PX |
1373 | if (unlikely(zap_skip_check_mapping(details, page))) |
1374 | continue; | |
5042db43 JG |
1375 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
1376 | rss[mm_counter(page)]--; | |
b756a3b5 AP |
1377 | |
1378 | if (is_device_private_entry(entry)) | |
1379 | page_remove_rmap(page, false); | |
1380 | ||
5042db43 JG |
1381 | put_page(page); |
1382 | continue; | |
1383 | } | |
1384 | ||
3e8715fd KS |
1385 | /* If details->check_mapping, we leave swap entries. */ |
1386 | if (unlikely(details)) | |
1da177e4 | 1387 | continue; |
b084d435 | 1388 | |
8a5f14a2 KS |
1389 | if (!non_swap_entry(entry)) |
1390 | rss[MM_SWAPENTS]--; | |
1391 | else if (is_migration_entry(entry)) { | |
1392 | struct page *page; | |
9f9f1acd | 1393 | |
af5cdaf8 | 1394 | page = pfn_swap_entry_to_page(entry); |
eca56ff9 | 1395 | rss[mm_counter(page)]--; |
b084d435 | 1396 | } |
8a5f14a2 KS |
1397 | if (unlikely(!free_swap_and_cache(entry))) |
1398 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 1399 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1400 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1401 | |
d559db08 | 1402 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1403 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1404 | |
1cf35d47 | 1405 | /* Do the actual TLB flush before dropping ptl */ |
fb7332a9 | 1406 | if (force_flush) |
1cf35d47 | 1407 | tlb_flush_mmu_tlbonly(tlb); |
1cf35d47 LT |
1408 | pte_unmap_unlock(start_pte, ptl); |
1409 | ||
1410 | /* | |
1411 | * If we forced a TLB flush (either due to running out of | |
1412 | * batch buffers or because we needed to flush dirty TLB | |
1413 | * entries before releasing the ptl), free the batched | |
1414 | * memory too. Restart if we didn't do everything. | |
1415 | */ | |
1416 | if (force_flush) { | |
1417 | force_flush = 0; | |
fa0aafb8 | 1418 | tlb_flush_mmu(tlb); |
7b167b68 MK |
1419 | } |
1420 | ||
1421 | if (addr != end) { | |
1422 | cond_resched(); | |
1423 | goto again; | |
d16dfc55 PZ |
1424 | } |
1425 | ||
51c6f666 | 1426 | return addr; |
1da177e4 LT |
1427 | } |
1428 | ||
51c6f666 | 1429 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1430 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1431 | unsigned long addr, unsigned long end, |
97a89413 | 1432 | struct zap_details *details) |
1da177e4 LT |
1433 | { |
1434 | pmd_t *pmd; | |
1435 | unsigned long next; | |
1436 | ||
1437 | pmd = pmd_offset(pud, addr); | |
1438 | do { | |
1439 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 1440 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
53406ed1 | 1441 | if (next - addr != HPAGE_PMD_SIZE) |
fd60775a | 1442 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
53406ed1 | 1443 | else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1444 | goto next; |
71e3aac0 | 1445 | /* fall through */ |
22061a1f HD |
1446 | } else if (details && details->single_page && |
1447 | PageTransCompound(details->single_page) && | |
1448 | next - addr == HPAGE_PMD_SIZE && pmd_none(*pmd)) { | |
1449 | spinlock_t *ptl = pmd_lock(tlb->mm, pmd); | |
1450 | /* | |
1451 | * Take and drop THP pmd lock so that we cannot return | |
1452 | * prematurely, while zap_huge_pmd() has cleared *pmd, | |
1453 | * but not yet decremented compound_mapcount(). | |
1454 | */ | |
1455 | spin_unlock(ptl); | |
71e3aac0 | 1456 | } |
22061a1f | 1457 | |
1a5a9906 AA |
1458 | /* |
1459 | * Here there can be other concurrent MADV_DONTNEED or | |
1460 | * trans huge page faults running, and if the pmd is | |
1461 | * none or trans huge it can change under us. This is | |
c1e8d7c6 | 1462 | * because MADV_DONTNEED holds the mmap_lock in read |
1a5a9906 AA |
1463 | * mode. |
1464 | */ | |
1465 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1466 | goto next; | |
97a89413 | 1467 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1468 | next: |
97a89413 PZ |
1469 | cond_resched(); |
1470 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1471 | |
1472 | return addr; | |
1da177e4 LT |
1473 | } |
1474 | ||
51c6f666 | 1475 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
c2febafc | 1476 | struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1477 | unsigned long addr, unsigned long end, |
97a89413 | 1478 | struct zap_details *details) |
1da177e4 LT |
1479 | { |
1480 | pud_t *pud; | |
1481 | unsigned long next; | |
1482 | ||
c2febafc | 1483 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1484 | do { |
1485 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1486 | if (pud_trans_huge(*pud) || pud_devmap(*pud)) { |
1487 | if (next - addr != HPAGE_PUD_SIZE) { | |
42fc5414 | 1488 | mmap_assert_locked(tlb->mm); |
a00cc7d9 MW |
1489 | split_huge_pud(vma, pud, addr); |
1490 | } else if (zap_huge_pud(tlb, vma, pud, addr)) | |
1491 | goto next; | |
1492 | /* fall through */ | |
1493 | } | |
97a89413 | 1494 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1495 | continue; |
97a89413 | 1496 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
a00cc7d9 MW |
1497 | next: |
1498 | cond_resched(); | |
97a89413 | 1499 | } while (pud++, addr = next, addr != end); |
51c6f666 RH |
1500 | |
1501 | return addr; | |
1da177e4 LT |
1502 | } |
1503 | ||
c2febafc KS |
1504 | static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, |
1505 | struct vm_area_struct *vma, pgd_t *pgd, | |
1506 | unsigned long addr, unsigned long end, | |
1507 | struct zap_details *details) | |
1508 | { | |
1509 | p4d_t *p4d; | |
1510 | unsigned long next; | |
1511 | ||
1512 | p4d = p4d_offset(pgd, addr); | |
1513 | do { | |
1514 | next = p4d_addr_end(addr, end); | |
1515 | if (p4d_none_or_clear_bad(p4d)) | |
1516 | continue; | |
1517 | next = zap_pud_range(tlb, vma, p4d, addr, next, details); | |
1518 | } while (p4d++, addr = next, addr != end); | |
1519 | ||
1520 | return addr; | |
1521 | } | |
1522 | ||
aac45363 | 1523 | void unmap_page_range(struct mmu_gather *tlb, |
038c7aa1 AV |
1524 | struct vm_area_struct *vma, |
1525 | unsigned long addr, unsigned long end, | |
1526 | struct zap_details *details) | |
1da177e4 LT |
1527 | { |
1528 | pgd_t *pgd; | |
1529 | unsigned long next; | |
1530 | ||
1da177e4 LT |
1531 | BUG_ON(addr >= end); |
1532 | tlb_start_vma(tlb, vma); | |
1533 | pgd = pgd_offset(vma->vm_mm, addr); | |
1534 | do { | |
1535 | next = pgd_addr_end(addr, end); | |
97a89413 | 1536 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1537 | continue; |
c2febafc | 1538 | next = zap_p4d_range(tlb, vma, pgd, addr, next, details); |
97a89413 | 1539 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
1540 | tlb_end_vma(tlb, vma); |
1541 | } | |
51c6f666 | 1542 | |
f5cc4eef AV |
1543 | |
1544 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1545 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1546 | unsigned long end_addr, |
f5cc4eef AV |
1547 | struct zap_details *details) |
1548 | { | |
1549 | unsigned long start = max(vma->vm_start, start_addr); | |
1550 | unsigned long end; | |
1551 | ||
1552 | if (start >= vma->vm_end) | |
1553 | return; | |
1554 | end = min(vma->vm_end, end_addr); | |
1555 | if (end <= vma->vm_start) | |
1556 | return; | |
1557 | ||
cbc91f71 SD |
1558 | if (vma->vm_file) |
1559 | uprobe_munmap(vma, start, end); | |
1560 | ||
b3b9c293 | 1561 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1562 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1563 | |
1564 | if (start != end) { | |
1565 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1566 | /* | |
1567 | * It is undesirable to test vma->vm_file as it | |
1568 | * should be non-null for valid hugetlb area. | |
1569 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1570 | * cleanup path of mmap_region. When |
f5cc4eef | 1571 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1572 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1573 | * before calling this function to clean up. |
1574 | * Since no pte has actually been setup, it is | |
1575 | * safe to do nothing in this case. | |
1576 | */ | |
24669e58 | 1577 | if (vma->vm_file) { |
83cde9e8 | 1578 | i_mmap_lock_write(vma->vm_file->f_mapping); |
d833352a | 1579 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
83cde9e8 | 1580 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
24669e58 | 1581 | } |
f5cc4eef AV |
1582 | } else |
1583 | unmap_page_range(tlb, vma, start, end, details); | |
1584 | } | |
1da177e4 LT |
1585 | } |
1586 | ||
1da177e4 LT |
1587 | /** |
1588 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1589 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1590 | * @vma: the starting vma |
1591 | * @start_addr: virtual address at which to start unmapping | |
1592 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1593 | * |
508034a3 | 1594 | * Unmap all pages in the vma list. |
1da177e4 | 1595 | * |
1da177e4 LT |
1596 | * Only addresses between `start' and `end' will be unmapped. |
1597 | * | |
1598 | * The VMA list must be sorted in ascending virtual address order. | |
1599 | * | |
1600 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1601 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1602 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1603 | * drops the lock and schedules. | |
1604 | */ | |
6e8bb019 | 1605 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1606 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1607 | unsigned long end_addr) |
1da177e4 | 1608 | { |
ac46d4f3 | 1609 | struct mmu_notifier_range range; |
1da177e4 | 1610 | |
6f4f13e8 JG |
1611 | mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, vma->vm_mm, |
1612 | start_addr, end_addr); | |
ac46d4f3 | 1613 | mmu_notifier_invalidate_range_start(&range); |
f5cc4eef | 1614 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1615 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
ac46d4f3 | 1616 | mmu_notifier_invalidate_range_end(&range); |
1da177e4 LT |
1617 | } |
1618 | ||
1619 | /** | |
1620 | * zap_page_range - remove user pages in a given range | |
1621 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1622 | * @start: starting address of pages to zap |
1da177e4 | 1623 | * @size: number of bytes to zap |
f5cc4eef AV |
1624 | * |
1625 | * Caller must protect the VMA list | |
1da177e4 | 1626 | */ |
7e027b14 | 1627 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
ecf1385d | 1628 | unsigned long size) |
1da177e4 | 1629 | { |
ac46d4f3 | 1630 | struct mmu_notifier_range range; |
d16dfc55 | 1631 | struct mmu_gather tlb; |
1da177e4 | 1632 | |
1da177e4 | 1633 | lru_add_drain(); |
7269f999 | 1634 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, vma->vm_mm, |
6f4f13e8 | 1635 | start, start + size); |
a72afd87 | 1636 | tlb_gather_mmu(&tlb, vma->vm_mm); |
ac46d4f3 JG |
1637 | update_hiwater_rss(vma->vm_mm); |
1638 | mmu_notifier_invalidate_range_start(&range); | |
1639 | for ( ; vma && vma->vm_start < range.end; vma = vma->vm_next) | |
1640 | unmap_single_vma(&tlb, vma, start, range.end, NULL); | |
1641 | mmu_notifier_invalidate_range_end(&range); | |
ae8eba8b | 1642 | tlb_finish_mmu(&tlb); |
1da177e4 LT |
1643 | } |
1644 | ||
f5cc4eef AV |
1645 | /** |
1646 | * zap_page_range_single - remove user pages in a given range | |
1647 | * @vma: vm_area_struct holding the applicable pages | |
1648 | * @address: starting address of pages to zap | |
1649 | * @size: number of bytes to zap | |
8a5f14a2 | 1650 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1651 | * |
1652 | * The range must fit into one VMA. | |
1da177e4 | 1653 | */ |
f5cc4eef | 1654 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1655 | unsigned long size, struct zap_details *details) |
1656 | { | |
ac46d4f3 | 1657 | struct mmu_notifier_range range; |
d16dfc55 | 1658 | struct mmu_gather tlb; |
1da177e4 | 1659 | |
1da177e4 | 1660 | lru_add_drain(); |
7269f999 | 1661 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, vma->vm_mm, |
6f4f13e8 | 1662 | address, address + size); |
a72afd87 | 1663 | tlb_gather_mmu(&tlb, vma->vm_mm); |
ac46d4f3 JG |
1664 | update_hiwater_rss(vma->vm_mm); |
1665 | mmu_notifier_invalidate_range_start(&range); | |
1666 | unmap_single_vma(&tlb, vma, address, range.end, details); | |
1667 | mmu_notifier_invalidate_range_end(&range); | |
ae8eba8b | 1668 | tlb_finish_mmu(&tlb); |
1da177e4 LT |
1669 | } |
1670 | ||
c627f9cc JS |
1671 | /** |
1672 | * zap_vma_ptes - remove ptes mapping the vma | |
1673 | * @vma: vm_area_struct holding ptes to be zapped | |
1674 | * @address: starting address of pages to zap | |
1675 | * @size: number of bytes to zap | |
1676 | * | |
1677 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1678 | * | |
1679 | * The entire address range must be fully contained within the vma. | |
1680 | * | |
c627f9cc | 1681 | */ |
27d036e3 | 1682 | void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, |
c627f9cc JS |
1683 | unsigned long size) |
1684 | { | |
1685 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1686 | !(vma->vm_flags & VM_PFNMAP)) | |
27d036e3 LR |
1687 | return; |
1688 | ||
f5cc4eef | 1689 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1690 | } |
1691 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1692 | ||
8cd3984d | 1693 | static pmd_t *walk_to_pmd(struct mm_struct *mm, unsigned long addr) |
c9cfcddf | 1694 | { |
c2febafc KS |
1695 | pgd_t *pgd; |
1696 | p4d_t *p4d; | |
1697 | pud_t *pud; | |
1698 | pmd_t *pmd; | |
1699 | ||
1700 | pgd = pgd_offset(mm, addr); | |
1701 | p4d = p4d_alloc(mm, pgd, addr); | |
1702 | if (!p4d) | |
1703 | return NULL; | |
1704 | pud = pud_alloc(mm, p4d, addr); | |
1705 | if (!pud) | |
1706 | return NULL; | |
1707 | pmd = pmd_alloc(mm, pud, addr); | |
1708 | if (!pmd) | |
1709 | return NULL; | |
1710 | ||
1711 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
8cd3984d AR |
1712 | return pmd; |
1713 | } | |
1714 | ||
1715 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, | |
1716 | spinlock_t **ptl) | |
1717 | { | |
1718 | pmd_t *pmd = walk_to_pmd(mm, addr); | |
1719 | ||
1720 | if (!pmd) | |
1721 | return NULL; | |
c2febafc | 1722 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
c9cfcddf LT |
1723 | } |
1724 | ||
8efd6f5b AR |
1725 | static int validate_page_before_insert(struct page *page) |
1726 | { | |
1727 | if (PageAnon(page) || PageSlab(page) || page_has_type(page)) | |
1728 | return -EINVAL; | |
1729 | flush_dcache_page(page); | |
1730 | return 0; | |
1731 | } | |
1732 | ||
1733 | static int insert_page_into_pte_locked(struct mm_struct *mm, pte_t *pte, | |
1734 | unsigned long addr, struct page *page, pgprot_t prot) | |
1735 | { | |
1736 | if (!pte_none(*pte)) | |
1737 | return -EBUSY; | |
1738 | /* Ok, finally just insert the thing.. */ | |
1739 | get_page(page); | |
1740 | inc_mm_counter_fast(mm, mm_counter_file(page)); | |
1741 | page_add_file_rmap(page, false); | |
1742 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); | |
1743 | return 0; | |
1744 | } | |
1745 | ||
238f58d8 LT |
1746 | /* |
1747 | * This is the old fallback for page remapping. | |
1748 | * | |
1749 | * For historical reasons, it only allows reserved pages. Only | |
1750 | * old drivers should use this, and they needed to mark their | |
1751 | * pages reserved for the old functions anyway. | |
1752 | */ | |
423bad60 NP |
1753 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1754 | struct page *page, pgprot_t prot) | |
238f58d8 | 1755 | { |
423bad60 | 1756 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1757 | int retval; |
c9cfcddf | 1758 | pte_t *pte; |
8a9f3ccd BS |
1759 | spinlock_t *ptl; |
1760 | ||
8efd6f5b AR |
1761 | retval = validate_page_before_insert(page); |
1762 | if (retval) | |
5b4e655e | 1763 | goto out; |
238f58d8 | 1764 | retval = -ENOMEM; |
c9cfcddf | 1765 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1766 | if (!pte) |
5b4e655e | 1767 | goto out; |
8efd6f5b | 1768 | retval = insert_page_into_pte_locked(mm, pte, addr, page, prot); |
238f58d8 LT |
1769 | pte_unmap_unlock(pte, ptl); |
1770 | out: | |
1771 | return retval; | |
1772 | } | |
1773 | ||
8cd3984d | 1774 | #ifdef pte_index |
7f70c2a6 | 1775 | static int insert_page_in_batch_locked(struct mm_struct *mm, pte_t *pte, |
8cd3984d AR |
1776 | unsigned long addr, struct page *page, pgprot_t prot) |
1777 | { | |
1778 | int err; | |
1779 | ||
1780 | if (!page_count(page)) | |
1781 | return -EINVAL; | |
1782 | err = validate_page_before_insert(page); | |
7f70c2a6 AR |
1783 | if (err) |
1784 | return err; | |
1785 | return insert_page_into_pte_locked(mm, pte, addr, page, prot); | |
8cd3984d AR |
1786 | } |
1787 | ||
1788 | /* insert_pages() amortizes the cost of spinlock operations | |
1789 | * when inserting pages in a loop. Arch *must* define pte_index. | |
1790 | */ | |
1791 | static int insert_pages(struct vm_area_struct *vma, unsigned long addr, | |
1792 | struct page **pages, unsigned long *num, pgprot_t prot) | |
1793 | { | |
1794 | pmd_t *pmd = NULL; | |
7f70c2a6 AR |
1795 | pte_t *start_pte, *pte; |
1796 | spinlock_t *pte_lock; | |
8cd3984d AR |
1797 | struct mm_struct *const mm = vma->vm_mm; |
1798 | unsigned long curr_page_idx = 0; | |
1799 | unsigned long remaining_pages_total = *num; | |
1800 | unsigned long pages_to_write_in_pmd; | |
1801 | int ret; | |
1802 | more: | |
1803 | ret = -EFAULT; | |
1804 | pmd = walk_to_pmd(mm, addr); | |
1805 | if (!pmd) | |
1806 | goto out; | |
1807 | ||
1808 | pages_to_write_in_pmd = min_t(unsigned long, | |
1809 | remaining_pages_total, PTRS_PER_PTE - pte_index(addr)); | |
1810 | ||
1811 | /* Allocate the PTE if necessary; takes PMD lock once only. */ | |
1812 | ret = -ENOMEM; | |
1813 | if (pte_alloc(mm, pmd)) | |
1814 | goto out; | |
8cd3984d AR |
1815 | |
1816 | while (pages_to_write_in_pmd) { | |
1817 | int pte_idx = 0; | |
1818 | const int batch_size = min_t(int, pages_to_write_in_pmd, 8); | |
1819 | ||
7f70c2a6 AR |
1820 | start_pte = pte_offset_map_lock(mm, pmd, addr, &pte_lock); |
1821 | for (pte = start_pte; pte_idx < batch_size; ++pte, ++pte_idx) { | |
1822 | int err = insert_page_in_batch_locked(mm, pte, | |
8cd3984d AR |
1823 | addr, pages[curr_page_idx], prot); |
1824 | if (unlikely(err)) { | |
7f70c2a6 | 1825 | pte_unmap_unlock(start_pte, pte_lock); |
8cd3984d AR |
1826 | ret = err; |
1827 | remaining_pages_total -= pte_idx; | |
1828 | goto out; | |
1829 | } | |
1830 | addr += PAGE_SIZE; | |
1831 | ++curr_page_idx; | |
1832 | } | |
7f70c2a6 | 1833 | pte_unmap_unlock(start_pte, pte_lock); |
8cd3984d AR |
1834 | pages_to_write_in_pmd -= batch_size; |
1835 | remaining_pages_total -= batch_size; | |
1836 | } | |
1837 | if (remaining_pages_total) | |
1838 | goto more; | |
1839 | ret = 0; | |
1840 | out: | |
1841 | *num = remaining_pages_total; | |
1842 | return ret; | |
1843 | } | |
1844 | #endif /* ifdef pte_index */ | |
1845 | ||
1846 | /** | |
1847 | * vm_insert_pages - insert multiple pages into user vma, batching the pmd lock. | |
1848 | * @vma: user vma to map to | |
1849 | * @addr: target start user address of these pages | |
1850 | * @pages: source kernel pages | |
1851 | * @num: in: number of pages to map. out: number of pages that were *not* | |
1852 | * mapped. (0 means all pages were successfully mapped). | |
1853 | * | |
1854 | * Preferred over vm_insert_page() when inserting multiple pages. | |
1855 | * | |
1856 | * In case of error, we may have mapped a subset of the provided | |
1857 | * pages. It is the caller's responsibility to account for this case. | |
1858 | * | |
1859 | * The same restrictions apply as in vm_insert_page(). | |
1860 | */ | |
1861 | int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr, | |
1862 | struct page **pages, unsigned long *num) | |
1863 | { | |
1864 | #ifdef pte_index | |
1865 | const unsigned long end_addr = addr + (*num * PAGE_SIZE) - 1; | |
1866 | ||
1867 | if (addr < vma->vm_start || end_addr >= vma->vm_end) | |
1868 | return -EFAULT; | |
1869 | if (!(vma->vm_flags & VM_MIXEDMAP)) { | |
d8ed45c5 | 1870 | BUG_ON(mmap_read_trylock(vma->vm_mm)); |
8cd3984d AR |
1871 | BUG_ON(vma->vm_flags & VM_PFNMAP); |
1872 | vma->vm_flags |= VM_MIXEDMAP; | |
1873 | } | |
1874 | /* Defer page refcount checking till we're about to map that page. */ | |
1875 | return insert_pages(vma, addr, pages, num, vma->vm_page_prot); | |
1876 | #else | |
1877 | unsigned long idx = 0, pgcount = *num; | |
45779b03 | 1878 | int err = -EINVAL; |
8cd3984d AR |
1879 | |
1880 | for (; idx < pgcount; ++idx) { | |
1881 | err = vm_insert_page(vma, addr + (PAGE_SIZE * idx), pages[idx]); | |
1882 | if (err) | |
1883 | break; | |
1884 | } | |
1885 | *num = pgcount - idx; | |
1886 | return err; | |
1887 | #endif /* ifdef pte_index */ | |
1888 | } | |
1889 | EXPORT_SYMBOL(vm_insert_pages); | |
1890 | ||
bfa5bf6d REB |
1891 | /** |
1892 | * vm_insert_page - insert single page into user vma | |
1893 | * @vma: user vma to map to | |
1894 | * @addr: target user address of this page | |
1895 | * @page: source kernel page | |
1896 | * | |
a145dd41 LT |
1897 | * This allows drivers to insert individual pages they've allocated |
1898 | * into a user vma. | |
1899 | * | |
1900 | * The page has to be a nice clean _individual_ kernel allocation. | |
1901 | * If you allocate a compound page, you need to have marked it as | |
1902 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1903 | * (see split_page()). |
a145dd41 LT |
1904 | * |
1905 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1906 | * took an arbitrary page protection parameter. This doesn't allow | |
1907 | * that. Your vma protection will have to be set up correctly, which | |
1908 | * means that if you want a shared writable mapping, you'd better | |
1909 | * ask for a shared writable mapping! | |
1910 | * | |
1911 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1912 | * |
1913 | * Usually this function is called from f_op->mmap() handler | |
c1e8d7c6 | 1914 | * under mm->mmap_lock write-lock, so it can change vma->vm_flags. |
4b6e1e37 KK |
1915 | * Caller must set VM_MIXEDMAP on vma if it wants to call this |
1916 | * function from other places, for example from page-fault handler. | |
a862f68a MR |
1917 | * |
1918 | * Return: %0 on success, negative error code otherwise. | |
a145dd41 | 1919 | */ |
423bad60 NP |
1920 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1921 | struct page *page) | |
a145dd41 LT |
1922 | { |
1923 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1924 | return -EFAULT; | |
1925 | if (!page_count(page)) | |
1926 | return -EINVAL; | |
4b6e1e37 | 1927 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
d8ed45c5 | 1928 | BUG_ON(mmap_read_trylock(vma->vm_mm)); |
4b6e1e37 KK |
1929 | BUG_ON(vma->vm_flags & VM_PFNMAP); |
1930 | vma->vm_flags |= VM_MIXEDMAP; | |
1931 | } | |
423bad60 | 1932 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1933 | } |
e3c3374f | 1934 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1935 | |
a667d745 SJ |
1936 | /* |
1937 | * __vm_map_pages - maps range of kernel pages into user vma | |
1938 | * @vma: user vma to map to | |
1939 | * @pages: pointer to array of source kernel pages | |
1940 | * @num: number of pages in page array | |
1941 | * @offset: user's requested vm_pgoff | |
1942 | * | |
1943 | * This allows drivers to map range of kernel pages into a user vma. | |
1944 | * | |
1945 | * Return: 0 on success and error code otherwise. | |
1946 | */ | |
1947 | static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages, | |
1948 | unsigned long num, unsigned long offset) | |
1949 | { | |
1950 | unsigned long count = vma_pages(vma); | |
1951 | unsigned long uaddr = vma->vm_start; | |
1952 | int ret, i; | |
1953 | ||
1954 | /* Fail if the user requested offset is beyond the end of the object */ | |
96756fcb | 1955 | if (offset >= num) |
a667d745 SJ |
1956 | return -ENXIO; |
1957 | ||
1958 | /* Fail if the user requested size exceeds available object size */ | |
1959 | if (count > num - offset) | |
1960 | return -ENXIO; | |
1961 | ||
1962 | for (i = 0; i < count; i++) { | |
1963 | ret = vm_insert_page(vma, uaddr, pages[offset + i]); | |
1964 | if (ret < 0) | |
1965 | return ret; | |
1966 | uaddr += PAGE_SIZE; | |
1967 | } | |
1968 | ||
1969 | return 0; | |
1970 | } | |
1971 | ||
1972 | /** | |
1973 | * vm_map_pages - maps range of kernel pages starts with non zero offset | |
1974 | * @vma: user vma to map to | |
1975 | * @pages: pointer to array of source kernel pages | |
1976 | * @num: number of pages in page array | |
1977 | * | |
1978 | * Maps an object consisting of @num pages, catering for the user's | |
1979 | * requested vm_pgoff | |
1980 | * | |
1981 | * If we fail to insert any page into the vma, the function will return | |
1982 | * immediately leaving any previously inserted pages present. Callers | |
1983 | * from the mmap handler may immediately return the error as their caller | |
1984 | * will destroy the vma, removing any successfully inserted pages. Other | |
1985 | * callers should make their own arrangements for calling unmap_region(). | |
1986 | * | |
1987 | * Context: Process context. Called by mmap handlers. | |
1988 | * Return: 0 on success and error code otherwise. | |
1989 | */ | |
1990 | int vm_map_pages(struct vm_area_struct *vma, struct page **pages, | |
1991 | unsigned long num) | |
1992 | { | |
1993 | return __vm_map_pages(vma, pages, num, vma->vm_pgoff); | |
1994 | } | |
1995 | EXPORT_SYMBOL(vm_map_pages); | |
1996 | ||
1997 | /** | |
1998 | * vm_map_pages_zero - map range of kernel pages starts with zero offset | |
1999 | * @vma: user vma to map to | |
2000 | * @pages: pointer to array of source kernel pages | |
2001 | * @num: number of pages in page array | |
2002 | * | |
2003 | * Similar to vm_map_pages(), except that it explicitly sets the offset | |
2004 | * to 0. This function is intended for the drivers that did not consider | |
2005 | * vm_pgoff. | |
2006 | * | |
2007 | * Context: Process context. Called by mmap handlers. | |
2008 | * Return: 0 on success and error code otherwise. | |
2009 | */ | |
2010 | int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, | |
2011 | unsigned long num) | |
2012 | { | |
2013 | return __vm_map_pages(vma, pages, num, 0); | |
2014 | } | |
2015 | EXPORT_SYMBOL(vm_map_pages_zero); | |
2016 | ||
9b5a8e00 | 2017 | static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
b2770da6 | 2018 | pfn_t pfn, pgprot_t prot, bool mkwrite) |
423bad60 NP |
2019 | { |
2020 | struct mm_struct *mm = vma->vm_mm; | |
423bad60 NP |
2021 | pte_t *pte, entry; |
2022 | spinlock_t *ptl; | |
2023 | ||
423bad60 NP |
2024 | pte = get_locked_pte(mm, addr, &ptl); |
2025 | if (!pte) | |
9b5a8e00 | 2026 | return VM_FAULT_OOM; |
b2770da6 RZ |
2027 | if (!pte_none(*pte)) { |
2028 | if (mkwrite) { | |
2029 | /* | |
2030 | * For read faults on private mappings the PFN passed | |
2031 | * in may not match the PFN we have mapped if the | |
2032 | * mapped PFN is a writeable COW page. In the mkwrite | |
2033 | * case we are creating a writable PTE for a shared | |
f2c57d91 JK |
2034 | * mapping and we expect the PFNs to match. If they |
2035 | * don't match, we are likely racing with block | |
2036 | * allocation and mapping invalidation so just skip the | |
2037 | * update. | |
b2770da6 | 2038 | */ |
f2c57d91 JK |
2039 | if (pte_pfn(*pte) != pfn_t_to_pfn(pfn)) { |
2040 | WARN_ON_ONCE(!is_zero_pfn(pte_pfn(*pte))); | |
b2770da6 | 2041 | goto out_unlock; |
f2c57d91 | 2042 | } |
cae85cb8 JK |
2043 | entry = pte_mkyoung(*pte); |
2044 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2045 | if (ptep_set_access_flags(vma, addr, pte, entry, 1)) | |
2046 | update_mmu_cache(vma, addr, pte); | |
2047 | } | |
2048 | goto out_unlock; | |
b2770da6 | 2049 | } |
423bad60 NP |
2050 | |
2051 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
2052 | if (pfn_t_devmap(pfn)) |
2053 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
2054 | else | |
2055 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
b2770da6 | 2056 | |
b2770da6 RZ |
2057 | if (mkwrite) { |
2058 | entry = pte_mkyoung(entry); | |
2059 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2060 | } | |
2061 | ||
423bad60 | 2062 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 2063 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 | 2064 | |
423bad60 NP |
2065 | out_unlock: |
2066 | pte_unmap_unlock(pte, ptl); | |
9b5a8e00 | 2067 | return VM_FAULT_NOPAGE; |
423bad60 NP |
2068 | } |
2069 | ||
f5e6d1d5 MW |
2070 | /** |
2071 | * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
2072 | * @vma: user vma to map to | |
2073 | * @addr: target user address of this page | |
2074 | * @pfn: source kernel pfn | |
2075 | * @pgprot: pgprot flags for the inserted page | |
2076 | * | |
a1a0aea5 | 2077 | * This is exactly like vmf_insert_pfn(), except that it allows drivers |
f5e6d1d5 MW |
2078 | * to override pgprot on a per-page basis. |
2079 | * | |
2080 | * This only makes sense for IO mappings, and it makes no sense for | |
2081 | * COW mappings. In general, using multiple vmas is preferable; | |
ae2b01f3 | 2082 | * vmf_insert_pfn_prot should only be used if using multiple VMAs is |
f5e6d1d5 MW |
2083 | * impractical. |
2084 | * | |
574c5b3d TH |
2085 | * See vmf_insert_mixed_prot() for a discussion of the implication of using |
2086 | * a value of @pgprot different from that of @vma->vm_page_prot. | |
2087 | * | |
ae2b01f3 | 2088 | * Context: Process context. May allocate using %GFP_KERNEL. |
f5e6d1d5 MW |
2089 | * Return: vm_fault_t value. |
2090 | */ | |
2091 | vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
2092 | unsigned long pfn, pgprot_t pgprot) | |
2093 | { | |
6d958546 MW |
2094 | /* |
2095 | * Technically, architectures with pte_special can avoid all these | |
2096 | * restrictions (same for remap_pfn_range). However we would like | |
2097 | * consistency in testing and feature parity among all, so we should | |
2098 | * try to keep these invariants in place for everybody. | |
2099 | */ | |
2100 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); | |
2101 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
2102 | (VM_PFNMAP|VM_MIXEDMAP)); | |
2103 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
2104 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
2105 | ||
2106 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
2107 | return VM_FAULT_SIGBUS; | |
2108 | ||
2109 | if (!pfn_modify_allowed(pfn, pgprot)) | |
2110 | return VM_FAULT_SIGBUS; | |
2111 | ||
2112 | track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); | |
2113 | ||
9b5a8e00 | 2114 | return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, |
6d958546 | 2115 | false); |
f5e6d1d5 MW |
2116 | } |
2117 | EXPORT_SYMBOL(vmf_insert_pfn_prot); | |
e0dc0d8f | 2118 | |
ae2b01f3 MW |
2119 | /** |
2120 | * vmf_insert_pfn - insert single pfn into user vma | |
2121 | * @vma: user vma to map to | |
2122 | * @addr: target user address of this page | |
2123 | * @pfn: source kernel pfn | |
2124 | * | |
2125 | * Similar to vm_insert_page, this allows drivers to insert individual pages | |
2126 | * they've allocated into a user vma. Same comments apply. | |
2127 | * | |
2128 | * This function should only be called from a vm_ops->fault handler, and | |
2129 | * in that case the handler should return the result of this function. | |
2130 | * | |
2131 | * vma cannot be a COW mapping. | |
2132 | * | |
2133 | * As this is called only for pages that do not currently exist, we | |
2134 | * do not need to flush old virtual caches or the TLB. | |
2135 | * | |
2136 | * Context: Process context. May allocate using %GFP_KERNEL. | |
2137 | * Return: vm_fault_t value. | |
2138 | */ | |
2139 | vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
2140 | unsigned long pfn) | |
2141 | { | |
2142 | return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
2143 | } | |
2144 | EXPORT_SYMBOL(vmf_insert_pfn); | |
2145 | ||
785a3fab DW |
2146 | static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) |
2147 | { | |
2148 | /* these checks mirror the abort conditions in vm_normal_page */ | |
2149 | if (vma->vm_flags & VM_MIXEDMAP) | |
2150 | return true; | |
2151 | if (pfn_t_devmap(pfn)) | |
2152 | return true; | |
2153 | if (pfn_t_special(pfn)) | |
2154 | return true; | |
2155 | if (is_zero_pfn(pfn_t_to_pfn(pfn))) | |
2156 | return true; | |
2157 | return false; | |
2158 | } | |
2159 | ||
79f3aa5b | 2160 | static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma, |
574c5b3d TH |
2161 | unsigned long addr, pfn_t pfn, pgprot_t pgprot, |
2162 | bool mkwrite) | |
423bad60 | 2163 | { |
79f3aa5b | 2164 | int err; |
87744ab3 | 2165 | |
785a3fab | 2166 | BUG_ON(!vm_mixed_ok(vma, pfn)); |
e0dc0d8f | 2167 | |
423bad60 | 2168 | if (addr < vma->vm_start || addr >= vma->vm_end) |
79f3aa5b | 2169 | return VM_FAULT_SIGBUS; |
308a047c BP |
2170 | |
2171 | track_pfn_insert(vma, &pgprot, pfn); | |
e0dc0d8f | 2172 | |
42e4089c | 2173 | if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) |
79f3aa5b | 2174 | return VM_FAULT_SIGBUS; |
42e4089c | 2175 | |
423bad60 NP |
2176 | /* |
2177 | * If we don't have pte special, then we have to use the pfn_valid() | |
2178 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
2179 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
2180 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
2181 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 2182 | */ |
00b3a331 LD |
2183 | if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && |
2184 | !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { | |
423bad60 NP |
2185 | struct page *page; |
2186 | ||
03fc2da6 DW |
2187 | /* |
2188 | * At this point we are committed to insert_page() | |
2189 | * regardless of whether the caller specified flags that | |
2190 | * result in pfn_t_has_page() == false. | |
2191 | */ | |
2192 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
79f3aa5b MW |
2193 | err = insert_page(vma, addr, page, pgprot); |
2194 | } else { | |
9b5a8e00 | 2195 | return insert_pfn(vma, addr, pfn, pgprot, mkwrite); |
423bad60 | 2196 | } |
b2770da6 | 2197 | |
5d747637 MW |
2198 | if (err == -ENOMEM) |
2199 | return VM_FAULT_OOM; | |
2200 | if (err < 0 && err != -EBUSY) | |
2201 | return VM_FAULT_SIGBUS; | |
2202 | ||
2203 | return VM_FAULT_NOPAGE; | |
e0dc0d8f | 2204 | } |
79f3aa5b | 2205 | |
574c5b3d TH |
2206 | /** |
2207 | * vmf_insert_mixed_prot - insert single pfn into user vma with specified pgprot | |
2208 | * @vma: user vma to map to | |
2209 | * @addr: target user address of this page | |
2210 | * @pfn: source kernel pfn | |
2211 | * @pgprot: pgprot flags for the inserted page | |
2212 | * | |
a1a0aea5 | 2213 | * This is exactly like vmf_insert_mixed(), except that it allows drivers |
574c5b3d TH |
2214 | * to override pgprot on a per-page basis. |
2215 | * | |
2216 | * Typically this function should be used by drivers to set caching- and | |
2217 | * encryption bits different than those of @vma->vm_page_prot, because | |
2218 | * the caching- or encryption mode may not be known at mmap() time. | |
2219 | * This is ok as long as @vma->vm_page_prot is not used by the core vm | |
2220 | * to set caching and encryption bits for those vmas (except for COW pages). | |
2221 | * This is ensured by core vm only modifying these page table entries using | |
2222 | * functions that don't touch caching- or encryption bits, using pte_modify() | |
2223 | * if needed. (See for example mprotect()). | |
2224 | * Also when new page-table entries are created, this is only done using the | |
2225 | * fault() callback, and never using the value of vma->vm_page_prot, | |
2226 | * except for page-table entries that point to anonymous pages as the result | |
2227 | * of COW. | |
2228 | * | |
2229 | * Context: Process context. May allocate using %GFP_KERNEL. | |
2230 | * Return: vm_fault_t value. | |
2231 | */ | |
2232 | vm_fault_t vmf_insert_mixed_prot(struct vm_area_struct *vma, unsigned long addr, | |
2233 | pfn_t pfn, pgprot_t pgprot) | |
2234 | { | |
2235 | return __vm_insert_mixed(vma, addr, pfn, pgprot, false); | |
2236 | } | |
5379e4dd | 2237 | EXPORT_SYMBOL(vmf_insert_mixed_prot); |
574c5b3d | 2238 | |
79f3aa5b MW |
2239 | vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
2240 | pfn_t pfn) | |
2241 | { | |
574c5b3d | 2242 | return __vm_insert_mixed(vma, addr, pfn, vma->vm_page_prot, false); |
79f3aa5b | 2243 | } |
5d747637 | 2244 | EXPORT_SYMBOL(vmf_insert_mixed); |
e0dc0d8f | 2245 | |
ab77dab4 SJ |
2246 | /* |
2247 | * If the insertion of PTE failed because someone else already added a | |
2248 | * different entry in the mean time, we treat that as success as we assume | |
2249 | * the same entry was actually inserted. | |
2250 | */ | |
ab77dab4 SJ |
2251 | vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, |
2252 | unsigned long addr, pfn_t pfn) | |
b2770da6 | 2253 | { |
574c5b3d | 2254 | return __vm_insert_mixed(vma, addr, pfn, vma->vm_page_prot, true); |
b2770da6 | 2255 | } |
ab77dab4 | 2256 | EXPORT_SYMBOL(vmf_insert_mixed_mkwrite); |
b2770da6 | 2257 | |
1da177e4 LT |
2258 | /* |
2259 | * maps a range of physical memory into the requested pages. the old | |
2260 | * mappings are removed. any references to nonexistent pages results | |
2261 | * in null mappings (currently treated as "copy-on-access") | |
2262 | */ | |
2263 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
2264 | unsigned long addr, unsigned long end, | |
2265 | unsigned long pfn, pgprot_t prot) | |
2266 | { | |
90a3e375 | 2267 | pte_t *pte, *mapped_pte; |
c74df32c | 2268 | spinlock_t *ptl; |
42e4089c | 2269 | int err = 0; |
1da177e4 | 2270 | |
90a3e375 | 2271 | mapped_pte = pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
2272 | if (!pte) |
2273 | return -ENOMEM; | |
6606c3e0 | 2274 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
2275 | do { |
2276 | BUG_ON(!pte_none(*pte)); | |
42e4089c AK |
2277 | if (!pfn_modify_allowed(pfn, prot)) { |
2278 | err = -EACCES; | |
2279 | break; | |
2280 | } | |
7e675137 | 2281 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
2282 | pfn++; |
2283 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 2284 | arch_leave_lazy_mmu_mode(); |
90a3e375 | 2285 | pte_unmap_unlock(mapped_pte, ptl); |
42e4089c | 2286 | return err; |
1da177e4 LT |
2287 | } |
2288 | ||
2289 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2290 | unsigned long addr, unsigned long end, | |
2291 | unsigned long pfn, pgprot_t prot) | |
2292 | { | |
2293 | pmd_t *pmd; | |
2294 | unsigned long next; | |
42e4089c | 2295 | int err; |
1da177e4 LT |
2296 | |
2297 | pfn -= addr >> PAGE_SHIFT; | |
2298 | pmd = pmd_alloc(mm, pud, addr); | |
2299 | if (!pmd) | |
2300 | return -ENOMEM; | |
f66055ab | 2301 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
2302 | do { |
2303 | next = pmd_addr_end(addr, end); | |
42e4089c AK |
2304 | err = remap_pte_range(mm, pmd, addr, next, |
2305 | pfn + (addr >> PAGE_SHIFT), prot); | |
2306 | if (err) | |
2307 | return err; | |
1da177e4 LT |
2308 | } while (pmd++, addr = next, addr != end); |
2309 | return 0; | |
2310 | } | |
2311 | ||
c2febafc | 2312 | static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, |
1da177e4 LT |
2313 | unsigned long addr, unsigned long end, |
2314 | unsigned long pfn, pgprot_t prot) | |
2315 | { | |
2316 | pud_t *pud; | |
2317 | unsigned long next; | |
42e4089c | 2318 | int err; |
1da177e4 LT |
2319 | |
2320 | pfn -= addr >> PAGE_SHIFT; | |
c2febafc | 2321 | pud = pud_alloc(mm, p4d, addr); |
1da177e4 LT |
2322 | if (!pud) |
2323 | return -ENOMEM; | |
2324 | do { | |
2325 | next = pud_addr_end(addr, end); | |
42e4089c AK |
2326 | err = remap_pmd_range(mm, pud, addr, next, |
2327 | pfn + (addr >> PAGE_SHIFT), prot); | |
2328 | if (err) | |
2329 | return err; | |
1da177e4 LT |
2330 | } while (pud++, addr = next, addr != end); |
2331 | return 0; | |
2332 | } | |
2333 | ||
c2febafc KS |
2334 | static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2335 | unsigned long addr, unsigned long end, | |
2336 | unsigned long pfn, pgprot_t prot) | |
2337 | { | |
2338 | p4d_t *p4d; | |
2339 | unsigned long next; | |
42e4089c | 2340 | int err; |
c2febafc KS |
2341 | |
2342 | pfn -= addr >> PAGE_SHIFT; | |
2343 | p4d = p4d_alloc(mm, pgd, addr); | |
2344 | if (!p4d) | |
2345 | return -ENOMEM; | |
2346 | do { | |
2347 | next = p4d_addr_end(addr, end); | |
42e4089c AK |
2348 | err = remap_pud_range(mm, p4d, addr, next, |
2349 | pfn + (addr >> PAGE_SHIFT), prot); | |
2350 | if (err) | |
2351 | return err; | |
c2febafc KS |
2352 | } while (p4d++, addr = next, addr != end); |
2353 | return 0; | |
2354 | } | |
2355 | ||
74ffa5a3 CH |
2356 | /* |
2357 | * Variant of remap_pfn_range that does not call track_pfn_remap. The caller | |
2358 | * must have pre-validated the caching bits of the pgprot_t. | |
bfa5bf6d | 2359 | */ |
74ffa5a3 CH |
2360 | int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr, |
2361 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1da177e4 LT |
2362 | { |
2363 | pgd_t *pgd; | |
2364 | unsigned long next; | |
2d15cab8 | 2365 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 LT |
2366 | struct mm_struct *mm = vma->vm_mm; |
2367 | int err; | |
2368 | ||
0c4123e3 AZ |
2369 | if (WARN_ON_ONCE(!PAGE_ALIGNED(addr))) |
2370 | return -EINVAL; | |
2371 | ||
1da177e4 LT |
2372 | /* |
2373 | * Physically remapped pages are special. Tell the | |
2374 | * rest of the world about it: | |
2375 | * VM_IO tells people not to look at these pages | |
2376 | * (accesses can have side effects). | |
6aab341e LT |
2377 | * VM_PFNMAP tells the core MM that the base pages are just |
2378 | * raw PFN mappings, and do not have a "struct page" associated | |
2379 | * with them. | |
314e51b9 KK |
2380 | * VM_DONTEXPAND |
2381 | * Disable vma merging and expanding with mremap(). | |
2382 | * VM_DONTDUMP | |
2383 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
2384 | * |
2385 | * There's a horrible special case to handle copy-on-write | |
2386 | * behaviour that some programs depend on. We mark the "original" | |
2387 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 2388 | * See vm_normal_page() for details. |
1da177e4 | 2389 | */ |
b3b9c293 KK |
2390 | if (is_cow_mapping(vma->vm_flags)) { |
2391 | if (addr != vma->vm_start || end != vma->vm_end) | |
2392 | return -EINVAL; | |
fb155c16 | 2393 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
2394 | } |
2395 | ||
314e51b9 | 2396 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
2397 | |
2398 | BUG_ON(addr >= end); | |
2399 | pfn -= addr >> PAGE_SHIFT; | |
2400 | pgd = pgd_offset(mm, addr); | |
2401 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
2402 | do { |
2403 | next = pgd_addr_end(addr, end); | |
c2febafc | 2404 | err = remap_p4d_range(mm, pgd, addr, next, |
1da177e4 LT |
2405 | pfn + (addr >> PAGE_SHIFT), prot); |
2406 | if (err) | |
74ffa5a3 | 2407 | return err; |
1da177e4 | 2408 | } while (pgd++, addr = next, addr != end); |
2ab64037 | 2409 | |
74ffa5a3 CH |
2410 | return 0; |
2411 | } | |
2412 | ||
2413 | /** | |
2414 | * remap_pfn_range - remap kernel memory to userspace | |
2415 | * @vma: user vma to map to | |
2416 | * @addr: target page aligned user address to start at | |
2417 | * @pfn: page frame number of kernel physical memory address | |
2418 | * @size: size of mapping area | |
2419 | * @prot: page protection flags for this mapping | |
2420 | * | |
2421 | * Note: this is only safe if the mm semaphore is held when called. | |
2422 | * | |
2423 | * Return: %0 on success, negative error code otherwise. | |
2424 | */ | |
2425 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, | |
2426 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
2427 | { | |
2428 | int err; | |
2429 | ||
2430 | err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); | |
2ab64037 | 2431 | if (err) |
74ffa5a3 | 2432 | return -EINVAL; |
2ab64037 | 2433 | |
74ffa5a3 CH |
2434 | err = remap_pfn_range_notrack(vma, addr, pfn, size, prot); |
2435 | if (err) | |
2436 | untrack_pfn(vma, pfn, PAGE_ALIGN(size)); | |
1da177e4 LT |
2437 | return err; |
2438 | } | |
2439 | EXPORT_SYMBOL(remap_pfn_range); | |
2440 | ||
b4cbb197 LT |
2441 | /** |
2442 | * vm_iomap_memory - remap memory to userspace | |
2443 | * @vma: user vma to map to | |
abd69b9e | 2444 | * @start: start of the physical memory to be mapped |
b4cbb197 LT |
2445 | * @len: size of area |
2446 | * | |
2447 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
2448 | * driver just needs to give us the physical memory range to be mapped, | |
2449 | * we'll figure out the rest from the vma information. | |
2450 | * | |
2451 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
2452 | * whatever write-combining details or similar. | |
a862f68a MR |
2453 | * |
2454 | * Return: %0 on success, negative error code otherwise. | |
b4cbb197 LT |
2455 | */ |
2456 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
2457 | { | |
2458 | unsigned long vm_len, pfn, pages; | |
2459 | ||
2460 | /* Check that the physical memory area passed in looks valid */ | |
2461 | if (start + len < start) | |
2462 | return -EINVAL; | |
2463 | /* | |
2464 | * You *really* shouldn't map things that aren't page-aligned, | |
2465 | * but we've historically allowed it because IO memory might | |
2466 | * just have smaller alignment. | |
2467 | */ | |
2468 | len += start & ~PAGE_MASK; | |
2469 | pfn = start >> PAGE_SHIFT; | |
2470 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
2471 | if (pfn + pages < pfn) | |
2472 | return -EINVAL; | |
2473 | ||
2474 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
2475 | if (vma->vm_pgoff > pages) | |
2476 | return -EINVAL; | |
2477 | pfn += vma->vm_pgoff; | |
2478 | pages -= vma->vm_pgoff; | |
2479 | ||
2480 | /* Can we fit all of the mapping? */ | |
2481 | vm_len = vma->vm_end - vma->vm_start; | |
2482 | if (vm_len >> PAGE_SHIFT > pages) | |
2483 | return -EINVAL; | |
2484 | ||
2485 | /* Ok, let it rip */ | |
2486 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
2487 | } | |
2488 | EXPORT_SYMBOL(vm_iomap_memory); | |
2489 | ||
aee16b3c JF |
2490 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
2491 | unsigned long addr, unsigned long end, | |
e80d3909 JR |
2492 | pte_fn_t fn, void *data, bool create, |
2493 | pgtbl_mod_mask *mask) | |
aee16b3c | 2494 | { |
8abb50c7 | 2495 | pte_t *pte, *mapped_pte; |
be1db475 | 2496 | int err = 0; |
3f649ab7 | 2497 | spinlock_t *ptl; |
aee16b3c | 2498 | |
be1db475 | 2499 | if (create) { |
8abb50c7 | 2500 | mapped_pte = pte = (mm == &init_mm) ? |
e80d3909 | 2501 | pte_alloc_kernel_track(pmd, addr, mask) : |
be1db475 DA |
2502 | pte_alloc_map_lock(mm, pmd, addr, &ptl); |
2503 | if (!pte) | |
2504 | return -ENOMEM; | |
2505 | } else { | |
8abb50c7 | 2506 | mapped_pte = pte = (mm == &init_mm) ? |
be1db475 DA |
2507 | pte_offset_kernel(pmd, addr) : |
2508 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
2509 | } | |
aee16b3c JF |
2510 | |
2511 | BUG_ON(pmd_huge(*pmd)); | |
2512 | ||
38e0edb1 JF |
2513 | arch_enter_lazy_mmu_mode(); |
2514 | ||
eeb4a05f CH |
2515 | if (fn) { |
2516 | do { | |
2517 | if (create || !pte_none(*pte)) { | |
2518 | err = fn(pte++, addr, data); | |
2519 | if (err) | |
2520 | break; | |
2521 | } | |
2522 | } while (addr += PAGE_SIZE, addr != end); | |
2523 | } | |
e80d3909 | 2524 | *mask |= PGTBL_PTE_MODIFIED; |
aee16b3c | 2525 | |
38e0edb1 JF |
2526 | arch_leave_lazy_mmu_mode(); |
2527 | ||
aee16b3c | 2528 | if (mm != &init_mm) |
8abb50c7 | 2529 | pte_unmap_unlock(mapped_pte, ptl); |
aee16b3c JF |
2530 | return err; |
2531 | } | |
2532 | ||
2533 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2534 | unsigned long addr, unsigned long end, | |
e80d3909 JR |
2535 | pte_fn_t fn, void *data, bool create, |
2536 | pgtbl_mod_mask *mask) | |
aee16b3c JF |
2537 | { |
2538 | pmd_t *pmd; | |
2539 | unsigned long next; | |
be1db475 | 2540 | int err = 0; |
aee16b3c | 2541 | |
ceb86879 AK |
2542 | BUG_ON(pud_huge(*pud)); |
2543 | ||
be1db475 | 2544 | if (create) { |
e80d3909 | 2545 | pmd = pmd_alloc_track(mm, pud, addr, mask); |
be1db475 DA |
2546 | if (!pmd) |
2547 | return -ENOMEM; | |
2548 | } else { | |
2549 | pmd = pmd_offset(pud, addr); | |
2550 | } | |
aee16b3c JF |
2551 | do { |
2552 | next = pmd_addr_end(addr, end); | |
0c95cba4 NP |
2553 | if (pmd_none(*pmd) && !create) |
2554 | continue; | |
2555 | if (WARN_ON_ONCE(pmd_leaf(*pmd))) | |
2556 | return -EINVAL; | |
2557 | if (!pmd_none(*pmd) && WARN_ON_ONCE(pmd_bad(*pmd))) { | |
2558 | if (!create) | |
2559 | continue; | |
2560 | pmd_clear_bad(pmd); | |
be1db475 | 2561 | } |
0c95cba4 NP |
2562 | err = apply_to_pte_range(mm, pmd, addr, next, |
2563 | fn, data, create, mask); | |
2564 | if (err) | |
2565 | break; | |
aee16b3c | 2566 | } while (pmd++, addr = next, addr != end); |
0c95cba4 | 2567 | |
aee16b3c JF |
2568 | return err; |
2569 | } | |
2570 | ||
c2febafc | 2571 | static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, |
aee16b3c | 2572 | unsigned long addr, unsigned long end, |
e80d3909 JR |
2573 | pte_fn_t fn, void *data, bool create, |
2574 | pgtbl_mod_mask *mask) | |
aee16b3c JF |
2575 | { |
2576 | pud_t *pud; | |
2577 | unsigned long next; | |
be1db475 | 2578 | int err = 0; |
aee16b3c | 2579 | |
be1db475 | 2580 | if (create) { |
e80d3909 | 2581 | pud = pud_alloc_track(mm, p4d, addr, mask); |
be1db475 DA |
2582 | if (!pud) |
2583 | return -ENOMEM; | |
2584 | } else { | |
2585 | pud = pud_offset(p4d, addr); | |
2586 | } | |
aee16b3c JF |
2587 | do { |
2588 | next = pud_addr_end(addr, end); | |
0c95cba4 NP |
2589 | if (pud_none(*pud) && !create) |
2590 | continue; | |
2591 | if (WARN_ON_ONCE(pud_leaf(*pud))) | |
2592 | return -EINVAL; | |
2593 | if (!pud_none(*pud) && WARN_ON_ONCE(pud_bad(*pud))) { | |
2594 | if (!create) | |
2595 | continue; | |
2596 | pud_clear_bad(pud); | |
be1db475 | 2597 | } |
0c95cba4 NP |
2598 | err = apply_to_pmd_range(mm, pud, addr, next, |
2599 | fn, data, create, mask); | |
2600 | if (err) | |
2601 | break; | |
aee16b3c | 2602 | } while (pud++, addr = next, addr != end); |
0c95cba4 | 2603 | |
aee16b3c JF |
2604 | return err; |
2605 | } | |
2606 | ||
c2febafc KS |
2607 | static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2608 | unsigned long addr, unsigned long end, | |
e80d3909 JR |
2609 | pte_fn_t fn, void *data, bool create, |
2610 | pgtbl_mod_mask *mask) | |
c2febafc KS |
2611 | { |
2612 | p4d_t *p4d; | |
2613 | unsigned long next; | |
be1db475 | 2614 | int err = 0; |
c2febafc | 2615 | |
be1db475 | 2616 | if (create) { |
e80d3909 | 2617 | p4d = p4d_alloc_track(mm, pgd, addr, mask); |
be1db475 DA |
2618 | if (!p4d) |
2619 | return -ENOMEM; | |
2620 | } else { | |
2621 | p4d = p4d_offset(pgd, addr); | |
2622 | } | |
c2febafc KS |
2623 | do { |
2624 | next = p4d_addr_end(addr, end); | |
0c95cba4 NP |
2625 | if (p4d_none(*p4d) && !create) |
2626 | continue; | |
2627 | if (WARN_ON_ONCE(p4d_leaf(*p4d))) | |
2628 | return -EINVAL; | |
2629 | if (!p4d_none(*p4d) && WARN_ON_ONCE(p4d_bad(*p4d))) { | |
2630 | if (!create) | |
2631 | continue; | |
2632 | p4d_clear_bad(p4d); | |
be1db475 | 2633 | } |
0c95cba4 NP |
2634 | err = apply_to_pud_range(mm, p4d, addr, next, |
2635 | fn, data, create, mask); | |
2636 | if (err) | |
2637 | break; | |
c2febafc | 2638 | } while (p4d++, addr = next, addr != end); |
0c95cba4 | 2639 | |
c2febafc KS |
2640 | return err; |
2641 | } | |
2642 | ||
be1db475 DA |
2643 | static int __apply_to_page_range(struct mm_struct *mm, unsigned long addr, |
2644 | unsigned long size, pte_fn_t fn, | |
2645 | void *data, bool create) | |
aee16b3c JF |
2646 | { |
2647 | pgd_t *pgd; | |
e80d3909 | 2648 | unsigned long start = addr, next; |
57250a5b | 2649 | unsigned long end = addr + size; |
e80d3909 | 2650 | pgtbl_mod_mask mask = 0; |
be1db475 | 2651 | int err = 0; |
aee16b3c | 2652 | |
9cb65bc3 MP |
2653 | if (WARN_ON(addr >= end)) |
2654 | return -EINVAL; | |
2655 | ||
aee16b3c JF |
2656 | pgd = pgd_offset(mm, addr); |
2657 | do { | |
2658 | next = pgd_addr_end(addr, end); | |
0c95cba4 | 2659 | if (pgd_none(*pgd) && !create) |
be1db475 | 2660 | continue; |
0c95cba4 NP |
2661 | if (WARN_ON_ONCE(pgd_leaf(*pgd))) |
2662 | return -EINVAL; | |
2663 | if (!pgd_none(*pgd) && WARN_ON_ONCE(pgd_bad(*pgd))) { | |
2664 | if (!create) | |
2665 | continue; | |
2666 | pgd_clear_bad(pgd); | |
2667 | } | |
2668 | err = apply_to_p4d_range(mm, pgd, addr, next, | |
2669 | fn, data, create, &mask); | |
aee16b3c JF |
2670 | if (err) |
2671 | break; | |
2672 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2673 | |
e80d3909 JR |
2674 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
2675 | arch_sync_kernel_mappings(start, start + size); | |
2676 | ||
aee16b3c JF |
2677 | return err; |
2678 | } | |
be1db475 DA |
2679 | |
2680 | /* | |
2681 | * Scan a region of virtual memory, filling in page tables as necessary | |
2682 | * and calling a provided function on each leaf page table. | |
2683 | */ | |
2684 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2685 | unsigned long size, pte_fn_t fn, void *data) | |
2686 | { | |
2687 | return __apply_to_page_range(mm, addr, size, fn, data, true); | |
2688 | } | |
aee16b3c JF |
2689 | EXPORT_SYMBOL_GPL(apply_to_page_range); |
2690 | ||
be1db475 DA |
2691 | /* |
2692 | * Scan a region of virtual memory, calling a provided function on | |
2693 | * each leaf page table where it exists. | |
2694 | * | |
2695 | * Unlike apply_to_page_range, this does _not_ fill in page tables | |
2696 | * where they are absent. | |
2697 | */ | |
2698 | int apply_to_existing_page_range(struct mm_struct *mm, unsigned long addr, | |
2699 | unsigned long size, pte_fn_t fn, void *data) | |
2700 | { | |
2701 | return __apply_to_page_range(mm, addr, size, fn, data, false); | |
2702 | } | |
2703 | EXPORT_SYMBOL_GPL(apply_to_existing_page_range); | |
2704 | ||
8f4e2101 | 2705 | /* |
9b4bdd2f KS |
2706 | * handle_pte_fault chooses page fault handler according to an entry which was |
2707 | * read non-atomically. Before making any commitment, on those architectures | |
2708 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
2709 | * parts, do_swap_page must check under lock before unmapping the pte and | |
2710 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2711 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2712 | */ |
2ca99358 | 2713 | static inline int pte_unmap_same(struct vm_fault *vmf) |
8f4e2101 HD |
2714 | { |
2715 | int same = 1; | |
923717cb | 2716 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPTION) |
8f4e2101 | 2717 | if (sizeof(pte_t) > sizeof(unsigned long)) { |
2ca99358 | 2718 | spinlock_t *ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); |
4c21e2f2 | 2719 | spin_lock(ptl); |
2ca99358 | 2720 | same = pte_same(*vmf->pte, vmf->orig_pte); |
4c21e2f2 | 2721 | spin_unlock(ptl); |
8f4e2101 HD |
2722 | } |
2723 | #endif | |
2ca99358 PX |
2724 | pte_unmap(vmf->pte); |
2725 | vmf->pte = NULL; | |
8f4e2101 HD |
2726 | return same; |
2727 | } | |
2728 | ||
83d116c5 JH |
2729 | static inline bool cow_user_page(struct page *dst, struct page *src, |
2730 | struct vm_fault *vmf) | |
6aab341e | 2731 | { |
83d116c5 JH |
2732 | bool ret; |
2733 | void *kaddr; | |
2734 | void __user *uaddr; | |
c3e5ea6e | 2735 | bool locked = false; |
83d116c5 JH |
2736 | struct vm_area_struct *vma = vmf->vma; |
2737 | struct mm_struct *mm = vma->vm_mm; | |
2738 | unsigned long addr = vmf->address; | |
2739 | ||
83d116c5 JH |
2740 | if (likely(src)) { |
2741 | copy_user_highpage(dst, src, addr, vma); | |
2742 | return true; | |
2743 | } | |
2744 | ||
6aab341e LT |
2745 | /* |
2746 | * If the source page was a PFN mapping, we don't have | |
2747 | * a "struct page" for it. We do a best-effort copy by | |
2748 | * just copying from the original user address. If that | |
2749 | * fails, we just zero-fill it. Live with it. | |
2750 | */ | |
83d116c5 JH |
2751 | kaddr = kmap_atomic(dst); |
2752 | uaddr = (void __user *)(addr & PAGE_MASK); | |
2753 | ||
2754 | /* | |
2755 | * On architectures with software "accessed" bits, we would | |
2756 | * take a double page fault, so mark it accessed here. | |
2757 | */ | |
c3e5ea6e | 2758 | if (arch_faults_on_old_pte() && !pte_young(vmf->orig_pte)) { |
83d116c5 | 2759 | pte_t entry; |
5d2a2dbb | 2760 | |
83d116c5 | 2761 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); |
c3e5ea6e | 2762 | locked = true; |
83d116c5 JH |
2763 | if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2764 | /* | |
2765 | * Other thread has already handled the fault | |
7df67697 | 2766 | * and update local tlb only |
83d116c5 | 2767 | */ |
7df67697 | 2768 | update_mmu_tlb(vma, addr, vmf->pte); |
83d116c5 JH |
2769 | ret = false; |
2770 | goto pte_unlock; | |
2771 | } | |
2772 | ||
2773 | entry = pte_mkyoung(vmf->orig_pte); | |
2774 | if (ptep_set_access_flags(vma, addr, vmf->pte, entry, 0)) | |
2775 | update_mmu_cache(vma, addr, vmf->pte); | |
2776 | } | |
2777 | ||
2778 | /* | |
2779 | * This really shouldn't fail, because the page is there | |
2780 | * in the page tables. But it might just be unreadable, | |
2781 | * in which case we just give up and fill the result with | |
2782 | * zeroes. | |
2783 | */ | |
2784 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { | |
c3e5ea6e KS |
2785 | if (locked) |
2786 | goto warn; | |
2787 | ||
2788 | /* Re-validate under PTL if the page is still mapped */ | |
2789 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); | |
2790 | locked = true; | |
2791 | if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) { | |
7df67697 BM |
2792 | /* The PTE changed under us, update local tlb */ |
2793 | update_mmu_tlb(vma, addr, vmf->pte); | |
c3e5ea6e KS |
2794 | ret = false; |
2795 | goto pte_unlock; | |
2796 | } | |
2797 | ||
5d2a2dbb | 2798 | /* |
985ba004 | 2799 | * The same page can be mapped back since last copy attempt. |
c3e5ea6e | 2800 | * Try to copy again under PTL. |
5d2a2dbb | 2801 | */ |
c3e5ea6e KS |
2802 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { |
2803 | /* | |
2804 | * Give a warn in case there can be some obscure | |
2805 | * use-case | |
2806 | */ | |
2807 | warn: | |
2808 | WARN_ON_ONCE(1); | |
2809 | clear_page(kaddr); | |
2810 | } | |
83d116c5 JH |
2811 | } |
2812 | ||
2813 | ret = true; | |
2814 | ||
2815 | pte_unlock: | |
c3e5ea6e | 2816 | if (locked) |
83d116c5 JH |
2817 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
2818 | kunmap_atomic(kaddr); | |
2819 | flush_dcache_page(dst); | |
2820 | ||
2821 | return ret; | |
6aab341e LT |
2822 | } |
2823 | ||
c20cd45e MH |
2824 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
2825 | { | |
2826 | struct file *vm_file = vma->vm_file; | |
2827 | ||
2828 | if (vm_file) | |
2829 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
2830 | ||
2831 | /* | |
2832 | * Special mappings (e.g. VDSO) do not have any file so fake | |
2833 | * a default GFP_KERNEL for them. | |
2834 | */ | |
2835 | return GFP_KERNEL; | |
2836 | } | |
2837 | ||
fb09a464 KS |
2838 | /* |
2839 | * Notify the address space that the page is about to become writable so that | |
2840 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2841 | * | |
2842 | * We do this without the lock held, so that it can sleep if it needs to. | |
2843 | */ | |
2b740303 | 2844 | static vm_fault_t do_page_mkwrite(struct vm_fault *vmf) |
fb09a464 | 2845 | { |
2b740303 | 2846 | vm_fault_t ret; |
38b8cb7f JK |
2847 | struct page *page = vmf->page; |
2848 | unsigned int old_flags = vmf->flags; | |
fb09a464 | 2849 | |
38b8cb7f | 2850 | vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
fb09a464 | 2851 | |
dc617f29 DW |
2852 | if (vmf->vma->vm_file && |
2853 | IS_SWAPFILE(vmf->vma->vm_file->f_mapping->host)) | |
2854 | return VM_FAULT_SIGBUS; | |
2855 | ||
11bac800 | 2856 | ret = vmf->vma->vm_ops->page_mkwrite(vmf); |
38b8cb7f JK |
2857 | /* Restore original flags so that caller is not surprised */ |
2858 | vmf->flags = old_flags; | |
fb09a464 KS |
2859 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2860 | return ret; | |
2861 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2862 | lock_page(page); | |
2863 | if (!page->mapping) { | |
2864 | unlock_page(page); | |
2865 | return 0; /* retry */ | |
2866 | } | |
2867 | ret |= VM_FAULT_LOCKED; | |
2868 | } else | |
2869 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2870 | return ret; | |
2871 | } | |
2872 | ||
97ba0c2b JK |
2873 | /* |
2874 | * Handle dirtying of a page in shared file mapping on a write fault. | |
2875 | * | |
2876 | * The function expects the page to be locked and unlocks it. | |
2877 | */ | |
89b15332 | 2878 | static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf) |
97ba0c2b | 2879 | { |
89b15332 | 2880 | struct vm_area_struct *vma = vmf->vma; |
97ba0c2b | 2881 | struct address_space *mapping; |
89b15332 | 2882 | struct page *page = vmf->page; |
97ba0c2b JK |
2883 | bool dirtied; |
2884 | bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; | |
2885 | ||
2886 | dirtied = set_page_dirty(page); | |
2887 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2888 | /* | |
2889 | * Take a local copy of the address_space - page.mapping may be zeroed | |
2890 | * by truncate after unlock_page(). The address_space itself remains | |
2891 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
2892 | * release semantics to prevent the compiler from undoing this copying. | |
2893 | */ | |
2894 | mapping = page_rmapping(page); | |
2895 | unlock_page(page); | |
2896 | ||
89b15332 JW |
2897 | if (!page_mkwrite) |
2898 | file_update_time(vma->vm_file); | |
2899 | ||
2900 | /* | |
2901 | * Throttle page dirtying rate down to writeback speed. | |
2902 | * | |
2903 | * mapping may be NULL here because some device drivers do not | |
2904 | * set page.mapping but still dirty their pages | |
2905 | * | |
c1e8d7c6 | 2906 | * Drop the mmap_lock before waiting on IO, if we can. The file |
89b15332 JW |
2907 | * is pinning the mapping, as per above. |
2908 | */ | |
97ba0c2b | 2909 | if ((dirtied || page_mkwrite) && mapping) { |
89b15332 JW |
2910 | struct file *fpin; |
2911 | ||
2912 | fpin = maybe_unlock_mmap_for_io(vmf, NULL); | |
97ba0c2b | 2913 | balance_dirty_pages_ratelimited(mapping); |
89b15332 JW |
2914 | if (fpin) { |
2915 | fput(fpin); | |
2916 | return VM_FAULT_RETRY; | |
2917 | } | |
97ba0c2b JK |
2918 | } |
2919 | ||
89b15332 | 2920 | return 0; |
97ba0c2b JK |
2921 | } |
2922 | ||
4e047f89 SR |
2923 | /* |
2924 | * Handle write page faults for pages that can be reused in the current vma | |
2925 | * | |
2926 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
2927 | * or due to us being the last reference standing to the page. In either | |
2928 | * case, all we need to do here is to mark the page as writable and update | |
2929 | * any related book-keeping. | |
2930 | */ | |
997dd98d | 2931 | static inline void wp_page_reuse(struct vm_fault *vmf) |
82b0f8c3 | 2932 | __releases(vmf->ptl) |
4e047f89 | 2933 | { |
82b0f8c3 | 2934 | struct vm_area_struct *vma = vmf->vma; |
a41b70d6 | 2935 | struct page *page = vmf->page; |
4e047f89 SR |
2936 | pte_t entry; |
2937 | /* | |
2938 | * Clear the pages cpupid information as the existing | |
2939 | * information potentially belongs to a now completely | |
2940 | * unrelated process. | |
2941 | */ | |
2942 | if (page) | |
2943 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2944 | ||
2994302b JK |
2945 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2946 | entry = pte_mkyoung(vmf->orig_pte); | |
4e047f89 | 2947 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
82b0f8c3 JK |
2948 | if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) |
2949 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2950 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
798a6b87 | 2951 | count_vm_event(PGREUSE); |
4e047f89 SR |
2952 | } |
2953 | ||
2f38ab2c SR |
2954 | /* |
2955 | * Handle the case of a page which we actually need to copy to a new page. | |
2956 | * | |
c1e8d7c6 | 2957 | * Called with mmap_lock locked and the old page referenced, but |
2f38ab2c SR |
2958 | * without the ptl held. |
2959 | * | |
2960 | * High level logic flow: | |
2961 | * | |
2962 | * - Allocate a page, copy the content of the old page to the new one. | |
2963 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2964 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2965 | * - If the pte is still the way we remember it, update the page table and all | |
2966 | * relevant references. This includes dropping the reference the page-table | |
2967 | * held to the old page, as well as updating the rmap. | |
2968 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2969 | */ | |
2b740303 | 2970 | static vm_fault_t wp_page_copy(struct vm_fault *vmf) |
2f38ab2c | 2971 | { |
82b0f8c3 | 2972 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2973 | struct mm_struct *mm = vma->vm_mm; |
a41b70d6 | 2974 | struct page *old_page = vmf->page; |
2f38ab2c | 2975 | struct page *new_page = NULL; |
2f38ab2c SR |
2976 | pte_t entry; |
2977 | int page_copied = 0; | |
ac46d4f3 | 2978 | struct mmu_notifier_range range; |
2f38ab2c SR |
2979 | |
2980 | if (unlikely(anon_vma_prepare(vma))) | |
2981 | goto oom; | |
2982 | ||
2994302b | 2983 | if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { |
82b0f8c3 JK |
2984 | new_page = alloc_zeroed_user_highpage_movable(vma, |
2985 | vmf->address); | |
2f38ab2c SR |
2986 | if (!new_page) |
2987 | goto oom; | |
2988 | } else { | |
bae473a4 | 2989 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
82b0f8c3 | 2990 | vmf->address); |
2f38ab2c SR |
2991 | if (!new_page) |
2992 | goto oom; | |
83d116c5 JH |
2993 | |
2994 | if (!cow_user_page(new_page, old_page, vmf)) { | |
2995 | /* | |
2996 | * COW failed, if the fault was solved by other, | |
2997 | * it's fine. If not, userspace would re-fault on | |
2998 | * the same address and we will handle the fault | |
2999 | * from the second attempt. | |
3000 | */ | |
3001 | put_page(new_page); | |
3002 | if (old_page) | |
3003 | put_page(old_page); | |
3004 | return 0; | |
3005 | } | |
2f38ab2c | 3006 | } |
2f38ab2c | 3007 | |
8f425e4e | 3008 | if (mem_cgroup_charge(page_folio(new_page), mm, GFP_KERNEL)) |
2f38ab2c | 3009 | goto oom_free_new; |
9d82c694 | 3010 | cgroup_throttle_swaprate(new_page, GFP_KERNEL); |
2f38ab2c | 3011 | |
eb3c24f3 MG |
3012 | __SetPageUptodate(new_page); |
3013 | ||
7269f999 | 3014 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, |
6f4f13e8 | 3015 | vmf->address & PAGE_MASK, |
ac46d4f3 JG |
3016 | (vmf->address & PAGE_MASK) + PAGE_SIZE); |
3017 | mmu_notifier_invalidate_range_start(&range); | |
2f38ab2c SR |
3018 | |
3019 | /* | |
3020 | * Re-check the pte - we dropped the lock | |
3021 | */ | |
82b0f8c3 | 3022 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); |
2994302b | 3023 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2f38ab2c SR |
3024 | if (old_page) { |
3025 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
3026 | dec_mm_counter_fast(mm, |
3027 | mm_counter_file(old_page)); | |
2f38ab2c SR |
3028 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
3029 | } | |
3030 | } else { | |
3031 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
3032 | } | |
2994302b | 3033 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2f38ab2c | 3034 | entry = mk_pte(new_page, vma->vm_page_prot); |
50c25ee9 | 3035 | entry = pte_sw_mkyoung(entry); |
2f38ab2c | 3036 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
111fe718 | 3037 | |
2f38ab2c SR |
3038 | /* |
3039 | * Clear the pte entry and flush it first, before updating the | |
111fe718 NP |
3040 | * pte with the new entry, to keep TLBs on different CPUs in |
3041 | * sync. This code used to set the new PTE then flush TLBs, but | |
3042 | * that left a window where the new PTE could be loaded into | |
3043 | * some TLBs while the old PTE remains in others. | |
2f38ab2c | 3044 | */ |
82b0f8c3 JK |
3045 | ptep_clear_flush_notify(vma, vmf->address, vmf->pte); |
3046 | page_add_new_anon_rmap(new_page, vma, vmf->address, false); | |
b518154e | 3047 | lru_cache_add_inactive_or_unevictable(new_page, vma); |
2f38ab2c SR |
3048 | /* |
3049 | * We call the notify macro here because, when using secondary | |
3050 | * mmu page tables (such as kvm shadow page tables), we want the | |
3051 | * new page to be mapped directly into the secondary page table. | |
3052 | */ | |
82b0f8c3 JK |
3053 | set_pte_at_notify(mm, vmf->address, vmf->pte, entry); |
3054 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2f38ab2c SR |
3055 | if (old_page) { |
3056 | /* | |
3057 | * Only after switching the pte to the new page may | |
3058 | * we remove the mapcount here. Otherwise another | |
3059 | * process may come and find the rmap count decremented | |
3060 | * before the pte is switched to the new page, and | |
3061 | * "reuse" the old page writing into it while our pte | |
3062 | * here still points into it and can be read by other | |
3063 | * threads. | |
3064 | * | |
3065 | * The critical issue is to order this | |
3066 | * page_remove_rmap with the ptp_clear_flush above. | |
3067 | * Those stores are ordered by (if nothing else,) | |
3068 | * the barrier present in the atomic_add_negative | |
3069 | * in page_remove_rmap. | |
3070 | * | |
3071 | * Then the TLB flush in ptep_clear_flush ensures that | |
3072 | * no process can access the old page before the | |
3073 | * decremented mapcount is visible. And the old page | |
3074 | * cannot be reused until after the decremented | |
3075 | * mapcount is visible. So transitively, TLBs to | |
3076 | * old page will be flushed before it can be reused. | |
3077 | */ | |
d281ee61 | 3078 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
3079 | } |
3080 | ||
3081 | /* Free the old page.. */ | |
3082 | new_page = old_page; | |
3083 | page_copied = 1; | |
3084 | } else { | |
7df67697 | 3085 | update_mmu_tlb(vma, vmf->address, vmf->pte); |
2f38ab2c SR |
3086 | } |
3087 | ||
3088 | if (new_page) | |
09cbfeaf | 3089 | put_page(new_page); |
2f38ab2c | 3090 | |
82b0f8c3 | 3091 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4645b9fe JG |
3092 | /* |
3093 | * No need to double call mmu_notifier->invalidate_range() callback as | |
3094 | * the above ptep_clear_flush_notify() did already call it. | |
3095 | */ | |
ac46d4f3 | 3096 | mmu_notifier_invalidate_range_only_end(&range); |
2f38ab2c SR |
3097 | if (old_page) { |
3098 | /* | |
3099 | * Don't let another task, with possibly unlocked vma, | |
3100 | * keep the mlocked page. | |
3101 | */ | |
3102 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
3103 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
3104 | if (PageMlocked(old_page)) |
3105 | munlock_vma_page(old_page); | |
2f38ab2c SR |
3106 | unlock_page(old_page); |
3107 | } | |
f4c4a3f4 HY |
3108 | if (page_copied) |
3109 | free_swap_cache(old_page); | |
09cbfeaf | 3110 | put_page(old_page); |
2f38ab2c SR |
3111 | } |
3112 | return page_copied ? VM_FAULT_WRITE : 0; | |
3113 | oom_free_new: | |
09cbfeaf | 3114 | put_page(new_page); |
2f38ab2c SR |
3115 | oom: |
3116 | if (old_page) | |
09cbfeaf | 3117 | put_page(old_page); |
2f38ab2c SR |
3118 | return VM_FAULT_OOM; |
3119 | } | |
3120 | ||
66a6197c JK |
3121 | /** |
3122 | * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE | |
3123 | * writeable once the page is prepared | |
3124 | * | |
3125 | * @vmf: structure describing the fault | |
3126 | * | |
3127 | * This function handles all that is needed to finish a write page fault in a | |
3128 | * shared mapping due to PTE being read-only once the mapped page is prepared. | |
a862f68a | 3129 | * It handles locking of PTE and modifying it. |
66a6197c JK |
3130 | * |
3131 | * The function expects the page to be locked or other protection against | |
3132 | * concurrent faults / writeback (such as DAX radix tree locks). | |
a862f68a | 3133 | * |
2797e79f | 3134 | * Return: %0 on success, %VM_FAULT_NOPAGE when PTE got changed before |
a862f68a | 3135 | * we acquired PTE lock. |
66a6197c | 3136 | */ |
2b740303 | 3137 | vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf) |
66a6197c JK |
3138 | { |
3139 | WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); | |
3140 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, | |
3141 | &vmf->ptl); | |
3142 | /* | |
3143 | * We might have raced with another page fault while we released the | |
3144 | * pte_offset_map_lock. | |
3145 | */ | |
3146 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { | |
7df67697 | 3147 | update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); |
66a6197c | 3148 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a19e2553 | 3149 | return VM_FAULT_NOPAGE; |
66a6197c JK |
3150 | } |
3151 | wp_page_reuse(vmf); | |
a19e2553 | 3152 | return 0; |
66a6197c JK |
3153 | } |
3154 | ||
dd906184 BH |
3155 | /* |
3156 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
3157 | * mapping | |
3158 | */ | |
2b740303 | 3159 | static vm_fault_t wp_pfn_shared(struct vm_fault *vmf) |
dd906184 | 3160 | { |
82b0f8c3 | 3161 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 3162 | |
dd906184 | 3163 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
2b740303 | 3164 | vm_fault_t ret; |
dd906184 | 3165 | |
82b0f8c3 | 3166 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
fe82221f | 3167 | vmf->flags |= FAULT_FLAG_MKWRITE; |
11bac800 | 3168 | ret = vma->vm_ops->pfn_mkwrite(vmf); |
2f89dc12 | 3169 | if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) |
dd906184 | 3170 | return ret; |
66a6197c | 3171 | return finish_mkwrite_fault(vmf); |
dd906184 | 3172 | } |
997dd98d JK |
3173 | wp_page_reuse(vmf); |
3174 | return VM_FAULT_WRITE; | |
dd906184 BH |
3175 | } |
3176 | ||
2b740303 | 3177 | static vm_fault_t wp_page_shared(struct vm_fault *vmf) |
82b0f8c3 | 3178 | __releases(vmf->ptl) |
93e478d4 | 3179 | { |
82b0f8c3 | 3180 | struct vm_area_struct *vma = vmf->vma; |
89b15332 | 3181 | vm_fault_t ret = VM_FAULT_WRITE; |
93e478d4 | 3182 | |
a41b70d6 | 3183 | get_page(vmf->page); |
93e478d4 | 3184 | |
93e478d4 | 3185 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2b740303 | 3186 | vm_fault_t tmp; |
93e478d4 | 3187 | |
82b0f8c3 | 3188 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
38b8cb7f | 3189 | tmp = do_page_mkwrite(vmf); |
93e478d4 SR |
3190 | if (unlikely(!tmp || (tmp & |
3191 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
a41b70d6 | 3192 | put_page(vmf->page); |
93e478d4 SR |
3193 | return tmp; |
3194 | } | |
66a6197c | 3195 | tmp = finish_mkwrite_fault(vmf); |
a19e2553 | 3196 | if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
a41b70d6 | 3197 | unlock_page(vmf->page); |
a41b70d6 | 3198 | put_page(vmf->page); |
66a6197c | 3199 | return tmp; |
93e478d4 | 3200 | } |
66a6197c JK |
3201 | } else { |
3202 | wp_page_reuse(vmf); | |
997dd98d | 3203 | lock_page(vmf->page); |
93e478d4 | 3204 | } |
89b15332 | 3205 | ret |= fault_dirty_shared_page(vmf); |
997dd98d | 3206 | put_page(vmf->page); |
93e478d4 | 3207 | |
89b15332 | 3208 | return ret; |
93e478d4 SR |
3209 | } |
3210 | ||
1da177e4 LT |
3211 | /* |
3212 | * This routine handles present pages, when users try to write | |
3213 | * to a shared page. It is done by copying the page to a new address | |
3214 | * and decrementing the shared-page counter for the old page. | |
3215 | * | |
1da177e4 LT |
3216 | * Note that this routine assumes that the protection checks have been |
3217 | * done by the caller (the low-level page fault routine in most cases). | |
3218 | * Thus we can safely just mark it writable once we've done any necessary | |
3219 | * COW. | |
3220 | * | |
3221 | * We also mark the page dirty at this point even though the page will | |
3222 | * change only once the write actually happens. This avoids a few races, | |
3223 | * and potentially makes it more efficient. | |
3224 | * | |
c1e8d7c6 | 3225 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
8f4e2101 | 3226 | * but allow concurrent faults), with pte both mapped and locked. |
c1e8d7c6 | 3227 | * We return with mmap_lock still held, but pte unmapped and unlocked. |
1da177e4 | 3228 | */ |
2b740303 | 3229 | static vm_fault_t do_wp_page(struct vm_fault *vmf) |
82b0f8c3 | 3230 | __releases(vmf->ptl) |
1da177e4 | 3231 | { |
82b0f8c3 | 3232 | struct vm_area_struct *vma = vmf->vma; |
1da177e4 | 3233 | |
292924b2 | 3234 | if (userfaultfd_pte_wp(vma, *vmf->pte)) { |
529b930b AA |
3235 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3236 | return handle_userfault(vmf, VM_UFFD_WP); | |
3237 | } | |
3238 | ||
6ce64428 NA |
3239 | /* |
3240 | * Userfaultfd write-protect can defer flushes. Ensure the TLB | |
3241 | * is flushed in this case before copying. | |
3242 | */ | |
3243 | if (unlikely(userfaultfd_wp(vmf->vma) && | |
3244 | mm_tlb_flush_pending(vmf->vma->vm_mm))) | |
3245 | flush_tlb_page(vmf->vma, vmf->address); | |
3246 | ||
a41b70d6 JK |
3247 | vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); |
3248 | if (!vmf->page) { | |
251b97f5 | 3249 | /* |
64e45507 PF |
3250 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
3251 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
3252 | * |
3253 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 3254 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
3255 | */ |
3256 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
3257 | (VM_WRITE|VM_SHARED)) | |
2994302b | 3258 | return wp_pfn_shared(vmf); |
2f38ab2c | 3259 | |
82b0f8c3 | 3260 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 3261 | return wp_page_copy(vmf); |
251b97f5 | 3262 | } |
1da177e4 | 3263 | |
d08b3851 | 3264 | /* |
ee6a6457 PZ |
3265 | * Take out anonymous pages first, anonymous shared vmas are |
3266 | * not dirty accountable. | |
d08b3851 | 3267 | */ |
52d1e606 | 3268 | if (PageAnon(vmf->page)) { |
09854ba9 LT |
3269 | struct page *page = vmf->page; |
3270 | ||
3271 | /* PageKsm() doesn't necessarily raise the page refcount */ | |
3272 | if (PageKsm(page) || page_count(page) != 1) | |
3273 | goto copy; | |
3274 | if (!trylock_page(page)) | |
3275 | goto copy; | |
3276 | if (PageKsm(page) || page_mapcount(page) != 1 || page_count(page) != 1) { | |
3277 | unlock_page(page); | |
52d1e606 | 3278 | goto copy; |
b009c024 | 3279 | } |
09854ba9 LT |
3280 | /* |
3281 | * Ok, we've got the only map reference, and the only | |
3282 | * page count reference, and the page is locked, | |
3283 | * it's dark out, and we're wearing sunglasses. Hit it. | |
3284 | */ | |
09854ba9 | 3285 | unlock_page(page); |
be068f29 | 3286 | wp_page_reuse(vmf); |
09854ba9 | 3287 | return VM_FAULT_WRITE; |
ee6a6457 | 3288 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 3289 | (VM_WRITE|VM_SHARED))) { |
a41b70d6 | 3290 | return wp_page_shared(vmf); |
1da177e4 | 3291 | } |
52d1e606 | 3292 | copy: |
1da177e4 LT |
3293 | /* |
3294 | * Ok, we need to copy. Oh, well.. | |
3295 | */ | |
a41b70d6 | 3296 | get_page(vmf->page); |
28766805 | 3297 | |
82b0f8c3 | 3298 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 3299 | return wp_page_copy(vmf); |
1da177e4 LT |
3300 | } |
3301 | ||
97a89413 | 3302 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
3303 | unsigned long start_addr, unsigned long end_addr, |
3304 | struct zap_details *details) | |
3305 | { | |
f5cc4eef | 3306 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
3307 | } |
3308 | ||
f808c13f | 3309 | static inline void unmap_mapping_range_tree(struct rb_root_cached *root, |
232a6a1c PX |
3310 | pgoff_t first_index, |
3311 | pgoff_t last_index, | |
1da177e4 LT |
3312 | struct zap_details *details) |
3313 | { | |
3314 | struct vm_area_struct *vma; | |
1da177e4 LT |
3315 | pgoff_t vba, vea, zba, zea; |
3316 | ||
232a6a1c | 3317 | vma_interval_tree_foreach(vma, root, first_index, last_index) { |
1da177e4 | 3318 | vba = vma->vm_pgoff; |
d6e93217 | 3319 | vea = vba + vma_pages(vma) - 1; |
232a6a1c | 3320 | zba = first_index; |
1da177e4 LT |
3321 | if (zba < vba) |
3322 | zba = vba; | |
232a6a1c | 3323 | zea = last_index; |
1da177e4 LT |
3324 | if (zea > vea) |
3325 | zea = vea; | |
3326 | ||
97a89413 | 3327 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
3328 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
3329 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 3330 | details); |
1da177e4 LT |
3331 | } |
3332 | } | |
3333 | ||
22061a1f HD |
3334 | /** |
3335 | * unmap_mapping_page() - Unmap single page from processes. | |
3336 | * @page: The locked page to be unmapped. | |
3337 | * | |
3338 | * Unmap this page from any userspace process which still has it mmaped. | |
3339 | * Typically, for efficiency, the range of nearby pages has already been | |
3340 | * unmapped by unmap_mapping_pages() or unmap_mapping_range(). But once | |
3341 | * truncation or invalidation holds the lock on a page, it may find that | |
3342 | * the page has been remapped again: and then uses unmap_mapping_page() | |
3343 | * to unmap it finally. | |
3344 | */ | |
3345 | void unmap_mapping_page(struct page *page) | |
3346 | { | |
3347 | struct address_space *mapping = page->mapping; | |
3348 | struct zap_details details = { }; | |
232a6a1c PX |
3349 | pgoff_t first_index; |
3350 | pgoff_t last_index; | |
22061a1f HD |
3351 | |
3352 | VM_BUG_ON(!PageLocked(page)); | |
3353 | VM_BUG_ON(PageTail(page)); | |
3354 | ||
232a6a1c PX |
3355 | first_index = page->index; |
3356 | last_index = page->index + thp_nr_pages(page) - 1; | |
3357 | ||
91b61ef3 | 3358 | details.zap_mapping = mapping; |
22061a1f HD |
3359 | details.single_page = page; |
3360 | ||
3361 | i_mmap_lock_write(mapping); | |
3362 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) | |
232a6a1c PX |
3363 | unmap_mapping_range_tree(&mapping->i_mmap, first_index, |
3364 | last_index, &details); | |
22061a1f HD |
3365 | i_mmap_unlock_write(mapping); |
3366 | } | |
3367 | ||
977fbdcd MW |
3368 | /** |
3369 | * unmap_mapping_pages() - Unmap pages from processes. | |
3370 | * @mapping: The address space containing pages to be unmapped. | |
3371 | * @start: Index of first page to be unmapped. | |
3372 | * @nr: Number of pages to be unmapped. 0 to unmap to end of file. | |
3373 | * @even_cows: Whether to unmap even private COWed pages. | |
3374 | * | |
3375 | * Unmap the pages in this address space from any userspace process which | |
3376 | * has them mmaped. Generally, you want to remove COWed pages as well when | |
3377 | * a file is being truncated, but not when invalidating pages from the page | |
3378 | * cache. | |
3379 | */ | |
3380 | void unmap_mapping_pages(struct address_space *mapping, pgoff_t start, | |
3381 | pgoff_t nr, bool even_cows) | |
3382 | { | |
3383 | struct zap_details details = { }; | |
232a6a1c PX |
3384 | pgoff_t first_index = start; |
3385 | pgoff_t last_index = start + nr - 1; | |
977fbdcd | 3386 | |
91b61ef3 | 3387 | details.zap_mapping = even_cows ? NULL : mapping; |
232a6a1c PX |
3388 | if (last_index < first_index) |
3389 | last_index = ULONG_MAX; | |
977fbdcd MW |
3390 | |
3391 | i_mmap_lock_write(mapping); | |
3392 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) | |
232a6a1c PX |
3393 | unmap_mapping_range_tree(&mapping->i_mmap, first_index, |
3394 | last_index, &details); | |
977fbdcd MW |
3395 | i_mmap_unlock_write(mapping); |
3396 | } | |
6e0e99d5 | 3397 | EXPORT_SYMBOL_GPL(unmap_mapping_pages); |
977fbdcd | 3398 | |
1da177e4 | 3399 | /** |
8a5f14a2 | 3400 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
977fbdcd | 3401 | * address_space corresponding to the specified byte range in the underlying |
8a5f14a2 KS |
3402 | * file. |
3403 | * | |
3d41088f | 3404 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
3405 | * @holebegin: byte in first page to unmap, relative to the start of |
3406 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 3407 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
3408 | * must keep the partial page. In contrast, we must get rid of |
3409 | * partial pages. | |
3410 | * @holelen: size of prospective hole in bytes. This will be rounded | |
3411 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
3412 | * end of the file. | |
3413 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
3414 | * but 0 when invalidating pagecache, don't throw away private data. | |
3415 | */ | |
3416 | void unmap_mapping_range(struct address_space *mapping, | |
3417 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
3418 | { | |
1da177e4 LT |
3419 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
3420 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
3421 | ||
3422 | /* Check for overflow. */ | |
3423 | if (sizeof(holelen) > sizeof(hlen)) { | |
3424 | long long holeend = | |
3425 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
3426 | if (holeend & ~(long long)ULONG_MAX) | |
3427 | hlen = ULONG_MAX - hba + 1; | |
3428 | } | |
3429 | ||
977fbdcd | 3430 | unmap_mapping_pages(mapping, hba, hlen, even_cows); |
1da177e4 LT |
3431 | } |
3432 | EXPORT_SYMBOL(unmap_mapping_range); | |
3433 | ||
b756a3b5 AP |
3434 | /* |
3435 | * Restore a potential device exclusive pte to a working pte entry | |
3436 | */ | |
3437 | static vm_fault_t remove_device_exclusive_entry(struct vm_fault *vmf) | |
3438 | { | |
3439 | struct page *page = vmf->page; | |
3440 | struct vm_area_struct *vma = vmf->vma; | |
3441 | struct mmu_notifier_range range; | |
3442 | ||
3443 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) | |
3444 | return VM_FAULT_RETRY; | |
3445 | mmu_notifier_range_init_owner(&range, MMU_NOTIFY_EXCLUSIVE, 0, vma, | |
3446 | vma->vm_mm, vmf->address & PAGE_MASK, | |
3447 | (vmf->address & PAGE_MASK) + PAGE_SIZE, NULL); | |
3448 | mmu_notifier_invalidate_range_start(&range); | |
3449 | ||
3450 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, | |
3451 | &vmf->ptl); | |
3452 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) | |
3453 | restore_exclusive_pte(vma, page, vmf->address, vmf->pte); | |
3454 | ||
3455 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3456 | unlock_page(page); | |
3457 | ||
3458 | mmu_notifier_invalidate_range_end(&range); | |
3459 | return 0; | |
3460 | } | |
3461 | ||
1da177e4 | 3462 | /* |
c1e8d7c6 | 3463 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
8f4e2101 | 3464 | * but allow concurrent faults), and pte mapped but not yet locked. |
9a95f3cf PC |
3465 | * We return with pte unmapped and unlocked. |
3466 | * | |
c1e8d7c6 | 3467 | * We return with the mmap_lock locked or unlocked in the same cases |
9a95f3cf | 3468 | * as does filemap_fault(). |
1da177e4 | 3469 | */ |
2b740303 | 3470 | vm_fault_t do_swap_page(struct vm_fault *vmf) |
1da177e4 | 3471 | { |
82b0f8c3 | 3472 | struct vm_area_struct *vma = vmf->vma; |
eaf649eb | 3473 | struct page *page = NULL, *swapcache; |
2799e775 | 3474 | struct swap_info_struct *si = NULL; |
65500d23 | 3475 | swp_entry_t entry; |
1da177e4 | 3476 | pte_t pte; |
d065bd81 | 3477 | int locked; |
ad8c2ee8 | 3478 | int exclusive = 0; |
2b740303 | 3479 | vm_fault_t ret = 0; |
aae466b0 | 3480 | void *shadow = NULL; |
1da177e4 | 3481 | |
2ca99358 | 3482 | if (!pte_unmap_same(vmf)) |
8f4e2101 | 3483 | goto out; |
65500d23 | 3484 | |
2994302b | 3485 | entry = pte_to_swp_entry(vmf->orig_pte); |
d1737fdb AK |
3486 | if (unlikely(non_swap_entry(entry))) { |
3487 | if (is_migration_entry(entry)) { | |
82b0f8c3 JK |
3488 | migration_entry_wait(vma->vm_mm, vmf->pmd, |
3489 | vmf->address); | |
b756a3b5 AP |
3490 | } else if (is_device_exclusive_entry(entry)) { |
3491 | vmf->page = pfn_swap_entry_to_page(entry); | |
3492 | ret = remove_device_exclusive_entry(vmf); | |
5042db43 | 3493 | } else if (is_device_private_entry(entry)) { |
af5cdaf8 | 3494 | vmf->page = pfn_swap_entry_to_page(entry); |
897e6365 | 3495 | ret = vmf->page->pgmap->ops->migrate_to_ram(vmf); |
d1737fdb AK |
3496 | } else if (is_hwpoison_entry(entry)) { |
3497 | ret = VM_FAULT_HWPOISON; | |
3498 | } else { | |
2994302b | 3499 | print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); |
d99be1a8 | 3500 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 3501 | } |
0697212a CL |
3502 | goto out; |
3503 | } | |
0bcac06f | 3504 | |
2799e775 ML |
3505 | /* Prevent swapoff from happening to us. */ |
3506 | si = get_swap_device(entry); | |
3507 | if (unlikely(!si)) | |
3508 | goto out; | |
0bcac06f | 3509 | |
3d1c7fd9 | 3510 | delayacct_set_flag(current, DELAYACCT_PF_SWAPIN); |
eaf649eb MK |
3511 | page = lookup_swap_cache(entry, vma, vmf->address); |
3512 | swapcache = page; | |
f8020772 | 3513 | |
1da177e4 | 3514 | if (!page) { |
a449bf58 QC |
3515 | if (data_race(si->flags & SWP_SYNCHRONOUS_IO) && |
3516 | __swap_count(entry) == 1) { | |
0bcac06f | 3517 | /* skip swapcache */ |
e9e9b7ec MK |
3518 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
3519 | vmf->address); | |
0bcac06f MK |
3520 | if (page) { |
3521 | __SetPageLocked(page); | |
3522 | __SetPageSwapBacked(page); | |
4c6355b2 | 3523 | |
0add0c77 SB |
3524 | if (mem_cgroup_swapin_charge_page(page, |
3525 | vma->vm_mm, GFP_KERNEL, entry)) { | |
545b1b07 | 3526 | ret = VM_FAULT_OOM; |
4c6355b2 | 3527 | goto out_page; |
545b1b07 | 3528 | } |
0add0c77 | 3529 | mem_cgroup_swapin_uncharge_swap(entry); |
4c6355b2 | 3530 | |
aae466b0 JK |
3531 | shadow = get_shadow_from_swap_cache(entry); |
3532 | if (shadow) | |
0995d7e5 MWO |
3533 | workingset_refault(page_folio(page), |
3534 | shadow); | |
0076f029 | 3535 | |
6058eaec | 3536 | lru_cache_add(page); |
0add0c77 SB |
3537 | |
3538 | /* To provide entry to swap_readpage() */ | |
3539 | set_page_private(page, entry.val); | |
0bcac06f | 3540 | swap_readpage(page, true); |
0add0c77 | 3541 | set_page_private(page, 0); |
0bcac06f | 3542 | } |
aa8d22a1 | 3543 | } else { |
e9e9b7ec MK |
3544 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
3545 | vmf); | |
aa8d22a1 | 3546 | swapcache = page; |
0bcac06f MK |
3547 | } |
3548 | ||
1da177e4 LT |
3549 | if (!page) { |
3550 | /* | |
8f4e2101 HD |
3551 | * Back out if somebody else faulted in this pte |
3552 | * while we released the pte lock. | |
1da177e4 | 3553 | */ |
82b0f8c3 JK |
3554 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
3555 | vmf->address, &vmf->ptl); | |
2994302b | 3556 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) |
1da177e4 | 3557 | ret = VM_FAULT_OOM; |
3d1c7fd9 | 3558 | delayacct_clear_flag(current, DELAYACCT_PF_SWAPIN); |
65500d23 | 3559 | goto unlock; |
1da177e4 LT |
3560 | } |
3561 | ||
3562 | /* Had to read the page from swap area: Major fault */ | |
3563 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 3564 | count_vm_event(PGMAJFAULT); |
2262185c | 3565 | count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); |
d1737fdb | 3566 | } else if (PageHWPoison(page)) { |
71f72525 WF |
3567 | /* |
3568 | * hwpoisoned dirty swapcache pages are kept for killing | |
3569 | * owner processes (which may be unknown at hwpoison time) | |
3570 | */ | |
d1737fdb | 3571 | ret = VM_FAULT_HWPOISON; |
3d1c7fd9 | 3572 | delayacct_clear_flag(current, DELAYACCT_PF_SWAPIN); |
4779cb31 | 3573 | goto out_release; |
1da177e4 LT |
3574 | } |
3575 | ||
82b0f8c3 | 3576 | locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags); |
e709ffd6 | 3577 | |
3d1c7fd9 | 3578 | delayacct_clear_flag(current, DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
3579 | if (!locked) { |
3580 | ret |= VM_FAULT_RETRY; | |
3581 | goto out_release; | |
3582 | } | |
073e587e | 3583 | |
4969c119 | 3584 | /* |
31c4a3d3 HD |
3585 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
3586 | * release the swapcache from under us. The page pin, and pte_same | |
3587 | * test below, are not enough to exclude that. Even if it is still | |
3588 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 3589 | */ |
0bcac06f MK |
3590 | if (unlikely((!PageSwapCache(page) || |
3591 | page_private(page) != entry.val)) && swapcache) | |
4969c119 AA |
3592 | goto out_page; |
3593 | ||
82b0f8c3 | 3594 | page = ksm_might_need_to_copy(page, vma, vmf->address); |
cbf86cfe HD |
3595 | if (unlikely(!page)) { |
3596 | ret = VM_FAULT_OOM; | |
3597 | page = swapcache; | |
cbf86cfe | 3598 | goto out_page; |
5ad64688 HD |
3599 | } |
3600 | ||
9d82c694 | 3601 | cgroup_throttle_swaprate(page, GFP_KERNEL); |
8a9f3ccd | 3602 | |
1da177e4 | 3603 | /* |
8f4e2101 | 3604 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 3605 | */ |
82b0f8c3 JK |
3606 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3607 | &vmf->ptl); | |
2994302b | 3608 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) |
b8107480 | 3609 | goto out_nomap; |
b8107480 KK |
3610 | |
3611 | if (unlikely(!PageUptodate(page))) { | |
3612 | ret = VM_FAULT_SIGBUS; | |
3613 | goto out_nomap; | |
1da177e4 LT |
3614 | } |
3615 | ||
8c7c6e34 KH |
3616 | /* |
3617 | * The page isn't present yet, go ahead with the fault. | |
3618 | * | |
3619 | * Be careful about the sequence of operations here. | |
3620 | * To get its accounting right, reuse_swap_page() must be called | |
3621 | * while the page is counted on swap but not yet in mapcount i.e. | |
3622 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
3623 | * must be called after the swap_free(), or it will never succeed. | |
8c7c6e34 | 3624 | */ |
1da177e4 | 3625 | |
bae473a4 KS |
3626 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
3627 | dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS); | |
1da177e4 | 3628 | pte = mk_pte(page, vma->vm_page_prot); |
82b0f8c3 | 3629 | if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { |
1da177e4 | 3630 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
82b0f8c3 | 3631 | vmf->flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 3632 | ret |= VM_FAULT_WRITE; |
d281ee61 | 3633 | exclusive = RMAP_EXCLUSIVE; |
1da177e4 | 3634 | } |
1da177e4 | 3635 | flush_icache_page(vma, page); |
2994302b | 3636 | if (pte_swp_soft_dirty(vmf->orig_pte)) |
179ef71c | 3637 | pte = pte_mksoft_dirty(pte); |
f45ec5ff PX |
3638 | if (pte_swp_uffd_wp(vmf->orig_pte)) { |
3639 | pte = pte_mkuffd_wp(pte); | |
3640 | pte = pte_wrprotect(pte); | |
3641 | } | |
82b0f8c3 | 3642 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); |
ca827d55 | 3643 | arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte); |
2994302b | 3644 | vmf->orig_pte = pte; |
0bcac06f MK |
3645 | |
3646 | /* ksm created a completely new copy */ | |
3647 | if (unlikely(page != swapcache && swapcache)) { | |
82b0f8c3 | 3648 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
b518154e | 3649 | lru_cache_add_inactive_or_unevictable(page, vma); |
0bcac06f MK |
3650 | } else { |
3651 | do_page_add_anon_rmap(page, vma, vmf->address, exclusive); | |
00501b53 | 3652 | } |
1da177e4 | 3653 | |
c475a8ab | 3654 | swap_free(entry); |
5ccc5aba VD |
3655 | if (mem_cgroup_swap_full(page) || |
3656 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 3657 | try_to_free_swap(page); |
c475a8ab | 3658 | unlock_page(page); |
0bcac06f | 3659 | if (page != swapcache && swapcache) { |
4969c119 AA |
3660 | /* |
3661 | * Hold the lock to avoid the swap entry to be reused | |
3662 | * until we take the PT lock for the pte_same() check | |
3663 | * (to avoid false positives from pte_same). For | |
3664 | * further safety release the lock after the swap_free | |
3665 | * so that the swap count won't change under a | |
3666 | * parallel locked swapcache. | |
3667 | */ | |
3668 | unlock_page(swapcache); | |
09cbfeaf | 3669 | put_page(swapcache); |
4969c119 | 3670 | } |
c475a8ab | 3671 | |
82b0f8c3 | 3672 | if (vmf->flags & FAULT_FLAG_WRITE) { |
2994302b | 3673 | ret |= do_wp_page(vmf); |
61469f1d HD |
3674 | if (ret & VM_FAULT_ERROR) |
3675 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
3676 | goto out; |
3677 | } | |
3678 | ||
3679 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3680 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3681 | unlock: |
82b0f8c3 | 3682 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
1da177e4 | 3683 | out: |
2799e775 ML |
3684 | if (si) |
3685 | put_swap_device(si); | |
1da177e4 | 3686 | return ret; |
b8107480 | 3687 | out_nomap: |
82b0f8c3 | 3688 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bc43f75c | 3689 | out_page: |
b8107480 | 3690 | unlock_page(page); |
4779cb31 | 3691 | out_release: |
09cbfeaf | 3692 | put_page(page); |
0bcac06f | 3693 | if (page != swapcache && swapcache) { |
4969c119 | 3694 | unlock_page(swapcache); |
09cbfeaf | 3695 | put_page(swapcache); |
4969c119 | 3696 | } |
2799e775 ML |
3697 | if (si) |
3698 | put_swap_device(si); | |
65500d23 | 3699 | return ret; |
1da177e4 LT |
3700 | } |
3701 | ||
3702 | /* | |
c1e8d7c6 | 3703 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
8f4e2101 | 3704 | * but allow concurrent faults), and pte mapped but not yet locked. |
c1e8d7c6 | 3705 | * We return with mmap_lock still held, but pte unmapped and unlocked. |
1da177e4 | 3706 | */ |
2b740303 | 3707 | static vm_fault_t do_anonymous_page(struct vm_fault *vmf) |
1da177e4 | 3708 | { |
82b0f8c3 | 3709 | struct vm_area_struct *vma = vmf->vma; |
8f4e2101 | 3710 | struct page *page; |
2b740303 | 3711 | vm_fault_t ret = 0; |
1da177e4 | 3712 | pte_t entry; |
1da177e4 | 3713 | |
6b7339f4 KS |
3714 | /* File mapping without ->vm_ops ? */ |
3715 | if (vma->vm_flags & VM_SHARED) | |
3716 | return VM_FAULT_SIGBUS; | |
3717 | ||
7267ec00 KS |
3718 | /* |
3719 | * Use pte_alloc() instead of pte_alloc_map(). We can't run | |
3720 | * pte_offset_map() on pmds where a huge pmd might be created | |
3721 | * from a different thread. | |
3722 | * | |
3e4e28c5 | 3723 | * pte_alloc_map() is safe to use under mmap_write_lock(mm) or when |
7267ec00 KS |
3724 | * parallel threads are excluded by other means. |
3725 | * | |
3e4e28c5 | 3726 | * Here we only have mmap_read_lock(mm). |
7267ec00 | 3727 | */ |
4cf58924 | 3728 | if (pte_alloc(vma->vm_mm, vmf->pmd)) |
7267ec00 KS |
3729 | return VM_FAULT_OOM; |
3730 | ||
f9ce0be7 | 3731 | /* See comment in handle_pte_fault() */ |
82b0f8c3 | 3732 | if (unlikely(pmd_trans_unstable(vmf->pmd))) |
7267ec00 KS |
3733 | return 0; |
3734 | ||
11ac5524 | 3735 | /* Use the zero-page for reads */ |
82b0f8c3 | 3736 | if (!(vmf->flags & FAULT_FLAG_WRITE) && |
bae473a4 | 3737 | !mm_forbids_zeropage(vma->vm_mm)) { |
82b0f8c3 | 3738 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), |
62eede62 | 3739 | vma->vm_page_prot)); |
82b0f8c3 JK |
3740 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
3741 | vmf->address, &vmf->ptl); | |
7df67697 BM |
3742 | if (!pte_none(*vmf->pte)) { |
3743 | update_mmu_tlb(vma, vmf->address, vmf->pte); | |
a13ea5b7 | 3744 | goto unlock; |
7df67697 | 3745 | } |
6b31d595 MH |
3746 | ret = check_stable_address_space(vma->vm_mm); |
3747 | if (ret) | |
3748 | goto unlock; | |
6b251fc9 AA |
3749 | /* Deliver the page fault to userland, check inside PT lock */ |
3750 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 JK |
3751 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3752 | return handle_userfault(vmf, VM_UFFD_MISSING); | |
6b251fc9 | 3753 | } |
a13ea5b7 HD |
3754 | goto setpte; |
3755 | } | |
3756 | ||
557ed1fa | 3757 | /* Allocate our own private page. */ |
557ed1fa NP |
3758 | if (unlikely(anon_vma_prepare(vma))) |
3759 | goto oom; | |
82b0f8c3 | 3760 | page = alloc_zeroed_user_highpage_movable(vma, vmf->address); |
557ed1fa NP |
3761 | if (!page) |
3762 | goto oom; | |
eb3c24f3 | 3763 | |
8f425e4e | 3764 | if (mem_cgroup_charge(page_folio(page), vma->vm_mm, GFP_KERNEL)) |
eb3c24f3 | 3765 | goto oom_free_page; |
9d82c694 | 3766 | cgroup_throttle_swaprate(page, GFP_KERNEL); |
eb3c24f3 | 3767 | |
52f37629 MK |
3768 | /* |
3769 | * The memory barrier inside __SetPageUptodate makes sure that | |
f4f5329d | 3770 | * preceding stores to the page contents become visible before |
52f37629 MK |
3771 | * the set_pte_at() write. |
3772 | */ | |
0ed361de | 3773 | __SetPageUptodate(page); |
8f4e2101 | 3774 | |
557ed1fa | 3775 | entry = mk_pte(page, vma->vm_page_prot); |
50c25ee9 | 3776 | entry = pte_sw_mkyoung(entry); |
1ac0cb5d HD |
3777 | if (vma->vm_flags & VM_WRITE) |
3778 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 3779 | |
82b0f8c3 JK |
3780 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3781 | &vmf->ptl); | |
7df67697 BM |
3782 | if (!pte_none(*vmf->pte)) { |
3783 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
557ed1fa | 3784 | goto release; |
7df67697 | 3785 | } |
9ba69294 | 3786 | |
6b31d595 MH |
3787 | ret = check_stable_address_space(vma->vm_mm); |
3788 | if (ret) | |
3789 | goto release; | |
3790 | ||
6b251fc9 AA |
3791 | /* Deliver the page fault to userland, check inside PT lock */ |
3792 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 | 3793 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
09cbfeaf | 3794 | put_page(page); |
82b0f8c3 | 3795 | return handle_userfault(vmf, VM_UFFD_MISSING); |
6b251fc9 AA |
3796 | } |
3797 | ||
bae473a4 | 3798 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3799 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
b518154e | 3800 | lru_cache_add_inactive_or_unevictable(page, vma); |
a13ea5b7 | 3801 | setpte: |
82b0f8c3 | 3802 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
1da177e4 LT |
3803 | |
3804 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3805 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3806 | unlock: |
82b0f8c3 | 3807 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
6b31d595 | 3808 | return ret; |
8f4e2101 | 3809 | release: |
09cbfeaf | 3810 | put_page(page); |
8f4e2101 | 3811 | goto unlock; |
8a9f3ccd | 3812 | oom_free_page: |
09cbfeaf | 3813 | put_page(page); |
65500d23 | 3814 | oom: |
1da177e4 LT |
3815 | return VM_FAULT_OOM; |
3816 | } | |
3817 | ||
9a95f3cf | 3818 | /* |
c1e8d7c6 | 3819 | * The mmap_lock must have been held on entry, and may have been |
9a95f3cf PC |
3820 | * released depending on flags and vma->vm_ops->fault() return value. |
3821 | * See filemap_fault() and __lock_page_retry(). | |
3822 | */ | |
2b740303 | 3823 | static vm_fault_t __do_fault(struct vm_fault *vmf) |
7eae74af | 3824 | { |
82b0f8c3 | 3825 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 3826 | vm_fault_t ret; |
7eae74af | 3827 | |
63f3655f MH |
3828 | /* |
3829 | * Preallocate pte before we take page_lock because this might lead to | |
3830 | * deadlocks for memcg reclaim which waits for pages under writeback: | |
3831 | * lock_page(A) | |
3832 | * SetPageWriteback(A) | |
3833 | * unlock_page(A) | |
3834 | * lock_page(B) | |
3835 | * lock_page(B) | |
d383807a | 3836 | * pte_alloc_one |
63f3655f MH |
3837 | * shrink_page_list |
3838 | * wait_on_page_writeback(A) | |
3839 | * SetPageWriteback(B) | |
3840 | * unlock_page(B) | |
3841 | * # flush A, B to clear the writeback | |
3842 | */ | |
3843 | if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) { | |
a7069ee3 | 3844 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); |
63f3655f MH |
3845 | if (!vmf->prealloc_pte) |
3846 | return VM_FAULT_OOM; | |
63f3655f MH |
3847 | } |
3848 | ||
11bac800 | 3849 | ret = vma->vm_ops->fault(vmf); |
3917048d | 3850 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | |
b1aa812b | 3851 | VM_FAULT_DONE_COW))) |
bc2466e4 | 3852 | return ret; |
7eae74af | 3853 | |
667240e0 | 3854 | if (unlikely(PageHWPoison(vmf->page))) { |
7eae74af | 3855 | if (ret & VM_FAULT_LOCKED) |
667240e0 JK |
3856 | unlock_page(vmf->page); |
3857 | put_page(vmf->page); | |
936ca80d | 3858 | vmf->page = NULL; |
7eae74af KS |
3859 | return VM_FAULT_HWPOISON; |
3860 | } | |
3861 | ||
3862 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
667240e0 | 3863 | lock_page(vmf->page); |
7eae74af | 3864 | else |
667240e0 | 3865 | VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); |
7eae74af | 3866 | |
7eae74af KS |
3867 | return ret; |
3868 | } | |
3869 | ||
396bcc52 | 3870 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
82b0f8c3 | 3871 | static void deposit_prealloc_pte(struct vm_fault *vmf) |
953c66c2 | 3872 | { |
82b0f8c3 | 3873 | struct vm_area_struct *vma = vmf->vma; |
953c66c2 | 3874 | |
82b0f8c3 | 3875 | pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
953c66c2 AK |
3876 | /* |
3877 | * We are going to consume the prealloc table, | |
3878 | * count that as nr_ptes. | |
3879 | */ | |
c4812909 | 3880 | mm_inc_nr_ptes(vma->vm_mm); |
7f2b6ce8 | 3881 | vmf->prealloc_pte = NULL; |
953c66c2 AK |
3882 | } |
3883 | ||
f9ce0be7 | 3884 | vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3885 | { |
82b0f8c3 JK |
3886 | struct vm_area_struct *vma = vmf->vma; |
3887 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3888 | unsigned long haddr = vmf->address & HPAGE_PMD_MASK; | |
10102459 | 3889 | pmd_t entry; |
2b740303 | 3890 | int i; |
d01ac3c3 | 3891 | vm_fault_t ret = VM_FAULT_FALLBACK; |
10102459 KS |
3892 | |
3893 | if (!transhuge_vma_suitable(vma, haddr)) | |
d01ac3c3 | 3894 | return ret; |
10102459 | 3895 | |
10102459 | 3896 | page = compound_head(page); |
d01ac3c3 MWO |
3897 | if (compound_order(page) != HPAGE_PMD_ORDER) |
3898 | return ret; | |
10102459 | 3899 | |
eac96c3e YS |
3900 | /* |
3901 | * Just backoff if any subpage of a THP is corrupted otherwise | |
3902 | * the corrupted page may mapped by PMD silently to escape the | |
3903 | * check. This kind of THP just can be PTE mapped. Access to | |
3904 | * the corrupted subpage should trigger SIGBUS as expected. | |
3905 | */ | |
3906 | if (unlikely(PageHasHWPoisoned(page))) | |
3907 | return ret; | |
3908 | ||
953c66c2 | 3909 | /* |
f0953a1b | 3910 | * Archs like ppc64 need additional space to store information |
953c66c2 AK |
3911 | * related to pte entry. Use the preallocated table for that. |
3912 | */ | |
82b0f8c3 | 3913 | if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { |
4cf58924 | 3914 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); |
82b0f8c3 | 3915 | if (!vmf->prealloc_pte) |
953c66c2 | 3916 | return VM_FAULT_OOM; |
953c66c2 AK |
3917 | } |
3918 | ||
82b0f8c3 JK |
3919 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); |
3920 | if (unlikely(!pmd_none(*vmf->pmd))) | |
10102459 KS |
3921 | goto out; |
3922 | ||
3923 | for (i = 0; i < HPAGE_PMD_NR; i++) | |
3924 | flush_icache_page(vma, page + i); | |
3925 | ||
3926 | entry = mk_huge_pmd(page, vma->vm_page_prot); | |
3927 | if (write) | |
f55e1014 | 3928 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); |
10102459 | 3929 | |
fadae295 | 3930 | add_mm_counter(vma->vm_mm, mm_counter_file(page), HPAGE_PMD_NR); |
10102459 | 3931 | page_add_file_rmap(page, true); |
953c66c2 AK |
3932 | /* |
3933 | * deposit and withdraw with pmd lock held | |
3934 | */ | |
3935 | if (arch_needs_pgtable_deposit()) | |
82b0f8c3 | 3936 | deposit_prealloc_pte(vmf); |
10102459 | 3937 | |
82b0f8c3 | 3938 | set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); |
10102459 | 3939 | |
82b0f8c3 | 3940 | update_mmu_cache_pmd(vma, haddr, vmf->pmd); |
10102459 KS |
3941 | |
3942 | /* fault is handled */ | |
3943 | ret = 0; | |
95ecedcd | 3944 | count_vm_event(THP_FILE_MAPPED); |
10102459 | 3945 | out: |
82b0f8c3 | 3946 | spin_unlock(vmf->ptl); |
10102459 KS |
3947 | return ret; |
3948 | } | |
3949 | #else | |
f9ce0be7 | 3950 | vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3951 | { |
f9ce0be7 | 3952 | return VM_FAULT_FALLBACK; |
10102459 KS |
3953 | } |
3954 | #endif | |
3955 | ||
9d3af4b4 | 3956 | void do_set_pte(struct vm_fault *vmf, struct page *page, unsigned long addr) |
3bb97794 | 3957 | { |
82b0f8c3 JK |
3958 | struct vm_area_struct *vma = vmf->vma; |
3959 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
9d3af4b4 | 3960 | bool prefault = vmf->address != addr; |
3bb97794 | 3961 | pte_t entry; |
7267ec00 | 3962 | |
3bb97794 KS |
3963 | flush_icache_page(vma, page); |
3964 | entry = mk_pte(page, vma->vm_page_prot); | |
46bdb427 WD |
3965 | |
3966 | if (prefault && arch_wants_old_prefaulted_pte()) | |
3967 | entry = pte_mkold(entry); | |
50c25ee9 TB |
3968 | else |
3969 | entry = pte_sw_mkyoung(entry); | |
46bdb427 | 3970 | |
3bb97794 KS |
3971 | if (write) |
3972 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
bae473a4 KS |
3973 | /* copy-on-write page */ |
3974 | if (write && !(vma->vm_flags & VM_SHARED)) { | |
3bb97794 | 3975 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
9d3af4b4 | 3976 | page_add_new_anon_rmap(page, vma, addr, false); |
b518154e | 3977 | lru_cache_add_inactive_or_unevictable(page, vma); |
3bb97794 | 3978 | } else { |
eca56ff9 | 3979 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
dd78fedd | 3980 | page_add_file_rmap(page, false); |
3bb97794 | 3981 | } |
9d3af4b4 | 3982 | set_pte_at(vma->vm_mm, addr, vmf->pte, entry); |
3bb97794 KS |
3983 | } |
3984 | ||
9118c0cb JK |
3985 | /** |
3986 | * finish_fault - finish page fault once we have prepared the page to fault | |
3987 | * | |
3988 | * @vmf: structure describing the fault | |
3989 | * | |
3990 | * This function handles all that is needed to finish a page fault once the | |
3991 | * page to fault in is prepared. It handles locking of PTEs, inserts PTE for | |
3992 | * given page, adds reverse page mapping, handles memcg charges and LRU | |
a862f68a | 3993 | * addition. |
9118c0cb JK |
3994 | * |
3995 | * The function expects the page to be locked and on success it consumes a | |
3996 | * reference of a page being mapped (for the PTE which maps it). | |
a862f68a MR |
3997 | * |
3998 | * Return: %0 on success, %VM_FAULT_ code in case of error. | |
9118c0cb | 3999 | */ |
2b740303 | 4000 | vm_fault_t finish_fault(struct vm_fault *vmf) |
9118c0cb | 4001 | { |
f9ce0be7 | 4002 | struct vm_area_struct *vma = vmf->vma; |
9118c0cb | 4003 | struct page *page; |
f9ce0be7 | 4004 | vm_fault_t ret; |
9118c0cb JK |
4005 | |
4006 | /* Did we COW the page? */ | |
f9ce0be7 | 4007 | if ((vmf->flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) |
9118c0cb JK |
4008 | page = vmf->cow_page; |
4009 | else | |
4010 | page = vmf->page; | |
6b31d595 MH |
4011 | |
4012 | /* | |
4013 | * check even for read faults because we might have lost our CoWed | |
4014 | * page | |
4015 | */ | |
f9ce0be7 KS |
4016 | if (!(vma->vm_flags & VM_SHARED)) { |
4017 | ret = check_stable_address_space(vma->vm_mm); | |
4018 | if (ret) | |
4019 | return ret; | |
4020 | } | |
4021 | ||
4022 | if (pmd_none(*vmf->pmd)) { | |
4023 | if (PageTransCompound(page)) { | |
4024 | ret = do_set_pmd(vmf, page); | |
4025 | if (ret != VM_FAULT_FALLBACK) | |
4026 | return ret; | |
4027 | } | |
4028 | ||
03c4f204 QZ |
4029 | if (vmf->prealloc_pte) |
4030 | pmd_install(vma->vm_mm, vmf->pmd, &vmf->prealloc_pte); | |
4031 | else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd))) | |
f9ce0be7 KS |
4032 | return VM_FAULT_OOM; |
4033 | } | |
4034 | ||
4035 | /* See comment in handle_pte_fault() */ | |
4036 | if (pmd_devmap_trans_unstable(vmf->pmd)) | |
4037 | return 0; | |
4038 | ||
4039 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, | |
4040 | vmf->address, &vmf->ptl); | |
4041 | ret = 0; | |
4042 | /* Re-check under ptl */ | |
4043 | if (likely(pte_none(*vmf->pte))) | |
9d3af4b4 | 4044 | do_set_pte(vmf, page, vmf->address); |
f9ce0be7 KS |
4045 | else |
4046 | ret = VM_FAULT_NOPAGE; | |
4047 | ||
4048 | update_mmu_tlb(vma, vmf->address, vmf->pte); | |
4049 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
9118c0cb JK |
4050 | return ret; |
4051 | } | |
4052 | ||
3a91053a KS |
4053 | static unsigned long fault_around_bytes __read_mostly = |
4054 | rounddown_pow_of_two(65536); | |
a9b0f861 | 4055 | |
a9b0f861 KS |
4056 | #ifdef CONFIG_DEBUG_FS |
4057 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 4058 | { |
a9b0f861 | 4059 | *val = fault_around_bytes; |
1592eef0 KS |
4060 | return 0; |
4061 | } | |
4062 | ||
b4903d6e | 4063 | /* |
da391d64 WK |
4064 | * fault_around_bytes must be rounded down to the nearest page order as it's |
4065 | * what do_fault_around() expects to see. | |
b4903d6e | 4066 | */ |
a9b0f861 | 4067 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 4068 | { |
a9b0f861 | 4069 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 4070 | return -EINVAL; |
b4903d6e AR |
4071 | if (val > PAGE_SIZE) |
4072 | fault_around_bytes = rounddown_pow_of_two(val); | |
4073 | else | |
4074 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
4075 | return 0; |
4076 | } | |
0a1345f8 | 4077 | DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, |
a9b0f861 | 4078 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
1592eef0 KS |
4079 | |
4080 | static int __init fault_around_debugfs(void) | |
4081 | { | |
d9f7979c GKH |
4082 | debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, |
4083 | &fault_around_bytes_fops); | |
1592eef0 KS |
4084 | return 0; |
4085 | } | |
4086 | late_initcall(fault_around_debugfs); | |
1592eef0 | 4087 | #endif |
8c6e50b0 | 4088 | |
1fdb412b KS |
4089 | /* |
4090 | * do_fault_around() tries to map few pages around the fault address. The hope | |
4091 | * is that the pages will be needed soon and this will lower the number of | |
4092 | * faults to handle. | |
4093 | * | |
4094 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
4095 | * not ready to be mapped: not up-to-date, locked, etc. | |
4096 | * | |
4097 | * This function is called with the page table lock taken. In the split ptlock | |
4098 | * case the page table lock only protects only those entries which belong to | |
4099 | * the page table corresponding to the fault address. | |
4100 | * | |
4101 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
4102 | * only once. | |
4103 | * | |
da391d64 WK |
4104 | * fault_around_bytes defines how many bytes we'll try to map. |
4105 | * do_fault_around() expects it to be set to a power of two less than or equal | |
4106 | * to PTRS_PER_PTE. | |
1fdb412b | 4107 | * |
da391d64 WK |
4108 | * The virtual address of the area that we map is naturally aligned to |
4109 | * fault_around_bytes rounded down to the machine page size | |
4110 | * (and therefore to page order). This way it's easier to guarantee | |
4111 | * that we don't cross page table boundaries. | |
1fdb412b | 4112 | */ |
2b740303 | 4113 | static vm_fault_t do_fault_around(struct vm_fault *vmf) |
8c6e50b0 | 4114 | { |
82b0f8c3 | 4115 | unsigned long address = vmf->address, nr_pages, mask; |
0721ec8b | 4116 | pgoff_t start_pgoff = vmf->pgoff; |
bae473a4 | 4117 | pgoff_t end_pgoff; |
2b740303 | 4118 | int off; |
8c6e50b0 | 4119 | |
4db0c3c2 | 4120 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
4121 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
4122 | ||
f9ce0be7 KS |
4123 | address = max(address & mask, vmf->vma->vm_start); |
4124 | off = ((vmf->address - address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); | |
bae473a4 | 4125 | start_pgoff -= off; |
8c6e50b0 KS |
4126 | |
4127 | /* | |
da391d64 WK |
4128 | * end_pgoff is either the end of the page table, the end of |
4129 | * the vma or nr_pages from start_pgoff, depending what is nearest. | |
8c6e50b0 | 4130 | */ |
bae473a4 | 4131 | end_pgoff = start_pgoff - |
f9ce0be7 | 4132 | ((address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + |
8c6e50b0 | 4133 | PTRS_PER_PTE - 1; |
82b0f8c3 | 4134 | end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1, |
bae473a4 | 4135 | start_pgoff + nr_pages - 1); |
8c6e50b0 | 4136 | |
82b0f8c3 | 4137 | if (pmd_none(*vmf->pmd)) { |
4cf58924 | 4138 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm); |
82b0f8c3 | 4139 | if (!vmf->prealloc_pte) |
f9ce0be7 | 4140 | return VM_FAULT_OOM; |
8c6e50b0 KS |
4141 | } |
4142 | ||
f9ce0be7 | 4143 | return vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff); |
8c6e50b0 KS |
4144 | } |
4145 | ||
2b740303 | 4146 | static vm_fault_t do_read_fault(struct vm_fault *vmf) |
e655fb29 | 4147 | { |
82b0f8c3 | 4148 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 4149 | vm_fault_t ret = 0; |
8c6e50b0 KS |
4150 | |
4151 | /* | |
4152 | * Let's call ->map_pages() first and use ->fault() as fallback | |
4153 | * if page by the offset is not ready to be mapped (cold cache or | |
4154 | * something). | |
4155 | */ | |
9b4bdd2f | 4156 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
c949b097 AR |
4157 | if (likely(!userfaultfd_minor(vmf->vma))) { |
4158 | ret = do_fault_around(vmf); | |
4159 | if (ret) | |
4160 | return ret; | |
4161 | } | |
8c6e50b0 | 4162 | } |
e655fb29 | 4163 | |
936ca80d | 4164 | ret = __do_fault(vmf); |
e655fb29 KS |
4165 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
4166 | return ret; | |
4167 | ||
9118c0cb | 4168 | ret |= finish_fault(vmf); |
936ca80d | 4169 | unlock_page(vmf->page); |
7267ec00 | 4170 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
936ca80d | 4171 | put_page(vmf->page); |
e655fb29 KS |
4172 | return ret; |
4173 | } | |
4174 | ||
2b740303 | 4175 | static vm_fault_t do_cow_fault(struct vm_fault *vmf) |
ec47c3b9 | 4176 | { |
82b0f8c3 | 4177 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 4178 | vm_fault_t ret; |
ec47c3b9 KS |
4179 | |
4180 | if (unlikely(anon_vma_prepare(vma))) | |
4181 | return VM_FAULT_OOM; | |
4182 | ||
936ca80d JK |
4183 | vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); |
4184 | if (!vmf->cow_page) | |
ec47c3b9 KS |
4185 | return VM_FAULT_OOM; |
4186 | ||
8f425e4e MWO |
4187 | if (mem_cgroup_charge(page_folio(vmf->cow_page), vma->vm_mm, |
4188 | GFP_KERNEL)) { | |
936ca80d | 4189 | put_page(vmf->cow_page); |
ec47c3b9 KS |
4190 | return VM_FAULT_OOM; |
4191 | } | |
9d82c694 | 4192 | cgroup_throttle_swaprate(vmf->cow_page, GFP_KERNEL); |
ec47c3b9 | 4193 | |
936ca80d | 4194 | ret = __do_fault(vmf); |
ec47c3b9 KS |
4195 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
4196 | goto uncharge_out; | |
3917048d JK |
4197 | if (ret & VM_FAULT_DONE_COW) |
4198 | return ret; | |
ec47c3b9 | 4199 | |
b1aa812b | 4200 | copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); |
936ca80d | 4201 | __SetPageUptodate(vmf->cow_page); |
ec47c3b9 | 4202 | |
9118c0cb | 4203 | ret |= finish_fault(vmf); |
b1aa812b JK |
4204 | unlock_page(vmf->page); |
4205 | put_page(vmf->page); | |
7267ec00 KS |
4206 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
4207 | goto uncharge_out; | |
ec47c3b9 KS |
4208 | return ret; |
4209 | uncharge_out: | |
936ca80d | 4210 | put_page(vmf->cow_page); |
ec47c3b9 KS |
4211 | return ret; |
4212 | } | |
4213 | ||
2b740303 | 4214 | static vm_fault_t do_shared_fault(struct vm_fault *vmf) |
1da177e4 | 4215 | { |
82b0f8c3 | 4216 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 4217 | vm_fault_t ret, tmp; |
1d65f86d | 4218 | |
936ca80d | 4219 | ret = __do_fault(vmf); |
7eae74af | 4220 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 4221 | return ret; |
1da177e4 LT |
4222 | |
4223 | /* | |
f0c6d4d2 KS |
4224 | * Check if the backing address space wants to know that the page is |
4225 | * about to become writable | |
1da177e4 | 4226 | */ |
fb09a464 | 4227 | if (vma->vm_ops->page_mkwrite) { |
936ca80d | 4228 | unlock_page(vmf->page); |
38b8cb7f | 4229 | tmp = do_page_mkwrite(vmf); |
fb09a464 KS |
4230 | if (unlikely(!tmp || |
4231 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
936ca80d | 4232 | put_page(vmf->page); |
fb09a464 | 4233 | return tmp; |
4294621f | 4234 | } |
fb09a464 KS |
4235 | } |
4236 | ||
9118c0cb | 4237 | ret |= finish_fault(vmf); |
7267ec00 KS |
4238 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
4239 | VM_FAULT_RETRY))) { | |
936ca80d JK |
4240 | unlock_page(vmf->page); |
4241 | put_page(vmf->page); | |
f0c6d4d2 | 4242 | return ret; |
1da177e4 | 4243 | } |
b827e496 | 4244 | |
89b15332 | 4245 | ret |= fault_dirty_shared_page(vmf); |
1d65f86d | 4246 | return ret; |
54cb8821 | 4247 | } |
d00806b1 | 4248 | |
9a95f3cf | 4249 | /* |
c1e8d7c6 | 4250 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
9a95f3cf | 4251 | * but allow concurrent faults). |
c1e8d7c6 | 4252 | * The mmap_lock may have been released depending on flags and our |
9138e47e | 4253 | * return value. See filemap_fault() and __folio_lock_or_retry(). |
c1e8d7c6 | 4254 | * If mmap_lock is released, vma may become invalid (for example |
fc8efd2d | 4255 | * by other thread calling munmap()). |
9a95f3cf | 4256 | */ |
2b740303 | 4257 | static vm_fault_t do_fault(struct vm_fault *vmf) |
54cb8821 | 4258 | { |
82b0f8c3 | 4259 | struct vm_area_struct *vma = vmf->vma; |
fc8efd2d | 4260 | struct mm_struct *vm_mm = vma->vm_mm; |
2b740303 | 4261 | vm_fault_t ret; |
54cb8821 | 4262 | |
ff09d7ec AK |
4263 | /* |
4264 | * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND | |
4265 | */ | |
4266 | if (!vma->vm_ops->fault) { | |
4267 | /* | |
4268 | * If we find a migration pmd entry or a none pmd entry, which | |
4269 | * should never happen, return SIGBUS | |
4270 | */ | |
4271 | if (unlikely(!pmd_present(*vmf->pmd))) | |
4272 | ret = VM_FAULT_SIGBUS; | |
4273 | else { | |
4274 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, | |
4275 | vmf->pmd, | |
4276 | vmf->address, | |
4277 | &vmf->ptl); | |
4278 | /* | |
4279 | * Make sure this is not a temporary clearing of pte | |
4280 | * by holding ptl and checking again. A R/M/W update | |
4281 | * of pte involves: take ptl, clearing the pte so that | |
4282 | * we don't have concurrent modification by hardware | |
4283 | * followed by an update. | |
4284 | */ | |
4285 | if (unlikely(pte_none(*vmf->pte))) | |
4286 | ret = VM_FAULT_SIGBUS; | |
4287 | else | |
4288 | ret = VM_FAULT_NOPAGE; | |
4289 | ||
4290 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4291 | } | |
4292 | } else if (!(vmf->flags & FAULT_FLAG_WRITE)) | |
b0b9b3df HD |
4293 | ret = do_read_fault(vmf); |
4294 | else if (!(vma->vm_flags & VM_SHARED)) | |
4295 | ret = do_cow_fault(vmf); | |
4296 | else | |
4297 | ret = do_shared_fault(vmf); | |
4298 | ||
4299 | /* preallocated pagetable is unused: free it */ | |
4300 | if (vmf->prealloc_pte) { | |
fc8efd2d | 4301 | pte_free(vm_mm, vmf->prealloc_pte); |
7f2b6ce8 | 4302 | vmf->prealloc_pte = NULL; |
b0b9b3df HD |
4303 | } |
4304 | return ret; | |
54cb8821 NP |
4305 | } |
4306 | ||
f4c0d836 YS |
4307 | int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
4308 | unsigned long addr, int page_nid, int *flags) | |
9532fec1 MG |
4309 | { |
4310 | get_page(page); | |
4311 | ||
4312 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 4313 | if (page_nid == numa_node_id()) { |
9532fec1 | 4314 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
4315 | *flags |= TNF_FAULT_LOCAL; |
4316 | } | |
9532fec1 MG |
4317 | |
4318 | return mpol_misplaced(page, vma, addr); | |
4319 | } | |
4320 | ||
2b740303 | 4321 | static vm_fault_t do_numa_page(struct vm_fault *vmf) |
d10e63f2 | 4322 | { |
82b0f8c3 | 4323 | struct vm_area_struct *vma = vmf->vma; |
4daae3b4 | 4324 | struct page *page = NULL; |
98fa15f3 | 4325 | int page_nid = NUMA_NO_NODE; |
90572890 | 4326 | int last_cpupid; |
cbee9f88 | 4327 | int target_nid; |
04a86453 | 4328 | pte_t pte, old_pte; |
288bc549 | 4329 | bool was_writable = pte_savedwrite(vmf->orig_pte); |
6688cc05 | 4330 | int flags = 0; |
d10e63f2 MG |
4331 | |
4332 | /* | |
166f61b9 TH |
4333 | * The "pte" at this point cannot be used safely without |
4334 | * validation through pte_unmap_same(). It's of NUMA type but | |
4335 | * the pfn may be screwed if the read is non atomic. | |
166f61b9 | 4336 | */ |
82b0f8c3 JK |
4337 | vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); |
4338 | spin_lock(vmf->ptl); | |
cee216a6 | 4339 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { |
82b0f8c3 | 4340 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
4341 | goto out; |
4342 | } | |
4343 | ||
b99a342d HY |
4344 | /* Get the normal PTE */ |
4345 | old_pte = ptep_get(vmf->pte); | |
04a86453 | 4346 | pte = pte_modify(old_pte, vma->vm_page_prot); |
d10e63f2 | 4347 | |
82b0f8c3 | 4348 | page = vm_normal_page(vma, vmf->address, pte); |
b99a342d HY |
4349 | if (!page) |
4350 | goto out_map; | |
d10e63f2 | 4351 | |
e81c4802 | 4352 | /* TODO: handle PTE-mapped THP */ |
b99a342d HY |
4353 | if (PageCompound(page)) |
4354 | goto out_map; | |
e81c4802 | 4355 | |
6688cc05 | 4356 | /* |
bea66fbd MG |
4357 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
4358 | * much anyway since they can be in shared cache state. This misses | |
4359 | * the case where a mapping is writable but the process never writes | |
4360 | * to it but pte_write gets cleared during protection updates and | |
4361 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
4362 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 4363 | */ |
b99a342d | 4364 | if (!was_writable) |
6688cc05 PZ |
4365 | flags |= TNF_NO_GROUP; |
4366 | ||
dabe1d99 RR |
4367 | /* |
4368 | * Flag if the page is shared between multiple address spaces. This | |
4369 | * is later used when determining whether to group tasks together | |
4370 | */ | |
4371 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
4372 | flags |= TNF_SHARED; | |
4373 | ||
90572890 | 4374 | last_cpupid = page_cpupid_last(page); |
8191acbd | 4375 | page_nid = page_to_nid(page); |
82b0f8c3 | 4376 | target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, |
bae473a4 | 4377 | &flags); |
98fa15f3 | 4378 | if (target_nid == NUMA_NO_NODE) { |
4daae3b4 | 4379 | put_page(page); |
b99a342d | 4380 | goto out_map; |
4daae3b4 | 4381 | } |
b99a342d | 4382 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
4383 | |
4384 | /* Migrate to the requested node */ | |
bf90ac19 | 4385 | if (migrate_misplaced_page(page, vma, target_nid)) { |
8191acbd | 4386 | page_nid = target_nid; |
6688cc05 | 4387 | flags |= TNF_MIGRATED; |
b99a342d | 4388 | } else { |
074c2381 | 4389 | flags |= TNF_MIGRATE_FAIL; |
b99a342d HY |
4390 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
4391 | spin_lock(vmf->ptl); | |
4392 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { | |
4393 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4394 | goto out; | |
4395 | } | |
4396 | goto out_map; | |
4397 | } | |
4daae3b4 MG |
4398 | |
4399 | out: | |
98fa15f3 | 4400 | if (page_nid != NUMA_NO_NODE) |
6688cc05 | 4401 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 | 4402 | return 0; |
b99a342d HY |
4403 | out_map: |
4404 | /* | |
4405 | * Make it present again, depending on how arch implements | |
4406 | * non-accessible ptes, some can allow access by kernel mode. | |
4407 | */ | |
4408 | old_pte = ptep_modify_prot_start(vma, vmf->address, vmf->pte); | |
4409 | pte = pte_modify(old_pte, vma->vm_page_prot); | |
4410 | pte = pte_mkyoung(pte); | |
4411 | if (was_writable) | |
4412 | pte = pte_mkwrite(pte); | |
4413 | ptep_modify_prot_commit(vma, vmf->address, vmf->pte, old_pte, pte); | |
4414 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
4415 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4416 | goto out; | |
d10e63f2 MG |
4417 | } |
4418 | ||
2b740303 | 4419 | static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 4420 | { |
f4200391 | 4421 | if (vma_is_anonymous(vmf->vma)) |
82b0f8c3 | 4422 | return do_huge_pmd_anonymous_page(vmf); |
a2d58167 | 4423 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 4424 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
b96375f7 MW |
4425 | return VM_FAULT_FALLBACK; |
4426 | } | |
4427 | ||
183f24aa | 4428 | /* `inline' is required to avoid gcc 4.1.2 build error */ |
5db4f15c | 4429 | static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 4430 | { |
529b930b | 4431 | if (vma_is_anonymous(vmf->vma)) { |
5db4f15c | 4432 | if (userfaultfd_huge_pmd_wp(vmf->vma, vmf->orig_pmd)) |
529b930b | 4433 | return handle_userfault(vmf, VM_UFFD_WP); |
5db4f15c | 4434 | return do_huge_pmd_wp_page(vmf); |
529b930b | 4435 | } |
327e9fd4 THV |
4436 | if (vmf->vma->vm_ops->huge_fault) { |
4437 | vm_fault_t ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); | |
4438 | ||
4439 | if (!(ret & VM_FAULT_FALLBACK)) | |
4440 | return ret; | |
4441 | } | |
af9e4d5f | 4442 | |
327e9fd4 | 4443 | /* COW or write-notify handled on pte level: split pmd. */ |
82b0f8c3 | 4444 | __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); |
af9e4d5f | 4445 | |
b96375f7 MW |
4446 | return VM_FAULT_FALLBACK; |
4447 | } | |
4448 | ||
2b740303 | 4449 | static vm_fault_t create_huge_pud(struct vm_fault *vmf) |
a00cc7d9 | 4450 | { |
327e9fd4 THV |
4451 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ |
4452 | defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) | |
a00cc7d9 MW |
4453 | /* No support for anonymous transparent PUD pages yet */ |
4454 | if (vma_is_anonymous(vmf->vma)) | |
327e9fd4 THV |
4455 | goto split; |
4456 | if (vmf->vma->vm_ops->huge_fault) { | |
4457 | vm_fault_t ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); | |
4458 | ||
4459 | if (!(ret & VM_FAULT_FALLBACK)) | |
4460 | return ret; | |
4461 | } | |
4462 | split: | |
4463 | /* COW or write-notify not handled on PUD level: split pud.*/ | |
4464 | __split_huge_pud(vmf->vma, vmf->pud, vmf->address); | |
a00cc7d9 MW |
4465 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
4466 | return VM_FAULT_FALLBACK; | |
4467 | } | |
4468 | ||
2b740303 | 4469 | static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) |
a00cc7d9 MW |
4470 | { |
4471 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4472 | /* No support for anonymous transparent PUD pages yet */ | |
4473 | if (vma_is_anonymous(vmf->vma)) | |
4474 | return VM_FAULT_FALLBACK; | |
4475 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 4476 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
4477 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
4478 | return VM_FAULT_FALLBACK; | |
4479 | } | |
4480 | ||
1da177e4 LT |
4481 | /* |
4482 | * These routines also need to handle stuff like marking pages dirty | |
4483 | * and/or accessed for architectures that don't do it in hardware (most | |
4484 | * RISC architectures). The early dirtying is also good on the i386. | |
4485 | * | |
4486 | * There is also a hook called "update_mmu_cache()" that architectures | |
4487 | * with external mmu caches can use to update those (ie the Sparc or | |
4488 | * PowerPC hashed page tables that act as extended TLBs). | |
4489 | * | |
c1e8d7c6 | 4490 | * We enter with non-exclusive mmap_lock (to exclude vma changes, but allow |
7267ec00 | 4491 | * concurrent faults). |
9a95f3cf | 4492 | * |
c1e8d7c6 | 4493 | * The mmap_lock may have been released depending on flags and our return value. |
9138e47e | 4494 | * See filemap_fault() and __folio_lock_or_retry(). |
1da177e4 | 4495 | */ |
2b740303 | 4496 | static vm_fault_t handle_pte_fault(struct vm_fault *vmf) |
1da177e4 LT |
4497 | { |
4498 | pte_t entry; | |
4499 | ||
82b0f8c3 | 4500 | if (unlikely(pmd_none(*vmf->pmd))) { |
7267ec00 KS |
4501 | /* |
4502 | * Leave __pte_alloc() until later: because vm_ops->fault may | |
4503 | * want to allocate huge page, and if we expose page table | |
4504 | * for an instant, it will be difficult to retract from | |
4505 | * concurrent faults and from rmap lookups. | |
4506 | */ | |
82b0f8c3 | 4507 | vmf->pte = NULL; |
7267ec00 | 4508 | } else { |
f9ce0be7 KS |
4509 | /* |
4510 | * If a huge pmd materialized under us just retry later. Use | |
4511 | * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead | |
4512 | * of pmd_trans_huge() to ensure the pmd didn't become | |
4513 | * pmd_trans_huge under us and then back to pmd_none, as a | |
4514 | * result of MADV_DONTNEED running immediately after a huge pmd | |
4515 | * fault in a different thread of this mm, in turn leading to a | |
4516 | * misleading pmd_trans_huge() retval. All we have to ensure is | |
4517 | * that it is a regular pmd that we can walk with | |
4518 | * pte_offset_map() and we can do that through an atomic read | |
4519 | * in C, which is what pmd_trans_unstable() provides. | |
4520 | */ | |
d0f0931d | 4521 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
4522 | return 0; |
4523 | /* | |
4524 | * A regular pmd is established and it can't morph into a huge | |
4525 | * pmd from under us anymore at this point because we hold the | |
c1e8d7c6 | 4526 | * mmap_lock read mode and khugepaged takes it in write mode. |
7267ec00 KS |
4527 | * So now it's safe to run pte_offset_map(). |
4528 | */ | |
82b0f8c3 | 4529 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
2994302b | 4530 | vmf->orig_pte = *vmf->pte; |
7267ec00 KS |
4531 | |
4532 | /* | |
4533 | * some architectures can have larger ptes than wordsize, | |
4534 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and | |
b03a0fe0 PM |
4535 | * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic |
4536 | * accesses. The code below just needs a consistent view | |
4537 | * for the ifs and we later double check anyway with the | |
7267ec00 KS |
4538 | * ptl lock held. So here a barrier will do. |
4539 | */ | |
4540 | barrier(); | |
2994302b | 4541 | if (pte_none(vmf->orig_pte)) { |
82b0f8c3 JK |
4542 | pte_unmap(vmf->pte); |
4543 | vmf->pte = NULL; | |
65500d23 | 4544 | } |
1da177e4 LT |
4545 | } |
4546 | ||
82b0f8c3 JK |
4547 | if (!vmf->pte) { |
4548 | if (vma_is_anonymous(vmf->vma)) | |
4549 | return do_anonymous_page(vmf); | |
7267ec00 | 4550 | else |
82b0f8c3 | 4551 | return do_fault(vmf); |
7267ec00 KS |
4552 | } |
4553 | ||
2994302b JK |
4554 | if (!pte_present(vmf->orig_pte)) |
4555 | return do_swap_page(vmf); | |
7267ec00 | 4556 | |
2994302b JK |
4557 | if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) |
4558 | return do_numa_page(vmf); | |
d10e63f2 | 4559 | |
82b0f8c3 JK |
4560 | vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); |
4561 | spin_lock(vmf->ptl); | |
2994302b | 4562 | entry = vmf->orig_pte; |
7df67697 BM |
4563 | if (unlikely(!pte_same(*vmf->pte, entry))) { |
4564 | update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); | |
8f4e2101 | 4565 | goto unlock; |
7df67697 | 4566 | } |
82b0f8c3 | 4567 | if (vmf->flags & FAULT_FLAG_WRITE) { |
f6f37321 | 4568 | if (!pte_write(entry)) |
2994302b | 4569 | return do_wp_page(vmf); |
1da177e4 LT |
4570 | entry = pte_mkdirty(entry); |
4571 | } | |
4572 | entry = pte_mkyoung(entry); | |
82b0f8c3 JK |
4573 | if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, |
4574 | vmf->flags & FAULT_FLAG_WRITE)) { | |
4575 | update_mmu_cache(vmf->vma, vmf->address, vmf->pte); | |
1a44e149 | 4576 | } else { |
b7333b58 YS |
4577 | /* Skip spurious TLB flush for retried page fault */ |
4578 | if (vmf->flags & FAULT_FLAG_TRIED) | |
4579 | goto unlock; | |
1a44e149 AA |
4580 | /* |
4581 | * This is needed only for protection faults but the arch code | |
4582 | * is not yet telling us if this is a protection fault or not. | |
4583 | * This still avoids useless tlb flushes for .text page faults | |
4584 | * with threads. | |
4585 | */ | |
82b0f8c3 JK |
4586 | if (vmf->flags & FAULT_FLAG_WRITE) |
4587 | flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); | |
1a44e149 | 4588 | } |
8f4e2101 | 4589 | unlock: |
82b0f8c3 | 4590 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 4591 | return 0; |
1da177e4 LT |
4592 | } |
4593 | ||
4594 | /* | |
4595 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf | 4596 | * |
c1e8d7c6 | 4597 | * The mmap_lock may have been released depending on flags and our |
9138e47e | 4598 | * return value. See filemap_fault() and __folio_lock_or_retry(). |
1da177e4 | 4599 | */ |
2b740303 SJ |
4600 | static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma, |
4601 | unsigned long address, unsigned int flags) | |
1da177e4 | 4602 | { |
82b0f8c3 | 4603 | struct vm_fault vmf = { |
bae473a4 | 4604 | .vma = vma, |
1a29d85e | 4605 | .address = address & PAGE_MASK, |
bae473a4 | 4606 | .flags = flags, |
0721ec8b | 4607 | .pgoff = linear_page_index(vma, address), |
667240e0 | 4608 | .gfp_mask = __get_fault_gfp_mask(vma), |
bae473a4 | 4609 | }; |
fde26bed | 4610 | unsigned int dirty = flags & FAULT_FLAG_WRITE; |
dcddffd4 | 4611 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 4612 | pgd_t *pgd; |
c2febafc | 4613 | p4d_t *p4d; |
2b740303 | 4614 | vm_fault_t ret; |
1da177e4 | 4615 | |
1da177e4 | 4616 | pgd = pgd_offset(mm, address); |
c2febafc KS |
4617 | p4d = p4d_alloc(mm, pgd, address); |
4618 | if (!p4d) | |
4619 | return VM_FAULT_OOM; | |
a00cc7d9 | 4620 | |
c2febafc | 4621 | vmf.pud = pud_alloc(mm, p4d, address); |
a00cc7d9 | 4622 | if (!vmf.pud) |
c74df32c | 4623 | return VM_FAULT_OOM; |
625110b5 | 4624 | retry_pud: |
7635d9cb | 4625 | if (pud_none(*vmf.pud) && __transparent_hugepage_enabled(vma)) { |
a00cc7d9 MW |
4626 | ret = create_huge_pud(&vmf); |
4627 | if (!(ret & VM_FAULT_FALLBACK)) | |
4628 | return ret; | |
4629 | } else { | |
4630 | pud_t orig_pud = *vmf.pud; | |
4631 | ||
4632 | barrier(); | |
4633 | if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { | |
a00cc7d9 | 4634 | |
a00cc7d9 MW |
4635 | /* NUMA case for anonymous PUDs would go here */ |
4636 | ||
f6f37321 | 4637 | if (dirty && !pud_write(orig_pud)) { |
a00cc7d9 MW |
4638 | ret = wp_huge_pud(&vmf, orig_pud); |
4639 | if (!(ret & VM_FAULT_FALLBACK)) | |
4640 | return ret; | |
4641 | } else { | |
4642 | huge_pud_set_accessed(&vmf, orig_pud); | |
4643 | return 0; | |
4644 | } | |
4645 | } | |
4646 | } | |
4647 | ||
4648 | vmf.pmd = pmd_alloc(mm, vmf.pud, address); | |
82b0f8c3 | 4649 | if (!vmf.pmd) |
c74df32c | 4650 | return VM_FAULT_OOM; |
625110b5 TH |
4651 | |
4652 | /* Huge pud page fault raced with pmd_alloc? */ | |
4653 | if (pud_trans_unstable(vmf.pud)) | |
4654 | goto retry_pud; | |
4655 | ||
7635d9cb | 4656 | if (pmd_none(*vmf.pmd) && __transparent_hugepage_enabled(vma)) { |
a2d58167 | 4657 | ret = create_huge_pmd(&vmf); |
c0292554 KS |
4658 | if (!(ret & VM_FAULT_FALLBACK)) |
4659 | return ret; | |
71e3aac0 | 4660 | } else { |
5db4f15c | 4661 | vmf.orig_pmd = *vmf.pmd; |
1f1d06c3 | 4662 | |
71e3aac0 | 4663 | barrier(); |
5db4f15c | 4664 | if (unlikely(is_swap_pmd(vmf.orig_pmd))) { |
84c3fc4e | 4665 | VM_BUG_ON(thp_migration_supported() && |
5db4f15c YS |
4666 | !is_pmd_migration_entry(vmf.orig_pmd)); |
4667 | if (is_pmd_migration_entry(vmf.orig_pmd)) | |
84c3fc4e ZY |
4668 | pmd_migration_entry_wait(mm, vmf.pmd); |
4669 | return 0; | |
4670 | } | |
5db4f15c YS |
4671 | if (pmd_trans_huge(vmf.orig_pmd) || pmd_devmap(vmf.orig_pmd)) { |
4672 | if (pmd_protnone(vmf.orig_pmd) && vma_is_accessible(vma)) | |
4673 | return do_huge_pmd_numa_page(&vmf); | |
d10e63f2 | 4674 | |
5db4f15c YS |
4675 | if (dirty && !pmd_write(vmf.orig_pmd)) { |
4676 | ret = wp_huge_pmd(&vmf); | |
9845cbbd KS |
4677 | if (!(ret & VM_FAULT_FALLBACK)) |
4678 | return ret; | |
a1dd450b | 4679 | } else { |
5db4f15c | 4680 | huge_pmd_set_accessed(&vmf); |
9845cbbd | 4681 | return 0; |
1f1d06c3 | 4682 | } |
71e3aac0 AA |
4683 | } |
4684 | } | |
4685 | ||
82b0f8c3 | 4686 | return handle_pte_fault(&vmf); |
1da177e4 LT |
4687 | } |
4688 | ||
bce617ed | 4689 | /** |
f0953a1b | 4690 | * mm_account_fault - Do page fault accounting |
bce617ed PX |
4691 | * |
4692 | * @regs: the pt_regs struct pointer. When set to NULL, will skip accounting | |
4693 | * of perf event counters, but we'll still do the per-task accounting to | |
4694 | * the task who triggered this page fault. | |
4695 | * @address: the faulted address. | |
4696 | * @flags: the fault flags. | |
4697 | * @ret: the fault retcode. | |
4698 | * | |
f0953a1b | 4699 | * This will take care of most of the page fault accounting. Meanwhile, it |
bce617ed | 4700 | * will also include the PERF_COUNT_SW_PAGE_FAULTS_[MAJ|MIN] perf counter |
f0953a1b | 4701 | * updates. However, note that the handling of PERF_COUNT_SW_PAGE_FAULTS should |
bce617ed PX |
4702 | * still be in per-arch page fault handlers at the entry of page fault. |
4703 | */ | |
4704 | static inline void mm_account_fault(struct pt_regs *regs, | |
4705 | unsigned long address, unsigned int flags, | |
4706 | vm_fault_t ret) | |
4707 | { | |
4708 | bool major; | |
4709 | ||
4710 | /* | |
4711 | * We don't do accounting for some specific faults: | |
4712 | * | |
4713 | * - Unsuccessful faults (e.g. when the address wasn't valid). That | |
4714 | * includes arch_vma_access_permitted() failing before reaching here. | |
4715 | * So this is not a "this many hardware page faults" counter. We | |
4716 | * should use the hw profiling for that. | |
4717 | * | |
4718 | * - Incomplete faults (VM_FAULT_RETRY). They will only be counted | |
4719 | * once they're completed. | |
4720 | */ | |
4721 | if (ret & (VM_FAULT_ERROR | VM_FAULT_RETRY)) | |
4722 | return; | |
4723 | ||
4724 | /* | |
4725 | * We define the fault as a major fault when the final successful fault | |
4726 | * is VM_FAULT_MAJOR, or if it retried (which implies that we couldn't | |
4727 | * handle it immediately previously). | |
4728 | */ | |
4729 | major = (ret & VM_FAULT_MAJOR) || (flags & FAULT_FLAG_TRIED); | |
4730 | ||
a2beb5f1 PX |
4731 | if (major) |
4732 | current->maj_flt++; | |
4733 | else | |
4734 | current->min_flt++; | |
4735 | ||
bce617ed | 4736 | /* |
a2beb5f1 PX |
4737 | * If the fault is done for GUP, regs will be NULL. We only do the |
4738 | * accounting for the per thread fault counters who triggered the | |
4739 | * fault, and we skip the perf event updates. | |
bce617ed PX |
4740 | */ |
4741 | if (!regs) | |
4742 | return; | |
4743 | ||
a2beb5f1 | 4744 | if (major) |
bce617ed | 4745 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); |
a2beb5f1 | 4746 | else |
bce617ed | 4747 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); |
bce617ed PX |
4748 | } |
4749 | ||
9a95f3cf PC |
4750 | /* |
4751 | * By the time we get here, we already hold the mm semaphore | |
4752 | * | |
c1e8d7c6 | 4753 | * The mmap_lock may have been released depending on flags and our |
9138e47e | 4754 | * return value. See filemap_fault() and __folio_lock_or_retry(). |
9a95f3cf | 4755 | */ |
2b740303 | 4756 | vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
bce617ed | 4757 | unsigned int flags, struct pt_regs *regs) |
519e5247 | 4758 | { |
2b740303 | 4759 | vm_fault_t ret; |
519e5247 JW |
4760 | |
4761 | __set_current_state(TASK_RUNNING); | |
4762 | ||
4763 | count_vm_event(PGFAULT); | |
2262185c | 4764 | count_memcg_event_mm(vma->vm_mm, PGFAULT); |
519e5247 JW |
4765 | |
4766 | /* do counter updates before entering really critical section. */ | |
4767 | check_sync_rss_stat(current); | |
4768 | ||
de0c799b LD |
4769 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
4770 | flags & FAULT_FLAG_INSTRUCTION, | |
4771 | flags & FAULT_FLAG_REMOTE)) | |
4772 | return VM_FAULT_SIGSEGV; | |
4773 | ||
519e5247 JW |
4774 | /* |
4775 | * Enable the memcg OOM handling for faults triggered in user | |
4776 | * space. Kernel faults are handled more gracefully. | |
4777 | */ | |
4778 | if (flags & FAULT_FLAG_USER) | |
29ef680a | 4779 | mem_cgroup_enter_user_fault(); |
519e5247 | 4780 | |
bae473a4 KS |
4781 | if (unlikely(is_vm_hugetlb_page(vma))) |
4782 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); | |
4783 | else | |
4784 | ret = __handle_mm_fault(vma, address, flags); | |
519e5247 | 4785 | |
49426420 | 4786 | if (flags & FAULT_FLAG_USER) { |
29ef680a | 4787 | mem_cgroup_exit_user_fault(); |
166f61b9 TH |
4788 | /* |
4789 | * The task may have entered a memcg OOM situation but | |
4790 | * if the allocation error was handled gracefully (no | |
4791 | * VM_FAULT_OOM), there is no need to kill anything. | |
4792 | * Just clean up the OOM state peacefully. | |
4793 | */ | |
4794 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
4795 | mem_cgroup_oom_synchronize(false); | |
49426420 | 4796 | } |
3812c8c8 | 4797 | |
bce617ed PX |
4798 | mm_account_fault(regs, address, flags, ret); |
4799 | ||
519e5247 JW |
4800 | return ret; |
4801 | } | |
e1d6d01a | 4802 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 4803 | |
90eceff1 KS |
4804 | #ifndef __PAGETABLE_P4D_FOLDED |
4805 | /* | |
4806 | * Allocate p4d page table. | |
4807 | * We've already handled the fast-path in-line. | |
4808 | */ | |
4809 | int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
4810 | { | |
4811 | p4d_t *new = p4d_alloc_one(mm, address); | |
4812 | if (!new) | |
4813 | return -ENOMEM; | |
4814 | ||
90eceff1 | 4815 | spin_lock(&mm->page_table_lock); |
ed33b5a6 | 4816 | if (pgd_present(*pgd)) { /* Another has populated it */ |
90eceff1 | 4817 | p4d_free(mm, new); |
ed33b5a6 QZ |
4818 | } else { |
4819 | smp_wmb(); /* See comment in pmd_install() */ | |
90eceff1 | 4820 | pgd_populate(mm, pgd, new); |
ed33b5a6 | 4821 | } |
90eceff1 KS |
4822 | spin_unlock(&mm->page_table_lock); |
4823 | return 0; | |
4824 | } | |
4825 | #endif /* __PAGETABLE_P4D_FOLDED */ | |
4826 | ||
1da177e4 LT |
4827 | #ifndef __PAGETABLE_PUD_FOLDED |
4828 | /* | |
4829 | * Allocate page upper directory. | |
872fec16 | 4830 | * We've already handled the fast-path in-line. |
1da177e4 | 4831 | */ |
c2febafc | 4832 | int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) |
1da177e4 | 4833 | { |
c74df32c HD |
4834 | pud_t *new = pud_alloc_one(mm, address); |
4835 | if (!new) | |
1bb3630e | 4836 | return -ENOMEM; |
1da177e4 | 4837 | |
872fec16 | 4838 | spin_lock(&mm->page_table_lock); |
b4e98d9a KS |
4839 | if (!p4d_present(*p4d)) { |
4840 | mm_inc_nr_puds(mm); | |
ed33b5a6 | 4841 | smp_wmb(); /* See comment in pmd_install() */ |
c2febafc | 4842 | p4d_populate(mm, p4d, new); |
b4e98d9a | 4843 | } else /* Another has populated it */ |
5e541973 | 4844 | pud_free(mm, new); |
c74df32c | 4845 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 4846 | return 0; |
1da177e4 LT |
4847 | } |
4848 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
4849 | ||
4850 | #ifndef __PAGETABLE_PMD_FOLDED | |
4851 | /* | |
4852 | * Allocate page middle directory. | |
872fec16 | 4853 | * We've already handled the fast-path in-line. |
1da177e4 | 4854 | */ |
1bb3630e | 4855 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 4856 | { |
a00cc7d9 | 4857 | spinlock_t *ptl; |
c74df32c HD |
4858 | pmd_t *new = pmd_alloc_one(mm, address); |
4859 | if (!new) | |
1bb3630e | 4860 | return -ENOMEM; |
1da177e4 | 4861 | |
a00cc7d9 | 4862 | ptl = pud_lock(mm, pud); |
dc6c9a35 KS |
4863 | if (!pud_present(*pud)) { |
4864 | mm_inc_nr_pmds(mm); | |
ed33b5a6 | 4865 | smp_wmb(); /* See comment in pmd_install() */ |
1bb3630e | 4866 | pud_populate(mm, pud, new); |
ed33b5a6 | 4867 | } else { /* Another has populated it */ |
5e541973 | 4868 | pmd_free(mm, new); |
ed33b5a6 | 4869 | } |
a00cc7d9 | 4870 | spin_unlock(ptl); |
1bb3630e | 4871 | return 0; |
e0f39591 | 4872 | } |
1da177e4 LT |
4873 | #endif /* __PAGETABLE_PMD_FOLDED */ |
4874 | ||
9fd6dad1 PB |
4875 | int follow_invalidate_pte(struct mm_struct *mm, unsigned long address, |
4876 | struct mmu_notifier_range *range, pte_t **ptepp, | |
4877 | pmd_t **pmdpp, spinlock_t **ptlp) | |
f8ad0f49 JW |
4878 | { |
4879 | pgd_t *pgd; | |
c2febafc | 4880 | p4d_t *p4d; |
f8ad0f49 JW |
4881 | pud_t *pud; |
4882 | pmd_t *pmd; | |
4883 | pte_t *ptep; | |
4884 | ||
4885 | pgd = pgd_offset(mm, address); | |
4886 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
4887 | goto out; | |
4888 | ||
c2febafc KS |
4889 | p4d = p4d_offset(pgd, address); |
4890 | if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) | |
4891 | goto out; | |
4892 | ||
4893 | pud = pud_offset(p4d, address); | |
f8ad0f49 JW |
4894 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
4895 | goto out; | |
4896 | ||
4897 | pmd = pmd_offset(pud, address); | |
f66055ab | 4898 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 | 4899 | |
09796395 RZ |
4900 | if (pmd_huge(*pmd)) { |
4901 | if (!pmdpp) | |
4902 | goto out; | |
4903 | ||
ac46d4f3 | 4904 | if (range) { |
7269f999 | 4905 | mmu_notifier_range_init(range, MMU_NOTIFY_CLEAR, 0, |
6f4f13e8 JG |
4906 | NULL, mm, address & PMD_MASK, |
4907 | (address & PMD_MASK) + PMD_SIZE); | |
ac46d4f3 | 4908 | mmu_notifier_invalidate_range_start(range); |
a4d1a885 | 4909 | } |
09796395 RZ |
4910 | *ptlp = pmd_lock(mm, pmd); |
4911 | if (pmd_huge(*pmd)) { | |
4912 | *pmdpp = pmd; | |
4913 | return 0; | |
4914 | } | |
4915 | spin_unlock(*ptlp); | |
ac46d4f3 JG |
4916 | if (range) |
4917 | mmu_notifier_invalidate_range_end(range); | |
09796395 RZ |
4918 | } |
4919 | ||
4920 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
f8ad0f49 JW |
4921 | goto out; |
4922 | ||
ac46d4f3 | 4923 | if (range) { |
7269f999 | 4924 | mmu_notifier_range_init(range, MMU_NOTIFY_CLEAR, 0, NULL, mm, |
6f4f13e8 JG |
4925 | address & PAGE_MASK, |
4926 | (address & PAGE_MASK) + PAGE_SIZE); | |
ac46d4f3 | 4927 | mmu_notifier_invalidate_range_start(range); |
a4d1a885 | 4928 | } |
f8ad0f49 | 4929 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
f8ad0f49 JW |
4930 | if (!pte_present(*ptep)) |
4931 | goto unlock; | |
4932 | *ptepp = ptep; | |
4933 | return 0; | |
4934 | unlock: | |
4935 | pte_unmap_unlock(ptep, *ptlp); | |
ac46d4f3 JG |
4936 | if (range) |
4937 | mmu_notifier_invalidate_range_end(range); | |
f8ad0f49 JW |
4938 | out: |
4939 | return -EINVAL; | |
4940 | } | |
4941 | ||
9fd6dad1 PB |
4942 | /** |
4943 | * follow_pte - look up PTE at a user virtual address | |
4944 | * @mm: the mm_struct of the target address space | |
4945 | * @address: user virtual address | |
4946 | * @ptepp: location to store found PTE | |
4947 | * @ptlp: location to store the lock for the PTE | |
4948 | * | |
4949 | * On a successful return, the pointer to the PTE is stored in @ptepp; | |
4950 | * the corresponding lock is taken and its location is stored in @ptlp. | |
4951 | * The contents of the PTE are only stable until @ptlp is released; | |
4952 | * any further use, if any, must be protected against invalidation | |
4953 | * with MMU notifiers. | |
4954 | * | |
4955 | * Only IO mappings and raw PFN mappings are allowed. The mmap semaphore | |
4956 | * should be taken for read. | |
4957 | * | |
4958 | * KVM uses this function. While it is arguably less bad than ``follow_pfn``, | |
4959 | * it is not a good general-purpose API. | |
4960 | * | |
4961 | * Return: zero on success, -ve otherwise. | |
4962 | */ | |
4963 | int follow_pte(struct mm_struct *mm, unsigned long address, | |
4964 | pte_t **ptepp, spinlock_t **ptlp) | |
4965 | { | |
4966 | return follow_invalidate_pte(mm, address, NULL, ptepp, NULL, ptlp); | |
4967 | } | |
4968 | EXPORT_SYMBOL_GPL(follow_pte); | |
4969 | ||
3b6748e2 JW |
4970 | /** |
4971 | * follow_pfn - look up PFN at a user virtual address | |
4972 | * @vma: memory mapping | |
4973 | * @address: user virtual address | |
4974 | * @pfn: location to store found PFN | |
4975 | * | |
4976 | * Only IO mappings and raw PFN mappings are allowed. | |
4977 | * | |
9fd6dad1 PB |
4978 | * This function does not allow the caller to read the permissions |
4979 | * of the PTE. Do not use it. | |
4980 | * | |
a862f68a | 4981 | * Return: zero and the pfn at @pfn on success, -ve otherwise. |
3b6748e2 JW |
4982 | */ |
4983 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
4984 | unsigned long *pfn) | |
4985 | { | |
4986 | int ret = -EINVAL; | |
4987 | spinlock_t *ptl; | |
4988 | pte_t *ptep; | |
4989 | ||
4990 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
4991 | return ret; | |
4992 | ||
9fd6dad1 | 4993 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); |
3b6748e2 JW |
4994 | if (ret) |
4995 | return ret; | |
4996 | *pfn = pte_pfn(*ptep); | |
4997 | pte_unmap_unlock(ptep, ptl); | |
4998 | return 0; | |
4999 | } | |
5000 | EXPORT_SYMBOL(follow_pfn); | |
5001 | ||
28b2ee20 | 5002 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 5003 | int follow_phys(struct vm_area_struct *vma, |
5004 | unsigned long address, unsigned int flags, | |
5005 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 5006 | { |
03668a4d | 5007 | int ret = -EINVAL; |
28b2ee20 RR |
5008 | pte_t *ptep, pte; |
5009 | spinlock_t *ptl; | |
28b2ee20 | 5010 | |
d87fe660 | 5011 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
5012 | goto out; | |
28b2ee20 | 5013 | |
9fd6dad1 | 5014 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 5015 | goto out; |
28b2ee20 | 5016 | pte = *ptep; |
03668a4d | 5017 | |
f6f37321 | 5018 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
28b2ee20 | 5019 | goto unlock; |
28b2ee20 RR |
5020 | |
5021 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 5022 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 5023 | |
03668a4d | 5024 | ret = 0; |
28b2ee20 RR |
5025 | unlock: |
5026 | pte_unmap_unlock(ptep, ptl); | |
5027 | out: | |
d87fe660 | 5028 | return ret; |
28b2ee20 RR |
5029 | } |
5030 | ||
96667f8a DV |
5031 | /** |
5032 | * generic_access_phys - generic implementation for iomem mmap access | |
5033 | * @vma: the vma to access | |
f0953a1b | 5034 | * @addr: userspace address, not relative offset within @vma |
96667f8a DV |
5035 | * @buf: buffer to read/write |
5036 | * @len: length of transfer | |
5037 | * @write: set to FOLL_WRITE when writing, otherwise reading | |
5038 | * | |
5039 | * This is a generic implementation for &vm_operations_struct.access for an | |
5040 | * iomem mapping. This callback is used by access_process_vm() when the @vma is | |
5041 | * not page based. | |
5042 | */ | |
28b2ee20 RR |
5043 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, |
5044 | void *buf, int len, int write) | |
5045 | { | |
5046 | resource_size_t phys_addr; | |
5047 | unsigned long prot = 0; | |
2bc7273b | 5048 | void __iomem *maddr; |
96667f8a DV |
5049 | pte_t *ptep, pte; |
5050 | spinlock_t *ptl; | |
5051 | int offset = offset_in_page(addr); | |
5052 | int ret = -EINVAL; | |
5053 | ||
5054 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
5055 | return -EINVAL; | |
5056 | ||
5057 | retry: | |
e913a8cd | 5058 | if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) |
96667f8a DV |
5059 | return -EINVAL; |
5060 | pte = *ptep; | |
5061 | pte_unmap_unlock(ptep, ptl); | |
28b2ee20 | 5062 | |
96667f8a DV |
5063 | prot = pgprot_val(pte_pgprot(pte)); |
5064 | phys_addr = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; | |
5065 | ||
5066 | if ((write & FOLL_WRITE) && !pte_write(pte)) | |
28b2ee20 RR |
5067 | return -EINVAL; |
5068 | ||
9cb12d7b | 5069 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
24eee1e4 | 5070 | if (!maddr) |
5071 | return -ENOMEM; | |
5072 | ||
e913a8cd | 5073 | if (follow_pte(vma->vm_mm, addr, &ptep, &ptl)) |
96667f8a DV |
5074 | goto out_unmap; |
5075 | ||
5076 | if (!pte_same(pte, *ptep)) { | |
5077 | pte_unmap_unlock(ptep, ptl); | |
5078 | iounmap(maddr); | |
5079 | ||
5080 | goto retry; | |
5081 | } | |
5082 | ||
28b2ee20 RR |
5083 | if (write) |
5084 | memcpy_toio(maddr + offset, buf, len); | |
5085 | else | |
5086 | memcpy_fromio(buf, maddr + offset, len); | |
96667f8a DV |
5087 | ret = len; |
5088 | pte_unmap_unlock(ptep, ptl); | |
5089 | out_unmap: | |
28b2ee20 RR |
5090 | iounmap(maddr); |
5091 | ||
96667f8a | 5092 | return ret; |
28b2ee20 | 5093 | } |
5a73633e | 5094 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
5095 | #endif |
5096 | ||
0ec76a11 | 5097 | /* |
d3f5ffca | 5098 | * Access another process' address space as given in mm. |
0ec76a11 | 5099 | */ |
d3f5ffca JH |
5100 | int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf, |
5101 | int len, unsigned int gup_flags) | |
0ec76a11 | 5102 | { |
0ec76a11 | 5103 | struct vm_area_struct *vma; |
0ec76a11 | 5104 | void *old_buf = buf; |
442486ec | 5105 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 5106 | |
d8ed45c5 | 5107 | if (mmap_read_lock_killable(mm)) |
1e426fe2 KK |
5108 | return 0; |
5109 | ||
183ff22b | 5110 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
5111 | while (len) { |
5112 | int bytes, ret, offset; | |
5113 | void *maddr; | |
28b2ee20 | 5114 | struct page *page = NULL; |
0ec76a11 | 5115 | |
64019a2e | 5116 | ret = get_user_pages_remote(mm, addr, 1, |
5b56d49f | 5117 | gup_flags, &page, &vma, NULL); |
28b2ee20 | 5118 | if (ret <= 0) { |
dbffcd03 RR |
5119 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
5120 | break; | |
5121 | #else | |
28b2ee20 RR |
5122 | /* |
5123 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
5124 | * we can access using slightly different code. | |
5125 | */ | |
3e418f98 LH |
5126 | vma = vma_lookup(mm, addr); |
5127 | if (!vma) | |
28b2ee20 RR |
5128 | break; |
5129 | if (vma->vm_ops && vma->vm_ops->access) | |
5130 | ret = vma->vm_ops->access(vma, addr, buf, | |
5131 | len, write); | |
5132 | if (ret <= 0) | |
28b2ee20 RR |
5133 | break; |
5134 | bytes = ret; | |
dbffcd03 | 5135 | #endif |
0ec76a11 | 5136 | } else { |
28b2ee20 RR |
5137 | bytes = len; |
5138 | offset = addr & (PAGE_SIZE-1); | |
5139 | if (bytes > PAGE_SIZE-offset) | |
5140 | bytes = PAGE_SIZE-offset; | |
5141 | ||
5142 | maddr = kmap(page); | |
5143 | if (write) { | |
5144 | copy_to_user_page(vma, page, addr, | |
5145 | maddr + offset, buf, bytes); | |
5146 | set_page_dirty_lock(page); | |
5147 | } else { | |
5148 | copy_from_user_page(vma, page, addr, | |
5149 | buf, maddr + offset, bytes); | |
5150 | } | |
5151 | kunmap(page); | |
09cbfeaf | 5152 | put_page(page); |
0ec76a11 | 5153 | } |
0ec76a11 DH |
5154 | len -= bytes; |
5155 | buf += bytes; | |
5156 | addr += bytes; | |
5157 | } | |
d8ed45c5 | 5158 | mmap_read_unlock(mm); |
0ec76a11 DH |
5159 | |
5160 | return buf - old_buf; | |
5161 | } | |
03252919 | 5162 | |
5ddd36b9 | 5163 | /** |
ae91dbfc | 5164 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
5165 | * @mm: the mm_struct of the target address space |
5166 | * @addr: start address to access | |
5167 | * @buf: source or destination buffer | |
5168 | * @len: number of bytes to transfer | |
6347e8d5 | 5169 | * @gup_flags: flags modifying lookup behaviour |
5ddd36b9 SW |
5170 | * |
5171 | * The caller must hold a reference on @mm. | |
a862f68a MR |
5172 | * |
5173 | * Return: number of bytes copied from source to destination. | |
5ddd36b9 SW |
5174 | */ |
5175 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 5176 | void *buf, int len, unsigned int gup_flags) |
5ddd36b9 | 5177 | { |
d3f5ffca | 5178 | return __access_remote_vm(mm, addr, buf, len, gup_flags); |
5ddd36b9 SW |
5179 | } |
5180 | ||
206cb636 SW |
5181 | /* |
5182 | * Access another process' address space. | |
5183 | * Source/target buffer must be kernel space, | |
5184 | * Do not walk the page table directly, use get_user_pages | |
5185 | */ | |
5186 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
f307ab6d | 5187 | void *buf, int len, unsigned int gup_flags) |
206cb636 SW |
5188 | { |
5189 | struct mm_struct *mm; | |
5190 | int ret; | |
5191 | ||
5192 | mm = get_task_mm(tsk); | |
5193 | if (!mm) | |
5194 | return 0; | |
5195 | ||
d3f5ffca | 5196 | ret = __access_remote_vm(mm, addr, buf, len, gup_flags); |
442486ec | 5197 | |
206cb636 SW |
5198 | mmput(mm); |
5199 | ||
5200 | return ret; | |
5201 | } | |
fcd35857 | 5202 | EXPORT_SYMBOL_GPL(access_process_vm); |
206cb636 | 5203 | |
03252919 AK |
5204 | /* |
5205 | * Print the name of a VMA. | |
5206 | */ | |
5207 | void print_vma_addr(char *prefix, unsigned long ip) | |
5208 | { | |
5209 | struct mm_struct *mm = current->mm; | |
5210 | struct vm_area_struct *vma; | |
5211 | ||
e8bff74a | 5212 | /* |
0a7f682d | 5213 | * we might be running from an atomic context so we cannot sleep |
e8bff74a | 5214 | */ |
d8ed45c5 | 5215 | if (!mmap_read_trylock(mm)) |
e8bff74a IM |
5216 | return; |
5217 | ||
03252919 AK |
5218 | vma = find_vma(mm, ip); |
5219 | if (vma && vma->vm_file) { | |
5220 | struct file *f = vma->vm_file; | |
0a7f682d | 5221 | char *buf = (char *)__get_free_page(GFP_NOWAIT); |
03252919 | 5222 | if (buf) { |
2fbc57c5 | 5223 | char *p; |
03252919 | 5224 | |
9bf39ab2 | 5225 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
5226 | if (IS_ERR(p)) |
5227 | p = "?"; | |
2fbc57c5 | 5228 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
5229 | vma->vm_start, |
5230 | vma->vm_end - vma->vm_start); | |
5231 | free_page((unsigned long)buf); | |
5232 | } | |
5233 | } | |
d8ed45c5 | 5234 | mmap_read_unlock(mm); |
03252919 | 5235 | } |
3ee1afa3 | 5236 | |
662bbcb2 | 5237 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 5238 | void __might_fault(const char *file, int line) |
3ee1afa3 | 5239 | { |
95156f00 PZ |
5240 | /* |
5241 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
c1e8d7c6 | 5242 | * holding the mmap_lock, this is safe because kernel memory doesn't |
95156f00 PZ |
5243 | * get paged out, therefore we'll never actually fault, and the |
5244 | * below annotations will generate false positives. | |
5245 | */ | |
db68ce10 | 5246 | if (uaccess_kernel()) |
95156f00 | 5247 | return; |
9ec23531 | 5248 | if (pagefault_disabled()) |
662bbcb2 | 5249 | return; |
42a38756 | 5250 | __might_sleep(file, line); |
9ec23531 | 5251 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
662bbcb2 | 5252 | if (current->mm) |
da1c55f1 | 5253 | might_lock_read(¤t->mm->mmap_lock); |
9ec23531 | 5254 | #endif |
3ee1afa3 | 5255 | } |
9ec23531 | 5256 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 5257 | #endif |
47ad8475 AA |
5258 | |
5259 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
c6ddfb6c HY |
5260 | /* |
5261 | * Process all subpages of the specified huge page with the specified | |
5262 | * operation. The target subpage will be processed last to keep its | |
5263 | * cache lines hot. | |
5264 | */ | |
5265 | static inline void process_huge_page( | |
5266 | unsigned long addr_hint, unsigned int pages_per_huge_page, | |
5267 | void (*process_subpage)(unsigned long addr, int idx, void *arg), | |
5268 | void *arg) | |
47ad8475 | 5269 | { |
c79b57e4 HY |
5270 | int i, n, base, l; |
5271 | unsigned long addr = addr_hint & | |
5272 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
47ad8475 | 5273 | |
c6ddfb6c | 5274 | /* Process target subpage last to keep its cache lines hot */ |
47ad8475 | 5275 | might_sleep(); |
c79b57e4 HY |
5276 | n = (addr_hint - addr) / PAGE_SIZE; |
5277 | if (2 * n <= pages_per_huge_page) { | |
c6ddfb6c | 5278 | /* If target subpage in first half of huge page */ |
c79b57e4 HY |
5279 | base = 0; |
5280 | l = n; | |
c6ddfb6c | 5281 | /* Process subpages at the end of huge page */ |
c79b57e4 HY |
5282 | for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { |
5283 | cond_resched(); | |
c6ddfb6c | 5284 | process_subpage(addr + i * PAGE_SIZE, i, arg); |
c79b57e4 HY |
5285 | } |
5286 | } else { | |
c6ddfb6c | 5287 | /* If target subpage in second half of huge page */ |
c79b57e4 HY |
5288 | base = pages_per_huge_page - 2 * (pages_per_huge_page - n); |
5289 | l = pages_per_huge_page - n; | |
c6ddfb6c | 5290 | /* Process subpages at the begin of huge page */ |
c79b57e4 HY |
5291 | for (i = 0; i < base; i++) { |
5292 | cond_resched(); | |
c6ddfb6c | 5293 | process_subpage(addr + i * PAGE_SIZE, i, arg); |
c79b57e4 HY |
5294 | } |
5295 | } | |
5296 | /* | |
c6ddfb6c HY |
5297 | * Process remaining subpages in left-right-left-right pattern |
5298 | * towards the target subpage | |
c79b57e4 HY |
5299 | */ |
5300 | for (i = 0; i < l; i++) { | |
5301 | int left_idx = base + i; | |
5302 | int right_idx = base + 2 * l - 1 - i; | |
5303 | ||
5304 | cond_resched(); | |
c6ddfb6c | 5305 | process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg); |
47ad8475 | 5306 | cond_resched(); |
c6ddfb6c | 5307 | process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg); |
47ad8475 AA |
5308 | } |
5309 | } | |
5310 | ||
c6ddfb6c HY |
5311 | static void clear_gigantic_page(struct page *page, |
5312 | unsigned long addr, | |
5313 | unsigned int pages_per_huge_page) | |
5314 | { | |
5315 | int i; | |
5316 | struct page *p = page; | |
5317 | ||
5318 | might_sleep(); | |
5319 | for (i = 0; i < pages_per_huge_page; | |
5320 | i++, p = mem_map_next(p, page, i)) { | |
5321 | cond_resched(); | |
5322 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
5323 | } | |
5324 | } | |
5325 | ||
5326 | static void clear_subpage(unsigned long addr, int idx, void *arg) | |
5327 | { | |
5328 | struct page *page = arg; | |
5329 | ||
5330 | clear_user_highpage(page + idx, addr); | |
5331 | } | |
5332 | ||
5333 | void clear_huge_page(struct page *page, | |
5334 | unsigned long addr_hint, unsigned int pages_per_huge_page) | |
5335 | { | |
5336 | unsigned long addr = addr_hint & | |
5337 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
5338 | ||
5339 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
5340 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
5341 | return; | |
5342 | } | |
5343 | ||
5344 | process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page); | |
5345 | } | |
5346 | ||
47ad8475 AA |
5347 | static void copy_user_gigantic_page(struct page *dst, struct page *src, |
5348 | unsigned long addr, | |
5349 | struct vm_area_struct *vma, | |
5350 | unsigned int pages_per_huge_page) | |
5351 | { | |
5352 | int i; | |
5353 | struct page *dst_base = dst; | |
5354 | struct page *src_base = src; | |
5355 | ||
5356 | for (i = 0; i < pages_per_huge_page; ) { | |
5357 | cond_resched(); | |
5358 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
5359 | ||
5360 | i++; | |
5361 | dst = mem_map_next(dst, dst_base, i); | |
5362 | src = mem_map_next(src, src_base, i); | |
5363 | } | |
5364 | } | |
5365 | ||
c9f4cd71 HY |
5366 | struct copy_subpage_arg { |
5367 | struct page *dst; | |
5368 | struct page *src; | |
5369 | struct vm_area_struct *vma; | |
5370 | }; | |
5371 | ||
5372 | static void copy_subpage(unsigned long addr, int idx, void *arg) | |
5373 | { | |
5374 | struct copy_subpage_arg *copy_arg = arg; | |
5375 | ||
5376 | copy_user_highpage(copy_arg->dst + idx, copy_arg->src + idx, | |
5377 | addr, copy_arg->vma); | |
5378 | } | |
5379 | ||
47ad8475 | 5380 | void copy_user_huge_page(struct page *dst, struct page *src, |
c9f4cd71 | 5381 | unsigned long addr_hint, struct vm_area_struct *vma, |
47ad8475 AA |
5382 | unsigned int pages_per_huge_page) |
5383 | { | |
c9f4cd71 HY |
5384 | unsigned long addr = addr_hint & |
5385 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
5386 | struct copy_subpage_arg arg = { | |
5387 | .dst = dst, | |
5388 | .src = src, | |
5389 | .vma = vma, | |
5390 | }; | |
47ad8475 AA |
5391 | |
5392 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
5393 | copy_user_gigantic_page(dst, src, addr, vma, | |
5394 | pages_per_huge_page); | |
5395 | return; | |
5396 | } | |
5397 | ||
c9f4cd71 | 5398 | process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg); |
47ad8475 | 5399 | } |
fa4d75c1 MK |
5400 | |
5401 | long copy_huge_page_from_user(struct page *dst_page, | |
5402 | const void __user *usr_src, | |
810a56b9 MK |
5403 | unsigned int pages_per_huge_page, |
5404 | bool allow_pagefault) | |
fa4d75c1 | 5405 | { |
fa4d75c1 MK |
5406 | void *page_kaddr; |
5407 | unsigned long i, rc = 0; | |
5408 | unsigned long ret_val = pages_per_huge_page * PAGE_SIZE; | |
3272cfc2 | 5409 | struct page *subpage = dst_page; |
fa4d75c1 | 5410 | |
3272cfc2 MK |
5411 | for (i = 0; i < pages_per_huge_page; |
5412 | i++, subpage = mem_map_next(subpage, dst_page, i)) { | |
810a56b9 | 5413 | if (allow_pagefault) |
3272cfc2 | 5414 | page_kaddr = kmap(subpage); |
810a56b9 | 5415 | else |
3272cfc2 | 5416 | page_kaddr = kmap_atomic(subpage); |
fa4d75c1 | 5417 | rc = copy_from_user(page_kaddr, |
b063e374 | 5418 | usr_src + i * PAGE_SIZE, PAGE_SIZE); |
810a56b9 | 5419 | if (allow_pagefault) |
3272cfc2 | 5420 | kunmap(subpage); |
810a56b9 MK |
5421 | else |
5422 | kunmap_atomic(page_kaddr); | |
fa4d75c1 MK |
5423 | |
5424 | ret_val -= (PAGE_SIZE - rc); | |
5425 | if (rc) | |
5426 | break; | |
5427 | ||
5428 | cond_resched(); | |
5429 | } | |
5430 | return ret_val; | |
5431 | } | |
47ad8475 | 5432 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
49076ec2 | 5433 | |
40b64acd | 5434 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
5435 | |
5436 | static struct kmem_cache *page_ptl_cachep; | |
5437 | ||
5438 | void __init ptlock_cache_init(void) | |
5439 | { | |
5440 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
5441 | SLAB_PANIC, NULL); | |
5442 | } | |
5443 | ||
539edb58 | 5444 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
5445 | { |
5446 | spinlock_t *ptl; | |
5447 | ||
b35f1819 | 5448 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
5449 | if (!ptl) |
5450 | return false; | |
539edb58 | 5451 | page->ptl = ptl; |
49076ec2 KS |
5452 | return true; |
5453 | } | |
5454 | ||
539edb58 | 5455 | void ptlock_free(struct page *page) |
49076ec2 | 5456 | { |
b35f1819 | 5457 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
5458 | } |
5459 | #endif |