Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : /*
3 : * linux/mm/memory.c
4 : *
5 : * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 : */
7 :
8 : /*
9 : * demand-loading started 01.12.91 - seems it is high on the list of
10 : * things wanted, and it should be easy to implement. - Linus
11 : */
12 :
13 : /*
14 : * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
15 : * pages started 02.12.91, seems to work. - Linus.
16 : *
17 : * Tested sharing by executing about 30 /bin/sh: under the old kernel it
18 : * would have taken more than the 6M I have free, but it worked well as
19 : * far as I could see.
20 : *
21 : * Also corrected some "invalidate()"s - I wasn't doing enough of them.
22 : */
23 :
24 : /*
25 : * Real VM (paging to/from disk) started 18.12.91. Much more work and
26 : * thought has to go into this. Oh, well..
27 : * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
28 : * Found it. Everything seems to work now.
29 : * 20.12.91 - Ok, making the swap-device changeable like the root.
30 : */
31 :
32 : /*
33 : * 05.04.94 - Multi-page memory management added for v1.1.
34 : * Idea by Alex Bligh (alex@cconcepts.co.uk)
35 : *
36 : * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
37 : * (Gerhard.Wichert@pdb.siemens.de)
38 : *
39 : * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
40 : */
41 :
42 : #include <linux/kernel_stat.h>
43 : #include <linux/mm.h>
44 : #include <linux/mm_inline.h>
45 : #include <linux/sched/mm.h>
46 : #include <linux/sched/coredump.h>
47 : #include <linux/sched/numa_balancing.h>
48 : #include <linux/sched/task.h>
49 : #include <linux/hugetlb.h>
50 : #include <linux/mman.h>
51 : #include <linux/swap.h>
52 : #include <linux/highmem.h>
53 : #include <linux/pagemap.h>
54 : #include <linux/memremap.h>
55 : #include <linux/kmsan.h>
56 : #include <linux/ksm.h>
57 : #include <linux/rmap.h>
58 : #include <linux/export.h>
59 : #include <linux/delayacct.h>
60 : #include <linux/init.h>
61 : #include <linux/pfn_t.h>
62 : #include <linux/writeback.h>
63 : #include <linux/memcontrol.h>
64 : #include <linux/mmu_notifier.h>
65 : #include <linux/swapops.h>
66 : #include <linux/elf.h>
67 : #include <linux/gfp.h>
68 : #include <linux/migrate.h>
69 : #include <linux/string.h>
70 : #include <linux/memory-tiers.h>
71 : #include <linux/debugfs.h>
72 : #include <linux/userfaultfd_k.h>
73 : #include <linux/dax.h>
74 : #include <linux/oom.h>
75 : #include <linux/numa.h>
76 : #include <linux/perf_event.h>
77 : #include <linux/ptrace.h>
78 : #include <linux/vmalloc.h>
79 : #include <linux/sched/sysctl.h>
80 :
81 : #include <trace/events/kmem.h>
82 :
83 : #include <asm/io.h>
84 : #include <asm/mmu_context.h>
85 : #include <asm/pgalloc.h>
86 : #include <linux/uaccess.h>
87 : #include <asm/tlb.h>
88 : #include <asm/tlbflush.h>
89 :
90 : #include "pgalloc-track.h"
91 : #include "internal.h"
92 : #include "swap.h"
93 :
94 : #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST)
95 : #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
96 : #endif
97 :
98 : #ifndef CONFIG_NUMA
99 : unsigned long max_mapnr;
100 : EXPORT_SYMBOL(max_mapnr);
101 :
102 : struct page *mem_map;
103 : EXPORT_SYMBOL(mem_map);
104 : #endif
105 :
106 : static vm_fault_t do_fault(struct vm_fault *vmf);
107 :
108 : /*
109 : * A number of key systems in x86 including ioremap() rely on the assumption
110 : * that high_memory defines the upper bound on direct map memory, then end
111 : * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
112 : * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
113 : * and ZONE_HIGHMEM.
114 : */
115 : void *high_memory;
116 : EXPORT_SYMBOL(high_memory);
117 :
118 : /*
119 : * Randomize the address space (stacks, mmaps, brk, etc.).
120 : *
121 : * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
122 : * as ancient (libc5 based) binaries can segfault. )
123 : */
124 : int randomize_va_space __read_mostly =
125 : #ifdef CONFIG_COMPAT_BRK
126 : 1;
127 : #else
128 : 2;
129 : #endif
130 :
131 : #ifndef arch_wants_old_prefaulted_pte
132 : static inline bool arch_wants_old_prefaulted_pte(void)
133 : {
134 : /*
135 : * Transitioning a PTE from 'old' to 'young' can be expensive on
136 : * some architectures, even if it's performed in hardware. By
137 : * default, "false" means prefaulted entries will be 'young'.
138 : */
139 : return false;
140 : }
141 : #endif
142 :
143 0 : static int __init disable_randmaps(char *s)
144 : {
145 0 : randomize_va_space = 0;
146 0 : return 1;
147 : }
148 : __setup("norandmaps", disable_randmaps);
149 :
150 : unsigned long zero_pfn __read_mostly;
151 : EXPORT_SYMBOL(zero_pfn);
152 :
153 : unsigned long highest_memmap_pfn __read_mostly;
154 :
155 : /*
156 : * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
157 : */
158 1 : static int __init init_zero_pfn(void)
159 : {
160 2 : zero_pfn = page_to_pfn(ZERO_PAGE(0));
161 1 : return 0;
162 : }
163 : early_initcall(init_zero_pfn);
164 :
165 0 : void mm_trace_rss_stat(struct mm_struct *mm, int member)
166 : {
167 0 : trace_rss_stat(mm, member);
168 0 : }
169 :
170 : /*
171 : * Note: this doesn't free the actual pages themselves. That
172 : * has been handled earlier when unmapping all the memory regions.
173 : */
174 0 : static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
175 : unsigned long addr)
176 : {
177 0 : pgtable_t token = pmd_pgtable(*pmd);
178 0 : pmd_clear(pmd);
179 0 : pte_free_tlb(tlb, token, addr);
180 0 : mm_dec_nr_ptes(tlb->mm);
181 0 : }
182 :
183 0 : static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
184 : unsigned long addr, unsigned long end,
185 : unsigned long floor, unsigned long ceiling)
186 : {
187 : pmd_t *pmd;
188 : unsigned long next;
189 : unsigned long start;
190 :
191 0 : start = addr;
192 0 : pmd = pmd_offset(pud, addr);
193 : do {
194 0 : next = pmd_addr_end(addr, end);
195 0 : if (pmd_none_or_clear_bad(pmd))
196 0 : continue;
197 0 : free_pte_range(tlb, pmd, addr);
198 0 : } while (pmd++, addr = next, addr != end);
199 :
200 0 : start &= PUD_MASK;
201 0 : if (start < floor)
202 : return;
203 0 : if (ceiling) {
204 0 : ceiling &= PUD_MASK;
205 0 : if (!ceiling)
206 : return;
207 : }
208 0 : if (end - 1 > ceiling - 1)
209 : return;
210 :
211 0 : pmd = pmd_offset(pud, start);
212 0 : pud_clear(pud);
213 0 : pmd_free_tlb(tlb, pmd, start);
214 0 : mm_dec_nr_pmds(tlb->mm);
215 : }
216 :
217 0 : static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d,
218 : unsigned long addr, unsigned long end,
219 : unsigned long floor, unsigned long ceiling)
220 : {
221 : pud_t *pud;
222 : unsigned long next;
223 : unsigned long start;
224 :
225 0 : start = addr;
226 0 : pud = pud_offset(p4d, addr);
227 : do {
228 0 : next = pud_addr_end(addr, end);
229 0 : if (pud_none_or_clear_bad(pud))
230 0 : continue;
231 0 : free_pmd_range(tlb, pud, addr, next, floor, ceiling);
232 0 : } while (pud++, addr = next, addr != end);
233 :
234 0 : start &= P4D_MASK;
235 : if (start < floor)
236 : return;
237 : if (ceiling) {
238 : ceiling &= P4D_MASK;
239 : if (!ceiling)
240 : return;
241 : }
242 : if (end - 1 > ceiling - 1)
243 : return;
244 :
245 : pud = pud_offset(p4d, start);
246 : p4d_clear(p4d);
247 : pud_free_tlb(tlb, pud, start);
248 : mm_dec_nr_puds(tlb->mm);
249 : }
250 :
251 : static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd,
252 : unsigned long addr, unsigned long end,
253 : unsigned long floor, unsigned long ceiling)
254 : {
255 : p4d_t *p4d;
256 : unsigned long next;
257 : unsigned long start;
258 :
259 0 : start = addr;
260 0 : p4d = p4d_offset(pgd, addr);
261 : do {
262 0 : next = p4d_addr_end(addr, end);
263 0 : if (p4d_none_or_clear_bad(p4d))
264 : continue;
265 0 : free_pud_range(tlb, p4d, addr, next, floor, ceiling);
266 0 : } while (p4d++, addr = next, addr != end);
267 :
268 0 : start &= PGDIR_MASK;
269 : if (start < floor)
270 : return;
271 : if (ceiling) {
272 : ceiling &= PGDIR_MASK;
273 : if (!ceiling)
274 : return;
275 : }
276 : if (end - 1 > ceiling - 1)
277 : return;
278 :
279 : p4d = p4d_offset(pgd, start);
280 : pgd_clear(pgd);
281 : p4d_free_tlb(tlb, p4d, start);
282 : }
283 :
284 : /*
285 : * This function frees user-level page tables of a process.
286 : */
287 0 : void free_pgd_range(struct mmu_gather *tlb,
288 : unsigned long addr, unsigned long end,
289 : unsigned long floor, unsigned long ceiling)
290 : {
291 : pgd_t *pgd;
292 : unsigned long next;
293 :
294 : /*
295 : * The next few lines have given us lots of grief...
296 : *
297 : * Why are we testing PMD* at this top level? Because often
298 : * there will be no work to do at all, and we'd prefer not to
299 : * go all the way down to the bottom just to discover that.
300 : *
301 : * Why all these "- 1"s? Because 0 represents both the bottom
302 : * of the address space and the top of it (using -1 for the
303 : * top wouldn't help much: the masks would do the wrong thing).
304 : * The rule is that addr 0 and floor 0 refer to the bottom of
305 : * the address space, but end 0 and ceiling 0 refer to the top
306 : * Comparisons need to use "end - 1" and "ceiling - 1" (though
307 : * that end 0 case should be mythical).
308 : *
309 : * Wherever addr is brought up or ceiling brought down, we must
310 : * be careful to reject "the opposite 0" before it confuses the
311 : * subsequent tests. But what about where end is brought down
312 : * by PMD_SIZE below? no, end can't go down to 0 there.
313 : *
314 : * Whereas we round start (addr) and ceiling down, by different
315 : * masks at different levels, in order to test whether a table
316 : * now has no other vmas using it, so can be freed, we don't
317 : * bother to round floor or end up - the tests don't need that.
318 : */
319 :
320 0 : addr &= PMD_MASK;
321 0 : if (addr < floor) {
322 0 : addr += PMD_SIZE;
323 0 : if (!addr)
324 : return;
325 : }
326 0 : if (ceiling) {
327 0 : ceiling &= PMD_MASK;
328 0 : if (!ceiling)
329 : return;
330 : }
331 0 : if (end - 1 > ceiling - 1)
332 0 : end -= PMD_SIZE;
333 0 : if (addr > end - 1)
334 : return;
335 : /*
336 : * We add page table cache pages with PAGE_SIZE,
337 : * (see pte_free_tlb()), flush the tlb if we need
338 : */
339 0 : tlb_change_page_size(tlb, PAGE_SIZE);
340 0 : pgd = pgd_offset(tlb->mm, addr);
341 : do {
342 0 : next = pgd_addr_end(addr, end);
343 0 : if (pgd_none_or_clear_bad(pgd))
344 : continue;
345 : free_p4d_range(tlb, pgd, addr, next, floor, ceiling);
346 0 : } while (pgd++, addr = next, addr != end);
347 : }
348 :
349 0 : void free_pgtables(struct mmu_gather *tlb, struct maple_tree *mt,
350 : struct vm_area_struct *vma, unsigned long floor,
351 : unsigned long ceiling)
352 : {
353 0 : MA_STATE(mas, mt, vma->vm_end, vma->vm_end);
354 :
355 : do {
356 0 : unsigned long addr = vma->vm_start;
357 : struct vm_area_struct *next;
358 :
359 : /*
360 : * Note: USER_PGTABLES_CEILING may be passed as ceiling and may
361 : * be 0. This will underflow and is okay.
362 : */
363 0 : next = mas_find(&mas, ceiling - 1);
364 :
365 : /*
366 : * Hide vma from rmap and truncate_pagecache before freeing
367 : * pgtables
368 : */
369 0 : unlink_anon_vmas(vma);
370 0 : unlink_file_vma(vma);
371 :
372 0 : if (is_vm_hugetlb_page(vma)) {
373 : hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
374 : floor, next ? next->vm_start : ceiling);
375 : } else {
376 : /*
377 : * Optimization: gather nearby vmas into one call down
378 : */
379 0 : while (next && next->vm_start <= vma->vm_end + PMD_SIZE
380 0 : && !is_vm_hugetlb_page(next)) {
381 0 : vma = next;
382 0 : next = mas_find(&mas, ceiling - 1);
383 0 : unlink_anon_vmas(vma);
384 0 : unlink_file_vma(vma);
385 : }
386 0 : free_pgd_range(tlb, addr, vma->vm_end,
387 : floor, next ? next->vm_start : ceiling);
388 : }
389 0 : vma = next;
390 0 : } while (vma);
391 0 : }
392 :
393 0 : void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte)
394 : {
395 0 : spinlock_t *ptl = pmd_lock(mm, pmd);
396 :
397 0 : if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
398 0 : mm_inc_nr_ptes(mm);
399 : /*
400 : * Ensure all pte setup (eg. pte page lock and page clearing) are
401 : * visible before the pte is made visible to other CPUs by being
402 : * put into page tables.
403 : *
404 : * The other side of the story is the pointer chasing in the page
405 : * table walking code (when walking the page table without locking;
406 : * ie. most of the time). Fortunately, these data accesses consist
407 : * of a chain of data-dependent loads, meaning most CPUs (alpha
408 : * being the notable exception) will already guarantee loads are
409 : * seen in-order. See the alpha page table accessors for the
410 : * smp_rmb() barriers in page table walking code.
411 : */
412 0 : smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
413 0 : pmd_populate(mm, pmd, *pte);
414 0 : *pte = NULL;
415 : }
416 0 : spin_unlock(ptl);
417 0 : }
418 :
419 0 : int __pte_alloc(struct mm_struct *mm, pmd_t *pmd)
420 : {
421 0 : pgtable_t new = pte_alloc_one(mm);
422 0 : if (!new)
423 : return -ENOMEM;
424 :
425 0 : pmd_install(mm, pmd, &new);
426 0 : if (new)
427 0 : pte_free(mm, new);
428 : return 0;
429 : }
430 :
431 17 : int __pte_alloc_kernel(pmd_t *pmd)
432 : {
433 34 : pte_t *new = pte_alloc_one_kernel(&init_mm);
434 17 : if (!new)
435 : return -ENOMEM;
436 :
437 17 : spin_lock(&init_mm.page_table_lock);
438 17 : if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
439 17 : smp_wmb(); /* See comment in pmd_install() */
440 17 : pmd_populate_kernel(&init_mm, pmd, new);
441 17 : new = NULL;
442 : }
443 17 : spin_unlock(&init_mm.page_table_lock);
444 17 : if (new)
445 0 : pte_free_kernel(&init_mm, new);
446 : return 0;
447 : }
448 :
449 : static inline void init_rss_vec(int *rss)
450 : {
451 0 : memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
452 : }
453 :
454 0 : static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
455 : {
456 : int i;
457 :
458 0 : if (current->mm == mm)
459 : sync_mm_rss(mm);
460 0 : for (i = 0; i < NR_MM_COUNTERS; i++)
461 0 : if (rss[i])
462 0 : add_mm_counter(mm, i, rss[i]);
463 0 : }
464 :
465 : /*
466 : * This function is called to print an error when a bad pte
467 : * is found. For example, we might have a PFN-mapped pte in
468 : * a region that doesn't allow it.
469 : *
470 : * The calling function must still handle the error.
471 : */
472 0 : static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
473 : pte_t pte, struct page *page)
474 : {
475 0 : pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
476 0 : p4d_t *p4d = p4d_offset(pgd, addr);
477 0 : pud_t *pud = pud_offset(p4d, addr);
478 0 : pmd_t *pmd = pmd_offset(pud, addr);
479 : struct address_space *mapping;
480 : pgoff_t index;
481 : static unsigned long resume;
482 : static unsigned long nr_shown;
483 : static unsigned long nr_unshown;
484 :
485 : /*
486 : * Allow a burst of 60 reports, then keep quiet for that minute;
487 : * or allow a steady drip of one report per second.
488 : */
489 0 : if (nr_shown == 60) {
490 0 : if (time_before(jiffies, resume)) {
491 0 : nr_unshown++;
492 0 : return;
493 : }
494 0 : if (nr_unshown) {
495 0 : pr_alert("BUG: Bad page map: %lu messages suppressed\n",
496 : nr_unshown);
497 0 : nr_unshown = 0;
498 : }
499 0 : nr_shown = 0;
500 : }
501 0 : if (nr_shown++ == 0)
502 0 : resume = jiffies + 60 * HZ;
503 :
504 0 : mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
505 0 : index = linear_page_index(vma, addr);
506 :
507 0 : pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
508 : current->comm,
509 : (long long)pte_val(pte), (long long)pmd_val(*pmd));
510 0 : if (page)
511 0 : dump_page(page, "bad pte");
512 0 : pr_alert("addr:%px vm_flags:%08lx anon_vma:%px mapping:%px index:%lx\n",
513 : (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
514 0 : pr_alert("file:%pD fault:%ps mmap:%ps read_folio:%ps\n",
515 : vma->vm_file,
516 : vma->vm_ops ? vma->vm_ops->fault : NULL,
517 : vma->vm_file ? vma->vm_file->f_op->mmap : NULL,
518 : mapping ? mapping->a_ops->read_folio : NULL);
519 0 : dump_stack();
520 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
521 : }
522 :
523 : /*
524 : * vm_normal_page -- This function gets the "struct page" associated with a pte.
525 : *
526 : * "Special" mappings do not wish to be associated with a "struct page" (either
527 : * it doesn't exist, or it exists but they don't want to touch it). In this
528 : * case, NULL is returned here. "Normal" mappings do have a struct page.
529 : *
530 : * There are 2 broad cases. Firstly, an architecture may define a pte_special()
531 : * pte bit, in which case this function is trivial. Secondly, an architecture
532 : * may not have a spare pte bit, which requires a more complicated scheme,
533 : * described below.
534 : *
535 : * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
536 : * special mapping (even if there are underlying and valid "struct pages").
537 : * COWed pages of a VM_PFNMAP are always normal.
538 : *
539 : * The way we recognize COWed pages within VM_PFNMAP mappings is through the
540 : * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
541 : * set, and the vm_pgoff will point to the first PFN mapped: thus every special
542 : * mapping will always honor the rule
543 : *
544 : * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
545 : *
546 : * And for normal mappings this is false.
547 : *
548 : * This restricts such mappings to be a linear translation from virtual address
549 : * to pfn. To get around this restriction, we allow arbitrary mappings so long
550 : * as the vma is not a COW mapping; in that case, we know that all ptes are
551 : * special (because none can have been COWed).
552 : *
553 : *
554 : * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
555 : *
556 : * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
557 : * page" backing, however the difference is that _all_ pages with a struct
558 : * page (that is, those where pfn_valid is true) are refcounted and considered
559 : * normal pages by the VM. The disadvantage is that pages are refcounted
560 : * (which can be slower and simply not an option for some PFNMAP users). The
561 : * advantage is that we don't have to follow the strict linearity rule of
562 : * PFNMAP mappings in order to support COWable mappings.
563 : *
564 : */
565 0 : struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
566 : pte_t pte)
567 : {
568 0 : unsigned long pfn = pte_pfn(pte);
569 :
570 : if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) {
571 : if (likely(!pte_special(pte)))
572 : goto check_pfn;
573 : if (vma->vm_ops && vma->vm_ops->find_special_page)
574 : return vma->vm_ops->find_special_page(vma, addr);
575 : if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
576 : return NULL;
577 : if (is_zero_pfn(pfn))
578 : return NULL;
579 : if (pte_devmap(pte))
580 : /*
581 : * NOTE: New users of ZONE_DEVICE will not set pte_devmap()
582 : * and will have refcounts incremented on their struct pages
583 : * when they are inserted into PTEs, thus they are safe to
584 : * return here. Legacy ZONE_DEVICE pages that set pte_devmap()
585 : * do not have refcounts. Example of legacy ZONE_DEVICE is
586 : * MEMORY_DEVICE_FS_DAX type in pmem or virtio_fs drivers.
587 : */
588 : return NULL;
589 :
590 : print_bad_pte(vma, addr, pte, NULL);
591 : return NULL;
592 : }
593 :
594 : /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */
595 :
596 0 : if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
597 0 : if (vma->vm_flags & VM_MIXEDMAP) {
598 0 : if (!pfn_valid(pfn))
599 : return NULL;
600 : goto out;
601 : } else {
602 : unsigned long off;
603 0 : off = (addr - vma->vm_start) >> PAGE_SHIFT;
604 0 : if (pfn == vma->vm_pgoff + off)
605 : return NULL;
606 0 : if (!is_cow_mapping(vma->vm_flags))
607 : return NULL;
608 : }
609 : }
610 :
611 0 : if (is_zero_pfn(pfn))
612 : return NULL;
613 :
614 : check_pfn:
615 0 : if (unlikely(pfn > highest_memmap_pfn)) {
616 0 : print_bad_pte(vma, addr, pte, NULL);
617 0 : return NULL;
618 : }
619 :
620 : /*
621 : * NOTE! We still have PageReserved() pages in the page tables.
622 : * eg. VDSO mappings can cause them to exist.
623 : */
624 : out:
625 0 : return pfn_to_page(pfn);
626 : }
627 :
628 0 : struct folio *vm_normal_folio(struct vm_area_struct *vma, unsigned long addr,
629 : pte_t pte)
630 : {
631 0 : struct page *page = vm_normal_page(vma, addr, pte);
632 :
633 0 : if (page)
634 0 : return page_folio(page);
635 : return NULL;
636 : }
637 :
638 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
639 : struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr,
640 : pmd_t pmd)
641 : {
642 : unsigned long pfn = pmd_pfn(pmd);
643 :
644 : /*
645 : * There is no pmd_special() but there may be special pmds, e.g.
646 : * in a direct-access (dax) mapping, so let's just replicate the
647 : * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here.
648 : */
649 : if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
650 : if (vma->vm_flags & VM_MIXEDMAP) {
651 : if (!pfn_valid(pfn))
652 : return NULL;
653 : goto out;
654 : } else {
655 : unsigned long off;
656 : off = (addr - vma->vm_start) >> PAGE_SHIFT;
657 : if (pfn == vma->vm_pgoff + off)
658 : return NULL;
659 : if (!is_cow_mapping(vma->vm_flags))
660 : return NULL;
661 : }
662 : }
663 :
664 : if (pmd_devmap(pmd))
665 : return NULL;
666 : if (is_huge_zero_pmd(pmd))
667 : return NULL;
668 : if (unlikely(pfn > highest_memmap_pfn))
669 : return NULL;
670 :
671 : /*
672 : * NOTE! We still have PageReserved() pages in the page tables.
673 : * eg. VDSO mappings can cause them to exist.
674 : */
675 : out:
676 : return pfn_to_page(pfn);
677 : }
678 : #endif
679 :
680 : static void restore_exclusive_pte(struct vm_area_struct *vma,
681 : struct page *page, unsigned long address,
682 : pte_t *ptep)
683 : {
684 : pte_t pte;
685 : swp_entry_t entry;
686 :
687 : pte = pte_mkold(mk_pte(page, READ_ONCE(vma->vm_page_prot)));
688 : if (pte_swp_soft_dirty(*ptep))
689 : pte = pte_mksoft_dirty(pte);
690 :
691 : entry = pte_to_swp_entry(*ptep);
692 : if (pte_swp_uffd_wp(*ptep))
693 : pte = pte_mkuffd_wp(pte);
694 : else if (is_writable_device_exclusive_entry(entry))
695 : pte = maybe_mkwrite(pte_mkdirty(pte), vma);
696 :
697 : VM_BUG_ON(pte_write(pte) && !(PageAnon(page) && PageAnonExclusive(page)));
698 :
699 : /*
700 : * No need to take a page reference as one was already
701 : * created when the swap entry was made.
702 : */
703 : if (PageAnon(page))
704 : page_add_anon_rmap(page, vma, address, RMAP_NONE);
705 : else
706 : /*
707 : * Currently device exclusive access only supports anonymous
708 : * memory so the entry shouldn't point to a filebacked page.
709 : */
710 : WARN_ON_ONCE(1);
711 :
712 : set_pte_at(vma->vm_mm, address, ptep, pte);
713 :
714 : /*
715 : * No need to invalidate - it was non-present before. However
716 : * secondary CPUs may have mappings that need invalidating.
717 : */
718 : update_mmu_cache(vma, address, ptep);
719 : }
720 :
721 : /*
722 : * Tries to restore an exclusive pte if the page lock can be acquired without
723 : * sleeping.
724 : */
725 : static int
726 : try_restore_exclusive_pte(pte_t *src_pte, struct vm_area_struct *vma,
727 : unsigned long addr)
728 : {
729 : swp_entry_t entry = pte_to_swp_entry(*src_pte);
730 : struct page *page = pfn_swap_entry_to_page(entry);
731 :
732 : if (trylock_page(page)) {
733 : restore_exclusive_pte(vma, page, addr, src_pte);
734 : unlock_page(page);
735 : return 0;
736 : }
737 :
738 : return -EBUSY;
739 : }
740 :
741 : /*
742 : * copy one vm_area from one task to the other. Assumes the page tables
743 : * already present in the new task to be cleared in the whole range
744 : * covered by this vma.
745 : */
746 :
747 : static unsigned long
748 0 : copy_nonpresent_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
749 : pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *dst_vma,
750 : struct vm_area_struct *src_vma, unsigned long addr, int *rss)
751 : {
752 0 : unsigned long vm_flags = dst_vma->vm_flags;
753 0 : pte_t pte = *src_pte;
754 : struct page *page;
755 0 : swp_entry_t entry = pte_to_swp_entry(pte);
756 :
757 0 : if (likely(!non_swap_entry(entry))) {
758 0 : if (swap_duplicate(entry) < 0)
759 : return -EIO;
760 :
761 : /* make sure dst_mm is on swapoff's mmlist. */
762 0 : if (unlikely(list_empty(&dst_mm->mmlist))) {
763 0 : spin_lock(&mmlist_lock);
764 0 : if (list_empty(&dst_mm->mmlist))
765 0 : list_add(&dst_mm->mmlist,
766 : &src_mm->mmlist);
767 : spin_unlock(&mmlist_lock);
768 : }
769 : /* Mark the swap entry as shared. */
770 0 : if (pte_swp_exclusive(*src_pte)) {
771 0 : pte = pte_swp_clear_exclusive(*src_pte);
772 0 : set_pte_at(src_mm, addr, src_pte, pte);
773 : }
774 0 : rss[MM_SWAPENTS]++;
775 0 : } else if (is_migration_entry(entry)) {
776 0 : page = pfn_swap_entry_to_page(entry);
777 :
778 0 : rss[mm_counter(page)]++;
779 :
780 0 : if (!is_readable_migration_entry(entry) &&
781 0 : is_cow_mapping(vm_flags)) {
782 : /*
783 : * COW mappings require pages in both parent and child
784 : * to be set to read. A previously exclusive entry is
785 : * now shared.
786 : */
787 0 : entry = make_readable_migration_entry(
788 : swp_offset(entry));
789 0 : pte = swp_entry_to_pte(entry);
790 0 : if (pte_swp_soft_dirty(*src_pte))
791 : pte = pte_swp_mksoft_dirty(pte);
792 : if (pte_swp_uffd_wp(*src_pte))
793 : pte = pte_swp_mkuffd_wp(pte);
794 0 : set_pte_at(src_mm, addr, src_pte, pte);
795 : }
796 0 : } else if (is_device_private_entry(entry)) {
797 : page = pfn_swap_entry_to_page(entry);
798 :
799 : /*
800 : * Update rss count even for unaddressable pages, as
801 : * they should treated just like normal pages in this
802 : * respect.
803 : *
804 : * We will likely want to have some new rss counters
805 : * for unaddressable pages, at some point. But for now
806 : * keep things as they are.
807 : */
808 : get_page(page);
809 : rss[mm_counter(page)]++;
810 : /* Cannot fail as these pages cannot get pinned. */
811 : BUG_ON(page_try_dup_anon_rmap(page, false, src_vma));
812 :
813 : /*
814 : * We do not preserve soft-dirty information, because so
815 : * far, checkpoint/restore is the only feature that
816 : * requires that. And checkpoint/restore does not work
817 : * when a device driver is involved (you cannot easily
818 : * save and restore device driver state).
819 : */
820 : if (is_writable_device_private_entry(entry) &&
821 : is_cow_mapping(vm_flags)) {
822 : entry = make_readable_device_private_entry(
823 : swp_offset(entry));
824 : pte = swp_entry_to_pte(entry);
825 : if (pte_swp_uffd_wp(*src_pte))
826 : pte = pte_swp_mkuffd_wp(pte);
827 : set_pte_at(src_mm, addr, src_pte, pte);
828 : }
829 0 : } else if (is_device_exclusive_entry(entry)) {
830 : /*
831 : * Make device exclusive entries present by restoring the
832 : * original entry then copying as for a present pte. Device
833 : * exclusive entries currently only support private writable
834 : * (ie. COW) mappings.
835 : */
836 : VM_BUG_ON(!is_cow_mapping(src_vma->vm_flags));
837 : if (try_restore_exclusive_pte(src_pte, src_vma, addr))
838 : return -EBUSY;
839 : return -ENOENT;
840 0 : } else if (is_pte_marker_entry(entry)) {
841 0 : if (is_swapin_error_entry(entry) || userfaultfd_wp(dst_vma))
842 0 : set_pte_at(dst_mm, addr, dst_pte, pte);
843 : return 0;
844 : }
845 0 : if (!userfaultfd_wp(dst_vma))
846 : pte = pte_swp_clear_uffd_wp(pte);
847 0 : set_pte_at(dst_mm, addr, dst_pte, pte);
848 : return 0;
849 : }
850 :
851 : /*
852 : * Copy a present and normal page.
853 : *
854 : * NOTE! The usual case is that this isn't required;
855 : * instead, the caller can just increase the page refcount
856 : * and re-use the pte the traditional way.
857 : *
858 : * And if we need a pre-allocated page but don't yet have
859 : * one, return a negative error to let the preallocation
860 : * code know so that it can do so outside the page table
861 : * lock.
862 : */
863 : static inline int
864 0 : copy_present_page(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma,
865 : pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss,
866 : struct folio **prealloc, struct page *page)
867 : {
868 : struct folio *new_folio;
869 : pte_t pte;
870 :
871 0 : new_folio = *prealloc;
872 0 : if (!new_folio)
873 : return -EAGAIN;
874 :
875 : /*
876 : * We have a prealloc page, all good! Take it
877 : * over and copy the page & arm it.
878 : */
879 0 : *prealloc = NULL;
880 0 : copy_user_highpage(&new_folio->page, page, addr, src_vma);
881 0 : __folio_mark_uptodate(new_folio);
882 0 : folio_add_new_anon_rmap(new_folio, dst_vma, addr);
883 0 : folio_add_lru_vma(new_folio, dst_vma);
884 0 : rss[MM_ANONPAGES]++;
885 :
886 : /* All done, just insert the new page copy in the child */
887 0 : pte = mk_pte(&new_folio->page, dst_vma->vm_page_prot);
888 0 : pte = maybe_mkwrite(pte_mkdirty(pte), dst_vma);
889 0 : if (userfaultfd_pte_wp(dst_vma, *src_pte))
890 : /* Uffd-wp needs to be delivered to dest pte as well */
891 : pte = pte_mkuffd_wp(pte);
892 0 : set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte);
893 : return 0;
894 : }
895 :
896 : /*
897 : * Copy one pte. Returns 0 if succeeded, or -EAGAIN if one preallocated page
898 : * is required to copy this pte.
899 : */
900 : static inline int
901 0 : copy_present_pte(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma,
902 : pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss,
903 : struct folio **prealloc)
904 : {
905 0 : struct mm_struct *src_mm = src_vma->vm_mm;
906 0 : unsigned long vm_flags = src_vma->vm_flags;
907 0 : pte_t pte = *src_pte;
908 : struct page *page;
909 : struct folio *folio;
910 :
911 0 : page = vm_normal_page(src_vma, addr, pte);
912 0 : if (page)
913 0 : folio = page_folio(page);
914 0 : if (page && folio_test_anon(folio)) {
915 : /*
916 : * If this page may have been pinned by the parent process,
917 : * copy the page immediately for the child so that we'll always
918 : * guarantee the pinned page won't be randomly replaced in the
919 : * future.
920 : */
921 0 : folio_get(folio);
922 0 : if (unlikely(page_try_dup_anon_rmap(page, false, src_vma))) {
923 : /* Page may be pinned, we have to copy. */
924 0 : folio_put(folio);
925 0 : return copy_present_page(dst_vma, src_vma, dst_pte, src_pte,
926 : addr, rss, prealloc, page);
927 : }
928 0 : rss[MM_ANONPAGES]++;
929 0 : } else if (page) {
930 0 : folio_get(folio);
931 0 : page_dup_file_rmap(page, false);
932 0 : rss[mm_counter_file(page)]++;
933 : }
934 :
935 : /*
936 : * If it's a COW mapping, write protect it both
937 : * in the parent and the child
938 : */
939 0 : if (is_cow_mapping(vm_flags) && pte_write(pte)) {
940 0 : ptep_set_wrprotect(src_mm, addr, src_pte);
941 : pte = pte_wrprotect(pte);
942 : }
943 : VM_BUG_ON(page && folio_test_anon(folio) && PageAnonExclusive(page));
944 :
945 : /*
946 : * If it's a shared mapping, mark it clean in
947 : * the child
948 : */
949 0 : if (vm_flags & VM_SHARED)
950 : pte = pte_mkclean(pte);
951 0 : pte = pte_mkold(pte);
952 :
953 0 : if (!userfaultfd_wp(dst_vma))
954 : pte = pte_clear_uffd_wp(pte);
955 :
956 0 : set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte);
957 : return 0;
958 : }
959 :
960 : static inline struct folio *page_copy_prealloc(struct mm_struct *src_mm,
961 : struct vm_area_struct *vma, unsigned long addr)
962 : {
963 : struct folio *new_folio;
964 :
965 0 : new_folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma, addr, false);
966 0 : if (!new_folio)
967 : return NULL;
968 :
969 0 : if (mem_cgroup_charge(new_folio, src_mm, GFP_KERNEL)) {
970 : folio_put(new_folio);
971 : return NULL;
972 : }
973 0 : cgroup_throttle_swaprate(&new_folio->page, GFP_KERNEL);
974 :
975 : return new_folio;
976 : }
977 :
978 : static int
979 0 : copy_pte_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma,
980 : pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr,
981 : unsigned long end)
982 : {
983 0 : struct mm_struct *dst_mm = dst_vma->vm_mm;
984 0 : struct mm_struct *src_mm = src_vma->vm_mm;
985 : pte_t *orig_src_pte, *orig_dst_pte;
986 : pte_t *src_pte, *dst_pte;
987 : spinlock_t *src_ptl, *dst_ptl;
988 0 : int progress, ret = 0;
989 : int rss[NR_MM_COUNTERS];
990 0 : swp_entry_t entry = (swp_entry_t){0};
991 0 : struct folio *prealloc = NULL;
992 :
993 : again:
994 0 : progress = 0;
995 0 : init_rss_vec(rss);
996 :
997 0 : dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
998 0 : if (!dst_pte) {
999 : ret = -ENOMEM;
1000 : goto out;
1001 : }
1002 0 : src_pte = pte_offset_map(src_pmd, addr);
1003 0 : src_ptl = pte_lockptr(src_mm, src_pmd);
1004 0 : spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
1005 0 : orig_src_pte = src_pte;
1006 0 : orig_dst_pte = dst_pte;
1007 : arch_enter_lazy_mmu_mode();
1008 :
1009 : do {
1010 : /*
1011 : * We are holding two locks at this point - either of them
1012 : * could generate latencies in another task on another CPU.
1013 : */
1014 0 : if (progress >= 32) {
1015 0 : progress = 0;
1016 0 : if (need_resched() ||
1017 : spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
1018 : break;
1019 : }
1020 0 : if (pte_none(*src_pte)) {
1021 0 : progress++;
1022 0 : continue;
1023 : }
1024 0 : if (unlikely(!pte_present(*src_pte))) {
1025 0 : ret = copy_nonpresent_pte(dst_mm, src_mm,
1026 : dst_pte, src_pte,
1027 : dst_vma, src_vma,
1028 : addr, rss);
1029 0 : if (ret == -EIO) {
1030 : entry = pte_to_swp_entry(*src_pte);
1031 : break;
1032 0 : } else if (ret == -EBUSY) {
1033 : break;
1034 0 : } else if (!ret) {
1035 0 : progress += 8;
1036 0 : continue;
1037 : }
1038 :
1039 : /*
1040 : * Device exclusive entry restored, continue by copying
1041 : * the now present pte.
1042 : */
1043 0 : WARN_ON_ONCE(ret != -ENOENT);
1044 : }
1045 : /* copy_present_pte() will clear `*prealloc' if consumed */
1046 0 : ret = copy_present_pte(dst_vma, src_vma, dst_pte, src_pte,
1047 : addr, rss, &prealloc);
1048 : /*
1049 : * If we need a pre-allocated page for this pte, drop the
1050 : * locks, allocate, and try again.
1051 : */
1052 0 : if (unlikely(ret == -EAGAIN))
1053 : break;
1054 0 : if (unlikely(prealloc)) {
1055 : /*
1056 : * pre-alloc page cannot be reused by next time so as
1057 : * to strictly follow mempolicy (e.g., alloc_page_vma()
1058 : * will allocate page according to address). This
1059 : * could only happen if one pinned pte changed.
1060 : */
1061 0 : folio_put(prealloc);
1062 0 : prealloc = NULL;
1063 : }
1064 0 : progress += 8;
1065 0 : } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
1066 :
1067 : arch_leave_lazy_mmu_mode();
1068 0 : spin_unlock(src_ptl);
1069 : pte_unmap(orig_src_pte);
1070 0 : add_mm_rss_vec(dst_mm, rss);
1071 0 : pte_unmap_unlock(orig_dst_pte, dst_ptl);
1072 0 : cond_resched();
1073 :
1074 0 : if (ret == -EIO) {
1075 : VM_WARN_ON_ONCE(!entry.val);
1076 0 : if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) {
1077 : ret = -ENOMEM;
1078 : goto out;
1079 : }
1080 0 : entry.val = 0;
1081 0 : } else if (ret == -EBUSY) {
1082 : goto out;
1083 0 : } else if (ret == -EAGAIN) {
1084 0 : prealloc = page_copy_prealloc(src_mm, src_vma, addr);
1085 0 : if (!prealloc)
1086 : return -ENOMEM;
1087 : } else if (ret) {
1088 : VM_WARN_ON_ONCE(1);
1089 : }
1090 :
1091 : /* We've captured and resolved the error. Reset, try again. */
1092 0 : ret = 0;
1093 :
1094 0 : if (addr != end)
1095 : goto again;
1096 : out:
1097 0 : if (unlikely(prealloc))
1098 0 : folio_put(prealloc);
1099 : return ret;
1100 : }
1101 :
1102 : static inline int
1103 0 : copy_pmd_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma,
1104 : pud_t *dst_pud, pud_t *src_pud, unsigned long addr,
1105 : unsigned long end)
1106 : {
1107 0 : struct mm_struct *dst_mm = dst_vma->vm_mm;
1108 0 : struct mm_struct *src_mm = src_vma->vm_mm;
1109 : pmd_t *src_pmd, *dst_pmd;
1110 : unsigned long next;
1111 :
1112 0 : dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
1113 0 : if (!dst_pmd)
1114 : return -ENOMEM;
1115 0 : src_pmd = pmd_offset(src_pud, addr);
1116 : do {
1117 0 : next = pmd_addr_end(addr, end);
1118 0 : if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd)
1119 0 : || pmd_devmap(*src_pmd)) {
1120 : int err;
1121 : VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, src_vma);
1122 : err = copy_huge_pmd(dst_mm, src_mm, dst_pmd, src_pmd,
1123 : addr, dst_vma, src_vma);
1124 : if (err == -ENOMEM)
1125 : return -ENOMEM;
1126 : if (!err)
1127 : continue;
1128 : /* fall through */
1129 : }
1130 0 : if (pmd_none_or_clear_bad(src_pmd))
1131 0 : continue;
1132 0 : if (copy_pte_range(dst_vma, src_vma, dst_pmd, src_pmd,
1133 : addr, next))
1134 : return -ENOMEM;
1135 0 : } while (dst_pmd++, src_pmd++, addr = next, addr != end);
1136 : return 0;
1137 : }
1138 :
1139 : static inline int
1140 0 : copy_pud_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma,
1141 : p4d_t *dst_p4d, p4d_t *src_p4d, unsigned long addr,
1142 : unsigned long end)
1143 : {
1144 0 : struct mm_struct *dst_mm = dst_vma->vm_mm;
1145 0 : struct mm_struct *src_mm = src_vma->vm_mm;
1146 : pud_t *src_pud, *dst_pud;
1147 : unsigned long next;
1148 :
1149 0 : dst_pud = pud_alloc(dst_mm, dst_p4d, addr);
1150 0 : if (!dst_pud)
1151 : return -ENOMEM;
1152 0 : src_pud = pud_offset(src_p4d, addr);
1153 : do {
1154 0 : next = pud_addr_end(addr, end);
1155 0 : if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) {
1156 : int err;
1157 :
1158 : VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, src_vma);
1159 : err = copy_huge_pud(dst_mm, src_mm,
1160 : dst_pud, src_pud, addr, src_vma);
1161 : if (err == -ENOMEM)
1162 : return -ENOMEM;
1163 : if (!err)
1164 : continue;
1165 : /* fall through */
1166 : }
1167 0 : if (pud_none_or_clear_bad(src_pud))
1168 0 : continue;
1169 0 : if (copy_pmd_range(dst_vma, src_vma, dst_pud, src_pud,
1170 : addr, next))
1171 : return -ENOMEM;
1172 0 : } while (dst_pud++, src_pud++, addr = next, addr != end);
1173 0 : return 0;
1174 : }
1175 :
1176 : static inline int
1177 : copy_p4d_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma,
1178 : pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long addr,
1179 : unsigned long end)
1180 : {
1181 0 : struct mm_struct *dst_mm = dst_vma->vm_mm;
1182 : p4d_t *src_p4d, *dst_p4d;
1183 : unsigned long next;
1184 :
1185 0 : dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr);
1186 0 : if (!dst_p4d)
1187 : return -ENOMEM;
1188 0 : src_p4d = p4d_offset(src_pgd, addr);
1189 : do {
1190 0 : next = p4d_addr_end(addr, end);
1191 0 : if (p4d_none_or_clear_bad(src_p4d))
1192 : continue;
1193 0 : if (copy_pud_range(dst_vma, src_vma, dst_p4d, src_p4d,
1194 : addr, next))
1195 : return -ENOMEM;
1196 0 : } while (dst_p4d++, src_p4d++, addr = next, addr != end);
1197 : return 0;
1198 : }
1199 :
1200 : /*
1201 : * Return true if the vma needs to copy the pgtable during this fork(). Return
1202 : * false when we can speed up fork() by allowing lazy page faults later until
1203 : * when the child accesses the memory range.
1204 : */
1205 : static bool
1206 : vma_needs_copy(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma)
1207 : {
1208 : /*
1209 : * Always copy pgtables when dst_vma has uffd-wp enabled even if it's
1210 : * file-backed (e.g. shmem). Because when uffd-wp is enabled, pgtable
1211 : * contains uffd-wp protection information, that's something we can't
1212 : * retrieve from page cache, and skip copying will lose those info.
1213 : */
1214 0 : if (userfaultfd_wp(dst_vma))
1215 : return true;
1216 :
1217 0 : if (src_vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
1218 : return true;
1219 :
1220 0 : if (src_vma->anon_vma)
1221 : return true;
1222 :
1223 : /*
1224 : * Don't copy ptes where a page fault will fill them correctly. Fork
1225 : * becomes much lighter when there are big shared or private readonly
1226 : * mappings. The tradeoff is that copy_page_range is more efficient
1227 : * than faulting.
1228 : */
1229 : return false;
1230 : }
1231 :
1232 : int
1233 0 : copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma)
1234 : {
1235 : pgd_t *src_pgd, *dst_pgd;
1236 : unsigned long next;
1237 0 : unsigned long addr = src_vma->vm_start;
1238 0 : unsigned long end = src_vma->vm_end;
1239 0 : struct mm_struct *dst_mm = dst_vma->vm_mm;
1240 0 : struct mm_struct *src_mm = src_vma->vm_mm;
1241 : struct mmu_notifier_range range;
1242 : bool is_cow;
1243 : int ret;
1244 :
1245 0 : if (!vma_needs_copy(dst_vma, src_vma))
1246 : return 0;
1247 :
1248 0 : if (is_vm_hugetlb_page(src_vma))
1249 : return copy_hugetlb_page_range(dst_mm, src_mm, dst_vma, src_vma);
1250 :
1251 : if (unlikely(src_vma->vm_flags & VM_PFNMAP)) {
1252 : /*
1253 : * We do not free on error cases below as remove_vma
1254 : * gets called on error from higher level routine
1255 : */
1256 : ret = track_pfn_copy(src_vma);
1257 : if (ret)
1258 : return ret;
1259 : }
1260 :
1261 : /*
1262 : * We need to invalidate the secondary MMU mappings only when
1263 : * there could be a permission downgrade on the ptes of the
1264 : * parent mm. And a permission downgrade will only happen if
1265 : * is_cow_mapping() returns true.
1266 : */
1267 0 : is_cow = is_cow_mapping(src_vma->vm_flags);
1268 :
1269 0 : if (is_cow) {
1270 0 : mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_PAGE,
1271 : 0, src_mm, addr, end);
1272 0 : mmu_notifier_invalidate_range_start(&range);
1273 : /*
1274 : * Disabling preemption is not needed for the write side, as
1275 : * the read side doesn't spin, but goes to the mmap_lock.
1276 : *
1277 : * Use the raw variant of the seqcount_t write API to avoid
1278 : * lockdep complaining about preemptibility.
1279 : */
1280 0 : mmap_assert_write_locked(src_mm);
1281 0 : raw_write_seqcount_begin(&src_mm->write_protect_seq);
1282 : }
1283 :
1284 0 : ret = 0;
1285 0 : dst_pgd = pgd_offset(dst_mm, addr);
1286 0 : src_pgd = pgd_offset(src_mm, addr);
1287 : do {
1288 0 : next = pgd_addr_end(addr, end);
1289 0 : if (pgd_none_or_clear_bad(src_pgd))
1290 : continue;
1291 0 : if (unlikely(copy_p4d_range(dst_vma, src_vma, dst_pgd, src_pgd,
1292 : addr, next))) {
1293 : ret = -ENOMEM;
1294 : break;
1295 : }
1296 0 : } while (dst_pgd++, src_pgd++, addr = next, addr != end);
1297 :
1298 0 : if (is_cow) {
1299 0 : raw_write_seqcount_end(&src_mm->write_protect_seq);
1300 0 : mmu_notifier_invalidate_range_end(&range);
1301 : }
1302 : return ret;
1303 : }
1304 :
1305 : /* Whether we should zap all COWed (private) pages too */
1306 : static inline bool should_zap_cows(struct zap_details *details)
1307 : {
1308 : /* By default, zap all pages */
1309 0 : if (!details)
1310 : return true;
1311 :
1312 : /* Or, we zap COWed pages only if the caller wants to */
1313 0 : return details->even_cows;
1314 : }
1315 :
1316 : /* Decides whether we should zap this page with the page pointer specified */
1317 0 : static inline bool should_zap_page(struct zap_details *details, struct page *page)
1318 : {
1319 : /* If we can make a decision without *page.. */
1320 0 : if (should_zap_cows(details))
1321 : return true;
1322 :
1323 : /* E.g. the caller passes NULL for the case of a zero page */
1324 0 : if (!page)
1325 : return true;
1326 :
1327 : /* Otherwise we should only zap non-anon pages */
1328 0 : return !PageAnon(page);
1329 : }
1330 :
1331 : static inline bool zap_drop_file_uffd_wp(struct zap_details *details)
1332 : {
1333 : if (!details)
1334 : return false;
1335 :
1336 : return details->zap_flags & ZAP_FLAG_DROP_MARKER;
1337 : }
1338 :
1339 : /*
1340 : * This function makes sure that we'll replace the none pte with an uffd-wp
1341 : * swap special pte marker when necessary. Must be with the pgtable lock held.
1342 : */
1343 : static inline void
1344 : zap_install_uffd_wp_if_needed(struct vm_area_struct *vma,
1345 : unsigned long addr, pte_t *pte,
1346 : struct zap_details *details, pte_t pteval)
1347 : {
1348 0 : if (zap_drop_file_uffd_wp(details))
1349 : return;
1350 :
1351 : pte_install_uffd_wp_if_needed(vma, addr, pte, pteval);
1352 : }
1353 :
1354 0 : static unsigned long zap_pte_range(struct mmu_gather *tlb,
1355 : struct vm_area_struct *vma, pmd_t *pmd,
1356 : unsigned long addr, unsigned long end,
1357 : struct zap_details *details)
1358 : {
1359 0 : struct mm_struct *mm = tlb->mm;
1360 0 : int force_flush = 0;
1361 : int rss[NR_MM_COUNTERS];
1362 : spinlock_t *ptl;
1363 : pte_t *start_pte;
1364 : pte_t *pte;
1365 : swp_entry_t entry;
1366 :
1367 0 : tlb_change_page_size(tlb, PAGE_SIZE);
1368 : again:
1369 0 : init_rss_vec(rss);
1370 0 : start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
1371 0 : pte = start_pte;
1372 0 : flush_tlb_batched_pending(mm);
1373 : arch_enter_lazy_mmu_mode();
1374 : do {
1375 0 : pte_t ptent = *pte;
1376 : struct page *page;
1377 :
1378 0 : if (pte_none(ptent))
1379 0 : continue;
1380 :
1381 0 : if (need_resched())
1382 : break;
1383 :
1384 0 : if (pte_present(ptent)) {
1385 : unsigned int delay_rmap;
1386 :
1387 0 : page = vm_normal_page(vma, addr, ptent);
1388 0 : if (unlikely(!should_zap_page(details, page)))
1389 0 : continue;
1390 0 : ptent = ptep_get_and_clear_full(mm, addr, pte,
1391 0 : tlb->fullmm);
1392 0 : tlb_remove_tlb_entry(tlb, pte, addr);
1393 0 : zap_install_uffd_wp_if_needed(vma, addr, pte, details,
1394 : ptent);
1395 0 : if (unlikely(!page))
1396 0 : continue;
1397 :
1398 0 : delay_rmap = 0;
1399 0 : if (!PageAnon(page)) {
1400 0 : if (pte_dirty(ptent)) {
1401 0 : set_page_dirty(page);
1402 : if (tlb_delay_rmap(tlb)) {
1403 : delay_rmap = 1;
1404 : force_flush = 1;
1405 : }
1406 : }
1407 0 : if (pte_young(ptent) && likely(vma_has_recency(vma)))
1408 0 : mark_page_accessed(page);
1409 : }
1410 0 : rss[mm_counter(page)]--;
1411 : if (!delay_rmap) {
1412 0 : page_remove_rmap(page, vma, false);
1413 0 : if (unlikely(page_mapcount(page) < 0))
1414 0 : print_bad_pte(vma, addr, ptent, page);
1415 : }
1416 0 : if (unlikely(__tlb_remove_page(tlb, page, delay_rmap))) {
1417 : force_flush = 1;
1418 : addr += PAGE_SIZE;
1419 : break;
1420 : }
1421 0 : continue;
1422 : }
1423 :
1424 0 : entry = pte_to_swp_entry(ptent);
1425 0 : if (is_device_private_entry(entry) ||
1426 0 : is_device_exclusive_entry(entry)) {
1427 : page = pfn_swap_entry_to_page(entry);
1428 : if (unlikely(!should_zap_page(details, page)))
1429 : continue;
1430 : /*
1431 : * Both device private/exclusive mappings should only
1432 : * work with anonymous page so far, so we don't need to
1433 : * consider uffd-wp bit when zap. For more information,
1434 : * see zap_install_uffd_wp_if_needed().
1435 : */
1436 : WARN_ON_ONCE(!vma_is_anonymous(vma));
1437 : rss[mm_counter(page)]--;
1438 : if (is_device_private_entry(entry))
1439 : page_remove_rmap(page, vma, false);
1440 : put_page(page);
1441 0 : } else if (!non_swap_entry(entry)) {
1442 : /* Genuine swap entry, hence a private anon page */
1443 0 : if (!should_zap_cows(details))
1444 0 : continue;
1445 0 : rss[MM_SWAPENTS]--;
1446 0 : if (unlikely(!free_swap_and_cache(entry)))
1447 0 : print_bad_pte(vma, addr, ptent, NULL);
1448 0 : } else if (is_migration_entry(entry)) {
1449 0 : page = pfn_swap_entry_to_page(entry);
1450 0 : if (!should_zap_page(details, page))
1451 0 : continue;
1452 0 : rss[mm_counter(page)]--;
1453 0 : } else if (pte_marker_entry_uffd_wp(entry)) {
1454 : /* Only drop the uffd-wp marker if explicitly requested */
1455 : if (!zap_drop_file_uffd_wp(details))
1456 : continue;
1457 0 : } else if (is_hwpoison_entry(entry) ||
1458 0 : is_swapin_error_entry(entry)) {
1459 0 : if (!should_zap_cows(details))
1460 0 : continue;
1461 : } else {
1462 : /* We should have covered all the swap entry types */
1463 0 : WARN_ON_ONCE(1);
1464 : }
1465 0 : pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
1466 0 : zap_install_uffd_wp_if_needed(vma, addr, pte, details, ptent);
1467 0 : } while (pte++, addr += PAGE_SIZE, addr != end);
1468 :
1469 0 : add_mm_rss_vec(mm, rss);
1470 : arch_leave_lazy_mmu_mode();
1471 :
1472 : /* Do the actual TLB flush before dropping ptl */
1473 0 : if (force_flush) {
1474 0 : tlb_flush_mmu_tlbonly(tlb);
1475 0 : tlb_flush_rmaps(tlb, vma);
1476 : }
1477 0 : pte_unmap_unlock(start_pte, ptl);
1478 :
1479 : /*
1480 : * If we forced a TLB flush (either due to running out of
1481 : * batch buffers or because we needed to flush dirty TLB
1482 : * entries before releasing the ptl), free the batched
1483 : * memory too. Restart if we didn't do everything.
1484 : */
1485 0 : if (force_flush) {
1486 0 : force_flush = 0;
1487 0 : tlb_flush_mmu(tlb);
1488 : }
1489 :
1490 0 : if (addr != end) {
1491 0 : cond_resched();
1492 0 : goto again;
1493 : }
1494 :
1495 0 : return addr;
1496 : }
1497 :
1498 0 : static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
1499 : struct vm_area_struct *vma, pud_t *pud,
1500 : unsigned long addr, unsigned long end,
1501 : struct zap_details *details)
1502 : {
1503 : pmd_t *pmd;
1504 : unsigned long next;
1505 :
1506 0 : pmd = pmd_offset(pud, addr);
1507 : do {
1508 0 : next = pmd_addr_end(addr, end);
1509 0 : if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) {
1510 : if (next - addr != HPAGE_PMD_SIZE)
1511 : __split_huge_pmd(vma, pmd, addr, false, NULL);
1512 : else if (zap_huge_pmd(tlb, vma, pmd, addr))
1513 : goto next;
1514 : /* fall through */
1515 : } else if (details && details->single_folio &&
1516 : folio_test_pmd_mappable(details->single_folio) &&
1517 : next - addr == HPAGE_PMD_SIZE && pmd_none(*pmd)) {
1518 : spinlock_t *ptl = pmd_lock(tlb->mm, pmd);
1519 : /*
1520 : * Take and drop THP pmd lock so that we cannot return
1521 : * prematurely, while zap_huge_pmd() has cleared *pmd,
1522 : * but not yet decremented compound_mapcount().
1523 : */
1524 : spin_unlock(ptl);
1525 : }
1526 :
1527 : /*
1528 : * Here there can be other concurrent MADV_DONTNEED or
1529 : * trans huge page faults running, and if the pmd is
1530 : * none or trans huge it can change under us. This is
1531 : * because MADV_DONTNEED holds the mmap_lock in read
1532 : * mode.
1533 : */
1534 0 : if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1535 : goto next;
1536 0 : next = zap_pte_range(tlb, vma, pmd, addr, next, details);
1537 : next:
1538 0 : cond_resched();
1539 0 : } while (pmd++, addr = next, addr != end);
1540 :
1541 0 : return addr;
1542 : }
1543 :
1544 0 : static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
1545 : struct vm_area_struct *vma, p4d_t *p4d,
1546 : unsigned long addr, unsigned long end,
1547 : struct zap_details *details)
1548 : {
1549 : pud_t *pud;
1550 : unsigned long next;
1551 :
1552 0 : pud = pud_offset(p4d, addr);
1553 : do {
1554 0 : next = pud_addr_end(addr, end);
1555 0 : if (pud_trans_huge(*pud) || pud_devmap(*pud)) {
1556 : if (next - addr != HPAGE_PUD_SIZE) {
1557 : mmap_assert_locked(tlb->mm);
1558 : split_huge_pud(vma, pud, addr);
1559 : } else if (zap_huge_pud(tlb, vma, pud, addr))
1560 : goto next;
1561 : /* fall through */
1562 : }
1563 0 : if (pud_none_or_clear_bad(pud))
1564 0 : continue;
1565 0 : next = zap_pmd_range(tlb, vma, pud, addr, next, details);
1566 : next:
1567 0 : cond_resched();
1568 0 : } while (pud++, addr = next, addr != end);
1569 :
1570 0 : return addr;
1571 : }
1572 :
1573 : static inline unsigned long zap_p4d_range(struct mmu_gather *tlb,
1574 : struct vm_area_struct *vma, pgd_t *pgd,
1575 : unsigned long addr, unsigned long end,
1576 : struct zap_details *details)
1577 : {
1578 : p4d_t *p4d;
1579 : unsigned long next;
1580 :
1581 : p4d = p4d_offset(pgd, addr);
1582 : do {
1583 0 : next = p4d_addr_end(addr, end);
1584 0 : if (p4d_none_or_clear_bad(p4d))
1585 : continue;
1586 0 : next = zap_pud_range(tlb, vma, p4d, addr, next, details);
1587 0 : } while (p4d++, addr = next, addr != end);
1588 :
1589 : return addr;
1590 : }
1591 :
1592 0 : void unmap_page_range(struct mmu_gather *tlb,
1593 : struct vm_area_struct *vma,
1594 : unsigned long addr, unsigned long end,
1595 : struct zap_details *details)
1596 : {
1597 : pgd_t *pgd;
1598 : unsigned long next;
1599 :
1600 0 : BUG_ON(addr >= end);
1601 0 : tlb_start_vma(tlb, vma);
1602 0 : pgd = pgd_offset(vma->vm_mm, addr);
1603 : do {
1604 0 : next = pgd_addr_end(addr, end);
1605 0 : if (pgd_none_or_clear_bad(pgd))
1606 : continue;
1607 0 : next = zap_p4d_range(tlb, vma, pgd, addr, next, details);
1608 0 : } while (pgd++, addr = next, addr != end);
1609 0 : tlb_end_vma(tlb, vma);
1610 0 : }
1611 :
1612 :
1613 0 : static void unmap_single_vma(struct mmu_gather *tlb,
1614 : struct vm_area_struct *vma, unsigned long start_addr,
1615 : unsigned long end_addr,
1616 : struct zap_details *details, bool mm_wr_locked)
1617 : {
1618 0 : unsigned long start = max(vma->vm_start, start_addr);
1619 : unsigned long end;
1620 :
1621 0 : if (start >= vma->vm_end)
1622 : return;
1623 0 : end = min(vma->vm_end, end_addr);
1624 0 : if (end <= vma->vm_start)
1625 : return;
1626 :
1627 : if (vma->vm_file)
1628 : uprobe_munmap(vma, start, end);
1629 :
1630 : if (unlikely(vma->vm_flags & VM_PFNMAP))
1631 : untrack_pfn(vma, 0, 0, mm_wr_locked);
1632 :
1633 0 : if (start != end) {
1634 0 : if (unlikely(is_vm_hugetlb_page(vma))) {
1635 : /*
1636 : * It is undesirable to test vma->vm_file as it
1637 : * should be non-null for valid hugetlb area.
1638 : * However, vm_file will be NULL in the error
1639 : * cleanup path of mmap_region. When
1640 : * hugetlbfs ->mmap method fails,
1641 : * mmap_region() nullifies vma->vm_file
1642 : * before calling this function to clean up.
1643 : * Since no pte has actually been setup, it is
1644 : * safe to do nothing in this case.
1645 : */
1646 : if (vma->vm_file) {
1647 : zap_flags_t zap_flags = details ?
1648 : details->zap_flags : 0;
1649 : __unmap_hugepage_range_final(tlb, vma, start, end,
1650 : NULL, zap_flags);
1651 : }
1652 : } else
1653 0 : unmap_page_range(tlb, vma, start, end, details);
1654 : }
1655 : }
1656 :
1657 : /**
1658 : * unmap_vmas - unmap a range of memory covered by a list of vma's
1659 : * @tlb: address of the caller's struct mmu_gather
1660 : * @mt: the maple tree
1661 : * @vma: the starting vma
1662 : * @start_addr: virtual address at which to start unmapping
1663 : * @end_addr: virtual address at which to end unmapping
1664 : *
1665 : * Unmap all pages in the vma list.
1666 : *
1667 : * Only addresses between `start' and `end' will be unmapped.
1668 : *
1669 : * The VMA list must be sorted in ascending virtual address order.
1670 : *
1671 : * unmap_vmas() assumes that the caller will flush the whole unmapped address
1672 : * range after unmap_vmas() returns. So the only responsibility here is to
1673 : * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1674 : * drops the lock and schedules.
1675 : */
1676 0 : void unmap_vmas(struct mmu_gather *tlb, struct maple_tree *mt,
1677 : struct vm_area_struct *vma, unsigned long start_addr,
1678 : unsigned long end_addr, bool mm_wr_locked)
1679 : {
1680 : struct mmu_notifier_range range;
1681 0 : struct zap_details details = {
1682 : .zap_flags = ZAP_FLAG_DROP_MARKER | ZAP_FLAG_UNMAP,
1683 : /* Careful - we need to zap private pages too! */
1684 : .even_cows = true,
1685 : };
1686 0 : MA_STATE(mas, mt, vma->vm_end, vma->vm_end);
1687 :
1688 : mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma->vm_mm,
1689 : start_addr, end_addr);
1690 : mmu_notifier_invalidate_range_start(&range);
1691 : do {
1692 0 : unmap_single_vma(tlb, vma, start_addr, end_addr, &details,
1693 : mm_wr_locked);
1694 0 : } while ((vma = mas_find(&mas, end_addr - 1)) != NULL);
1695 0 : mmu_notifier_invalidate_range_end(&range);
1696 0 : }
1697 :
1698 : /**
1699 : * zap_page_range_single - remove user pages in a given range
1700 : * @vma: vm_area_struct holding the applicable pages
1701 : * @address: starting address of pages to zap
1702 : * @size: number of bytes to zap
1703 : * @details: details of shared cache invalidation
1704 : *
1705 : * The range must fit into one VMA.
1706 : */
1707 0 : void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
1708 : unsigned long size, struct zap_details *details)
1709 : {
1710 0 : const unsigned long end = address + size;
1711 : struct mmu_notifier_range range;
1712 : struct mmu_gather tlb;
1713 :
1714 0 : lru_add_drain();
1715 0 : mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma->vm_mm,
1716 : address, end);
1717 0 : if (is_vm_hugetlb_page(vma))
1718 : adjust_range_if_pmd_sharing_possible(vma, &range.start,
1719 : &range.end);
1720 0 : tlb_gather_mmu(&tlb, vma->vm_mm);
1721 0 : update_hiwater_rss(vma->vm_mm);
1722 0 : mmu_notifier_invalidate_range_start(&range);
1723 : /*
1724 : * unmap 'address-end' not 'range.start-range.end' as range
1725 : * could have been expanded for hugetlb pmd sharing.
1726 : */
1727 0 : unmap_single_vma(&tlb, vma, address, end, details, false);
1728 0 : mmu_notifier_invalidate_range_end(&range);
1729 0 : tlb_finish_mmu(&tlb);
1730 0 : }
1731 :
1732 : /**
1733 : * zap_vma_ptes - remove ptes mapping the vma
1734 : * @vma: vm_area_struct holding ptes to be zapped
1735 : * @address: starting address of pages to zap
1736 : * @size: number of bytes to zap
1737 : *
1738 : * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1739 : *
1740 : * The entire address range must be fully contained within the vma.
1741 : *
1742 : */
1743 0 : void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1744 : unsigned long size)
1745 : {
1746 0 : if (!range_in_vma(vma, address, address + size) ||
1747 0 : !(vma->vm_flags & VM_PFNMAP))
1748 : return;
1749 :
1750 0 : zap_page_range_single(vma, address, size, NULL);
1751 : }
1752 : EXPORT_SYMBOL_GPL(zap_vma_ptes);
1753 :
1754 0 : static pmd_t *walk_to_pmd(struct mm_struct *mm, unsigned long addr)
1755 : {
1756 : pgd_t *pgd;
1757 : p4d_t *p4d;
1758 : pud_t *pud;
1759 : pmd_t *pmd;
1760 :
1761 0 : pgd = pgd_offset(mm, addr);
1762 0 : p4d = p4d_alloc(mm, pgd, addr);
1763 0 : if (!p4d)
1764 : return NULL;
1765 0 : pud = pud_alloc(mm, p4d, addr);
1766 : if (!pud)
1767 : return NULL;
1768 0 : pmd = pmd_alloc(mm, pud, addr);
1769 0 : if (!pmd)
1770 : return NULL;
1771 :
1772 : VM_BUG_ON(pmd_trans_huge(*pmd));
1773 0 : return pmd;
1774 : }
1775 :
1776 0 : pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
1777 : spinlock_t **ptl)
1778 : {
1779 0 : pmd_t *pmd = walk_to_pmd(mm, addr);
1780 :
1781 0 : if (!pmd)
1782 : return NULL;
1783 0 : return pte_alloc_map_lock(mm, pmd, addr, ptl);
1784 : }
1785 :
1786 0 : static int validate_page_before_insert(struct page *page)
1787 : {
1788 0 : if (PageAnon(page) || PageSlab(page) || page_has_type(page))
1789 : return -EINVAL;
1790 : flush_dcache_page(page);
1791 : return 0;
1792 : }
1793 :
1794 0 : static int insert_page_into_pte_locked(struct vm_area_struct *vma, pte_t *pte,
1795 : unsigned long addr, struct page *page, pgprot_t prot)
1796 : {
1797 0 : if (!pte_none(*pte))
1798 : return -EBUSY;
1799 : /* Ok, finally just insert the thing.. */
1800 0 : get_page(page);
1801 0 : inc_mm_counter(vma->vm_mm, mm_counter_file(page));
1802 0 : page_add_file_rmap(page, vma, false);
1803 0 : set_pte_at(vma->vm_mm, addr, pte, mk_pte(page, prot));
1804 : return 0;
1805 : }
1806 :
1807 : /*
1808 : * This is the old fallback for page remapping.
1809 : *
1810 : * For historical reasons, it only allows reserved pages. Only
1811 : * old drivers should use this, and they needed to mark their
1812 : * pages reserved for the old functions anyway.
1813 : */
1814 0 : static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1815 : struct page *page, pgprot_t prot)
1816 : {
1817 : int retval;
1818 : pte_t *pte;
1819 : spinlock_t *ptl;
1820 :
1821 0 : retval = validate_page_before_insert(page);
1822 0 : if (retval)
1823 : goto out;
1824 0 : retval = -ENOMEM;
1825 0 : pte = get_locked_pte(vma->vm_mm, addr, &ptl);
1826 0 : if (!pte)
1827 : goto out;
1828 0 : retval = insert_page_into_pte_locked(vma, pte, addr, page, prot);
1829 0 : pte_unmap_unlock(pte, ptl);
1830 : out:
1831 0 : return retval;
1832 : }
1833 :
1834 : #ifdef pte_index
1835 0 : static int insert_page_in_batch_locked(struct vm_area_struct *vma, pte_t *pte,
1836 : unsigned long addr, struct page *page, pgprot_t prot)
1837 : {
1838 : int err;
1839 :
1840 0 : if (!page_count(page))
1841 : return -EINVAL;
1842 0 : err = validate_page_before_insert(page);
1843 0 : if (err)
1844 : return err;
1845 0 : return insert_page_into_pte_locked(vma, pte, addr, page, prot);
1846 : }
1847 :
1848 : /* insert_pages() amortizes the cost of spinlock operations
1849 : * when inserting pages in a loop. Arch *must* define pte_index.
1850 : */
1851 0 : static int insert_pages(struct vm_area_struct *vma, unsigned long addr,
1852 : struct page **pages, unsigned long *num, pgprot_t prot)
1853 : {
1854 0 : pmd_t *pmd = NULL;
1855 : pte_t *start_pte, *pte;
1856 : spinlock_t *pte_lock;
1857 0 : struct mm_struct *const mm = vma->vm_mm;
1858 0 : unsigned long curr_page_idx = 0;
1859 0 : unsigned long remaining_pages_total = *num;
1860 : unsigned long pages_to_write_in_pmd;
1861 : int ret;
1862 : more:
1863 0 : ret = -EFAULT;
1864 0 : pmd = walk_to_pmd(mm, addr);
1865 0 : if (!pmd)
1866 : goto out;
1867 :
1868 0 : pages_to_write_in_pmd = min_t(unsigned long,
1869 : remaining_pages_total, PTRS_PER_PTE - pte_index(addr));
1870 :
1871 : /* Allocate the PTE if necessary; takes PMD lock once only. */
1872 0 : ret = -ENOMEM;
1873 0 : if (pte_alloc(mm, pmd))
1874 : goto out;
1875 :
1876 0 : while (pages_to_write_in_pmd) {
1877 0 : int pte_idx = 0;
1878 0 : const int batch_size = min_t(int, pages_to_write_in_pmd, 8);
1879 :
1880 0 : start_pte = pte_offset_map_lock(mm, pmd, addr, &pte_lock);
1881 0 : for (pte = start_pte; pte_idx < batch_size; ++pte, ++pte_idx) {
1882 0 : int err = insert_page_in_batch_locked(vma, pte,
1883 0 : addr, pages[curr_page_idx], prot);
1884 0 : if (unlikely(err)) {
1885 0 : pte_unmap_unlock(start_pte, pte_lock);
1886 0 : ret = err;
1887 0 : remaining_pages_total -= pte_idx;
1888 0 : goto out;
1889 : }
1890 0 : addr += PAGE_SIZE;
1891 0 : ++curr_page_idx;
1892 : }
1893 0 : pte_unmap_unlock(start_pte, pte_lock);
1894 0 : pages_to_write_in_pmd -= batch_size;
1895 0 : remaining_pages_total -= batch_size;
1896 : }
1897 0 : if (remaining_pages_total)
1898 : goto more;
1899 : ret = 0;
1900 : out:
1901 0 : *num = remaining_pages_total;
1902 0 : return ret;
1903 : }
1904 : #endif /* ifdef pte_index */
1905 :
1906 : /**
1907 : * vm_insert_pages - insert multiple pages into user vma, batching the pmd lock.
1908 : * @vma: user vma to map to
1909 : * @addr: target start user address of these pages
1910 : * @pages: source kernel pages
1911 : * @num: in: number of pages to map. out: number of pages that were *not*
1912 : * mapped. (0 means all pages were successfully mapped).
1913 : *
1914 : * Preferred over vm_insert_page() when inserting multiple pages.
1915 : *
1916 : * In case of error, we may have mapped a subset of the provided
1917 : * pages. It is the caller's responsibility to account for this case.
1918 : *
1919 : * The same restrictions apply as in vm_insert_page().
1920 : */
1921 0 : int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr,
1922 : struct page **pages, unsigned long *num)
1923 : {
1924 : #ifdef pte_index
1925 0 : const unsigned long end_addr = addr + (*num * PAGE_SIZE) - 1;
1926 :
1927 0 : if (addr < vma->vm_start || end_addr >= vma->vm_end)
1928 : return -EFAULT;
1929 0 : if (!(vma->vm_flags & VM_MIXEDMAP)) {
1930 0 : BUG_ON(mmap_read_trylock(vma->vm_mm));
1931 0 : BUG_ON(vma->vm_flags & VM_PFNMAP);
1932 0 : vm_flags_set(vma, VM_MIXEDMAP);
1933 : }
1934 : /* Defer page refcount checking till we're about to map that page. */
1935 0 : return insert_pages(vma, addr, pages, num, vma->vm_page_prot);
1936 : #else
1937 : unsigned long idx = 0, pgcount = *num;
1938 : int err = -EINVAL;
1939 :
1940 : for (; idx < pgcount; ++idx) {
1941 : err = vm_insert_page(vma, addr + (PAGE_SIZE * idx), pages[idx]);
1942 : if (err)
1943 : break;
1944 : }
1945 : *num = pgcount - idx;
1946 : return err;
1947 : #endif /* ifdef pte_index */
1948 : }
1949 : EXPORT_SYMBOL(vm_insert_pages);
1950 :
1951 : /**
1952 : * vm_insert_page - insert single page into user vma
1953 : * @vma: user vma to map to
1954 : * @addr: target user address of this page
1955 : * @page: source kernel page
1956 : *
1957 : * This allows drivers to insert individual pages they've allocated
1958 : * into a user vma.
1959 : *
1960 : * The page has to be a nice clean _individual_ kernel allocation.
1961 : * If you allocate a compound page, you need to have marked it as
1962 : * such (__GFP_COMP), or manually just split the page up yourself
1963 : * (see split_page()).
1964 : *
1965 : * NOTE! Traditionally this was done with "remap_pfn_range()" which
1966 : * took an arbitrary page protection parameter. This doesn't allow
1967 : * that. Your vma protection will have to be set up correctly, which
1968 : * means that if you want a shared writable mapping, you'd better
1969 : * ask for a shared writable mapping!
1970 : *
1971 : * The page does not need to be reserved.
1972 : *
1973 : * Usually this function is called from f_op->mmap() handler
1974 : * under mm->mmap_lock write-lock, so it can change vma->vm_flags.
1975 : * Caller must set VM_MIXEDMAP on vma if it wants to call this
1976 : * function from other places, for example from page-fault handler.
1977 : *
1978 : * Return: %0 on success, negative error code otherwise.
1979 : */
1980 0 : int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1981 : struct page *page)
1982 : {
1983 0 : if (addr < vma->vm_start || addr >= vma->vm_end)
1984 : return -EFAULT;
1985 0 : if (!page_count(page))
1986 : return -EINVAL;
1987 0 : if (!(vma->vm_flags & VM_MIXEDMAP)) {
1988 0 : BUG_ON(mmap_read_trylock(vma->vm_mm));
1989 0 : BUG_ON(vma->vm_flags & VM_PFNMAP);
1990 0 : vm_flags_set(vma, VM_MIXEDMAP);
1991 : }
1992 0 : return insert_page(vma, addr, page, vma->vm_page_prot);
1993 : }
1994 : EXPORT_SYMBOL(vm_insert_page);
1995 :
1996 : /*
1997 : * __vm_map_pages - maps range of kernel pages into user vma
1998 : * @vma: user vma to map to
1999 : * @pages: pointer to array of source kernel pages
2000 : * @num: number of pages in page array
2001 : * @offset: user's requested vm_pgoff
2002 : *
2003 : * This allows drivers to map range of kernel pages into a user vma.
2004 : *
2005 : * Return: 0 on success and error code otherwise.
2006 : */
2007 0 : static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages,
2008 : unsigned long num, unsigned long offset)
2009 : {
2010 0 : unsigned long count = vma_pages(vma);
2011 0 : unsigned long uaddr = vma->vm_start;
2012 : int ret, i;
2013 :
2014 : /* Fail if the user requested offset is beyond the end of the object */
2015 0 : if (offset >= num)
2016 : return -ENXIO;
2017 :
2018 : /* Fail if the user requested size exceeds available object size */
2019 0 : if (count > num - offset)
2020 : return -ENXIO;
2021 :
2022 0 : for (i = 0; i < count; i++) {
2023 0 : ret = vm_insert_page(vma, uaddr, pages[offset + i]);
2024 0 : if (ret < 0)
2025 : return ret;
2026 0 : uaddr += PAGE_SIZE;
2027 : }
2028 :
2029 : return 0;
2030 : }
2031 :
2032 : /**
2033 : * vm_map_pages - maps range of kernel pages starts with non zero offset
2034 : * @vma: user vma to map to
2035 : * @pages: pointer to array of source kernel pages
2036 : * @num: number of pages in page array
2037 : *
2038 : * Maps an object consisting of @num pages, catering for the user's
2039 : * requested vm_pgoff
2040 : *
2041 : * If we fail to insert any page into the vma, the function will return
2042 : * immediately leaving any previously inserted pages present. Callers
2043 : * from the mmap handler may immediately return the error as their caller
2044 : * will destroy the vma, removing any successfully inserted pages. Other
2045 : * callers should make their own arrangements for calling unmap_region().
2046 : *
2047 : * Context: Process context. Called by mmap handlers.
2048 : * Return: 0 on success and error code otherwise.
2049 : */
2050 0 : int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
2051 : unsigned long num)
2052 : {
2053 0 : return __vm_map_pages(vma, pages, num, vma->vm_pgoff);
2054 : }
2055 : EXPORT_SYMBOL(vm_map_pages);
2056 :
2057 : /**
2058 : * vm_map_pages_zero - map range of kernel pages starts with zero offset
2059 : * @vma: user vma to map to
2060 : * @pages: pointer to array of source kernel pages
2061 : * @num: number of pages in page array
2062 : *
2063 : * Similar to vm_map_pages(), except that it explicitly sets the offset
2064 : * to 0. This function is intended for the drivers that did not consider
2065 : * vm_pgoff.
2066 : *
2067 : * Context: Process context. Called by mmap handlers.
2068 : * Return: 0 on success and error code otherwise.
2069 : */
2070 0 : int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
2071 : unsigned long num)
2072 : {
2073 0 : return __vm_map_pages(vma, pages, num, 0);
2074 : }
2075 : EXPORT_SYMBOL(vm_map_pages_zero);
2076 :
2077 0 : static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr,
2078 : pfn_t pfn, pgprot_t prot, bool mkwrite)
2079 : {
2080 0 : struct mm_struct *mm = vma->vm_mm;
2081 : pte_t *pte, entry;
2082 : spinlock_t *ptl;
2083 :
2084 0 : pte = get_locked_pte(mm, addr, &ptl);
2085 0 : if (!pte)
2086 : return VM_FAULT_OOM;
2087 0 : if (!pte_none(*pte)) {
2088 0 : if (mkwrite) {
2089 : /*
2090 : * For read faults on private mappings the PFN passed
2091 : * in may not match the PFN we have mapped if the
2092 : * mapped PFN is a writeable COW page. In the mkwrite
2093 : * case we are creating a writable PTE for a shared
2094 : * mapping and we expect the PFNs to match. If they
2095 : * don't match, we are likely racing with block
2096 : * allocation and mapping invalidation so just skip the
2097 : * update.
2098 : */
2099 0 : if (pte_pfn(*pte) != pfn_t_to_pfn(pfn)) {
2100 0 : WARN_ON_ONCE(!is_zero_pfn(pte_pfn(*pte)));
2101 : goto out_unlock;
2102 : }
2103 0 : entry = pte_mkyoung(*pte);
2104 0 : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2105 0 : if (ptep_set_access_flags(vma, addr, pte, entry, 1))
2106 : update_mmu_cache(vma, addr, pte);
2107 : }
2108 : goto out_unlock;
2109 : }
2110 :
2111 : /* Ok, finally just insert the thing.. */
2112 0 : if (pfn_t_devmap(pfn))
2113 : entry = pte_mkdevmap(pfn_t_pte(pfn, prot));
2114 : else
2115 0 : entry = pte_mkspecial(pfn_t_pte(pfn, prot));
2116 :
2117 0 : if (mkwrite) {
2118 0 : entry = pte_mkyoung(entry);
2119 0 : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2120 : }
2121 :
2122 0 : set_pte_at(mm, addr, pte, entry);
2123 : update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
2124 :
2125 : out_unlock:
2126 0 : pte_unmap_unlock(pte, ptl);
2127 0 : return VM_FAULT_NOPAGE;
2128 : }
2129 :
2130 : /**
2131 : * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot
2132 : * @vma: user vma to map to
2133 : * @addr: target user address of this page
2134 : * @pfn: source kernel pfn
2135 : * @pgprot: pgprot flags for the inserted page
2136 : *
2137 : * This is exactly like vmf_insert_pfn(), except that it allows drivers
2138 : * to override pgprot on a per-page basis.
2139 : *
2140 : * This only makes sense for IO mappings, and it makes no sense for
2141 : * COW mappings. In general, using multiple vmas is preferable;
2142 : * vmf_insert_pfn_prot should only be used if using multiple VMAs is
2143 : * impractical.
2144 : *
2145 : * See vmf_insert_mixed_prot() for a discussion of the implication of using
2146 : * a value of @pgprot different from that of @vma->vm_page_prot.
2147 : *
2148 : * Context: Process context. May allocate using %GFP_KERNEL.
2149 : * Return: vm_fault_t value.
2150 : */
2151 0 : vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr,
2152 : unsigned long pfn, pgprot_t pgprot)
2153 : {
2154 : /*
2155 : * Technically, architectures with pte_special can avoid all these
2156 : * restrictions (same for remap_pfn_range). However we would like
2157 : * consistency in testing and feature parity among all, so we should
2158 : * try to keep these invariants in place for everybody.
2159 : */
2160 0 : BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
2161 0 : BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
2162 : (VM_PFNMAP|VM_MIXEDMAP));
2163 0 : BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
2164 0 : BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
2165 :
2166 0 : if (addr < vma->vm_start || addr >= vma->vm_end)
2167 : return VM_FAULT_SIGBUS;
2168 :
2169 0 : if (!pfn_modify_allowed(pfn, pgprot))
2170 : return VM_FAULT_SIGBUS;
2171 :
2172 0 : track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV));
2173 :
2174 0 : return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot,
2175 : false);
2176 : }
2177 : EXPORT_SYMBOL(vmf_insert_pfn_prot);
2178 :
2179 : /**
2180 : * vmf_insert_pfn - insert single pfn into user vma
2181 : * @vma: user vma to map to
2182 : * @addr: target user address of this page
2183 : * @pfn: source kernel pfn
2184 : *
2185 : * Similar to vm_insert_page, this allows drivers to insert individual pages
2186 : * they've allocated into a user vma. Same comments apply.
2187 : *
2188 : * This function should only be called from a vm_ops->fault handler, and
2189 : * in that case the handler should return the result of this function.
2190 : *
2191 : * vma cannot be a COW mapping.
2192 : *
2193 : * As this is called only for pages that do not currently exist, we
2194 : * do not need to flush old virtual caches or the TLB.
2195 : *
2196 : * Context: Process context. May allocate using %GFP_KERNEL.
2197 : * Return: vm_fault_t value.
2198 : */
2199 0 : vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
2200 : unsigned long pfn)
2201 : {
2202 0 : return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot);
2203 : }
2204 : EXPORT_SYMBOL(vmf_insert_pfn);
2205 :
2206 : static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn)
2207 : {
2208 : /* these checks mirror the abort conditions in vm_normal_page */
2209 0 : if (vma->vm_flags & VM_MIXEDMAP)
2210 : return true;
2211 0 : if (pfn_t_devmap(pfn))
2212 : return true;
2213 0 : if (pfn_t_special(pfn))
2214 : return true;
2215 0 : if (is_zero_pfn(pfn_t_to_pfn(pfn)))
2216 : return true;
2217 : return false;
2218 : }
2219 :
2220 0 : static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma,
2221 : unsigned long addr, pfn_t pfn, pgprot_t pgprot,
2222 : bool mkwrite)
2223 : {
2224 : int err;
2225 :
2226 0 : BUG_ON(!vm_mixed_ok(vma, pfn));
2227 :
2228 0 : if (addr < vma->vm_start || addr >= vma->vm_end)
2229 : return VM_FAULT_SIGBUS;
2230 :
2231 0 : track_pfn_insert(vma, &pgprot, pfn);
2232 :
2233 0 : if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot))
2234 : return VM_FAULT_SIGBUS;
2235 :
2236 : /*
2237 : * If we don't have pte special, then we have to use the pfn_valid()
2238 : * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
2239 : * refcount the page if pfn_valid is true (hence insert_page rather
2240 : * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
2241 : * without pte special, it would there be refcounted as a normal page.
2242 : */
2243 : if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) &&
2244 0 : !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) {
2245 : struct page *page;
2246 :
2247 : /*
2248 : * At this point we are committed to insert_page()
2249 : * regardless of whether the caller specified flags that
2250 : * result in pfn_t_has_page() == false.
2251 : */
2252 0 : page = pfn_to_page(pfn_t_to_pfn(pfn));
2253 0 : err = insert_page(vma, addr, page, pgprot);
2254 : } else {
2255 0 : return insert_pfn(vma, addr, pfn, pgprot, mkwrite);
2256 : }
2257 :
2258 0 : if (err == -ENOMEM)
2259 : return VM_FAULT_OOM;
2260 0 : if (err < 0 && err != -EBUSY)
2261 : return VM_FAULT_SIGBUS;
2262 :
2263 0 : return VM_FAULT_NOPAGE;
2264 : }
2265 :
2266 : /**
2267 : * vmf_insert_mixed_prot - insert single pfn into user vma with specified pgprot
2268 : * @vma: user vma to map to
2269 : * @addr: target user address of this page
2270 : * @pfn: source kernel pfn
2271 : * @pgprot: pgprot flags for the inserted page
2272 : *
2273 : * This is exactly like vmf_insert_mixed(), except that it allows drivers
2274 : * to override pgprot on a per-page basis.
2275 : *
2276 : * Typically this function should be used by drivers to set caching- and
2277 : * encryption bits different than those of @vma->vm_page_prot, because
2278 : * the caching- or encryption mode may not be known at mmap() time.
2279 : * This is ok as long as @vma->vm_page_prot is not used by the core vm
2280 : * to set caching and encryption bits for those vmas (except for COW pages).
2281 : * This is ensured by core vm only modifying these page table entries using
2282 : * functions that don't touch caching- or encryption bits, using pte_modify()
2283 : * if needed. (See for example mprotect()).
2284 : * Also when new page-table entries are created, this is only done using the
2285 : * fault() callback, and never using the value of vma->vm_page_prot,
2286 : * except for page-table entries that point to anonymous pages as the result
2287 : * of COW.
2288 : *
2289 : * Context: Process context. May allocate using %GFP_KERNEL.
2290 : * Return: vm_fault_t value.
2291 : */
2292 0 : vm_fault_t vmf_insert_mixed_prot(struct vm_area_struct *vma, unsigned long addr,
2293 : pfn_t pfn, pgprot_t pgprot)
2294 : {
2295 0 : return __vm_insert_mixed(vma, addr, pfn, pgprot, false);
2296 : }
2297 : EXPORT_SYMBOL(vmf_insert_mixed_prot);
2298 :
2299 0 : vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
2300 : pfn_t pfn)
2301 : {
2302 0 : return __vm_insert_mixed(vma, addr, pfn, vma->vm_page_prot, false);
2303 : }
2304 : EXPORT_SYMBOL(vmf_insert_mixed);
2305 :
2306 : /*
2307 : * If the insertion of PTE failed because someone else already added a
2308 : * different entry in the mean time, we treat that as success as we assume
2309 : * the same entry was actually inserted.
2310 : */
2311 0 : vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma,
2312 : unsigned long addr, pfn_t pfn)
2313 : {
2314 0 : return __vm_insert_mixed(vma, addr, pfn, vma->vm_page_prot, true);
2315 : }
2316 : EXPORT_SYMBOL(vmf_insert_mixed_mkwrite);
2317 :
2318 : /*
2319 : * maps a range of physical memory into the requested pages. the old
2320 : * mappings are removed. any references to nonexistent pages results
2321 : * in null mappings (currently treated as "copy-on-access")
2322 : */
2323 0 : static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
2324 : unsigned long addr, unsigned long end,
2325 : unsigned long pfn, pgprot_t prot)
2326 : {
2327 : pte_t *pte, *mapped_pte;
2328 : spinlock_t *ptl;
2329 0 : int err = 0;
2330 :
2331 0 : mapped_pte = pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
2332 0 : if (!pte)
2333 : return -ENOMEM;
2334 : arch_enter_lazy_mmu_mode();
2335 : do {
2336 0 : BUG_ON(!pte_none(*pte));
2337 0 : if (!pfn_modify_allowed(pfn, prot)) {
2338 : err = -EACCES;
2339 : break;
2340 : }
2341 0 : set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
2342 0 : pfn++;
2343 0 : } while (pte++, addr += PAGE_SIZE, addr != end);
2344 : arch_leave_lazy_mmu_mode();
2345 0 : pte_unmap_unlock(mapped_pte, ptl);
2346 0 : return err;
2347 : }
2348 :
2349 0 : static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
2350 : unsigned long addr, unsigned long end,
2351 : unsigned long pfn, pgprot_t prot)
2352 : {
2353 : pmd_t *pmd;
2354 : unsigned long next;
2355 : int err;
2356 :
2357 0 : pfn -= addr >> PAGE_SHIFT;
2358 0 : pmd = pmd_alloc(mm, pud, addr);
2359 0 : if (!pmd)
2360 : return -ENOMEM;
2361 : VM_BUG_ON(pmd_trans_huge(*pmd));
2362 : do {
2363 0 : next = pmd_addr_end(addr, end);
2364 0 : err = remap_pte_range(mm, pmd, addr, next,
2365 0 : pfn + (addr >> PAGE_SHIFT), prot);
2366 0 : if (err)
2367 : return err;
2368 0 : } while (pmd++, addr = next, addr != end);
2369 : return 0;
2370 : }
2371 :
2372 : static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d,
2373 : unsigned long addr, unsigned long end,
2374 : unsigned long pfn, pgprot_t prot)
2375 : {
2376 : pud_t *pud;
2377 : unsigned long next;
2378 : int err;
2379 :
2380 0 : pfn -= addr >> PAGE_SHIFT;
2381 0 : pud = pud_alloc(mm, p4d, addr);
2382 : if (!pud)
2383 : return -ENOMEM;
2384 : do {
2385 0 : next = pud_addr_end(addr, end);
2386 0 : err = remap_pmd_range(mm, pud, addr, next,
2387 : pfn + (addr >> PAGE_SHIFT), prot);
2388 0 : if (err)
2389 : return err;
2390 0 : } while (pud++, addr = next, addr != end);
2391 : return 0;
2392 : }
2393 :
2394 0 : static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd,
2395 : unsigned long addr, unsigned long end,
2396 : unsigned long pfn, pgprot_t prot)
2397 : {
2398 : p4d_t *p4d;
2399 : unsigned long next;
2400 : int err;
2401 :
2402 0 : pfn -= addr >> PAGE_SHIFT;
2403 0 : p4d = p4d_alloc(mm, pgd, addr);
2404 0 : if (!p4d)
2405 : return -ENOMEM;
2406 : do {
2407 0 : next = p4d_addr_end(addr, end);
2408 0 : err = remap_pud_range(mm, p4d, addr, next,
2409 : pfn + (addr >> PAGE_SHIFT), prot);
2410 0 : if (err)
2411 : return err;
2412 0 : } while (p4d++, addr = next, addr != end);
2413 0 : return 0;
2414 : }
2415 :
2416 : /*
2417 : * Variant of remap_pfn_range that does not call track_pfn_remap. The caller
2418 : * must have pre-validated the caching bits of the pgprot_t.
2419 : */
2420 0 : int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr,
2421 : unsigned long pfn, unsigned long size, pgprot_t prot)
2422 : {
2423 : pgd_t *pgd;
2424 : unsigned long next;
2425 0 : unsigned long end = addr + PAGE_ALIGN(size);
2426 0 : struct mm_struct *mm = vma->vm_mm;
2427 : int err;
2428 :
2429 0 : if (WARN_ON_ONCE(!PAGE_ALIGNED(addr)))
2430 : return -EINVAL;
2431 :
2432 : /*
2433 : * Physically remapped pages are special. Tell the
2434 : * rest of the world about it:
2435 : * VM_IO tells people not to look at these pages
2436 : * (accesses can have side effects).
2437 : * VM_PFNMAP tells the core MM that the base pages are just
2438 : * raw PFN mappings, and do not have a "struct page" associated
2439 : * with them.
2440 : * VM_DONTEXPAND
2441 : * Disable vma merging and expanding with mremap().
2442 : * VM_DONTDUMP
2443 : * Omit vma from core dump, even when VM_IO turned off.
2444 : *
2445 : * There's a horrible special case to handle copy-on-write
2446 : * behaviour that some programs depend on. We mark the "original"
2447 : * un-COW'ed pages by matching them up with "vma->vm_pgoff".
2448 : * See vm_normal_page() for details.
2449 : */
2450 0 : if (is_cow_mapping(vma->vm_flags)) {
2451 0 : if (addr != vma->vm_start || end != vma->vm_end)
2452 : return -EINVAL;
2453 0 : vma->vm_pgoff = pfn;
2454 : }
2455 :
2456 0 : vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP);
2457 :
2458 0 : BUG_ON(addr >= end);
2459 0 : pfn -= addr >> PAGE_SHIFT;
2460 0 : pgd = pgd_offset(mm, addr);
2461 0 : flush_cache_range(vma, addr, end);
2462 : do {
2463 0 : next = pgd_addr_end(addr, end);
2464 0 : err = remap_p4d_range(mm, pgd, addr, next,
2465 0 : pfn + (addr >> PAGE_SHIFT), prot);
2466 0 : if (err)
2467 : return err;
2468 0 : } while (pgd++, addr = next, addr != end);
2469 :
2470 : return 0;
2471 : }
2472 :
2473 : /**
2474 : * remap_pfn_range - remap kernel memory to userspace
2475 : * @vma: user vma to map to
2476 : * @addr: target page aligned user address to start at
2477 : * @pfn: page frame number of kernel physical memory address
2478 : * @size: size of mapping area
2479 : * @prot: page protection flags for this mapping
2480 : *
2481 : * Note: this is only safe if the mm semaphore is held when called.
2482 : *
2483 : * Return: %0 on success, negative error code otherwise.
2484 : */
2485 0 : int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
2486 : unsigned long pfn, unsigned long size, pgprot_t prot)
2487 : {
2488 : int err;
2489 :
2490 0 : err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size));
2491 : if (err)
2492 : return -EINVAL;
2493 :
2494 0 : err = remap_pfn_range_notrack(vma, addr, pfn, size, prot);
2495 : if (err)
2496 : untrack_pfn(vma, pfn, PAGE_ALIGN(size), true);
2497 : return err;
2498 : }
2499 : EXPORT_SYMBOL(remap_pfn_range);
2500 :
2501 : /**
2502 : * vm_iomap_memory - remap memory to userspace
2503 : * @vma: user vma to map to
2504 : * @start: start of the physical memory to be mapped
2505 : * @len: size of area
2506 : *
2507 : * This is a simplified io_remap_pfn_range() for common driver use. The
2508 : * driver just needs to give us the physical memory range to be mapped,
2509 : * we'll figure out the rest from the vma information.
2510 : *
2511 : * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get
2512 : * whatever write-combining details or similar.
2513 : *
2514 : * Return: %0 on success, negative error code otherwise.
2515 : */
2516 0 : int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
2517 : {
2518 : unsigned long vm_len, pfn, pages;
2519 :
2520 : /* Check that the physical memory area passed in looks valid */
2521 0 : if (start + len < start)
2522 : return -EINVAL;
2523 : /*
2524 : * You *really* shouldn't map things that aren't page-aligned,
2525 : * but we've historically allowed it because IO memory might
2526 : * just have smaller alignment.
2527 : */
2528 0 : len += start & ~PAGE_MASK;
2529 0 : pfn = start >> PAGE_SHIFT;
2530 0 : pages = (len + ~PAGE_MASK) >> PAGE_SHIFT;
2531 0 : if (pfn + pages < pfn)
2532 : return -EINVAL;
2533 :
2534 : /* We start the mapping 'vm_pgoff' pages into the area */
2535 0 : if (vma->vm_pgoff > pages)
2536 : return -EINVAL;
2537 0 : pfn += vma->vm_pgoff;
2538 0 : pages -= vma->vm_pgoff;
2539 :
2540 : /* Can we fit all of the mapping? */
2541 0 : vm_len = vma->vm_end - vma->vm_start;
2542 0 : if (vm_len >> PAGE_SHIFT > pages)
2543 : return -EINVAL;
2544 :
2545 : /* Ok, let it rip */
2546 0 : return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
2547 : }
2548 : EXPORT_SYMBOL(vm_iomap_memory);
2549 :
2550 0 : static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
2551 : unsigned long addr, unsigned long end,
2552 : pte_fn_t fn, void *data, bool create,
2553 : pgtbl_mod_mask *mask)
2554 : {
2555 : pte_t *pte, *mapped_pte;
2556 0 : int err = 0;
2557 : spinlock_t *ptl;
2558 :
2559 0 : if (create) {
2560 0 : mapped_pte = pte = (mm == &init_mm) ?
2561 0 : pte_alloc_kernel_track(pmd, addr, mask) :
2562 0 : pte_alloc_map_lock(mm, pmd, addr, &ptl);
2563 0 : if (!pte)
2564 : return -ENOMEM;
2565 : } else {
2566 : mapped_pte = pte = (mm == &init_mm) ?
2567 0 : pte_offset_kernel(pmd, addr) :
2568 0 : pte_offset_map_lock(mm, pmd, addr, &ptl);
2569 : }
2570 :
2571 0 : BUG_ON(pmd_huge(*pmd));
2572 :
2573 : arch_enter_lazy_mmu_mode();
2574 :
2575 0 : if (fn) {
2576 : do {
2577 0 : if (create || !pte_none(*pte)) {
2578 0 : err = fn(pte++, addr, data);
2579 0 : if (err)
2580 : break;
2581 : }
2582 0 : } while (addr += PAGE_SIZE, addr != end);
2583 : }
2584 0 : *mask |= PGTBL_PTE_MODIFIED;
2585 :
2586 : arch_leave_lazy_mmu_mode();
2587 :
2588 0 : if (mm != &init_mm)
2589 0 : pte_unmap_unlock(mapped_pte, ptl);
2590 : return err;
2591 : }
2592 :
2593 0 : static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
2594 : unsigned long addr, unsigned long end,
2595 : pte_fn_t fn, void *data, bool create,
2596 : pgtbl_mod_mask *mask)
2597 : {
2598 : pmd_t *pmd;
2599 : unsigned long next;
2600 0 : int err = 0;
2601 :
2602 0 : BUG_ON(pud_huge(*pud));
2603 :
2604 0 : if (create) {
2605 0 : pmd = pmd_alloc_track(mm, pud, addr, mask);
2606 0 : if (!pmd)
2607 : return -ENOMEM;
2608 : } else {
2609 0 : pmd = pmd_offset(pud, addr);
2610 : }
2611 : do {
2612 0 : next = pmd_addr_end(addr, end);
2613 0 : if (pmd_none(*pmd) && !create)
2614 0 : continue;
2615 0 : if (WARN_ON_ONCE(pmd_leaf(*pmd)))
2616 : return -EINVAL;
2617 0 : if (!pmd_none(*pmd) && WARN_ON_ONCE(pmd_bad(*pmd))) {
2618 0 : if (!create)
2619 0 : continue;
2620 0 : pmd_clear_bad(pmd);
2621 : }
2622 0 : err = apply_to_pte_range(mm, pmd, addr, next,
2623 : fn, data, create, mask);
2624 0 : if (err)
2625 : break;
2626 0 : } while (pmd++, addr = next, addr != end);
2627 :
2628 : return err;
2629 : }
2630 :
2631 0 : static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d,
2632 : unsigned long addr, unsigned long end,
2633 : pte_fn_t fn, void *data, bool create,
2634 : pgtbl_mod_mask *mask)
2635 : {
2636 : pud_t *pud;
2637 : unsigned long next;
2638 0 : int err = 0;
2639 :
2640 0 : if (create) {
2641 0 : pud = pud_alloc_track(mm, p4d, addr, mask);
2642 0 : if (!pud)
2643 : return -ENOMEM;
2644 : } else {
2645 : pud = pud_offset(p4d, addr);
2646 : }
2647 : do {
2648 0 : next = pud_addr_end(addr, end);
2649 0 : if (pud_none(*pud) && !create)
2650 0 : continue;
2651 0 : if (WARN_ON_ONCE(pud_leaf(*pud)))
2652 : return -EINVAL;
2653 0 : if (!pud_none(*pud) && WARN_ON_ONCE(pud_bad(*pud))) {
2654 0 : if (!create)
2655 0 : continue;
2656 : pud_clear_bad(pud);
2657 : }
2658 0 : err = apply_to_pmd_range(mm, pud, addr, next,
2659 : fn, data, create, mask);
2660 0 : if (err)
2661 : break;
2662 0 : } while (pud++, addr = next, addr != end);
2663 :
2664 : return err;
2665 : }
2666 :
2667 : static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd,
2668 : unsigned long addr, unsigned long end,
2669 : pte_fn_t fn, void *data, bool create,
2670 : pgtbl_mod_mask *mask)
2671 : {
2672 : p4d_t *p4d;
2673 : unsigned long next;
2674 0 : int err = 0;
2675 :
2676 0 : if (create) {
2677 0 : p4d = p4d_alloc_track(mm, pgd, addr, mask);
2678 0 : if (!p4d)
2679 : return -ENOMEM;
2680 : } else {
2681 : p4d = p4d_offset(pgd, addr);
2682 : }
2683 : do {
2684 0 : next = p4d_addr_end(addr, end);
2685 0 : if (p4d_none(*p4d) && !create)
2686 : continue;
2687 0 : if (WARN_ON_ONCE(p4d_leaf(*p4d)))
2688 : return -EINVAL;
2689 0 : if (!p4d_none(*p4d) && WARN_ON_ONCE(p4d_bad(*p4d))) {
2690 : if (!create)
2691 : continue;
2692 : p4d_clear_bad(p4d);
2693 : }
2694 0 : err = apply_to_pud_range(mm, p4d, addr, next,
2695 : fn, data, create, mask);
2696 : if (err)
2697 : break;
2698 : } while (p4d++, addr = next, addr != end);
2699 :
2700 : return err;
2701 : }
2702 :
2703 0 : static int __apply_to_page_range(struct mm_struct *mm, unsigned long addr,
2704 : unsigned long size, pte_fn_t fn,
2705 : void *data, bool create)
2706 : {
2707 : pgd_t *pgd;
2708 0 : unsigned long start = addr, next;
2709 0 : unsigned long end = addr + size;
2710 0 : pgtbl_mod_mask mask = 0;
2711 0 : int err = 0;
2712 :
2713 0 : if (WARN_ON(addr >= end))
2714 : return -EINVAL;
2715 :
2716 0 : pgd = pgd_offset(mm, addr);
2717 : do {
2718 0 : next = pgd_addr_end(addr, end);
2719 0 : if (pgd_none(*pgd) && !create)
2720 : continue;
2721 0 : if (WARN_ON_ONCE(pgd_leaf(*pgd)))
2722 : return -EINVAL;
2723 0 : if (!pgd_none(*pgd) && WARN_ON_ONCE(pgd_bad(*pgd))) {
2724 : if (!create)
2725 : continue;
2726 : pgd_clear_bad(pgd);
2727 : }
2728 0 : err = apply_to_p4d_range(mm, pgd, addr, next,
2729 : fn, data, create, &mask);
2730 0 : if (err)
2731 : break;
2732 0 : } while (pgd++, addr = next, addr != end);
2733 :
2734 : if (mask & ARCH_PAGE_TABLE_SYNC_MASK)
2735 : arch_sync_kernel_mappings(start, start + size);
2736 :
2737 : return err;
2738 : }
2739 :
2740 : /*
2741 : * Scan a region of virtual memory, filling in page tables as necessary
2742 : * and calling a provided function on each leaf page table.
2743 : */
2744 0 : int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
2745 : unsigned long size, pte_fn_t fn, void *data)
2746 : {
2747 0 : return __apply_to_page_range(mm, addr, size, fn, data, true);
2748 : }
2749 : EXPORT_SYMBOL_GPL(apply_to_page_range);
2750 :
2751 : /*
2752 : * Scan a region of virtual memory, calling a provided function on
2753 : * each leaf page table where it exists.
2754 : *
2755 : * Unlike apply_to_page_range, this does _not_ fill in page tables
2756 : * where they are absent.
2757 : */
2758 0 : int apply_to_existing_page_range(struct mm_struct *mm, unsigned long addr,
2759 : unsigned long size, pte_fn_t fn, void *data)
2760 : {
2761 0 : return __apply_to_page_range(mm, addr, size, fn, data, false);
2762 : }
2763 : EXPORT_SYMBOL_GPL(apply_to_existing_page_range);
2764 :
2765 : /*
2766 : * handle_pte_fault chooses page fault handler according to an entry which was
2767 : * read non-atomically. Before making any commitment, on those architectures
2768 : * or configurations (e.g. i386 with PAE) which might give a mix of unmatched
2769 : * parts, do_swap_page must check under lock before unmapping the pte and
2770 : * proceeding (but do_wp_page is only called after already making such a check;
2771 : * and do_anonymous_page can safely check later on).
2772 : */
2773 : static inline int pte_unmap_same(struct vm_fault *vmf)
2774 : {
2775 0 : int same = 1;
2776 : #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPTION)
2777 : if (sizeof(pte_t) > sizeof(unsigned long)) {
2778 : spinlock_t *ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
2779 : spin_lock(ptl);
2780 : same = pte_same(*vmf->pte, vmf->orig_pte);
2781 : spin_unlock(ptl);
2782 : }
2783 : #endif
2784 : pte_unmap(vmf->pte);
2785 0 : vmf->pte = NULL;
2786 : return same;
2787 : }
2788 :
2789 : /*
2790 : * Return:
2791 : * 0: copied succeeded
2792 : * -EHWPOISON: copy failed due to hwpoison in source page
2793 : * -EAGAIN: copied failed (some other reason)
2794 : */
2795 0 : static inline int __wp_page_copy_user(struct page *dst, struct page *src,
2796 : struct vm_fault *vmf)
2797 : {
2798 : int ret;
2799 : void *kaddr;
2800 : void __user *uaddr;
2801 0 : bool locked = false;
2802 0 : struct vm_area_struct *vma = vmf->vma;
2803 0 : struct mm_struct *mm = vma->vm_mm;
2804 0 : unsigned long addr = vmf->address;
2805 :
2806 0 : if (likely(src)) {
2807 0 : if (copy_mc_user_highpage(dst, src, addr, vma)) {
2808 : memory_failure_queue(page_to_pfn(src), 0);
2809 : return -EHWPOISON;
2810 : }
2811 0 : return 0;
2812 : }
2813 :
2814 : /*
2815 : * If the source page was a PFN mapping, we don't have
2816 : * a "struct page" for it. We do a best-effort copy by
2817 : * just copying from the original user address. If that
2818 : * fails, we just zero-fill it. Live with it.
2819 : */
2820 0 : kaddr = kmap_atomic(dst);
2821 0 : uaddr = (void __user *)(addr & PAGE_MASK);
2822 :
2823 : /*
2824 : * On architectures with software "accessed" bits, we would
2825 : * take a double page fault, so mark it accessed here.
2826 : */
2827 0 : if (!arch_has_hw_pte_young() && !pte_young(vmf->orig_pte)) {
2828 : pte_t entry;
2829 :
2830 0 : vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl);
2831 0 : locked = true;
2832 0 : if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) {
2833 : /*
2834 : * Other thread has already handled the fault
2835 : * and update local tlb only
2836 : */
2837 0 : update_mmu_tlb(vma, addr, vmf->pte);
2838 0 : ret = -EAGAIN;
2839 0 : goto pte_unlock;
2840 : }
2841 :
2842 0 : entry = pte_mkyoung(vmf->orig_pte);
2843 0 : if (ptep_set_access_flags(vma, addr, vmf->pte, entry, 0))
2844 : update_mmu_cache(vma, addr, vmf->pte);
2845 : }
2846 :
2847 : /*
2848 : * This really shouldn't fail, because the page is there
2849 : * in the page tables. But it might just be unreadable,
2850 : * in which case we just give up and fill the result with
2851 : * zeroes.
2852 : */
2853 0 : if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) {
2854 0 : if (locked)
2855 : goto warn;
2856 :
2857 : /* Re-validate under PTL if the page is still mapped */
2858 0 : vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl);
2859 0 : locked = true;
2860 0 : if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) {
2861 : /* The PTE changed under us, update local tlb */
2862 : update_mmu_tlb(vma, addr, vmf->pte);
2863 : ret = -EAGAIN;
2864 : goto pte_unlock;
2865 : }
2866 :
2867 : /*
2868 : * The same page can be mapped back since last copy attempt.
2869 : * Try to copy again under PTL.
2870 : */
2871 0 : if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) {
2872 : /*
2873 : * Give a warn in case there can be some obscure
2874 : * use-case
2875 : */
2876 : warn:
2877 0 : WARN_ON_ONCE(1);
2878 0 : clear_page(kaddr);
2879 : }
2880 : }
2881 :
2882 : ret = 0;
2883 :
2884 : pte_unlock:
2885 0 : if (locked)
2886 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
2887 0 : kunmap_atomic(kaddr);
2888 0 : flush_dcache_page(dst);
2889 :
2890 0 : return ret;
2891 : }
2892 :
2893 : static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma)
2894 : {
2895 0 : struct file *vm_file = vma->vm_file;
2896 :
2897 0 : if (vm_file)
2898 0 : return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO;
2899 :
2900 : /*
2901 : * Special mappings (e.g. VDSO) do not have any file so fake
2902 : * a default GFP_KERNEL for them.
2903 : */
2904 : return GFP_KERNEL;
2905 : }
2906 :
2907 : /*
2908 : * Notify the address space that the page is about to become writable so that
2909 : * it can prohibit this or wait for the page to get into an appropriate state.
2910 : *
2911 : * We do this without the lock held, so that it can sleep if it needs to.
2912 : */
2913 0 : static vm_fault_t do_page_mkwrite(struct vm_fault *vmf)
2914 : {
2915 : vm_fault_t ret;
2916 0 : struct page *page = vmf->page;
2917 0 : unsigned int old_flags = vmf->flags;
2918 :
2919 0 : vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2920 :
2921 0 : if (vmf->vma->vm_file &&
2922 0 : IS_SWAPFILE(vmf->vma->vm_file->f_mapping->host))
2923 : return VM_FAULT_SIGBUS;
2924 :
2925 0 : ret = vmf->vma->vm_ops->page_mkwrite(vmf);
2926 : /* Restore original flags so that caller is not surprised */
2927 0 : vmf->flags = old_flags;
2928 0 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2929 : return ret;
2930 0 : if (unlikely(!(ret & VM_FAULT_LOCKED))) {
2931 0 : lock_page(page);
2932 0 : if (!page->mapping) {
2933 0 : unlock_page(page);
2934 0 : return 0; /* retry */
2935 : }
2936 0 : ret |= VM_FAULT_LOCKED;
2937 : } else
2938 : VM_BUG_ON_PAGE(!PageLocked(page), page);
2939 : return ret;
2940 : }
2941 :
2942 : /*
2943 : * Handle dirtying of a page in shared file mapping on a write fault.
2944 : *
2945 : * The function expects the page to be locked and unlocks it.
2946 : */
2947 0 : static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf)
2948 : {
2949 0 : struct vm_area_struct *vma = vmf->vma;
2950 : struct address_space *mapping;
2951 0 : struct page *page = vmf->page;
2952 : bool dirtied;
2953 0 : bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite;
2954 :
2955 0 : dirtied = set_page_dirty(page);
2956 : VM_BUG_ON_PAGE(PageAnon(page), page);
2957 : /*
2958 : * Take a local copy of the address_space - page.mapping may be zeroed
2959 : * by truncate after unlock_page(). The address_space itself remains
2960 : * pinned by vma->vm_file's reference. We rely on unlock_page()'s
2961 : * release semantics to prevent the compiler from undoing this copying.
2962 : */
2963 0 : mapping = page_rmapping(page);
2964 0 : unlock_page(page);
2965 :
2966 0 : if (!page_mkwrite)
2967 0 : file_update_time(vma->vm_file);
2968 :
2969 : /*
2970 : * Throttle page dirtying rate down to writeback speed.
2971 : *
2972 : * mapping may be NULL here because some device drivers do not
2973 : * set page.mapping but still dirty their pages
2974 : *
2975 : * Drop the mmap_lock before waiting on IO, if we can. The file
2976 : * is pinning the mapping, as per above.
2977 : */
2978 0 : if ((dirtied || page_mkwrite) && mapping) {
2979 : struct file *fpin;
2980 :
2981 0 : fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2982 0 : balance_dirty_pages_ratelimited(mapping);
2983 0 : if (fpin) {
2984 0 : fput(fpin);
2985 0 : return VM_FAULT_COMPLETED;
2986 : }
2987 : }
2988 :
2989 : return 0;
2990 : }
2991 :
2992 : /*
2993 : * Handle write page faults for pages that can be reused in the current vma
2994 : *
2995 : * This can happen either due to the mapping being with the VM_SHARED flag,
2996 : * or due to us being the last reference standing to the page. In either
2997 : * case, all we need to do here is to mark the page as writable and update
2998 : * any related book-keeping.
2999 : */
3000 0 : static inline void wp_page_reuse(struct vm_fault *vmf)
3001 : __releases(vmf->ptl)
3002 : {
3003 0 : struct vm_area_struct *vma = vmf->vma;
3004 0 : struct page *page = vmf->page;
3005 : pte_t entry;
3006 :
3007 : VM_BUG_ON(!(vmf->flags & FAULT_FLAG_WRITE));
3008 : VM_BUG_ON(page && PageAnon(page) && !PageAnonExclusive(page));
3009 :
3010 : /*
3011 : * Clear the pages cpupid information as the existing
3012 : * information potentially belongs to a now completely
3013 : * unrelated process.
3014 : */
3015 : if (page)
3016 : page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1);
3017 :
3018 0 : flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
3019 0 : entry = pte_mkyoung(vmf->orig_pte);
3020 0 : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
3021 0 : if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1))
3022 : update_mmu_cache(vma, vmf->address, vmf->pte);
3023 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3024 0 : count_vm_event(PGREUSE);
3025 0 : }
3026 :
3027 : /*
3028 : * Handle the case of a page which we actually need to copy to a new page,
3029 : * either due to COW or unsharing.
3030 : *
3031 : * Called with mmap_lock locked and the old page referenced, but
3032 : * without the ptl held.
3033 : *
3034 : * High level logic flow:
3035 : *
3036 : * - Allocate a page, copy the content of the old page to the new one.
3037 : * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc.
3038 : * - Take the PTL. If the pte changed, bail out and release the allocated page
3039 : * - If the pte is still the way we remember it, update the page table and all
3040 : * relevant references. This includes dropping the reference the page-table
3041 : * held to the old page, as well as updating the rmap.
3042 : * - In any case, unlock the PTL and drop the reference we took to the old page.
3043 : */
3044 0 : static vm_fault_t wp_page_copy(struct vm_fault *vmf)
3045 : {
3046 0 : const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE;
3047 0 : struct vm_area_struct *vma = vmf->vma;
3048 0 : struct mm_struct *mm = vma->vm_mm;
3049 0 : struct folio *old_folio = NULL;
3050 0 : struct folio *new_folio = NULL;
3051 : pte_t entry;
3052 0 : int page_copied = 0;
3053 : struct mmu_notifier_range range;
3054 : int ret;
3055 :
3056 : delayacct_wpcopy_start();
3057 :
3058 0 : if (vmf->page)
3059 0 : old_folio = page_folio(vmf->page);
3060 0 : if (unlikely(anon_vma_prepare(vma)))
3061 : goto oom;
3062 :
3063 0 : if (is_zero_pfn(pte_pfn(vmf->orig_pte))) {
3064 0 : new_folio = vma_alloc_zeroed_movable_folio(vma, vmf->address);
3065 0 : if (!new_folio)
3066 : goto oom;
3067 : } else {
3068 0 : new_folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma,
3069 : vmf->address, false);
3070 0 : if (!new_folio)
3071 : goto oom;
3072 :
3073 0 : ret = __wp_page_copy_user(&new_folio->page, vmf->page, vmf);
3074 0 : if (ret) {
3075 : /*
3076 : * COW failed, if the fault was solved by other,
3077 : * it's fine. If not, userspace would re-fault on
3078 : * the same address and we will handle the fault
3079 : * from the second attempt.
3080 : * The -EHWPOISON case will not be retried.
3081 : */
3082 0 : folio_put(new_folio);
3083 0 : if (old_folio)
3084 : folio_put(old_folio);
3085 :
3086 : delayacct_wpcopy_end();
3087 0 : return ret == -EHWPOISON ? VM_FAULT_HWPOISON : 0;
3088 : }
3089 : kmsan_copy_page_meta(&new_folio->page, vmf->page);
3090 : }
3091 :
3092 0 : if (mem_cgroup_charge(new_folio, mm, GFP_KERNEL))
3093 : goto oom_free_new;
3094 0 : cgroup_throttle_swaprate(&new_folio->page, GFP_KERNEL);
3095 :
3096 0 : __folio_mark_uptodate(new_folio);
3097 :
3098 0 : mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, mm,
3099 : vmf->address & PAGE_MASK,
3100 : (vmf->address & PAGE_MASK) + PAGE_SIZE);
3101 0 : mmu_notifier_invalidate_range_start(&range);
3102 :
3103 : /*
3104 : * Re-check the pte - we dropped the lock
3105 : */
3106 0 : vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl);
3107 0 : if (likely(pte_same(*vmf->pte, vmf->orig_pte))) {
3108 0 : if (old_folio) {
3109 0 : if (!folio_test_anon(old_folio)) {
3110 0 : dec_mm_counter(mm, mm_counter_file(&old_folio->page));
3111 : inc_mm_counter(mm, MM_ANONPAGES);
3112 : }
3113 : } else {
3114 : inc_mm_counter(mm, MM_ANONPAGES);
3115 : }
3116 0 : flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
3117 0 : entry = mk_pte(&new_folio->page, vma->vm_page_prot);
3118 : entry = pte_sw_mkyoung(entry);
3119 0 : if (unlikely(unshare)) {
3120 : if (pte_soft_dirty(vmf->orig_pte))
3121 : entry = pte_mksoft_dirty(entry);
3122 : if (pte_uffd_wp(vmf->orig_pte))
3123 : entry = pte_mkuffd_wp(entry);
3124 : } else {
3125 0 : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
3126 : }
3127 :
3128 : /*
3129 : * Clear the pte entry and flush it first, before updating the
3130 : * pte with the new entry, to keep TLBs on different CPUs in
3131 : * sync. This code used to set the new PTE then flush TLBs, but
3132 : * that left a window where the new PTE could be loaded into
3133 : * some TLBs while the old PTE remains in others.
3134 : */
3135 0 : ptep_clear_flush_notify(vma, vmf->address, vmf->pte);
3136 0 : folio_add_new_anon_rmap(new_folio, vma, vmf->address);
3137 0 : folio_add_lru_vma(new_folio, vma);
3138 : /*
3139 : * We call the notify macro here because, when using secondary
3140 : * mmu page tables (such as kvm shadow page tables), we want the
3141 : * new page to be mapped directly into the secondary page table.
3142 : */
3143 0 : BUG_ON(unshare && pte_write(entry));
3144 0 : set_pte_at_notify(mm, vmf->address, vmf->pte, entry);
3145 : update_mmu_cache(vma, vmf->address, vmf->pte);
3146 0 : if (old_folio) {
3147 : /*
3148 : * Only after switching the pte to the new page may
3149 : * we remove the mapcount here. Otherwise another
3150 : * process may come and find the rmap count decremented
3151 : * before the pte is switched to the new page, and
3152 : * "reuse" the old page writing into it while our pte
3153 : * here still points into it and can be read by other
3154 : * threads.
3155 : *
3156 : * The critical issue is to order this
3157 : * page_remove_rmap with the ptp_clear_flush above.
3158 : * Those stores are ordered by (if nothing else,)
3159 : * the barrier present in the atomic_add_negative
3160 : * in page_remove_rmap.
3161 : *
3162 : * Then the TLB flush in ptep_clear_flush ensures that
3163 : * no process can access the old page before the
3164 : * decremented mapcount is visible. And the old page
3165 : * cannot be reused until after the decremented
3166 : * mapcount is visible. So transitively, TLBs to
3167 : * old page will be flushed before it can be reused.
3168 : */
3169 0 : page_remove_rmap(vmf->page, vma, false);
3170 : }
3171 :
3172 : /* Free the old page.. */
3173 : new_folio = old_folio;
3174 : page_copied = 1;
3175 : } else {
3176 : update_mmu_tlb(vma, vmf->address, vmf->pte);
3177 : }
3178 :
3179 0 : if (new_folio)
3180 : folio_put(new_folio);
3181 :
3182 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3183 : /*
3184 : * No need to double call mmu_notifier->invalidate_range() callback as
3185 : * the above ptep_clear_flush_notify() did already call it.
3186 : */
3187 0 : mmu_notifier_invalidate_range_only_end(&range);
3188 0 : if (old_folio) {
3189 0 : if (page_copied)
3190 0 : free_swap_cache(&old_folio->page);
3191 : folio_put(old_folio);
3192 : }
3193 :
3194 : delayacct_wpcopy_end();
3195 : return 0;
3196 : oom_free_new:
3197 : folio_put(new_folio);
3198 : oom:
3199 0 : if (old_folio)
3200 : folio_put(old_folio);
3201 :
3202 : delayacct_wpcopy_end();
3203 : return VM_FAULT_OOM;
3204 : }
3205 :
3206 : /**
3207 : * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE
3208 : * writeable once the page is prepared
3209 : *
3210 : * @vmf: structure describing the fault
3211 : *
3212 : * This function handles all that is needed to finish a write page fault in a
3213 : * shared mapping due to PTE being read-only once the mapped page is prepared.
3214 : * It handles locking of PTE and modifying it.
3215 : *
3216 : * The function expects the page to be locked or other protection against
3217 : * concurrent faults / writeback (such as DAX radix tree locks).
3218 : *
3219 : * Return: %0 on success, %VM_FAULT_NOPAGE when PTE got changed before
3220 : * we acquired PTE lock.
3221 : */
3222 0 : vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf)
3223 : {
3224 0 : WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED));
3225 0 : vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address,
3226 : &vmf->ptl);
3227 : /*
3228 : * We might have raced with another page fault while we released the
3229 : * pte_offset_map_lock.
3230 : */
3231 0 : if (!pte_same(*vmf->pte, vmf->orig_pte)) {
3232 0 : update_mmu_tlb(vmf->vma, vmf->address, vmf->pte);
3233 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3234 0 : return VM_FAULT_NOPAGE;
3235 : }
3236 0 : wp_page_reuse(vmf);
3237 0 : return 0;
3238 : }
3239 :
3240 : /*
3241 : * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED
3242 : * mapping
3243 : */
3244 0 : static vm_fault_t wp_pfn_shared(struct vm_fault *vmf)
3245 : {
3246 0 : struct vm_area_struct *vma = vmf->vma;
3247 :
3248 0 : if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) {
3249 : vm_fault_t ret;
3250 :
3251 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3252 0 : vmf->flags |= FAULT_FLAG_MKWRITE;
3253 0 : ret = vma->vm_ops->pfn_mkwrite(vmf);
3254 0 : if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))
3255 : return ret;
3256 0 : return finish_mkwrite_fault(vmf);
3257 : }
3258 0 : wp_page_reuse(vmf);
3259 0 : return 0;
3260 : }
3261 :
3262 0 : static vm_fault_t wp_page_shared(struct vm_fault *vmf)
3263 : __releases(vmf->ptl)
3264 : {
3265 0 : struct vm_area_struct *vma = vmf->vma;
3266 0 : vm_fault_t ret = 0;
3267 :
3268 0 : get_page(vmf->page);
3269 :
3270 0 : if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
3271 : vm_fault_t tmp;
3272 :
3273 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3274 0 : tmp = do_page_mkwrite(vmf);
3275 0 : if (unlikely(!tmp || (tmp &
3276 : (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
3277 0 : put_page(vmf->page);
3278 0 : return tmp;
3279 : }
3280 0 : tmp = finish_mkwrite_fault(vmf);
3281 0 : if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
3282 0 : unlock_page(vmf->page);
3283 0 : put_page(vmf->page);
3284 0 : return tmp;
3285 : }
3286 : } else {
3287 0 : wp_page_reuse(vmf);
3288 0 : lock_page(vmf->page);
3289 : }
3290 0 : ret |= fault_dirty_shared_page(vmf);
3291 0 : put_page(vmf->page);
3292 :
3293 0 : return ret;
3294 : }
3295 :
3296 : /*
3297 : * This routine handles present pages, when
3298 : * * users try to write to a shared page (FAULT_FLAG_WRITE)
3299 : * * GUP wants to take a R/O pin on a possibly shared anonymous page
3300 : * (FAULT_FLAG_UNSHARE)
3301 : *
3302 : * It is done by copying the page to a new address and decrementing the
3303 : * shared-page counter for the old page.
3304 : *
3305 : * Note that this routine assumes that the protection checks have been
3306 : * done by the caller (the low-level page fault routine in most cases).
3307 : * Thus, with FAULT_FLAG_WRITE, we can safely just mark it writable once we've
3308 : * done any necessary COW.
3309 : *
3310 : * In case of FAULT_FLAG_WRITE, we also mark the page dirty at this point even
3311 : * though the page will change only once the write actually happens. This
3312 : * avoids a few races, and potentially makes it more efficient.
3313 : *
3314 : * We enter with non-exclusive mmap_lock (to exclude vma changes,
3315 : * but allow concurrent faults), with pte both mapped and locked.
3316 : * We return with mmap_lock still held, but pte unmapped and unlocked.
3317 : */
3318 0 : static vm_fault_t do_wp_page(struct vm_fault *vmf)
3319 : __releases(vmf->ptl)
3320 : {
3321 0 : const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE;
3322 0 : struct vm_area_struct *vma = vmf->vma;
3323 0 : struct folio *folio = NULL;
3324 :
3325 : if (likely(!unshare)) {
3326 : if (userfaultfd_pte_wp(vma, *vmf->pte)) {
3327 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3328 : return handle_userfault(vmf, VM_UFFD_WP);
3329 : }
3330 :
3331 : /*
3332 : * Userfaultfd write-protect can defer flushes. Ensure the TLB
3333 : * is flushed in this case before copying.
3334 : */
3335 : if (unlikely(userfaultfd_wp(vmf->vma) &&
3336 : mm_tlb_flush_pending(vmf->vma->vm_mm)))
3337 : flush_tlb_page(vmf->vma, vmf->address);
3338 : }
3339 :
3340 0 : vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte);
3341 :
3342 : /*
3343 : * Shared mapping: we are guaranteed to have VM_WRITE and
3344 : * FAULT_FLAG_WRITE set at this point.
3345 : */
3346 0 : if (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) {
3347 : /*
3348 : * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a
3349 : * VM_PFNMAP VMA.
3350 : *
3351 : * We should not cow pages in a shared writeable mapping.
3352 : * Just mark the pages writable and/or call ops->pfn_mkwrite.
3353 : */
3354 0 : if (!vmf->page)
3355 0 : return wp_pfn_shared(vmf);
3356 0 : return wp_page_shared(vmf);
3357 : }
3358 :
3359 0 : if (vmf->page)
3360 0 : folio = page_folio(vmf->page);
3361 :
3362 : /*
3363 : * Private mapping: create an exclusive anonymous page copy if reuse
3364 : * is impossible. We might miss VM_WRITE for FOLL_FORCE handling.
3365 : */
3366 0 : if (folio && folio_test_anon(folio)) {
3367 : /*
3368 : * If the page is exclusive to this process we must reuse the
3369 : * page without further checks.
3370 : */
3371 0 : if (PageAnonExclusive(vmf->page))
3372 : goto reuse;
3373 :
3374 : /*
3375 : * We have to verify under folio lock: these early checks are
3376 : * just an optimization to avoid locking the folio and freeing
3377 : * the swapcache if there is little hope that we can reuse.
3378 : *
3379 : * KSM doesn't necessarily raise the folio refcount.
3380 : */
3381 0 : if (folio_test_ksm(folio) || folio_ref_count(folio) > 3)
3382 : goto copy;
3383 0 : if (!folio_test_lru(folio))
3384 : /*
3385 : * Note: We cannot easily detect+handle references from
3386 : * remote LRU pagevecs or references to LRU folios.
3387 : */
3388 0 : lru_add_drain();
3389 0 : if (folio_ref_count(folio) > 1 + folio_test_swapcache(folio))
3390 : goto copy;
3391 0 : if (!folio_trylock(folio))
3392 : goto copy;
3393 0 : if (folio_test_swapcache(folio))
3394 0 : folio_free_swap(folio);
3395 0 : if (folio_test_ksm(folio) || folio_ref_count(folio) != 1) {
3396 0 : folio_unlock(folio);
3397 0 : goto copy;
3398 : }
3399 : /*
3400 : * Ok, we've got the only folio reference from our mapping
3401 : * and the folio is locked, it's dark out, and we're wearing
3402 : * sunglasses. Hit it.
3403 : */
3404 0 : page_move_anon_rmap(vmf->page, vma);
3405 0 : folio_unlock(folio);
3406 : reuse:
3407 0 : if (unlikely(unshare)) {
3408 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3409 0 : return 0;
3410 : }
3411 0 : wp_page_reuse(vmf);
3412 0 : return 0;
3413 : }
3414 : copy:
3415 : /*
3416 : * Ok, we need to copy. Oh, well..
3417 : */
3418 0 : if (folio)
3419 : folio_get(folio);
3420 :
3421 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3422 : #ifdef CONFIG_KSM
3423 : if (folio && folio_test_ksm(folio))
3424 : count_vm_event(COW_KSM);
3425 : #endif
3426 0 : return wp_page_copy(vmf);
3427 : }
3428 :
3429 : static void unmap_mapping_range_vma(struct vm_area_struct *vma,
3430 : unsigned long start_addr, unsigned long end_addr,
3431 : struct zap_details *details)
3432 : {
3433 0 : zap_page_range_single(vma, start_addr, end_addr - start_addr, details);
3434 : }
3435 :
3436 0 : static inline void unmap_mapping_range_tree(struct rb_root_cached *root,
3437 : pgoff_t first_index,
3438 : pgoff_t last_index,
3439 : struct zap_details *details)
3440 : {
3441 : struct vm_area_struct *vma;
3442 : pgoff_t vba, vea, zba, zea;
3443 :
3444 0 : vma_interval_tree_foreach(vma, root, first_index, last_index) {
3445 0 : vba = vma->vm_pgoff;
3446 0 : vea = vba + vma_pages(vma) - 1;
3447 0 : zba = max(first_index, vba);
3448 0 : zea = min(last_index, vea);
3449 :
3450 0 : unmap_mapping_range_vma(vma,
3451 0 : ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
3452 0 : ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
3453 : details);
3454 : }
3455 0 : }
3456 :
3457 : /**
3458 : * unmap_mapping_folio() - Unmap single folio from processes.
3459 : * @folio: The locked folio to be unmapped.
3460 : *
3461 : * Unmap this folio from any userspace process which still has it mmaped.
3462 : * Typically, for efficiency, the range of nearby pages has already been
3463 : * unmapped by unmap_mapping_pages() or unmap_mapping_range(). But once
3464 : * truncation or invalidation holds the lock on a folio, it may find that
3465 : * the page has been remapped again: and then uses unmap_mapping_folio()
3466 : * to unmap it finally.
3467 : */
3468 0 : void unmap_mapping_folio(struct folio *folio)
3469 : {
3470 0 : struct address_space *mapping = folio->mapping;
3471 0 : struct zap_details details = { };
3472 : pgoff_t first_index;
3473 : pgoff_t last_index;
3474 :
3475 : VM_BUG_ON(!folio_test_locked(folio));
3476 :
3477 0 : first_index = folio->index;
3478 0 : last_index = folio->index + folio_nr_pages(folio) - 1;
3479 :
3480 : details.even_cows = false;
3481 0 : details.single_folio = folio;
3482 0 : details.zap_flags = ZAP_FLAG_DROP_MARKER;
3483 :
3484 0 : i_mmap_lock_read(mapping);
3485 0 : if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)))
3486 0 : unmap_mapping_range_tree(&mapping->i_mmap, first_index,
3487 : last_index, &details);
3488 0 : i_mmap_unlock_read(mapping);
3489 0 : }
3490 :
3491 : /**
3492 : * unmap_mapping_pages() - Unmap pages from processes.
3493 : * @mapping: The address space containing pages to be unmapped.
3494 : * @start: Index of first page to be unmapped.
3495 : * @nr: Number of pages to be unmapped. 0 to unmap to end of file.
3496 : * @even_cows: Whether to unmap even private COWed pages.
3497 : *
3498 : * Unmap the pages in this address space from any userspace process which
3499 : * has them mmaped. Generally, you want to remove COWed pages as well when
3500 : * a file is being truncated, but not when invalidating pages from the page
3501 : * cache.
3502 : */
3503 0 : void unmap_mapping_pages(struct address_space *mapping, pgoff_t start,
3504 : pgoff_t nr, bool even_cows)
3505 : {
3506 0 : struct zap_details details = { };
3507 0 : pgoff_t first_index = start;
3508 0 : pgoff_t last_index = start + nr - 1;
3509 :
3510 0 : details.even_cows = even_cows;
3511 0 : if (last_index < first_index)
3512 0 : last_index = ULONG_MAX;
3513 :
3514 0 : i_mmap_lock_read(mapping);
3515 0 : if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)))
3516 0 : unmap_mapping_range_tree(&mapping->i_mmap, first_index,
3517 : last_index, &details);
3518 0 : i_mmap_unlock_read(mapping);
3519 0 : }
3520 : EXPORT_SYMBOL_GPL(unmap_mapping_pages);
3521 :
3522 : /**
3523 : * unmap_mapping_range - unmap the portion of all mmaps in the specified
3524 : * address_space corresponding to the specified byte range in the underlying
3525 : * file.
3526 : *
3527 : * @mapping: the address space containing mmaps to be unmapped.
3528 : * @holebegin: byte in first page to unmap, relative to the start of
3529 : * the underlying file. This will be rounded down to a PAGE_SIZE
3530 : * boundary. Note that this is different from truncate_pagecache(), which
3531 : * must keep the partial page. In contrast, we must get rid of
3532 : * partial pages.
3533 : * @holelen: size of prospective hole in bytes. This will be rounded
3534 : * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
3535 : * end of the file.
3536 : * @even_cows: 1 when truncating a file, unmap even private COWed pages;
3537 : * but 0 when invalidating pagecache, don't throw away private data.
3538 : */
3539 0 : void unmap_mapping_range(struct address_space *mapping,
3540 : loff_t const holebegin, loff_t const holelen, int even_cows)
3541 : {
3542 0 : pgoff_t hba = holebegin >> PAGE_SHIFT;
3543 0 : pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
3544 :
3545 : /* Check for overflow. */
3546 : if (sizeof(holelen) > sizeof(hlen)) {
3547 : long long holeend =
3548 : (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
3549 : if (holeend & ~(long long)ULONG_MAX)
3550 : hlen = ULONG_MAX - hba + 1;
3551 : }
3552 :
3553 0 : unmap_mapping_pages(mapping, hba, hlen, even_cows);
3554 0 : }
3555 : EXPORT_SYMBOL(unmap_mapping_range);
3556 :
3557 : /*
3558 : * Restore a potential device exclusive pte to a working pte entry
3559 : */
3560 : static vm_fault_t remove_device_exclusive_entry(struct vm_fault *vmf)
3561 : {
3562 : struct folio *folio = page_folio(vmf->page);
3563 : struct vm_area_struct *vma = vmf->vma;
3564 : struct mmu_notifier_range range;
3565 :
3566 : if (!folio_lock_or_retry(folio, vma->vm_mm, vmf->flags))
3567 : return VM_FAULT_RETRY;
3568 : mmu_notifier_range_init_owner(&range, MMU_NOTIFY_EXCLUSIVE, 0,
3569 : vma->vm_mm, vmf->address & PAGE_MASK,
3570 : (vmf->address & PAGE_MASK) + PAGE_SIZE, NULL);
3571 : mmu_notifier_invalidate_range_start(&range);
3572 :
3573 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
3574 : &vmf->ptl);
3575 : if (likely(pte_same(*vmf->pte, vmf->orig_pte)))
3576 : restore_exclusive_pte(vma, vmf->page, vmf->address, vmf->pte);
3577 :
3578 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3579 : folio_unlock(folio);
3580 :
3581 : mmu_notifier_invalidate_range_end(&range);
3582 : return 0;
3583 : }
3584 :
3585 0 : static inline bool should_try_to_free_swap(struct folio *folio,
3586 : struct vm_area_struct *vma,
3587 : unsigned int fault_flags)
3588 : {
3589 0 : if (!folio_test_swapcache(folio))
3590 : return false;
3591 0 : if (mem_cgroup_swap_full(folio) || (vma->vm_flags & VM_LOCKED) ||
3592 0 : folio_test_mlocked(folio))
3593 : return true;
3594 : /*
3595 : * If we want to map a page that's in the swapcache writable, we
3596 : * have to detect via the refcount if we're really the exclusive
3597 : * user. Try freeing the swapcache to get rid of the swapcache
3598 : * reference only in case it's likely that we'll be the exlusive user.
3599 : */
3600 0 : return (fault_flags & FAULT_FLAG_WRITE) && !folio_test_ksm(folio) &&
3601 0 : folio_ref_count(folio) == 2;
3602 : }
3603 :
3604 : static vm_fault_t pte_marker_clear(struct vm_fault *vmf)
3605 : {
3606 : vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
3607 : vmf->address, &vmf->ptl);
3608 : /*
3609 : * Be careful so that we will only recover a special uffd-wp pte into a
3610 : * none pte. Otherwise it means the pte could have changed, so retry.
3611 : *
3612 : * This should also cover the case where e.g. the pte changed
3613 : * quickly from a PTE_MARKER_UFFD_WP into PTE_MARKER_SWAPIN_ERROR.
3614 : * So is_pte_marker() check is not enough to safely drop the pte.
3615 : */
3616 : if (pte_same(vmf->orig_pte, *vmf->pte))
3617 : pte_clear(vmf->vma->vm_mm, vmf->address, vmf->pte);
3618 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3619 : return 0;
3620 : }
3621 :
3622 : /*
3623 : * This is actually a page-missing access, but with uffd-wp special pte
3624 : * installed. It means this pte was wr-protected before being unmapped.
3625 : */
3626 : static vm_fault_t pte_marker_handle_uffd_wp(struct vm_fault *vmf)
3627 : {
3628 : /*
3629 : * Just in case there're leftover special ptes even after the region
3630 : * got unregistered - we can simply clear them.
3631 : */
3632 : if (unlikely(!userfaultfd_wp(vmf->vma) || vma_is_anonymous(vmf->vma)))
3633 : return pte_marker_clear(vmf);
3634 :
3635 : /* do_fault() can handle pte markers too like none pte */
3636 : return do_fault(vmf);
3637 : }
3638 :
3639 0 : static vm_fault_t handle_pte_marker(struct vm_fault *vmf)
3640 : {
3641 0 : swp_entry_t entry = pte_to_swp_entry(vmf->orig_pte);
3642 0 : unsigned long marker = pte_marker_get(entry);
3643 :
3644 : /*
3645 : * PTE markers should never be empty. If anything weird happened,
3646 : * the best thing to do is to kill the process along with its mm.
3647 : */
3648 0 : if (WARN_ON_ONCE(!marker))
3649 : return VM_FAULT_SIGBUS;
3650 :
3651 : /* Higher priority than uffd-wp when data corrupted */
3652 : if (marker & PTE_MARKER_SWAPIN_ERROR)
3653 : return VM_FAULT_SIGBUS;
3654 :
3655 : if (pte_marker_entry_uffd_wp(entry))
3656 : return pte_marker_handle_uffd_wp(vmf);
3657 :
3658 : /* This is an unknown pte marker */
3659 : return VM_FAULT_SIGBUS;
3660 : }
3661 :
3662 : /*
3663 : * We enter with non-exclusive mmap_lock (to exclude vma changes,
3664 : * but allow concurrent faults), and pte mapped but not yet locked.
3665 : * We return with pte unmapped and unlocked.
3666 : *
3667 : * We return with the mmap_lock locked or unlocked in the same cases
3668 : * as does filemap_fault().
3669 : */
3670 0 : vm_fault_t do_swap_page(struct vm_fault *vmf)
3671 : {
3672 0 : struct vm_area_struct *vma = vmf->vma;
3673 0 : struct folio *swapcache, *folio = NULL;
3674 : struct page *page;
3675 0 : struct swap_info_struct *si = NULL;
3676 0 : rmap_t rmap_flags = RMAP_NONE;
3677 0 : bool exclusive = false;
3678 : swp_entry_t entry;
3679 : pte_t pte;
3680 : int locked;
3681 0 : vm_fault_t ret = 0;
3682 0 : void *shadow = NULL;
3683 :
3684 0 : if (!pte_unmap_same(vmf))
3685 : goto out;
3686 :
3687 0 : entry = pte_to_swp_entry(vmf->orig_pte);
3688 0 : if (unlikely(non_swap_entry(entry))) {
3689 0 : if (is_migration_entry(entry)) {
3690 0 : migration_entry_wait(vma->vm_mm, vmf->pmd,
3691 : vmf->address);
3692 0 : } else if (is_device_exclusive_entry(entry)) {
3693 : vmf->page = pfn_swap_entry_to_page(entry);
3694 : ret = remove_device_exclusive_entry(vmf);
3695 0 : } else if (is_device_private_entry(entry)) {
3696 : vmf->page = pfn_swap_entry_to_page(entry);
3697 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
3698 : vmf->address, &vmf->ptl);
3699 : if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) {
3700 : spin_unlock(vmf->ptl);
3701 : goto out;
3702 : }
3703 :
3704 : /*
3705 : * Get a page reference while we know the page can't be
3706 : * freed.
3707 : */
3708 : get_page(vmf->page);
3709 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3710 : ret = vmf->page->pgmap->ops->migrate_to_ram(vmf);
3711 : put_page(vmf->page);
3712 0 : } else if (is_hwpoison_entry(entry)) {
3713 : ret = VM_FAULT_HWPOISON;
3714 0 : } else if (is_pte_marker_entry(entry)) {
3715 0 : ret = handle_pte_marker(vmf);
3716 : } else {
3717 0 : print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL);
3718 0 : ret = VM_FAULT_SIGBUS;
3719 : }
3720 : goto out;
3721 : }
3722 :
3723 : /* Prevent swapoff from happening to us. */
3724 0 : si = get_swap_device(entry);
3725 0 : if (unlikely(!si))
3726 : goto out;
3727 :
3728 0 : folio = swap_cache_get_folio(entry, vma, vmf->address);
3729 0 : if (folio)
3730 0 : page = folio_file_page(folio, swp_offset(entry));
3731 0 : swapcache = folio;
3732 :
3733 0 : if (!folio) {
3734 0 : if (data_race(si->flags & SWP_SYNCHRONOUS_IO) &&
3735 0 : __swap_count(entry) == 1) {
3736 : /* skip swapcache */
3737 0 : folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0,
3738 : vma, vmf->address, false);
3739 0 : page = &folio->page;
3740 0 : if (folio) {
3741 0 : __folio_set_locked(folio);
3742 0 : __folio_set_swapbacked(folio);
3743 :
3744 0 : if (mem_cgroup_swapin_charge_folio(folio,
3745 : vma->vm_mm, GFP_KERNEL,
3746 : entry)) {
3747 : ret = VM_FAULT_OOM;
3748 : goto out_page;
3749 : }
3750 0 : mem_cgroup_swapin_uncharge_swap(entry);
3751 :
3752 0 : shadow = get_shadow_from_swap_cache(entry);
3753 0 : if (shadow)
3754 0 : workingset_refault(folio, shadow);
3755 :
3756 0 : folio_add_lru(folio);
3757 :
3758 : /* To provide entry to swap_readpage() */
3759 0 : folio_set_swap_entry(folio, entry);
3760 0 : swap_readpage(page, true, NULL);
3761 0 : folio->private = NULL;
3762 : }
3763 : } else {
3764 0 : page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
3765 : vmf);
3766 0 : if (page)
3767 0 : folio = page_folio(page);
3768 : swapcache = folio;
3769 : }
3770 :
3771 0 : if (!folio) {
3772 : /*
3773 : * Back out if somebody else faulted in this pte
3774 : * while we released the pte lock.
3775 : */
3776 0 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
3777 : vmf->address, &vmf->ptl);
3778 0 : if (likely(pte_same(*vmf->pte, vmf->orig_pte)))
3779 0 : ret = VM_FAULT_OOM;
3780 : goto unlock;
3781 : }
3782 :
3783 : /* Had to read the page from swap area: Major fault */
3784 0 : ret = VM_FAULT_MAJOR;
3785 0 : count_vm_event(PGMAJFAULT);
3786 0 : count_memcg_event_mm(vma->vm_mm, PGMAJFAULT);
3787 : } else if (PageHWPoison(page)) {
3788 : /*
3789 : * hwpoisoned dirty swapcache pages are kept for killing
3790 : * owner processes (which may be unknown at hwpoison time)
3791 : */
3792 : ret = VM_FAULT_HWPOISON;
3793 : goto out_release;
3794 : }
3795 :
3796 0 : locked = folio_lock_or_retry(folio, vma->vm_mm, vmf->flags);
3797 :
3798 0 : if (!locked) {
3799 0 : ret |= VM_FAULT_RETRY;
3800 0 : goto out_release;
3801 : }
3802 :
3803 0 : if (swapcache) {
3804 : /*
3805 : * Make sure folio_free_swap() or swapoff did not release the
3806 : * swapcache from under us. The page pin, and pte_same test
3807 : * below, are not enough to exclude that. Even if it is still
3808 : * swapcache, we need to check that the page's swap has not
3809 : * changed.
3810 : */
3811 0 : if (unlikely(!folio_test_swapcache(folio) ||
3812 : page_private(page) != entry.val))
3813 : goto out_page;
3814 :
3815 : /*
3816 : * KSM sometimes has to copy on read faults, for example, if
3817 : * page->index of !PageKSM() pages would be nonlinear inside the
3818 : * anon VMA -- PageKSM() is lost on actual swapout.
3819 : */
3820 0 : page = ksm_might_need_to_copy(page, vma, vmf->address);
3821 0 : if (unlikely(!page)) {
3822 : ret = VM_FAULT_OOM;
3823 : goto out_page;
3824 0 : } else if (unlikely(PTR_ERR(page) == -EHWPOISON)) {
3825 : ret = VM_FAULT_HWPOISON;
3826 : goto out_page;
3827 : }
3828 0 : folio = page_folio(page);
3829 :
3830 : /*
3831 : * If we want to map a page that's in the swapcache writable, we
3832 : * have to detect via the refcount if we're really the exclusive
3833 : * owner. Try removing the extra reference from the local LRU
3834 : * pagevecs if required.
3835 : */
3836 0 : if ((vmf->flags & FAULT_FLAG_WRITE) && folio == swapcache &&
3837 0 : !folio_test_ksm(folio) && !folio_test_lru(folio))
3838 0 : lru_add_drain();
3839 : }
3840 :
3841 0 : cgroup_throttle_swaprate(page, GFP_KERNEL);
3842 :
3843 : /*
3844 : * Back out if somebody else already faulted in this pte.
3845 : */
3846 0 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
3847 : &vmf->ptl);
3848 0 : if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte)))
3849 : goto out_nomap;
3850 :
3851 0 : if (unlikely(!folio_test_uptodate(folio))) {
3852 : ret = VM_FAULT_SIGBUS;
3853 : goto out_nomap;
3854 : }
3855 :
3856 : /*
3857 : * PG_anon_exclusive reuses PG_mappedtodisk for anon pages. A swap pte
3858 : * must never point at an anonymous page in the swapcache that is
3859 : * PG_anon_exclusive. Sanity check that this holds and especially, that
3860 : * no filesystem set PG_mappedtodisk on a page in the swapcache. Sanity
3861 : * check after taking the PT lock and making sure that nobody
3862 : * concurrently faulted in this page and set PG_anon_exclusive.
3863 : */
3864 0 : BUG_ON(!folio_test_anon(folio) && folio_test_mappedtodisk(folio));
3865 0 : BUG_ON(folio_test_anon(folio) && PageAnonExclusive(page));
3866 :
3867 : /*
3868 : * Check under PT lock (to protect against concurrent fork() sharing
3869 : * the swap entry concurrently) for certainly exclusive pages.
3870 : */
3871 0 : if (!folio_test_ksm(folio)) {
3872 0 : exclusive = pte_swp_exclusive(vmf->orig_pte);
3873 0 : if (folio != swapcache) {
3874 : /*
3875 : * We have a fresh page that is not exposed to the
3876 : * swapcache -> certainly exclusive.
3877 : */
3878 : exclusive = true;
3879 0 : } else if (exclusive && folio_test_writeback(folio) &&
3880 0 : data_race(si->flags & SWP_STABLE_WRITES)) {
3881 : /*
3882 : * This is tricky: not all swap backends support
3883 : * concurrent page modifications while under writeback.
3884 : *
3885 : * So if we stumble over such a page in the swapcache
3886 : * we must not set the page exclusive, otherwise we can
3887 : * map it writable without further checks and modify it
3888 : * while still under writeback.
3889 : *
3890 : * For these problematic swap backends, simply drop the
3891 : * exclusive marker: this is perfectly fine as we start
3892 : * writeback only if we fully unmapped the page and
3893 : * there are no unexpected references on the page after
3894 : * unmapping succeeded. After fully unmapped, no
3895 : * further GUP references (FOLL_GET and FOLL_PIN) can
3896 : * appear, so dropping the exclusive marker and mapping
3897 : * it only R/O is fine.
3898 : */
3899 0 : exclusive = false;
3900 : }
3901 : }
3902 :
3903 : /*
3904 : * Remove the swap entry and conditionally try to free up the swapcache.
3905 : * We're already holding a reference on the page but haven't mapped it
3906 : * yet.
3907 : */
3908 0 : swap_free(entry);
3909 0 : if (should_try_to_free_swap(folio, vma, vmf->flags))
3910 0 : folio_free_swap(folio);
3911 :
3912 0 : inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
3913 0 : dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
3914 0 : pte = mk_pte(page, vma->vm_page_prot);
3915 :
3916 : /*
3917 : * Same logic as in do_wp_page(); however, optimize for pages that are
3918 : * certainly not shared either because we just allocated them without
3919 : * exposing them to the swapcache or because the swap entry indicates
3920 : * exclusivity.
3921 : */
3922 0 : if (!folio_test_ksm(folio) &&
3923 0 : (exclusive || folio_ref_count(folio) == 1)) {
3924 0 : if (vmf->flags & FAULT_FLAG_WRITE) {
3925 0 : pte = maybe_mkwrite(pte_mkdirty(pte), vma);
3926 0 : vmf->flags &= ~FAULT_FLAG_WRITE;
3927 : }
3928 : rmap_flags |= RMAP_EXCLUSIVE;
3929 : }
3930 0 : flush_icache_page(vma, page);
3931 0 : if (pte_swp_soft_dirty(vmf->orig_pte))
3932 : pte = pte_mksoft_dirty(pte);
3933 : if (pte_swp_uffd_wp(vmf->orig_pte))
3934 : pte = pte_mkuffd_wp(pte);
3935 0 : vmf->orig_pte = pte;
3936 :
3937 : /* ksm created a completely new copy */
3938 0 : if (unlikely(folio != swapcache && swapcache)) {
3939 0 : page_add_new_anon_rmap(page, vma, vmf->address);
3940 0 : folio_add_lru_vma(folio, vma);
3941 : } else {
3942 0 : page_add_anon_rmap(page, vma, vmf->address, rmap_flags);
3943 : }
3944 :
3945 : VM_BUG_ON(!folio_test_anon(folio) ||
3946 : (pte_write(pte) && !PageAnonExclusive(page)));
3947 0 : set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte);
3948 0 : arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte);
3949 :
3950 0 : folio_unlock(folio);
3951 0 : if (folio != swapcache && swapcache) {
3952 : /*
3953 : * Hold the lock to avoid the swap entry to be reused
3954 : * until we take the PT lock for the pte_same() check
3955 : * (to avoid false positives from pte_same). For
3956 : * further safety release the lock after the swap_free
3957 : * so that the swap count won't change under a
3958 : * parallel locked swapcache.
3959 : */
3960 0 : folio_unlock(swapcache);
3961 : folio_put(swapcache);
3962 : }
3963 :
3964 0 : if (vmf->flags & FAULT_FLAG_WRITE) {
3965 0 : ret |= do_wp_page(vmf);
3966 0 : if (ret & VM_FAULT_ERROR)
3967 0 : ret &= VM_FAULT_ERROR;
3968 : goto out;
3969 : }
3970 :
3971 : /* No need to invalidate - it was non-present before */
3972 : update_mmu_cache(vma, vmf->address, vmf->pte);
3973 : unlock:
3974 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3975 : out:
3976 0 : if (si)
3977 : put_swap_device(si);
3978 : return ret;
3979 : out_nomap:
3980 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3981 : out_page:
3982 0 : folio_unlock(folio);
3983 : out_release:
3984 0 : folio_put(folio);
3985 0 : if (folio != swapcache && swapcache) {
3986 0 : folio_unlock(swapcache);
3987 : folio_put(swapcache);
3988 : }
3989 0 : if (si)
3990 : put_swap_device(si);
3991 : return ret;
3992 : }
3993 :
3994 : /*
3995 : * We enter with non-exclusive mmap_lock (to exclude vma changes,
3996 : * but allow concurrent faults), and pte mapped but not yet locked.
3997 : * We return with mmap_lock still held, but pte unmapped and unlocked.
3998 : */
3999 0 : static vm_fault_t do_anonymous_page(struct vm_fault *vmf)
4000 : {
4001 0 : struct vm_area_struct *vma = vmf->vma;
4002 : struct folio *folio;
4003 0 : vm_fault_t ret = 0;
4004 : pte_t entry;
4005 :
4006 : /* File mapping without ->vm_ops ? */
4007 0 : if (vma->vm_flags & VM_SHARED)
4008 : return VM_FAULT_SIGBUS;
4009 :
4010 : /*
4011 : * Use pte_alloc() instead of pte_alloc_map(). We can't run
4012 : * pte_offset_map() on pmds where a huge pmd might be created
4013 : * from a different thread.
4014 : *
4015 : * pte_alloc_map() is safe to use under mmap_write_lock(mm) or when
4016 : * parallel threads are excluded by other means.
4017 : *
4018 : * Here we only have mmap_read_lock(mm).
4019 : */
4020 0 : if (pte_alloc(vma->vm_mm, vmf->pmd))
4021 : return VM_FAULT_OOM;
4022 :
4023 : /* See comment in handle_pte_fault() */
4024 0 : if (unlikely(pmd_trans_unstable(vmf->pmd)))
4025 : return 0;
4026 :
4027 : /* Use the zero-page for reads */
4028 0 : if (!(vmf->flags & FAULT_FLAG_WRITE) &&
4029 : !mm_forbids_zeropage(vma->vm_mm)) {
4030 0 : entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address),
4031 : vma->vm_page_prot));
4032 0 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
4033 : vmf->address, &vmf->ptl);
4034 0 : if (!pte_none(*vmf->pte)) {
4035 : update_mmu_tlb(vma, vmf->address, vmf->pte);
4036 : goto unlock;
4037 : }
4038 0 : ret = check_stable_address_space(vma->vm_mm);
4039 0 : if (ret)
4040 : goto unlock;
4041 : /* Deliver the page fault to userland, check inside PT lock */
4042 : if (userfaultfd_missing(vma)) {
4043 : pte_unmap_unlock(vmf->pte, vmf->ptl);
4044 : return handle_userfault(vmf, VM_UFFD_MISSING);
4045 : }
4046 : goto setpte;
4047 : }
4048 :
4049 : /* Allocate our own private page. */
4050 0 : if (unlikely(anon_vma_prepare(vma)))
4051 : goto oom;
4052 0 : folio = vma_alloc_zeroed_movable_folio(vma, vmf->address);
4053 0 : if (!folio)
4054 : goto oom;
4055 :
4056 0 : if (mem_cgroup_charge(folio, vma->vm_mm, GFP_KERNEL))
4057 : goto oom_free_page;
4058 0 : cgroup_throttle_swaprate(&folio->page, GFP_KERNEL);
4059 :
4060 : /*
4061 : * The memory barrier inside __folio_mark_uptodate makes sure that
4062 : * preceding stores to the page contents become visible before
4063 : * the set_pte_at() write.
4064 : */
4065 0 : __folio_mark_uptodate(folio);
4066 :
4067 0 : entry = mk_pte(&folio->page, vma->vm_page_prot);
4068 : entry = pte_sw_mkyoung(entry);
4069 0 : if (vma->vm_flags & VM_WRITE)
4070 0 : entry = pte_mkwrite(pte_mkdirty(entry));
4071 :
4072 0 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
4073 : &vmf->ptl);
4074 0 : if (!pte_none(*vmf->pte)) {
4075 : update_mmu_tlb(vma, vmf->address, vmf->pte);
4076 : goto release;
4077 : }
4078 :
4079 0 : ret = check_stable_address_space(vma->vm_mm);
4080 0 : if (ret)
4081 : goto release;
4082 :
4083 : /* Deliver the page fault to userland, check inside PT lock */
4084 0 : if (userfaultfd_missing(vma)) {
4085 : pte_unmap_unlock(vmf->pte, vmf->ptl);
4086 : folio_put(folio);
4087 : return handle_userfault(vmf, VM_UFFD_MISSING);
4088 : }
4089 :
4090 0 : inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
4091 0 : folio_add_new_anon_rmap(folio, vma, vmf->address);
4092 0 : folio_add_lru_vma(folio, vma);
4093 : setpte:
4094 0 : set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry);
4095 :
4096 : /* No need to invalidate - it was non-present before */
4097 : update_mmu_cache(vma, vmf->address, vmf->pte);
4098 : unlock:
4099 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
4100 0 : return ret;
4101 : release:
4102 : folio_put(folio);
4103 : goto unlock;
4104 : oom_free_page:
4105 : folio_put(folio);
4106 : oom:
4107 : return VM_FAULT_OOM;
4108 : }
4109 :
4110 : /*
4111 : * The mmap_lock must have been held on entry, and may have been
4112 : * released depending on flags and vma->vm_ops->fault() return value.
4113 : * See filemap_fault() and __lock_page_retry().
4114 : */
4115 0 : static vm_fault_t __do_fault(struct vm_fault *vmf)
4116 : {
4117 0 : struct vm_area_struct *vma = vmf->vma;
4118 : vm_fault_t ret;
4119 :
4120 : /*
4121 : * Preallocate pte before we take page_lock because this might lead to
4122 : * deadlocks for memcg reclaim which waits for pages under writeback:
4123 : * lock_page(A)
4124 : * SetPageWriteback(A)
4125 : * unlock_page(A)
4126 : * lock_page(B)
4127 : * lock_page(B)
4128 : * pte_alloc_one
4129 : * shrink_page_list
4130 : * wait_on_page_writeback(A)
4131 : * SetPageWriteback(B)
4132 : * unlock_page(B)
4133 : * # flush A, B to clear the writeback
4134 : */
4135 0 : if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) {
4136 0 : vmf->prealloc_pte = pte_alloc_one(vma->vm_mm);
4137 0 : if (!vmf->prealloc_pte)
4138 : return VM_FAULT_OOM;
4139 : }
4140 :
4141 0 : ret = vma->vm_ops->fault(vmf);
4142 0 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY |
4143 : VM_FAULT_DONE_COW)))
4144 : return ret;
4145 :
4146 0 : if (unlikely(PageHWPoison(vmf->page))) {
4147 : struct page *page = vmf->page;
4148 : vm_fault_t poisonret = VM_FAULT_HWPOISON;
4149 : if (ret & VM_FAULT_LOCKED) {
4150 : if (page_mapped(page))
4151 : unmap_mapping_pages(page_mapping(page),
4152 : page->index, 1, false);
4153 : /* Retry if a clean page was removed from the cache. */
4154 : if (invalidate_inode_page(page))
4155 : poisonret = VM_FAULT_NOPAGE;
4156 : unlock_page(page);
4157 : }
4158 : put_page(page);
4159 : vmf->page = NULL;
4160 : return poisonret;
4161 : }
4162 :
4163 0 : if (unlikely(!(ret & VM_FAULT_LOCKED)))
4164 0 : lock_page(vmf->page);
4165 : else
4166 : VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page);
4167 :
4168 : return ret;
4169 : }
4170 :
4171 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4172 : static void deposit_prealloc_pte(struct vm_fault *vmf)
4173 : {
4174 : struct vm_area_struct *vma = vmf->vma;
4175 :
4176 : pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte);
4177 : /*
4178 : * We are going to consume the prealloc table,
4179 : * count that as nr_ptes.
4180 : */
4181 : mm_inc_nr_ptes(vma->vm_mm);
4182 : vmf->prealloc_pte = NULL;
4183 : }
4184 :
4185 : vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page)
4186 : {
4187 : struct vm_area_struct *vma = vmf->vma;
4188 : bool write = vmf->flags & FAULT_FLAG_WRITE;
4189 : unsigned long haddr = vmf->address & HPAGE_PMD_MASK;
4190 : pmd_t entry;
4191 : int i;
4192 : vm_fault_t ret = VM_FAULT_FALLBACK;
4193 :
4194 : if (!transhuge_vma_suitable(vma, haddr))
4195 : return ret;
4196 :
4197 : page = compound_head(page);
4198 : if (compound_order(page) != HPAGE_PMD_ORDER)
4199 : return ret;
4200 :
4201 : /*
4202 : * Just backoff if any subpage of a THP is corrupted otherwise
4203 : * the corrupted page may mapped by PMD silently to escape the
4204 : * check. This kind of THP just can be PTE mapped. Access to
4205 : * the corrupted subpage should trigger SIGBUS as expected.
4206 : */
4207 : if (unlikely(PageHasHWPoisoned(page)))
4208 : return ret;
4209 :
4210 : /*
4211 : * Archs like ppc64 need additional space to store information
4212 : * related to pte entry. Use the preallocated table for that.
4213 : */
4214 : if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) {
4215 : vmf->prealloc_pte = pte_alloc_one(vma->vm_mm);
4216 : if (!vmf->prealloc_pte)
4217 : return VM_FAULT_OOM;
4218 : }
4219 :
4220 : vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd);
4221 : if (unlikely(!pmd_none(*vmf->pmd)))
4222 : goto out;
4223 :
4224 : for (i = 0; i < HPAGE_PMD_NR; i++)
4225 : flush_icache_page(vma, page + i);
4226 :
4227 : entry = mk_huge_pmd(page, vma->vm_page_prot);
4228 : if (write)
4229 : entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
4230 :
4231 : add_mm_counter(vma->vm_mm, mm_counter_file(page), HPAGE_PMD_NR);
4232 : page_add_file_rmap(page, vma, true);
4233 :
4234 : /*
4235 : * deposit and withdraw with pmd lock held
4236 : */
4237 : if (arch_needs_pgtable_deposit())
4238 : deposit_prealloc_pte(vmf);
4239 :
4240 : set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry);
4241 :
4242 : update_mmu_cache_pmd(vma, haddr, vmf->pmd);
4243 :
4244 : /* fault is handled */
4245 : ret = 0;
4246 : count_vm_event(THP_FILE_MAPPED);
4247 : out:
4248 : spin_unlock(vmf->ptl);
4249 : return ret;
4250 : }
4251 : #else
4252 0 : vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page)
4253 : {
4254 0 : return VM_FAULT_FALLBACK;
4255 : }
4256 : #endif
4257 :
4258 0 : void do_set_pte(struct vm_fault *vmf, struct page *page, unsigned long addr)
4259 : {
4260 0 : struct vm_area_struct *vma = vmf->vma;
4261 0 : bool uffd_wp = pte_marker_uffd_wp(vmf->orig_pte);
4262 0 : bool write = vmf->flags & FAULT_FLAG_WRITE;
4263 0 : bool prefault = vmf->address != addr;
4264 : pte_t entry;
4265 :
4266 0 : flush_icache_page(vma, page);
4267 0 : entry = mk_pte(page, vma->vm_page_prot);
4268 :
4269 : if (prefault && arch_wants_old_prefaulted_pte())
4270 : entry = pte_mkold(entry);
4271 : else
4272 : entry = pte_sw_mkyoung(entry);
4273 :
4274 0 : if (write)
4275 0 : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
4276 : if (unlikely(uffd_wp))
4277 : entry = pte_mkuffd_wp(entry);
4278 : /* copy-on-write page */
4279 0 : if (write && !(vma->vm_flags & VM_SHARED)) {
4280 0 : inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
4281 0 : page_add_new_anon_rmap(page, vma, addr);
4282 0 : lru_cache_add_inactive_or_unevictable(page, vma);
4283 : } else {
4284 0 : inc_mm_counter(vma->vm_mm, mm_counter_file(page));
4285 0 : page_add_file_rmap(page, vma, false);
4286 : }
4287 0 : set_pte_at(vma->vm_mm, addr, vmf->pte, entry);
4288 0 : }
4289 :
4290 : static bool vmf_pte_changed(struct vm_fault *vmf)
4291 : {
4292 0 : if (vmf->flags & FAULT_FLAG_ORIG_PTE_VALID)
4293 0 : return !pte_same(*vmf->pte, vmf->orig_pte);
4294 :
4295 0 : return !pte_none(*vmf->pte);
4296 : }
4297 :
4298 : /**
4299 : * finish_fault - finish page fault once we have prepared the page to fault
4300 : *
4301 : * @vmf: structure describing the fault
4302 : *
4303 : * This function handles all that is needed to finish a page fault once the
4304 : * page to fault in is prepared. It handles locking of PTEs, inserts PTE for
4305 : * given page, adds reverse page mapping, handles memcg charges and LRU
4306 : * addition.
4307 : *
4308 : * The function expects the page to be locked and on success it consumes a
4309 : * reference of a page being mapped (for the PTE which maps it).
4310 : *
4311 : * Return: %0 on success, %VM_FAULT_ code in case of error.
4312 : */
4313 0 : vm_fault_t finish_fault(struct vm_fault *vmf)
4314 : {
4315 0 : struct vm_area_struct *vma = vmf->vma;
4316 : struct page *page;
4317 : vm_fault_t ret;
4318 :
4319 : /* Did we COW the page? */
4320 0 : if ((vmf->flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED))
4321 0 : page = vmf->cow_page;
4322 : else
4323 0 : page = vmf->page;
4324 :
4325 : /*
4326 : * check even for read faults because we might have lost our CoWed
4327 : * page
4328 : */
4329 0 : if (!(vma->vm_flags & VM_SHARED)) {
4330 0 : ret = check_stable_address_space(vma->vm_mm);
4331 0 : if (ret)
4332 : return ret;
4333 : }
4334 :
4335 0 : if (pmd_none(*vmf->pmd)) {
4336 0 : if (PageTransCompound(page)) {
4337 : ret = do_set_pmd(vmf, page);
4338 : if (ret != VM_FAULT_FALLBACK)
4339 : return ret;
4340 : }
4341 :
4342 0 : if (vmf->prealloc_pte)
4343 0 : pmd_install(vma->vm_mm, vmf->pmd, &vmf->prealloc_pte);
4344 0 : else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd)))
4345 : return VM_FAULT_OOM;
4346 : }
4347 :
4348 : /*
4349 : * See comment in handle_pte_fault() for how this scenario happens, we
4350 : * need to return NOPAGE so that we drop this page.
4351 : */
4352 0 : if (pmd_devmap_trans_unstable(vmf->pmd))
4353 : return VM_FAULT_NOPAGE;
4354 :
4355 0 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
4356 : vmf->address, &vmf->ptl);
4357 :
4358 : /* Re-check under ptl */
4359 0 : if (likely(!vmf_pte_changed(vmf))) {
4360 0 : do_set_pte(vmf, page, vmf->address);
4361 :
4362 : /* no need to invalidate: a not-present page won't be cached */
4363 : update_mmu_cache(vma, vmf->address, vmf->pte);
4364 :
4365 0 : ret = 0;
4366 : } else {
4367 : update_mmu_tlb(vma, vmf->address, vmf->pte);
4368 : ret = VM_FAULT_NOPAGE;
4369 : }
4370 :
4371 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
4372 0 : return ret;
4373 : }
4374 :
4375 : static unsigned long fault_around_bytes __read_mostly =
4376 : rounddown_pow_of_two(65536);
4377 :
4378 : #ifdef CONFIG_DEBUG_FS
4379 : static int fault_around_bytes_get(void *data, u64 *val)
4380 : {
4381 : *val = fault_around_bytes;
4382 : return 0;
4383 : }
4384 :
4385 : /*
4386 : * fault_around_bytes must be rounded down to the nearest page order as it's
4387 : * what do_fault_around() expects to see.
4388 : */
4389 : static int fault_around_bytes_set(void *data, u64 val)
4390 : {
4391 : if (val / PAGE_SIZE > PTRS_PER_PTE)
4392 : return -EINVAL;
4393 : if (val > PAGE_SIZE)
4394 : fault_around_bytes = rounddown_pow_of_two(val);
4395 : else
4396 : fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */
4397 : return 0;
4398 : }
4399 : DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops,
4400 : fault_around_bytes_get, fault_around_bytes_set, "%llu\n");
4401 :
4402 : static int __init fault_around_debugfs(void)
4403 : {
4404 : debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL,
4405 : &fault_around_bytes_fops);
4406 : return 0;
4407 : }
4408 : late_initcall(fault_around_debugfs);
4409 : #endif
4410 :
4411 : /*
4412 : * do_fault_around() tries to map few pages around the fault address. The hope
4413 : * is that the pages will be needed soon and this will lower the number of
4414 : * faults to handle.
4415 : *
4416 : * It uses vm_ops->map_pages() to map the pages, which skips the page if it's
4417 : * not ready to be mapped: not up-to-date, locked, etc.
4418 : *
4419 : * This function doesn't cross the VMA boundaries, in order to call map_pages()
4420 : * only once.
4421 : *
4422 : * fault_around_bytes defines how many bytes we'll try to map.
4423 : * do_fault_around() expects it to be set to a power of two less than or equal
4424 : * to PTRS_PER_PTE.
4425 : *
4426 : * The virtual address of the area that we map is naturally aligned to
4427 : * fault_around_bytes rounded down to the machine page size
4428 : * (and therefore to page order). This way it's easier to guarantee
4429 : * that we don't cross page table boundaries.
4430 : */
4431 0 : static vm_fault_t do_fault_around(struct vm_fault *vmf)
4432 : {
4433 0 : unsigned long address = vmf->address, nr_pages, mask;
4434 0 : pgoff_t start_pgoff = vmf->pgoff;
4435 : pgoff_t end_pgoff;
4436 : int off;
4437 :
4438 0 : nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT;
4439 0 : mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK;
4440 :
4441 0 : address = max(address & mask, vmf->vma->vm_start);
4442 0 : off = ((vmf->address - address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
4443 0 : start_pgoff -= off;
4444 :
4445 : /*
4446 : * end_pgoff is either the end of the page table, the end of
4447 : * the vma or nr_pages from start_pgoff, depending what is nearest.
4448 : */
4449 0 : end_pgoff = start_pgoff -
4450 0 : ((address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) +
4451 : PTRS_PER_PTE - 1;
4452 0 : end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1,
4453 : start_pgoff + nr_pages - 1);
4454 :
4455 0 : if (pmd_none(*vmf->pmd)) {
4456 0 : vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm);
4457 0 : if (!vmf->prealloc_pte)
4458 : return VM_FAULT_OOM;
4459 : }
4460 :
4461 0 : return vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff);
4462 : }
4463 :
4464 : /* Return true if we should do read fault-around, false otherwise */
4465 : static inline bool should_fault_around(struct vm_fault *vmf)
4466 : {
4467 : /* No ->map_pages? No way to fault around... */
4468 0 : if (!vmf->vma->vm_ops->map_pages)
4469 : return false;
4470 :
4471 0 : if (uffd_disable_fault_around(vmf->vma))
4472 : return false;
4473 :
4474 0 : return fault_around_bytes >> PAGE_SHIFT > 1;
4475 : }
4476 :
4477 0 : static vm_fault_t do_read_fault(struct vm_fault *vmf)
4478 : {
4479 0 : vm_fault_t ret = 0;
4480 :
4481 : /*
4482 : * Let's call ->map_pages() first and use ->fault() as fallback
4483 : * if page by the offset is not ready to be mapped (cold cache or
4484 : * something).
4485 : */
4486 0 : if (should_fault_around(vmf)) {
4487 0 : ret = do_fault_around(vmf);
4488 0 : if (ret)
4489 : return ret;
4490 : }
4491 :
4492 0 : ret = __do_fault(vmf);
4493 0 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
4494 : return ret;
4495 :
4496 0 : ret |= finish_fault(vmf);
4497 0 : unlock_page(vmf->page);
4498 0 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
4499 0 : put_page(vmf->page);
4500 : return ret;
4501 : }
4502 :
4503 0 : static vm_fault_t do_cow_fault(struct vm_fault *vmf)
4504 : {
4505 0 : struct vm_area_struct *vma = vmf->vma;
4506 : vm_fault_t ret;
4507 :
4508 0 : if (unlikely(anon_vma_prepare(vma)))
4509 : return VM_FAULT_OOM;
4510 :
4511 0 : vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address);
4512 0 : if (!vmf->cow_page)
4513 : return VM_FAULT_OOM;
4514 :
4515 0 : if (mem_cgroup_charge(page_folio(vmf->cow_page), vma->vm_mm,
4516 : GFP_KERNEL)) {
4517 : put_page(vmf->cow_page);
4518 : return VM_FAULT_OOM;
4519 : }
4520 0 : cgroup_throttle_swaprate(vmf->cow_page, GFP_KERNEL);
4521 :
4522 0 : ret = __do_fault(vmf);
4523 0 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
4524 : goto uncharge_out;
4525 0 : if (ret & VM_FAULT_DONE_COW)
4526 : return ret;
4527 :
4528 0 : copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma);
4529 0 : __SetPageUptodate(vmf->cow_page);
4530 :
4531 0 : ret |= finish_fault(vmf);
4532 0 : unlock_page(vmf->page);
4533 0 : put_page(vmf->page);
4534 0 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
4535 : goto uncharge_out;
4536 : return ret;
4537 : uncharge_out:
4538 0 : put_page(vmf->cow_page);
4539 0 : return ret;
4540 : }
4541 :
4542 0 : static vm_fault_t do_shared_fault(struct vm_fault *vmf)
4543 : {
4544 0 : struct vm_area_struct *vma = vmf->vma;
4545 : vm_fault_t ret, tmp;
4546 :
4547 0 : ret = __do_fault(vmf);
4548 0 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
4549 : return ret;
4550 :
4551 : /*
4552 : * Check if the backing address space wants to know that the page is
4553 : * about to become writable
4554 : */
4555 0 : if (vma->vm_ops->page_mkwrite) {
4556 0 : unlock_page(vmf->page);
4557 0 : tmp = do_page_mkwrite(vmf);
4558 0 : if (unlikely(!tmp ||
4559 : (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
4560 0 : put_page(vmf->page);
4561 0 : return tmp;
4562 : }
4563 : }
4564 :
4565 0 : ret |= finish_fault(vmf);
4566 0 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |
4567 : VM_FAULT_RETRY))) {
4568 0 : unlock_page(vmf->page);
4569 0 : put_page(vmf->page);
4570 0 : return ret;
4571 : }
4572 :
4573 0 : ret |= fault_dirty_shared_page(vmf);
4574 0 : return ret;
4575 : }
4576 :
4577 : /*
4578 : * We enter with non-exclusive mmap_lock (to exclude vma changes,
4579 : * but allow concurrent faults).
4580 : * The mmap_lock may have been released depending on flags and our
4581 : * return value. See filemap_fault() and __folio_lock_or_retry().
4582 : * If mmap_lock is released, vma may become invalid (for example
4583 : * by other thread calling munmap()).
4584 : */
4585 0 : static vm_fault_t do_fault(struct vm_fault *vmf)
4586 : {
4587 0 : struct vm_area_struct *vma = vmf->vma;
4588 0 : struct mm_struct *vm_mm = vma->vm_mm;
4589 : vm_fault_t ret;
4590 :
4591 : /*
4592 : * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND
4593 : */
4594 0 : if (!vma->vm_ops->fault) {
4595 : /*
4596 : * If we find a migration pmd entry or a none pmd entry, which
4597 : * should never happen, return SIGBUS
4598 : */
4599 0 : if (unlikely(!pmd_present(*vmf->pmd)))
4600 : ret = VM_FAULT_SIGBUS;
4601 : else {
4602 0 : vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm,
4603 : vmf->pmd,
4604 : vmf->address,
4605 : &vmf->ptl);
4606 : /*
4607 : * Make sure this is not a temporary clearing of pte
4608 : * by holding ptl and checking again. A R/M/W update
4609 : * of pte involves: take ptl, clearing the pte so that
4610 : * we don't have concurrent modification by hardware
4611 : * followed by an update.
4612 : */
4613 0 : if (unlikely(pte_none(*vmf->pte)))
4614 : ret = VM_FAULT_SIGBUS;
4615 : else
4616 0 : ret = VM_FAULT_NOPAGE;
4617 :
4618 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
4619 : }
4620 0 : } else if (!(vmf->flags & FAULT_FLAG_WRITE))
4621 0 : ret = do_read_fault(vmf);
4622 0 : else if (!(vma->vm_flags & VM_SHARED))
4623 0 : ret = do_cow_fault(vmf);
4624 : else
4625 0 : ret = do_shared_fault(vmf);
4626 :
4627 : /* preallocated pagetable is unused: free it */
4628 0 : if (vmf->prealloc_pte) {
4629 0 : pte_free(vm_mm, vmf->prealloc_pte);
4630 0 : vmf->prealloc_pte = NULL;
4631 : }
4632 0 : return ret;
4633 : }
4634 :
4635 0 : int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
4636 : unsigned long addr, int page_nid, int *flags)
4637 : {
4638 0 : get_page(page);
4639 :
4640 : count_vm_numa_event(NUMA_HINT_FAULTS);
4641 0 : if (page_nid == numa_node_id()) {
4642 : count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
4643 0 : *flags |= TNF_FAULT_LOCAL;
4644 : }
4645 :
4646 0 : return mpol_misplaced(page, vma, addr);
4647 : }
4648 :
4649 : static vm_fault_t do_numa_page(struct vm_fault *vmf)
4650 : {
4651 : struct vm_area_struct *vma = vmf->vma;
4652 : struct page *page = NULL;
4653 : int page_nid = NUMA_NO_NODE;
4654 : bool writable = false;
4655 : int last_cpupid;
4656 : int target_nid;
4657 : pte_t pte, old_pte;
4658 : int flags = 0;
4659 :
4660 : /*
4661 : * The "pte" at this point cannot be used safely without
4662 : * validation through pte_unmap_same(). It's of NUMA type but
4663 : * the pfn may be screwed if the read is non atomic.
4664 : */
4665 : vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd);
4666 : spin_lock(vmf->ptl);
4667 : if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) {
4668 : pte_unmap_unlock(vmf->pte, vmf->ptl);
4669 : goto out;
4670 : }
4671 :
4672 : /* Get the normal PTE */
4673 : old_pte = ptep_get(vmf->pte);
4674 : pte = pte_modify(old_pte, vma->vm_page_prot);
4675 :
4676 : /*
4677 : * Detect now whether the PTE could be writable; this information
4678 : * is only valid while holding the PT lock.
4679 : */
4680 : writable = pte_write(pte);
4681 : if (!writable && vma_wants_manual_pte_write_upgrade(vma) &&
4682 : can_change_pte_writable(vma, vmf->address, pte))
4683 : writable = true;
4684 :
4685 : page = vm_normal_page(vma, vmf->address, pte);
4686 : if (!page || is_zone_device_page(page))
4687 : goto out_map;
4688 :
4689 : /* TODO: handle PTE-mapped THP */
4690 : if (PageCompound(page))
4691 : goto out_map;
4692 :
4693 : /*
4694 : * Avoid grouping on RO pages in general. RO pages shouldn't hurt as
4695 : * much anyway since they can be in shared cache state. This misses
4696 : * the case where a mapping is writable but the process never writes
4697 : * to it but pte_write gets cleared during protection updates and
4698 : * pte_dirty has unpredictable behaviour between PTE scan updates,
4699 : * background writeback, dirty balancing and application behaviour.
4700 : */
4701 : if (!writable)
4702 : flags |= TNF_NO_GROUP;
4703 :
4704 : /*
4705 : * Flag if the page is shared between multiple address spaces. This
4706 : * is later used when determining whether to group tasks together
4707 : */
4708 : if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED))
4709 : flags |= TNF_SHARED;
4710 :
4711 : page_nid = page_to_nid(page);
4712 : /*
4713 : * For memory tiering mode, cpupid of slow memory page is used
4714 : * to record page access time. So use default value.
4715 : */
4716 : if ((sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) &&
4717 : !node_is_toptier(page_nid))
4718 : last_cpupid = (-1 & LAST_CPUPID_MASK);
4719 : else
4720 : last_cpupid = page_cpupid_last(page);
4721 : target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid,
4722 : &flags);
4723 : if (target_nid == NUMA_NO_NODE) {
4724 : put_page(page);
4725 : goto out_map;
4726 : }
4727 : pte_unmap_unlock(vmf->pte, vmf->ptl);
4728 : writable = false;
4729 :
4730 : /* Migrate to the requested node */
4731 : if (migrate_misplaced_page(page, vma, target_nid)) {
4732 : page_nid = target_nid;
4733 : flags |= TNF_MIGRATED;
4734 : } else {
4735 : flags |= TNF_MIGRATE_FAIL;
4736 : vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
4737 : spin_lock(vmf->ptl);
4738 : if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) {
4739 : pte_unmap_unlock(vmf->pte, vmf->ptl);
4740 : goto out;
4741 : }
4742 : goto out_map;
4743 : }
4744 :
4745 : out:
4746 : if (page_nid != NUMA_NO_NODE)
4747 : task_numa_fault(last_cpupid, page_nid, 1, flags);
4748 : return 0;
4749 : out_map:
4750 : /*
4751 : * Make it present again, depending on how arch implements
4752 : * non-accessible ptes, some can allow access by kernel mode.
4753 : */
4754 : old_pte = ptep_modify_prot_start(vma, vmf->address, vmf->pte);
4755 : pte = pte_modify(old_pte, vma->vm_page_prot);
4756 : pte = pte_mkyoung(pte);
4757 : if (writable)
4758 : pte = pte_mkwrite(pte);
4759 : ptep_modify_prot_commit(vma, vmf->address, vmf->pte, old_pte, pte);
4760 : update_mmu_cache(vma, vmf->address, vmf->pte);
4761 : pte_unmap_unlock(vmf->pte, vmf->ptl);
4762 : goto out;
4763 : }
4764 :
4765 : static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf)
4766 : {
4767 : if (vma_is_anonymous(vmf->vma))
4768 : return do_huge_pmd_anonymous_page(vmf);
4769 : if (vmf->vma->vm_ops->huge_fault)
4770 : return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD);
4771 : return VM_FAULT_FALLBACK;
4772 : }
4773 :
4774 : /* `inline' is required to avoid gcc 4.1.2 build error */
4775 : static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf)
4776 : {
4777 : const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE;
4778 : vm_fault_t ret;
4779 :
4780 : if (vma_is_anonymous(vmf->vma)) {
4781 : if (likely(!unshare) &&
4782 : userfaultfd_huge_pmd_wp(vmf->vma, vmf->orig_pmd))
4783 : return handle_userfault(vmf, VM_UFFD_WP);
4784 : return do_huge_pmd_wp_page(vmf);
4785 : }
4786 :
4787 : if (vmf->vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) {
4788 : if (vmf->vma->vm_ops->huge_fault) {
4789 : ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD);
4790 : if (!(ret & VM_FAULT_FALLBACK))
4791 : return ret;
4792 : }
4793 : }
4794 :
4795 : /* COW or write-notify handled on pte level: split pmd. */
4796 : __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL);
4797 :
4798 : return VM_FAULT_FALLBACK;
4799 : }
4800 :
4801 : static vm_fault_t create_huge_pud(struct vm_fault *vmf)
4802 : {
4803 : #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
4804 : defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
4805 : /* No support for anonymous transparent PUD pages yet */
4806 : if (vma_is_anonymous(vmf->vma))
4807 : return VM_FAULT_FALLBACK;
4808 : if (vmf->vma->vm_ops->huge_fault)
4809 : return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD);
4810 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
4811 : return VM_FAULT_FALLBACK;
4812 : }
4813 :
4814 : static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud)
4815 : {
4816 : #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
4817 : defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
4818 : vm_fault_t ret;
4819 :
4820 : /* No support for anonymous transparent PUD pages yet */
4821 : if (vma_is_anonymous(vmf->vma))
4822 : goto split;
4823 : if (vmf->vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) {
4824 : if (vmf->vma->vm_ops->huge_fault) {
4825 : ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD);
4826 : if (!(ret & VM_FAULT_FALLBACK))
4827 : return ret;
4828 : }
4829 : }
4830 : split:
4831 : /* COW or write-notify not handled on PUD level: split pud.*/
4832 : __split_huge_pud(vmf->vma, vmf->pud, vmf->address);
4833 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
4834 : return VM_FAULT_FALLBACK;
4835 : }
4836 :
4837 : /*
4838 : * These routines also need to handle stuff like marking pages dirty
4839 : * and/or accessed for architectures that don't do it in hardware (most
4840 : * RISC architectures). The early dirtying is also good on the i386.
4841 : *
4842 : * There is also a hook called "update_mmu_cache()" that architectures
4843 : * with external mmu caches can use to update those (ie the Sparc or
4844 : * PowerPC hashed page tables that act as extended TLBs).
4845 : *
4846 : * We enter with non-exclusive mmap_lock (to exclude vma changes, but allow
4847 : * concurrent faults).
4848 : *
4849 : * The mmap_lock may have been released depending on flags and our return value.
4850 : * See filemap_fault() and __folio_lock_or_retry().
4851 : */
4852 0 : static vm_fault_t handle_pte_fault(struct vm_fault *vmf)
4853 : {
4854 : pte_t entry;
4855 :
4856 0 : if (unlikely(pmd_none(*vmf->pmd))) {
4857 : /*
4858 : * Leave __pte_alloc() until later: because vm_ops->fault may
4859 : * want to allocate huge page, and if we expose page table
4860 : * for an instant, it will be difficult to retract from
4861 : * concurrent faults and from rmap lookups.
4862 : */
4863 0 : vmf->pte = NULL;
4864 0 : vmf->flags &= ~FAULT_FLAG_ORIG_PTE_VALID;
4865 : } else {
4866 : /*
4867 : * If a huge pmd materialized under us just retry later. Use
4868 : * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead
4869 : * of pmd_trans_huge() to ensure the pmd didn't become
4870 : * pmd_trans_huge under us and then back to pmd_none, as a
4871 : * result of MADV_DONTNEED running immediately after a huge pmd
4872 : * fault in a different thread of this mm, in turn leading to a
4873 : * misleading pmd_trans_huge() retval. All we have to ensure is
4874 : * that it is a regular pmd that we can walk with
4875 : * pte_offset_map() and we can do that through an atomic read
4876 : * in C, which is what pmd_trans_unstable() provides.
4877 : */
4878 0 : if (pmd_devmap_trans_unstable(vmf->pmd))
4879 : return 0;
4880 : /*
4881 : * A regular pmd is established and it can't morph into a huge
4882 : * pmd from under us anymore at this point because we hold the
4883 : * mmap_lock read mode and khugepaged takes it in write mode.
4884 : * So now it's safe to run pte_offset_map().
4885 : */
4886 0 : vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
4887 0 : vmf->orig_pte = *vmf->pte;
4888 0 : vmf->flags |= FAULT_FLAG_ORIG_PTE_VALID;
4889 :
4890 : /*
4891 : * some architectures can have larger ptes than wordsize,
4892 : * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and
4893 : * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic
4894 : * accesses. The code below just needs a consistent view
4895 : * for the ifs and we later double check anyway with the
4896 : * ptl lock held. So here a barrier will do.
4897 : */
4898 0 : barrier();
4899 0 : if (pte_none(vmf->orig_pte)) {
4900 : pte_unmap(vmf->pte);
4901 0 : vmf->pte = NULL;
4902 : }
4903 : }
4904 :
4905 0 : if (!vmf->pte) {
4906 0 : if (vma_is_anonymous(vmf->vma))
4907 0 : return do_anonymous_page(vmf);
4908 : else
4909 0 : return do_fault(vmf);
4910 : }
4911 :
4912 0 : if (!pte_present(vmf->orig_pte))
4913 0 : return do_swap_page(vmf);
4914 :
4915 0 : if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma))
4916 : return do_numa_page(vmf);
4917 :
4918 0 : vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
4919 0 : spin_lock(vmf->ptl);
4920 0 : entry = vmf->orig_pte;
4921 0 : if (unlikely(!pte_same(*vmf->pte, entry))) {
4922 : update_mmu_tlb(vmf->vma, vmf->address, vmf->pte);
4923 : goto unlock;
4924 : }
4925 0 : if (vmf->flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) {
4926 0 : if (!pte_write(entry))
4927 0 : return do_wp_page(vmf);
4928 0 : else if (likely(vmf->flags & FAULT_FLAG_WRITE))
4929 : entry = pte_mkdirty(entry);
4930 : }
4931 0 : entry = pte_mkyoung(entry);
4932 0 : if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry,
4933 0 : vmf->flags & FAULT_FLAG_WRITE)) {
4934 : update_mmu_cache(vmf->vma, vmf->address, vmf->pte);
4935 : } else {
4936 : /* Skip spurious TLB flush for retried page fault */
4937 0 : if (vmf->flags & FAULT_FLAG_TRIED)
4938 : goto unlock;
4939 : /*
4940 : * This is needed only for protection faults but the arch code
4941 : * is not yet telling us if this is a protection fault or not.
4942 : * This still avoids useless tlb flushes for .text page faults
4943 : * with threads.
4944 : */
4945 0 : if (vmf->flags & FAULT_FLAG_WRITE)
4946 0 : flush_tlb_fix_spurious_fault(vmf->vma, vmf->address);
4947 : }
4948 : unlock:
4949 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
4950 0 : return 0;
4951 : }
4952 :
4953 : /*
4954 : * By the time we get here, we already hold the mm semaphore
4955 : *
4956 : * The mmap_lock may have been released depending on flags and our
4957 : * return value. See filemap_fault() and __folio_lock_or_retry().
4958 : */
4959 0 : static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma,
4960 : unsigned long address, unsigned int flags)
4961 : {
4962 0 : struct vm_fault vmf = {
4963 : .vma = vma,
4964 0 : .address = address & PAGE_MASK,
4965 : .real_address = address,
4966 : .flags = flags,
4967 0 : .pgoff = linear_page_index(vma, address),
4968 0 : .gfp_mask = __get_fault_gfp_mask(vma),
4969 : };
4970 0 : struct mm_struct *mm = vma->vm_mm;
4971 0 : unsigned long vm_flags = vma->vm_flags;
4972 : pgd_t *pgd;
4973 : p4d_t *p4d;
4974 : vm_fault_t ret;
4975 :
4976 0 : pgd = pgd_offset(mm, address);
4977 0 : p4d = p4d_alloc(mm, pgd, address);
4978 0 : if (!p4d)
4979 : return VM_FAULT_OOM;
4980 :
4981 0 : vmf.pud = pud_alloc(mm, p4d, address);
4982 : if (!vmf.pud)
4983 : return VM_FAULT_OOM;
4984 : retry_pud:
4985 : if (pud_none(*vmf.pud) &&
4986 : hugepage_vma_check(vma, vm_flags, false, true, true)) {
4987 : ret = create_huge_pud(&vmf);
4988 : if (!(ret & VM_FAULT_FALLBACK))
4989 : return ret;
4990 : } else {
4991 : pud_t orig_pud = *vmf.pud;
4992 :
4993 0 : barrier();
4994 0 : if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) {
4995 :
4996 : /*
4997 : * TODO once we support anonymous PUDs: NUMA case and
4998 : * FAULT_FLAG_UNSHARE handling.
4999 : */
5000 : if ((flags & FAULT_FLAG_WRITE) && !pud_write(orig_pud)) {
5001 : ret = wp_huge_pud(&vmf, orig_pud);
5002 : if (!(ret & VM_FAULT_FALLBACK))
5003 : return ret;
5004 : } else {
5005 : huge_pud_set_accessed(&vmf, orig_pud);
5006 : return 0;
5007 : }
5008 : }
5009 : }
5010 :
5011 0 : vmf.pmd = pmd_alloc(mm, vmf.pud, address);
5012 0 : if (!vmf.pmd)
5013 : return VM_FAULT_OOM;
5014 :
5015 : /* Huge pud page fault raced with pmd_alloc? */
5016 0 : if (pud_trans_unstable(vmf.pud))
5017 : goto retry_pud;
5018 :
5019 : if (pmd_none(*vmf.pmd) &&
5020 : hugepage_vma_check(vma, vm_flags, false, true, true)) {
5021 : ret = create_huge_pmd(&vmf);
5022 : if (!(ret & VM_FAULT_FALLBACK))
5023 : return ret;
5024 : } else {
5025 0 : vmf.orig_pmd = *vmf.pmd;
5026 :
5027 0 : barrier();
5028 0 : if (unlikely(is_swap_pmd(vmf.orig_pmd))) {
5029 : VM_BUG_ON(thp_migration_supported() &&
5030 : !is_pmd_migration_entry(vmf.orig_pmd));
5031 : if (is_pmd_migration_entry(vmf.orig_pmd))
5032 : pmd_migration_entry_wait(mm, vmf.pmd);
5033 : return 0;
5034 : }
5035 0 : if (pmd_trans_huge(vmf.orig_pmd) || pmd_devmap(vmf.orig_pmd)) {
5036 : if (pmd_protnone(vmf.orig_pmd) && vma_is_accessible(vma))
5037 : return do_huge_pmd_numa_page(&vmf);
5038 :
5039 : if ((flags & (FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE)) &&
5040 : !pmd_write(vmf.orig_pmd)) {
5041 : ret = wp_huge_pmd(&vmf);
5042 : if (!(ret & VM_FAULT_FALLBACK))
5043 : return ret;
5044 : } else {
5045 : huge_pmd_set_accessed(&vmf);
5046 : return 0;
5047 : }
5048 : }
5049 : }
5050 :
5051 0 : return handle_pte_fault(&vmf);
5052 : }
5053 :
5054 : /**
5055 : * mm_account_fault - Do page fault accounting
5056 : *
5057 : * @regs: the pt_regs struct pointer. When set to NULL, will skip accounting
5058 : * of perf event counters, but we'll still do the per-task accounting to
5059 : * the task who triggered this page fault.
5060 : * @address: the faulted address.
5061 : * @flags: the fault flags.
5062 : * @ret: the fault retcode.
5063 : *
5064 : * This will take care of most of the page fault accounting. Meanwhile, it
5065 : * will also include the PERF_COUNT_SW_PAGE_FAULTS_[MAJ|MIN] perf counter
5066 : * updates. However, note that the handling of PERF_COUNT_SW_PAGE_FAULTS should
5067 : * still be in per-arch page fault handlers at the entry of page fault.
5068 : */
5069 0 : static inline void mm_account_fault(struct pt_regs *regs,
5070 : unsigned long address, unsigned int flags,
5071 : vm_fault_t ret)
5072 : {
5073 : bool major;
5074 :
5075 : /*
5076 : * We don't do accounting for some specific faults:
5077 : *
5078 : * - Unsuccessful faults (e.g. when the address wasn't valid). That
5079 : * includes arch_vma_access_permitted() failing before reaching here.
5080 : * So this is not a "this many hardware page faults" counter. We
5081 : * should use the hw profiling for that.
5082 : *
5083 : * - Incomplete faults (VM_FAULT_RETRY). They will only be counted
5084 : * once they're completed.
5085 : */
5086 0 : if (ret & (VM_FAULT_ERROR | VM_FAULT_RETRY))
5087 : return;
5088 :
5089 : /*
5090 : * We define the fault as a major fault when the final successful fault
5091 : * is VM_FAULT_MAJOR, or if it retried (which implies that we couldn't
5092 : * handle it immediately previously).
5093 : */
5094 0 : major = (ret & VM_FAULT_MAJOR) || (flags & FAULT_FLAG_TRIED);
5095 :
5096 0 : if (major)
5097 0 : current->maj_flt++;
5098 : else
5099 0 : current->min_flt++;
5100 :
5101 : /*
5102 : * If the fault is done for GUP, regs will be NULL. We only do the
5103 : * accounting for the per thread fault counters who triggered the
5104 : * fault, and we skip the perf event updates.
5105 : */
5106 : if (!regs)
5107 : return;
5108 :
5109 : if (major)
5110 : perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
5111 : else
5112 : perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
5113 : }
5114 :
5115 : #ifdef CONFIG_LRU_GEN
5116 : static void lru_gen_enter_fault(struct vm_area_struct *vma)
5117 : {
5118 : /* the LRU algorithm only applies to accesses with recency */
5119 : current->in_lru_fault = vma_has_recency(vma);
5120 : }
5121 :
5122 : static void lru_gen_exit_fault(void)
5123 : {
5124 : current->in_lru_fault = false;
5125 : }
5126 : #else
5127 : static void lru_gen_enter_fault(struct vm_area_struct *vma)
5128 : {
5129 : }
5130 :
5131 : static void lru_gen_exit_fault(void)
5132 : {
5133 : }
5134 : #endif /* CONFIG_LRU_GEN */
5135 :
5136 0 : static vm_fault_t sanitize_fault_flags(struct vm_area_struct *vma,
5137 : unsigned int *flags)
5138 : {
5139 0 : if (unlikely(*flags & FAULT_FLAG_UNSHARE)) {
5140 0 : if (WARN_ON_ONCE(*flags & FAULT_FLAG_WRITE))
5141 : return VM_FAULT_SIGSEGV;
5142 : /*
5143 : * FAULT_FLAG_UNSHARE only applies to COW mappings. Let's
5144 : * just treat it like an ordinary read-fault otherwise.
5145 : */
5146 0 : if (!is_cow_mapping(vma->vm_flags))
5147 0 : *flags &= ~FAULT_FLAG_UNSHARE;
5148 0 : } else if (*flags & FAULT_FLAG_WRITE) {
5149 : /* Write faults on read-only mappings are impossible ... */
5150 0 : if (WARN_ON_ONCE(!(vma->vm_flags & VM_MAYWRITE)))
5151 : return VM_FAULT_SIGSEGV;
5152 : /* ... and FOLL_FORCE only applies to COW mappings. */
5153 0 : if (WARN_ON_ONCE(!(vma->vm_flags & VM_WRITE) &&
5154 : !is_cow_mapping(vma->vm_flags)))
5155 : return VM_FAULT_SIGSEGV;
5156 : }
5157 : return 0;
5158 : }
5159 :
5160 : /*
5161 : * By the time we get here, we already hold the mm semaphore
5162 : *
5163 : * The mmap_lock may have been released depending on flags and our
5164 : * return value. See filemap_fault() and __folio_lock_or_retry().
5165 : */
5166 0 : vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
5167 : unsigned int flags, struct pt_regs *regs)
5168 : {
5169 : vm_fault_t ret;
5170 :
5171 0 : __set_current_state(TASK_RUNNING);
5172 :
5173 0 : count_vm_event(PGFAULT);
5174 0 : count_memcg_event_mm(vma->vm_mm, PGFAULT);
5175 :
5176 0 : ret = sanitize_fault_flags(vma, &flags);
5177 0 : if (ret)
5178 : return ret;
5179 :
5180 0 : if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
5181 0 : flags & FAULT_FLAG_INSTRUCTION,
5182 0 : flags & FAULT_FLAG_REMOTE))
5183 : return VM_FAULT_SIGSEGV;
5184 :
5185 : /*
5186 : * Enable the memcg OOM handling for faults triggered in user
5187 : * space. Kernel faults are handled more gracefully.
5188 : */
5189 : if (flags & FAULT_FLAG_USER)
5190 : mem_cgroup_enter_user_fault();
5191 :
5192 0 : lru_gen_enter_fault(vma);
5193 :
5194 0 : if (unlikely(is_vm_hugetlb_page(vma)))
5195 : ret = hugetlb_fault(vma->vm_mm, vma, address, flags);
5196 : else
5197 0 : ret = __handle_mm_fault(vma, address, flags);
5198 :
5199 : lru_gen_exit_fault();
5200 :
5201 0 : if (flags & FAULT_FLAG_USER) {
5202 : mem_cgroup_exit_user_fault();
5203 : /*
5204 : * The task may have entered a memcg OOM situation but
5205 : * if the allocation error was handled gracefully (no
5206 : * VM_FAULT_OOM), there is no need to kill anything.
5207 : * Just clean up the OOM state peacefully.
5208 : */
5209 0 : if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
5210 : mem_cgroup_oom_synchronize(false);
5211 : }
5212 :
5213 0 : mm_account_fault(regs, address, flags, ret);
5214 :
5215 0 : return ret;
5216 : }
5217 : EXPORT_SYMBOL_GPL(handle_mm_fault);
5218 :
5219 : #ifndef __PAGETABLE_P4D_FOLDED
5220 : /*
5221 : * Allocate p4d page table.
5222 : * We've already handled the fast-path in-line.
5223 : */
5224 : int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
5225 : {
5226 : p4d_t *new = p4d_alloc_one(mm, address);
5227 : if (!new)
5228 : return -ENOMEM;
5229 :
5230 : spin_lock(&mm->page_table_lock);
5231 : if (pgd_present(*pgd)) { /* Another has populated it */
5232 : p4d_free(mm, new);
5233 : } else {
5234 : smp_wmb(); /* See comment in pmd_install() */
5235 : pgd_populate(mm, pgd, new);
5236 : }
5237 : spin_unlock(&mm->page_table_lock);
5238 : return 0;
5239 : }
5240 : #endif /* __PAGETABLE_P4D_FOLDED */
5241 :
5242 : #ifndef __PAGETABLE_PUD_FOLDED
5243 : /*
5244 : * Allocate page upper directory.
5245 : * We've already handled the fast-path in-line.
5246 : */
5247 : int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address)
5248 : {
5249 : pud_t *new = pud_alloc_one(mm, address);
5250 : if (!new)
5251 : return -ENOMEM;
5252 :
5253 : spin_lock(&mm->page_table_lock);
5254 : if (!p4d_present(*p4d)) {
5255 : mm_inc_nr_puds(mm);
5256 : smp_wmb(); /* See comment in pmd_install() */
5257 : p4d_populate(mm, p4d, new);
5258 : } else /* Another has populated it */
5259 : pud_free(mm, new);
5260 : spin_unlock(&mm->page_table_lock);
5261 : return 0;
5262 : }
5263 : #endif /* __PAGETABLE_PUD_FOLDED */
5264 :
5265 : #ifndef __PAGETABLE_PMD_FOLDED
5266 : /*
5267 : * Allocate page middle directory.
5268 : * We've already handled the fast-path in-line.
5269 : */
5270 1 : int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
5271 : {
5272 : spinlock_t *ptl;
5273 1 : pmd_t *new = pmd_alloc_one(mm, address);
5274 1 : if (!new)
5275 : return -ENOMEM;
5276 :
5277 2 : ptl = pud_lock(mm, pud);
5278 1 : if (!pud_present(*pud)) {
5279 1 : mm_inc_nr_pmds(mm);
5280 1 : smp_wmb(); /* See comment in pmd_install() */
5281 1 : pud_populate(mm, pud, new);
5282 : } else { /* Another has populated it */
5283 0 : pmd_free(mm, new);
5284 : }
5285 1 : spin_unlock(ptl);
5286 1 : return 0;
5287 : }
5288 : #endif /* __PAGETABLE_PMD_FOLDED */
5289 :
5290 : /**
5291 : * follow_pte - look up PTE at a user virtual address
5292 : * @mm: the mm_struct of the target address space
5293 : * @address: user virtual address
5294 : * @ptepp: location to store found PTE
5295 : * @ptlp: location to store the lock for the PTE
5296 : *
5297 : * On a successful return, the pointer to the PTE is stored in @ptepp;
5298 : * the corresponding lock is taken and its location is stored in @ptlp.
5299 : * The contents of the PTE are only stable until @ptlp is released;
5300 : * any further use, if any, must be protected against invalidation
5301 : * with MMU notifiers.
5302 : *
5303 : * Only IO mappings and raw PFN mappings are allowed. The mmap semaphore
5304 : * should be taken for read.
5305 : *
5306 : * KVM uses this function. While it is arguably less bad than ``follow_pfn``,
5307 : * it is not a good general-purpose API.
5308 : *
5309 : * Return: zero on success, -ve otherwise.
5310 : */
5311 0 : int follow_pte(struct mm_struct *mm, unsigned long address,
5312 : pte_t **ptepp, spinlock_t **ptlp)
5313 : {
5314 : pgd_t *pgd;
5315 : p4d_t *p4d;
5316 : pud_t *pud;
5317 : pmd_t *pmd;
5318 : pte_t *ptep;
5319 :
5320 0 : pgd = pgd_offset(mm, address);
5321 : if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
5322 : goto out;
5323 :
5324 0 : p4d = p4d_offset(pgd, address);
5325 : if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d)))
5326 : goto out;
5327 :
5328 0 : pud = pud_offset(p4d, address);
5329 0 : if (pud_none(*pud) || unlikely(pud_bad(*pud)))
5330 : goto out;
5331 :
5332 0 : pmd = pmd_offset(pud, address);
5333 : VM_BUG_ON(pmd_trans_huge(*pmd));
5334 :
5335 0 : if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
5336 : goto out;
5337 :
5338 0 : ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
5339 0 : if (!pte_present(*ptep))
5340 : goto unlock;
5341 0 : *ptepp = ptep;
5342 0 : return 0;
5343 : unlock:
5344 0 : pte_unmap_unlock(ptep, *ptlp);
5345 : out:
5346 : return -EINVAL;
5347 : }
5348 : EXPORT_SYMBOL_GPL(follow_pte);
5349 :
5350 : /**
5351 : * follow_pfn - look up PFN at a user virtual address
5352 : * @vma: memory mapping
5353 : * @address: user virtual address
5354 : * @pfn: location to store found PFN
5355 : *
5356 : * Only IO mappings and raw PFN mappings are allowed.
5357 : *
5358 : * This function does not allow the caller to read the permissions
5359 : * of the PTE. Do not use it.
5360 : *
5361 : * Return: zero and the pfn at @pfn on success, -ve otherwise.
5362 : */
5363 0 : int follow_pfn(struct vm_area_struct *vma, unsigned long address,
5364 : unsigned long *pfn)
5365 : {
5366 0 : int ret = -EINVAL;
5367 : spinlock_t *ptl;
5368 : pte_t *ptep;
5369 :
5370 0 : if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
5371 : return ret;
5372 :
5373 0 : ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
5374 0 : if (ret)
5375 : return ret;
5376 0 : *pfn = pte_pfn(*ptep);
5377 0 : pte_unmap_unlock(ptep, ptl);
5378 0 : return 0;
5379 : }
5380 : EXPORT_SYMBOL(follow_pfn);
5381 :
5382 : #ifdef CONFIG_HAVE_IOREMAP_PROT
5383 : int follow_phys(struct vm_area_struct *vma,
5384 : unsigned long address, unsigned int flags,
5385 : unsigned long *prot, resource_size_t *phys)
5386 : {
5387 : int ret = -EINVAL;
5388 : pte_t *ptep, pte;
5389 : spinlock_t *ptl;
5390 :
5391 : if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
5392 : goto out;
5393 :
5394 : if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
5395 : goto out;
5396 : pte = *ptep;
5397 :
5398 : if ((flags & FOLL_WRITE) && !pte_write(pte))
5399 : goto unlock;
5400 :
5401 : *prot = pgprot_val(pte_pgprot(pte));
5402 : *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
5403 :
5404 : ret = 0;
5405 : unlock:
5406 : pte_unmap_unlock(ptep, ptl);
5407 : out:
5408 : return ret;
5409 : }
5410 :
5411 : /**
5412 : * generic_access_phys - generic implementation for iomem mmap access
5413 : * @vma: the vma to access
5414 : * @addr: userspace address, not relative offset within @vma
5415 : * @buf: buffer to read/write
5416 : * @len: length of transfer
5417 : * @write: set to FOLL_WRITE when writing, otherwise reading
5418 : *
5419 : * This is a generic implementation for &vm_operations_struct.access for an
5420 : * iomem mapping. This callback is used by access_process_vm() when the @vma is
5421 : * not page based.
5422 : */
5423 : int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
5424 : void *buf, int len, int write)
5425 : {
5426 : resource_size_t phys_addr;
5427 : unsigned long prot = 0;
5428 : void __iomem *maddr;
5429 : pte_t *ptep, pte;
5430 : spinlock_t *ptl;
5431 : int offset = offset_in_page(addr);
5432 : int ret = -EINVAL;
5433 :
5434 : if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
5435 : return -EINVAL;
5436 :
5437 : retry:
5438 : if (follow_pte(vma->vm_mm, addr, &ptep, &ptl))
5439 : return -EINVAL;
5440 : pte = *ptep;
5441 : pte_unmap_unlock(ptep, ptl);
5442 :
5443 : prot = pgprot_val(pte_pgprot(pte));
5444 : phys_addr = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
5445 :
5446 : if ((write & FOLL_WRITE) && !pte_write(pte))
5447 : return -EINVAL;
5448 :
5449 : maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot);
5450 : if (!maddr)
5451 : return -ENOMEM;
5452 :
5453 : if (follow_pte(vma->vm_mm, addr, &ptep, &ptl))
5454 : goto out_unmap;
5455 :
5456 : if (!pte_same(pte, *ptep)) {
5457 : pte_unmap_unlock(ptep, ptl);
5458 : iounmap(maddr);
5459 :
5460 : goto retry;
5461 : }
5462 :
5463 : if (write)
5464 : memcpy_toio(maddr + offset, buf, len);
5465 : else
5466 : memcpy_fromio(buf, maddr + offset, len);
5467 : ret = len;
5468 : pte_unmap_unlock(ptep, ptl);
5469 : out_unmap:
5470 : iounmap(maddr);
5471 :
5472 : return ret;
5473 : }
5474 : EXPORT_SYMBOL_GPL(generic_access_phys);
5475 : #endif
5476 :
5477 : /*
5478 : * Access another process' address space as given in mm.
5479 : */
5480 0 : int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf,
5481 : int len, unsigned int gup_flags)
5482 : {
5483 : struct vm_area_struct *vma;
5484 0 : void *old_buf = buf;
5485 0 : int write = gup_flags & FOLL_WRITE;
5486 :
5487 0 : if (mmap_read_lock_killable(mm))
5488 : return 0;
5489 :
5490 : /* ignore errors, just check how much was successfully transferred */
5491 0 : while (len) {
5492 : int bytes, ret, offset;
5493 : void *maddr;
5494 0 : struct page *page = NULL;
5495 :
5496 0 : ret = get_user_pages_remote(mm, addr, 1,
5497 : gup_flags, &page, &vma, NULL);
5498 0 : if (ret <= 0) {
5499 : #ifndef CONFIG_HAVE_IOREMAP_PROT
5500 : break;
5501 : #else
5502 : /*
5503 : * Check if this is a VM_IO | VM_PFNMAP VMA, which
5504 : * we can access using slightly different code.
5505 : */
5506 : vma = vma_lookup(mm, addr);
5507 : if (!vma)
5508 : break;
5509 : if (vma->vm_ops && vma->vm_ops->access)
5510 : ret = vma->vm_ops->access(vma, addr, buf,
5511 : len, write);
5512 : if (ret <= 0)
5513 : break;
5514 : bytes = ret;
5515 : #endif
5516 : } else {
5517 0 : bytes = len;
5518 0 : offset = addr & (PAGE_SIZE-1);
5519 0 : if (bytes > PAGE_SIZE-offset)
5520 0 : bytes = PAGE_SIZE-offset;
5521 :
5522 0 : maddr = kmap(page);
5523 0 : if (write) {
5524 0 : copy_to_user_page(vma, page, addr,
5525 : maddr + offset, buf, bytes);
5526 0 : set_page_dirty_lock(page);
5527 : } else {
5528 0 : copy_from_user_page(vma, page, addr,
5529 : buf, maddr + offset, bytes);
5530 : }
5531 0 : kunmap(page);
5532 0 : put_page(page);
5533 : }
5534 0 : len -= bytes;
5535 0 : buf += bytes;
5536 0 : addr += bytes;
5537 : }
5538 0 : mmap_read_unlock(mm);
5539 :
5540 0 : return buf - old_buf;
5541 : }
5542 :
5543 : /**
5544 : * access_remote_vm - access another process' address space
5545 : * @mm: the mm_struct of the target address space
5546 : * @addr: start address to access
5547 : * @buf: source or destination buffer
5548 : * @len: number of bytes to transfer
5549 : * @gup_flags: flags modifying lookup behaviour
5550 : *
5551 : * The caller must hold a reference on @mm.
5552 : *
5553 : * Return: number of bytes copied from source to destination.
5554 : */
5555 0 : int access_remote_vm(struct mm_struct *mm, unsigned long addr,
5556 : void *buf, int len, unsigned int gup_flags)
5557 : {
5558 0 : return __access_remote_vm(mm, addr, buf, len, gup_flags);
5559 : }
5560 :
5561 : /*
5562 : * Access another process' address space.
5563 : * Source/target buffer must be kernel space,
5564 : * Do not walk the page table directly, use get_user_pages
5565 : */
5566 0 : int access_process_vm(struct task_struct *tsk, unsigned long addr,
5567 : void *buf, int len, unsigned int gup_flags)
5568 : {
5569 : struct mm_struct *mm;
5570 : int ret;
5571 :
5572 0 : mm = get_task_mm(tsk);
5573 0 : if (!mm)
5574 : return 0;
5575 :
5576 0 : ret = __access_remote_vm(mm, addr, buf, len, gup_flags);
5577 :
5578 0 : mmput(mm);
5579 :
5580 0 : return ret;
5581 : }
5582 : EXPORT_SYMBOL_GPL(access_process_vm);
5583 :
5584 : /*
5585 : * Print the name of a VMA.
5586 : */
5587 0 : void print_vma_addr(char *prefix, unsigned long ip)
5588 : {
5589 0 : struct mm_struct *mm = current->mm;
5590 : struct vm_area_struct *vma;
5591 :
5592 : /*
5593 : * we might be running from an atomic context so we cannot sleep
5594 : */
5595 0 : if (!mmap_read_trylock(mm))
5596 : return;
5597 :
5598 0 : vma = find_vma(mm, ip);
5599 0 : if (vma && vma->vm_file) {
5600 0 : struct file *f = vma->vm_file;
5601 0 : char *buf = (char *)__get_free_page(GFP_NOWAIT);
5602 0 : if (buf) {
5603 : char *p;
5604 :
5605 0 : p = file_path(f, buf, PAGE_SIZE);
5606 0 : if (IS_ERR(p))
5607 0 : p = "?";
5608 0 : printk("%s%s[%lx+%lx]", prefix, kbasename(p),
5609 : vma->vm_start,
5610 : vma->vm_end - vma->vm_start);
5611 0 : free_page((unsigned long)buf);
5612 : }
5613 : }
5614 : mmap_read_unlock(mm);
5615 : }
5616 :
5617 : #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)
5618 : void __might_fault(const char *file, int line)
5619 : {
5620 : if (pagefault_disabled())
5621 : return;
5622 : __might_sleep(file, line);
5623 : #if defined(CONFIG_DEBUG_ATOMIC_SLEEP)
5624 : if (current->mm)
5625 : might_lock_read(¤t->mm->mmap_lock);
5626 : #endif
5627 : }
5628 : EXPORT_SYMBOL(__might_fault);
5629 : #endif
5630 :
5631 : #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
5632 : /*
5633 : * Process all subpages of the specified huge page with the specified
5634 : * operation. The target subpage will be processed last to keep its
5635 : * cache lines hot.
5636 : */
5637 : static inline void process_huge_page(
5638 : unsigned long addr_hint, unsigned int pages_per_huge_page,
5639 : void (*process_subpage)(unsigned long addr, int idx, void *arg),
5640 : void *arg)
5641 : {
5642 : int i, n, base, l;
5643 : unsigned long addr = addr_hint &
5644 : ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1);
5645 :
5646 : /* Process target subpage last to keep its cache lines hot */
5647 : might_sleep();
5648 : n = (addr_hint - addr) / PAGE_SIZE;
5649 : if (2 * n <= pages_per_huge_page) {
5650 : /* If target subpage in first half of huge page */
5651 : base = 0;
5652 : l = n;
5653 : /* Process subpages at the end of huge page */
5654 : for (i = pages_per_huge_page - 1; i >= 2 * n; i--) {
5655 : cond_resched();
5656 : process_subpage(addr + i * PAGE_SIZE, i, arg);
5657 : }
5658 : } else {
5659 : /* If target subpage in second half of huge page */
5660 : base = pages_per_huge_page - 2 * (pages_per_huge_page - n);
5661 : l = pages_per_huge_page - n;
5662 : /* Process subpages at the begin of huge page */
5663 : for (i = 0; i < base; i++) {
5664 : cond_resched();
5665 : process_subpage(addr + i * PAGE_SIZE, i, arg);
5666 : }
5667 : }
5668 : /*
5669 : * Process remaining subpages in left-right-left-right pattern
5670 : * towards the target subpage
5671 : */
5672 : for (i = 0; i < l; i++) {
5673 : int left_idx = base + i;
5674 : int right_idx = base + 2 * l - 1 - i;
5675 :
5676 : cond_resched();
5677 : process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg);
5678 : cond_resched();
5679 : process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg);
5680 : }
5681 : }
5682 :
5683 : static void clear_gigantic_page(struct page *page,
5684 : unsigned long addr,
5685 : unsigned int pages_per_huge_page)
5686 : {
5687 : int i;
5688 : struct page *p;
5689 :
5690 : might_sleep();
5691 : for (i = 0; i < pages_per_huge_page; i++) {
5692 : p = nth_page(page, i);
5693 : cond_resched();
5694 : clear_user_highpage(p, addr + i * PAGE_SIZE);
5695 : }
5696 : }
5697 :
5698 : static void clear_subpage(unsigned long addr, int idx, void *arg)
5699 : {
5700 : struct page *page = arg;
5701 :
5702 : clear_user_highpage(page + idx, addr);
5703 : }
5704 :
5705 : void clear_huge_page(struct page *page,
5706 : unsigned long addr_hint, unsigned int pages_per_huge_page)
5707 : {
5708 : unsigned long addr = addr_hint &
5709 : ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1);
5710 :
5711 : if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
5712 : clear_gigantic_page(page, addr, pages_per_huge_page);
5713 : return;
5714 : }
5715 :
5716 : process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page);
5717 : }
5718 :
5719 : static void copy_user_gigantic_page(struct page *dst, struct page *src,
5720 : unsigned long addr,
5721 : struct vm_area_struct *vma,
5722 : unsigned int pages_per_huge_page)
5723 : {
5724 : int i;
5725 : struct page *dst_base = dst;
5726 : struct page *src_base = src;
5727 :
5728 : for (i = 0; i < pages_per_huge_page; i++) {
5729 : dst = nth_page(dst_base, i);
5730 : src = nth_page(src_base, i);
5731 :
5732 : cond_resched();
5733 : copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
5734 : }
5735 : }
5736 :
5737 : struct copy_subpage_arg {
5738 : struct page *dst;
5739 : struct page *src;
5740 : struct vm_area_struct *vma;
5741 : };
5742 :
5743 : static void copy_subpage(unsigned long addr, int idx, void *arg)
5744 : {
5745 : struct copy_subpage_arg *copy_arg = arg;
5746 :
5747 : copy_user_highpage(copy_arg->dst + idx, copy_arg->src + idx,
5748 : addr, copy_arg->vma);
5749 : }
5750 :
5751 : void copy_user_huge_page(struct page *dst, struct page *src,
5752 : unsigned long addr_hint, struct vm_area_struct *vma,
5753 : unsigned int pages_per_huge_page)
5754 : {
5755 : unsigned long addr = addr_hint &
5756 : ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1);
5757 : struct copy_subpage_arg arg = {
5758 : .dst = dst,
5759 : .src = src,
5760 : .vma = vma,
5761 : };
5762 :
5763 : if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
5764 : copy_user_gigantic_page(dst, src, addr, vma,
5765 : pages_per_huge_page);
5766 : return;
5767 : }
5768 :
5769 : process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg);
5770 : }
5771 :
5772 : long copy_huge_page_from_user(struct page *dst_page,
5773 : const void __user *usr_src,
5774 : unsigned int pages_per_huge_page,
5775 : bool allow_pagefault)
5776 : {
5777 : void *page_kaddr;
5778 : unsigned long i, rc = 0;
5779 : unsigned long ret_val = pages_per_huge_page * PAGE_SIZE;
5780 : struct page *subpage;
5781 :
5782 : for (i = 0; i < pages_per_huge_page; i++) {
5783 : subpage = nth_page(dst_page, i);
5784 : if (allow_pagefault)
5785 : page_kaddr = kmap(subpage);
5786 : else
5787 : page_kaddr = kmap_atomic(subpage);
5788 : rc = copy_from_user(page_kaddr,
5789 : usr_src + i * PAGE_SIZE, PAGE_SIZE);
5790 : if (allow_pagefault)
5791 : kunmap(subpage);
5792 : else
5793 : kunmap_atomic(page_kaddr);
5794 :
5795 : ret_val -= (PAGE_SIZE - rc);
5796 : if (rc)
5797 : break;
5798 :
5799 : flush_dcache_page(subpage);
5800 :
5801 : cond_resched();
5802 : }
5803 : return ret_val;
5804 : }
5805 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
5806 :
5807 : #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS
5808 :
5809 : static struct kmem_cache *page_ptl_cachep;
5810 :
5811 : void __init ptlock_cache_init(void)
5812 : {
5813 : page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0,
5814 : SLAB_PANIC, NULL);
5815 : }
5816 :
5817 : bool ptlock_alloc(struct page *page)
5818 : {
5819 : spinlock_t *ptl;
5820 :
5821 : ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL);
5822 : if (!ptl)
5823 : return false;
5824 : page->ptl = ptl;
5825 : return true;
5826 : }
5827 :
5828 : void ptlock_free(struct page *page)
5829 : {
5830 : kmem_cache_free(page_ptl_cachep, page->ptl);
5831 : }
5832 : #endif
|
