Commit | Line | Data |
---|---|---|
d61ea1cb | 1 | |
457c8996 | 2 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
3 | /* |
4 | * linux/mm/memory.c | |
5 | * | |
6 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
7 | */ | |
8 | ||
9 | /* | |
10 | * demand-loading started 01.12.91 - seems it is high on the list of | |
11 | * things wanted, and it should be easy to implement. - Linus | |
12 | */ | |
13 | ||
14 | /* | |
15 | * Ok, demand-loading was easy, shared pages a little bit tricker. Shared | |
16 | * pages started 02.12.91, seems to work. - Linus. | |
17 | * | |
18 | * Tested sharing by executing about 30 /bin/sh: under the old kernel it | |
19 | * would have taken more than the 6M I have free, but it worked well as | |
20 | * far as I could see. | |
21 | * | |
22 | * Also corrected some "invalidate()"s - I wasn't doing enough of them. | |
23 | */ | |
24 | ||
25 | /* | |
26 | * Real VM (paging to/from disk) started 18.12.91. Much more work and | |
27 | * thought has to go into this. Oh, well.. | |
28 | * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. | |
29 | * Found it. Everything seems to work now. | |
30 | * 20.12.91 - Ok, making the swap-device changeable like the root. | |
31 | */ | |
32 | ||
33 | /* | |
34 | * 05.04.94 - Multi-page memory management added for v1.1. | |
166f61b9 | 35 | * Idea by Alex Bligh (alex@cconcepts.co.uk) |
1da177e4 LT |
36 | * |
37 | * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG | |
38 | * (Gerhard.Wichert@pdb.siemens.de) | |
39 | * | |
40 | * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) | |
41 | */ | |
42 | ||
43 | #include <linux/kernel_stat.h> | |
44 | #include <linux/mm.h> | |
36090def | 45 | #include <linux/mm_inline.h> |
6e84f315 | 46 | #include <linux/sched/mm.h> |
f7ccbae4 | 47 | #include <linux/sched/coredump.h> |
6a3827d7 | 48 | #include <linux/sched/numa_balancing.h> |
29930025 | 49 | #include <linux/sched/task.h> |
1da177e4 LT |
50 | #include <linux/hugetlb.h> |
51 | #include <linux/mman.h> | |
52 | #include <linux/swap.h> | |
53 | #include <linux/highmem.h> | |
54 | #include <linux/pagemap.h> | |
5042db43 | 55 | #include <linux/memremap.h> |
b073d7f8 | 56 | #include <linux/kmsan.h> |
9a840895 | 57 | #include <linux/ksm.h> |
1da177e4 | 58 | #include <linux/rmap.h> |
b95f1b31 | 59 | #include <linux/export.h> |
0ff92245 | 60 | #include <linux/delayacct.h> |
1da177e4 | 61 | #include <linux/init.h> |
01c8f1c4 | 62 | #include <linux/pfn_t.h> |
edc79b2a | 63 | #include <linux/writeback.h> |
8a9f3ccd | 64 | #include <linux/memcontrol.h> |
cddb8a5c | 65 | #include <linux/mmu_notifier.h> |
3dc14741 HD |
66 | #include <linux/swapops.h> |
67 | #include <linux/elf.h> | |
5a0e3ad6 | 68 | #include <linux/gfp.h> |
4daae3b4 | 69 | #include <linux/migrate.h> |
2fbc57c5 | 70 | #include <linux/string.h> |
467b171a | 71 | #include <linux/memory-tiers.h> |
1592eef0 | 72 | #include <linux/debugfs.h> |
6b251fc9 | 73 | #include <linux/userfaultfd_k.h> |
bc2466e4 | 74 | #include <linux/dax.h> |
6b31d595 | 75 | #include <linux/oom.h> |
98fa15f3 | 76 | #include <linux/numa.h> |
bce617ed PX |
77 | #include <linux/perf_event.h> |
78 | #include <linux/ptrace.h> | |
e80d3909 | 79 | #include <linux/vmalloc.h> |
33024536 | 80 | #include <linux/sched/sysctl.h> |
1da177e4 | 81 | |
b3d1411b JFG |
82 | #include <trace/events/kmem.h> |
83 | ||
6952b61d | 84 | #include <asm/io.h> |
33a709b2 | 85 | #include <asm/mmu_context.h> |
1da177e4 | 86 | #include <asm/pgalloc.h> |
7c0f6ba6 | 87 | #include <linux/uaccess.h> |
1da177e4 LT |
88 | #include <asm/tlb.h> |
89 | #include <asm/tlbflush.h> | |
1da177e4 | 90 | |
e80d3909 | 91 | #include "pgalloc-track.h" |
42b77728 | 92 | #include "internal.h" |
014bb1de | 93 | #include "swap.h" |
42b77728 | 94 | |
af27d940 | 95 | #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST) |
90572890 | 96 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. |
75980e97 PZ |
97 | #endif |
98 | ||
a9ee6cf5 | 99 | #ifndef CONFIG_NUMA |
1da177e4 | 100 | unsigned long max_mapnr; |
1da177e4 | 101 | EXPORT_SYMBOL(max_mapnr); |
166f61b9 TH |
102 | |
103 | struct page *mem_map; | |
1da177e4 LT |
104 | EXPORT_SYMBOL(mem_map); |
105 | #endif | |
106 | ||
5c041f5d | 107 | static vm_fault_t do_fault(struct vm_fault *vmf); |
2bad466c PX |
108 | static vm_fault_t do_anonymous_page(struct vm_fault *vmf); |
109 | static bool vmf_pte_changed(struct vm_fault *vmf); | |
110 | ||
111 | /* | |
112 | * Return true if the original pte was a uffd-wp pte marker (so the pte was | |
113 | * wr-protected). | |
114 | */ | |
6ed31ba3 | 115 | static __always_inline bool vmf_orig_pte_uffd_wp(struct vm_fault *vmf) |
2bad466c | 116 | { |
6ed31ba3 KW |
117 | if (!userfaultfd_wp(vmf->vma)) |
118 | return false; | |
2bad466c PX |
119 | if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID)) |
120 | return false; | |
121 | ||
122 | return pte_marker_uffd_wp(vmf->orig_pte); | |
123 | } | |
5c041f5d | 124 | |
1da177e4 LT |
125 | /* |
126 | * A number of key systems in x86 including ioremap() rely on the assumption | |
127 | * that high_memory defines the upper bound on direct map memory, then end | |
e99fb98d | 128 | * of ZONE_NORMAL. |
1da177e4 | 129 | */ |
166f61b9 | 130 | void *high_memory; |
1da177e4 | 131 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 132 | |
32a93233 IM |
133 | /* |
134 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
135 | * | |
136 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
137 | * as ancient (libc5 based) binaries can segfault. ) | |
138 | */ | |
139 | int randomize_va_space __read_mostly = | |
140 | #ifdef CONFIG_COMPAT_BRK | |
141 | 1; | |
142 | #else | |
143 | 2; | |
144 | #endif | |
a62eaf15 | 145 | |
46bdb427 WD |
146 | #ifndef arch_wants_old_prefaulted_pte |
147 | static inline bool arch_wants_old_prefaulted_pte(void) | |
148 | { | |
149 | /* | |
150 | * Transitioning a PTE from 'old' to 'young' can be expensive on | |
151 | * some architectures, even if it's performed in hardware. By | |
152 | * default, "false" means prefaulted entries will be 'young'. | |
153 | */ | |
154 | return false; | |
155 | } | |
156 | #endif | |
157 | ||
a62eaf15 AK |
158 | static int __init disable_randmaps(char *s) |
159 | { | |
160 | randomize_va_space = 0; | |
9b41046c | 161 | return 1; |
a62eaf15 AK |
162 | } |
163 | __setup("norandmaps", disable_randmaps); | |
164 | ||
62eede62 | 165 | unsigned long zero_pfn __read_mostly; |
0b70068e AB |
166 | EXPORT_SYMBOL(zero_pfn); |
167 | ||
166f61b9 TH |
168 | unsigned long highest_memmap_pfn __read_mostly; |
169 | ||
a13ea5b7 HD |
170 | /* |
171 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
172 | */ | |
173 | static int __init init_zero_pfn(void) | |
174 | { | |
175 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
176 | return 0; | |
177 | } | |
e720e7d0 | 178 | early_initcall(init_zero_pfn); |
a62eaf15 | 179 | |
f1a79412 | 180 | void mm_trace_rss_stat(struct mm_struct *mm, int member) |
b3d1411b | 181 | { |
f1a79412 | 182 | trace_rss_stat(mm, member); |
b3d1411b | 183 | } |
d559db08 | 184 | |
1da177e4 LT |
185 | /* |
186 | * Note: this doesn't free the actual pages themselves. That | |
187 | * has been handled earlier when unmapping all the memory regions. | |
188 | */ | |
9e1b32ca BH |
189 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
190 | unsigned long addr) | |
1da177e4 | 191 | { |
2f569afd | 192 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 193 | pmd_clear(pmd); |
9e1b32ca | 194 | pte_free_tlb(tlb, token, addr); |
c4812909 | 195 | mm_dec_nr_ptes(tlb->mm); |
1da177e4 LT |
196 | } |
197 | ||
e0da382c HD |
198 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
199 | unsigned long addr, unsigned long end, | |
200 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
201 | { |
202 | pmd_t *pmd; | |
203 | unsigned long next; | |
e0da382c | 204 | unsigned long start; |
1da177e4 | 205 | |
e0da382c | 206 | start = addr; |
1da177e4 | 207 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
208 | do { |
209 | next = pmd_addr_end(addr, end); | |
210 | if (pmd_none_or_clear_bad(pmd)) | |
211 | continue; | |
9e1b32ca | 212 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
213 | } while (pmd++, addr = next, addr != end); |
214 | ||
e0da382c HD |
215 | start &= PUD_MASK; |
216 | if (start < floor) | |
217 | return; | |
218 | if (ceiling) { | |
219 | ceiling &= PUD_MASK; | |
220 | if (!ceiling) | |
221 | return; | |
1da177e4 | 222 | } |
e0da382c HD |
223 | if (end - 1 > ceiling - 1) |
224 | return; | |
225 | ||
226 | pmd = pmd_offset(pud, start); | |
227 | pud_clear(pud); | |
9e1b32ca | 228 | pmd_free_tlb(tlb, pmd, start); |
dc6c9a35 | 229 | mm_dec_nr_pmds(tlb->mm); |
1da177e4 LT |
230 | } |
231 | ||
c2febafc | 232 | static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, |
e0da382c HD |
233 | unsigned long addr, unsigned long end, |
234 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
235 | { |
236 | pud_t *pud; | |
237 | unsigned long next; | |
e0da382c | 238 | unsigned long start; |
1da177e4 | 239 | |
e0da382c | 240 | start = addr; |
c2febafc | 241 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
242 | do { |
243 | next = pud_addr_end(addr, end); | |
244 | if (pud_none_or_clear_bad(pud)) | |
245 | continue; | |
e0da382c | 246 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
247 | } while (pud++, addr = next, addr != end); |
248 | ||
c2febafc KS |
249 | start &= P4D_MASK; |
250 | if (start < floor) | |
251 | return; | |
252 | if (ceiling) { | |
253 | ceiling &= P4D_MASK; | |
254 | if (!ceiling) | |
255 | return; | |
256 | } | |
257 | if (end - 1 > ceiling - 1) | |
258 | return; | |
259 | ||
260 | pud = pud_offset(p4d, start); | |
261 | p4d_clear(p4d); | |
262 | pud_free_tlb(tlb, pud, start); | |
b4e98d9a | 263 | mm_dec_nr_puds(tlb->mm); |
c2febafc KS |
264 | } |
265 | ||
266 | static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, | |
267 | unsigned long addr, unsigned long end, | |
268 | unsigned long floor, unsigned long ceiling) | |
269 | { | |
270 | p4d_t *p4d; | |
271 | unsigned long next; | |
272 | unsigned long start; | |
273 | ||
274 | start = addr; | |
275 | p4d = p4d_offset(pgd, addr); | |
276 | do { | |
277 | next = p4d_addr_end(addr, end); | |
278 | if (p4d_none_or_clear_bad(p4d)) | |
279 | continue; | |
280 | free_pud_range(tlb, p4d, addr, next, floor, ceiling); | |
281 | } while (p4d++, addr = next, addr != end); | |
282 | ||
e0da382c HD |
283 | start &= PGDIR_MASK; |
284 | if (start < floor) | |
285 | return; | |
286 | if (ceiling) { | |
287 | ceiling &= PGDIR_MASK; | |
288 | if (!ceiling) | |
289 | return; | |
1da177e4 | 290 | } |
e0da382c HD |
291 | if (end - 1 > ceiling - 1) |
292 | return; | |
293 | ||
c2febafc | 294 | p4d = p4d_offset(pgd, start); |
e0da382c | 295 | pgd_clear(pgd); |
c2febafc | 296 | p4d_free_tlb(tlb, p4d, start); |
1da177e4 LT |
297 | } |
298 | ||
299 | /* | |
e0da382c | 300 | * This function frees user-level page tables of a process. |
1da177e4 | 301 | */ |
42b77728 | 302 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
303 | unsigned long addr, unsigned long end, |
304 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
305 | { |
306 | pgd_t *pgd; | |
307 | unsigned long next; | |
e0da382c HD |
308 | |
309 | /* | |
310 | * The next few lines have given us lots of grief... | |
311 | * | |
312 | * Why are we testing PMD* at this top level? Because often | |
313 | * there will be no work to do at all, and we'd prefer not to | |
314 | * go all the way down to the bottom just to discover that. | |
315 | * | |
316 | * Why all these "- 1"s? Because 0 represents both the bottom | |
317 | * of the address space and the top of it (using -1 for the | |
318 | * top wouldn't help much: the masks would do the wrong thing). | |
319 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
320 | * the address space, but end 0 and ceiling 0 refer to the top | |
321 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
322 | * that end 0 case should be mythical). | |
323 | * | |
324 | * Wherever addr is brought up or ceiling brought down, we must | |
325 | * be careful to reject "the opposite 0" before it confuses the | |
326 | * subsequent tests. But what about where end is brought down | |
327 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
328 | * | |
329 | * Whereas we round start (addr) and ceiling down, by different | |
330 | * masks at different levels, in order to test whether a table | |
331 | * now has no other vmas using it, so can be freed, we don't | |
332 | * bother to round floor or end up - the tests don't need that. | |
333 | */ | |
1da177e4 | 334 | |
e0da382c HD |
335 | addr &= PMD_MASK; |
336 | if (addr < floor) { | |
337 | addr += PMD_SIZE; | |
338 | if (!addr) | |
339 | return; | |
340 | } | |
341 | if (ceiling) { | |
342 | ceiling &= PMD_MASK; | |
343 | if (!ceiling) | |
344 | return; | |
345 | } | |
346 | if (end - 1 > ceiling - 1) | |
347 | end -= PMD_SIZE; | |
348 | if (addr > end - 1) | |
349 | return; | |
07e32661 AK |
350 | /* |
351 | * We add page table cache pages with PAGE_SIZE, | |
352 | * (see pte_free_tlb()), flush the tlb if we need | |
353 | */ | |
ed6a7935 | 354 | tlb_change_page_size(tlb, PAGE_SIZE); |
42b77728 | 355 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
356 | do { |
357 | next = pgd_addr_end(addr, end); | |
358 | if (pgd_none_or_clear_bad(pgd)) | |
359 | continue; | |
c2febafc | 360 | free_p4d_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 361 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
362 | } |
363 | ||
fd892593 | 364 | void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas, |
763ecb03 | 365 | struct vm_area_struct *vma, unsigned long floor, |
98e51a22 | 366 | unsigned long ceiling, bool mm_wr_locked) |
e0da382c | 367 | { |
763ecb03 | 368 | do { |
e0da382c | 369 | unsigned long addr = vma->vm_start; |
763ecb03 LH |
370 | struct vm_area_struct *next; |
371 | ||
372 | /* | |
373 | * Note: USER_PGTABLES_CEILING may be passed as ceiling and may | |
374 | * be 0. This will underflow and is okay. | |
375 | */ | |
fd892593 | 376 | next = mas_find(mas, ceiling - 1); |
d2406291 PZ |
377 | if (unlikely(xa_is_zero(next))) |
378 | next = NULL; | |
e0da382c | 379 | |
8f4f8c16 | 380 | /* |
25d9e2d1 | 381 | * Hide vma from rmap and truncate_pagecache before freeing |
382 | * pgtables | |
8f4f8c16 | 383 | */ |
98e51a22 SB |
384 | if (mm_wr_locked) |
385 | vma_start_write(vma); | |
5beb4930 | 386 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
387 | unlink_file_vma(vma); |
388 | ||
9da61aef | 389 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 390 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
166f61b9 | 391 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 HD |
392 | } else { |
393 | /* | |
394 | * Optimization: gather nearby vmas into one call down | |
395 | */ | |
396 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 397 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 | 398 | vma = next; |
fd892593 | 399 | next = mas_find(mas, ceiling - 1); |
d2406291 PZ |
400 | if (unlikely(xa_is_zero(next))) |
401 | next = NULL; | |
98e51a22 SB |
402 | if (mm_wr_locked) |
403 | vma_start_write(vma); | |
5beb4930 | 404 | unlink_anon_vmas(vma); |
8f4f8c16 | 405 | unlink_file_vma(vma); |
3bf5ee95 HD |
406 | } |
407 | free_pgd_range(tlb, addr, vma->vm_end, | |
166f61b9 | 408 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 | 409 | } |
e0da382c | 410 | vma = next; |
763ecb03 | 411 | } while (vma); |
1da177e4 LT |
412 | } |
413 | ||
03c4f204 | 414 | void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte) |
1da177e4 | 415 | { |
03c4f204 | 416 | spinlock_t *ptl = pmd_lock(mm, pmd); |
1bb3630e | 417 | |
8ac1f832 | 418 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
c4812909 | 419 | mm_inc_nr_ptes(mm); |
ed33b5a6 QZ |
420 | /* |
421 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
422 | * visible before the pte is made visible to other CPUs by being | |
423 | * put into page tables. | |
424 | * | |
425 | * The other side of the story is the pointer chasing in the page | |
426 | * table walking code (when walking the page table without locking; | |
427 | * ie. most of the time). Fortunately, these data accesses consist | |
428 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
429 | * being the notable exception) will already guarantee loads are | |
430 | * seen in-order. See the alpha page table accessors for the | |
431 | * smp_rmb() barriers in page table walking code. | |
432 | */ | |
433 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
03c4f204 QZ |
434 | pmd_populate(mm, pmd, *pte); |
435 | *pte = NULL; | |
4b471e88 | 436 | } |
c4088ebd | 437 | spin_unlock(ptl); |
03c4f204 QZ |
438 | } |
439 | ||
4cf58924 | 440 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd) |
1da177e4 | 441 | { |
4cf58924 | 442 | pgtable_t new = pte_alloc_one(mm); |
1bb3630e HD |
443 | if (!new) |
444 | return -ENOMEM; | |
445 | ||
03c4f204 | 446 | pmd_install(mm, pmd, &new); |
2f569afd MS |
447 | if (new) |
448 | pte_free(mm, new); | |
1bb3630e | 449 | return 0; |
1da177e4 LT |
450 | } |
451 | ||
4cf58924 | 452 | int __pte_alloc_kernel(pmd_t *pmd) |
1da177e4 | 453 | { |
4cf58924 | 454 | pte_t *new = pte_alloc_one_kernel(&init_mm); |
1bb3630e HD |
455 | if (!new) |
456 | return -ENOMEM; | |
457 | ||
458 | spin_lock(&init_mm.page_table_lock); | |
8ac1f832 | 459 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
ed33b5a6 | 460 | smp_wmb(); /* See comment in pmd_install() */ |
1bb3630e | 461 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 462 | new = NULL; |
4b471e88 | 463 | } |
1bb3630e | 464 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
465 | if (new) |
466 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 467 | return 0; |
1da177e4 LT |
468 | } |
469 | ||
d559db08 KH |
470 | static inline void init_rss_vec(int *rss) |
471 | { | |
472 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
473 | } | |
474 | ||
475 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 476 | { |
d559db08 KH |
477 | int i; |
478 | ||
479 | for (i = 0; i < NR_MM_COUNTERS; i++) | |
480 | if (rss[i]) | |
481 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
482 | } |
483 | ||
b5810039 | 484 | /* |
6aab341e LT |
485 | * This function is called to print an error when a bad pte |
486 | * is found. For example, we might have a PFN-mapped pte in | |
487 | * a region that doesn't allow it. | |
b5810039 NP |
488 | * |
489 | * The calling function must still handle the error. | |
490 | */ | |
3dc14741 HD |
491 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
492 | pte_t pte, struct page *page) | |
b5810039 | 493 | { |
3dc14741 | 494 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
c2febafc KS |
495 | p4d_t *p4d = p4d_offset(pgd, addr); |
496 | pud_t *pud = pud_offset(p4d, addr); | |
3dc14741 HD |
497 | pmd_t *pmd = pmd_offset(pud, addr); |
498 | struct address_space *mapping; | |
499 | pgoff_t index; | |
d936cf9b HD |
500 | static unsigned long resume; |
501 | static unsigned long nr_shown; | |
502 | static unsigned long nr_unshown; | |
503 | ||
504 | /* | |
505 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
506 | * or allow a steady drip of one report per second. | |
507 | */ | |
508 | if (nr_shown == 60) { | |
509 | if (time_before(jiffies, resume)) { | |
510 | nr_unshown++; | |
511 | return; | |
512 | } | |
513 | if (nr_unshown) { | |
1170532b JP |
514 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
515 | nr_unshown); | |
d936cf9b HD |
516 | nr_unshown = 0; |
517 | } | |
518 | nr_shown = 0; | |
519 | } | |
520 | if (nr_shown++ == 0) | |
521 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
522 | |
523 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
524 | index = linear_page_index(vma, addr); | |
525 | ||
1170532b JP |
526 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
527 | current->comm, | |
528 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 529 | if (page) |
f0b791a3 | 530 | dump_page(page, "bad pte"); |
6aa9b8b2 | 531 | pr_alert("addr:%px vm_flags:%08lx anon_vma:%px mapping:%px index:%lx\n", |
1170532b | 532 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); |
7e0a1265 | 533 | pr_alert("file:%pD fault:%ps mmap:%ps read_folio:%ps\n", |
2682582a KK |
534 | vma->vm_file, |
535 | vma->vm_ops ? vma->vm_ops->fault : NULL, | |
536 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, | |
7e0a1265 | 537 | mapping ? mapping->a_ops->read_folio : NULL); |
b5810039 | 538 | dump_stack(); |
373d4d09 | 539 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
540 | } |
541 | ||
ee498ed7 | 542 | /* |
7e675137 | 543 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 544 | * |
7e675137 NP |
545 | * "Special" mappings do not wish to be associated with a "struct page" (either |
546 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
547 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 548 | * |
7e675137 NP |
549 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
550 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
551 | * may not have a spare pte bit, which requires a more complicated scheme, | |
552 | * described below. | |
553 | * | |
554 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
555 | * special mapping (even if there are underlying and valid "struct pages"). | |
556 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 557 | * |
b379d790 JH |
558 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
559 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
560 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
561 | * mapping will always honor the rule | |
6aab341e LT |
562 | * |
563 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
564 | * | |
7e675137 NP |
565 | * And for normal mappings this is false. |
566 | * | |
567 | * This restricts such mappings to be a linear translation from virtual address | |
568 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
569 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
570 | * special (because none can have been COWed). | |
b379d790 | 571 | * |
b379d790 | 572 | * |
7e675137 | 573 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
574 | * |
575 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
576 | * page" backing, however the difference is that _all_ pages with a struct | |
577 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
578 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
579 | * (which can be slower and simply not an option for some PFNMAP users). The | |
580 | * advantage is that we don't have to follow the strict linearity rule of | |
581 | * PFNMAP mappings in order to support COWable mappings. | |
582 | * | |
ee498ed7 | 583 | */ |
25b2995a CH |
584 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
585 | pte_t pte) | |
ee498ed7 | 586 | { |
22b31eec | 587 | unsigned long pfn = pte_pfn(pte); |
7e675137 | 588 | |
00b3a331 | 589 | if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) { |
b38af472 | 590 | if (likely(!pte_special(pte))) |
22b31eec | 591 | goto check_pfn; |
667a0a06 DV |
592 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
593 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
594 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
595 | return NULL; | |
df6ad698 JG |
596 | if (is_zero_pfn(pfn)) |
597 | return NULL; | |
e1fb4a08 | 598 | if (pte_devmap(pte)) |
3218f871 AS |
599 | /* |
600 | * NOTE: New users of ZONE_DEVICE will not set pte_devmap() | |
601 | * and will have refcounts incremented on their struct pages | |
602 | * when they are inserted into PTEs, thus they are safe to | |
603 | * return here. Legacy ZONE_DEVICE pages that set pte_devmap() | |
604 | * do not have refcounts. Example of legacy ZONE_DEVICE is | |
605 | * MEMORY_DEVICE_FS_DAX type in pmem or virtio_fs drivers. | |
606 | */ | |
e1fb4a08 DJ |
607 | return NULL; |
608 | ||
df6ad698 | 609 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
610 | return NULL; |
611 | } | |
612 | ||
00b3a331 | 613 | /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */ |
7e675137 | 614 | |
b379d790 JH |
615 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
616 | if (vma->vm_flags & VM_MIXEDMAP) { | |
617 | if (!pfn_valid(pfn)) | |
618 | return NULL; | |
619 | goto out; | |
620 | } else { | |
7e675137 NP |
621 | unsigned long off; |
622 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
623 | if (pfn == vma->vm_pgoff + off) |
624 | return NULL; | |
625 | if (!is_cow_mapping(vma->vm_flags)) | |
626 | return NULL; | |
627 | } | |
6aab341e LT |
628 | } |
629 | ||
b38af472 HD |
630 | if (is_zero_pfn(pfn)) |
631 | return NULL; | |
00b3a331 | 632 | |
22b31eec HD |
633 | check_pfn: |
634 | if (unlikely(pfn > highest_memmap_pfn)) { | |
635 | print_bad_pte(vma, addr, pte, NULL); | |
636 | return NULL; | |
637 | } | |
6aab341e LT |
638 | |
639 | /* | |
7e675137 | 640 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 641 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 642 | */ |
b379d790 | 643 | out: |
6aab341e | 644 | return pfn_to_page(pfn); |
ee498ed7 HD |
645 | } |
646 | ||
318e9342 VMO |
647 | struct folio *vm_normal_folio(struct vm_area_struct *vma, unsigned long addr, |
648 | pte_t pte) | |
649 | { | |
650 | struct page *page = vm_normal_page(vma, addr, pte); | |
651 | ||
652 | if (page) | |
653 | return page_folio(page); | |
654 | return NULL; | |
655 | } | |
656 | ||
28093f9f GS |
657 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
658 | struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, | |
659 | pmd_t pmd) | |
660 | { | |
661 | unsigned long pfn = pmd_pfn(pmd); | |
662 | ||
663 | /* | |
664 | * There is no pmd_special() but there may be special pmds, e.g. | |
665 | * in a direct-access (dax) mapping, so let's just replicate the | |
00b3a331 | 666 | * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here. |
28093f9f GS |
667 | */ |
668 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { | |
669 | if (vma->vm_flags & VM_MIXEDMAP) { | |
670 | if (!pfn_valid(pfn)) | |
671 | return NULL; | |
672 | goto out; | |
673 | } else { | |
674 | unsigned long off; | |
675 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
676 | if (pfn == vma->vm_pgoff + off) | |
677 | return NULL; | |
678 | if (!is_cow_mapping(vma->vm_flags)) | |
679 | return NULL; | |
680 | } | |
681 | } | |
682 | ||
e1fb4a08 DJ |
683 | if (pmd_devmap(pmd)) |
684 | return NULL; | |
3cde287b | 685 | if (is_huge_zero_pmd(pmd)) |
28093f9f GS |
686 | return NULL; |
687 | if (unlikely(pfn > highest_memmap_pfn)) | |
688 | return NULL; | |
689 | ||
690 | /* | |
691 | * NOTE! We still have PageReserved() pages in the page tables. | |
692 | * eg. VDSO mappings can cause them to exist. | |
693 | */ | |
694 | out: | |
695 | return pfn_to_page(pfn); | |
696 | } | |
65610453 KW |
697 | |
698 | struct folio *vm_normal_folio_pmd(struct vm_area_struct *vma, | |
699 | unsigned long addr, pmd_t pmd) | |
700 | { | |
701 | struct page *page = vm_normal_page_pmd(vma, addr, pmd); | |
702 | ||
703 | if (page) | |
704 | return page_folio(page); | |
705 | return NULL; | |
706 | } | |
28093f9f GS |
707 | #endif |
708 | ||
b756a3b5 AP |
709 | static void restore_exclusive_pte(struct vm_area_struct *vma, |
710 | struct page *page, unsigned long address, | |
711 | pte_t *ptep) | |
712 | { | |
b832a354 | 713 | struct folio *folio = page_folio(page); |
c33c7948 | 714 | pte_t orig_pte; |
b756a3b5 AP |
715 | pte_t pte; |
716 | swp_entry_t entry; | |
717 | ||
c33c7948 | 718 | orig_pte = ptep_get(ptep); |
b756a3b5 | 719 | pte = pte_mkold(mk_pte(page, READ_ONCE(vma->vm_page_prot))); |
c33c7948 | 720 | if (pte_swp_soft_dirty(orig_pte)) |
b756a3b5 AP |
721 | pte = pte_mksoft_dirty(pte); |
722 | ||
c33c7948 RR |
723 | entry = pte_to_swp_entry(orig_pte); |
724 | if (pte_swp_uffd_wp(orig_pte)) | |
b756a3b5 AP |
725 | pte = pte_mkuffd_wp(pte); |
726 | else if (is_writable_device_exclusive_entry(entry)) | |
727 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); | |
728 | ||
b832a354 DH |
729 | VM_BUG_ON_FOLIO(pte_write(pte) && (!folio_test_anon(folio) && |
730 | PageAnonExclusive(page)), folio); | |
6c287605 | 731 | |
b756a3b5 AP |
732 | /* |
733 | * No need to take a page reference as one was already | |
734 | * created when the swap entry was made. | |
735 | */ | |
b832a354 DH |
736 | if (folio_test_anon(folio)) |
737 | folio_add_anon_rmap_pte(folio, page, vma, address, RMAP_NONE); | |
b756a3b5 AP |
738 | else |
739 | /* | |
740 | * Currently device exclusive access only supports anonymous | |
741 | * memory so the entry shouldn't point to a filebacked page. | |
742 | */ | |
4d8ff640 | 743 | WARN_ON_ONCE(1); |
b756a3b5 | 744 | |
1eba86c0 PT |
745 | set_pte_at(vma->vm_mm, address, ptep, pte); |
746 | ||
b756a3b5 AP |
747 | /* |
748 | * No need to invalidate - it was non-present before. However | |
749 | * secondary CPUs may have mappings that need invalidating. | |
750 | */ | |
751 | update_mmu_cache(vma, address, ptep); | |
752 | } | |
753 | ||
754 | /* | |
755 | * Tries to restore an exclusive pte if the page lock can be acquired without | |
756 | * sleeping. | |
757 | */ | |
758 | static int | |
759 | try_restore_exclusive_pte(pte_t *src_pte, struct vm_area_struct *vma, | |
760 | unsigned long addr) | |
761 | { | |
c33c7948 | 762 | swp_entry_t entry = pte_to_swp_entry(ptep_get(src_pte)); |
b756a3b5 AP |
763 | struct page *page = pfn_swap_entry_to_page(entry); |
764 | ||
765 | if (trylock_page(page)) { | |
766 | restore_exclusive_pte(vma, page, addr, src_pte); | |
767 | unlock_page(page); | |
768 | return 0; | |
769 | } | |
770 | ||
771 | return -EBUSY; | |
772 | } | |
773 | ||
1da177e4 LT |
774 | /* |
775 | * copy one vm_area from one task to the other. Assumes the page tables | |
776 | * already present in the new task to be cleared in the whole range | |
777 | * covered by this vma. | |
1da177e4 LT |
778 | */ |
779 | ||
df3a57d1 LT |
780 | static unsigned long |
781 | copy_nonpresent_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
8f34f1ea PX |
782 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *dst_vma, |
783 | struct vm_area_struct *src_vma, unsigned long addr, int *rss) | |
1da177e4 | 784 | { |
8f34f1ea | 785 | unsigned long vm_flags = dst_vma->vm_flags; |
c33c7948 RR |
786 | pte_t orig_pte = ptep_get(src_pte); |
787 | pte_t pte = orig_pte; | |
08e7795e | 788 | struct folio *folio; |
1da177e4 | 789 | struct page *page; |
c33c7948 | 790 | swp_entry_t entry = pte_to_swp_entry(orig_pte); |
df3a57d1 LT |
791 | |
792 | if (likely(!non_swap_entry(entry))) { | |
793 | if (swap_duplicate(entry) < 0) | |
9a5cc85c | 794 | return -EIO; |
df3a57d1 LT |
795 | |
796 | /* make sure dst_mm is on swapoff's mmlist. */ | |
797 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
798 | spin_lock(&mmlist_lock); | |
799 | if (list_empty(&dst_mm->mmlist)) | |
800 | list_add(&dst_mm->mmlist, | |
801 | &src_mm->mmlist); | |
802 | spin_unlock(&mmlist_lock); | |
803 | } | |
1493a191 | 804 | /* Mark the swap entry as shared. */ |
c33c7948 RR |
805 | if (pte_swp_exclusive(orig_pte)) { |
806 | pte = pte_swp_clear_exclusive(orig_pte); | |
1493a191 DH |
807 | set_pte_at(src_mm, addr, src_pte, pte); |
808 | } | |
df3a57d1 LT |
809 | rss[MM_SWAPENTS]++; |
810 | } else if (is_migration_entry(entry)) { | |
530c2a0d | 811 | folio = pfn_swap_entry_folio(entry); |
1da177e4 | 812 | |
a23f517b | 813 | rss[mm_counter(folio)]++; |
5042db43 | 814 | |
6c287605 | 815 | if (!is_readable_migration_entry(entry) && |
df3a57d1 | 816 | is_cow_mapping(vm_flags)) { |
5042db43 | 817 | /* |
6c287605 DH |
818 | * COW mappings require pages in both parent and child |
819 | * to be set to read. A previously exclusive entry is | |
820 | * now shared. | |
5042db43 | 821 | */ |
4dd845b5 AP |
822 | entry = make_readable_migration_entry( |
823 | swp_offset(entry)); | |
df3a57d1 | 824 | pte = swp_entry_to_pte(entry); |
c33c7948 | 825 | if (pte_swp_soft_dirty(orig_pte)) |
df3a57d1 | 826 | pte = pte_swp_mksoft_dirty(pte); |
c33c7948 | 827 | if (pte_swp_uffd_wp(orig_pte)) |
df3a57d1 LT |
828 | pte = pte_swp_mkuffd_wp(pte); |
829 | set_pte_at(src_mm, addr, src_pte, pte); | |
830 | } | |
831 | } else if (is_device_private_entry(entry)) { | |
af5cdaf8 | 832 | page = pfn_swap_entry_to_page(entry); |
08e7795e | 833 | folio = page_folio(page); |
5042db43 | 834 | |
df3a57d1 LT |
835 | /* |
836 | * Update rss count even for unaddressable pages, as | |
837 | * they should treated just like normal pages in this | |
838 | * respect. | |
839 | * | |
840 | * We will likely want to have some new rss counters | |
841 | * for unaddressable pages, at some point. But for now | |
842 | * keep things as they are. | |
843 | */ | |
08e7795e | 844 | folio_get(folio); |
a23f517b | 845 | rss[mm_counter(folio)]++; |
fb3d824d | 846 | /* Cannot fail as these pages cannot get pinned. */ |
08e7795e | 847 | folio_try_dup_anon_rmap_pte(folio, page, src_vma); |
df3a57d1 LT |
848 | |
849 | /* | |
850 | * We do not preserve soft-dirty information, because so | |
851 | * far, checkpoint/restore is the only feature that | |
852 | * requires that. And checkpoint/restore does not work | |
853 | * when a device driver is involved (you cannot easily | |
854 | * save and restore device driver state). | |
855 | */ | |
4dd845b5 | 856 | if (is_writable_device_private_entry(entry) && |
df3a57d1 | 857 | is_cow_mapping(vm_flags)) { |
4dd845b5 AP |
858 | entry = make_readable_device_private_entry( |
859 | swp_offset(entry)); | |
df3a57d1 | 860 | pte = swp_entry_to_pte(entry); |
c33c7948 | 861 | if (pte_swp_uffd_wp(orig_pte)) |
df3a57d1 LT |
862 | pte = pte_swp_mkuffd_wp(pte); |
863 | set_pte_at(src_mm, addr, src_pte, pte); | |
1da177e4 | 864 | } |
b756a3b5 AP |
865 | } else if (is_device_exclusive_entry(entry)) { |
866 | /* | |
867 | * Make device exclusive entries present by restoring the | |
868 | * original entry then copying as for a present pte. Device | |
869 | * exclusive entries currently only support private writable | |
870 | * (ie. COW) mappings. | |
871 | */ | |
872 | VM_BUG_ON(!is_cow_mapping(src_vma->vm_flags)); | |
873 | if (try_restore_exclusive_pte(src_pte, src_vma, addr)) | |
874 | return -EBUSY; | |
875 | return -ENOENT; | |
c56d1b62 | 876 | } else if (is_pte_marker_entry(entry)) { |
af19487f AR |
877 | pte_marker marker = copy_pte_marker(entry, dst_vma); |
878 | ||
879 | if (marker) | |
880 | set_pte_at(dst_mm, addr, dst_pte, | |
881 | make_pte_marker(marker)); | |
c56d1b62 | 882 | return 0; |
1da177e4 | 883 | } |
8f34f1ea PX |
884 | if (!userfaultfd_wp(dst_vma)) |
885 | pte = pte_swp_clear_uffd_wp(pte); | |
df3a57d1 LT |
886 | set_pte_at(dst_mm, addr, dst_pte, pte); |
887 | return 0; | |
888 | } | |
889 | ||
70e806e4 | 890 | /* |
b51ad4f8 | 891 | * Copy a present and normal page. |
70e806e4 | 892 | * |
b51ad4f8 DH |
893 | * NOTE! The usual case is that this isn't required; |
894 | * instead, the caller can just increase the page refcount | |
895 | * and re-use the pte the traditional way. | |
70e806e4 PX |
896 | * |
897 | * And if we need a pre-allocated page but don't yet have | |
898 | * one, return a negative error to let the preallocation | |
899 | * code know so that it can do so outside the page table | |
900 | * lock. | |
901 | */ | |
902 | static inline int | |
c78f4636 PX |
903 | copy_present_page(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, |
904 | pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, | |
edf50470 | 905 | struct folio **prealloc, struct page *page) |
70e806e4 | 906 | { |
edf50470 | 907 | struct folio *new_folio; |
b51ad4f8 | 908 | pte_t pte; |
70e806e4 | 909 | |
edf50470 MWO |
910 | new_folio = *prealloc; |
911 | if (!new_folio) | |
70e806e4 PX |
912 | return -EAGAIN; |
913 | ||
914 | /* | |
915 | * We have a prealloc page, all good! Take it | |
916 | * over and copy the page & arm it. | |
917 | */ | |
918 | *prealloc = NULL; | |
edf50470 MWO |
919 | copy_user_highpage(&new_folio->page, page, addr, src_vma); |
920 | __folio_mark_uptodate(new_folio); | |
921 | folio_add_new_anon_rmap(new_folio, dst_vma, addr); | |
922 | folio_add_lru_vma(new_folio, dst_vma); | |
923 | rss[MM_ANONPAGES]++; | |
70e806e4 PX |
924 | |
925 | /* All done, just insert the new page copy in the child */ | |
edf50470 | 926 | pte = mk_pte(&new_folio->page, dst_vma->vm_page_prot); |
c78f4636 | 927 | pte = maybe_mkwrite(pte_mkdirty(pte), dst_vma); |
c33c7948 | 928 | if (userfaultfd_pte_wp(dst_vma, ptep_get(src_pte))) |
8f34f1ea | 929 | /* Uffd-wp needs to be delivered to dest pte as well */ |
f1eb1bac | 930 | pte = pte_mkuffd_wp(pte); |
c78f4636 | 931 | set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); |
70e806e4 PX |
932 | return 0; |
933 | } | |
934 | ||
f8d93776 | 935 | static __always_inline void __copy_present_ptes(struct vm_area_struct *dst_vma, |
23ed1908 | 936 | struct vm_area_struct *src_vma, pte_t *dst_pte, pte_t *src_pte, |
f8d93776 | 937 | pte_t pte, unsigned long addr, int nr) |
23ed1908 DH |
938 | { |
939 | struct mm_struct *src_mm = src_vma->vm_mm; | |
940 | ||
941 | /* If it's a COW mapping, write protect it both processes. */ | |
942 | if (is_cow_mapping(src_vma->vm_flags) && pte_write(pte)) { | |
f8d93776 | 943 | wrprotect_ptes(src_mm, addr, src_pte, nr); |
23ed1908 DH |
944 | pte = pte_wrprotect(pte); |
945 | } | |
946 | ||
947 | /* If it's a shared mapping, mark it clean in the child. */ | |
948 | if (src_vma->vm_flags & VM_SHARED) | |
949 | pte = pte_mkclean(pte); | |
950 | pte = pte_mkold(pte); | |
951 | ||
952 | if (!userfaultfd_wp(dst_vma)) | |
953 | pte = pte_clear_uffd_wp(pte); | |
954 | ||
f8d93776 DH |
955 | set_ptes(dst_vma->vm_mm, addr, dst_pte, pte, nr); |
956 | } | |
957 | ||
70e806e4 | 958 | /* |
f8d93776 DH |
959 | * Copy one present PTE, trying to batch-process subsequent PTEs that map |
960 | * consecutive pages of the same folio by copying them as well. | |
961 | * | |
962 | * Returns -EAGAIN if one preallocated page is required to copy the next PTE. | |
963 | * Otherwise, returns the number of copied PTEs (at least 1). | |
70e806e4 PX |
964 | */ |
965 | static inline int | |
f8d93776 | 966 | copy_present_ptes(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, |
53723298 | 967 | pte_t *dst_pte, pte_t *src_pte, pte_t pte, unsigned long addr, |
f8d93776 | 968 | int max_nr, int *rss, struct folio **prealloc) |
df3a57d1 | 969 | { |
df3a57d1 | 970 | struct page *page; |
14ddee41 | 971 | struct folio *folio; |
d7c0e5f7 | 972 | bool any_writable; |
25365e10 | 973 | fpb_t flags = 0; |
f8d93776 | 974 | int err, nr; |
df3a57d1 | 975 | |
c78f4636 | 976 | page = vm_normal_page(src_vma, addr, pte); |
23ed1908 DH |
977 | if (unlikely(!page)) |
978 | goto copy_pte; | |
979 | ||
980 | folio = page_folio(page); | |
f8d93776 DH |
981 | |
982 | /* | |
983 | * If we likely have to copy, just don't bother with batching. Make | |
984 | * sure that the common "small folio" case is as fast as possible | |
985 | * by keeping the batching logic separate. | |
986 | */ | |
987 | if (unlikely(!*prealloc && folio_test_large(folio) && max_nr != 1)) { | |
25365e10 DH |
988 | if (src_vma->vm_flags & VM_SHARED) |
989 | flags |= FPB_IGNORE_DIRTY; | |
990 | if (!vma_soft_dirty_enabled(src_vma)) | |
991 | flags |= FPB_IGNORE_SOFT_DIRTY; | |
992 | ||
d7c0e5f7 | 993 | nr = folio_pte_batch(folio, addr, src_pte, pte, max_nr, flags, |
96ebdb03 | 994 | &any_writable, NULL, NULL); |
f8d93776 DH |
995 | folio_ref_add(folio, nr); |
996 | if (folio_test_anon(folio)) { | |
997 | if (unlikely(folio_try_dup_anon_rmap_ptes(folio, page, | |
998 | nr, src_vma))) { | |
999 | folio_ref_sub(folio, nr); | |
1000 | return -EAGAIN; | |
1001 | } | |
1002 | rss[MM_ANONPAGES] += nr; | |
1003 | VM_WARN_ON_FOLIO(PageAnonExclusive(page), folio); | |
1004 | } else { | |
1005 | folio_dup_file_rmap_ptes(folio, page, nr); | |
1006 | rss[mm_counter_file(folio)] += nr; | |
1007 | } | |
d7c0e5f7 DH |
1008 | if (any_writable) |
1009 | pte = pte_mkwrite(pte, src_vma); | |
f8d93776 DH |
1010 | __copy_present_ptes(dst_vma, src_vma, dst_pte, src_pte, pte, |
1011 | addr, nr); | |
1012 | return nr; | |
1013 | } | |
1014 | ||
23ed1908 DH |
1015 | folio_get(folio); |
1016 | if (folio_test_anon(folio)) { | |
b51ad4f8 DH |
1017 | /* |
1018 | * If this page may have been pinned by the parent process, | |
1019 | * copy the page immediately for the child so that we'll always | |
1020 | * guarantee the pinned page won't be randomly replaced in the | |
1021 | * future. | |
1022 | */ | |
08e7795e | 1023 | if (unlikely(folio_try_dup_anon_rmap_pte(folio, page, src_vma))) { |
14ddee41 MWO |
1024 | /* Page may be pinned, we have to copy. */ |
1025 | folio_put(folio); | |
f8d93776 DH |
1026 | err = copy_present_page(dst_vma, src_vma, dst_pte, src_pte, |
1027 | addr, rss, prealloc, page); | |
1028 | return err ? err : 1; | |
fb3d824d | 1029 | } |
edf50470 | 1030 | rss[MM_ANONPAGES]++; |
23ed1908 DH |
1031 | VM_WARN_ON_FOLIO(PageAnonExclusive(page), folio); |
1032 | } else { | |
d8ef5e31 | 1033 | folio_dup_file_rmap_pte(folio, page); |
6b27cc6c | 1034 | rss[mm_counter_file(folio)]++; |
70e806e4 PX |
1035 | } |
1036 | ||
23ed1908 | 1037 | copy_pte: |
f8d93776 DH |
1038 | __copy_present_ptes(dst_vma, src_vma, dst_pte, src_pte, pte, addr, 1); |
1039 | return 1; | |
70e806e4 PX |
1040 | } |
1041 | ||
294de6d8 KW |
1042 | static inline struct folio *folio_prealloc(struct mm_struct *src_mm, |
1043 | struct vm_area_struct *vma, unsigned long addr, bool need_zero) | |
70e806e4 | 1044 | { |
edf50470 | 1045 | struct folio *new_folio; |
70e806e4 | 1046 | |
294de6d8 KW |
1047 | if (need_zero) |
1048 | new_folio = vma_alloc_zeroed_movable_folio(vma, addr); | |
1049 | else | |
1050 | new_folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma, | |
1051 | addr, false); | |
1052 | ||
edf50470 | 1053 | if (!new_folio) |
70e806e4 PX |
1054 | return NULL; |
1055 | ||
edf50470 MWO |
1056 | if (mem_cgroup_charge(new_folio, src_mm, GFP_KERNEL)) { |
1057 | folio_put(new_folio); | |
70e806e4 | 1058 | return NULL; |
6aab341e | 1059 | } |
e601ded4 | 1060 | folio_throttle_swaprate(new_folio, GFP_KERNEL); |
ae859762 | 1061 | |
edf50470 | 1062 | return new_folio; |
1da177e4 LT |
1063 | } |
1064 | ||
c78f4636 PX |
1065 | static int |
1066 | copy_pte_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1067 | pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr, | |
1068 | unsigned long end) | |
1da177e4 | 1069 | { |
c78f4636 PX |
1070 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
1071 | struct mm_struct *src_mm = src_vma->vm_mm; | |
c36987e2 | 1072 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 1073 | pte_t *src_pte, *dst_pte; |
c33c7948 | 1074 | pte_t ptent; |
c74df32c | 1075 | spinlock_t *src_ptl, *dst_ptl; |
f8d93776 | 1076 | int progress, max_nr, ret = 0; |
d559db08 | 1077 | int rss[NR_MM_COUNTERS]; |
570a335b | 1078 | swp_entry_t entry = (swp_entry_t){0}; |
edf50470 | 1079 | struct folio *prealloc = NULL; |
f8d93776 | 1080 | int nr; |
1da177e4 LT |
1081 | |
1082 | again: | |
70e806e4 | 1083 | progress = 0; |
d559db08 KH |
1084 | init_rss_vec(rss); |
1085 | ||
3db82b93 HD |
1086 | /* |
1087 | * copy_pmd_range()'s prior pmd_none_or_clear_bad(src_pmd), and the | |
1088 | * error handling here, assume that exclusive mmap_lock on dst and src | |
1089 | * protects anon from unexpected THP transitions; with shmem and file | |
1090 | * protected by mmap_lock-less collapse skipping areas with anon_vma | |
1091 | * (whereas vma_needs_copy() skips areas without anon_vma). A rework | |
1092 | * can remove such assumptions later, but this is good enough for now. | |
1093 | */ | |
c74df32c | 1094 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
70e806e4 PX |
1095 | if (!dst_pte) { |
1096 | ret = -ENOMEM; | |
1097 | goto out; | |
1098 | } | |
3db82b93 HD |
1099 | src_pte = pte_offset_map_nolock(src_mm, src_pmd, addr, &src_ptl); |
1100 | if (!src_pte) { | |
1101 | pte_unmap_unlock(dst_pte, dst_ptl); | |
1102 | /* ret == 0 */ | |
1103 | goto out; | |
1104 | } | |
f20dc5f7 | 1105 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
1106 | orig_src_pte = src_pte; |
1107 | orig_dst_pte = dst_pte; | |
6606c3e0 | 1108 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 1109 | |
1da177e4 | 1110 | do { |
f8d93776 DH |
1111 | nr = 1; |
1112 | ||
1da177e4 LT |
1113 | /* |
1114 | * We are holding two locks at this point - either of them | |
1115 | * could generate latencies in another task on another CPU. | |
1116 | */ | |
e040f218 HD |
1117 | if (progress >= 32) { |
1118 | progress = 0; | |
1119 | if (need_resched() || | |
95c354fe | 1120 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
1121 | break; |
1122 | } | |
c33c7948 RR |
1123 | ptent = ptep_get(src_pte); |
1124 | if (pte_none(ptent)) { | |
1da177e4 LT |
1125 | progress++; |
1126 | continue; | |
1127 | } | |
c33c7948 | 1128 | if (unlikely(!pte_present(ptent))) { |
9a5cc85c AP |
1129 | ret = copy_nonpresent_pte(dst_mm, src_mm, |
1130 | dst_pte, src_pte, | |
1131 | dst_vma, src_vma, | |
1132 | addr, rss); | |
1133 | if (ret == -EIO) { | |
c33c7948 | 1134 | entry = pte_to_swp_entry(ptep_get(src_pte)); |
79a1971c | 1135 | break; |
b756a3b5 AP |
1136 | } else if (ret == -EBUSY) { |
1137 | break; | |
1138 | } else if (!ret) { | |
1139 | progress += 8; | |
1140 | continue; | |
9a5cc85c | 1141 | } |
53723298 DH |
1142 | ptent = ptep_get(src_pte); |
1143 | VM_WARN_ON_ONCE(!pte_present(ptent)); | |
b756a3b5 AP |
1144 | |
1145 | /* | |
1146 | * Device exclusive entry restored, continue by copying | |
1147 | * the now present pte. | |
1148 | */ | |
1149 | WARN_ON_ONCE(ret != -ENOENT); | |
79a1971c | 1150 | } |
f8d93776 DH |
1151 | /* copy_present_ptes() will clear `*prealloc' if consumed */ |
1152 | max_nr = (end - addr) / PAGE_SIZE; | |
1153 | ret = copy_present_ptes(dst_vma, src_vma, dst_pte, src_pte, | |
1154 | ptent, addr, max_nr, rss, &prealloc); | |
70e806e4 PX |
1155 | /* |
1156 | * If we need a pre-allocated page for this pte, drop the | |
1157 | * locks, allocate, and try again. | |
1158 | */ | |
1159 | if (unlikely(ret == -EAGAIN)) | |
1160 | break; | |
1161 | if (unlikely(prealloc)) { | |
1162 | /* | |
1163 | * pre-alloc page cannot be reused by next time so as | |
1164 | * to strictly follow mempolicy (e.g., alloc_page_vma() | |
1165 | * will allocate page according to address). This | |
1166 | * could only happen if one pinned pte changed. | |
1167 | */ | |
edf50470 | 1168 | folio_put(prealloc); |
70e806e4 PX |
1169 | prealloc = NULL; |
1170 | } | |
f8d93776 DH |
1171 | nr = ret; |
1172 | progress += 8 * nr; | |
1173 | } while (dst_pte += nr, src_pte += nr, addr += PAGE_SIZE * nr, | |
1174 | addr != end); | |
1da177e4 | 1175 | |
6606c3e0 | 1176 | arch_leave_lazy_mmu_mode(); |
3db82b93 | 1177 | pte_unmap_unlock(orig_src_pte, src_ptl); |
d559db08 | 1178 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 1179 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 1180 | cond_resched(); |
570a335b | 1181 | |
9a5cc85c AP |
1182 | if (ret == -EIO) { |
1183 | VM_WARN_ON_ONCE(!entry.val); | |
70e806e4 PX |
1184 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) { |
1185 | ret = -ENOMEM; | |
1186 | goto out; | |
1187 | } | |
1188 | entry.val = 0; | |
b756a3b5 AP |
1189 | } else if (ret == -EBUSY) { |
1190 | goto out; | |
9a5cc85c | 1191 | } else if (ret == -EAGAIN) { |
294de6d8 | 1192 | prealloc = folio_prealloc(src_mm, src_vma, addr, false); |
70e806e4 | 1193 | if (!prealloc) |
570a335b | 1194 | return -ENOMEM; |
f8d93776 | 1195 | } else if (ret < 0) { |
9a5cc85c | 1196 | VM_WARN_ON_ONCE(1); |
570a335b | 1197 | } |
9a5cc85c AP |
1198 | |
1199 | /* We've captured and resolved the error. Reset, try again. */ | |
1200 | ret = 0; | |
1201 | ||
1da177e4 LT |
1202 | if (addr != end) |
1203 | goto again; | |
70e806e4 PX |
1204 | out: |
1205 | if (unlikely(prealloc)) | |
edf50470 | 1206 | folio_put(prealloc); |
70e806e4 | 1207 | return ret; |
1da177e4 LT |
1208 | } |
1209 | ||
c78f4636 PX |
1210 | static inline int |
1211 | copy_pmd_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1212 | pud_t *dst_pud, pud_t *src_pud, unsigned long addr, | |
1213 | unsigned long end) | |
1da177e4 | 1214 | { |
c78f4636 PX |
1215 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
1216 | struct mm_struct *src_mm = src_vma->vm_mm; | |
1da177e4 LT |
1217 | pmd_t *src_pmd, *dst_pmd; |
1218 | unsigned long next; | |
1219 | ||
1220 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
1221 | if (!dst_pmd) | |
1222 | return -ENOMEM; | |
1223 | src_pmd = pmd_offset(src_pud, addr); | |
1224 | do { | |
1225 | next = pmd_addr_end(addr, end); | |
84c3fc4e ZY |
1226 | if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) |
1227 | || pmd_devmap(*src_pmd)) { | |
71e3aac0 | 1228 | int err; |
c78f4636 | 1229 | VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, src_vma); |
8f34f1ea PX |
1230 | err = copy_huge_pmd(dst_mm, src_mm, dst_pmd, src_pmd, |
1231 | addr, dst_vma, src_vma); | |
71e3aac0 AA |
1232 | if (err == -ENOMEM) |
1233 | return -ENOMEM; | |
1234 | if (!err) | |
1235 | continue; | |
1236 | /* fall through */ | |
1237 | } | |
1da177e4 LT |
1238 | if (pmd_none_or_clear_bad(src_pmd)) |
1239 | continue; | |
c78f4636 PX |
1240 | if (copy_pte_range(dst_vma, src_vma, dst_pmd, src_pmd, |
1241 | addr, next)) | |
1da177e4 LT |
1242 | return -ENOMEM; |
1243 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
1244 | return 0; | |
1245 | } | |
1246 | ||
c78f4636 PX |
1247 | static inline int |
1248 | copy_pud_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1249 | p4d_t *dst_p4d, p4d_t *src_p4d, unsigned long addr, | |
1250 | unsigned long end) | |
1da177e4 | 1251 | { |
c78f4636 PX |
1252 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
1253 | struct mm_struct *src_mm = src_vma->vm_mm; | |
1da177e4 LT |
1254 | pud_t *src_pud, *dst_pud; |
1255 | unsigned long next; | |
1256 | ||
c2febafc | 1257 | dst_pud = pud_alloc(dst_mm, dst_p4d, addr); |
1da177e4 LT |
1258 | if (!dst_pud) |
1259 | return -ENOMEM; | |
c2febafc | 1260 | src_pud = pud_offset(src_p4d, addr); |
1da177e4 LT |
1261 | do { |
1262 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1263 | if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { |
1264 | int err; | |
1265 | ||
c78f4636 | 1266 | VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, src_vma); |
a00cc7d9 | 1267 | err = copy_huge_pud(dst_mm, src_mm, |
c78f4636 | 1268 | dst_pud, src_pud, addr, src_vma); |
a00cc7d9 MW |
1269 | if (err == -ENOMEM) |
1270 | return -ENOMEM; | |
1271 | if (!err) | |
1272 | continue; | |
1273 | /* fall through */ | |
1274 | } | |
1da177e4 LT |
1275 | if (pud_none_or_clear_bad(src_pud)) |
1276 | continue; | |
c78f4636 PX |
1277 | if (copy_pmd_range(dst_vma, src_vma, dst_pud, src_pud, |
1278 | addr, next)) | |
1da177e4 LT |
1279 | return -ENOMEM; |
1280 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
1281 | return 0; | |
1282 | } | |
1283 | ||
c78f4636 PX |
1284 | static inline int |
1285 | copy_p4d_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, | |
1286 | pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long addr, | |
1287 | unsigned long end) | |
c2febafc | 1288 | { |
c78f4636 | 1289 | struct mm_struct *dst_mm = dst_vma->vm_mm; |
c2febafc KS |
1290 | p4d_t *src_p4d, *dst_p4d; |
1291 | unsigned long next; | |
1292 | ||
1293 | dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); | |
1294 | if (!dst_p4d) | |
1295 | return -ENOMEM; | |
1296 | src_p4d = p4d_offset(src_pgd, addr); | |
1297 | do { | |
1298 | next = p4d_addr_end(addr, end); | |
1299 | if (p4d_none_or_clear_bad(src_p4d)) | |
1300 | continue; | |
c78f4636 PX |
1301 | if (copy_pud_range(dst_vma, src_vma, dst_p4d, src_p4d, |
1302 | addr, next)) | |
c2febafc KS |
1303 | return -ENOMEM; |
1304 | } while (dst_p4d++, src_p4d++, addr = next, addr != end); | |
1305 | return 0; | |
1306 | } | |
1307 | ||
c56d1b62 PX |
1308 | /* |
1309 | * Return true if the vma needs to copy the pgtable during this fork(). Return | |
1310 | * false when we can speed up fork() by allowing lazy page faults later until | |
1311 | * when the child accesses the memory range. | |
1312 | */ | |
bc70fbf2 | 1313 | static bool |
c56d1b62 PX |
1314 | vma_needs_copy(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) |
1315 | { | |
1316 | /* | |
1317 | * Always copy pgtables when dst_vma has uffd-wp enabled even if it's | |
1318 | * file-backed (e.g. shmem). Because when uffd-wp is enabled, pgtable | |
1319 | * contains uffd-wp protection information, that's something we can't | |
1320 | * retrieve from page cache, and skip copying will lose those info. | |
1321 | */ | |
1322 | if (userfaultfd_wp(dst_vma)) | |
1323 | return true; | |
1324 | ||
bcd51a3c | 1325 | if (src_vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
c56d1b62 PX |
1326 | return true; |
1327 | ||
1328 | if (src_vma->anon_vma) | |
1329 | return true; | |
1330 | ||
1331 | /* | |
1332 | * Don't copy ptes where a page fault will fill them correctly. Fork | |
1333 | * becomes much lighter when there are big shared or private readonly | |
1334 | * mappings. The tradeoff is that copy_page_range is more efficient | |
1335 | * than faulting. | |
1336 | */ | |
1337 | return false; | |
1338 | } | |
1339 | ||
c78f4636 PX |
1340 | int |
1341 | copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) | |
1da177e4 LT |
1342 | { |
1343 | pgd_t *src_pgd, *dst_pgd; | |
1344 | unsigned long next; | |
c78f4636 PX |
1345 | unsigned long addr = src_vma->vm_start; |
1346 | unsigned long end = src_vma->vm_end; | |
1347 | struct mm_struct *dst_mm = dst_vma->vm_mm; | |
1348 | struct mm_struct *src_mm = src_vma->vm_mm; | |
ac46d4f3 | 1349 | struct mmu_notifier_range range; |
2ec74c3e | 1350 | bool is_cow; |
cddb8a5c | 1351 | int ret; |
1da177e4 | 1352 | |
c56d1b62 | 1353 | if (!vma_needs_copy(dst_vma, src_vma)) |
0661a336 | 1354 | return 0; |
d992895b | 1355 | |
c78f4636 | 1356 | if (is_vm_hugetlb_page(src_vma)) |
bc70fbf2 | 1357 | return copy_hugetlb_page_range(dst_mm, src_mm, dst_vma, src_vma); |
1da177e4 | 1358 | |
c78f4636 | 1359 | if (unlikely(src_vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1360 | /* |
1361 | * We do not free on error cases below as remove_vma | |
1362 | * gets called on error from higher level routine | |
1363 | */ | |
c78f4636 | 1364 | ret = track_pfn_copy(src_vma); |
2ab64037 | 1365 | if (ret) |
1366 | return ret; | |
1367 | } | |
1368 | ||
cddb8a5c AA |
1369 | /* |
1370 | * We need to invalidate the secondary MMU mappings only when | |
1371 | * there could be a permission downgrade on the ptes of the | |
1372 | * parent mm. And a permission downgrade will only happen if | |
1373 | * is_cow_mapping() returns true. | |
1374 | */ | |
c78f4636 | 1375 | is_cow = is_cow_mapping(src_vma->vm_flags); |
ac46d4f3 JG |
1376 | |
1377 | if (is_cow) { | |
7269f999 | 1378 | mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_PAGE, |
7d4a8be0 | 1379 | 0, src_mm, addr, end); |
ac46d4f3 | 1380 | mmu_notifier_invalidate_range_start(&range); |
57efa1fe JG |
1381 | /* |
1382 | * Disabling preemption is not needed for the write side, as | |
1383 | * the read side doesn't spin, but goes to the mmap_lock. | |
1384 | * | |
1385 | * Use the raw variant of the seqcount_t write API to avoid | |
1386 | * lockdep complaining about preemptibility. | |
1387 | */ | |
e727bfd5 | 1388 | vma_assert_write_locked(src_vma); |
57efa1fe | 1389 | raw_write_seqcount_begin(&src_mm->write_protect_seq); |
ac46d4f3 | 1390 | } |
cddb8a5c AA |
1391 | |
1392 | ret = 0; | |
1da177e4 LT |
1393 | dst_pgd = pgd_offset(dst_mm, addr); |
1394 | src_pgd = pgd_offset(src_mm, addr); | |
1395 | do { | |
1396 | next = pgd_addr_end(addr, end); | |
1397 | if (pgd_none_or_clear_bad(src_pgd)) | |
1398 | continue; | |
c78f4636 PX |
1399 | if (unlikely(copy_p4d_range(dst_vma, src_vma, dst_pgd, src_pgd, |
1400 | addr, next))) { | |
d155df53 | 1401 | untrack_pfn_clear(dst_vma); |
cddb8a5c AA |
1402 | ret = -ENOMEM; |
1403 | break; | |
1404 | } | |
1da177e4 | 1405 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1406 | |
57efa1fe JG |
1407 | if (is_cow) { |
1408 | raw_write_seqcount_end(&src_mm->write_protect_seq); | |
ac46d4f3 | 1409 | mmu_notifier_invalidate_range_end(&range); |
57efa1fe | 1410 | } |
cddb8a5c | 1411 | return ret; |
1da177e4 LT |
1412 | } |
1413 | ||
5abfd71d PX |
1414 | /* Whether we should zap all COWed (private) pages too */ |
1415 | static inline bool should_zap_cows(struct zap_details *details) | |
1416 | { | |
1417 | /* By default, zap all pages */ | |
1418 | if (!details) | |
1419 | return true; | |
1420 | ||
1421 | /* Or, we zap COWed pages only if the caller wants to */ | |
2e148f1e | 1422 | return details->even_cows; |
5abfd71d PX |
1423 | } |
1424 | ||
eabafaaa KW |
1425 | /* Decides whether we should zap this folio with the folio pointer specified */ |
1426 | static inline bool should_zap_folio(struct zap_details *details, | |
1427 | struct folio *folio) | |
3506659e | 1428 | { |
eabafaaa | 1429 | /* If we can make a decision without *folio.. */ |
5abfd71d | 1430 | if (should_zap_cows(details)) |
254ab940 | 1431 | return true; |
5abfd71d | 1432 | |
eabafaaa KW |
1433 | /* Otherwise we should only zap non-anon folios */ |
1434 | return !folio_test_anon(folio); | |
3506659e MWO |
1435 | } |
1436 | ||
999dad82 PX |
1437 | static inline bool zap_drop_file_uffd_wp(struct zap_details *details) |
1438 | { | |
1439 | if (!details) | |
1440 | return false; | |
1441 | ||
1442 | return details->zap_flags & ZAP_FLAG_DROP_MARKER; | |
1443 | } | |
1444 | ||
1445 | /* | |
1446 | * This function makes sure that we'll replace the none pte with an uffd-wp | |
1447 | * swap special pte marker when necessary. Must be with the pgtable lock held. | |
1448 | */ | |
1449 | static inline void | |
1450 | zap_install_uffd_wp_if_needed(struct vm_area_struct *vma, | |
10ebac4f | 1451 | unsigned long addr, pte_t *pte, int nr, |
999dad82 PX |
1452 | struct zap_details *details, pte_t pteval) |
1453 | { | |
2bad466c PX |
1454 | /* Zap on anonymous always means dropping everything */ |
1455 | if (vma_is_anonymous(vma)) | |
1456 | return; | |
1457 | ||
999dad82 PX |
1458 | if (zap_drop_file_uffd_wp(details)) |
1459 | return; | |
1460 | ||
10ebac4f DH |
1461 | for (;;) { |
1462 | /* the PFN in the PTE is irrelevant. */ | |
1463 | pte_install_uffd_wp_if_needed(vma, addr, pte, pteval); | |
1464 | if (--nr == 0) | |
1465 | break; | |
1466 | pte++; | |
1467 | addr += PAGE_SIZE; | |
1468 | } | |
999dad82 PX |
1469 | } |
1470 | ||
10ebac4f | 1471 | static __always_inline void zap_present_folio_ptes(struct mmu_gather *tlb, |
2b42a7e5 | 1472 | struct vm_area_struct *vma, struct folio *folio, |
10ebac4f DH |
1473 | struct page *page, pte_t *pte, pte_t ptent, unsigned int nr, |
1474 | unsigned long addr, struct zap_details *details, int *rss, | |
1475 | bool *force_flush, bool *force_break) | |
789753e1 DH |
1476 | { |
1477 | struct mm_struct *mm = tlb->mm; | |
789753e1 | 1478 | bool delay_rmap = false; |
789753e1 DH |
1479 | |
1480 | if (!folio_test_anon(folio)) { | |
10ebac4f | 1481 | ptent = get_and_clear_full_ptes(mm, addr, pte, nr, tlb->fullmm); |
789753e1 DH |
1482 | if (pte_dirty(ptent)) { |
1483 | folio_mark_dirty(folio); | |
1484 | if (tlb_delay_rmap(tlb)) { | |
1485 | delay_rmap = true; | |
1486 | *force_flush = true; | |
1487 | } | |
1488 | } | |
1489 | if (pte_young(ptent) && likely(vma_has_recency(vma))) | |
1490 | folio_mark_accessed(folio); | |
10ebac4f | 1491 | rss[mm_counter(folio)] -= nr; |
d11838ed DH |
1492 | } else { |
1493 | /* We don't need up-to-date accessed/dirty bits. */ | |
10ebac4f DH |
1494 | clear_full_ptes(mm, addr, pte, nr, tlb->fullmm); |
1495 | rss[MM_ANONPAGES] -= nr; | |
789753e1 | 1496 | } |
10ebac4f | 1497 | /* Checking a single PTE in a batch is sufficient. */ |
d11838ed | 1498 | arch_check_zapped_pte(vma, ptent); |
10ebac4f | 1499 | tlb_remove_tlb_entries(tlb, pte, nr, addr); |
d11838ed | 1500 | if (unlikely(userfaultfd_pte_wp(vma, ptent))) |
10ebac4f DH |
1501 | zap_install_uffd_wp_if_needed(vma, addr, pte, nr, details, |
1502 | ptent); | |
d11838ed | 1503 | |
789753e1 | 1504 | if (!delay_rmap) { |
10ebac4f DH |
1505 | folio_remove_rmap_ptes(folio, page, nr, vma); |
1506 | ||
3aeea4fc | 1507 | if (unlikely(folio_mapcount(folio) < 0)) |
789753e1 DH |
1508 | print_bad_pte(vma, addr, ptent, page); |
1509 | } | |
ba42b524 YJN |
1510 | |
1511 | if (want_init_mlocked_on_free() && folio_test_mlocked(folio) && | |
1512 | !delay_rmap && folio_test_anon(folio)) { | |
1513 | kernel_init_pages(page, folio_nr_pages(folio)); | |
1514 | } | |
1515 | ||
10ebac4f | 1516 | if (unlikely(__tlb_remove_folio_pages(tlb, page, nr, delay_rmap))) { |
789753e1 DH |
1517 | *force_flush = true; |
1518 | *force_break = true; | |
1519 | } | |
1520 | } | |
1521 | ||
10ebac4f DH |
1522 | /* |
1523 | * Zap or skip at least one present PTE, trying to batch-process subsequent | |
1524 | * PTEs that map consecutive pages of the same folio. | |
1525 | * | |
1526 | * Returns the number of processed (skipped or zapped) PTEs (at least 1). | |
1527 | */ | |
1528 | static inline int zap_present_ptes(struct mmu_gather *tlb, | |
2b42a7e5 | 1529 | struct vm_area_struct *vma, pte_t *pte, pte_t ptent, |
10ebac4f DH |
1530 | unsigned int max_nr, unsigned long addr, |
1531 | struct zap_details *details, int *rss, bool *force_flush, | |
1532 | bool *force_break) | |
2b42a7e5 | 1533 | { |
10ebac4f | 1534 | const fpb_t fpb_flags = FPB_IGNORE_DIRTY | FPB_IGNORE_SOFT_DIRTY; |
2b42a7e5 DH |
1535 | struct mm_struct *mm = tlb->mm; |
1536 | struct folio *folio; | |
1537 | struct page *page; | |
10ebac4f | 1538 | int nr; |
2b42a7e5 DH |
1539 | |
1540 | page = vm_normal_page(vma, addr, ptent); | |
1541 | if (!page) { | |
1542 | /* We don't need up-to-date accessed/dirty bits. */ | |
1543 | ptep_get_and_clear_full(mm, addr, pte, tlb->fullmm); | |
1544 | arch_check_zapped_pte(vma, ptent); | |
1545 | tlb_remove_tlb_entry(tlb, pte, addr); | |
f8572367 PX |
1546 | if (userfaultfd_pte_wp(vma, ptent)) |
1547 | zap_install_uffd_wp_if_needed(vma, addr, pte, 1, | |
1548 | details, ptent); | |
2b42a7e5 | 1549 | ksm_might_unmap_zero_page(mm, ptent); |
10ebac4f | 1550 | return 1; |
2b42a7e5 DH |
1551 | } |
1552 | ||
1553 | folio = page_folio(page); | |
1554 | if (unlikely(!should_zap_folio(details, folio))) | |
10ebac4f DH |
1555 | return 1; |
1556 | ||
1557 | /* | |
1558 | * Make sure that the common "small folio" case is as fast as possible | |
1559 | * by keeping the batching logic separate. | |
1560 | */ | |
1561 | if (unlikely(folio_test_large(folio) && max_nr != 1)) { | |
1562 | nr = folio_pte_batch(folio, addr, pte, ptent, max_nr, fpb_flags, | |
96ebdb03 | 1563 | NULL, NULL, NULL); |
10ebac4f DH |
1564 | |
1565 | zap_present_folio_ptes(tlb, vma, folio, page, pte, ptent, nr, | |
1566 | addr, details, rss, force_flush, | |
1567 | force_break); | |
1568 | return nr; | |
1569 | } | |
1570 | zap_present_folio_ptes(tlb, vma, folio, page, pte, ptent, 1, addr, | |
1571 | details, rss, force_flush, force_break); | |
1572 | return 1; | |
2b42a7e5 DH |
1573 | } |
1574 | ||
51c6f666 | 1575 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1576 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1577 | unsigned long addr, unsigned long end, |
97a89413 | 1578 | struct zap_details *details) |
1da177e4 | 1579 | { |
789753e1 | 1580 | bool force_flush = false, force_break = false; |
b5810039 | 1581 | struct mm_struct *mm = tlb->mm; |
d559db08 | 1582 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1583 | spinlock_t *ptl; |
5f1a1907 | 1584 | pte_t *start_pte; |
97a89413 | 1585 | pte_t *pte; |
8a5f14a2 | 1586 | swp_entry_t entry; |
10ebac4f | 1587 | int nr; |
d559db08 | 1588 | |
ed6a7935 | 1589 | tlb_change_page_size(tlb, PAGE_SIZE); |
e303297e | 1590 | init_rss_vec(rss); |
3db82b93 HD |
1591 | start_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1592 | if (!pte) | |
1593 | return addr; | |
1594 | ||
3ea27719 | 1595 | flush_tlb_batched_pending(mm); |
6606c3e0 | 1596 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 1597 | do { |
c33c7948 | 1598 | pte_t ptent = ptep_get(pte); |
789753e1 | 1599 | struct folio *folio; |
8018db85 | 1600 | struct page *page; |
10ebac4f | 1601 | int max_nr; |
8018db85 | 1602 | |
10ebac4f | 1603 | nr = 1; |
166f61b9 | 1604 | if (pte_none(ptent)) |
1da177e4 | 1605 | continue; |
6f5e6b9e | 1606 | |
7b167b68 MK |
1607 | if (need_resched()) |
1608 | break; | |
1609 | ||
1da177e4 | 1610 | if (pte_present(ptent)) { |
10ebac4f DH |
1611 | max_nr = (end - addr) / PAGE_SIZE; |
1612 | nr = zap_present_ptes(tlb, vma, pte, ptent, max_nr, | |
1613 | addr, details, rss, &force_flush, | |
1614 | &force_break); | |
789753e1 | 1615 | if (unlikely(force_break)) { |
10ebac4f | 1616 | addr += nr * PAGE_SIZE; |
d16dfc55 | 1617 | break; |
1cf35d47 | 1618 | } |
1da177e4 LT |
1619 | continue; |
1620 | } | |
5042db43 JG |
1621 | |
1622 | entry = pte_to_swp_entry(ptent); | |
b756a3b5 AP |
1623 | if (is_device_private_entry(entry) || |
1624 | is_device_exclusive_entry(entry)) { | |
8018db85 | 1625 | page = pfn_swap_entry_to_page(entry); |
c4626503 | 1626 | folio = page_folio(page); |
eabafaaa | 1627 | if (unlikely(!should_zap_folio(details, folio))) |
91b61ef3 | 1628 | continue; |
999dad82 PX |
1629 | /* |
1630 | * Both device private/exclusive mappings should only | |
1631 | * work with anonymous page so far, so we don't need to | |
1632 | * consider uffd-wp bit when zap. For more information, | |
1633 | * see zap_install_uffd_wp_if_needed(). | |
1634 | */ | |
1635 | WARN_ON_ONCE(!vma_is_anonymous(vma)); | |
a23f517b | 1636 | rss[mm_counter(folio)]--; |
b756a3b5 | 1637 | if (is_device_private_entry(entry)) |
c4626503 DH |
1638 | folio_remove_rmap_pte(folio, page, vma); |
1639 | folio_put(folio); | |
8018db85 | 1640 | } else if (!non_swap_entry(entry)) { |
a62fb92a RR |
1641 | max_nr = (end - addr) / PAGE_SIZE; |
1642 | nr = swap_pte_batch(pte, max_nr, ptent); | |
1643 | /* Genuine swap entries, hence a private anon pages */ | |
5abfd71d PX |
1644 | if (!should_zap_cows(details)) |
1645 | continue; | |
a62fb92a RR |
1646 | rss[MM_SWAPENTS] -= nr; |
1647 | free_swap_and_cache_nr(entry, nr); | |
5abfd71d | 1648 | } else if (is_migration_entry(entry)) { |
eabafaaa KW |
1649 | folio = pfn_swap_entry_folio(entry); |
1650 | if (!should_zap_folio(details, folio)) | |
5abfd71d | 1651 | continue; |
a23f517b | 1652 | rss[mm_counter(folio)]--; |
999dad82 | 1653 | } else if (pte_marker_entry_uffd_wp(entry)) { |
2bad466c PX |
1654 | /* |
1655 | * For anon: always drop the marker; for file: only | |
1656 | * drop the marker if explicitly requested. | |
1657 | */ | |
1658 | if (!vma_is_anonymous(vma) && | |
1659 | !zap_drop_file_uffd_wp(details)) | |
999dad82 | 1660 | continue; |
9f186f9e | 1661 | } else if (is_hwpoison_entry(entry) || |
af19487f | 1662 | is_poisoned_swp_entry(entry)) { |
5abfd71d PX |
1663 | if (!should_zap_cows(details)) |
1664 | continue; | |
1665 | } else { | |
1666 | /* We should have covered all the swap entry types */ | |
727d16f1 | 1667 | pr_alert("unrecognized swap entry 0x%lx\n", entry.val); |
5abfd71d | 1668 | WARN_ON_ONCE(1); |
b084d435 | 1669 | } |
a62fb92a RR |
1670 | clear_not_present_full_ptes(mm, addr, pte, nr, tlb->fullmm); |
1671 | zap_install_uffd_wp_if_needed(vma, addr, pte, nr, details, ptent); | |
10ebac4f | 1672 | } while (pte += nr, addr += PAGE_SIZE * nr, addr != end); |
ae859762 | 1673 | |
d559db08 | 1674 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1675 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1676 | |
1cf35d47 | 1677 | /* Do the actual TLB flush before dropping ptl */ |
5df397de | 1678 | if (force_flush) { |
1cf35d47 | 1679 | tlb_flush_mmu_tlbonly(tlb); |
f036c818 | 1680 | tlb_flush_rmaps(tlb, vma); |
5df397de | 1681 | } |
1cf35d47 LT |
1682 | pte_unmap_unlock(start_pte, ptl); |
1683 | ||
1684 | /* | |
1685 | * If we forced a TLB flush (either due to running out of | |
1686 | * batch buffers or because we needed to flush dirty TLB | |
1687 | * entries before releasing the ptl), free the batched | |
3db82b93 | 1688 | * memory too. Come back again if we didn't do everything. |
1cf35d47 | 1689 | */ |
3db82b93 | 1690 | if (force_flush) |
fa0aafb8 | 1691 | tlb_flush_mmu(tlb); |
d16dfc55 | 1692 | |
51c6f666 | 1693 | return addr; |
1da177e4 LT |
1694 | } |
1695 | ||
51c6f666 | 1696 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1697 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1698 | unsigned long addr, unsigned long end, |
97a89413 | 1699 | struct zap_details *details) |
1da177e4 LT |
1700 | { |
1701 | pmd_t *pmd; | |
1702 | unsigned long next; | |
1703 | ||
1704 | pmd = pmd_offset(pud, addr); | |
1705 | do { | |
1706 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 1707 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
53406ed1 | 1708 | if (next - addr != HPAGE_PMD_SIZE) |
fd60775a | 1709 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
3db82b93 HD |
1710 | else if (zap_huge_pmd(tlb, vma, pmd, addr)) { |
1711 | addr = next; | |
1712 | continue; | |
1713 | } | |
71e3aac0 | 1714 | /* fall through */ |
3506659e MWO |
1715 | } else if (details && details->single_folio && |
1716 | folio_test_pmd_mappable(details->single_folio) && | |
22061a1f HD |
1717 | next - addr == HPAGE_PMD_SIZE && pmd_none(*pmd)) { |
1718 | spinlock_t *ptl = pmd_lock(tlb->mm, pmd); | |
1719 | /* | |
1720 | * Take and drop THP pmd lock so that we cannot return | |
1721 | * prematurely, while zap_huge_pmd() has cleared *pmd, | |
1722 | * but not yet decremented compound_mapcount(). | |
1723 | */ | |
1724 | spin_unlock(ptl); | |
71e3aac0 | 1725 | } |
3db82b93 HD |
1726 | if (pmd_none(*pmd)) { |
1727 | addr = next; | |
1728 | continue; | |
1729 | } | |
1730 | addr = zap_pte_range(tlb, vma, pmd, addr, next, details); | |
1731 | if (addr != next) | |
1732 | pmd--; | |
1733 | } while (pmd++, cond_resched(), addr != end); | |
51c6f666 RH |
1734 | |
1735 | return addr; | |
1da177e4 LT |
1736 | } |
1737 | ||
51c6f666 | 1738 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
c2febafc | 1739 | struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1740 | unsigned long addr, unsigned long end, |
97a89413 | 1741 | struct zap_details *details) |
1da177e4 LT |
1742 | { |
1743 | pud_t *pud; | |
1744 | unsigned long next; | |
1745 | ||
c2febafc | 1746 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1747 | do { |
1748 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1749 | if (pud_trans_huge(*pud) || pud_devmap(*pud)) { |
1750 | if (next - addr != HPAGE_PUD_SIZE) { | |
42fc5414 | 1751 | mmap_assert_locked(tlb->mm); |
a00cc7d9 MW |
1752 | split_huge_pud(vma, pud, addr); |
1753 | } else if (zap_huge_pud(tlb, vma, pud, addr)) | |
1754 | goto next; | |
1755 | /* fall through */ | |
1756 | } | |
97a89413 | 1757 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1758 | continue; |
97a89413 | 1759 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
a00cc7d9 MW |
1760 | next: |
1761 | cond_resched(); | |
97a89413 | 1762 | } while (pud++, addr = next, addr != end); |
51c6f666 RH |
1763 | |
1764 | return addr; | |
1da177e4 LT |
1765 | } |
1766 | ||
c2febafc KS |
1767 | static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, |
1768 | struct vm_area_struct *vma, pgd_t *pgd, | |
1769 | unsigned long addr, unsigned long end, | |
1770 | struct zap_details *details) | |
1771 | { | |
1772 | p4d_t *p4d; | |
1773 | unsigned long next; | |
1774 | ||
1775 | p4d = p4d_offset(pgd, addr); | |
1776 | do { | |
1777 | next = p4d_addr_end(addr, end); | |
1778 | if (p4d_none_or_clear_bad(p4d)) | |
1779 | continue; | |
1780 | next = zap_pud_range(tlb, vma, p4d, addr, next, details); | |
1781 | } while (p4d++, addr = next, addr != end); | |
1782 | ||
1783 | return addr; | |
1784 | } | |
1785 | ||
aac45363 | 1786 | void unmap_page_range(struct mmu_gather *tlb, |
038c7aa1 AV |
1787 | struct vm_area_struct *vma, |
1788 | unsigned long addr, unsigned long end, | |
1789 | struct zap_details *details) | |
1da177e4 LT |
1790 | { |
1791 | pgd_t *pgd; | |
1792 | unsigned long next; | |
1793 | ||
1da177e4 LT |
1794 | BUG_ON(addr >= end); |
1795 | tlb_start_vma(tlb, vma); | |
1796 | pgd = pgd_offset(vma->vm_mm, addr); | |
1797 | do { | |
1798 | next = pgd_addr_end(addr, end); | |
97a89413 | 1799 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1800 | continue; |
c2febafc | 1801 | next = zap_p4d_range(tlb, vma, pgd, addr, next, details); |
97a89413 | 1802 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
1803 | tlb_end_vma(tlb, vma); |
1804 | } | |
51c6f666 | 1805 | |
f5cc4eef AV |
1806 | |
1807 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1808 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1809 | unsigned long end_addr, |
68f48381 | 1810 | struct zap_details *details, bool mm_wr_locked) |
f5cc4eef AV |
1811 | { |
1812 | unsigned long start = max(vma->vm_start, start_addr); | |
1813 | unsigned long end; | |
1814 | ||
1815 | if (start >= vma->vm_end) | |
1816 | return; | |
1817 | end = min(vma->vm_end, end_addr); | |
1818 | if (end <= vma->vm_start) | |
1819 | return; | |
1820 | ||
cbc91f71 SD |
1821 | if (vma->vm_file) |
1822 | uprobe_munmap(vma, start, end); | |
1823 | ||
b3b9c293 | 1824 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
68f48381 | 1825 | untrack_pfn(vma, 0, 0, mm_wr_locked); |
f5cc4eef AV |
1826 | |
1827 | if (start != end) { | |
1828 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1829 | /* | |
1830 | * It is undesirable to test vma->vm_file as it | |
1831 | * should be non-null for valid hugetlb area. | |
1832 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1833 | * cleanup path of mmap_region. When |
f5cc4eef | 1834 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1835 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1836 | * before calling this function to clean up. |
1837 | * Since no pte has actually been setup, it is | |
1838 | * safe to do nothing in this case. | |
1839 | */ | |
24669e58 | 1840 | if (vma->vm_file) { |
05e90bd0 PX |
1841 | zap_flags_t zap_flags = details ? |
1842 | details->zap_flags : 0; | |
2820b0f0 | 1843 | __unmap_hugepage_range(tlb, vma, start, end, |
05e90bd0 | 1844 | NULL, zap_flags); |
24669e58 | 1845 | } |
f5cc4eef AV |
1846 | } else |
1847 | unmap_page_range(tlb, vma, start, end, details); | |
1848 | } | |
1da177e4 LT |
1849 | } |
1850 | ||
1da177e4 LT |
1851 | /** |
1852 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1853 | * @tlb: address of the caller's struct mmu_gather |
6e412203 | 1854 | * @mas: the maple state |
1da177e4 LT |
1855 | * @vma: the starting vma |
1856 | * @start_addr: virtual address at which to start unmapping | |
1857 | * @end_addr: virtual address at which to end unmapping | |
6e412203 | 1858 | * @tree_end: The maximum index to check |
809ef83c | 1859 | * @mm_wr_locked: lock flag |
1da177e4 | 1860 | * |
508034a3 | 1861 | * Unmap all pages in the vma list. |
1da177e4 | 1862 | * |
1da177e4 LT |
1863 | * Only addresses between `start' and `end' will be unmapped. |
1864 | * | |
1865 | * The VMA list must be sorted in ascending virtual address order. | |
1866 | * | |
1867 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1868 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1869 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1870 | * drops the lock and schedules. | |
1871 | */ | |
fd892593 | 1872 | void unmap_vmas(struct mmu_gather *tlb, struct ma_state *mas, |
1da177e4 | 1873 | struct vm_area_struct *vma, unsigned long start_addr, |
fd892593 LH |
1874 | unsigned long end_addr, unsigned long tree_end, |
1875 | bool mm_wr_locked) | |
1da177e4 | 1876 | { |
ac46d4f3 | 1877 | struct mmu_notifier_range range; |
999dad82 | 1878 | struct zap_details details = { |
04ada095 | 1879 | .zap_flags = ZAP_FLAG_DROP_MARKER | ZAP_FLAG_UNMAP, |
999dad82 PX |
1880 | /* Careful - we need to zap private pages too! */ |
1881 | .even_cows = true, | |
1882 | }; | |
1da177e4 | 1883 | |
7d4a8be0 | 1884 | mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma->vm_mm, |
6f4f13e8 | 1885 | start_addr, end_addr); |
ac46d4f3 | 1886 | mmu_notifier_invalidate_range_start(&range); |
763ecb03 | 1887 | do { |
2820b0f0 RR |
1888 | unsigned long start = start_addr; |
1889 | unsigned long end = end_addr; | |
1890 | hugetlb_zap_begin(vma, &start, &end); | |
1891 | unmap_single_vma(tlb, vma, start, end, &details, | |
68f48381 | 1892 | mm_wr_locked); |
2820b0f0 | 1893 | hugetlb_zap_end(vma, &details); |
d2406291 PZ |
1894 | vma = mas_find(mas, tree_end - 1); |
1895 | } while (vma && likely(!xa_is_zero(vma))); | |
ac46d4f3 | 1896 | mmu_notifier_invalidate_range_end(&range); |
1da177e4 LT |
1897 | } |
1898 | ||
f5cc4eef AV |
1899 | /** |
1900 | * zap_page_range_single - remove user pages in a given range | |
1901 | * @vma: vm_area_struct holding the applicable pages | |
1902 | * @address: starting address of pages to zap | |
1903 | * @size: number of bytes to zap | |
8a5f14a2 | 1904 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1905 | * |
1906 | * The range must fit into one VMA. | |
1da177e4 | 1907 | */ |
21b85b09 | 1908 | void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1909 | unsigned long size, struct zap_details *details) |
1910 | { | |
21b85b09 | 1911 | const unsigned long end = address + size; |
ac46d4f3 | 1912 | struct mmu_notifier_range range; |
d16dfc55 | 1913 | struct mmu_gather tlb; |
1da177e4 | 1914 | |
1da177e4 | 1915 | lru_add_drain(); |
7d4a8be0 | 1916 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma->vm_mm, |
21b85b09 | 1917 | address, end); |
2820b0f0 | 1918 | hugetlb_zap_begin(vma, &range.start, &range.end); |
a72afd87 | 1919 | tlb_gather_mmu(&tlb, vma->vm_mm); |
ac46d4f3 JG |
1920 | update_hiwater_rss(vma->vm_mm); |
1921 | mmu_notifier_invalidate_range_start(&range); | |
21b85b09 MK |
1922 | /* |
1923 | * unmap 'address-end' not 'range.start-range.end' as range | |
1924 | * could have been expanded for hugetlb pmd sharing. | |
1925 | */ | |
68f48381 | 1926 | unmap_single_vma(&tlb, vma, address, end, details, false); |
ac46d4f3 | 1927 | mmu_notifier_invalidate_range_end(&range); |
ae8eba8b | 1928 | tlb_finish_mmu(&tlb); |
2820b0f0 | 1929 | hugetlb_zap_end(vma, details); |
1da177e4 LT |
1930 | } |
1931 | ||
c627f9cc JS |
1932 | /** |
1933 | * zap_vma_ptes - remove ptes mapping the vma | |
1934 | * @vma: vm_area_struct holding ptes to be zapped | |
1935 | * @address: starting address of pages to zap | |
1936 | * @size: number of bytes to zap | |
1937 | * | |
1938 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1939 | * | |
1940 | * The entire address range must be fully contained within the vma. | |
1941 | * | |
c627f9cc | 1942 | */ |
27d036e3 | 1943 | void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, |
c627f9cc JS |
1944 | unsigned long size) |
1945 | { | |
88a35912 | 1946 | if (!range_in_vma(vma, address, address + size) || |
c627f9cc | 1947 | !(vma->vm_flags & VM_PFNMAP)) |
27d036e3 LR |
1948 | return; |
1949 | ||
f5cc4eef | 1950 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1951 | } |
1952 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1953 | ||
8cd3984d | 1954 | static pmd_t *walk_to_pmd(struct mm_struct *mm, unsigned long addr) |
c9cfcddf | 1955 | { |
c2febafc KS |
1956 | pgd_t *pgd; |
1957 | p4d_t *p4d; | |
1958 | pud_t *pud; | |
1959 | pmd_t *pmd; | |
1960 | ||
1961 | pgd = pgd_offset(mm, addr); | |
1962 | p4d = p4d_alloc(mm, pgd, addr); | |
1963 | if (!p4d) | |
1964 | return NULL; | |
1965 | pud = pud_alloc(mm, p4d, addr); | |
1966 | if (!pud) | |
1967 | return NULL; | |
1968 | pmd = pmd_alloc(mm, pud, addr); | |
1969 | if (!pmd) | |
1970 | return NULL; | |
1971 | ||
1972 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
8cd3984d AR |
1973 | return pmd; |
1974 | } | |
1975 | ||
1976 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, | |
1977 | spinlock_t **ptl) | |
1978 | { | |
1979 | pmd_t *pmd = walk_to_pmd(mm, addr); | |
1980 | ||
1981 | if (!pmd) | |
1982 | return NULL; | |
c2febafc | 1983 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
c9cfcddf LT |
1984 | } |
1985 | ||
8efd6f5b AR |
1986 | static int validate_page_before_insert(struct page *page) |
1987 | { | |
f8b6187d KW |
1988 | struct folio *folio = page_folio(page); |
1989 | ||
1990 | if (folio_test_anon(folio) || folio_test_slab(folio) || | |
1991 | page_has_type(page)) | |
8efd6f5b | 1992 | return -EINVAL; |
f8b6187d | 1993 | flush_dcache_folio(folio); |
8efd6f5b AR |
1994 | return 0; |
1995 | } | |
1996 | ||
cea86fe2 | 1997 | static int insert_page_into_pte_locked(struct vm_area_struct *vma, pte_t *pte, |
8efd6f5b AR |
1998 | unsigned long addr, struct page *page, pgprot_t prot) |
1999 | { | |
ef37b2ea DH |
2000 | struct folio *folio = page_folio(page); |
2001 | ||
c33c7948 | 2002 | if (!pte_none(ptep_get(pte))) |
8efd6f5b AR |
2003 | return -EBUSY; |
2004 | /* Ok, finally just insert the thing.. */ | |
ef37b2ea | 2005 | folio_get(folio); |
6b27cc6c | 2006 | inc_mm_counter(vma->vm_mm, mm_counter_file(folio)); |
ef37b2ea | 2007 | folio_add_file_rmap_pte(folio, page, vma); |
cea86fe2 | 2008 | set_pte_at(vma->vm_mm, addr, pte, mk_pte(page, prot)); |
8efd6f5b AR |
2009 | return 0; |
2010 | } | |
2011 | ||
238f58d8 LT |
2012 | /* |
2013 | * This is the old fallback for page remapping. | |
2014 | * | |
2015 | * For historical reasons, it only allows reserved pages. Only | |
2016 | * old drivers should use this, and they needed to mark their | |
2017 | * pages reserved for the old functions anyway. | |
2018 | */ | |
423bad60 NP |
2019 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
2020 | struct page *page, pgprot_t prot) | |
238f58d8 LT |
2021 | { |
2022 | int retval; | |
c9cfcddf | 2023 | pte_t *pte; |
8a9f3ccd BS |
2024 | spinlock_t *ptl; |
2025 | ||
8efd6f5b AR |
2026 | retval = validate_page_before_insert(page); |
2027 | if (retval) | |
5b4e655e | 2028 | goto out; |
238f58d8 | 2029 | retval = -ENOMEM; |
cea86fe2 | 2030 | pte = get_locked_pte(vma->vm_mm, addr, &ptl); |
238f58d8 | 2031 | if (!pte) |
5b4e655e | 2032 | goto out; |
cea86fe2 | 2033 | retval = insert_page_into_pte_locked(vma, pte, addr, page, prot); |
238f58d8 LT |
2034 | pte_unmap_unlock(pte, ptl); |
2035 | out: | |
2036 | return retval; | |
2037 | } | |
2038 | ||
cea86fe2 | 2039 | static int insert_page_in_batch_locked(struct vm_area_struct *vma, pte_t *pte, |
8cd3984d AR |
2040 | unsigned long addr, struct page *page, pgprot_t prot) |
2041 | { | |
2042 | int err; | |
2043 | ||
2044 | if (!page_count(page)) | |
2045 | return -EINVAL; | |
2046 | err = validate_page_before_insert(page); | |
7f70c2a6 AR |
2047 | if (err) |
2048 | return err; | |
cea86fe2 | 2049 | return insert_page_into_pte_locked(vma, pte, addr, page, prot); |
8cd3984d AR |
2050 | } |
2051 | ||
2052 | /* insert_pages() amortizes the cost of spinlock operations | |
bb7dbaaf | 2053 | * when inserting pages in a loop. |
8cd3984d AR |
2054 | */ |
2055 | static int insert_pages(struct vm_area_struct *vma, unsigned long addr, | |
2056 | struct page **pages, unsigned long *num, pgprot_t prot) | |
2057 | { | |
2058 | pmd_t *pmd = NULL; | |
7f70c2a6 AR |
2059 | pte_t *start_pte, *pte; |
2060 | spinlock_t *pte_lock; | |
8cd3984d AR |
2061 | struct mm_struct *const mm = vma->vm_mm; |
2062 | unsigned long curr_page_idx = 0; | |
2063 | unsigned long remaining_pages_total = *num; | |
2064 | unsigned long pages_to_write_in_pmd; | |
2065 | int ret; | |
2066 | more: | |
2067 | ret = -EFAULT; | |
2068 | pmd = walk_to_pmd(mm, addr); | |
2069 | if (!pmd) | |
2070 | goto out; | |
2071 | ||
2072 | pages_to_write_in_pmd = min_t(unsigned long, | |
2073 | remaining_pages_total, PTRS_PER_PTE - pte_index(addr)); | |
2074 | ||
2075 | /* Allocate the PTE if necessary; takes PMD lock once only. */ | |
2076 | ret = -ENOMEM; | |
2077 | if (pte_alloc(mm, pmd)) | |
2078 | goto out; | |
8cd3984d AR |
2079 | |
2080 | while (pages_to_write_in_pmd) { | |
2081 | int pte_idx = 0; | |
2082 | const int batch_size = min_t(int, pages_to_write_in_pmd, 8); | |
2083 | ||
7f70c2a6 | 2084 | start_pte = pte_offset_map_lock(mm, pmd, addr, &pte_lock); |
3db82b93 HD |
2085 | if (!start_pte) { |
2086 | ret = -EFAULT; | |
2087 | goto out; | |
2088 | } | |
7f70c2a6 | 2089 | for (pte = start_pte; pte_idx < batch_size; ++pte, ++pte_idx) { |
cea86fe2 | 2090 | int err = insert_page_in_batch_locked(vma, pte, |
8cd3984d AR |
2091 | addr, pages[curr_page_idx], prot); |
2092 | if (unlikely(err)) { | |
7f70c2a6 | 2093 | pte_unmap_unlock(start_pte, pte_lock); |
8cd3984d AR |
2094 | ret = err; |
2095 | remaining_pages_total -= pte_idx; | |
2096 | goto out; | |
2097 | } | |
2098 | addr += PAGE_SIZE; | |
2099 | ++curr_page_idx; | |
2100 | } | |
7f70c2a6 | 2101 | pte_unmap_unlock(start_pte, pte_lock); |
8cd3984d AR |
2102 | pages_to_write_in_pmd -= batch_size; |
2103 | remaining_pages_total -= batch_size; | |
2104 | } | |
2105 | if (remaining_pages_total) | |
2106 | goto more; | |
2107 | ret = 0; | |
2108 | out: | |
2109 | *num = remaining_pages_total; | |
2110 | return ret; | |
2111 | } | |
8cd3984d AR |
2112 | |
2113 | /** | |
2114 | * vm_insert_pages - insert multiple pages into user vma, batching the pmd lock. | |
2115 | * @vma: user vma to map to | |
2116 | * @addr: target start user address of these pages | |
2117 | * @pages: source kernel pages | |
2118 | * @num: in: number of pages to map. out: number of pages that were *not* | |
2119 | * mapped. (0 means all pages were successfully mapped). | |
2120 | * | |
2121 | * Preferred over vm_insert_page() when inserting multiple pages. | |
2122 | * | |
2123 | * In case of error, we may have mapped a subset of the provided | |
2124 | * pages. It is the caller's responsibility to account for this case. | |
2125 | * | |
2126 | * The same restrictions apply as in vm_insert_page(). | |
2127 | */ | |
2128 | int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr, | |
2129 | struct page **pages, unsigned long *num) | |
2130 | { | |
8cd3984d AR |
2131 | const unsigned long end_addr = addr + (*num * PAGE_SIZE) - 1; |
2132 | ||
2133 | if (addr < vma->vm_start || end_addr >= vma->vm_end) | |
2134 | return -EFAULT; | |
2135 | if (!(vma->vm_flags & VM_MIXEDMAP)) { | |
d8ed45c5 | 2136 | BUG_ON(mmap_read_trylock(vma->vm_mm)); |
8cd3984d | 2137 | BUG_ON(vma->vm_flags & VM_PFNMAP); |
1c71222e | 2138 | vm_flags_set(vma, VM_MIXEDMAP); |
8cd3984d AR |
2139 | } |
2140 | /* Defer page refcount checking till we're about to map that page. */ | |
2141 | return insert_pages(vma, addr, pages, num, vma->vm_page_prot); | |
8cd3984d AR |
2142 | } |
2143 | EXPORT_SYMBOL(vm_insert_pages); | |
2144 | ||
bfa5bf6d REB |
2145 | /** |
2146 | * vm_insert_page - insert single page into user vma | |
2147 | * @vma: user vma to map to | |
2148 | * @addr: target user address of this page | |
2149 | * @page: source kernel page | |
2150 | * | |
a145dd41 LT |
2151 | * This allows drivers to insert individual pages they've allocated |
2152 | * into a user vma. | |
2153 | * | |
2154 | * The page has to be a nice clean _individual_ kernel allocation. | |
2155 | * If you allocate a compound page, you need to have marked it as | |
2156 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 2157 | * (see split_page()). |
a145dd41 LT |
2158 | * |
2159 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
2160 | * took an arbitrary page protection parameter. This doesn't allow | |
2161 | * that. Your vma protection will have to be set up correctly, which | |
2162 | * means that if you want a shared writable mapping, you'd better | |
2163 | * ask for a shared writable mapping! | |
2164 | * | |
2165 | * The page does not need to be reserved. | |
4b6e1e37 KK |
2166 | * |
2167 | * Usually this function is called from f_op->mmap() handler | |
c1e8d7c6 | 2168 | * under mm->mmap_lock write-lock, so it can change vma->vm_flags. |
4b6e1e37 KK |
2169 | * Caller must set VM_MIXEDMAP on vma if it wants to call this |
2170 | * function from other places, for example from page-fault handler. | |
a862f68a MR |
2171 | * |
2172 | * Return: %0 on success, negative error code otherwise. | |
a145dd41 | 2173 | */ |
423bad60 NP |
2174 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
2175 | struct page *page) | |
a145dd41 LT |
2176 | { |
2177 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
2178 | return -EFAULT; | |
2179 | if (!page_count(page)) | |
2180 | return -EINVAL; | |
4b6e1e37 | 2181 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
d8ed45c5 | 2182 | BUG_ON(mmap_read_trylock(vma->vm_mm)); |
4b6e1e37 | 2183 | BUG_ON(vma->vm_flags & VM_PFNMAP); |
1c71222e | 2184 | vm_flags_set(vma, VM_MIXEDMAP); |
4b6e1e37 | 2185 | } |
423bad60 | 2186 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 2187 | } |
e3c3374f | 2188 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 2189 | |
a667d745 SJ |
2190 | /* |
2191 | * __vm_map_pages - maps range of kernel pages into user vma | |
2192 | * @vma: user vma to map to | |
2193 | * @pages: pointer to array of source kernel pages | |
2194 | * @num: number of pages in page array | |
2195 | * @offset: user's requested vm_pgoff | |
2196 | * | |
2197 | * This allows drivers to map range of kernel pages into a user vma. | |
2198 | * | |
2199 | * Return: 0 on success and error code otherwise. | |
2200 | */ | |
2201 | static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages, | |
2202 | unsigned long num, unsigned long offset) | |
2203 | { | |
2204 | unsigned long count = vma_pages(vma); | |
2205 | unsigned long uaddr = vma->vm_start; | |
2206 | int ret, i; | |
2207 | ||
2208 | /* Fail if the user requested offset is beyond the end of the object */ | |
96756fcb | 2209 | if (offset >= num) |
a667d745 SJ |
2210 | return -ENXIO; |
2211 | ||
2212 | /* Fail if the user requested size exceeds available object size */ | |
2213 | if (count > num - offset) | |
2214 | return -ENXIO; | |
2215 | ||
2216 | for (i = 0; i < count; i++) { | |
2217 | ret = vm_insert_page(vma, uaddr, pages[offset + i]); | |
2218 | if (ret < 0) | |
2219 | return ret; | |
2220 | uaddr += PAGE_SIZE; | |
2221 | } | |
2222 | ||
2223 | return 0; | |
2224 | } | |
2225 | ||
2226 | /** | |
2227 | * vm_map_pages - maps range of kernel pages starts with non zero offset | |
2228 | * @vma: user vma to map to | |
2229 | * @pages: pointer to array of source kernel pages | |
2230 | * @num: number of pages in page array | |
2231 | * | |
2232 | * Maps an object consisting of @num pages, catering for the user's | |
2233 | * requested vm_pgoff | |
2234 | * | |
2235 | * If we fail to insert any page into the vma, the function will return | |
2236 | * immediately leaving any previously inserted pages present. Callers | |
2237 | * from the mmap handler may immediately return the error as their caller | |
2238 | * will destroy the vma, removing any successfully inserted pages. Other | |
2239 | * callers should make their own arrangements for calling unmap_region(). | |
2240 | * | |
2241 | * Context: Process context. Called by mmap handlers. | |
2242 | * Return: 0 on success and error code otherwise. | |
2243 | */ | |
2244 | int vm_map_pages(struct vm_area_struct *vma, struct page **pages, | |
2245 | unsigned long num) | |
2246 | { | |
2247 | return __vm_map_pages(vma, pages, num, vma->vm_pgoff); | |
2248 | } | |
2249 | EXPORT_SYMBOL(vm_map_pages); | |
2250 | ||
2251 | /** | |
2252 | * vm_map_pages_zero - map range of kernel pages starts with zero offset | |
2253 | * @vma: user vma to map to | |
2254 | * @pages: pointer to array of source kernel pages | |
2255 | * @num: number of pages in page array | |
2256 | * | |
2257 | * Similar to vm_map_pages(), except that it explicitly sets the offset | |
2258 | * to 0. This function is intended for the drivers that did not consider | |
2259 | * vm_pgoff. | |
2260 | * | |
2261 | * Context: Process context. Called by mmap handlers. | |
2262 | * Return: 0 on success and error code otherwise. | |
2263 | */ | |
2264 | int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, | |
2265 | unsigned long num) | |
2266 | { | |
2267 | return __vm_map_pages(vma, pages, num, 0); | |
2268 | } | |
2269 | EXPORT_SYMBOL(vm_map_pages_zero); | |
2270 | ||
9b5a8e00 | 2271 | static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
b2770da6 | 2272 | pfn_t pfn, pgprot_t prot, bool mkwrite) |
423bad60 NP |
2273 | { |
2274 | struct mm_struct *mm = vma->vm_mm; | |
423bad60 NP |
2275 | pte_t *pte, entry; |
2276 | spinlock_t *ptl; | |
2277 | ||
423bad60 NP |
2278 | pte = get_locked_pte(mm, addr, &ptl); |
2279 | if (!pte) | |
9b5a8e00 | 2280 | return VM_FAULT_OOM; |
c33c7948 RR |
2281 | entry = ptep_get(pte); |
2282 | if (!pte_none(entry)) { | |
b2770da6 RZ |
2283 | if (mkwrite) { |
2284 | /* | |
2285 | * For read faults on private mappings the PFN passed | |
2286 | * in may not match the PFN we have mapped if the | |
2287 | * mapped PFN is a writeable COW page. In the mkwrite | |
2288 | * case we are creating a writable PTE for a shared | |
f2c57d91 JK |
2289 | * mapping and we expect the PFNs to match. If they |
2290 | * don't match, we are likely racing with block | |
2291 | * allocation and mapping invalidation so just skip the | |
2292 | * update. | |
b2770da6 | 2293 | */ |
c33c7948 RR |
2294 | if (pte_pfn(entry) != pfn_t_to_pfn(pfn)) { |
2295 | WARN_ON_ONCE(!is_zero_pfn(pte_pfn(entry))); | |
b2770da6 | 2296 | goto out_unlock; |
f2c57d91 | 2297 | } |
c33c7948 | 2298 | entry = pte_mkyoung(entry); |
cae85cb8 JK |
2299 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
2300 | if (ptep_set_access_flags(vma, addr, pte, entry, 1)) | |
2301 | update_mmu_cache(vma, addr, pte); | |
2302 | } | |
2303 | goto out_unlock; | |
b2770da6 | 2304 | } |
423bad60 NP |
2305 | |
2306 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
2307 | if (pfn_t_devmap(pfn)) |
2308 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
2309 | else | |
2310 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
b2770da6 | 2311 | |
b2770da6 RZ |
2312 | if (mkwrite) { |
2313 | entry = pte_mkyoung(entry); | |
2314 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2315 | } | |
2316 | ||
423bad60 | 2317 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 2318 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 | 2319 | |
423bad60 NP |
2320 | out_unlock: |
2321 | pte_unmap_unlock(pte, ptl); | |
9b5a8e00 | 2322 | return VM_FAULT_NOPAGE; |
423bad60 NP |
2323 | } |
2324 | ||
f5e6d1d5 MW |
2325 | /** |
2326 | * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
2327 | * @vma: user vma to map to | |
2328 | * @addr: target user address of this page | |
2329 | * @pfn: source kernel pfn | |
2330 | * @pgprot: pgprot flags for the inserted page | |
2331 | * | |
a1a0aea5 | 2332 | * This is exactly like vmf_insert_pfn(), except that it allows drivers |
f5e6d1d5 MW |
2333 | * to override pgprot on a per-page basis. |
2334 | * | |
2335 | * This only makes sense for IO mappings, and it makes no sense for | |
2336 | * COW mappings. In general, using multiple vmas is preferable; | |
ae2b01f3 | 2337 | * vmf_insert_pfn_prot should only be used if using multiple VMAs is |
f5e6d1d5 MW |
2338 | * impractical. |
2339 | * | |
28d8b812 LS |
2340 | * pgprot typically only differs from @vma->vm_page_prot when drivers set |
2341 | * caching- and encryption bits different than those of @vma->vm_page_prot, | |
2342 | * because the caching- or encryption mode may not be known at mmap() time. | |
2343 | * | |
2344 | * This is ok as long as @vma->vm_page_prot is not used by the core vm | |
2345 | * to set caching and encryption bits for those vmas (except for COW pages). | |
2346 | * This is ensured by core vm only modifying these page table entries using | |
2347 | * functions that don't touch caching- or encryption bits, using pte_modify() | |
2348 | * if needed. (See for example mprotect()). | |
2349 | * | |
2350 | * Also when new page-table entries are created, this is only done using the | |
2351 | * fault() callback, and never using the value of vma->vm_page_prot, | |
2352 | * except for page-table entries that point to anonymous pages as the result | |
2353 | * of COW. | |
574c5b3d | 2354 | * |
ae2b01f3 | 2355 | * Context: Process context. May allocate using %GFP_KERNEL. |
f5e6d1d5 MW |
2356 | * Return: vm_fault_t value. |
2357 | */ | |
2358 | vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
2359 | unsigned long pfn, pgprot_t pgprot) | |
2360 | { | |
6d958546 MW |
2361 | /* |
2362 | * Technically, architectures with pte_special can avoid all these | |
2363 | * restrictions (same for remap_pfn_range). However we would like | |
2364 | * consistency in testing and feature parity among all, so we should | |
2365 | * try to keep these invariants in place for everybody. | |
2366 | */ | |
2367 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); | |
2368 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
2369 | (VM_PFNMAP|VM_MIXEDMAP)); | |
2370 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
2371 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
2372 | ||
2373 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
2374 | return VM_FAULT_SIGBUS; | |
2375 | ||
2376 | if (!pfn_modify_allowed(pfn, pgprot)) | |
2377 | return VM_FAULT_SIGBUS; | |
2378 | ||
2379 | track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); | |
2380 | ||
9b5a8e00 | 2381 | return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, |
6d958546 | 2382 | false); |
f5e6d1d5 MW |
2383 | } |
2384 | EXPORT_SYMBOL(vmf_insert_pfn_prot); | |
e0dc0d8f | 2385 | |
ae2b01f3 MW |
2386 | /** |
2387 | * vmf_insert_pfn - insert single pfn into user vma | |
2388 | * @vma: user vma to map to | |
2389 | * @addr: target user address of this page | |
2390 | * @pfn: source kernel pfn | |
2391 | * | |
2392 | * Similar to vm_insert_page, this allows drivers to insert individual pages | |
2393 | * they've allocated into a user vma. Same comments apply. | |
2394 | * | |
2395 | * This function should only be called from a vm_ops->fault handler, and | |
2396 | * in that case the handler should return the result of this function. | |
2397 | * | |
2398 | * vma cannot be a COW mapping. | |
2399 | * | |
2400 | * As this is called only for pages that do not currently exist, we | |
2401 | * do not need to flush old virtual caches or the TLB. | |
2402 | * | |
2403 | * Context: Process context. May allocate using %GFP_KERNEL. | |
2404 | * Return: vm_fault_t value. | |
2405 | */ | |
2406 | vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
2407 | unsigned long pfn) | |
2408 | { | |
2409 | return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
2410 | } | |
2411 | EXPORT_SYMBOL(vmf_insert_pfn); | |
2412 | ||
785a3fab DW |
2413 | static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) |
2414 | { | |
2415 | /* these checks mirror the abort conditions in vm_normal_page */ | |
2416 | if (vma->vm_flags & VM_MIXEDMAP) | |
2417 | return true; | |
2418 | if (pfn_t_devmap(pfn)) | |
2419 | return true; | |
2420 | if (pfn_t_special(pfn)) | |
2421 | return true; | |
2422 | if (is_zero_pfn(pfn_t_to_pfn(pfn))) | |
2423 | return true; | |
2424 | return false; | |
2425 | } | |
2426 | ||
79f3aa5b | 2427 | static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma, |
28d8b812 | 2428 | unsigned long addr, pfn_t pfn, bool mkwrite) |
423bad60 | 2429 | { |
28d8b812 | 2430 | pgprot_t pgprot = vma->vm_page_prot; |
79f3aa5b | 2431 | int err; |
87744ab3 | 2432 | |
785a3fab | 2433 | BUG_ON(!vm_mixed_ok(vma, pfn)); |
e0dc0d8f | 2434 | |
423bad60 | 2435 | if (addr < vma->vm_start || addr >= vma->vm_end) |
79f3aa5b | 2436 | return VM_FAULT_SIGBUS; |
308a047c BP |
2437 | |
2438 | track_pfn_insert(vma, &pgprot, pfn); | |
e0dc0d8f | 2439 | |
42e4089c | 2440 | if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) |
79f3aa5b | 2441 | return VM_FAULT_SIGBUS; |
42e4089c | 2442 | |
423bad60 NP |
2443 | /* |
2444 | * If we don't have pte special, then we have to use the pfn_valid() | |
2445 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
2446 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
2447 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
2448 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 2449 | */ |
00b3a331 LD |
2450 | if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && |
2451 | !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { | |
423bad60 NP |
2452 | struct page *page; |
2453 | ||
03fc2da6 DW |
2454 | /* |
2455 | * At this point we are committed to insert_page() | |
2456 | * regardless of whether the caller specified flags that | |
2457 | * result in pfn_t_has_page() == false. | |
2458 | */ | |
2459 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
79f3aa5b MW |
2460 | err = insert_page(vma, addr, page, pgprot); |
2461 | } else { | |
9b5a8e00 | 2462 | return insert_pfn(vma, addr, pfn, pgprot, mkwrite); |
423bad60 | 2463 | } |
b2770da6 | 2464 | |
5d747637 MW |
2465 | if (err == -ENOMEM) |
2466 | return VM_FAULT_OOM; | |
2467 | if (err < 0 && err != -EBUSY) | |
2468 | return VM_FAULT_SIGBUS; | |
2469 | ||
2470 | return VM_FAULT_NOPAGE; | |
e0dc0d8f | 2471 | } |
79f3aa5b MW |
2472 | |
2473 | vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, | |
2474 | pfn_t pfn) | |
2475 | { | |
28d8b812 | 2476 | return __vm_insert_mixed(vma, addr, pfn, false); |
79f3aa5b | 2477 | } |
5d747637 | 2478 | EXPORT_SYMBOL(vmf_insert_mixed); |
e0dc0d8f | 2479 | |
ab77dab4 SJ |
2480 | /* |
2481 | * If the insertion of PTE failed because someone else already added a | |
2482 | * different entry in the mean time, we treat that as success as we assume | |
2483 | * the same entry was actually inserted. | |
2484 | */ | |
ab77dab4 SJ |
2485 | vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, |
2486 | unsigned long addr, pfn_t pfn) | |
b2770da6 | 2487 | { |
28d8b812 | 2488 | return __vm_insert_mixed(vma, addr, pfn, true); |
b2770da6 | 2489 | } |
ab77dab4 | 2490 | EXPORT_SYMBOL(vmf_insert_mixed_mkwrite); |
b2770da6 | 2491 | |
1da177e4 LT |
2492 | /* |
2493 | * maps a range of physical memory into the requested pages. the old | |
2494 | * mappings are removed. any references to nonexistent pages results | |
2495 | * in null mappings (currently treated as "copy-on-access") | |
2496 | */ | |
2497 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
2498 | unsigned long addr, unsigned long end, | |
2499 | unsigned long pfn, pgprot_t prot) | |
2500 | { | |
90a3e375 | 2501 | pte_t *pte, *mapped_pte; |
c74df32c | 2502 | spinlock_t *ptl; |
42e4089c | 2503 | int err = 0; |
1da177e4 | 2504 | |
90a3e375 | 2505 | mapped_pte = pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
2506 | if (!pte) |
2507 | return -ENOMEM; | |
6606c3e0 | 2508 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 2509 | do { |
c33c7948 | 2510 | BUG_ON(!pte_none(ptep_get(pte))); |
42e4089c AK |
2511 | if (!pfn_modify_allowed(pfn, prot)) { |
2512 | err = -EACCES; | |
2513 | break; | |
2514 | } | |
7e675137 | 2515 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
2516 | pfn++; |
2517 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 2518 | arch_leave_lazy_mmu_mode(); |
90a3e375 | 2519 | pte_unmap_unlock(mapped_pte, ptl); |
42e4089c | 2520 | return err; |
1da177e4 LT |
2521 | } |
2522 | ||
2523 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2524 | unsigned long addr, unsigned long end, | |
2525 | unsigned long pfn, pgprot_t prot) | |
2526 | { | |
2527 | pmd_t *pmd; | |
2528 | unsigned long next; | |
42e4089c | 2529 | int err; |
1da177e4 LT |
2530 | |
2531 | pfn -= addr >> PAGE_SHIFT; | |
2532 | pmd = pmd_alloc(mm, pud, addr); | |
2533 | if (!pmd) | |
2534 | return -ENOMEM; | |
f66055ab | 2535 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
2536 | do { |
2537 | next = pmd_addr_end(addr, end); | |
42e4089c AK |
2538 | err = remap_pte_range(mm, pmd, addr, next, |
2539 | pfn + (addr >> PAGE_SHIFT), prot); | |
2540 | if (err) | |
2541 | return err; | |
1da177e4 LT |
2542 | } while (pmd++, addr = next, addr != end); |
2543 | return 0; | |
2544 | } | |
2545 | ||
c2febafc | 2546 | static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, |
1da177e4 LT |
2547 | unsigned long addr, unsigned long end, |
2548 | unsigned long pfn, pgprot_t prot) | |
2549 | { | |
2550 | pud_t *pud; | |
2551 | unsigned long next; | |
42e4089c | 2552 | int err; |
1da177e4 LT |
2553 | |
2554 | pfn -= addr >> PAGE_SHIFT; | |
c2febafc | 2555 | pud = pud_alloc(mm, p4d, addr); |
1da177e4 LT |
2556 | if (!pud) |
2557 | return -ENOMEM; | |
2558 | do { | |
2559 | next = pud_addr_end(addr, end); | |
42e4089c AK |
2560 | err = remap_pmd_range(mm, pud, addr, next, |
2561 | pfn + (addr >> PAGE_SHIFT), prot); | |
2562 | if (err) | |
2563 | return err; | |
1da177e4 LT |
2564 | } while (pud++, addr = next, addr != end); |
2565 | return 0; | |
2566 | } | |
2567 | ||
c2febafc KS |
2568 | static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2569 | unsigned long addr, unsigned long end, | |
2570 | unsigned long pfn, pgprot_t prot) | |
2571 | { | |
2572 | p4d_t *p4d; | |
2573 | unsigned long next; | |
42e4089c | 2574 | int err; |
c2febafc KS |
2575 | |
2576 | pfn -= addr >> PAGE_SHIFT; | |
2577 | p4d = p4d_alloc(mm, pgd, addr); | |
2578 | if (!p4d) | |
2579 | return -ENOMEM; | |
2580 | do { | |
2581 | next = p4d_addr_end(addr, end); | |
42e4089c AK |
2582 | err = remap_pud_range(mm, p4d, addr, next, |
2583 | pfn + (addr >> PAGE_SHIFT), prot); | |
2584 | if (err) | |
2585 | return err; | |
c2febafc KS |
2586 | } while (p4d++, addr = next, addr != end); |
2587 | return 0; | |
2588 | } | |
2589 | ||
74ffa5a3 CH |
2590 | /* |
2591 | * Variant of remap_pfn_range that does not call track_pfn_remap. The caller | |
2592 | * must have pre-validated the caching bits of the pgprot_t. | |
bfa5bf6d | 2593 | */ |
74ffa5a3 CH |
2594 | int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr, |
2595 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1da177e4 LT |
2596 | { |
2597 | pgd_t *pgd; | |
2598 | unsigned long next; | |
2d15cab8 | 2599 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 LT |
2600 | struct mm_struct *mm = vma->vm_mm; |
2601 | int err; | |
2602 | ||
0c4123e3 AZ |
2603 | if (WARN_ON_ONCE(!PAGE_ALIGNED(addr))) |
2604 | return -EINVAL; | |
2605 | ||
1da177e4 LT |
2606 | /* |
2607 | * Physically remapped pages are special. Tell the | |
2608 | * rest of the world about it: | |
2609 | * VM_IO tells people not to look at these pages | |
2610 | * (accesses can have side effects). | |
6aab341e LT |
2611 | * VM_PFNMAP tells the core MM that the base pages are just |
2612 | * raw PFN mappings, and do not have a "struct page" associated | |
2613 | * with them. | |
314e51b9 KK |
2614 | * VM_DONTEXPAND |
2615 | * Disable vma merging and expanding with mremap(). | |
2616 | * VM_DONTDUMP | |
2617 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
2618 | * |
2619 | * There's a horrible special case to handle copy-on-write | |
2620 | * behaviour that some programs depend on. We mark the "original" | |
2621 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 2622 | * See vm_normal_page() for details. |
1da177e4 | 2623 | */ |
b3b9c293 KK |
2624 | if (is_cow_mapping(vma->vm_flags)) { |
2625 | if (addr != vma->vm_start || end != vma->vm_end) | |
2626 | return -EINVAL; | |
fb155c16 | 2627 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
2628 | } |
2629 | ||
1c71222e | 2630 | vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP); |
1da177e4 LT |
2631 | |
2632 | BUG_ON(addr >= end); | |
2633 | pfn -= addr >> PAGE_SHIFT; | |
2634 | pgd = pgd_offset(mm, addr); | |
2635 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
2636 | do { |
2637 | next = pgd_addr_end(addr, end); | |
c2febafc | 2638 | err = remap_p4d_range(mm, pgd, addr, next, |
1da177e4 LT |
2639 | pfn + (addr >> PAGE_SHIFT), prot); |
2640 | if (err) | |
74ffa5a3 | 2641 | return err; |
1da177e4 | 2642 | } while (pgd++, addr = next, addr != end); |
2ab64037 | 2643 | |
74ffa5a3 CH |
2644 | return 0; |
2645 | } | |
2646 | ||
2647 | /** | |
2648 | * remap_pfn_range - remap kernel memory to userspace | |
2649 | * @vma: user vma to map to | |
2650 | * @addr: target page aligned user address to start at | |
2651 | * @pfn: page frame number of kernel physical memory address | |
2652 | * @size: size of mapping area | |
2653 | * @prot: page protection flags for this mapping | |
2654 | * | |
2655 | * Note: this is only safe if the mm semaphore is held when called. | |
2656 | * | |
2657 | * Return: %0 on success, negative error code otherwise. | |
2658 | */ | |
2659 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, | |
2660 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
2661 | { | |
2662 | int err; | |
2663 | ||
2664 | err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); | |
2ab64037 | 2665 | if (err) |
74ffa5a3 | 2666 | return -EINVAL; |
2ab64037 | 2667 | |
74ffa5a3 CH |
2668 | err = remap_pfn_range_notrack(vma, addr, pfn, size, prot); |
2669 | if (err) | |
68f48381 | 2670 | untrack_pfn(vma, pfn, PAGE_ALIGN(size), true); |
1da177e4 LT |
2671 | return err; |
2672 | } | |
2673 | EXPORT_SYMBOL(remap_pfn_range); | |
2674 | ||
b4cbb197 LT |
2675 | /** |
2676 | * vm_iomap_memory - remap memory to userspace | |
2677 | * @vma: user vma to map to | |
abd69b9e | 2678 | * @start: start of the physical memory to be mapped |
b4cbb197 LT |
2679 | * @len: size of area |
2680 | * | |
2681 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
2682 | * driver just needs to give us the physical memory range to be mapped, | |
2683 | * we'll figure out the rest from the vma information. | |
2684 | * | |
2685 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
2686 | * whatever write-combining details or similar. | |
a862f68a MR |
2687 | * |
2688 | * Return: %0 on success, negative error code otherwise. | |
b4cbb197 LT |
2689 | */ |
2690 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
2691 | { | |
2692 | unsigned long vm_len, pfn, pages; | |
2693 | ||
2694 | /* Check that the physical memory area passed in looks valid */ | |
2695 | if (start + len < start) | |
2696 | return -EINVAL; | |
2697 | /* | |
2698 | * You *really* shouldn't map things that aren't page-aligned, | |
2699 | * but we've historically allowed it because IO memory might | |
2700 | * just have smaller alignment. | |
2701 | */ | |
2702 | len += start & ~PAGE_MASK; | |
2703 | pfn = start >> PAGE_SHIFT; | |
2704 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
2705 | if (pfn + pages < pfn) | |
2706 | return -EINVAL; | |
2707 | ||
2708 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
2709 | if (vma->vm_pgoff > pages) | |
2710 | return -EINVAL; | |
2711 | pfn += vma->vm_pgoff; | |
2712 | pages -= vma->vm_pgoff; | |
2713 | ||
2714 | /* Can we fit all of the mapping? */ | |
2715 | vm_len = vma->vm_end - vma->vm_start; | |
2716 | if (vm_len >> PAGE_SHIFT > pages) | |
2717 | return -EINVAL; | |
2718 | ||
2719 | /* Ok, let it rip */ | |
2720 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
2721 | } | |
2722 | EXPORT_SYMBOL(vm_iomap_memory); | |
2723 | ||
aee16b3c JF |
2724 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
2725 | unsigned long addr, unsigned long end, | |
e80d3909 JR |
2726 | pte_fn_t fn, void *data, bool create, |
2727 | pgtbl_mod_mask *mask) | |
aee16b3c | 2728 | { |
8abb50c7 | 2729 | pte_t *pte, *mapped_pte; |
be1db475 | 2730 | int err = 0; |
3f649ab7 | 2731 | spinlock_t *ptl; |
aee16b3c | 2732 | |
be1db475 | 2733 | if (create) { |
8abb50c7 | 2734 | mapped_pte = pte = (mm == &init_mm) ? |
e80d3909 | 2735 | pte_alloc_kernel_track(pmd, addr, mask) : |
be1db475 DA |
2736 | pte_alloc_map_lock(mm, pmd, addr, &ptl); |
2737 | if (!pte) | |
2738 | return -ENOMEM; | |
2739 | } else { | |
8abb50c7 | 2740 | mapped_pte = pte = (mm == &init_mm) ? |
be1db475 DA |
2741 | pte_offset_kernel(pmd, addr) : |
2742 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
3db82b93 HD |
2743 | if (!pte) |
2744 | return -EINVAL; | |
be1db475 | 2745 | } |
aee16b3c | 2746 | |
38e0edb1 JF |
2747 | arch_enter_lazy_mmu_mode(); |
2748 | ||
eeb4a05f CH |
2749 | if (fn) { |
2750 | do { | |
c33c7948 | 2751 | if (create || !pte_none(ptep_get(pte))) { |
eeb4a05f CH |
2752 | err = fn(pte++, addr, data); |
2753 | if (err) | |
2754 | break; | |
2755 | } | |
2756 | } while (addr += PAGE_SIZE, addr != end); | |
2757 | } | |
e80d3909 | 2758 | *mask |= PGTBL_PTE_MODIFIED; |
aee16b3c | 2759 | |
38e0edb1 JF |
2760 | arch_leave_lazy_mmu_mode(); |
2761 | ||
aee16b3c | 2762 | if (mm != &init_mm) |
8abb50c7 | 2763 | pte_unmap_unlock(mapped_pte, ptl); |
aee16b3c JF |
2764 | return err; |
2765 | } | |
2766 | ||
2767 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2768 | unsigned long addr, unsigned long end, | |
e80d3909 JR |
2769 | pte_fn_t fn, void *data, bool create, |
2770 | pgtbl_mod_mask *mask) | |
aee16b3c JF |
2771 | { |
2772 | pmd_t *pmd; | |
2773 | unsigned long next; | |
be1db475 | 2774 | int err = 0; |
aee16b3c | 2775 | |
1965e933 | 2776 | BUG_ON(pud_leaf(*pud)); |
ceb86879 | 2777 | |
be1db475 | 2778 | if (create) { |
e80d3909 | 2779 | pmd = pmd_alloc_track(mm, pud, addr, mask); |
be1db475 DA |
2780 | if (!pmd) |
2781 | return -ENOMEM; | |
2782 | } else { | |
2783 | pmd = pmd_offset(pud, addr); | |
2784 | } | |
aee16b3c JF |
2785 | do { |
2786 | next = pmd_addr_end(addr, end); | |
0c95cba4 NP |
2787 | if (pmd_none(*pmd) && !create) |
2788 | continue; | |
2789 | if (WARN_ON_ONCE(pmd_leaf(*pmd))) | |
2790 | return -EINVAL; | |
2791 | if (!pmd_none(*pmd) && WARN_ON_ONCE(pmd_bad(*pmd))) { | |
2792 | if (!create) | |
2793 | continue; | |
2794 | pmd_clear_bad(pmd); | |
be1db475 | 2795 | } |
0c95cba4 NP |
2796 | err = apply_to_pte_range(mm, pmd, addr, next, |
2797 | fn, data, create, mask); | |
2798 | if (err) | |
2799 | break; | |
aee16b3c | 2800 | } while (pmd++, addr = next, addr != end); |
0c95cba4 | 2801 | |
aee16b3c JF |
2802 | return err; |
2803 | } | |
2804 | ||
c2febafc | 2805 | static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, |
aee16b3c | 2806 | unsigned long addr, unsigned long end, |
e80d3909 JR |
2807 | pte_fn_t fn, void *data, bool create, |
2808 | pgtbl_mod_mask *mask) | |
aee16b3c JF |
2809 | { |
2810 | pud_t *pud; | |
2811 | unsigned long next; | |
be1db475 | 2812 | int err = 0; |
aee16b3c | 2813 | |
be1db475 | 2814 | if (create) { |
e80d3909 | 2815 | pud = pud_alloc_track(mm, p4d, addr, mask); |
be1db475 DA |
2816 | if (!pud) |
2817 | return -ENOMEM; | |
2818 | } else { | |
2819 | pud = pud_offset(p4d, addr); | |
2820 | } | |
aee16b3c JF |
2821 | do { |
2822 | next = pud_addr_end(addr, end); | |
0c95cba4 NP |
2823 | if (pud_none(*pud) && !create) |
2824 | continue; | |
2825 | if (WARN_ON_ONCE(pud_leaf(*pud))) | |
2826 | return -EINVAL; | |
2827 | if (!pud_none(*pud) && WARN_ON_ONCE(pud_bad(*pud))) { | |
2828 | if (!create) | |
2829 | continue; | |
2830 | pud_clear_bad(pud); | |
be1db475 | 2831 | } |
0c95cba4 NP |
2832 | err = apply_to_pmd_range(mm, pud, addr, next, |
2833 | fn, data, create, mask); | |
2834 | if (err) | |
2835 | break; | |
aee16b3c | 2836 | } while (pud++, addr = next, addr != end); |
0c95cba4 | 2837 | |
aee16b3c JF |
2838 | return err; |
2839 | } | |
2840 | ||
c2febafc KS |
2841 | static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2842 | unsigned long addr, unsigned long end, | |
e80d3909 JR |
2843 | pte_fn_t fn, void *data, bool create, |
2844 | pgtbl_mod_mask *mask) | |
c2febafc KS |
2845 | { |
2846 | p4d_t *p4d; | |
2847 | unsigned long next; | |
be1db475 | 2848 | int err = 0; |
c2febafc | 2849 | |
be1db475 | 2850 | if (create) { |
e80d3909 | 2851 | p4d = p4d_alloc_track(mm, pgd, addr, mask); |
be1db475 DA |
2852 | if (!p4d) |
2853 | return -ENOMEM; | |
2854 | } else { | |
2855 | p4d = p4d_offset(pgd, addr); | |
2856 | } | |
c2febafc KS |
2857 | do { |
2858 | next = p4d_addr_end(addr, end); | |
0c95cba4 NP |
2859 | if (p4d_none(*p4d) && !create) |
2860 | continue; | |
2861 | if (WARN_ON_ONCE(p4d_leaf(*p4d))) | |
2862 | return -EINVAL; | |
2863 | if (!p4d_none(*p4d) && WARN_ON_ONCE(p4d_bad(*p4d))) { | |
2864 | if (!create) | |
2865 | continue; | |
2866 | p4d_clear_bad(p4d); | |
be1db475 | 2867 | } |
0c95cba4 NP |
2868 | err = apply_to_pud_range(mm, p4d, addr, next, |
2869 | fn, data, create, mask); | |
2870 | if (err) | |
2871 | break; | |
c2febafc | 2872 | } while (p4d++, addr = next, addr != end); |
0c95cba4 | 2873 | |
c2febafc KS |
2874 | return err; |
2875 | } | |
2876 | ||
be1db475 DA |
2877 | static int __apply_to_page_range(struct mm_struct *mm, unsigned long addr, |
2878 | unsigned long size, pte_fn_t fn, | |
2879 | void *data, bool create) | |
aee16b3c JF |
2880 | { |
2881 | pgd_t *pgd; | |
e80d3909 | 2882 | unsigned long start = addr, next; |
57250a5b | 2883 | unsigned long end = addr + size; |
e80d3909 | 2884 | pgtbl_mod_mask mask = 0; |
be1db475 | 2885 | int err = 0; |
aee16b3c | 2886 | |
9cb65bc3 MP |
2887 | if (WARN_ON(addr >= end)) |
2888 | return -EINVAL; | |
2889 | ||
aee16b3c JF |
2890 | pgd = pgd_offset(mm, addr); |
2891 | do { | |
2892 | next = pgd_addr_end(addr, end); | |
0c95cba4 | 2893 | if (pgd_none(*pgd) && !create) |
be1db475 | 2894 | continue; |
0c95cba4 NP |
2895 | if (WARN_ON_ONCE(pgd_leaf(*pgd))) |
2896 | return -EINVAL; | |
2897 | if (!pgd_none(*pgd) && WARN_ON_ONCE(pgd_bad(*pgd))) { | |
2898 | if (!create) | |
2899 | continue; | |
2900 | pgd_clear_bad(pgd); | |
2901 | } | |
2902 | err = apply_to_p4d_range(mm, pgd, addr, next, | |
2903 | fn, data, create, &mask); | |
aee16b3c JF |
2904 | if (err) |
2905 | break; | |
2906 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2907 | |
e80d3909 JR |
2908 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
2909 | arch_sync_kernel_mappings(start, start + size); | |
2910 | ||
aee16b3c JF |
2911 | return err; |
2912 | } | |
be1db475 DA |
2913 | |
2914 | /* | |
2915 | * Scan a region of virtual memory, filling in page tables as necessary | |
2916 | * and calling a provided function on each leaf page table. | |
2917 | */ | |
2918 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2919 | unsigned long size, pte_fn_t fn, void *data) | |
2920 | { | |
2921 | return __apply_to_page_range(mm, addr, size, fn, data, true); | |
2922 | } | |
aee16b3c JF |
2923 | EXPORT_SYMBOL_GPL(apply_to_page_range); |
2924 | ||
be1db475 DA |
2925 | /* |
2926 | * Scan a region of virtual memory, calling a provided function on | |
2927 | * each leaf page table where it exists. | |
2928 | * | |
2929 | * Unlike apply_to_page_range, this does _not_ fill in page tables | |
2930 | * where they are absent. | |
2931 | */ | |
2932 | int apply_to_existing_page_range(struct mm_struct *mm, unsigned long addr, | |
2933 | unsigned long size, pte_fn_t fn, void *data) | |
2934 | { | |
2935 | return __apply_to_page_range(mm, addr, size, fn, data, false); | |
2936 | } | |
2937 | EXPORT_SYMBOL_GPL(apply_to_existing_page_range); | |
2938 | ||
8f4e2101 | 2939 | /* |
9b4bdd2f KS |
2940 | * handle_pte_fault chooses page fault handler according to an entry which was |
2941 | * read non-atomically. Before making any commitment, on those architectures | |
2942 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
2943 | * parts, do_swap_page must check under lock before unmapping the pte and | |
2944 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2945 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2946 | */ |
2ca99358 | 2947 | static inline int pte_unmap_same(struct vm_fault *vmf) |
8f4e2101 HD |
2948 | { |
2949 | int same = 1; | |
923717cb | 2950 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPTION) |
8f4e2101 | 2951 | if (sizeof(pte_t) > sizeof(unsigned long)) { |
c7ad0880 | 2952 | spin_lock(vmf->ptl); |
c33c7948 | 2953 | same = pte_same(ptep_get(vmf->pte), vmf->orig_pte); |
c7ad0880 | 2954 | spin_unlock(vmf->ptl); |
8f4e2101 HD |
2955 | } |
2956 | #endif | |
2ca99358 PX |
2957 | pte_unmap(vmf->pte); |
2958 | vmf->pte = NULL; | |
8f4e2101 HD |
2959 | return same; |
2960 | } | |
2961 | ||
a873dfe1 TL |
2962 | /* |
2963 | * Return: | |
2964 | * 0: copied succeeded | |
2965 | * -EHWPOISON: copy failed due to hwpoison in source page | |
2966 | * -EAGAIN: copied failed (some other reason) | |
2967 | */ | |
2968 | static inline int __wp_page_copy_user(struct page *dst, struct page *src, | |
2969 | struct vm_fault *vmf) | |
6aab341e | 2970 | { |
a873dfe1 | 2971 | int ret; |
83d116c5 JH |
2972 | void *kaddr; |
2973 | void __user *uaddr; | |
83d116c5 JH |
2974 | struct vm_area_struct *vma = vmf->vma; |
2975 | struct mm_struct *mm = vma->vm_mm; | |
2976 | unsigned long addr = vmf->address; | |
2977 | ||
83d116c5 | 2978 | if (likely(src)) { |
d302c239 TL |
2979 | if (copy_mc_user_highpage(dst, src, addr, vma)) { |
2980 | memory_failure_queue(page_to_pfn(src), 0); | |
a873dfe1 | 2981 | return -EHWPOISON; |
d302c239 | 2982 | } |
a873dfe1 | 2983 | return 0; |
83d116c5 JH |
2984 | } |
2985 | ||
6aab341e LT |
2986 | /* |
2987 | * If the source page was a PFN mapping, we don't have | |
2988 | * a "struct page" for it. We do a best-effort copy by | |
2989 | * just copying from the original user address. If that | |
2990 | * fails, we just zero-fill it. Live with it. | |
2991 | */ | |
24d2613a FDF |
2992 | kaddr = kmap_local_page(dst); |
2993 | pagefault_disable(); | |
83d116c5 JH |
2994 | uaddr = (void __user *)(addr & PAGE_MASK); |
2995 | ||
2996 | /* | |
2997 | * On architectures with software "accessed" bits, we would | |
2998 | * take a double page fault, so mark it accessed here. | |
2999 | */ | |
3db82b93 | 3000 | vmf->pte = NULL; |
e1fd09e3 | 3001 | if (!arch_has_hw_pte_young() && !pte_young(vmf->orig_pte)) { |
83d116c5 | 3002 | pte_t entry; |
5d2a2dbb | 3003 | |
83d116c5 | 3004 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); |
c33c7948 | 3005 | if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { |
83d116c5 JH |
3006 | /* |
3007 | * Other thread has already handled the fault | |
7df67697 | 3008 | * and update local tlb only |
83d116c5 | 3009 | */ |
a92cbb82 HD |
3010 | if (vmf->pte) |
3011 | update_mmu_tlb(vma, addr, vmf->pte); | |
a873dfe1 | 3012 | ret = -EAGAIN; |
83d116c5 JH |
3013 | goto pte_unlock; |
3014 | } | |
3015 | ||
3016 | entry = pte_mkyoung(vmf->orig_pte); | |
3017 | if (ptep_set_access_flags(vma, addr, vmf->pte, entry, 0)) | |
5003a2bd | 3018 | update_mmu_cache_range(vmf, vma, addr, vmf->pte, 1); |
83d116c5 JH |
3019 | } |
3020 | ||
3021 | /* | |
3022 | * This really shouldn't fail, because the page is there | |
3023 | * in the page tables. But it might just be unreadable, | |
3024 | * in which case we just give up and fill the result with | |
3025 | * zeroes. | |
3026 | */ | |
3027 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { | |
3db82b93 | 3028 | if (vmf->pte) |
c3e5ea6e KS |
3029 | goto warn; |
3030 | ||
3031 | /* Re-validate under PTL if the page is still mapped */ | |
3032 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); | |
c33c7948 | 3033 | if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { |
7df67697 | 3034 | /* The PTE changed under us, update local tlb */ |
a92cbb82 HD |
3035 | if (vmf->pte) |
3036 | update_mmu_tlb(vma, addr, vmf->pte); | |
a873dfe1 | 3037 | ret = -EAGAIN; |
c3e5ea6e KS |
3038 | goto pte_unlock; |
3039 | } | |
3040 | ||
5d2a2dbb | 3041 | /* |
985ba004 | 3042 | * The same page can be mapped back since last copy attempt. |
c3e5ea6e | 3043 | * Try to copy again under PTL. |
5d2a2dbb | 3044 | */ |
c3e5ea6e KS |
3045 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { |
3046 | /* | |
3047 | * Give a warn in case there can be some obscure | |
3048 | * use-case | |
3049 | */ | |
3050 | warn: | |
3051 | WARN_ON_ONCE(1); | |
3052 | clear_page(kaddr); | |
3053 | } | |
83d116c5 JH |
3054 | } |
3055 | ||
a873dfe1 | 3056 | ret = 0; |
83d116c5 JH |
3057 | |
3058 | pte_unlock: | |
3db82b93 | 3059 | if (vmf->pte) |
83d116c5 | 3060 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
24d2613a FDF |
3061 | pagefault_enable(); |
3062 | kunmap_local(kaddr); | |
83d116c5 JH |
3063 | flush_dcache_page(dst); |
3064 | ||
3065 | return ret; | |
6aab341e LT |
3066 | } |
3067 | ||
c20cd45e MH |
3068 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
3069 | { | |
3070 | struct file *vm_file = vma->vm_file; | |
3071 | ||
3072 | if (vm_file) | |
3073 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
3074 | ||
3075 | /* | |
3076 | * Special mappings (e.g. VDSO) do not have any file so fake | |
3077 | * a default GFP_KERNEL for them. | |
3078 | */ | |
3079 | return GFP_KERNEL; | |
3080 | } | |
3081 | ||
fb09a464 KS |
3082 | /* |
3083 | * Notify the address space that the page is about to become writable so that | |
3084 | * it can prohibit this or wait for the page to get into an appropriate state. | |
3085 | * | |
3086 | * We do this without the lock held, so that it can sleep if it needs to. | |
3087 | */ | |
86aa6998 | 3088 | static vm_fault_t do_page_mkwrite(struct vm_fault *vmf, struct folio *folio) |
fb09a464 | 3089 | { |
2b740303 | 3090 | vm_fault_t ret; |
38b8cb7f | 3091 | unsigned int old_flags = vmf->flags; |
fb09a464 | 3092 | |
38b8cb7f | 3093 | vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
fb09a464 | 3094 | |
dc617f29 DW |
3095 | if (vmf->vma->vm_file && |
3096 | IS_SWAPFILE(vmf->vma->vm_file->f_mapping->host)) | |
3097 | return VM_FAULT_SIGBUS; | |
3098 | ||
11bac800 | 3099 | ret = vmf->vma->vm_ops->page_mkwrite(vmf); |
38b8cb7f JK |
3100 | /* Restore original flags so that caller is not surprised */ |
3101 | vmf->flags = old_flags; | |
fb09a464 KS |
3102 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
3103 | return ret; | |
3104 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
3d243659 SK |
3105 | folio_lock(folio); |
3106 | if (!folio->mapping) { | |
3107 | folio_unlock(folio); | |
fb09a464 KS |
3108 | return 0; /* retry */ |
3109 | } | |
3110 | ret |= VM_FAULT_LOCKED; | |
3111 | } else | |
3d243659 | 3112 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
fb09a464 KS |
3113 | return ret; |
3114 | } | |
3115 | ||
97ba0c2b JK |
3116 | /* |
3117 | * Handle dirtying of a page in shared file mapping on a write fault. | |
3118 | * | |
3119 | * The function expects the page to be locked and unlocks it. | |
3120 | */ | |
89b15332 | 3121 | static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf) |
97ba0c2b | 3122 | { |
89b15332 | 3123 | struct vm_area_struct *vma = vmf->vma; |
97ba0c2b | 3124 | struct address_space *mapping; |
15b4919a | 3125 | struct folio *folio = page_folio(vmf->page); |
97ba0c2b JK |
3126 | bool dirtied; |
3127 | bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; | |
3128 | ||
15b4919a Z |
3129 | dirtied = folio_mark_dirty(folio); |
3130 | VM_BUG_ON_FOLIO(folio_test_anon(folio), folio); | |
97ba0c2b | 3131 | /* |
15b4919a Z |
3132 | * Take a local copy of the address_space - folio.mapping may be zeroed |
3133 | * by truncate after folio_unlock(). The address_space itself remains | |
3134 | * pinned by vma->vm_file's reference. We rely on folio_unlock()'s | |
97ba0c2b JK |
3135 | * release semantics to prevent the compiler from undoing this copying. |
3136 | */ | |
15b4919a Z |
3137 | mapping = folio_raw_mapping(folio); |
3138 | folio_unlock(folio); | |
97ba0c2b | 3139 | |
89b15332 JW |
3140 | if (!page_mkwrite) |
3141 | file_update_time(vma->vm_file); | |
3142 | ||
3143 | /* | |
3144 | * Throttle page dirtying rate down to writeback speed. | |
3145 | * | |
3146 | * mapping may be NULL here because some device drivers do not | |
3147 | * set page.mapping but still dirty their pages | |
3148 | * | |
c1e8d7c6 | 3149 | * Drop the mmap_lock before waiting on IO, if we can. The file |
89b15332 JW |
3150 | * is pinning the mapping, as per above. |
3151 | */ | |
97ba0c2b | 3152 | if ((dirtied || page_mkwrite) && mapping) { |
89b15332 JW |
3153 | struct file *fpin; |
3154 | ||
3155 | fpin = maybe_unlock_mmap_for_io(vmf, NULL); | |
97ba0c2b | 3156 | balance_dirty_pages_ratelimited(mapping); |
89b15332 JW |
3157 | if (fpin) { |
3158 | fput(fpin); | |
d9272525 | 3159 | return VM_FAULT_COMPLETED; |
89b15332 | 3160 | } |
97ba0c2b JK |
3161 | } |
3162 | ||
89b15332 | 3163 | return 0; |
97ba0c2b JK |
3164 | } |
3165 | ||
4e047f89 SR |
3166 | /* |
3167 | * Handle write page faults for pages that can be reused in the current vma | |
3168 | * | |
3169 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
3170 | * or due to us being the last reference standing to the page. In either | |
3171 | * case, all we need to do here is to mark the page as writable and update | |
3172 | * any related book-keeping. | |
3173 | */ | |
a86bc96b | 3174 | static inline void wp_page_reuse(struct vm_fault *vmf, struct folio *folio) |
82b0f8c3 | 3175 | __releases(vmf->ptl) |
4e047f89 | 3176 | { |
82b0f8c3 | 3177 | struct vm_area_struct *vma = vmf->vma; |
4e047f89 | 3178 | pte_t entry; |
6c287605 | 3179 | |
c89357e2 | 3180 | VM_BUG_ON(!(vmf->flags & FAULT_FLAG_WRITE)); |
6c287605 | 3181 | |
c2c3b514 KW |
3182 | if (folio) { |
3183 | VM_BUG_ON(folio_test_anon(folio) && | |
3184 | !PageAnonExclusive(vmf->page)); | |
3185 | /* | |
3186 | * Clear the folio's cpupid information as the existing | |
3187 | * information potentially belongs to a now completely | |
3188 | * unrelated process. | |
3189 | */ | |
3190 | folio_xchg_last_cpupid(folio, (1 << LAST_CPUPID_SHIFT) - 1); | |
3191 | } | |
4e047f89 | 3192 | |
2994302b JK |
3193 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
3194 | entry = pte_mkyoung(vmf->orig_pte); | |
4e047f89 | 3195 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
82b0f8c3 | 3196 | if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) |
5003a2bd | 3197 | update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1); |
82b0f8c3 | 3198 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
798a6b87 | 3199 | count_vm_event(PGREUSE); |
4e047f89 SR |
3200 | } |
3201 | ||
4ed43798 MWO |
3202 | /* |
3203 | * We could add a bitflag somewhere, but for now, we know that all | |
3204 | * vm_ops that have a ->map_pages have been audited and don't need | |
3205 | * the mmap_lock to be held. | |
3206 | */ | |
3207 | static inline vm_fault_t vmf_can_call_fault(const struct vm_fault *vmf) | |
3208 | { | |
3209 | struct vm_area_struct *vma = vmf->vma; | |
3210 | ||
3211 | if (vma->vm_ops->map_pages || !(vmf->flags & FAULT_FLAG_VMA_LOCK)) | |
3212 | return 0; | |
3213 | vma_end_read(vma); | |
3214 | return VM_FAULT_RETRY; | |
3215 | } | |
3216 | ||
a373baed MWO |
3217 | /** |
3218 | * vmf_anon_prepare - Prepare to handle an anonymous fault. | |
3219 | * @vmf: The vm_fault descriptor passed from the fault handler. | |
3220 | * | |
3221 | * When preparing to insert an anonymous page into a VMA from a | |
3222 | * fault handler, call this function rather than anon_vma_prepare(). | |
3223 | * If this vma does not already have an associated anon_vma and we are | |
3224 | * only protected by the per-VMA lock, the caller must retry with the | |
3225 | * mmap_lock held. __anon_vma_prepare() will look at adjacent VMAs to | |
3226 | * determine if this VMA can share its anon_vma, and that's not safe to | |
3227 | * do with only the per-VMA lock held for this VMA. | |
3228 | * | |
3229 | * Return: 0 if fault handling can proceed. Any other value should be | |
3230 | * returned to the caller. | |
3231 | */ | |
997f0ecb | 3232 | vm_fault_t vmf_anon_prepare(struct vm_fault *vmf) |
164b06f2 MWO |
3233 | { |
3234 | struct vm_area_struct *vma = vmf->vma; | |
737019cf | 3235 | vm_fault_t ret = 0; |
164b06f2 MWO |
3236 | |
3237 | if (likely(vma->anon_vma)) | |
3238 | return 0; | |
3239 | if (vmf->flags & FAULT_FLAG_VMA_LOCK) { | |
737019cf MWO |
3240 | if (!mmap_read_trylock(vma->vm_mm)) { |
3241 | vma_end_read(vma); | |
3242 | return VM_FAULT_RETRY; | |
3243 | } | |
164b06f2 MWO |
3244 | } |
3245 | if (__anon_vma_prepare(vma)) | |
737019cf MWO |
3246 | ret = VM_FAULT_OOM; |
3247 | if (vmf->flags & FAULT_FLAG_VMA_LOCK) | |
3248 | mmap_read_unlock(vma->vm_mm); | |
3249 | return ret; | |
164b06f2 MWO |
3250 | } |
3251 | ||
2f38ab2c | 3252 | /* |
c89357e2 DH |
3253 | * Handle the case of a page which we actually need to copy to a new page, |
3254 | * either due to COW or unsharing. | |
2f38ab2c | 3255 | * |
c1e8d7c6 | 3256 | * Called with mmap_lock locked and the old page referenced, but |
2f38ab2c SR |
3257 | * without the ptl held. |
3258 | * | |
3259 | * High level logic flow: | |
3260 | * | |
3261 | * - Allocate a page, copy the content of the old page to the new one. | |
3262 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
3263 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
3264 | * - If the pte is still the way we remember it, update the page table and all | |
3265 | * relevant references. This includes dropping the reference the page-table | |
3266 | * held to the old page, as well as updating the rmap. | |
3267 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
3268 | */ | |
2b740303 | 3269 | static vm_fault_t wp_page_copy(struct vm_fault *vmf) |
2f38ab2c | 3270 | { |
c89357e2 | 3271 | const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; |
82b0f8c3 | 3272 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 3273 | struct mm_struct *mm = vma->vm_mm; |
28d41a48 MWO |
3274 | struct folio *old_folio = NULL; |
3275 | struct folio *new_folio = NULL; | |
2f38ab2c SR |
3276 | pte_t entry; |
3277 | int page_copied = 0; | |
ac46d4f3 | 3278 | struct mmu_notifier_range range; |
164b06f2 | 3279 | vm_fault_t ret; |
cf503cc6 | 3280 | bool pfn_is_zero; |
2f38ab2c | 3281 | |
662ce1dc YY |
3282 | delayacct_wpcopy_start(); |
3283 | ||
28d41a48 MWO |
3284 | if (vmf->page) |
3285 | old_folio = page_folio(vmf->page); | |
164b06f2 MWO |
3286 | ret = vmf_anon_prepare(vmf); |
3287 | if (unlikely(ret)) | |
3288 | goto out; | |
2f38ab2c | 3289 | |
cf503cc6 KW |
3290 | pfn_is_zero = is_zero_pfn(pte_pfn(vmf->orig_pte)); |
3291 | new_folio = folio_prealloc(mm, vma, vmf->address, pfn_is_zero); | |
3292 | if (!new_folio) | |
3293 | goto oom; | |
3294 | ||
3295 | if (!pfn_is_zero) { | |
164b06f2 | 3296 | int err; |
83d116c5 | 3297 | |
164b06f2 MWO |
3298 | err = __wp_page_copy_user(&new_folio->page, vmf->page, vmf); |
3299 | if (err) { | |
83d116c5 JH |
3300 | /* |
3301 | * COW failed, if the fault was solved by other, | |
3302 | * it's fine. If not, userspace would re-fault on | |
3303 | * the same address and we will handle the fault | |
3304 | * from the second attempt. | |
a873dfe1 | 3305 | * The -EHWPOISON case will not be retried. |
83d116c5 | 3306 | */ |
28d41a48 MWO |
3307 | folio_put(new_folio); |
3308 | if (old_folio) | |
3309 | folio_put(old_folio); | |
662ce1dc YY |
3310 | |
3311 | delayacct_wpcopy_end(); | |
164b06f2 | 3312 | return err == -EHWPOISON ? VM_FAULT_HWPOISON : 0; |
83d116c5 | 3313 | } |
28d41a48 | 3314 | kmsan_copy_page_meta(&new_folio->page, vmf->page); |
2f38ab2c | 3315 | } |
2f38ab2c | 3316 | |
28d41a48 | 3317 | __folio_mark_uptodate(new_folio); |
eb3c24f3 | 3318 | |
7d4a8be0 | 3319 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, mm, |
6f4f13e8 | 3320 | vmf->address & PAGE_MASK, |
ac46d4f3 JG |
3321 | (vmf->address & PAGE_MASK) + PAGE_SIZE); |
3322 | mmu_notifier_invalidate_range_start(&range); | |
2f38ab2c SR |
3323 | |
3324 | /* | |
3325 | * Re-check the pte - we dropped the lock | |
3326 | */ | |
82b0f8c3 | 3327 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); |
c33c7948 | 3328 | if (likely(vmf->pte && pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { |
28d41a48 MWO |
3329 | if (old_folio) { |
3330 | if (!folio_test_anon(old_folio)) { | |
6b27cc6c | 3331 | dec_mm_counter(mm, mm_counter_file(old_folio)); |
f1a79412 | 3332 | inc_mm_counter(mm, MM_ANONPAGES); |
2f38ab2c SR |
3333 | } |
3334 | } else { | |
6080d19f | 3335 | ksm_might_unmap_zero_page(mm, vmf->orig_pte); |
f1a79412 | 3336 | inc_mm_counter(mm, MM_ANONPAGES); |
2f38ab2c | 3337 | } |
2994302b | 3338 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
28d41a48 | 3339 | entry = mk_pte(&new_folio->page, vma->vm_page_prot); |
50c25ee9 | 3340 | entry = pte_sw_mkyoung(entry); |
c89357e2 DH |
3341 | if (unlikely(unshare)) { |
3342 | if (pte_soft_dirty(vmf->orig_pte)) | |
3343 | entry = pte_mksoft_dirty(entry); | |
3344 | if (pte_uffd_wp(vmf->orig_pte)) | |
3345 | entry = pte_mkuffd_wp(entry); | |
3346 | } else { | |
3347 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
3348 | } | |
111fe718 | 3349 | |
2f38ab2c SR |
3350 | /* |
3351 | * Clear the pte entry and flush it first, before updating the | |
111fe718 NP |
3352 | * pte with the new entry, to keep TLBs on different CPUs in |
3353 | * sync. This code used to set the new PTE then flush TLBs, but | |
3354 | * that left a window where the new PTE could be loaded into | |
3355 | * some TLBs while the old PTE remains in others. | |
2f38ab2c | 3356 | */ |
ec8832d0 | 3357 | ptep_clear_flush(vma, vmf->address, vmf->pte); |
28d41a48 MWO |
3358 | folio_add_new_anon_rmap(new_folio, vma, vmf->address); |
3359 | folio_add_lru_vma(new_folio, vma); | |
c89357e2 | 3360 | BUG_ON(unshare && pte_write(entry)); |
f7842747 | 3361 | set_pte_at(mm, vmf->address, vmf->pte, entry); |
5003a2bd | 3362 | update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1); |
28d41a48 | 3363 | if (old_folio) { |
2f38ab2c SR |
3364 | /* |
3365 | * Only after switching the pte to the new page may | |
3366 | * we remove the mapcount here. Otherwise another | |
3367 | * process may come and find the rmap count decremented | |
3368 | * before the pte is switched to the new page, and | |
3369 | * "reuse" the old page writing into it while our pte | |
3370 | * here still points into it and can be read by other | |
3371 | * threads. | |
3372 | * | |
3373 | * The critical issue is to order this | |
c4626503 DH |
3374 | * folio_remove_rmap_pte() with the ptp_clear_flush |
3375 | * above. Those stores are ordered by (if nothing else,) | |
2f38ab2c | 3376 | * the barrier present in the atomic_add_negative |
c4626503 | 3377 | * in folio_remove_rmap_pte(); |
2f38ab2c SR |
3378 | * |
3379 | * Then the TLB flush in ptep_clear_flush ensures that | |
3380 | * no process can access the old page before the | |
3381 | * decremented mapcount is visible. And the old page | |
3382 | * cannot be reused until after the decremented | |
3383 | * mapcount is visible. So transitively, TLBs to | |
3384 | * old page will be flushed before it can be reused. | |
3385 | */ | |
c4626503 | 3386 | folio_remove_rmap_pte(old_folio, vmf->page, vma); |
2f38ab2c SR |
3387 | } |
3388 | ||
3389 | /* Free the old page.. */ | |
28d41a48 | 3390 | new_folio = old_folio; |
2f38ab2c | 3391 | page_copied = 1; |
3db82b93 HD |
3392 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3393 | } else if (vmf->pte) { | |
7df67697 | 3394 | update_mmu_tlb(vma, vmf->address, vmf->pte); |
3db82b93 | 3395 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
2f38ab2c SR |
3396 | } |
3397 | ||
ec8832d0 | 3398 | mmu_notifier_invalidate_range_end(&range); |
3db82b93 HD |
3399 | |
3400 | if (new_folio) | |
3401 | folio_put(new_folio); | |
28d41a48 | 3402 | if (old_folio) { |
f4c4a3f4 | 3403 | if (page_copied) |
63b77499 | 3404 | free_swap_cache(old_folio); |
28d41a48 | 3405 | folio_put(old_folio); |
2f38ab2c | 3406 | } |
662ce1dc YY |
3407 | |
3408 | delayacct_wpcopy_end(); | |
cb8d8633 | 3409 | return 0; |
2f38ab2c | 3410 | oom: |
164b06f2 MWO |
3411 | ret = VM_FAULT_OOM; |
3412 | out: | |
28d41a48 MWO |
3413 | if (old_folio) |
3414 | folio_put(old_folio); | |
662ce1dc YY |
3415 | |
3416 | delayacct_wpcopy_end(); | |
164b06f2 | 3417 | return ret; |
2f38ab2c SR |
3418 | } |
3419 | ||
66a6197c JK |
3420 | /** |
3421 | * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE | |
3422 | * writeable once the page is prepared | |
3423 | * | |
3424 | * @vmf: structure describing the fault | |
a86bc96b | 3425 | * @folio: the folio of vmf->page |
66a6197c JK |
3426 | * |
3427 | * This function handles all that is needed to finish a write page fault in a | |
3428 | * shared mapping due to PTE being read-only once the mapped page is prepared. | |
a862f68a | 3429 | * It handles locking of PTE and modifying it. |
66a6197c JK |
3430 | * |
3431 | * The function expects the page to be locked or other protection against | |
3432 | * concurrent faults / writeback (such as DAX radix tree locks). | |
a862f68a | 3433 | * |
2797e79f | 3434 | * Return: %0 on success, %VM_FAULT_NOPAGE when PTE got changed before |
a862f68a | 3435 | * we acquired PTE lock. |
66a6197c | 3436 | */ |
a86bc96b | 3437 | static vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf, struct folio *folio) |
66a6197c JK |
3438 | { |
3439 | WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); | |
3440 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, | |
3441 | &vmf->ptl); | |
3db82b93 HD |
3442 | if (!vmf->pte) |
3443 | return VM_FAULT_NOPAGE; | |
66a6197c JK |
3444 | /* |
3445 | * We might have raced with another page fault while we released the | |
3446 | * pte_offset_map_lock. | |
3447 | */ | |
c33c7948 | 3448 | if (!pte_same(ptep_get(vmf->pte), vmf->orig_pte)) { |
7df67697 | 3449 | update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); |
66a6197c | 3450 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a19e2553 | 3451 | return VM_FAULT_NOPAGE; |
66a6197c | 3452 | } |
a86bc96b | 3453 | wp_page_reuse(vmf, folio); |
a19e2553 | 3454 | return 0; |
66a6197c JK |
3455 | } |
3456 | ||
dd906184 BH |
3457 | /* |
3458 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
3459 | * mapping | |
3460 | */ | |
2b740303 | 3461 | static vm_fault_t wp_pfn_shared(struct vm_fault *vmf) |
dd906184 | 3462 | { |
82b0f8c3 | 3463 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 3464 | |
dd906184 | 3465 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
2b740303 | 3466 | vm_fault_t ret; |
dd906184 | 3467 | |
82b0f8c3 | 3468 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4a68fef1 MWO |
3469 | ret = vmf_can_call_fault(vmf); |
3470 | if (ret) | |
3471 | return ret; | |
063e60d8 | 3472 | |
fe82221f | 3473 | vmf->flags |= FAULT_FLAG_MKWRITE; |
11bac800 | 3474 | ret = vma->vm_ops->pfn_mkwrite(vmf); |
2f89dc12 | 3475 | if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) |
dd906184 | 3476 | return ret; |
a86bc96b | 3477 | return finish_mkwrite_fault(vmf, NULL); |
dd906184 | 3478 | } |
a86bc96b | 3479 | wp_page_reuse(vmf, NULL); |
cb8d8633 | 3480 | return 0; |
dd906184 BH |
3481 | } |
3482 | ||
5a97858b | 3483 | static vm_fault_t wp_page_shared(struct vm_fault *vmf, struct folio *folio) |
82b0f8c3 | 3484 | __releases(vmf->ptl) |
93e478d4 | 3485 | { |
82b0f8c3 | 3486 | struct vm_area_struct *vma = vmf->vma; |
cb8d8633 | 3487 | vm_fault_t ret = 0; |
93e478d4 | 3488 | |
5a97858b | 3489 | folio_get(folio); |
93e478d4 | 3490 | |
93e478d4 | 3491 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2b740303 | 3492 | vm_fault_t tmp; |
93e478d4 | 3493 | |
82b0f8c3 | 3494 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4a68fef1 MWO |
3495 | tmp = vmf_can_call_fault(vmf); |
3496 | if (tmp) { | |
063e60d8 | 3497 | folio_put(folio); |
4a68fef1 | 3498 | return tmp; |
063e60d8 MWO |
3499 | } |
3500 | ||
86aa6998 | 3501 | tmp = do_page_mkwrite(vmf, folio); |
93e478d4 SR |
3502 | if (unlikely(!tmp || (tmp & |
3503 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
5a97858b | 3504 | folio_put(folio); |
93e478d4 SR |
3505 | return tmp; |
3506 | } | |
a86bc96b | 3507 | tmp = finish_mkwrite_fault(vmf, folio); |
a19e2553 | 3508 | if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
5a97858b SK |
3509 | folio_unlock(folio); |
3510 | folio_put(folio); | |
66a6197c | 3511 | return tmp; |
93e478d4 | 3512 | } |
66a6197c | 3513 | } else { |
a86bc96b | 3514 | wp_page_reuse(vmf, folio); |
5a97858b | 3515 | folio_lock(folio); |
93e478d4 | 3516 | } |
89b15332 | 3517 | ret |= fault_dirty_shared_page(vmf); |
5a97858b | 3518 | folio_put(folio); |
93e478d4 | 3519 | |
89b15332 | 3520 | return ret; |
93e478d4 SR |
3521 | } |
3522 | ||
dec078cc DH |
3523 | static bool wp_can_reuse_anon_folio(struct folio *folio, |
3524 | struct vm_area_struct *vma) | |
3525 | { | |
cd197c3a BS |
3526 | /* |
3527 | * We could currently only reuse a subpage of a large folio if no | |
3528 | * other subpages of the large folios are still mapped. However, | |
3529 | * let's just consistently not reuse subpages even if we could | |
3530 | * reuse in that scenario, and give back a large folio a bit | |
3531 | * sooner. | |
3532 | */ | |
3533 | if (folio_test_large(folio)) | |
3534 | return false; | |
3535 | ||
dec078cc DH |
3536 | /* |
3537 | * We have to verify under folio lock: these early checks are | |
3538 | * just an optimization to avoid locking the folio and freeing | |
3539 | * the swapcache if there is little hope that we can reuse. | |
3540 | * | |
3541 | * KSM doesn't necessarily raise the folio refcount. | |
3542 | */ | |
3543 | if (folio_test_ksm(folio) || folio_ref_count(folio) > 3) | |
3544 | return false; | |
3545 | if (!folio_test_lru(folio)) | |
3546 | /* | |
3547 | * We cannot easily detect+handle references from | |
3548 | * remote LRU caches or references to LRU folios. | |
3549 | */ | |
3550 | lru_add_drain(); | |
3551 | if (folio_ref_count(folio) > 1 + folio_test_swapcache(folio)) | |
3552 | return false; | |
3553 | if (!folio_trylock(folio)) | |
3554 | return false; | |
3555 | if (folio_test_swapcache(folio)) | |
3556 | folio_free_swap(folio); | |
3557 | if (folio_test_ksm(folio) || folio_ref_count(folio) != 1) { | |
3558 | folio_unlock(folio); | |
3559 | return false; | |
3560 | } | |
3561 | /* | |
3562 | * Ok, we've got the only folio reference from our mapping | |
3563 | * and the folio is locked, it's dark out, and we're wearing | |
3564 | * sunglasses. Hit it. | |
3565 | */ | |
3566 | folio_move_anon_rmap(folio, vma); | |
3567 | folio_unlock(folio); | |
3568 | return true; | |
3569 | } | |
3570 | ||
1da177e4 | 3571 | /* |
c89357e2 DH |
3572 | * This routine handles present pages, when |
3573 | * * users try to write to a shared page (FAULT_FLAG_WRITE) | |
3574 | * * GUP wants to take a R/O pin on a possibly shared anonymous page | |
3575 | * (FAULT_FLAG_UNSHARE) | |
3576 | * | |
3577 | * It is done by copying the page to a new address and decrementing the | |
3578 | * shared-page counter for the old page. | |
1da177e4 | 3579 | * |
1da177e4 LT |
3580 | * Note that this routine assumes that the protection checks have been |
3581 | * done by the caller (the low-level page fault routine in most cases). | |
c89357e2 DH |
3582 | * Thus, with FAULT_FLAG_WRITE, we can safely just mark it writable once we've |
3583 | * done any necessary COW. | |
1da177e4 | 3584 | * |
c89357e2 DH |
3585 | * In case of FAULT_FLAG_WRITE, we also mark the page dirty at this point even |
3586 | * though the page will change only once the write actually happens. This | |
3587 | * avoids a few races, and potentially makes it more efficient. | |
1da177e4 | 3588 | * |
c1e8d7c6 | 3589 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
8f4e2101 | 3590 | * but allow concurrent faults), with pte both mapped and locked. |
c1e8d7c6 | 3591 | * We return with mmap_lock still held, but pte unmapped and unlocked. |
1da177e4 | 3592 | */ |
2b740303 | 3593 | static vm_fault_t do_wp_page(struct vm_fault *vmf) |
82b0f8c3 | 3594 | __releases(vmf->ptl) |
1da177e4 | 3595 | { |
c89357e2 | 3596 | const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; |
82b0f8c3 | 3597 | struct vm_area_struct *vma = vmf->vma; |
b9086fde | 3598 | struct folio *folio = NULL; |
d61ea1cb | 3599 | pte_t pte; |
1da177e4 | 3600 | |
c89357e2 | 3601 | if (likely(!unshare)) { |
c33c7948 | 3602 | if (userfaultfd_pte_wp(vma, ptep_get(vmf->pte))) { |
d61ea1cb PX |
3603 | if (!userfaultfd_wp_async(vma)) { |
3604 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3605 | return handle_userfault(vmf, VM_UFFD_WP); | |
3606 | } | |
3607 | ||
3608 | /* | |
3609 | * Nothing needed (cache flush, TLB invalidations, | |
3610 | * etc.) because we're only removing the uffd-wp bit, | |
3611 | * which is completely invisible to the user. | |
3612 | */ | |
3613 | pte = pte_clear_uffd_wp(ptep_get(vmf->pte)); | |
3614 | ||
3615 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); | |
3616 | /* | |
3617 | * Update this to be prepared for following up CoW | |
3618 | * handling | |
3619 | */ | |
3620 | vmf->orig_pte = pte; | |
c89357e2 DH |
3621 | } |
3622 | ||
3623 | /* | |
3624 | * Userfaultfd write-protect can defer flushes. Ensure the TLB | |
3625 | * is flushed in this case before copying. | |
3626 | */ | |
3627 | if (unlikely(userfaultfd_wp(vmf->vma) && | |
3628 | mm_tlb_flush_pending(vmf->vma->vm_mm))) | |
3629 | flush_tlb_page(vmf->vma, vmf->address); | |
3630 | } | |
6ce64428 | 3631 | |
a41b70d6 | 3632 | vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); |
c89357e2 | 3633 | |
5a97858b SK |
3634 | if (vmf->page) |
3635 | folio = page_folio(vmf->page); | |
3636 | ||
b9086fde DH |
3637 | /* |
3638 | * Shared mapping: we are guaranteed to have VM_WRITE and | |
3639 | * FAULT_FLAG_WRITE set at this point. | |
3640 | */ | |
3641 | if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { | |
251b97f5 | 3642 | /* |
64e45507 PF |
3643 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
3644 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
3645 | * |
3646 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 3647 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 | 3648 | */ |
b9086fde | 3649 | if (!vmf->page) |
2994302b | 3650 | return wp_pfn_shared(vmf); |
5a97858b | 3651 | return wp_page_shared(vmf, folio); |
251b97f5 | 3652 | } |
1da177e4 | 3653 | |
d08b3851 | 3654 | /* |
b9086fde DH |
3655 | * Private mapping: create an exclusive anonymous page copy if reuse |
3656 | * is impossible. We might miss VM_WRITE for FOLL_FORCE handling. | |
dec078cc DH |
3657 | * |
3658 | * If we encounter a page that is marked exclusive, we must reuse | |
3659 | * the page without further checks. | |
d08b3851 | 3660 | */ |
dec078cc DH |
3661 | if (folio && folio_test_anon(folio) && |
3662 | (PageAnonExclusive(vmf->page) || wp_can_reuse_anon_folio(folio, vma))) { | |
3663 | if (!PageAnonExclusive(vmf->page)) | |
3664 | SetPageAnonExclusive(vmf->page); | |
c89357e2 DH |
3665 | if (unlikely(unshare)) { |
3666 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3667 | return 0; | |
3668 | } | |
a86bc96b | 3669 | wp_page_reuse(vmf, folio); |
cb8d8633 | 3670 | return 0; |
1da177e4 | 3671 | } |
1da177e4 LT |
3672 | /* |
3673 | * Ok, we need to copy. Oh, well.. | |
3674 | */ | |
b9086fde DH |
3675 | if (folio) |
3676 | folio_get(folio); | |
28766805 | 3677 | |
82b0f8c3 | 3678 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
94bfe85b | 3679 | #ifdef CONFIG_KSM |
b9086fde | 3680 | if (folio && folio_test_ksm(folio)) |
94bfe85b YY |
3681 | count_vm_event(COW_KSM); |
3682 | #endif | |
a41b70d6 | 3683 | return wp_page_copy(vmf); |
1da177e4 LT |
3684 | } |
3685 | ||
97a89413 | 3686 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
3687 | unsigned long start_addr, unsigned long end_addr, |
3688 | struct zap_details *details) | |
3689 | { | |
f5cc4eef | 3690 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
3691 | } |
3692 | ||
f808c13f | 3693 | static inline void unmap_mapping_range_tree(struct rb_root_cached *root, |
232a6a1c PX |
3694 | pgoff_t first_index, |
3695 | pgoff_t last_index, | |
1da177e4 LT |
3696 | struct zap_details *details) |
3697 | { | |
3698 | struct vm_area_struct *vma; | |
1da177e4 LT |
3699 | pgoff_t vba, vea, zba, zea; |
3700 | ||
232a6a1c | 3701 | vma_interval_tree_foreach(vma, root, first_index, last_index) { |
1da177e4 | 3702 | vba = vma->vm_pgoff; |
d6e93217 | 3703 | vea = vba + vma_pages(vma) - 1; |
f9871da9 ML |
3704 | zba = max(first_index, vba); |
3705 | zea = min(last_index, vea); | |
1da177e4 | 3706 | |
97a89413 | 3707 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
3708 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
3709 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 3710 | details); |
1da177e4 LT |
3711 | } |
3712 | } | |
3713 | ||
22061a1f | 3714 | /** |
3506659e MWO |
3715 | * unmap_mapping_folio() - Unmap single folio from processes. |
3716 | * @folio: The locked folio to be unmapped. | |
22061a1f | 3717 | * |
3506659e | 3718 | * Unmap this folio from any userspace process which still has it mmaped. |
22061a1f HD |
3719 | * Typically, for efficiency, the range of nearby pages has already been |
3720 | * unmapped by unmap_mapping_pages() or unmap_mapping_range(). But once | |
3506659e MWO |
3721 | * truncation or invalidation holds the lock on a folio, it may find that |
3722 | * the page has been remapped again: and then uses unmap_mapping_folio() | |
22061a1f HD |
3723 | * to unmap it finally. |
3724 | */ | |
3506659e | 3725 | void unmap_mapping_folio(struct folio *folio) |
22061a1f | 3726 | { |
3506659e | 3727 | struct address_space *mapping = folio->mapping; |
22061a1f | 3728 | struct zap_details details = { }; |
232a6a1c PX |
3729 | pgoff_t first_index; |
3730 | pgoff_t last_index; | |
22061a1f | 3731 | |
3506659e | 3732 | VM_BUG_ON(!folio_test_locked(folio)); |
22061a1f | 3733 | |
3506659e | 3734 | first_index = folio->index; |
87b11f86 | 3735 | last_index = folio_next_index(folio) - 1; |
232a6a1c | 3736 | |
2e148f1e | 3737 | details.even_cows = false; |
3506659e | 3738 | details.single_folio = folio; |
999dad82 | 3739 | details.zap_flags = ZAP_FLAG_DROP_MARKER; |
22061a1f | 3740 | |
2c865995 | 3741 | i_mmap_lock_read(mapping); |
22061a1f | 3742 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) |
232a6a1c PX |
3743 | unmap_mapping_range_tree(&mapping->i_mmap, first_index, |
3744 | last_index, &details); | |
2c865995 | 3745 | i_mmap_unlock_read(mapping); |
22061a1f HD |
3746 | } |
3747 | ||
977fbdcd MW |
3748 | /** |
3749 | * unmap_mapping_pages() - Unmap pages from processes. | |
3750 | * @mapping: The address space containing pages to be unmapped. | |
3751 | * @start: Index of first page to be unmapped. | |
3752 | * @nr: Number of pages to be unmapped. 0 to unmap to end of file. | |
3753 | * @even_cows: Whether to unmap even private COWed pages. | |
3754 | * | |
3755 | * Unmap the pages in this address space from any userspace process which | |
3756 | * has them mmaped. Generally, you want to remove COWed pages as well when | |
3757 | * a file is being truncated, but not when invalidating pages from the page | |
3758 | * cache. | |
3759 | */ | |
3760 | void unmap_mapping_pages(struct address_space *mapping, pgoff_t start, | |
3761 | pgoff_t nr, bool even_cows) | |
3762 | { | |
3763 | struct zap_details details = { }; | |
232a6a1c PX |
3764 | pgoff_t first_index = start; |
3765 | pgoff_t last_index = start + nr - 1; | |
977fbdcd | 3766 | |
2e148f1e | 3767 | details.even_cows = even_cows; |
232a6a1c PX |
3768 | if (last_index < first_index) |
3769 | last_index = ULONG_MAX; | |
977fbdcd | 3770 | |
2c865995 | 3771 | i_mmap_lock_read(mapping); |
977fbdcd | 3772 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) |
232a6a1c PX |
3773 | unmap_mapping_range_tree(&mapping->i_mmap, first_index, |
3774 | last_index, &details); | |
2c865995 | 3775 | i_mmap_unlock_read(mapping); |
977fbdcd | 3776 | } |
6e0e99d5 | 3777 | EXPORT_SYMBOL_GPL(unmap_mapping_pages); |
977fbdcd | 3778 | |
1da177e4 | 3779 | /** |
8a5f14a2 | 3780 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
977fbdcd | 3781 | * address_space corresponding to the specified byte range in the underlying |
8a5f14a2 KS |
3782 | * file. |
3783 | * | |
3d41088f | 3784 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
3785 | * @holebegin: byte in first page to unmap, relative to the start of |
3786 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 3787 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
3788 | * must keep the partial page. In contrast, we must get rid of |
3789 | * partial pages. | |
3790 | * @holelen: size of prospective hole in bytes. This will be rounded | |
3791 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
3792 | * end of the file. | |
3793 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
3794 | * but 0 when invalidating pagecache, don't throw away private data. | |
3795 | */ | |
3796 | void unmap_mapping_range(struct address_space *mapping, | |
3797 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
3798 | { | |
9eab0421 JX |
3799 | pgoff_t hba = (pgoff_t)(holebegin) >> PAGE_SHIFT; |
3800 | pgoff_t hlen = ((pgoff_t)(holelen) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1da177e4 LT |
3801 | |
3802 | /* Check for overflow. */ | |
3803 | if (sizeof(holelen) > sizeof(hlen)) { | |
3804 | long long holeend = | |
3805 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
3806 | if (holeend & ~(long long)ULONG_MAX) | |
3807 | hlen = ULONG_MAX - hba + 1; | |
3808 | } | |
3809 | ||
977fbdcd | 3810 | unmap_mapping_pages(mapping, hba, hlen, even_cows); |
1da177e4 LT |
3811 | } |
3812 | EXPORT_SYMBOL(unmap_mapping_range); | |
3813 | ||
b756a3b5 AP |
3814 | /* |
3815 | * Restore a potential device exclusive pte to a working pte entry | |
3816 | */ | |
3817 | static vm_fault_t remove_device_exclusive_entry(struct vm_fault *vmf) | |
3818 | { | |
19672a9e | 3819 | struct folio *folio = page_folio(vmf->page); |
b756a3b5 AP |
3820 | struct vm_area_struct *vma = vmf->vma; |
3821 | struct mmu_notifier_range range; | |
fdc724d6 | 3822 | vm_fault_t ret; |
b756a3b5 | 3823 | |
7c7b9629 AP |
3824 | /* |
3825 | * We need a reference to lock the folio because we don't hold | |
3826 | * the PTL so a racing thread can remove the device-exclusive | |
3827 | * entry and unmap it. If the folio is free the entry must | |
3828 | * have been removed already. If it happens to have already | |
3829 | * been re-allocated after being freed all we do is lock and | |
3830 | * unlock it. | |
3831 | */ | |
3832 | if (!folio_try_get(folio)) | |
3833 | return 0; | |
3834 | ||
fdc724d6 SB |
3835 | ret = folio_lock_or_retry(folio, vmf); |
3836 | if (ret) { | |
7c7b9629 | 3837 | folio_put(folio); |
fdc724d6 | 3838 | return ret; |
7c7b9629 | 3839 | } |
7d4a8be0 | 3840 | mmu_notifier_range_init_owner(&range, MMU_NOTIFY_EXCLUSIVE, 0, |
b756a3b5 AP |
3841 | vma->vm_mm, vmf->address & PAGE_MASK, |
3842 | (vmf->address & PAGE_MASK) + PAGE_SIZE, NULL); | |
3843 | mmu_notifier_invalidate_range_start(&range); | |
3844 | ||
3845 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, | |
3846 | &vmf->ptl); | |
c33c7948 | 3847 | if (likely(vmf->pte && pte_same(ptep_get(vmf->pte), vmf->orig_pte))) |
19672a9e | 3848 | restore_exclusive_pte(vma, vmf->page, vmf->address, vmf->pte); |
b756a3b5 | 3849 | |
3db82b93 HD |
3850 | if (vmf->pte) |
3851 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
19672a9e | 3852 | folio_unlock(folio); |
7c7b9629 | 3853 | folio_put(folio); |
b756a3b5 AP |
3854 | |
3855 | mmu_notifier_invalidate_range_end(&range); | |
3856 | return 0; | |
3857 | } | |
3858 | ||
a160e537 | 3859 | static inline bool should_try_to_free_swap(struct folio *folio, |
c145e0b4 DH |
3860 | struct vm_area_struct *vma, |
3861 | unsigned int fault_flags) | |
3862 | { | |
a160e537 | 3863 | if (!folio_test_swapcache(folio)) |
c145e0b4 | 3864 | return false; |
9202d527 | 3865 | if (mem_cgroup_swap_full(folio) || (vma->vm_flags & VM_LOCKED) || |
a160e537 | 3866 | folio_test_mlocked(folio)) |
c145e0b4 DH |
3867 | return true; |
3868 | /* | |
3869 | * If we want to map a page that's in the swapcache writable, we | |
3870 | * have to detect via the refcount if we're really the exclusive | |
3871 | * user. Try freeing the swapcache to get rid of the swapcache | |
3872 | * reference only in case it's likely that we'll be the exlusive user. | |
3873 | */ | |
a160e537 MWO |
3874 | return (fault_flags & FAULT_FLAG_WRITE) && !folio_test_ksm(folio) && |
3875 | folio_ref_count(folio) == 2; | |
c145e0b4 DH |
3876 | } |
3877 | ||
9c28a205 PX |
3878 | static vm_fault_t pte_marker_clear(struct vm_fault *vmf) |
3879 | { | |
3880 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, | |
3881 | vmf->address, &vmf->ptl); | |
3db82b93 HD |
3882 | if (!vmf->pte) |
3883 | return 0; | |
9c28a205 PX |
3884 | /* |
3885 | * Be careful so that we will only recover a special uffd-wp pte into a | |
3886 | * none pte. Otherwise it means the pte could have changed, so retry. | |
7e3ce3f8 PX |
3887 | * |
3888 | * This should also cover the case where e.g. the pte changed | |
af19487f | 3889 | * quickly from a PTE_MARKER_UFFD_WP into PTE_MARKER_POISONED. |
7e3ce3f8 | 3890 | * So is_pte_marker() check is not enough to safely drop the pte. |
9c28a205 | 3891 | */ |
c33c7948 | 3892 | if (pte_same(vmf->orig_pte, ptep_get(vmf->pte))) |
9c28a205 PX |
3893 | pte_clear(vmf->vma->vm_mm, vmf->address, vmf->pte); |
3894 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3895 | return 0; | |
3896 | } | |
3897 | ||
2bad466c PX |
3898 | static vm_fault_t do_pte_missing(struct vm_fault *vmf) |
3899 | { | |
3900 | if (vma_is_anonymous(vmf->vma)) | |
3901 | return do_anonymous_page(vmf); | |
3902 | else | |
3903 | return do_fault(vmf); | |
3904 | } | |
3905 | ||
9c28a205 PX |
3906 | /* |
3907 | * This is actually a page-missing access, but with uffd-wp special pte | |
3908 | * installed. It means this pte was wr-protected before being unmapped. | |
3909 | */ | |
3910 | static vm_fault_t pte_marker_handle_uffd_wp(struct vm_fault *vmf) | |
3911 | { | |
3912 | /* | |
3913 | * Just in case there're leftover special ptes even after the region | |
7a079ba2 | 3914 | * got unregistered - we can simply clear them. |
9c28a205 | 3915 | */ |
2bad466c | 3916 | if (unlikely(!userfaultfd_wp(vmf->vma))) |
9c28a205 PX |
3917 | return pte_marker_clear(vmf); |
3918 | ||
2bad466c | 3919 | return do_pte_missing(vmf); |
9c28a205 PX |
3920 | } |
3921 | ||
5c041f5d PX |
3922 | static vm_fault_t handle_pte_marker(struct vm_fault *vmf) |
3923 | { | |
3924 | swp_entry_t entry = pte_to_swp_entry(vmf->orig_pte); | |
3925 | unsigned long marker = pte_marker_get(entry); | |
3926 | ||
3927 | /* | |
ca92ea3d PX |
3928 | * PTE markers should never be empty. If anything weird happened, |
3929 | * the best thing to do is to kill the process along with its mm. | |
5c041f5d | 3930 | */ |
ca92ea3d | 3931 | if (WARN_ON_ONCE(!marker)) |
5c041f5d PX |
3932 | return VM_FAULT_SIGBUS; |
3933 | ||
15520a3f | 3934 | /* Higher priority than uffd-wp when data corrupted */ |
af19487f AR |
3935 | if (marker & PTE_MARKER_POISONED) |
3936 | return VM_FAULT_HWPOISON; | |
15520a3f | 3937 | |
9c28a205 PX |
3938 | if (pte_marker_entry_uffd_wp(entry)) |
3939 | return pte_marker_handle_uffd_wp(vmf); | |
3940 | ||
3941 | /* This is an unknown pte marker */ | |
3942 | return VM_FAULT_SIGBUS; | |
5c041f5d PX |
3943 | } |
3944 | ||
1da177e4 | 3945 | /* |
c1e8d7c6 | 3946 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
8f4e2101 | 3947 | * but allow concurrent faults), and pte mapped but not yet locked. |
9a95f3cf PC |
3948 | * We return with pte unmapped and unlocked. |
3949 | * | |
c1e8d7c6 | 3950 | * We return with the mmap_lock locked or unlocked in the same cases |
9a95f3cf | 3951 | * as does filemap_fault(). |
1da177e4 | 3952 | */ |
2b740303 | 3953 | vm_fault_t do_swap_page(struct vm_fault *vmf) |
1da177e4 | 3954 | { |
82b0f8c3 | 3955 | struct vm_area_struct *vma = vmf->vma; |
d4f9565a MWO |
3956 | struct folio *swapcache, *folio = NULL; |
3957 | struct page *page; | |
2799e775 | 3958 | struct swap_info_struct *si = NULL; |
14f9135d | 3959 | rmap_t rmap_flags = RMAP_NONE; |
13ddaf26 | 3960 | bool need_clear_cache = false; |
1493a191 | 3961 | bool exclusive = false; |
65500d23 | 3962 | swp_entry_t entry; |
1da177e4 | 3963 | pte_t pte; |
2b740303 | 3964 | vm_fault_t ret = 0; |
aae466b0 | 3965 | void *shadow = NULL; |
1da177e4 | 3966 | |
2ca99358 | 3967 | if (!pte_unmap_same(vmf)) |
8f4e2101 | 3968 | goto out; |
65500d23 | 3969 | |
2994302b | 3970 | entry = pte_to_swp_entry(vmf->orig_pte); |
d1737fdb AK |
3971 | if (unlikely(non_swap_entry(entry))) { |
3972 | if (is_migration_entry(entry)) { | |
82b0f8c3 JK |
3973 | migration_entry_wait(vma->vm_mm, vmf->pmd, |
3974 | vmf->address); | |
b756a3b5 AP |
3975 | } else if (is_device_exclusive_entry(entry)) { |
3976 | vmf->page = pfn_swap_entry_to_page(entry); | |
3977 | ret = remove_device_exclusive_entry(vmf); | |
5042db43 | 3978 | } else if (is_device_private_entry(entry)) { |
1235ccd0 SB |
3979 | if (vmf->flags & FAULT_FLAG_VMA_LOCK) { |
3980 | /* | |
3981 | * migrate_to_ram is not yet ready to operate | |
3982 | * under VMA lock. | |
3983 | */ | |
3984 | vma_end_read(vma); | |
3985 | ret = VM_FAULT_RETRY; | |
3986 | goto out; | |
3987 | } | |
3988 | ||
af5cdaf8 | 3989 | vmf->page = pfn_swap_entry_to_page(entry); |
16ce101d AP |
3990 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
3991 | vmf->address, &vmf->ptl); | |
3db82b93 | 3992 | if (unlikely(!vmf->pte || |
c33c7948 RR |
3993 | !pte_same(ptep_get(vmf->pte), |
3994 | vmf->orig_pte))) | |
3b65f437 | 3995 | goto unlock; |
16ce101d AP |
3996 | |
3997 | /* | |
3998 | * Get a page reference while we know the page can't be | |
3999 | * freed. | |
4000 | */ | |
4001 | get_page(vmf->page); | |
4002 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4a955bed | 4003 | ret = vmf->page->pgmap->ops->migrate_to_ram(vmf); |
16ce101d | 4004 | put_page(vmf->page); |
d1737fdb AK |
4005 | } else if (is_hwpoison_entry(entry)) { |
4006 | ret = VM_FAULT_HWPOISON; | |
5c041f5d PX |
4007 | } else if (is_pte_marker_entry(entry)) { |
4008 | ret = handle_pte_marker(vmf); | |
d1737fdb | 4009 | } else { |
2994302b | 4010 | print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); |
d99be1a8 | 4011 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 4012 | } |
0697212a CL |
4013 | goto out; |
4014 | } | |
0bcac06f | 4015 | |
2799e775 ML |
4016 | /* Prevent swapoff from happening to us. */ |
4017 | si = get_swap_device(entry); | |
4018 | if (unlikely(!si)) | |
4019 | goto out; | |
0bcac06f | 4020 | |
5a423081 MWO |
4021 | folio = swap_cache_get_folio(entry, vma, vmf->address); |
4022 | if (folio) | |
4023 | page = folio_file_page(folio, swp_offset(entry)); | |
d4f9565a | 4024 | swapcache = folio; |
f8020772 | 4025 | |
d4f9565a | 4026 | if (!folio) { |
a449bf58 QC |
4027 | if (data_race(si->flags & SWP_SYNCHRONOUS_IO) && |
4028 | __swap_count(entry) == 1) { | |
13ddaf26 KS |
4029 | /* |
4030 | * Prevent parallel swapin from proceeding with | |
4031 | * the cache flag. Otherwise, another thread may | |
4032 | * finish swapin first, free the entry, and swapout | |
4033 | * reusing the same entry. It's undetectable as | |
4034 | * pte_same() returns true due to entry reuse. | |
4035 | */ | |
4036 | if (swapcache_prepare(entry)) { | |
4037 | /* Relax a bit to prevent rapid repeated page faults */ | |
4038 | schedule_timeout_uninterruptible(1); | |
4039 | goto out; | |
4040 | } | |
4041 | need_clear_cache = true; | |
4042 | ||
0bcac06f | 4043 | /* skip swapcache */ |
63ad4add MWO |
4044 | folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, |
4045 | vma, vmf->address, false); | |
4046 | page = &folio->page; | |
4047 | if (folio) { | |
4048 | __folio_set_locked(folio); | |
4049 | __folio_set_swapbacked(folio); | |
4c6355b2 | 4050 | |
65995918 | 4051 | if (mem_cgroup_swapin_charge_folio(folio, |
63ad4add MWO |
4052 | vma->vm_mm, GFP_KERNEL, |
4053 | entry)) { | |
545b1b07 | 4054 | ret = VM_FAULT_OOM; |
4c6355b2 | 4055 | goto out_page; |
545b1b07 | 4056 | } |
0add0c77 | 4057 | mem_cgroup_swapin_uncharge_swap(entry); |
4c6355b2 | 4058 | |
aae466b0 JK |
4059 | shadow = get_shadow_from_swap_cache(entry); |
4060 | if (shadow) | |
63ad4add | 4061 | workingset_refault(folio, shadow); |
0076f029 | 4062 | |
63ad4add | 4063 | folio_add_lru(folio); |
0add0c77 | 4064 | |
c9bdf768 | 4065 | /* To provide entry to swap_read_folio() */ |
3d2c9087 | 4066 | folio->swap = entry; |
c9bdf768 | 4067 | swap_read_folio(folio, true, NULL); |
63ad4add | 4068 | folio->private = NULL; |
0bcac06f | 4069 | } |
aa8d22a1 | 4070 | } else { |
e9e9b7ec MK |
4071 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
4072 | vmf); | |
63ad4add MWO |
4073 | if (page) |
4074 | folio = page_folio(page); | |
d4f9565a | 4075 | swapcache = folio; |
0bcac06f MK |
4076 | } |
4077 | ||
d4f9565a | 4078 | if (!folio) { |
1da177e4 | 4079 | /* |
8f4e2101 HD |
4080 | * Back out if somebody else faulted in this pte |
4081 | * while we released the pte lock. | |
1da177e4 | 4082 | */ |
82b0f8c3 JK |
4083 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
4084 | vmf->address, &vmf->ptl); | |
c33c7948 RR |
4085 | if (likely(vmf->pte && |
4086 | pte_same(ptep_get(vmf->pte), vmf->orig_pte))) | |
1da177e4 | 4087 | ret = VM_FAULT_OOM; |
65500d23 | 4088 | goto unlock; |
1da177e4 LT |
4089 | } |
4090 | ||
4091 | /* Had to read the page from swap area: Major fault */ | |
4092 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 4093 | count_vm_event(PGMAJFAULT); |
2262185c | 4094 | count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); |
d1737fdb | 4095 | } else if (PageHWPoison(page)) { |
71f72525 WF |
4096 | /* |
4097 | * hwpoisoned dirty swapcache pages are kept for killing | |
4098 | * owner processes (which may be unknown at hwpoison time) | |
4099 | */ | |
d1737fdb | 4100 | ret = VM_FAULT_HWPOISON; |
4779cb31 | 4101 | goto out_release; |
1da177e4 LT |
4102 | } |
4103 | ||
fdc724d6 SB |
4104 | ret |= folio_lock_or_retry(folio, vmf); |
4105 | if (ret & VM_FAULT_RETRY) | |
d065bd81 | 4106 | goto out_release; |
073e587e | 4107 | |
84d60fdd DH |
4108 | if (swapcache) { |
4109 | /* | |
3b344157 | 4110 | * Make sure folio_free_swap() or swapoff did not release the |
84d60fdd DH |
4111 | * swapcache from under us. The page pin, and pte_same test |
4112 | * below, are not enough to exclude that. Even if it is still | |
4113 | * swapcache, we need to check that the page's swap has not | |
4114 | * changed. | |
4115 | */ | |
63ad4add | 4116 | if (unlikely(!folio_test_swapcache(folio) || |
cfeed8ff | 4117 | page_swap_entry(page).val != entry.val)) |
84d60fdd DH |
4118 | goto out_page; |
4119 | ||
4120 | /* | |
4121 | * KSM sometimes has to copy on read faults, for example, if | |
4122 | * page->index of !PageKSM() pages would be nonlinear inside the | |
4123 | * anon VMA -- PageKSM() is lost on actual swapout. | |
4124 | */ | |
96db66d9 MWO |
4125 | folio = ksm_might_need_to_copy(folio, vma, vmf->address); |
4126 | if (unlikely(!folio)) { | |
84d60fdd | 4127 | ret = VM_FAULT_OOM; |
96db66d9 | 4128 | folio = swapcache; |
84d60fdd | 4129 | goto out_page; |
96db66d9 | 4130 | } else if (unlikely(folio == ERR_PTR(-EHWPOISON))) { |
6b970599 | 4131 | ret = VM_FAULT_HWPOISON; |
96db66d9 | 4132 | folio = swapcache; |
6b970599 | 4133 | goto out_page; |
84d60fdd | 4134 | } |
96db66d9 MWO |
4135 | if (folio != swapcache) |
4136 | page = folio_page(folio, 0); | |
c145e0b4 DH |
4137 | |
4138 | /* | |
4139 | * If we want to map a page that's in the swapcache writable, we | |
4140 | * have to detect via the refcount if we're really the exclusive | |
4141 | * owner. Try removing the extra reference from the local LRU | |
1fec6890 | 4142 | * caches if required. |
c145e0b4 | 4143 | */ |
d4f9565a | 4144 | if ((vmf->flags & FAULT_FLAG_WRITE) && folio == swapcache && |
63ad4add | 4145 | !folio_test_ksm(folio) && !folio_test_lru(folio)) |
c145e0b4 | 4146 | lru_add_drain(); |
5ad64688 HD |
4147 | } |
4148 | ||
4231f842 | 4149 | folio_throttle_swaprate(folio, GFP_KERNEL); |
8a9f3ccd | 4150 | |
1da177e4 | 4151 | /* |
8f4e2101 | 4152 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 4153 | */ |
82b0f8c3 JK |
4154 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
4155 | &vmf->ptl); | |
c33c7948 | 4156 | if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte))) |
b8107480 | 4157 | goto out_nomap; |
b8107480 | 4158 | |
63ad4add | 4159 | if (unlikely(!folio_test_uptodate(folio))) { |
b8107480 KK |
4160 | ret = VM_FAULT_SIGBUS; |
4161 | goto out_nomap; | |
1da177e4 LT |
4162 | } |
4163 | ||
78fbe906 DH |
4164 | /* |
4165 | * PG_anon_exclusive reuses PG_mappedtodisk for anon pages. A swap pte | |
4166 | * must never point at an anonymous page in the swapcache that is | |
4167 | * PG_anon_exclusive. Sanity check that this holds and especially, that | |
4168 | * no filesystem set PG_mappedtodisk on a page in the swapcache. Sanity | |
4169 | * check after taking the PT lock and making sure that nobody | |
4170 | * concurrently faulted in this page and set PG_anon_exclusive. | |
4171 | */ | |
63ad4add MWO |
4172 | BUG_ON(!folio_test_anon(folio) && folio_test_mappedtodisk(folio)); |
4173 | BUG_ON(folio_test_anon(folio) && PageAnonExclusive(page)); | |
78fbe906 | 4174 | |
1493a191 DH |
4175 | /* |
4176 | * Check under PT lock (to protect against concurrent fork() sharing | |
4177 | * the swap entry concurrently) for certainly exclusive pages. | |
4178 | */ | |
63ad4add | 4179 | if (!folio_test_ksm(folio)) { |
1493a191 | 4180 | exclusive = pte_swp_exclusive(vmf->orig_pte); |
d4f9565a | 4181 | if (folio != swapcache) { |
1493a191 DH |
4182 | /* |
4183 | * We have a fresh page that is not exposed to the | |
4184 | * swapcache -> certainly exclusive. | |
4185 | */ | |
4186 | exclusive = true; | |
63ad4add | 4187 | } else if (exclusive && folio_test_writeback(folio) && |
eacde327 | 4188 | data_race(si->flags & SWP_STABLE_WRITES)) { |
1493a191 DH |
4189 | /* |
4190 | * This is tricky: not all swap backends support | |
4191 | * concurrent page modifications while under writeback. | |
4192 | * | |
4193 | * So if we stumble over such a page in the swapcache | |
4194 | * we must not set the page exclusive, otherwise we can | |
4195 | * map it writable without further checks and modify it | |
4196 | * while still under writeback. | |
4197 | * | |
4198 | * For these problematic swap backends, simply drop the | |
4199 | * exclusive marker: this is perfectly fine as we start | |
4200 | * writeback only if we fully unmapped the page and | |
4201 | * there are no unexpected references on the page after | |
4202 | * unmapping succeeded. After fully unmapped, no | |
4203 | * further GUP references (FOLL_GET and FOLL_PIN) can | |
4204 | * appear, so dropping the exclusive marker and mapping | |
4205 | * it only R/O is fine. | |
4206 | */ | |
4207 | exclusive = false; | |
4208 | } | |
4209 | } | |
4210 | ||
6dca4ac6 PC |
4211 | /* |
4212 | * Some architectures may have to restore extra metadata to the page | |
4213 | * when reading from swap. This metadata may be indexed by swap entry | |
4214 | * so this must be called before swap_free(). | |
4215 | */ | |
f238b8c3 | 4216 | arch_swap_restore(folio_swap(entry, folio), folio); |
6dca4ac6 | 4217 | |
8c7c6e34 | 4218 | /* |
c145e0b4 DH |
4219 | * Remove the swap entry and conditionally try to free up the swapcache. |
4220 | * We're already holding a reference on the page but haven't mapped it | |
4221 | * yet. | |
8c7c6e34 | 4222 | */ |
c145e0b4 | 4223 | swap_free(entry); |
a160e537 MWO |
4224 | if (should_try_to_free_swap(folio, vma, vmf->flags)) |
4225 | folio_free_swap(folio); | |
1da177e4 | 4226 | |
f1a79412 SB |
4227 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
4228 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); | |
1da177e4 | 4229 | pte = mk_pte(page, vma->vm_page_prot); |
c145e0b4 DH |
4230 | |
4231 | /* | |
1493a191 DH |
4232 | * Same logic as in do_wp_page(); however, optimize for pages that are |
4233 | * certainly not shared either because we just allocated them without | |
4234 | * exposing them to the swapcache or because the swap entry indicates | |
4235 | * exclusivity. | |
c145e0b4 | 4236 | */ |
63ad4add MWO |
4237 | if (!folio_test_ksm(folio) && |
4238 | (exclusive || folio_ref_count(folio) == 1)) { | |
6c287605 DH |
4239 | if (vmf->flags & FAULT_FLAG_WRITE) { |
4240 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); | |
4241 | vmf->flags &= ~FAULT_FLAG_WRITE; | |
6c287605 | 4242 | } |
14f9135d | 4243 | rmap_flags |= RMAP_EXCLUSIVE; |
1da177e4 | 4244 | } |
1da177e4 | 4245 | flush_icache_page(vma, page); |
2994302b | 4246 | if (pte_swp_soft_dirty(vmf->orig_pte)) |
179ef71c | 4247 | pte = pte_mksoft_dirty(pte); |
f1eb1bac | 4248 | if (pte_swp_uffd_wp(vmf->orig_pte)) |
f45ec5ff | 4249 | pte = pte_mkuffd_wp(pte); |
2994302b | 4250 | vmf->orig_pte = pte; |
0bcac06f MK |
4251 | |
4252 | /* ksm created a completely new copy */ | |
d4f9565a | 4253 | if (unlikely(folio != swapcache && swapcache)) { |
2853b66b | 4254 | folio_add_new_anon_rmap(folio, vma, vmf->address); |
63ad4add | 4255 | folio_add_lru_vma(folio, vma); |
0bcac06f | 4256 | } else { |
b832a354 DH |
4257 | folio_add_anon_rmap_pte(folio, page, vma, vmf->address, |
4258 | rmap_flags); | |
00501b53 | 4259 | } |
1da177e4 | 4260 | |
63ad4add MWO |
4261 | VM_BUG_ON(!folio_test_anon(folio) || |
4262 | (pte_write(pte) && !PageAnonExclusive(page))); | |
1eba86c0 PT |
4263 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); |
4264 | arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte); | |
4265 | ||
63ad4add | 4266 | folio_unlock(folio); |
d4f9565a | 4267 | if (folio != swapcache && swapcache) { |
4969c119 AA |
4268 | /* |
4269 | * Hold the lock to avoid the swap entry to be reused | |
4270 | * until we take the PT lock for the pte_same() check | |
4271 | * (to avoid false positives from pte_same). For | |
4272 | * further safety release the lock after the swap_free | |
4273 | * so that the swap count won't change under a | |
4274 | * parallel locked swapcache. | |
4275 | */ | |
d4f9565a MWO |
4276 | folio_unlock(swapcache); |
4277 | folio_put(swapcache); | |
4969c119 | 4278 | } |
c475a8ab | 4279 | |
82b0f8c3 | 4280 | if (vmf->flags & FAULT_FLAG_WRITE) { |
2994302b | 4281 | ret |= do_wp_page(vmf); |
61469f1d HD |
4282 | if (ret & VM_FAULT_ERROR) |
4283 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
4284 | goto out; |
4285 | } | |
4286 | ||
4287 | /* No need to invalidate - it was non-present before */ | |
5003a2bd | 4288 | update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1); |
65500d23 | 4289 | unlock: |
3db82b93 HD |
4290 | if (vmf->pte) |
4291 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
1da177e4 | 4292 | out: |
13ddaf26 KS |
4293 | /* Clear the swap cache pin for direct swapin after PTL unlock */ |
4294 | if (need_clear_cache) | |
4295 | swapcache_clear(si, entry); | |
2799e775 ML |
4296 | if (si) |
4297 | put_swap_device(si); | |
1da177e4 | 4298 | return ret; |
b8107480 | 4299 | out_nomap: |
3db82b93 HD |
4300 | if (vmf->pte) |
4301 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
bc43f75c | 4302 | out_page: |
63ad4add | 4303 | folio_unlock(folio); |
4779cb31 | 4304 | out_release: |
63ad4add | 4305 | folio_put(folio); |
d4f9565a MWO |
4306 | if (folio != swapcache && swapcache) { |
4307 | folio_unlock(swapcache); | |
4308 | folio_put(swapcache); | |
4969c119 | 4309 | } |
13ddaf26 KS |
4310 | if (need_clear_cache) |
4311 | swapcache_clear(si, entry); | |
2799e775 ML |
4312 | if (si) |
4313 | put_swap_device(si); | |
65500d23 | 4314 | return ret; |
1da177e4 LT |
4315 | } |
4316 | ||
19eaf449 RR |
4317 | static bool pte_range_none(pte_t *pte, int nr_pages) |
4318 | { | |
4319 | int i; | |
4320 | ||
4321 | for (i = 0; i < nr_pages; i++) { | |
4322 | if (!pte_none(ptep_get_lockless(pte + i))) | |
4323 | return false; | |
4324 | } | |
4325 | ||
4326 | return true; | |
4327 | } | |
4328 | ||
4329 | static struct folio *alloc_anon_folio(struct vm_fault *vmf) | |
4330 | { | |
19eaf449 | 4331 | struct vm_area_struct *vma = vmf->vma; |
085ff35e | 4332 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
19eaf449 RR |
4333 | unsigned long orders; |
4334 | struct folio *folio; | |
4335 | unsigned long addr; | |
4336 | pte_t *pte; | |
4337 | gfp_t gfp; | |
4338 | int order; | |
4339 | ||
4340 | /* | |
4341 | * If uffd is active for the vma we need per-page fault fidelity to | |
4342 | * maintain the uffd semantics. | |
4343 | */ | |
4344 | if (unlikely(userfaultfd_armed(vma))) | |
4345 | goto fallback; | |
4346 | ||
4347 | /* | |
4348 | * Get a list of all the (large) orders below PMD_ORDER that are enabled | |
4349 | * for this vma. Then filter out the orders that can't be allocated over | |
4350 | * the faulting address and still be fully contained in the vma. | |
4351 | */ | |
e0ffb29b MW |
4352 | orders = thp_vma_allowable_orders(vma, vma->vm_flags, |
4353 | TVA_IN_PF | TVA_ENFORCE_SYSFS, BIT(PMD_ORDER) - 1); | |
19eaf449 RR |
4354 | orders = thp_vma_suitable_orders(vma, vmf->address, orders); |
4355 | ||
4356 | if (!orders) | |
4357 | goto fallback; | |
4358 | ||
4359 | pte = pte_offset_map(vmf->pmd, vmf->address & PMD_MASK); | |
4360 | if (!pte) | |
4361 | return ERR_PTR(-EAGAIN); | |
4362 | ||
4363 | /* | |
4364 | * Find the highest order where the aligned range is completely | |
4365 | * pte_none(). Note that all remaining orders will be completely | |
4366 | * pte_none(). | |
4367 | */ | |
4368 | order = highest_order(orders); | |
4369 | while (orders) { | |
4370 | addr = ALIGN_DOWN(vmf->address, PAGE_SIZE << order); | |
4371 | if (pte_range_none(pte + pte_index(addr), 1 << order)) | |
4372 | break; | |
4373 | order = next_order(&orders, order); | |
4374 | } | |
4375 | ||
4376 | pte_unmap(pte); | |
4377 | ||
68dbcf48 BS |
4378 | if (!orders) |
4379 | goto fallback; | |
4380 | ||
19eaf449 RR |
4381 | /* Try allocating the highest of the remaining orders. */ |
4382 | gfp = vma_thp_gfp_mask(vma); | |
4383 | while (orders) { | |
4384 | addr = ALIGN_DOWN(vmf->address, PAGE_SIZE << order); | |
4385 | folio = vma_alloc_folio(gfp, order, vma, addr, true); | |
4386 | if (folio) { | |
085ff35e | 4387 | if (mem_cgroup_charge(folio, vma->vm_mm, gfp)) { |
ec33687c | 4388 | count_mthp_stat(order, MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE); |
085ff35e KW |
4389 | folio_put(folio); |
4390 | goto next; | |
4391 | } | |
4392 | folio_throttle_swaprate(folio, gfp); | |
19eaf449 RR |
4393 | clear_huge_page(&folio->page, vmf->address, 1 << order); |
4394 | return folio; | |
4395 | } | |
085ff35e | 4396 | next: |
ec33687c | 4397 | count_mthp_stat(order, MTHP_STAT_ANON_FAULT_FALLBACK); |
19eaf449 RR |
4398 | order = next_order(&orders, order); |
4399 | } | |
4400 | ||
4401 | fallback: | |
4402 | #endif | |
085ff35e | 4403 | return folio_prealloc(vma->vm_mm, vma, vmf->address, true); |
19eaf449 RR |
4404 | } |
4405 | ||
1da177e4 | 4406 | /* |
c1e8d7c6 | 4407 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
8f4e2101 | 4408 | * but allow concurrent faults), and pte mapped but not yet locked. |
c1e8d7c6 | 4409 | * We return with mmap_lock still held, but pte unmapped and unlocked. |
1da177e4 | 4410 | */ |
2b740303 | 4411 | static vm_fault_t do_anonymous_page(struct vm_fault *vmf) |
1da177e4 | 4412 | { |
82b0f8c3 | 4413 | struct vm_area_struct *vma = vmf->vma; |
19eaf449 | 4414 | unsigned long addr = vmf->address; |
6bc56a4d | 4415 | struct folio *folio; |
2b740303 | 4416 | vm_fault_t ret = 0; |
19eaf449 | 4417 | int nr_pages = 1; |
1da177e4 | 4418 | pte_t entry; |
19eaf449 | 4419 | int i; |
1da177e4 | 4420 | |
6b7339f4 KS |
4421 | /* File mapping without ->vm_ops ? */ |
4422 | if (vma->vm_flags & VM_SHARED) | |
4423 | return VM_FAULT_SIGBUS; | |
4424 | ||
7267ec00 | 4425 | /* |
3db82b93 HD |
4426 | * Use pte_alloc() instead of pte_alloc_map(), so that OOM can |
4427 | * be distinguished from a transient failure of pte_offset_map(). | |
7267ec00 | 4428 | */ |
4cf58924 | 4429 | if (pte_alloc(vma->vm_mm, vmf->pmd)) |
7267ec00 KS |
4430 | return VM_FAULT_OOM; |
4431 | ||
11ac5524 | 4432 | /* Use the zero-page for reads */ |
82b0f8c3 | 4433 | if (!(vmf->flags & FAULT_FLAG_WRITE) && |
bae473a4 | 4434 | !mm_forbids_zeropage(vma->vm_mm)) { |
82b0f8c3 | 4435 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), |
62eede62 | 4436 | vma->vm_page_prot)); |
82b0f8c3 JK |
4437 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
4438 | vmf->address, &vmf->ptl); | |
3db82b93 HD |
4439 | if (!vmf->pte) |
4440 | goto unlock; | |
2bad466c | 4441 | if (vmf_pte_changed(vmf)) { |
7df67697 | 4442 | update_mmu_tlb(vma, vmf->address, vmf->pte); |
a13ea5b7 | 4443 | goto unlock; |
7df67697 | 4444 | } |
6b31d595 MH |
4445 | ret = check_stable_address_space(vma->vm_mm); |
4446 | if (ret) | |
4447 | goto unlock; | |
6b251fc9 AA |
4448 | /* Deliver the page fault to userland, check inside PT lock */ |
4449 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 JK |
4450 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4451 | return handle_userfault(vmf, VM_UFFD_MISSING); | |
6b251fc9 | 4452 | } |
a13ea5b7 HD |
4453 | goto setpte; |
4454 | } | |
4455 | ||
557ed1fa | 4456 | /* Allocate our own private page. */ |
a373baed MWO |
4457 | ret = vmf_anon_prepare(vmf); |
4458 | if (ret) | |
4459 | return ret; | |
19eaf449 RR |
4460 | /* Returns NULL on OOM or ERR_PTR(-EAGAIN) if we must retry the fault */ |
4461 | folio = alloc_anon_folio(vmf); | |
4462 | if (IS_ERR(folio)) | |
4463 | return 0; | |
6bc56a4d | 4464 | if (!folio) |
557ed1fa | 4465 | goto oom; |
eb3c24f3 | 4466 | |
19eaf449 RR |
4467 | nr_pages = folio_nr_pages(folio); |
4468 | addr = ALIGN_DOWN(vmf->address, nr_pages * PAGE_SIZE); | |
4469 | ||
52f37629 | 4470 | /* |
cb3184de | 4471 | * The memory barrier inside __folio_mark_uptodate makes sure that |
f4f5329d | 4472 | * preceding stores to the page contents become visible before |
52f37629 MK |
4473 | * the set_pte_at() write. |
4474 | */ | |
cb3184de | 4475 | __folio_mark_uptodate(folio); |
8f4e2101 | 4476 | |
cb3184de | 4477 | entry = mk_pte(&folio->page, vma->vm_page_prot); |
50c25ee9 | 4478 | entry = pte_sw_mkyoung(entry); |
1ac0cb5d | 4479 | if (vma->vm_flags & VM_WRITE) |
161e393c | 4480 | entry = pte_mkwrite(pte_mkdirty(entry), vma); |
1da177e4 | 4481 | |
19eaf449 | 4482 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); |
3db82b93 HD |
4483 | if (!vmf->pte) |
4484 | goto release; | |
19eaf449 RR |
4485 | if (nr_pages == 1 && vmf_pte_changed(vmf)) { |
4486 | update_mmu_tlb(vma, addr, vmf->pte); | |
4487 | goto release; | |
4488 | } else if (nr_pages > 1 && !pte_range_none(vmf->pte, nr_pages)) { | |
4489 | for (i = 0; i < nr_pages; i++) | |
4490 | update_mmu_tlb(vma, addr + PAGE_SIZE * i, vmf->pte + i); | |
557ed1fa | 4491 | goto release; |
7df67697 | 4492 | } |
9ba69294 | 4493 | |
6b31d595 MH |
4494 | ret = check_stable_address_space(vma->vm_mm); |
4495 | if (ret) | |
4496 | goto release; | |
4497 | ||
6b251fc9 AA |
4498 | /* Deliver the page fault to userland, check inside PT lock */ |
4499 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 | 4500 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
cb3184de | 4501 | folio_put(folio); |
82b0f8c3 | 4502 | return handle_userfault(vmf, VM_UFFD_MISSING); |
6b251fc9 AA |
4503 | } |
4504 | ||
19eaf449 RR |
4505 | folio_ref_add(folio, nr_pages - 1); |
4506 | add_mm_counter(vma->vm_mm, MM_ANONPAGES, nr_pages); | |
ec33687c BS |
4507 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4508 | count_mthp_stat(folio_order(folio), MTHP_STAT_ANON_FAULT_ALLOC); | |
4509 | #endif | |
19eaf449 | 4510 | folio_add_new_anon_rmap(folio, vma, addr); |
cb3184de | 4511 | folio_add_lru_vma(folio, vma); |
a13ea5b7 | 4512 | setpte: |
6ed31ba3 | 4513 | if (vmf_orig_pte_uffd_wp(vmf)) |
2bad466c | 4514 | entry = pte_mkuffd_wp(entry); |
19eaf449 | 4515 | set_ptes(vma->vm_mm, addr, vmf->pte, entry, nr_pages); |
1da177e4 LT |
4516 | |
4517 | /* No need to invalidate - it was non-present before */ | |
19eaf449 | 4518 | update_mmu_cache_range(vmf, vma, addr, vmf->pte, nr_pages); |
65500d23 | 4519 | unlock: |
3db82b93 HD |
4520 | if (vmf->pte) |
4521 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
6b31d595 | 4522 | return ret; |
8f4e2101 | 4523 | release: |
cb3184de | 4524 | folio_put(folio); |
8f4e2101 | 4525 | goto unlock; |
65500d23 | 4526 | oom: |
1da177e4 LT |
4527 | return VM_FAULT_OOM; |
4528 | } | |
4529 | ||
9a95f3cf | 4530 | /* |
c1e8d7c6 | 4531 | * The mmap_lock must have been held on entry, and may have been |
9a95f3cf PC |
4532 | * released depending on flags and vma->vm_ops->fault() return value. |
4533 | * See filemap_fault() and __lock_page_retry(). | |
4534 | */ | |
2b740303 | 4535 | static vm_fault_t __do_fault(struct vm_fault *vmf) |
7eae74af | 4536 | { |
82b0f8c3 | 4537 | struct vm_area_struct *vma = vmf->vma; |
01d1e0e6 | 4538 | struct folio *folio; |
2b740303 | 4539 | vm_fault_t ret; |
7eae74af | 4540 | |
63f3655f MH |
4541 | /* |
4542 | * Preallocate pte before we take page_lock because this might lead to | |
4543 | * deadlocks for memcg reclaim which waits for pages under writeback: | |
4544 | * lock_page(A) | |
4545 | * SetPageWriteback(A) | |
4546 | * unlock_page(A) | |
4547 | * lock_page(B) | |
4548 | * lock_page(B) | |
d383807a | 4549 | * pte_alloc_one |
63f3655f MH |
4550 | * shrink_page_list |
4551 | * wait_on_page_writeback(A) | |
4552 | * SetPageWriteback(B) | |
4553 | * unlock_page(B) | |
4554 | * # flush A, B to clear the writeback | |
4555 | */ | |
4556 | if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) { | |
a7069ee3 | 4557 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); |
63f3655f MH |
4558 | if (!vmf->prealloc_pte) |
4559 | return VM_FAULT_OOM; | |
63f3655f MH |
4560 | } |
4561 | ||
11bac800 | 4562 | ret = vma->vm_ops->fault(vmf); |
3917048d | 4563 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | |
b1aa812b | 4564 | VM_FAULT_DONE_COW))) |
bc2466e4 | 4565 | return ret; |
7eae74af | 4566 | |
01d1e0e6 | 4567 | folio = page_folio(vmf->page); |
667240e0 | 4568 | if (unlikely(PageHWPoison(vmf->page))) { |
e53ac737 RR |
4569 | vm_fault_t poisonret = VM_FAULT_HWPOISON; |
4570 | if (ret & VM_FAULT_LOCKED) { | |
01d1e0e6 MWO |
4571 | if (page_mapped(vmf->page)) |
4572 | unmap_mapping_folio(folio); | |
4573 | /* Retry if a clean folio was removed from the cache. */ | |
4574 | if (mapping_evict_folio(folio->mapping, folio)) | |
3149c79f | 4575 | poisonret = VM_FAULT_NOPAGE; |
01d1e0e6 | 4576 | folio_unlock(folio); |
e53ac737 | 4577 | } |
01d1e0e6 | 4578 | folio_put(folio); |
936ca80d | 4579 | vmf->page = NULL; |
e53ac737 | 4580 | return poisonret; |
7eae74af KS |
4581 | } |
4582 | ||
4583 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
01d1e0e6 | 4584 | folio_lock(folio); |
7eae74af | 4585 | else |
01d1e0e6 | 4586 | VM_BUG_ON_PAGE(!folio_test_locked(folio), vmf->page); |
7eae74af | 4587 | |
7eae74af KS |
4588 | return ret; |
4589 | } | |
4590 | ||
396bcc52 | 4591 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
82b0f8c3 | 4592 | static void deposit_prealloc_pte(struct vm_fault *vmf) |
953c66c2 | 4593 | { |
82b0f8c3 | 4594 | struct vm_area_struct *vma = vmf->vma; |
953c66c2 | 4595 | |
82b0f8c3 | 4596 | pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
953c66c2 AK |
4597 | /* |
4598 | * We are going to consume the prealloc table, | |
4599 | * count that as nr_ptes. | |
4600 | */ | |
c4812909 | 4601 | mm_inc_nr_ptes(vma->vm_mm); |
7f2b6ce8 | 4602 | vmf->prealloc_pte = NULL; |
953c66c2 AK |
4603 | } |
4604 | ||
f9ce0be7 | 4605 | vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 4606 | { |
ef37b2ea | 4607 | struct folio *folio = page_folio(page); |
82b0f8c3 JK |
4608 | struct vm_area_struct *vma = vmf->vma; |
4609 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
4610 | unsigned long haddr = vmf->address & HPAGE_PMD_MASK; | |
10102459 | 4611 | pmd_t entry; |
d01ac3c3 | 4612 | vm_fault_t ret = VM_FAULT_FALLBACK; |
10102459 | 4613 | |
3485b883 | 4614 | if (!thp_vma_suitable_order(vma, haddr, PMD_ORDER)) |
d01ac3c3 | 4615 | return ret; |
10102459 | 4616 | |
ef37b2ea | 4617 | if (page != &folio->page || folio_order(folio) != HPAGE_PMD_ORDER) |
d01ac3c3 | 4618 | return ret; |
10102459 | 4619 | |
eac96c3e YS |
4620 | /* |
4621 | * Just backoff if any subpage of a THP is corrupted otherwise | |
4622 | * the corrupted page may mapped by PMD silently to escape the | |
4623 | * check. This kind of THP just can be PTE mapped. Access to | |
4624 | * the corrupted subpage should trigger SIGBUS as expected. | |
4625 | */ | |
ef37b2ea | 4626 | if (unlikely(folio_test_has_hwpoisoned(folio))) |
eac96c3e YS |
4627 | return ret; |
4628 | ||
953c66c2 | 4629 | /* |
f0953a1b | 4630 | * Archs like ppc64 need additional space to store information |
953c66c2 AK |
4631 | * related to pte entry. Use the preallocated table for that. |
4632 | */ | |
82b0f8c3 | 4633 | if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { |
4cf58924 | 4634 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); |
82b0f8c3 | 4635 | if (!vmf->prealloc_pte) |
953c66c2 | 4636 | return VM_FAULT_OOM; |
953c66c2 AK |
4637 | } |
4638 | ||
82b0f8c3 JK |
4639 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); |
4640 | if (unlikely(!pmd_none(*vmf->pmd))) | |
10102459 KS |
4641 | goto out; |
4642 | ||
9f1f5b60 | 4643 | flush_icache_pages(vma, page, HPAGE_PMD_NR); |
10102459 KS |
4644 | |
4645 | entry = mk_huge_pmd(page, vma->vm_page_prot); | |
4646 | if (write) | |
f55e1014 | 4647 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); |
10102459 | 4648 | |
6b27cc6c | 4649 | add_mm_counter(vma->vm_mm, mm_counter_file(folio), HPAGE_PMD_NR); |
ef37b2ea | 4650 | folio_add_file_rmap_pmd(folio, page, vma); |
cea86fe2 | 4651 | |
953c66c2 AK |
4652 | /* |
4653 | * deposit and withdraw with pmd lock held | |
4654 | */ | |
4655 | if (arch_needs_pgtable_deposit()) | |
82b0f8c3 | 4656 | deposit_prealloc_pte(vmf); |
10102459 | 4657 | |
82b0f8c3 | 4658 | set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); |
10102459 | 4659 | |
82b0f8c3 | 4660 | update_mmu_cache_pmd(vma, haddr, vmf->pmd); |
10102459 KS |
4661 | |
4662 | /* fault is handled */ | |
4663 | ret = 0; | |
95ecedcd | 4664 | count_vm_event(THP_FILE_MAPPED); |
10102459 | 4665 | out: |
82b0f8c3 | 4666 | spin_unlock(vmf->ptl); |
10102459 KS |
4667 | return ret; |
4668 | } | |
4669 | #else | |
f9ce0be7 | 4670 | vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 4671 | { |
f9ce0be7 | 4672 | return VM_FAULT_FALLBACK; |
10102459 KS |
4673 | } |
4674 | #endif | |
4675 | ||
3bd786f7 YF |
4676 | /** |
4677 | * set_pte_range - Set a range of PTEs to point to pages in a folio. | |
4678 | * @vmf: Fault decription. | |
4679 | * @folio: The folio that contains @page. | |
4680 | * @page: The first page to create a PTE for. | |
4681 | * @nr: The number of PTEs to create. | |
4682 | * @addr: The first address to create a PTE for. | |
4683 | */ | |
4684 | void set_pte_range(struct vm_fault *vmf, struct folio *folio, | |
4685 | struct page *page, unsigned int nr, unsigned long addr) | |
3bb97794 | 4686 | { |
82b0f8c3 JK |
4687 | struct vm_area_struct *vma = vmf->vma; |
4688 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3bd786f7 | 4689 | bool prefault = in_range(vmf->address, addr, nr * PAGE_SIZE); |
3bb97794 | 4690 | pte_t entry; |
7267ec00 | 4691 | |
3bd786f7 | 4692 | flush_icache_pages(vma, page, nr); |
3bb97794 | 4693 | entry = mk_pte(page, vma->vm_page_prot); |
46bdb427 WD |
4694 | |
4695 | if (prefault && arch_wants_old_prefaulted_pte()) | |
4696 | entry = pte_mkold(entry); | |
50c25ee9 TB |
4697 | else |
4698 | entry = pte_sw_mkyoung(entry); | |
46bdb427 | 4699 | |
3bb97794 KS |
4700 | if (write) |
4701 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
6ed31ba3 | 4702 | if (unlikely(vmf_orig_pte_uffd_wp(vmf))) |
f1eb1bac | 4703 | entry = pte_mkuffd_wp(entry); |
bae473a4 KS |
4704 | /* copy-on-write page */ |
4705 | if (write && !(vma->vm_flags & VM_SHARED)) { | |
3bd786f7 YF |
4706 | VM_BUG_ON_FOLIO(nr != 1, folio); |
4707 | folio_add_new_anon_rmap(folio, vma, addr); | |
4708 | folio_add_lru_vma(folio, vma); | |
3bb97794 | 4709 | } else { |
68f03208 | 4710 | folio_add_file_rmap_ptes(folio, page, nr, vma); |
3bb97794 | 4711 | } |
3bd786f7 YF |
4712 | set_ptes(vma->vm_mm, addr, vmf->pte, entry, nr); |
4713 | ||
4714 | /* no need to invalidate: a not-present page won't be cached */ | |
4715 | update_mmu_cache_range(vmf, vma, addr, vmf->pte, nr); | |
3bb97794 KS |
4716 | } |
4717 | ||
f46f2ade PX |
4718 | static bool vmf_pte_changed(struct vm_fault *vmf) |
4719 | { | |
4720 | if (vmf->flags & FAULT_FLAG_ORIG_PTE_VALID) | |
c33c7948 | 4721 | return !pte_same(ptep_get(vmf->pte), vmf->orig_pte); |
f46f2ade | 4722 | |
c33c7948 | 4723 | return !pte_none(ptep_get(vmf->pte)); |
f46f2ade PX |
4724 | } |
4725 | ||
9118c0cb JK |
4726 | /** |
4727 | * finish_fault - finish page fault once we have prepared the page to fault | |
4728 | * | |
4729 | * @vmf: structure describing the fault | |
4730 | * | |
4731 | * This function handles all that is needed to finish a page fault once the | |
4732 | * page to fault in is prepared. It handles locking of PTEs, inserts PTE for | |
4733 | * given page, adds reverse page mapping, handles memcg charges and LRU | |
a862f68a | 4734 | * addition. |
9118c0cb JK |
4735 | * |
4736 | * The function expects the page to be locked and on success it consumes a | |
4737 | * reference of a page being mapped (for the PTE which maps it). | |
a862f68a MR |
4738 | * |
4739 | * Return: %0 on success, %VM_FAULT_ code in case of error. | |
9118c0cb | 4740 | */ |
2b740303 | 4741 | vm_fault_t finish_fault(struct vm_fault *vmf) |
9118c0cb | 4742 | { |
f9ce0be7 | 4743 | struct vm_area_struct *vma = vmf->vma; |
9118c0cb | 4744 | struct page *page; |
f9ce0be7 | 4745 | vm_fault_t ret; |
1f2d8b44 KW |
4746 | bool is_cow = (vmf->flags & FAULT_FLAG_WRITE) && |
4747 | !(vma->vm_flags & VM_SHARED); | |
9118c0cb JK |
4748 | |
4749 | /* Did we COW the page? */ | |
1f2d8b44 | 4750 | if (is_cow) |
9118c0cb JK |
4751 | page = vmf->cow_page; |
4752 | else | |
4753 | page = vmf->page; | |
6b31d595 MH |
4754 | |
4755 | /* | |
4756 | * check even for read faults because we might have lost our CoWed | |
4757 | * page | |
4758 | */ | |
f9ce0be7 KS |
4759 | if (!(vma->vm_flags & VM_SHARED)) { |
4760 | ret = check_stable_address_space(vma->vm_mm); | |
4761 | if (ret) | |
4762 | return ret; | |
4763 | } | |
4764 | ||
4765 | if (pmd_none(*vmf->pmd)) { | |
4766 | if (PageTransCompound(page)) { | |
4767 | ret = do_set_pmd(vmf, page); | |
4768 | if (ret != VM_FAULT_FALLBACK) | |
4769 | return ret; | |
4770 | } | |
4771 | ||
03c4f204 QZ |
4772 | if (vmf->prealloc_pte) |
4773 | pmd_install(vma->vm_mm, vmf->pmd, &vmf->prealloc_pte); | |
4774 | else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd))) | |
f9ce0be7 KS |
4775 | return VM_FAULT_OOM; |
4776 | } | |
4777 | ||
f9ce0be7 KS |
4778 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
4779 | vmf->address, &vmf->ptl); | |
3db82b93 HD |
4780 | if (!vmf->pte) |
4781 | return VM_FAULT_NOPAGE; | |
70427f6e | 4782 | |
f9ce0be7 | 4783 | /* Re-check under ptl */ |
70427f6e | 4784 | if (likely(!vmf_pte_changed(vmf))) { |
3bd786f7 | 4785 | struct folio *folio = page_folio(page); |
1f2d8b44 | 4786 | int type = is_cow ? MM_ANONPAGES : mm_counter_file(folio); |
70427f6e | 4787 | |
3bd786f7 | 4788 | set_pte_range(vmf, folio, page, 1, vmf->address); |
1f2d8b44 | 4789 | add_mm_counter(vma->vm_mm, type, 1); |
70427f6e SA |
4790 | ret = 0; |
4791 | } else { | |
4792 | update_mmu_tlb(vma, vmf->address, vmf->pte); | |
f9ce0be7 | 4793 | ret = VM_FAULT_NOPAGE; |
70427f6e | 4794 | } |
f9ce0be7 | 4795 | |
f9ce0be7 | 4796 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
9118c0cb JK |
4797 | return ret; |
4798 | } | |
4799 | ||
53d36a56 LS |
4800 | static unsigned long fault_around_pages __read_mostly = |
4801 | 65536 >> PAGE_SHIFT; | |
a9b0f861 | 4802 | |
a9b0f861 KS |
4803 | #ifdef CONFIG_DEBUG_FS |
4804 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 4805 | { |
53d36a56 | 4806 | *val = fault_around_pages << PAGE_SHIFT; |
1592eef0 KS |
4807 | return 0; |
4808 | } | |
4809 | ||
b4903d6e | 4810 | /* |
da391d64 WK |
4811 | * fault_around_bytes must be rounded down to the nearest page order as it's |
4812 | * what do_fault_around() expects to see. | |
b4903d6e | 4813 | */ |
a9b0f861 | 4814 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 4815 | { |
a9b0f861 | 4816 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 4817 | return -EINVAL; |
53d36a56 LS |
4818 | |
4819 | /* | |
4820 | * The minimum value is 1 page, however this results in no fault-around | |
4821 | * at all. See should_fault_around(). | |
4822 | */ | |
5aa598a7 KW |
4823 | val = max(val, PAGE_SIZE); |
4824 | fault_around_pages = rounddown_pow_of_two(val) >> PAGE_SHIFT; | |
53d36a56 | 4825 | |
1592eef0 KS |
4826 | return 0; |
4827 | } | |
0a1345f8 | 4828 | DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, |
a9b0f861 | 4829 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
1592eef0 KS |
4830 | |
4831 | static int __init fault_around_debugfs(void) | |
4832 | { | |
d9f7979c GKH |
4833 | debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, |
4834 | &fault_around_bytes_fops); | |
1592eef0 KS |
4835 | return 0; |
4836 | } | |
4837 | late_initcall(fault_around_debugfs); | |
1592eef0 | 4838 | #endif |
8c6e50b0 | 4839 | |
1fdb412b KS |
4840 | /* |
4841 | * do_fault_around() tries to map few pages around the fault address. The hope | |
4842 | * is that the pages will be needed soon and this will lower the number of | |
4843 | * faults to handle. | |
4844 | * | |
4845 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
4846 | * not ready to be mapped: not up-to-date, locked, etc. | |
4847 | * | |
9042599e LS |
4848 | * This function doesn't cross VMA or page table boundaries, in order to call |
4849 | * map_pages() and acquire a PTE lock only once. | |
1fdb412b | 4850 | * |
53d36a56 | 4851 | * fault_around_pages defines how many pages we'll try to map. |
da391d64 WK |
4852 | * do_fault_around() expects it to be set to a power of two less than or equal |
4853 | * to PTRS_PER_PTE. | |
1fdb412b | 4854 | * |
da391d64 | 4855 | * The virtual address of the area that we map is naturally aligned to |
53d36a56 | 4856 | * fault_around_pages * PAGE_SIZE rounded down to the machine page size |
da391d64 WK |
4857 | * (and therefore to page order). This way it's easier to guarantee |
4858 | * that we don't cross page table boundaries. | |
1fdb412b | 4859 | */ |
2b740303 | 4860 | static vm_fault_t do_fault_around(struct vm_fault *vmf) |
8c6e50b0 | 4861 | { |
53d36a56 | 4862 | pgoff_t nr_pages = READ_ONCE(fault_around_pages); |
9042599e LS |
4863 | pgoff_t pte_off = pte_index(vmf->address); |
4864 | /* The page offset of vmf->address within the VMA. */ | |
4865 | pgoff_t vma_off = vmf->pgoff - vmf->vma->vm_pgoff; | |
4866 | pgoff_t from_pte, to_pte; | |
58ef47ef | 4867 | vm_fault_t ret; |
8c6e50b0 | 4868 | |
9042599e LS |
4869 | /* The PTE offset of the start address, clamped to the VMA. */ |
4870 | from_pte = max(ALIGN_DOWN(pte_off, nr_pages), | |
4871 | pte_off - min(pte_off, vma_off)); | |
aecd6f44 | 4872 | |
9042599e LS |
4873 | /* The PTE offset of the end address, clamped to the VMA and PTE. */ |
4874 | to_pte = min3(from_pte + nr_pages, (pgoff_t)PTRS_PER_PTE, | |
4875 | pte_off + vma_pages(vmf->vma) - vma_off) - 1; | |
8c6e50b0 | 4876 | |
82b0f8c3 | 4877 | if (pmd_none(*vmf->pmd)) { |
4cf58924 | 4878 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm); |
82b0f8c3 | 4879 | if (!vmf->prealloc_pte) |
f9ce0be7 | 4880 | return VM_FAULT_OOM; |
8c6e50b0 KS |
4881 | } |
4882 | ||
58ef47ef MWO |
4883 | rcu_read_lock(); |
4884 | ret = vmf->vma->vm_ops->map_pages(vmf, | |
4885 | vmf->pgoff + from_pte - pte_off, | |
4886 | vmf->pgoff + to_pte - pte_off); | |
4887 | rcu_read_unlock(); | |
4888 | ||
4889 | return ret; | |
8c6e50b0 KS |
4890 | } |
4891 | ||
9c28a205 PX |
4892 | /* Return true if we should do read fault-around, false otherwise */ |
4893 | static inline bool should_fault_around(struct vm_fault *vmf) | |
4894 | { | |
4895 | /* No ->map_pages? No way to fault around... */ | |
4896 | if (!vmf->vma->vm_ops->map_pages) | |
4897 | return false; | |
4898 | ||
4899 | if (uffd_disable_fault_around(vmf->vma)) | |
4900 | return false; | |
4901 | ||
53d36a56 LS |
4902 | /* A single page implies no faulting 'around' at all. */ |
4903 | return fault_around_pages > 1; | |
9c28a205 PX |
4904 | } |
4905 | ||
2b740303 | 4906 | static vm_fault_t do_read_fault(struct vm_fault *vmf) |
e655fb29 | 4907 | { |
2b740303 | 4908 | vm_fault_t ret = 0; |
22d1e68f | 4909 | struct folio *folio; |
8c6e50b0 KS |
4910 | |
4911 | /* | |
4912 | * Let's call ->map_pages() first and use ->fault() as fallback | |
4913 | * if page by the offset is not ready to be mapped (cold cache or | |
4914 | * something). | |
4915 | */ | |
9c28a205 PX |
4916 | if (should_fault_around(vmf)) { |
4917 | ret = do_fault_around(vmf); | |
4918 | if (ret) | |
4919 | return ret; | |
8c6e50b0 | 4920 | } |
e655fb29 | 4921 | |
12214eba MWO |
4922 | ret = vmf_can_call_fault(vmf); |
4923 | if (ret) | |
4924 | return ret; | |
f5617ffe | 4925 | |
936ca80d | 4926 | ret = __do_fault(vmf); |
e655fb29 KS |
4927 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
4928 | return ret; | |
4929 | ||
9118c0cb | 4930 | ret |= finish_fault(vmf); |
22d1e68f SK |
4931 | folio = page_folio(vmf->page); |
4932 | folio_unlock(folio); | |
7267ec00 | 4933 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
22d1e68f | 4934 | folio_put(folio); |
e655fb29 KS |
4935 | return ret; |
4936 | } | |
4937 | ||
2b740303 | 4938 | static vm_fault_t do_cow_fault(struct vm_fault *vmf) |
ec47c3b9 | 4939 | { |
82b0f8c3 | 4940 | struct vm_area_struct *vma = vmf->vma; |
e4621e70 | 4941 | struct folio *folio; |
2b740303 | 4942 | vm_fault_t ret; |
ec47c3b9 | 4943 | |
4de8c93a MWO |
4944 | ret = vmf_can_call_fault(vmf); |
4945 | if (!ret) | |
4946 | ret = vmf_anon_prepare(vmf); | |
4947 | if (ret) | |
4948 | return ret; | |
ec47c3b9 | 4949 | |
e4621e70 KW |
4950 | folio = folio_prealloc(vma->vm_mm, vma, vmf->address, false); |
4951 | if (!folio) | |
ec47c3b9 KS |
4952 | return VM_FAULT_OOM; |
4953 | ||
e4621e70 | 4954 | vmf->cow_page = &folio->page; |
ec47c3b9 | 4955 | |
936ca80d | 4956 | ret = __do_fault(vmf); |
ec47c3b9 KS |
4957 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
4958 | goto uncharge_out; | |
3917048d JK |
4959 | if (ret & VM_FAULT_DONE_COW) |
4960 | return ret; | |
ec47c3b9 | 4961 | |
b1aa812b | 4962 | copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); |
e4621e70 | 4963 | __folio_mark_uptodate(folio); |
ec47c3b9 | 4964 | |
9118c0cb | 4965 | ret |= finish_fault(vmf); |
b1aa812b JK |
4966 | unlock_page(vmf->page); |
4967 | put_page(vmf->page); | |
7267ec00 KS |
4968 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
4969 | goto uncharge_out; | |
ec47c3b9 KS |
4970 | return ret; |
4971 | uncharge_out: | |
e4621e70 | 4972 | folio_put(folio); |
ec47c3b9 KS |
4973 | return ret; |
4974 | } | |
4975 | ||
2b740303 | 4976 | static vm_fault_t do_shared_fault(struct vm_fault *vmf) |
1da177e4 | 4977 | { |
82b0f8c3 | 4978 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 4979 | vm_fault_t ret, tmp; |
6f609b7e | 4980 | struct folio *folio; |
1d65f86d | 4981 | |
4ed43798 MWO |
4982 | ret = vmf_can_call_fault(vmf); |
4983 | if (ret) | |
4984 | return ret; | |
1d65f86d | 4985 | |
936ca80d | 4986 | ret = __do_fault(vmf); |
7eae74af | 4987 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 4988 | return ret; |
1da177e4 | 4989 | |
6f609b7e SK |
4990 | folio = page_folio(vmf->page); |
4991 | ||
1da177e4 | 4992 | /* |
f0c6d4d2 KS |
4993 | * Check if the backing address space wants to know that the page is |
4994 | * about to become writable | |
1da177e4 | 4995 | */ |
fb09a464 | 4996 | if (vma->vm_ops->page_mkwrite) { |
6f609b7e | 4997 | folio_unlock(folio); |
86aa6998 | 4998 | tmp = do_page_mkwrite(vmf, folio); |
fb09a464 KS |
4999 | if (unlikely(!tmp || |
5000 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
6f609b7e | 5001 | folio_put(folio); |
fb09a464 | 5002 | return tmp; |
4294621f | 5003 | } |
fb09a464 KS |
5004 | } |
5005 | ||
9118c0cb | 5006 | ret |= finish_fault(vmf); |
7267ec00 KS |
5007 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
5008 | VM_FAULT_RETRY))) { | |
6f609b7e SK |
5009 | folio_unlock(folio); |
5010 | folio_put(folio); | |
f0c6d4d2 | 5011 | return ret; |
1da177e4 | 5012 | } |
b827e496 | 5013 | |
89b15332 | 5014 | ret |= fault_dirty_shared_page(vmf); |
1d65f86d | 5015 | return ret; |
54cb8821 | 5016 | } |
d00806b1 | 5017 | |
9a95f3cf | 5018 | /* |
c1e8d7c6 | 5019 | * We enter with non-exclusive mmap_lock (to exclude vma changes, |
9a95f3cf | 5020 | * but allow concurrent faults). |
c1e8d7c6 | 5021 | * The mmap_lock may have been released depending on flags and our |
9138e47e | 5022 | * return value. See filemap_fault() and __folio_lock_or_retry(). |
c1e8d7c6 | 5023 | * If mmap_lock is released, vma may become invalid (for example |
fc8efd2d | 5024 | * by other thread calling munmap()). |
9a95f3cf | 5025 | */ |
2b740303 | 5026 | static vm_fault_t do_fault(struct vm_fault *vmf) |
54cb8821 | 5027 | { |
82b0f8c3 | 5028 | struct vm_area_struct *vma = vmf->vma; |
fc8efd2d | 5029 | struct mm_struct *vm_mm = vma->vm_mm; |
2b740303 | 5030 | vm_fault_t ret; |
54cb8821 | 5031 | |
ff09d7ec AK |
5032 | /* |
5033 | * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND | |
5034 | */ | |
5035 | if (!vma->vm_ops->fault) { | |
3db82b93 HD |
5036 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, |
5037 | vmf->address, &vmf->ptl); | |
5038 | if (unlikely(!vmf->pte)) | |
ff09d7ec AK |
5039 | ret = VM_FAULT_SIGBUS; |
5040 | else { | |
ff09d7ec AK |
5041 | /* |
5042 | * Make sure this is not a temporary clearing of pte | |
5043 | * by holding ptl and checking again. A R/M/W update | |
5044 | * of pte involves: take ptl, clearing the pte so that | |
5045 | * we don't have concurrent modification by hardware | |
5046 | * followed by an update. | |
5047 | */ | |
c33c7948 | 5048 | if (unlikely(pte_none(ptep_get(vmf->pte)))) |
ff09d7ec AK |
5049 | ret = VM_FAULT_SIGBUS; |
5050 | else | |
5051 | ret = VM_FAULT_NOPAGE; | |
5052 | ||
5053 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
5054 | } | |
5055 | } else if (!(vmf->flags & FAULT_FLAG_WRITE)) | |
b0b9b3df HD |
5056 | ret = do_read_fault(vmf); |
5057 | else if (!(vma->vm_flags & VM_SHARED)) | |
5058 | ret = do_cow_fault(vmf); | |
5059 | else | |
5060 | ret = do_shared_fault(vmf); | |
5061 | ||
5062 | /* preallocated pagetable is unused: free it */ | |
5063 | if (vmf->prealloc_pte) { | |
fc8efd2d | 5064 | pte_free(vm_mm, vmf->prealloc_pte); |
7f2b6ce8 | 5065 | vmf->prealloc_pte = NULL; |
b0b9b3df HD |
5066 | } |
5067 | return ret; | |
54cb8821 NP |
5068 | } |
5069 | ||
f8fd525b | 5070 | int numa_migrate_prep(struct folio *folio, struct vm_fault *vmf, |
f4c0d836 | 5071 | unsigned long addr, int page_nid, int *flags) |
9532fec1 | 5072 | { |
f8fd525b DT |
5073 | struct vm_area_struct *vma = vmf->vma; |
5074 | ||
cda6d936 | 5075 | folio_get(folio); |
9532fec1 | 5076 | |
fc137c0d R |
5077 | /* Record the current PID acceesing VMA */ |
5078 | vma_set_access_pid_bit(vma); | |
5079 | ||
9532fec1 | 5080 | count_vm_numa_event(NUMA_HINT_FAULTS); |
04bb2f94 | 5081 | if (page_nid == numa_node_id()) { |
9532fec1 | 5082 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
5083 | *flags |= TNF_FAULT_LOCAL; |
5084 | } | |
9532fec1 | 5085 | |
f8fd525b | 5086 | return mpol_misplaced(folio, vmf, addr); |
9532fec1 MG |
5087 | } |
5088 | ||
6b0ed7b3 | 5089 | static void numa_rebuild_single_mapping(struct vm_fault *vmf, struct vm_area_struct *vma, |
d2136d74 | 5090 | unsigned long fault_addr, pte_t *fault_pte, |
6b0ed7b3 BW |
5091 | bool writable) |
5092 | { | |
5093 | pte_t pte, old_pte; | |
5094 | ||
d2136d74 | 5095 | old_pte = ptep_modify_prot_start(vma, fault_addr, fault_pte); |
6b0ed7b3 BW |
5096 | pte = pte_modify(old_pte, vma->vm_page_prot); |
5097 | pte = pte_mkyoung(pte); | |
5098 | if (writable) | |
5099 | pte = pte_mkwrite(pte, vma); | |
d2136d74 BW |
5100 | ptep_modify_prot_commit(vma, fault_addr, fault_pte, old_pte, pte); |
5101 | update_mmu_cache_range(vmf, vma, fault_addr, fault_pte, 1); | |
5102 | } | |
5103 | ||
5104 | static void numa_rebuild_large_mapping(struct vm_fault *vmf, struct vm_area_struct *vma, | |
5105 | struct folio *folio, pte_t fault_pte, | |
5106 | bool ignore_writable, bool pte_write_upgrade) | |
5107 | { | |
5108 | int nr = pte_pfn(fault_pte) - folio_pfn(folio); | |
5109 | unsigned long start = max(vmf->address - nr * PAGE_SIZE, vma->vm_start); | |
5110 | unsigned long end = min(vmf->address + (folio_nr_pages(folio) - nr) * PAGE_SIZE, vma->vm_end); | |
5111 | pte_t *start_ptep = vmf->pte - (vmf->address - start) / PAGE_SIZE; | |
5112 | unsigned long addr; | |
5113 | ||
5114 | /* Restore all PTEs' mapping of the large folio */ | |
5115 | for (addr = start; addr != end; start_ptep++, addr += PAGE_SIZE) { | |
5116 | pte_t ptent = ptep_get(start_ptep); | |
5117 | bool writable = false; | |
5118 | ||
5119 | if (!pte_present(ptent) || !pte_protnone(ptent)) | |
5120 | continue; | |
5121 | ||
5122 | if (pfn_folio(pte_pfn(ptent)) != folio) | |
5123 | continue; | |
5124 | ||
5125 | if (!ignore_writable) { | |
5126 | ptent = pte_modify(ptent, vma->vm_page_prot); | |
5127 | writable = pte_write(ptent); | |
5128 | if (!writable && pte_write_upgrade && | |
5129 | can_change_pte_writable(vma, addr, ptent)) | |
5130 | writable = true; | |
5131 | } | |
5132 | ||
5133 | numa_rebuild_single_mapping(vmf, vma, addr, start_ptep, writable); | |
5134 | } | |
9532fec1 MG |
5135 | } |
5136 | ||
2b740303 | 5137 | static vm_fault_t do_numa_page(struct vm_fault *vmf) |
d10e63f2 | 5138 | { |
82b0f8c3 | 5139 | struct vm_area_struct *vma = vmf->vma; |
6695cf68 KW |
5140 | struct folio *folio = NULL; |
5141 | int nid = NUMA_NO_NODE; | |
d2136d74 BW |
5142 | bool writable = false, ignore_writable = false; |
5143 | bool pte_write_upgrade = vma_wants_manual_pte_write_upgrade(vma); | |
90572890 | 5144 | int last_cpupid; |
cbee9f88 | 5145 | int target_nid; |
04a86453 | 5146 | pte_t pte, old_pte; |
d2136d74 | 5147 | int flags = 0, nr_pages; |
d10e63f2 MG |
5148 | |
5149 | /* | |
6c1b748e JH |
5150 | * The pte cannot be used safely until we verify, while holding the page |
5151 | * table lock, that its contents have not changed during fault handling. | |
166f61b9 | 5152 | */ |
82b0f8c3 | 5153 | spin_lock(vmf->ptl); |
6c1b748e JH |
5154 | /* Read the live PTE from the page tables: */ |
5155 | old_pte = ptep_get(vmf->pte); | |
5156 | ||
5157 | if (unlikely(!pte_same(old_pte, vmf->orig_pte))) { | |
82b0f8c3 | 5158 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
5159 | goto out; |
5160 | } | |
5161 | ||
04a86453 | 5162 | pte = pte_modify(old_pte, vma->vm_page_prot); |
d10e63f2 | 5163 | |
6a56ccbc DH |
5164 | /* |
5165 | * Detect now whether the PTE could be writable; this information | |
5166 | * is only valid while holding the PT lock. | |
5167 | */ | |
5168 | writable = pte_write(pte); | |
d2136d74 | 5169 | if (!writable && pte_write_upgrade && |
6a56ccbc DH |
5170 | can_change_pte_writable(vma, vmf->address, pte)) |
5171 | writable = true; | |
5172 | ||
6695cf68 KW |
5173 | folio = vm_normal_folio(vma, vmf->address, pte); |
5174 | if (!folio || folio_is_zone_device(folio)) | |
b99a342d | 5175 | goto out_map; |
d10e63f2 | 5176 | |
6688cc05 | 5177 | /* |
bea66fbd MG |
5178 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
5179 | * much anyway since they can be in shared cache state. This misses | |
5180 | * the case where a mapping is writable but the process never writes | |
5181 | * to it but pte_write gets cleared during protection updates and | |
5182 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
5183 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 5184 | */ |
6a56ccbc | 5185 | if (!writable) |
6688cc05 PZ |
5186 | flags |= TNF_NO_GROUP; |
5187 | ||
dabe1d99 | 5188 | /* |
6695cf68 | 5189 | * Flag if the folio is shared between multiple address spaces. This |
dabe1d99 RR |
5190 | * is later used when determining whether to group tasks together |
5191 | */ | |
ebb34f78 | 5192 | if (folio_likely_mapped_shared(folio) && (vma->vm_flags & VM_SHARED)) |
dabe1d99 RR |
5193 | flags |= TNF_SHARED; |
5194 | ||
6695cf68 | 5195 | nid = folio_nid(folio); |
d2136d74 | 5196 | nr_pages = folio_nr_pages(folio); |
33024536 HY |
5197 | /* |
5198 | * For memory tiering mode, cpupid of slow memory page is used | |
5199 | * to record page access time. So use default value. | |
5200 | */ | |
5201 | if ((sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) && | |
6695cf68 | 5202 | !node_is_toptier(nid)) |
33024536 HY |
5203 | last_cpupid = (-1 & LAST_CPUPID_MASK); |
5204 | else | |
67b33e3f | 5205 | last_cpupid = folio_last_cpupid(folio); |
f8fd525b | 5206 | target_nid = numa_migrate_prep(folio, vmf, vmf->address, nid, &flags); |
98fa15f3 | 5207 | if (target_nid == NUMA_NO_NODE) { |
6695cf68 | 5208 | folio_put(folio); |
b99a342d | 5209 | goto out_map; |
4daae3b4 | 5210 | } |
b99a342d | 5211 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
6a56ccbc | 5212 | writable = false; |
d2136d74 | 5213 | ignore_writable = true; |
4daae3b4 MG |
5214 | |
5215 | /* Migrate to the requested node */ | |
6695cf68 KW |
5216 | if (migrate_misplaced_folio(folio, vma, target_nid)) { |
5217 | nid = target_nid; | |
6688cc05 | 5218 | flags |= TNF_MIGRATED; |
b99a342d | 5219 | } else { |
074c2381 | 5220 | flags |= TNF_MIGRATE_FAIL; |
c7ad0880 HD |
5221 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
5222 | vmf->address, &vmf->ptl); | |
5223 | if (unlikely(!vmf->pte)) | |
5224 | goto out; | |
c33c7948 | 5225 | if (unlikely(!pte_same(ptep_get(vmf->pte), vmf->orig_pte))) { |
b99a342d HY |
5226 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
5227 | goto out; | |
5228 | } | |
5229 | goto out_map; | |
5230 | } | |
4daae3b4 MG |
5231 | |
5232 | out: | |
6695cf68 | 5233 | if (nid != NUMA_NO_NODE) |
d2136d74 | 5234 | task_numa_fault(last_cpupid, nid, nr_pages, flags); |
d10e63f2 | 5235 | return 0; |
b99a342d HY |
5236 | out_map: |
5237 | /* | |
5238 | * Make it present again, depending on how arch implements | |
5239 | * non-accessible ptes, some can allow access by kernel mode. | |
5240 | */ | |
d2136d74 BW |
5241 | if (folio && folio_test_large(folio)) |
5242 | numa_rebuild_large_mapping(vmf, vma, folio, pte, ignore_writable, | |
5243 | pte_write_upgrade); | |
5244 | else | |
5245 | numa_rebuild_single_mapping(vmf, vma, vmf->address, vmf->pte, | |
5246 | writable); | |
b99a342d HY |
5247 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
5248 | goto out; | |
d10e63f2 MG |
5249 | } |
5250 | ||
2b740303 | 5251 | static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 5252 | { |
8f5fd0e1 MWO |
5253 | struct vm_area_struct *vma = vmf->vma; |
5254 | if (vma_is_anonymous(vma)) | |
82b0f8c3 | 5255 | return do_huge_pmd_anonymous_page(vmf); |
40d49a3c | 5256 | if (vma->vm_ops->huge_fault) |
1d024e7a | 5257 | return vma->vm_ops->huge_fault(vmf, PMD_ORDER); |
b96375f7 MW |
5258 | return VM_FAULT_FALLBACK; |
5259 | } | |
5260 | ||
183f24aa | 5261 | /* `inline' is required to avoid gcc 4.1.2 build error */ |
5db4f15c | 5262 | static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 5263 | { |
8f5fd0e1 | 5264 | struct vm_area_struct *vma = vmf->vma; |
c89357e2 | 5265 | const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; |
aea06577 | 5266 | vm_fault_t ret; |
c89357e2 | 5267 | |
8f5fd0e1 | 5268 | if (vma_is_anonymous(vma)) { |
c89357e2 | 5269 | if (likely(!unshare) && |
d61ea1cb PX |
5270 | userfaultfd_huge_pmd_wp(vma, vmf->orig_pmd)) { |
5271 | if (userfaultfd_wp_async(vmf->vma)) | |
5272 | goto split; | |
529b930b | 5273 | return handle_userfault(vmf, VM_UFFD_WP); |
d61ea1cb | 5274 | } |
5db4f15c | 5275 | return do_huge_pmd_wp_page(vmf); |
529b930b | 5276 | } |
327e9fd4 | 5277 | |
8f5fd0e1 MWO |
5278 | if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { |
5279 | if (vma->vm_ops->huge_fault) { | |
1d024e7a | 5280 | ret = vma->vm_ops->huge_fault(vmf, PMD_ORDER); |
aea06577 DH |
5281 | if (!(ret & VM_FAULT_FALLBACK)) |
5282 | return ret; | |
5283 | } | |
327e9fd4 | 5284 | } |
af9e4d5f | 5285 | |
d61ea1cb | 5286 | split: |
327e9fd4 | 5287 | /* COW or write-notify handled on pte level: split pmd. */ |
8f5fd0e1 | 5288 | __split_huge_pmd(vma, vmf->pmd, vmf->address, false, NULL); |
af9e4d5f | 5289 | |
b96375f7 MW |
5290 | return VM_FAULT_FALLBACK; |
5291 | } | |
5292 | ||
2b740303 | 5293 | static vm_fault_t create_huge_pud(struct vm_fault *vmf) |
a00cc7d9 | 5294 | { |
14c99d65 GJ |
5295 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ |
5296 | defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) | |
c4fd825e | 5297 | struct vm_area_struct *vma = vmf->vma; |
14c99d65 | 5298 | /* No support for anonymous transparent PUD pages yet */ |
c4fd825e | 5299 | if (vma_is_anonymous(vma)) |
14c99d65 | 5300 | return VM_FAULT_FALLBACK; |
40d49a3c | 5301 | if (vma->vm_ops->huge_fault) |
1d024e7a | 5302 | return vma->vm_ops->huge_fault(vmf, PUD_ORDER); |
14c99d65 GJ |
5303 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
5304 | return VM_FAULT_FALLBACK; | |
5305 | } | |
5306 | ||
5307 | static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) | |
5308 | { | |
327e9fd4 THV |
5309 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ |
5310 | defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) | |
c4fd825e | 5311 | struct vm_area_struct *vma = vmf->vma; |
aea06577 DH |
5312 | vm_fault_t ret; |
5313 | ||
a00cc7d9 | 5314 | /* No support for anonymous transparent PUD pages yet */ |
c4fd825e | 5315 | if (vma_is_anonymous(vma)) |
327e9fd4 | 5316 | goto split; |
c4fd825e MWO |
5317 | if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) { |
5318 | if (vma->vm_ops->huge_fault) { | |
1d024e7a | 5319 | ret = vma->vm_ops->huge_fault(vmf, PUD_ORDER); |
aea06577 DH |
5320 | if (!(ret & VM_FAULT_FALLBACK)) |
5321 | return ret; | |
5322 | } | |
327e9fd4 THV |
5323 | } |
5324 | split: | |
5325 | /* COW or write-notify not handled on PUD level: split pud.*/ | |
c4fd825e | 5326 | __split_huge_pud(vma, vmf->pud, vmf->address); |
14c99d65 | 5327 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */ |
a00cc7d9 MW |
5328 | return VM_FAULT_FALLBACK; |
5329 | } | |
5330 | ||
1da177e4 LT |
5331 | /* |
5332 | * These routines also need to handle stuff like marking pages dirty | |
5333 | * and/or accessed for architectures that don't do it in hardware (most | |
5334 | * RISC architectures). The early dirtying is also good on the i386. | |
5335 | * | |
5336 | * There is also a hook called "update_mmu_cache()" that architectures | |
5337 | * with external mmu caches can use to update those (ie the Sparc or | |
5338 | * PowerPC hashed page tables that act as extended TLBs). | |
5339 | * | |
c1e8d7c6 | 5340 | * We enter with non-exclusive mmap_lock (to exclude vma changes, but allow |
7267ec00 | 5341 | * concurrent faults). |
9a95f3cf | 5342 | * |
c1e8d7c6 | 5343 | * The mmap_lock may have been released depending on flags and our return value. |
9138e47e | 5344 | * See filemap_fault() and __folio_lock_or_retry(). |
1da177e4 | 5345 | */ |
2b740303 | 5346 | static vm_fault_t handle_pte_fault(struct vm_fault *vmf) |
1da177e4 LT |
5347 | { |
5348 | pte_t entry; | |
5349 | ||
82b0f8c3 | 5350 | if (unlikely(pmd_none(*vmf->pmd))) { |
7267ec00 KS |
5351 | /* |
5352 | * Leave __pte_alloc() until later: because vm_ops->fault may | |
5353 | * want to allocate huge page, and if we expose page table | |
5354 | * for an instant, it will be difficult to retract from | |
5355 | * concurrent faults and from rmap lookups. | |
5356 | */ | |
82b0f8c3 | 5357 | vmf->pte = NULL; |
f46f2ade | 5358 | vmf->flags &= ~FAULT_FLAG_ORIG_PTE_VALID; |
7267ec00 | 5359 | } else { |
7267ec00 KS |
5360 | /* |
5361 | * A regular pmd is established and it can't morph into a huge | |
c7ad0880 HD |
5362 | * pmd by anon khugepaged, since that takes mmap_lock in write |
5363 | * mode; but shmem or file collapse to THP could still morph | |
5364 | * it into a huge pmd: just retry later if so. | |
7267ec00 | 5365 | */ |
c7ad0880 HD |
5366 | vmf->pte = pte_offset_map_nolock(vmf->vma->vm_mm, vmf->pmd, |
5367 | vmf->address, &vmf->ptl); | |
5368 | if (unlikely(!vmf->pte)) | |
5369 | return 0; | |
26e1a0c3 | 5370 | vmf->orig_pte = ptep_get_lockless(vmf->pte); |
f46f2ade | 5371 | vmf->flags |= FAULT_FLAG_ORIG_PTE_VALID; |
7267ec00 | 5372 | |
2994302b | 5373 | if (pte_none(vmf->orig_pte)) { |
82b0f8c3 JK |
5374 | pte_unmap(vmf->pte); |
5375 | vmf->pte = NULL; | |
65500d23 | 5376 | } |
1da177e4 LT |
5377 | } |
5378 | ||
2bad466c PX |
5379 | if (!vmf->pte) |
5380 | return do_pte_missing(vmf); | |
7267ec00 | 5381 | |
2994302b JK |
5382 | if (!pte_present(vmf->orig_pte)) |
5383 | return do_swap_page(vmf); | |
7267ec00 | 5384 | |
2994302b JK |
5385 | if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) |
5386 | return do_numa_page(vmf); | |
d10e63f2 | 5387 | |
82b0f8c3 | 5388 | spin_lock(vmf->ptl); |
2994302b | 5389 | entry = vmf->orig_pte; |
c33c7948 | 5390 | if (unlikely(!pte_same(ptep_get(vmf->pte), entry))) { |
7df67697 | 5391 | update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); |
8f4e2101 | 5392 | goto unlock; |
7df67697 | 5393 | } |
c89357e2 | 5394 | if (vmf->flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) { |
f6f37321 | 5395 | if (!pte_write(entry)) |
2994302b | 5396 | return do_wp_page(vmf); |
c89357e2 DH |
5397 | else if (likely(vmf->flags & FAULT_FLAG_WRITE)) |
5398 | entry = pte_mkdirty(entry); | |
1da177e4 LT |
5399 | } |
5400 | entry = pte_mkyoung(entry); | |
82b0f8c3 JK |
5401 | if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, |
5402 | vmf->flags & FAULT_FLAG_WRITE)) { | |
5003a2bd MWO |
5403 | update_mmu_cache_range(vmf, vmf->vma, vmf->address, |
5404 | vmf->pte, 1); | |
1a44e149 | 5405 | } else { |
b7333b58 YS |
5406 | /* Skip spurious TLB flush for retried page fault */ |
5407 | if (vmf->flags & FAULT_FLAG_TRIED) | |
5408 | goto unlock; | |
1a44e149 AA |
5409 | /* |
5410 | * This is needed only for protection faults but the arch code | |
5411 | * is not yet telling us if this is a protection fault or not. | |
5412 | * This still avoids useless tlb flushes for .text page faults | |
5413 | * with threads. | |
5414 | */ | |
82b0f8c3 | 5415 | if (vmf->flags & FAULT_FLAG_WRITE) |
99c29133 GS |
5416 | flush_tlb_fix_spurious_fault(vmf->vma, vmf->address, |
5417 | vmf->pte); | |
1a44e149 | 5418 | } |
8f4e2101 | 5419 | unlock: |
82b0f8c3 | 5420 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 5421 | return 0; |
1da177e4 LT |
5422 | } |
5423 | ||
5424 | /* | |
4ec31152 MWO |
5425 | * On entry, we hold either the VMA lock or the mmap_lock |
5426 | * (FAULT_FLAG_VMA_LOCK tells you which). If VM_FAULT_RETRY is set in | |
5427 | * the result, the mmap_lock is not held on exit. See filemap_fault() | |
5428 | * and __folio_lock_or_retry(). | |
1da177e4 | 5429 | */ |
2b740303 SJ |
5430 | static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma, |
5431 | unsigned long address, unsigned int flags) | |
1da177e4 | 5432 | { |
82b0f8c3 | 5433 | struct vm_fault vmf = { |
bae473a4 | 5434 | .vma = vma, |
1a29d85e | 5435 | .address = address & PAGE_MASK, |
824ddc60 | 5436 | .real_address = address, |
bae473a4 | 5437 | .flags = flags, |
0721ec8b | 5438 | .pgoff = linear_page_index(vma, address), |
667240e0 | 5439 | .gfp_mask = __get_fault_gfp_mask(vma), |
bae473a4 | 5440 | }; |
dcddffd4 | 5441 | struct mm_struct *mm = vma->vm_mm; |
7da4e2cb | 5442 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 | 5443 | pgd_t *pgd; |
c2febafc | 5444 | p4d_t *p4d; |
2b740303 | 5445 | vm_fault_t ret; |
1da177e4 | 5446 | |
1da177e4 | 5447 | pgd = pgd_offset(mm, address); |
c2febafc KS |
5448 | p4d = p4d_alloc(mm, pgd, address); |
5449 | if (!p4d) | |
5450 | return VM_FAULT_OOM; | |
a00cc7d9 | 5451 | |
c2febafc | 5452 | vmf.pud = pud_alloc(mm, p4d, address); |
a00cc7d9 | 5453 | if (!vmf.pud) |
c74df32c | 5454 | return VM_FAULT_OOM; |
625110b5 | 5455 | retry_pud: |
7da4e2cb | 5456 | if (pud_none(*vmf.pud) && |
e0ffb29b MW |
5457 | thp_vma_allowable_order(vma, vm_flags, |
5458 | TVA_IN_PF | TVA_ENFORCE_SYSFS, PUD_ORDER)) { | |
a00cc7d9 MW |
5459 | ret = create_huge_pud(&vmf); |
5460 | if (!(ret & VM_FAULT_FALLBACK)) | |
5461 | return ret; | |
5462 | } else { | |
5463 | pud_t orig_pud = *vmf.pud; | |
5464 | ||
5465 | barrier(); | |
5466 | if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { | |
a00cc7d9 | 5467 | |
c89357e2 DH |
5468 | /* |
5469 | * TODO once we support anonymous PUDs: NUMA case and | |
5470 | * FAULT_FLAG_UNSHARE handling. | |
5471 | */ | |
5472 | if ((flags & FAULT_FLAG_WRITE) && !pud_write(orig_pud)) { | |
a00cc7d9 MW |
5473 | ret = wp_huge_pud(&vmf, orig_pud); |
5474 | if (!(ret & VM_FAULT_FALLBACK)) | |
5475 | return ret; | |
5476 | } else { | |
5477 | huge_pud_set_accessed(&vmf, orig_pud); | |
5478 | return 0; | |
5479 | } | |
5480 | } | |
5481 | } | |
5482 | ||
5483 | vmf.pmd = pmd_alloc(mm, vmf.pud, address); | |
82b0f8c3 | 5484 | if (!vmf.pmd) |
c74df32c | 5485 | return VM_FAULT_OOM; |
625110b5 TH |
5486 | |
5487 | /* Huge pud page fault raced with pmd_alloc? */ | |
5488 | if (pud_trans_unstable(vmf.pud)) | |
5489 | goto retry_pud; | |
5490 | ||
7da4e2cb | 5491 | if (pmd_none(*vmf.pmd) && |
e0ffb29b MW |
5492 | thp_vma_allowable_order(vma, vm_flags, |
5493 | TVA_IN_PF | TVA_ENFORCE_SYSFS, PMD_ORDER)) { | |
a2d58167 | 5494 | ret = create_huge_pmd(&vmf); |
c0292554 KS |
5495 | if (!(ret & VM_FAULT_FALLBACK)) |
5496 | return ret; | |
71e3aac0 | 5497 | } else { |
26e1a0c3 | 5498 | vmf.orig_pmd = pmdp_get_lockless(vmf.pmd); |
1f1d06c3 | 5499 | |
5db4f15c | 5500 | if (unlikely(is_swap_pmd(vmf.orig_pmd))) { |
84c3fc4e | 5501 | VM_BUG_ON(thp_migration_supported() && |
5db4f15c YS |
5502 | !is_pmd_migration_entry(vmf.orig_pmd)); |
5503 | if (is_pmd_migration_entry(vmf.orig_pmd)) | |
84c3fc4e ZY |
5504 | pmd_migration_entry_wait(mm, vmf.pmd); |
5505 | return 0; | |
5506 | } | |
5db4f15c YS |
5507 | if (pmd_trans_huge(vmf.orig_pmd) || pmd_devmap(vmf.orig_pmd)) { |
5508 | if (pmd_protnone(vmf.orig_pmd) && vma_is_accessible(vma)) | |
5509 | return do_huge_pmd_numa_page(&vmf); | |
d10e63f2 | 5510 | |
c89357e2 DH |
5511 | if ((flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) && |
5512 | !pmd_write(vmf.orig_pmd)) { | |
5db4f15c | 5513 | ret = wp_huge_pmd(&vmf); |
9845cbbd KS |
5514 | if (!(ret & VM_FAULT_FALLBACK)) |
5515 | return ret; | |
a1dd450b | 5516 | } else { |
5db4f15c | 5517 | huge_pmd_set_accessed(&vmf); |
9845cbbd | 5518 | return 0; |
1f1d06c3 | 5519 | } |
71e3aac0 AA |
5520 | } |
5521 | } | |
5522 | ||
82b0f8c3 | 5523 | return handle_pte_fault(&vmf); |
1da177e4 LT |
5524 | } |
5525 | ||
bce617ed | 5526 | /** |
f0953a1b | 5527 | * mm_account_fault - Do page fault accounting |
809ef83c | 5528 | * @mm: mm from which memcg should be extracted. It can be NULL. |
bce617ed PX |
5529 | * @regs: the pt_regs struct pointer. When set to NULL, will skip accounting |
5530 | * of perf event counters, but we'll still do the per-task accounting to | |
5531 | * the task who triggered this page fault. | |
5532 | * @address: the faulted address. | |
5533 | * @flags: the fault flags. | |
5534 | * @ret: the fault retcode. | |
5535 | * | |
f0953a1b | 5536 | * This will take care of most of the page fault accounting. Meanwhile, it |
bce617ed | 5537 | * will also include the PERF_COUNT_SW_PAGE_FAULTS_[MAJ|MIN] perf counter |
f0953a1b | 5538 | * updates. However, note that the handling of PERF_COUNT_SW_PAGE_FAULTS should |
bce617ed PX |
5539 | * still be in per-arch page fault handlers at the entry of page fault. |
5540 | */ | |
53156443 | 5541 | static inline void mm_account_fault(struct mm_struct *mm, struct pt_regs *regs, |
bce617ed PX |
5542 | unsigned long address, unsigned int flags, |
5543 | vm_fault_t ret) | |
5544 | { | |
5545 | bool major; | |
5546 | ||
53156443 SB |
5547 | /* Incomplete faults will be accounted upon completion. */ |
5548 | if (ret & VM_FAULT_RETRY) | |
5549 | return; | |
5550 | ||
bce617ed | 5551 | /* |
53156443 SB |
5552 | * To preserve the behavior of older kernels, PGFAULT counters record |
5553 | * both successful and failed faults, as opposed to perf counters, | |
5554 | * which ignore failed cases. | |
bce617ed | 5555 | */ |
53156443 SB |
5556 | count_vm_event(PGFAULT); |
5557 | count_memcg_event_mm(mm, PGFAULT); | |
5558 | ||
5559 | /* | |
5560 | * Do not account for unsuccessful faults (e.g. when the address wasn't | |
5561 | * valid). That includes arch_vma_access_permitted() failing before | |
5562 | * reaching here. So this is not a "this many hardware page faults" | |
5563 | * counter. We should use the hw profiling for that. | |
5564 | */ | |
5565 | if (ret & VM_FAULT_ERROR) | |
bce617ed PX |
5566 | return; |
5567 | ||
5568 | /* | |
5569 | * We define the fault as a major fault when the final successful fault | |
5570 | * is VM_FAULT_MAJOR, or if it retried (which implies that we couldn't | |
5571 | * handle it immediately previously). | |
5572 | */ | |
5573 | major = (ret & VM_FAULT_MAJOR) || (flags & FAULT_FLAG_TRIED); | |
5574 | ||
a2beb5f1 PX |
5575 | if (major) |
5576 | current->maj_flt++; | |
5577 | else | |
5578 | current->min_flt++; | |
5579 | ||
bce617ed | 5580 | /* |
a2beb5f1 PX |
5581 | * If the fault is done for GUP, regs will be NULL. We only do the |
5582 | * accounting for the per thread fault counters who triggered the | |
5583 | * fault, and we skip the perf event updates. | |
bce617ed PX |
5584 | */ |
5585 | if (!regs) | |
5586 | return; | |
5587 | ||
a2beb5f1 | 5588 | if (major) |
bce617ed | 5589 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); |
a2beb5f1 | 5590 | else |
bce617ed | 5591 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); |
bce617ed PX |
5592 | } |
5593 | ||
ec1c86b2 YZ |
5594 | #ifdef CONFIG_LRU_GEN |
5595 | static void lru_gen_enter_fault(struct vm_area_struct *vma) | |
5596 | { | |
8788f678 YZ |
5597 | /* the LRU algorithm only applies to accesses with recency */ |
5598 | current->in_lru_fault = vma_has_recency(vma); | |
ec1c86b2 YZ |
5599 | } |
5600 | ||
5601 | static void lru_gen_exit_fault(void) | |
5602 | { | |
5603 | current->in_lru_fault = false; | |
5604 | } | |
5605 | #else | |
5606 | static void lru_gen_enter_fault(struct vm_area_struct *vma) | |
5607 | { | |
5608 | } | |
5609 | ||
5610 | static void lru_gen_exit_fault(void) | |
5611 | { | |
5612 | } | |
5613 | #endif /* CONFIG_LRU_GEN */ | |
5614 | ||
cdc5021c DH |
5615 | static vm_fault_t sanitize_fault_flags(struct vm_area_struct *vma, |
5616 | unsigned int *flags) | |
5617 | { | |
5618 | if (unlikely(*flags & FAULT_FLAG_UNSHARE)) { | |
5619 | if (WARN_ON_ONCE(*flags & FAULT_FLAG_WRITE)) | |
5620 | return VM_FAULT_SIGSEGV; | |
5621 | /* | |
5622 | * FAULT_FLAG_UNSHARE only applies to COW mappings. Let's | |
5623 | * just treat it like an ordinary read-fault otherwise. | |
5624 | */ | |
5625 | if (!is_cow_mapping(vma->vm_flags)) | |
5626 | *flags &= ~FAULT_FLAG_UNSHARE; | |
79881fed DH |
5627 | } else if (*flags & FAULT_FLAG_WRITE) { |
5628 | /* Write faults on read-only mappings are impossible ... */ | |
5629 | if (WARN_ON_ONCE(!(vma->vm_flags & VM_MAYWRITE))) | |
5630 | return VM_FAULT_SIGSEGV; | |
5631 | /* ... and FOLL_FORCE only applies to COW mappings. */ | |
5632 | if (WARN_ON_ONCE(!(vma->vm_flags & VM_WRITE) && | |
5633 | !is_cow_mapping(vma->vm_flags))) | |
5634 | return VM_FAULT_SIGSEGV; | |
cdc5021c | 5635 | } |
4089eef0 SB |
5636 | #ifdef CONFIG_PER_VMA_LOCK |
5637 | /* | |
5638 | * Per-VMA locks can't be used with FAULT_FLAG_RETRY_NOWAIT because of | |
5639 | * the assumption that lock is dropped on VM_FAULT_RETRY. | |
5640 | */ | |
5641 | if (WARN_ON_ONCE((*flags & | |
5642 | (FAULT_FLAG_VMA_LOCK | FAULT_FLAG_RETRY_NOWAIT)) == | |
5643 | (FAULT_FLAG_VMA_LOCK | FAULT_FLAG_RETRY_NOWAIT))) | |
5644 | return VM_FAULT_SIGSEGV; | |
5645 | #endif | |
5646 | ||
cdc5021c DH |
5647 | return 0; |
5648 | } | |
5649 | ||
9a95f3cf PC |
5650 | /* |
5651 | * By the time we get here, we already hold the mm semaphore | |
5652 | * | |
c1e8d7c6 | 5653 | * The mmap_lock may have been released depending on flags and our |
9138e47e | 5654 | * return value. See filemap_fault() and __folio_lock_or_retry(). |
9a95f3cf | 5655 | */ |
2b740303 | 5656 | vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
bce617ed | 5657 | unsigned int flags, struct pt_regs *regs) |
519e5247 | 5658 | { |
53156443 SB |
5659 | /* If the fault handler drops the mmap_lock, vma may be freed */ |
5660 | struct mm_struct *mm = vma->vm_mm; | |
2b740303 | 5661 | vm_fault_t ret; |
519e5247 JW |
5662 | |
5663 | __set_current_state(TASK_RUNNING); | |
5664 | ||
cdc5021c DH |
5665 | ret = sanitize_fault_flags(vma, &flags); |
5666 | if (ret) | |
53156443 | 5667 | goto out; |
cdc5021c | 5668 | |
de0c799b LD |
5669 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
5670 | flags & FAULT_FLAG_INSTRUCTION, | |
53156443 SB |
5671 | flags & FAULT_FLAG_REMOTE)) { |
5672 | ret = VM_FAULT_SIGSEGV; | |
5673 | goto out; | |
5674 | } | |
de0c799b | 5675 | |
519e5247 JW |
5676 | /* |
5677 | * Enable the memcg OOM handling for faults triggered in user | |
5678 | * space. Kernel faults are handled more gracefully. | |
5679 | */ | |
5680 | if (flags & FAULT_FLAG_USER) | |
29ef680a | 5681 | mem_cgroup_enter_user_fault(); |
519e5247 | 5682 | |
ec1c86b2 YZ |
5683 | lru_gen_enter_fault(vma); |
5684 | ||
bae473a4 KS |
5685 | if (unlikely(is_vm_hugetlb_page(vma))) |
5686 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); | |
5687 | else | |
5688 | ret = __handle_mm_fault(vma, address, flags); | |
519e5247 | 5689 | |
ec1c86b2 YZ |
5690 | lru_gen_exit_fault(); |
5691 | ||
49426420 | 5692 | if (flags & FAULT_FLAG_USER) { |
29ef680a | 5693 | mem_cgroup_exit_user_fault(); |
166f61b9 TH |
5694 | /* |
5695 | * The task may have entered a memcg OOM situation but | |
5696 | * if the allocation error was handled gracefully (no | |
5697 | * VM_FAULT_OOM), there is no need to kill anything. | |
5698 | * Just clean up the OOM state peacefully. | |
5699 | */ | |
5700 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
5701 | mem_cgroup_oom_synchronize(false); | |
49426420 | 5702 | } |
53156443 SB |
5703 | out: |
5704 | mm_account_fault(mm, regs, address, flags, ret); | |
bce617ed | 5705 | |
519e5247 JW |
5706 | return ret; |
5707 | } | |
e1d6d01a | 5708 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 5709 | |
c2508ec5 LT |
5710 | #ifdef CONFIG_LOCK_MM_AND_FIND_VMA |
5711 | #include <linux/extable.h> | |
5712 | ||
5713 | static inline bool get_mmap_lock_carefully(struct mm_struct *mm, struct pt_regs *regs) | |
5714 | { | |
4542057e | 5715 | if (likely(mmap_read_trylock(mm))) |
c2508ec5 | 5716 | return true; |
c2508ec5 LT |
5717 | |
5718 | if (regs && !user_mode(regs)) { | |
8fa50708 | 5719 | unsigned long ip = exception_ip(regs); |
c2508ec5 LT |
5720 | if (!search_exception_tables(ip)) |
5721 | return false; | |
5722 | } | |
5723 | ||
eda00472 | 5724 | return !mmap_read_lock_killable(mm); |
c2508ec5 LT |
5725 | } |
5726 | ||
5727 | static inline bool mmap_upgrade_trylock(struct mm_struct *mm) | |
5728 | { | |
5729 | /* | |
5730 | * We don't have this operation yet. | |
5731 | * | |
5732 | * It should be easy enough to do: it's basically a | |
5733 | * atomic_long_try_cmpxchg_acquire() | |
5734 | * from RWSEM_READER_BIAS -> RWSEM_WRITER_LOCKED, but | |
5735 | * it also needs the proper lockdep magic etc. | |
5736 | */ | |
5737 | return false; | |
5738 | } | |
5739 | ||
5740 | static inline bool upgrade_mmap_lock_carefully(struct mm_struct *mm, struct pt_regs *regs) | |
5741 | { | |
5742 | mmap_read_unlock(mm); | |
5743 | if (regs && !user_mode(regs)) { | |
8fa50708 | 5744 | unsigned long ip = exception_ip(regs); |
c2508ec5 LT |
5745 | if (!search_exception_tables(ip)) |
5746 | return false; | |
5747 | } | |
eda00472 | 5748 | return !mmap_write_lock_killable(mm); |
c2508ec5 LT |
5749 | } |
5750 | ||
5751 | /* | |
5752 | * Helper for page fault handling. | |
5753 | * | |
5754 | * This is kind of equivalend to "mmap_read_lock()" followed | |
5755 | * by "find_extend_vma()", except it's a lot more careful about | |
5756 | * the locking (and will drop the lock on failure). | |
5757 | * | |
5758 | * For example, if we have a kernel bug that causes a page | |
5759 | * fault, we don't want to just use mmap_read_lock() to get | |
5760 | * the mm lock, because that would deadlock if the bug were | |
5761 | * to happen while we're holding the mm lock for writing. | |
5762 | * | |
5763 | * So this checks the exception tables on kernel faults in | |
5764 | * order to only do this all for instructions that are actually | |
5765 | * expected to fault. | |
5766 | * | |
5767 | * We can also actually take the mm lock for writing if we | |
5768 | * need to extend the vma, which helps the VM layer a lot. | |
5769 | */ | |
5770 | struct vm_area_struct *lock_mm_and_find_vma(struct mm_struct *mm, | |
5771 | unsigned long addr, struct pt_regs *regs) | |
5772 | { | |
5773 | struct vm_area_struct *vma; | |
5774 | ||
5775 | if (!get_mmap_lock_carefully(mm, regs)) | |
5776 | return NULL; | |
5777 | ||
5778 | vma = find_vma(mm, addr); | |
5779 | if (likely(vma && (vma->vm_start <= addr))) | |
5780 | return vma; | |
5781 | ||
5782 | /* | |
5783 | * Well, dang. We might still be successful, but only | |
5784 | * if we can extend a vma to do so. | |
5785 | */ | |
5786 | if (!vma || !(vma->vm_flags & VM_GROWSDOWN)) { | |
5787 | mmap_read_unlock(mm); | |
5788 | return NULL; | |
5789 | } | |
5790 | ||
5791 | /* | |
5792 | * We can try to upgrade the mmap lock atomically, | |
5793 | * in which case we can continue to use the vma | |
5794 | * we already looked up. | |
5795 | * | |
5796 | * Otherwise we'll have to drop the mmap lock and | |
5797 | * re-take it, and also look up the vma again, | |
5798 | * re-checking it. | |
5799 | */ | |
5800 | if (!mmap_upgrade_trylock(mm)) { | |
5801 | if (!upgrade_mmap_lock_carefully(mm, regs)) | |
5802 | return NULL; | |
5803 | ||
5804 | vma = find_vma(mm, addr); | |
5805 | if (!vma) | |
5806 | goto fail; | |
5807 | if (vma->vm_start <= addr) | |
5808 | goto success; | |
5809 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
5810 | goto fail; | |
5811 | } | |
5812 | ||
8d7071af | 5813 | if (expand_stack_locked(vma, addr)) |
c2508ec5 LT |
5814 | goto fail; |
5815 | ||
5816 | success: | |
5817 | mmap_write_downgrade(mm); | |
5818 | return vma; | |
5819 | ||
5820 | fail: | |
5821 | mmap_write_unlock(mm); | |
5822 | return NULL; | |
5823 | } | |
5824 | #endif | |
5825 | ||
50ee3253 SB |
5826 | #ifdef CONFIG_PER_VMA_LOCK |
5827 | /* | |
5828 | * Lookup and lock a VMA under RCU protection. Returned VMA is guaranteed to be | |
5829 | * stable and not isolated. If the VMA is not found or is being modified the | |
5830 | * function returns NULL. | |
5831 | */ | |
5832 | struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm, | |
5833 | unsigned long address) | |
5834 | { | |
5835 | MA_STATE(mas, &mm->mm_mt, address, address); | |
5836 | struct vm_area_struct *vma; | |
5837 | ||
5838 | rcu_read_lock(); | |
5839 | retry: | |
5840 | vma = mas_walk(&mas); | |
5841 | if (!vma) | |
5842 | goto inval; | |
5843 | ||
50ee3253 SB |
5844 | if (!vma_start_read(vma)) |
5845 | goto inval; | |
5846 | ||
5847 | /* Check since vm_start/vm_end might change before we lock the VMA */ | |
657b5146 JH |
5848 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) |
5849 | goto inval_end_read; | |
50ee3253 SB |
5850 | |
5851 | /* Check if the VMA got isolated after we found it */ | |
5852 | if (vma->detached) { | |
5853 | vma_end_read(vma); | |
52f23865 | 5854 | count_vm_vma_lock_event(VMA_LOCK_MISS); |
50ee3253 SB |
5855 | /* The area was replaced with another one */ |
5856 | goto retry; | |
5857 | } | |
5858 | ||
5859 | rcu_read_unlock(); | |
5860 | return vma; | |
657b5146 JH |
5861 | |
5862 | inval_end_read: | |
5863 | vma_end_read(vma); | |
50ee3253 SB |
5864 | inval: |
5865 | rcu_read_unlock(); | |
52f23865 | 5866 | count_vm_vma_lock_event(VMA_LOCK_ABORT); |
50ee3253 SB |
5867 | return NULL; |
5868 | } | |
5869 | #endif /* CONFIG_PER_VMA_LOCK */ | |
5870 | ||
90eceff1 KS |
5871 | #ifndef __PAGETABLE_P4D_FOLDED |
5872 | /* | |
5873 | * Allocate p4d page table. | |
5874 | * We've already handled the fast-path in-line. | |
5875 | */ | |
5876 | int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
5877 | { | |
5878 | p4d_t *new = p4d_alloc_one(mm, address); | |
5879 | if (!new) | |
5880 | return -ENOMEM; | |
5881 | ||
90eceff1 | 5882 | spin_lock(&mm->page_table_lock); |
ed33b5a6 | 5883 | if (pgd_present(*pgd)) { /* Another has populated it */ |
90eceff1 | 5884 | p4d_free(mm, new); |
ed33b5a6 QZ |
5885 | } else { |
5886 | smp_wmb(); /* See comment in pmd_install() */ | |
90eceff1 | 5887 | pgd_populate(mm, pgd, new); |
ed33b5a6 | 5888 | } |
90eceff1 KS |
5889 | spin_unlock(&mm->page_table_lock); |
5890 | return 0; | |
5891 | } | |
5892 | #endif /* __PAGETABLE_P4D_FOLDED */ | |
5893 | ||
1da177e4 LT |
5894 | #ifndef __PAGETABLE_PUD_FOLDED |
5895 | /* | |
5896 | * Allocate page upper directory. | |
872fec16 | 5897 | * We've already handled the fast-path in-line. |
1da177e4 | 5898 | */ |
c2febafc | 5899 | int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) |
1da177e4 | 5900 | { |
c74df32c HD |
5901 | pud_t *new = pud_alloc_one(mm, address); |
5902 | if (!new) | |
1bb3630e | 5903 | return -ENOMEM; |
1da177e4 | 5904 | |
872fec16 | 5905 | spin_lock(&mm->page_table_lock); |
b4e98d9a KS |
5906 | if (!p4d_present(*p4d)) { |
5907 | mm_inc_nr_puds(mm); | |
ed33b5a6 | 5908 | smp_wmb(); /* See comment in pmd_install() */ |
c2febafc | 5909 | p4d_populate(mm, p4d, new); |
b4e98d9a | 5910 | } else /* Another has populated it */ |
5e541973 | 5911 | pud_free(mm, new); |
c74df32c | 5912 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 5913 | return 0; |
1da177e4 LT |
5914 | } |
5915 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
5916 | ||
5917 | #ifndef __PAGETABLE_PMD_FOLDED | |
5918 | /* | |
5919 | * Allocate page middle directory. | |
872fec16 | 5920 | * We've already handled the fast-path in-line. |
1da177e4 | 5921 | */ |
1bb3630e | 5922 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 5923 | { |
a00cc7d9 | 5924 | spinlock_t *ptl; |
c74df32c HD |
5925 | pmd_t *new = pmd_alloc_one(mm, address); |
5926 | if (!new) | |
1bb3630e | 5927 | return -ENOMEM; |
1da177e4 | 5928 | |
a00cc7d9 | 5929 | ptl = pud_lock(mm, pud); |
dc6c9a35 KS |
5930 | if (!pud_present(*pud)) { |
5931 | mm_inc_nr_pmds(mm); | |
ed33b5a6 | 5932 | smp_wmb(); /* See comment in pmd_install() */ |
1bb3630e | 5933 | pud_populate(mm, pud, new); |
ed33b5a6 | 5934 | } else { /* Another has populated it */ |
5e541973 | 5935 | pmd_free(mm, new); |
ed33b5a6 | 5936 | } |
a00cc7d9 | 5937 | spin_unlock(ptl); |
1bb3630e | 5938 | return 0; |
e0f39591 | 5939 | } |
1da177e4 LT |
5940 | #endif /* __PAGETABLE_PMD_FOLDED */ |
5941 | ||
0e5e64c0 MS |
5942 | /** |
5943 | * follow_pte - look up PTE at a user virtual address | |
29ae7d96 | 5944 | * @vma: the memory mapping |
0e5e64c0 MS |
5945 | * @address: user virtual address |
5946 | * @ptepp: location to store found PTE | |
5947 | * @ptlp: location to store the lock for the PTE | |
5948 | * | |
5949 | * On a successful return, the pointer to the PTE is stored in @ptepp; | |
5950 | * the corresponding lock is taken and its location is stored in @ptlp. | |
c5541ba3 DH |
5951 | * |
5952 | * The contents of the PTE are only stable until @ptlp is released using | |
5953 | * pte_unmap_unlock(). This function will fail if the PTE is non-present. | |
5954 | * Present PTEs may include PTEs that map refcounted pages, such as | |
5955 | * anonymous folios in COW mappings. | |
5956 | * | |
5957 | * Callers must be careful when relying on PTE content after | |
5958 | * pte_unmap_unlock(). Especially if the PTE maps a refcounted page, | |
5959 | * callers must protect against invalidation with MMU notifiers; otherwise | |
5960 | * access to the PFN at a later point in time can trigger use-after-free. | |
0e5e64c0 MS |
5961 | * |
5962 | * Only IO mappings and raw PFN mappings are allowed. The mmap semaphore | |
5963 | * should be taken for read. | |
5964 | * | |
c5541ba3 | 5965 | * This function must not be used to modify PTE content. |
0e5e64c0 MS |
5966 | * |
5967 | * Return: zero on success, -ve otherwise. | |
5968 | */ | |
29ae7d96 | 5969 | int follow_pte(struct vm_area_struct *vma, unsigned long address, |
0e5e64c0 | 5970 | pte_t **ptepp, spinlock_t **ptlp) |
f8ad0f49 | 5971 | { |
29ae7d96 | 5972 | struct mm_struct *mm = vma->vm_mm; |
f8ad0f49 | 5973 | pgd_t *pgd; |
c2febafc | 5974 | p4d_t *p4d; |
f8ad0f49 JW |
5975 | pud_t *pud; |
5976 | pmd_t *pmd; | |
5977 | pte_t *ptep; | |
5978 | ||
c5541ba3 DH |
5979 | mmap_assert_locked(mm); |
5980 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) | |
5981 | goto out; | |
5982 | ||
29ae7d96 DH |
5983 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
5984 | goto out; | |
5985 | ||
f8ad0f49 JW |
5986 | pgd = pgd_offset(mm, address); |
5987 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
5988 | goto out; | |
5989 | ||
c2febafc KS |
5990 | p4d = p4d_offset(pgd, address); |
5991 | if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) | |
5992 | goto out; | |
5993 | ||
5994 | pud = pud_offset(p4d, address); | |
f8ad0f49 JW |
5995 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
5996 | goto out; | |
5997 | ||
5998 | pmd = pmd_offset(pud, address); | |
f66055ab | 5999 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 | 6000 | |
f8ad0f49 | 6001 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
3db82b93 HD |
6002 | if (!ptep) |
6003 | goto out; | |
c33c7948 | 6004 | if (!pte_present(ptep_get(ptep))) |
f8ad0f49 JW |
6005 | goto unlock; |
6006 | *ptepp = ptep; | |
6007 | return 0; | |
6008 | unlock: | |
6009 | pte_unmap_unlock(ptep, *ptlp); | |
6010 | out: | |
6011 | return -EINVAL; | |
6012 | } | |
9fd6dad1 PB |
6013 | EXPORT_SYMBOL_GPL(follow_pte); |
6014 | ||
28b2ee20 | 6015 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
96667f8a DV |
6016 | /** |
6017 | * generic_access_phys - generic implementation for iomem mmap access | |
6018 | * @vma: the vma to access | |
f0953a1b | 6019 | * @addr: userspace address, not relative offset within @vma |
96667f8a DV |
6020 | * @buf: buffer to read/write |
6021 | * @len: length of transfer | |
6022 | * @write: set to FOLL_WRITE when writing, otherwise reading | |
6023 | * | |
6024 | * This is a generic implementation for &vm_operations_struct.access for an | |
6025 | * iomem mapping. This callback is used by access_process_vm() when the @vma is | |
6026 | * not page based. | |
6027 | */ | |
28b2ee20 RR |
6028 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, |
6029 | void *buf, int len, int write) | |
6030 | { | |
6031 | resource_size_t phys_addr; | |
6032 | unsigned long prot = 0; | |
2bc7273b | 6033 | void __iomem *maddr; |
96667f8a DV |
6034 | pte_t *ptep, pte; |
6035 | spinlock_t *ptl; | |
6036 | int offset = offset_in_page(addr); | |
6037 | int ret = -EINVAL; | |
6038 | ||
96667f8a | 6039 | retry: |
29ae7d96 | 6040 | if (follow_pte(vma, addr, &ptep, &ptl)) |
96667f8a | 6041 | return -EINVAL; |
c33c7948 | 6042 | pte = ptep_get(ptep); |
96667f8a | 6043 | pte_unmap_unlock(ptep, ptl); |
28b2ee20 | 6044 | |
96667f8a DV |
6045 | prot = pgprot_val(pte_pgprot(pte)); |
6046 | phys_addr = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; | |
6047 | ||
6048 | if ((write & FOLL_WRITE) && !pte_write(pte)) | |
28b2ee20 RR |
6049 | return -EINVAL; |
6050 | ||
9cb12d7b | 6051 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
24eee1e4 | 6052 | if (!maddr) |
6053 | return -ENOMEM; | |
6054 | ||
29ae7d96 | 6055 | if (follow_pte(vma, addr, &ptep, &ptl)) |
96667f8a DV |
6056 | goto out_unmap; |
6057 | ||
c33c7948 | 6058 | if (!pte_same(pte, ptep_get(ptep))) { |
96667f8a DV |
6059 | pte_unmap_unlock(ptep, ptl); |
6060 | iounmap(maddr); | |
6061 | ||
6062 | goto retry; | |
6063 | } | |
6064 | ||
28b2ee20 RR |
6065 | if (write) |
6066 | memcpy_toio(maddr + offset, buf, len); | |
6067 | else | |
6068 | memcpy_fromio(buf, maddr + offset, len); | |
96667f8a DV |
6069 | ret = len; |
6070 | pte_unmap_unlock(ptep, ptl); | |
6071 | out_unmap: | |
28b2ee20 RR |
6072 | iounmap(maddr); |
6073 | ||
96667f8a | 6074 | return ret; |
28b2ee20 | 6075 | } |
5a73633e | 6076 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
6077 | #endif |
6078 | ||
0ec76a11 | 6079 | /* |
d3f5ffca | 6080 | * Access another process' address space as given in mm. |
0ec76a11 | 6081 | */ |
c43cfa42 LS |
6082 | static int __access_remote_vm(struct mm_struct *mm, unsigned long addr, |
6083 | void *buf, int len, unsigned int gup_flags) | |
0ec76a11 | 6084 | { |
0ec76a11 | 6085 | void *old_buf = buf; |
442486ec | 6086 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 6087 | |
d8ed45c5 | 6088 | if (mmap_read_lock_killable(mm)) |
1e426fe2 KK |
6089 | return 0; |
6090 | ||
22883973 KS |
6091 | /* Untag the address before looking up the VMA */ |
6092 | addr = untagged_addr_remote(mm, addr); | |
6093 | ||
eee9c708 LT |
6094 | /* Avoid triggering the temporary warning in __get_user_pages */ |
6095 | if (!vma_lookup(mm, addr) && !expand_stack(mm, addr)) | |
6096 | return 0; | |
6097 | ||
183ff22b | 6098 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 | 6099 | while (len) { |
ca5e8632 | 6100 | int bytes, offset; |
0ec76a11 | 6101 | void *maddr; |
ca5e8632 LS |
6102 | struct vm_area_struct *vma = NULL; |
6103 | struct page *page = get_user_page_vma_remote(mm, addr, | |
6104 | gup_flags, &vma); | |
0ec76a11 | 6105 | |
6a1960b8 | 6106 | if (IS_ERR(page)) { |
9471f1f2 LT |
6107 | /* We might need to expand the stack to access it */ |
6108 | vma = vma_lookup(mm, addr); | |
6109 | if (!vma) { | |
6110 | vma = expand_stack(mm, addr); | |
6111 | ||
6112 | /* mmap_lock was dropped on failure */ | |
6113 | if (!vma) | |
6114 | return buf - old_buf; | |
6115 | ||
6116 | /* Try again if stack expansion worked */ | |
6117 | continue; | |
6118 | } | |
6119 | ||
28b2ee20 RR |
6120 | /* |
6121 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
6122 | * we can access using slightly different code. | |
6123 | */ | |
9471f1f2 LT |
6124 | bytes = 0; |
6125 | #ifdef CONFIG_HAVE_IOREMAP_PROT | |
28b2ee20 | 6126 | if (vma->vm_ops && vma->vm_ops->access) |
9471f1f2 LT |
6127 | bytes = vma->vm_ops->access(vma, addr, buf, |
6128 | len, write); | |
dbffcd03 | 6129 | #endif |
9471f1f2 LT |
6130 | if (bytes <= 0) |
6131 | break; | |
0ec76a11 | 6132 | } else { |
28b2ee20 RR |
6133 | bytes = len; |
6134 | offset = addr & (PAGE_SIZE-1); | |
6135 | if (bytes > PAGE_SIZE-offset) | |
6136 | bytes = PAGE_SIZE-offset; | |
6137 | ||
f7ef5fe7 | 6138 | maddr = kmap_local_page(page); |
28b2ee20 RR |
6139 | if (write) { |
6140 | copy_to_user_page(vma, page, addr, | |
6141 | maddr + offset, buf, bytes); | |
6142 | set_page_dirty_lock(page); | |
6143 | } else { | |
6144 | copy_from_user_page(vma, page, addr, | |
6145 | buf, maddr + offset, bytes); | |
6146 | } | |
f7ef5fe7 | 6147 | unmap_and_put_page(page, maddr); |
0ec76a11 | 6148 | } |
0ec76a11 DH |
6149 | len -= bytes; |
6150 | buf += bytes; | |
6151 | addr += bytes; | |
6152 | } | |
d8ed45c5 | 6153 | mmap_read_unlock(mm); |
0ec76a11 DH |
6154 | |
6155 | return buf - old_buf; | |
6156 | } | |
03252919 | 6157 | |
5ddd36b9 | 6158 | /** |
ae91dbfc | 6159 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
6160 | * @mm: the mm_struct of the target address space |
6161 | * @addr: start address to access | |
6162 | * @buf: source or destination buffer | |
6163 | * @len: number of bytes to transfer | |
6347e8d5 | 6164 | * @gup_flags: flags modifying lookup behaviour |
5ddd36b9 SW |
6165 | * |
6166 | * The caller must hold a reference on @mm. | |
a862f68a MR |
6167 | * |
6168 | * Return: number of bytes copied from source to destination. | |
5ddd36b9 SW |
6169 | */ |
6170 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 6171 | void *buf, int len, unsigned int gup_flags) |
5ddd36b9 | 6172 | { |
d3f5ffca | 6173 | return __access_remote_vm(mm, addr, buf, len, gup_flags); |
5ddd36b9 SW |
6174 | } |
6175 | ||
206cb636 SW |
6176 | /* |
6177 | * Access another process' address space. | |
6178 | * Source/target buffer must be kernel space, | |
6179 | * Do not walk the page table directly, use get_user_pages | |
6180 | */ | |
6181 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
f307ab6d | 6182 | void *buf, int len, unsigned int gup_flags) |
206cb636 SW |
6183 | { |
6184 | struct mm_struct *mm; | |
6185 | int ret; | |
6186 | ||
6187 | mm = get_task_mm(tsk); | |
6188 | if (!mm) | |
6189 | return 0; | |
6190 | ||
d3f5ffca | 6191 | ret = __access_remote_vm(mm, addr, buf, len, gup_flags); |
442486ec | 6192 | |
206cb636 SW |
6193 | mmput(mm); |
6194 | ||
6195 | return ret; | |
6196 | } | |
fcd35857 | 6197 | EXPORT_SYMBOL_GPL(access_process_vm); |
206cb636 | 6198 | |
03252919 AK |
6199 | /* |
6200 | * Print the name of a VMA. | |
6201 | */ | |
6202 | void print_vma_addr(char *prefix, unsigned long ip) | |
6203 | { | |
6204 | struct mm_struct *mm = current->mm; | |
6205 | struct vm_area_struct *vma; | |
6206 | ||
e8bff74a | 6207 | /* |
0a7f682d | 6208 | * we might be running from an atomic context so we cannot sleep |
e8bff74a | 6209 | */ |
d8ed45c5 | 6210 | if (!mmap_read_trylock(mm)) |
e8bff74a IM |
6211 | return; |
6212 | ||
de7e71ef | 6213 | vma = vma_lookup(mm, ip); |
03252919 AK |
6214 | if (vma && vma->vm_file) { |
6215 | struct file *f = vma->vm_file; | |
de7e71ef LT |
6216 | ip -= vma->vm_start; |
6217 | ip += vma->vm_pgoff << PAGE_SHIFT; | |
6218 | printk("%s%pD[%lx,%lx+%lx]", prefix, f, ip, | |
6219 | vma->vm_start, | |
6220 | vma->vm_end - vma->vm_start); | |
03252919 | 6221 | } |
d8ed45c5 | 6222 | mmap_read_unlock(mm); |
03252919 | 6223 | } |
3ee1afa3 | 6224 | |
662bbcb2 | 6225 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 6226 | void __might_fault(const char *file, int line) |
3ee1afa3 | 6227 | { |
9ec23531 | 6228 | if (pagefault_disabled()) |
662bbcb2 | 6229 | return; |
42a38756 | 6230 | __might_sleep(file, line); |
9ec23531 | 6231 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
662bbcb2 | 6232 | if (current->mm) |
da1c55f1 | 6233 | might_lock_read(¤t->mm->mmap_lock); |
9ec23531 | 6234 | #endif |
3ee1afa3 | 6235 | } |
9ec23531 | 6236 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 6237 | #endif |
47ad8475 AA |
6238 | |
6239 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
c6ddfb6c HY |
6240 | /* |
6241 | * Process all subpages of the specified huge page with the specified | |
6242 | * operation. The target subpage will be processed last to keep its | |
6243 | * cache lines hot. | |
6244 | */ | |
1cb9dc4b | 6245 | static inline int process_huge_page( |
c6ddfb6c | 6246 | unsigned long addr_hint, unsigned int pages_per_huge_page, |
1cb9dc4b | 6247 | int (*process_subpage)(unsigned long addr, int idx, void *arg), |
c6ddfb6c | 6248 | void *arg) |
47ad8475 | 6249 | { |
1cb9dc4b | 6250 | int i, n, base, l, ret; |
c79b57e4 HY |
6251 | unsigned long addr = addr_hint & |
6252 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
47ad8475 | 6253 | |
c6ddfb6c | 6254 | /* Process target subpage last to keep its cache lines hot */ |
47ad8475 | 6255 | might_sleep(); |
c79b57e4 HY |
6256 | n = (addr_hint - addr) / PAGE_SIZE; |
6257 | if (2 * n <= pages_per_huge_page) { | |
c6ddfb6c | 6258 | /* If target subpage in first half of huge page */ |
c79b57e4 HY |
6259 | base = 0; |
6260 | l = n; | |
c6ddfb6c | 6261 | /* Process subpages at the end of huge page */ |
c79b57e4 HY |
6262 | for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { |
6263 | cond_resched(); | |
1cb9dc4b LS |
6264 | ret = process_subpage(addr + i * PAGE_SIZE, i, arg); |
6265 | if (ret) | |
6266 | return ret; | |
c79b57e4 HY |
6267 | } |
6268 | } else { | |
c6ddfb6c | 6269 | /* If target subpage in second half of huge page */ |
c79b57e4 HY |
6270 | base = pages_per_huge_page - 2 * (pages_per_huge_page - n); |
6271 | l = pages_per_huge_page - n; | |
c6ddfb6c | 6272 | /* Process subpages at the begin of huge page */ |
c79b57e4 HY |
6273 | for (i = 0; i < base; i++) { |
6274 | cond_resched(); | |
1cb9dc4b LS |
6275 | ret = process_subpage(addr + i * PAGE_SIZE, i, arg); |
6276 | if (ret) | |
6277 | return ret; | |
c79b57e4 HY |
6278 | } |
6279 | } | |
6280 | /* | |
c6ddfb6c HY |
6281 | * Process remaining subpages in left-right-left-right pattern |
6282 | * towards the target subpage | |
c79b57e4 HY |
6283 | */ |
6284 | for (i = 0; i < l; i++) { | |
6285 | int left_idx = base + i; | |
6286 | int right_idx = base + 2 * l - 1 - i; | |
6287 | ||
6288 | cond_resched(); | |
1cb9dc4b LS |
6289 | ret = process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg); |
6290 | if (ret) | |
6291 | return ret; | |
47ad8475 | 6292 | cond_resched(); |
1cb9dc4b LS |
6293 | ret = process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg); |
6294 | if (ret) | |
6295 | return ret; | |
47ad8475 | 6296 | } |
1cb9dc4b | 6297 | return 0; |
47ad8475 AA |
6298 | } |
6299 | ||
c6ddfb6c HY |
6300 | static void clear_gigantic_page(struct page *page, |
6301 | unsigned long addr, | |
6302 | unsigned int pages_per_huge_page) | |
6303 | { | |
6304 | int i; | |
14455eab | 6305 | struct page *p; |
c6ddfb6c HY |
6306 | |
6307 | might_sleep(); | |
14455eab CL |
6308 | for (i = 0; i < pages_per_huge_page; i++) { |
6309 | p = nth_page(page, i); | |
c6ddfb6c HY |
6310 | cond_resched(); |
6311 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
6312 | } | |
6313 | } | |
6314 | ||
1cb9dc4b | 6315 | static int clear_subpage(unsigned long addr, int idx, void *arg) |
c6ddfb6c HY |
6316 | { |
6317 | struct page *page = arg; | |
6318 | ||
21fff064 | 6319 | clear_user_highpage(nth_page(page, idx), addr); |
1cb9dc4b | 6320 | return 0; |
c6ddfb6c HY |
6321 | } |
6322 | ||
6323 | void clear_huge_page(struct page *page, | |
6324 | unsigned long addr_hint, unsigned int pages_per_huge_page) | |
6325 | { | |
6326 | unsigned long addr = addr_hint & | |
6327 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
6328 | ||
6329 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
6330 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
6331 | return; | |
6332 | } | |
6333 | ||
6334 | process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page); | |
6335 | } | |
6336 | ||
1cb9dc4b | 6337 | static int copy_user_gigantic_page(struct folio *dst, struct folio *src, |
c0e8150e Z |
6338 | unsigned long addr, |
6339 | struct vm_area_struct *vma, | |
6340 | unsigned int pages_per_huge_page) | |
47ad8475 AA |
6341 | { |
6342 | int i; | |
c0e8150e Z |
6343 | struct page *dst_page; |
6344 | struct page *src_page; | |
47ad8475 | 6345 | |
14455eab | 6346 | for (i = 0; i < pages_per_huge_page; i++) { |
c0e8150e Z |
6347 | dst_page = folio_page(dst, i); |
6348 | src_page = folio_page(src, i); | |
14455eab | 6349 | |
47ad8475 | 6350 | cond_resched(); |
1cb9dc4b LS |
6351 | if (copy_mc_user_highpage(dst_page, src_page, |
6352 | addr + i*PAGE_SIZE, vma)) { | |
6353 | memory_failure_queue(page_to_pfn(src_page), 0); | |
6354 | return -EHWPOISON; | |
6355 | } | |
47ad8475 | 6356 | } |
1cb9dc4b | 6357 | return 0; |
47ad8475 AA |
6358 | } |
6359 | ||
c9f4cd71 HY |
6360 | struct copy_subpage_arg { |
6361 | struct page *dst; | |
6362 | struct page *src; | |
6363 | struct vm_area_struct *vma; | |
6364 | }; | |
6365 | ||
1cb9dc4b | 6366 | static int copy_subpage(unsigned long addr, int idx, void *arg) |
c9f4cd71 HY |
6367 | { |
6368 | struct copy_subpage_arg *copy_arg = arg; | |
21fff064 KW |
6369 | struct page *dst = nth_page(copy_arg->dst, idx); |
6370 | struct page *src = nth_page(copy_arg->src, idx); | |
c9f4cd71 | 6371 | |
21fff064 KW |
6372 | if (copy_mc_user_highpage(dst, src, addr, copy_arg->vma)) { |
6373 | memory_failure_queue(page_to_pfn(src), 0); | |
1cb9dc4b LS |
6374 | return -EHWPOISON; |
6375 | } | |
6376 | return 0; | |
c9f4cd71 HY |
6377 | } |
6378 | ||
1cb9dc4b LS |
6379 | int copy_user_large_folio(struct folio *dst, struct folio *src, |
6380 | unsigned long addr_hint, struct vm_area_struct *vma) | |
47ad8475 | 6381 | { |
c0e8150e | 6382 | unsigned int pages_per_huge_page = folio_nr_pages(dst); |
c9f4cd71 HY |
6383 | unsigned long addr = addr_hint & |
6384 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
6385 | struct copy_subpage_arg arg = { | |
c0e8150e Z |
6386 | .dst = &dst->page, |
6387 | .src = &src->page, | |
c9f4cd71 HY |
6388 | .vma = vma, |
6389 | }; | |
47ad8475 | 6390 | |
1cb9dc4b LS |
6391 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) |
6392 | return copy_user_gigantic_page(dst, src, addr, vma, | |
6393 | pages_per_huge_page); | |
47ad8475 | 6394 | |
1cb9dc4b | 6395 | return process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg); |
47ad8475 | 6396 | } |
fa4d75c1 | 6397 | |
e87340ca Z |
6398 | long copy_folio_from_user(struct folio *dst_folio, |
6399 | const void __user *usr_src, | |
6400 | bool allow_pagefault) | |
fa4d75c1 | 6401 | { |
e87340ca | 6402 | void *kaddr; |
fa4d75c1 | 6403 | unsigned long i, rc = 0; |
e87340ca Z |
6404 | unsigned int nr_pages = folio_nr_pages(dst_folio); |
6405 | unsigned long ret_val = nr_pages * PAGE_SIZE; | |
14455eab | 6406 | struct page *subpage; |
fa4d75c1 | 6407 | |
e87340ca Z |
6408 | for (i = 0; i < nr_pages; i++) { |
6409 | subpage = folio_page(dst_folio, i); | |
6410 | kaddr = kmap_local_page(subpage); | |
0d508c1f Z |
6411 | if (!allow_pagefault) |
6412 | pagefault_disable(); | |
e87340ca | 6413 | rc = copy_from_user(kaddr, usr_src + i * PAGE_SIZE, PAGE_SIZE); |
0d508c1f Z |
6414 | if (!allow_pagefault) |
6415 | pagefault_enable(); | |
e87340ca | 6416 | kunmap_local(kaddr); |
fa4d75c1 MK |
6417 | |
6418 | ret_val -= (PAGE_SIZE - rc); | |
6419 | if (rc) | |
6420 | break; | |
6421 | ||
e763243c MS |
6422 | flush_dcache_page(subpage); |
6423 | ||
fa4d75c1 MK |
6424 | cond_resched(); |
6425 | } | |
6426 | return ret_val; | |
6427 | } | |
47ad8475 | 6428 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
49076ec2 | 6429 | |
40b64acd | 6430 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
6431 | |
6432 | static struct kmem_cache *page_ptl_cachep; | |
6433 | ||
6434 | void __init ptlock_cache_init(void) | |
6435 | { | |
6436 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
6437 | SLAB_PANIC, NULL); | |
6438 | } | |
6439 | ||
f5ecca06 | 6440 | bool ptlock_alloc(struct ptdesc *ptdesc) |
49076ec2 KS |
6441 | { |
6442 | spinlock_t *ptl; | |
6443 | ||
b35f1819 | 6444 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
6445 | if (!ptl) |
6446 | return false; | |
f5ecca06 | 6447 | ptdesc->ptl = ptl; |
49076ec2 KS |
6448 | return true; |
6449 | } | |
6450 | ||
6ed1b8a0 | 6451 | void ptlock_free(struct ptdesc *ptdesc) |
49076ec2 | 6452 | { |
6ed1b8a0 | 6453 | kmem_cache_free(page_ptl_cachep, ptdesc->ptl); |
49076ec2 KS |
6454 | } |
6455 | #endif | |
239e9a90 PX |
6456 | |
6457 | void vma_pgtable_walk_begin(struct vm_area_struct *vma) | |
6458 | { | |
6459 | if (is_vm_hugetlb_page(vma)) | |
6460 | hugetlb_vma_lock_read(vma); | |
6461 | } | |
6462 | ||
6463 | void vma_pgtable_walk_end(struct vm_area_struct *vma) | |
6464 | { | |
6465 | if (is_vm_hugetlb_page(vma)) | |
6466 | hugetlb_vma_unlock_read(vma); | |
6467 | } |