Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : #include <linux/kernel.h>
3 : #include <linux/errno.h>
4 : #include <linux/err.h>
5 : #include <linux/spinlock.h>
6 :
7 : #include <linux/mm.h>
8 : #include <linux/memremap.h>
9 : #include <linux/pagemap.h>
10 : #include <linux/rmap.h>
11 : #include <linux/swap.h>
12 : #include <linux/swapops.h>
13 : #include <linux/secretmem.h>
14 :
15 : #include <linux/sched/signal.h>
16 : #include <linux/rwsem.h>
17 : #include <linux/hugetlb.h>
18 : #include <linux/migrate.h>
19 : #include <linux/mm_inline.h>
20 : #include <linux/sched/mm.h>
21 : #include <linux/shmem_fs.h>
22 :
23 : #include <asm/mmu_context.h>
24 : #include <asm/tlbflush.h>
25 :
26 : #include "internal.h"
27 :
28 : struct follow_page_context {
29 : struct dev_pagemap *pgmap;
30 : unsigned int page_mask;
31 : };
32 :
33 : static inline void sanity_check_pinned_pages(struct page **pages,
34 : unsigned long npages)
35 : {
36 : if (!IS_ENABLED(CONFIG_DEBUG_VM))
37 : return;
38 :
39 : /*
40 : * We only pin anonymous pages if they are exclusive. Once pinned, we
41 : * can no longer turn them possibly shared and PageAnonExclusive() will
42 : * stick around until the page is freed.
43 : *
44 : * We'd like to verify that our pinned anonymous pages are still mapped
45 : * exclusively. The issue with anon THP is that we don't know how
46 : * they are/were mapped when pinning them. However, for anon
47 : * THP we can assume that either the given page (PTE-mapped THP) or
48 : * the head page (PMD-mapped THP) should be PageAnonExclusive(). If
49 : * neither is the case, there is certainly something wrong.
50 : */
51 : for (; npages; npages--, pages++) {
52 : struct page *page = *pages;
53 : struct folio *folio = page_folio(page);
54 :
55 : if (is_zero_page(page) ||
56 : !folio_test_anon(folio))
57 : continue;
58 : if (!folio_test_large(folio) || folio_test_hugetlb(folio))
59 : VM_BUG_ON_PAGE(!PageAnonExclusive(&folio->page), page);
60 : else
61 : /* Either a PTE-mapped or a PMD-mapped THP. */
62 : VM_BUG_ON_PAGE(!PageAnonExclusive(&folio->page) &&
63 : !PageAnonExclusive(page), page);
64 : }
65 : }
66 :
67 : /*
68 : * Return the folio with ref appropriately incremented,
69 : * or NULL if that failed.
70 : */
71 0 : static inline struct folio *try_get_folio(struct page *page, int refs)
72 : {
73 : struct folio *folio;
74 :
75 : retry:
76 0 : folio = page_folio(page);
77 0 : if (WARN_ON_ONCE(folio_ref_count(folio) < 0))
78 : return NULL;
79 0 : if (unlikely(!folio_ref_try_add_rcu(folio, refs)))
80 : return NULL;
81 :
82 : /*
83 : * At this point we have a stable reference to the folio; but it
84 : * could be that between calling page_folio() and the refcount
85 : * increment, the folio was split, in which case we'd end up
86 : * holding a reference on a folio that has nothing to do with the page
87 : * we were given anymore.
88 : * So now that the folio is stable, recheck that the page still
89 : * belongs to this folio.
90 : */
91 0 : if (unlikely(page_folio(page) != folio)) {
92 0 : if (!put_devmap_managed_page_refs(&folio->page, refs))
93 : folio_put_refs(folio, refs);
94 : goto retry;
95 : }
96 :
97 : return folio;
98 : }
99 :
100 : /**
101 : * try_grab_folio() - Attempt to get or pin a folio.
102 : * @page: pointer to page to be grabbed
103 : * @refs: the value to (effectively) add to the folio's refcount
104 : * @flags: gup flags: these are the FOLL_* flag values.
105 : *
106 : * "grab" names in this file mean, "look at flags to decide whether to use
107 : * FOLL_PIN or FOLL_GET behavior, when incrementing the folio's refcount.
108 : *
109 : * Either FOLL_PIN or FOLL_GET (or neither) must be set, but not both at the
110 : * same time. (That's true throughout the get_user_pages*() and
111 : * pin_user_pages*() APIs.) Cases:
112 : *
113 : * FOLL_GET: folio's refcount will be incremented by @refs.
114 : *
115 : * FOLL_PIN on large folios: folio's refcount will be incremented by
116 : * @refs, and its pincount will be incremented by @refs.
117 : *
118 : * FOLL_PIN on single-page folios: folio's refcount will be incremented by
119 : * @refs * GUP_PIN_COUNTING_BIAS.
120 : *
121 : * Return: The folio containing @page (with refcount appropriately
122 : * incremented) for success, or NULL upon failure. If neither FOLL_GET
123 : * nor FOLL_PIN was set, that's considered failure, and furthermore,
124 : * a likely bug in the caller, so a warning is also emitted.
125 : */
126 0 : struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags)
127 : {
128 : struct folio *folio;
129 :
130 0 : if (WARN_ON_ONCE((flags & (FOLL_GET | FOLL_PIN)) == 0))
131 : return NULL;
132 :
133 : if (unlikely(!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page)))
134 : return NULL;
135 :
136 0 : if (flags & FOLL_GET)
137 0 : return try_get_folio(page, refs);
138 :
139 : /* FOLL_PIN is set */
140 :
141 : /*
142 : * Don't take a pin on the zero page - it's not going anywhere
143 : * and it is used in a *lot* of places.
144 : */
145 0 : if (is_zero_page(page))
146 0 : return page_folio(page);
147 :
148 0 : folio = try_get_folio(page, refs);
149 0 : if (!folio)
150 : return NULL;
151 :
152 : /*
153 : * Can't do FOLL_LONGTERM + FOLL_PIN gup fast path if not in a
154 : * right zone, so fail and let the caller fall back to the slow
155 : * path.
156 : */
157 0 : if (unlikely((flags & FOLL_LONGTERM) &&
158 : !folio_is_longterm_pinnable(folio))) {
159 0 : if (!put_devmap_managed_page_refs(&folio->page, refs))
160 : folio_put_refs(folio, refs);
161 : return NULL;
162 : }
163 :
164 : /*
165 : * When pinning a large folio, use an exact count to track it.
166 : *
167 : * However, be sure to *also* increment the normal folio
168 : * refcount field at least once, so that the folio really
169 : * is pinned. That's why the refcount from the earlier
170 : * try_get_folio() is left intact.
171 : */
172 0 : if (folio_test_large(folio))
173 0 : atomic_add(refs, &folio->_pincount);
174 : else
175 0 : folio_ref_add(folio,
176 0 : refs * (GUP_PIN_COUNTING_BIAS - 1));
177 : /*
178 : * Adjust the pincount before re-checking the PTE for changes.
179 : * This is essentially a smp_mb() and is paired with a memory
180 : * barrier in page_try_share_anon_rmap().
181 : */
182 0 : smp_mb__after_atomic();
183 :
184 0 : node_stat_mod_folio(folio, NR_FOLL_PIN_ACQUIRED, refs);
185 :
186 0 : return folio;
187 : }
188 :
189 0 : static void gup_put_folio(struct folio *folio, int refs, unsigned int flags)
190 : {
191 0 : if (flags & FOLL_PIN) {
192 0 : if (is_zero_folio(folio))
193 : return;
194 0 : node_stat_mod_folio(folio, NR_FOLL_PIN_RELEASED, refs);
195 0 : if (folio_test_large(folio))
196 0 : atomic_sub(refs, &folio->_pincount);
197 : else
198 0 : refs *= GUP_PIN_COUNTING_BIAS;
199 : }
200 :
201 0 : if (!put_devmap_managed_page_refs(&folio->page, refs))
202 : folio_put_refs(folio, refs);
203 : }
204 :
205 : /**
206 : * try_grab_page() - elevate a page's refcount by a flag-dependent amount
207 : * @page: pointer to page to be grabbed
208 : * @flags: gup flags: these are the FOLL_* flag values.
209 : *
210 : * This might not do anything at all, depending on the flags argument.
211 : *
212 : * "grab" names in this file mean, "look at flags to decide whether to use
213 : * FOLL_PIN or FOLL_GET behavior, when incrementing the page's refcount.
214 : *
215 : * Either FOLL_PIN or FOLL_GET (or neither) may be set, but not both at the same
216 : * time. Cases: please see the try_grab_folio() documentation, with
217 : * "refs=1".
218 : *
219 : * Return: 0 for success, or if no action was required (if neither FOLL_PIN
220 : * nor FOLL_GET was set, nothing is done). A negative error code for failure:
221 : *
222 : * -ENOMEM FOLL_GET or FOLL_PIN was set, but the page could not
223 : * be grabbed.
224 : */
225 0 : int __must_check try_grab_page(struct page *page, unsigned int flags)
226 : {
227 0 : struct folio *folio = page_folio(page);
228 :
229 0 : if (WARN_ON_ONCE(folio_ref_count(folio) <= 0))
230 : return -ENOMEM;
231 :
232 : if (unlikely(!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page)))
233 : return -EREMOTEIO;
234 :
235 0 : if (flags & FOLL_GET)
236 : folio_ref_inc(folio);
237 0 : else if (flags & FOLL_PIN) {
238 : /*
239 : * Don't take a pin on the zero page - it's not going anywhere
240 : * and it is used in a *lot* of places.
241 : */
242 0 : if (is_zero_page(page))
243 : return 0;
244 :
245 : /*
246 : * Similar to try_grab_folio(): be sure to *also*
247 : * increment the normal page refcount field at least once,
248 : * so that the page really is pinned.
249 : */
250 0 : if (folio_test_large(folio)) {
251 0 : folio_ref_add(folio, 1);
252 0 : atomic_add(1, &folio->_pincount);
253 : } else {
254 : folio_ref_add(folio, GUP_PIN_COUNTING_BIAS);
255 : }
256 :
257 0 : node_stat_mod_folio(folio, NR_FOLL_PIN_ACQUIRED, 1);
258 : }
259 :
260 : return 0;
261 : }
262 :
263 : /**
264 : * unpin_user_page() - release a dma-pinned page
265 : * @page: pointer to page to be released
266 : *
267 : * Pages that were pinned via pin_user_pages*() must be released via either
268 : * unpin_user_page(), or one of the unpin_user_pages*() routines. This is so
269 : * that such pages can be separately tracked and uniquely handled. In
270 : * particular, interactions with RDMA and filesystems need special handling.
271 : */
272 0 : void unpin_user_page(struct page *page)
273 : {
274 0 : sanity_check_pinned_pages(&page, 1);
275 0 : gup_put_folio(page_folio(page), 1, FOLL_PIN);
276 0 : }
277 : EXPORT_SYMBOL(unpin_user_page);
278 :
279 : /**
280 : * folio_add_pin - Try to get an additional pin on a pinned folio
281 : * @folio: The folio to be pinned
282 : *
283 : * Get an additional pin on a folio we already have a pin on. Makes no change
284 : * if the folio is a zero_page.
285 : */
286 0 : void folio_add_pin(struct folio *folio)
287 : {
288 0 : if (is_zero_folio(folio))
289 : return;
290 :
291 : /*
292 : * Similar to try_grab_folio(): be sure to *also* increment the normal
293 : * page refcount field at least once, so that the page really is
294 : * pinned.
295 : */
296 0 : if (folio_test_large(folio)) {
297 0 : WARN_ON_ONCE(atomic_read(&folio->_pincount) < 1);
298 0 : folio_ref_inc(folio);
299 0 : atomic_inc(&folio->_pincount);
300 : } else {
301 0 : WARN_ON_ONCE(folio_ref_count(folio) < GUP_PIN_COUNTING_BIAS);
302 : folio_ref_add(folio, GUP_PIN_COUNTING_BIAS);
303 : }
304 : }
305 :
306 0 : static inline struct folio *gup_folio_range_next(struct page *start,
307 : unsigned long npages, unsigned long i, unsigned int *ntails)
308 : {
309 0 : struct page *next = nth_page(start, i);
310 0 : struct folio *folio = page_folio(next);
311 0 : unsigned int nr = 1;
312 :
313 0 : if (folio_test_large(folio))
314 0 : nr = min_t(unsigned int, npages - i,
315 : folio_nr_pages(folio) - folio_page_idx(folio, next));
316 :
317 0 : *ntails = nr;
318 0 : return folio;
319 : }
320 :
321 0 : static inline struct folio *gup_folio_next(struct page **list,
322 : unsigned long npages, unsigned long i, unsigned int *ntails)
323 : {
324 0 : struct folio *folio = page_folio(list[i]);
325 : unsigned int nr;
326 :
327 0 : for (nr = i + 1; nr < npages; nr++) {
328 0 : if (page_folio(list[nr]) != folio)
329 : break;
330 : }
331 :
332 0 : *ntails = nr - i;
333 0 : return folio;
334 : }
335 :
336 : /**
337 : * unpin_user_pages_dirty_lock() - release and optionally dirty gup-pinned pages
338 : * @pages: array of pages to be maybe marked dirty, and definitely released.
339 : * @npages: number of pages in the @pages array.
340 : * @make_dirty: whether to mark the pages dirty
341 : *
342 : * "gup-pinned page" refers to a page that has had one of the get_user_pages()
343 : * variants called on that page.
344 : *
345 : * For each page in the @pages array, make that page (or its head page, if a
346 : * compound page) dirty, if @make_dirty is true, and if the page was previously
347 : * listed as clean. In any case, releases all pages using unpin_user_page(),
348 : * possibly via unpin_user_pages(), for the non-dirty case.
349 : *
350 : * Please see the unpin_user_page() documentation for details.
351 : *
352 : * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is
353 : * required, then the caller should a) verify that this is really correct,
354 : * because _lock() is usually required, and b) hand code it:
355 : * set_page_dirty_lock(), unpin_user_page().
356 : *
357 : */
358 0 : void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages,
359 : bool make_dirty)
360 : {
361 : unsigned long i;
362 : struct folio *folio;
363 : unsigned int nr;
364 :
365 0 : if (!make_dirty) {
366 0 : unpin_user_pages(pages, npages);
367 0 : return;
368 : }
369 :
370 : sanity_check_pinned_pages(pages, npages);
371 0 : for (i = 0; i < npages; i += nr) {
372 0 : folio = gup_folio_next(pages, npages, i, &nr);
373 : /*
374 : * Checking PageDirty at this point may race with
375 : * clear_page_dirty_for_io(), but that's OK. Two key
376 : * cases:
377 : *
378 : * 1) This code sees the page as already dirty, so it
379 : * skips the call to set_page_dirty(). That could happen
380 : * because clear_page_dirty_for_io() called
381 : * page_mkclean(), followed by set_page_dirty().
382 : * However, now the page is going to get written back,
383 : * which meets the original intention of setting it
384 : * dirty, so all is well: clear_page_dirty_for_io() goes
385 : * on to call TestClearPageDirty(), and write the page
386 : * back.
387 : *
388 : * 2) This code sees the page as clean, so it calls
389 : * set_page_dirty(). The page stays dirty, despite being
390 : * written back, so it gets written back again in the
391 : * next writeback cycle. This is harmless.
392 : */
393 0 : if (!folio_test_dirty(folio)) {
394 0 : folio_lock(folio);
395 0 : folio_mark_dirty(folio);
396 0 : folio_unlock(folio);
397 : }
398 0 : gup_put_folio(folio, nr, FOLL_PIN);
399 : }
400 : }
401 : EXPORT_SYMBOL(unpin_user_pages_dirty_lock);
402 :
403 : /**
404 : * unpin_user_page_range_dirty_lock() - release and optionally dirty
405 : * gup-pinned page range
406 : *
407 : * @page: the starting page of a range maybe marked dirty, and definitely released.
408 : * @npages: number of consecutive pages to release.
409 : * @make_dirty: whether to mark the pages dirty
410 : *
411 : * "gup-pinned page range" refers to a range of pages that has had one of the
412 : * pin_user_pages() variants called on that page.
413 : *
414 : * For the page ranges defined by [page .. page+npages], make that range (or
415 : * its head pages, if a compound page) dirty, if @make_dirty is true, and if the
416 : * page range was previously listed as clean.
417 : *
418 : * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is
419 : * required, then the caller should a) verify that this is really correct,
420 : * because _lock() is usually required, and b) hand code it:
421 : * set_page_dirty_lock(), unpin_user_page().
422 : *
423 : */
424 0 : void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages,
425 : bool make_dirty)
426 : {
427 : unsigned long i;
428 : struct folio *folio;
429 : unsigned int nr;
430 :
431 0 : for (i = 0; i < npages; i += nr) {
432 0 : folio = gup_folio_range_next(page, npages, i, &nr);
433 0 : if (make_dirty && !folio_test_dirty(folio)) {
434 0 : folio_lock(folio);
435 0 : folio_mark_dirty(folio);
436 0 : folio_unlock(folio);
437 : }
438 0 : gup_put_folio(folio, nr, FOLL_PIN);
439 : }
440 0 : }
441 : EXPORT_SYMBOL(unpin_user_page_range_dirty_lock);
442 :
443 : static void unpin_user_pages_lockless(struct page **pages, unsigned long npages)
444 : {
445 : unsigned long i;
446 : struct folio *folio;
447 : unsigned int nr;
448 :
449 : /*
450 : * Don't perform any sanity checks because we might have raced with
451 : * fork() and some anonymous pages might now actually be shared --
452 : * which is why we're unpinning after all.
453 : */
454 : for (i = 0; i < npages; i += nr) {
455 : folio = gup_folio_next(pages, npages, i, &nr);
456 : gup_put_folio(folio, nr, FOLL_PIN);
457 : }
458 : }
459 :
460 : /**
461 : * unpin_user_pages() - release an array of gup-pinned pages.
462 : * @pages: array of pages to be marked dirty and released.
463 : * @npages: number of pages in the @pages array.
464 : *
465 : * For each page in the @pages array, release the page using unpin_user_page().
466 : *
467 : * Please see the unpin_user_page() documentation for details.
468 : */
469 0 : void unpin_user_pages(struct page **pages, unsigned long npages)
470 : {
471 : unsigned long i;
472 : struct folio *folio;
473 : unsigned int nr;
474 :
475 : /*
476 : * If this WARN_ON() fires, then the system *might* be leaking pages (by
477 : * leaving them pinned), but probably not. More likely, gup/pup returned
478 : * a hard -ERRNO error to the caller, who erroneously passed it here.
479 : */
480 0 : if (WARN_ON(IS_ERR_VALUE(npages)))
481 0 : return;
482 :
483 : sanity_check_pinned_pages(pages, npages);
484 0 : for (i = 0; i < npages; i += nr) {
485 0 : folio = gup_folio_next(pages, npages, i, &nr);
486 0 : gup_put_folio(folio, nr, FOLL_PIN);
487 : }
488 : }
489 : EXPORT_SYMBOL(unpin_user_pages);
490 :
491 : /*
492 : * Set the MMF_HAS_PINNED if not set yet; after set it'll be there for the mm's
493 : * lifecycle. Avoid setting the bit unless necessary, or it might cause write
494 : * cache bouncing on large SMP machines for concurrent pinned gups.
495 : */
496 0 : static inline void mm_set_has_pinned_flag(unsigned long *mm_flags)
497 : {
498 0 : if (!test_bit(MMF_HAS_PINNED, mm_flags))
499 : set_bit(MMF_HAS_PINNED, mm_flags);
500 0 : }
501 :
502 : #ifdef CONFIG_MMU
503 : static struct page *no_page_table(struct vm_area_struct *vma,
504 : unsigned int flags)
505 : {
506 : /*
507 : * When core dumping an enormous anonymous area that nobody
508 : * has touched so far, we don't want to allocate unnecessary pages or
509 : * page tables. Return error instead of NULL to skip handle_mm_fault,
510 : * then get_dump_page() will return NULL to leave a hole in the dump.
511 : * But we can only make this optimization where a hole would surely
512 : * be zero-filled if handle_mm_fault() actually did handle it.
513 : */
514 0 : if ((flags & FOLL_DUMP) &&
515 0 : (vma_is_anonymous(vma) || !vma->vm_ops->fault))
516 : return ERR_PTR(-EFAULT);
517 : return NULL;
518 : }
519 :
520 0 : static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
521 : pte_t *pte, unsigned int flags)
522 : {
523 0 : if (flags & FOLL_TOUCH) {
524 0 : pte_t orig_entry = ptep_get(pte);
525 0 : pte_t entry = orig_entry;
526 :
527 0 : if (flags & FOLL_WRITE)
528 : entry = pte_mkdirty(entry);
529 0 : entry = pte_mkyoung(entry);
530 :
531 0 : if (!pte_same(orig_entry, entry)) {
532 0 : set_pte_at(vma->vm_mm, address, pte, entry);
533 : update_mmu_cache(vma, address, pte);
534 : }
535 : }
536 :
537 : /* Proper page table entry exists, but no corresponding struct page */
538 0 : return -EEXIST;
539 : }
540 :
541 : /* FOLL_FORCE can write to even unwritable PTEs in COW mappings. */
542 0 : static inline bool can_follow_write_pte(pte_t pte, struct page *page,
543 : struct vm_area_struct *vma,
544 : unsigned int flags)
545 : {
546 : /* If the pte is writable, we can write to the page. */
547 0 : if (pte_write(pte))
548 : return true;
549 :
550 : /* Maybe FOLL_FORCE is set to override it? */
551 0 : if (!(flags & FOLL_FORCE))
552 : return false;
553 :
554 : /* But FOLL_FORCE has no effect on shared mappings */
555 0 : if (vma->vm_flags & (VM_MAYSHARE | VM_SHARED))
556 : return false;
557 :
558 : /* ... or read-only private ones */
559 0 : if (!(vma->vm_flags & VM_MAYWRITE))
560 : return false;
561 :
562 : /* ... or already writable ones that just need to take a write fault */
563 0 : if (vma->vm_flags & VM_WRITE)
564 : return false;
565 :
566 : /*
567 : * See can_change_pte_writable(): we broke COW and could map the page
568 : * writable if we have an exclusive anonymous page ...
569 : */
570 0 : if (!page || !PageAnon(page) || !PageAnonExclusive(page))
571 : return false;
572 :
573 : /* ... and a write-fault isn't required for other reasons. */
574 : if (vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte))
575 : return false;
576 : return !userfaultfd_pte_wp(vma, pte);
577 : }
578 :
579 0 : static struct page *follow_page_pte(struct vm_area_struct *vma,
580 : unsigned long address, pmd_t *pmd, unsigned int flags,
581 : struct dev_pagemap **pgmap)
582 : {
583 0 : struct mm_struct *mm = vma->vm_mm;
584 : struct page *page;
585 : spinlock_t *ptl;
586 : pte_t *ptep, pte;
587 : int ret;
588 :
589 : /* FOLL_GET and FOLL_PIN are mutually exclusive. */
590 0 : if (WARN_ON_ONCE((flags & (FOLL_PIN | FOLL_GET)) ==
591 : (FOLL_PIN | FOLL_GET)))
592 : return ERR_PTR(-EINVAL);
593 :
594 0 : ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
595 0 : if (!ptep)
596 : return no_page_table(vma, flags);
597 0 : pte = ptep_get(ptep);
598 0 : if (!pte_present(pte))
599 : goto no_page;
600 0 : if (pte_protnone(pte) && !gup_can_follow_protnone(flags))
601 : goto no_page;
602 :
603 0 : page = vm_normal_page(vma, address, pte);
604 :
605 : /*
606 : * We only care about anon pages in can_follow_write_pte() and don't
607 : * have to worry about pte_devmap() because they are never anon.
608 : */
609 0 : if ((flags & FOLL_WRITE) &&
610 0 : !can_follow_write_pte(pte, page, vma, flags)) {
611 : page = NULL;
612 : goto out;
613 : }
614 :
615 : if (!page && pte_devmap(pte) && (flags & (FOLL_GET | FOLL_PIN))) {
616 : /*
617 : * Only return device mapping pages in the FOLL_GET or FOLL_PIN
618 : * case since they are only valid while holding the pgmap
619 : * reference.
620 : */
621 : *pgmap = get_dev_pagemap(pte_pfn(pte), *pgmap);
622 : if (*pgmap)
623 : page = pte_page(pte);
624 : else
625 : goto no_page;
626 0 : } else if (unlikely(!page)) {
627 0 : if (flags & FOLL_DUMP) {
628 : /* Avoid special (like zero) pages in core dumps */
629 : page = ERR_PTR(-EFAULT);
630 : goto out;
631 : }
632 :
633 0 : if (is_zero_pfn(pte_pfn(pte))) {
634 0 : page = pte_page(pte);
635 : } else {
636 0 : ret = follow_pfn_pte(vma, address, ptep, flags);
637 0 : page = ERR_PTR(ret);
638 : goto out;
639 : }
640 : }
641 :
642 0 : if (!pte_write(pte) && gup_must_unshare(vma, flags, page)) {
643 : page = ERR_PTR(-EMLINK);
644 : goto out;
645 : }
646 :
647 : VM_BUG_ON_PAGE((flags & FOLL_PIN) && PageAnon(page) &&
648 : !PageAnonExclusive(page), page);
649 :
650 : /* try_grab_page() does nothing unless FOLL_GET or FOLL_PIN is set. */
651 0 : ret = try_grab_page(page, flags);
652 0 : if (unlikely(ret)) {
653 0 : page = ERR_PTR(ret);
654 : goto out;
655 : }
656 :
657 : /*
658 : * We need to make the page accessible if and only if we are going
659 : * to access its content (the FOLL_PIN case). Please see
660 : * Documentation/core-api/pin_user_pages.rst for details.
661 : */
662 : if (flags & FOLL_PIN) {
663 : ret = arch_make_page_accessible(page);
664 : if (ret) {
665 : unpin_user_page(page);
666 : page = ERR_PTR(ret);
667 : goto out;
668 : }
669 : }
670 0 : if (flags & FOLL_TOUCH) {
671 0 : if ((flags & FOLL_WRITE) &&
672 0 : !pte_dirty(pte) && !PageDirty(page))
673 0 : set_page_dirty(page);
674 : /*
675 : * pte_mkyoung() would be more correct here, but atomic care
676 : * is needed to avoid losing the dirty bit: it is easier to use
677 : * mark_page_accessed().
678 : */
679 0 : mark_page_accessed(page);
680 : }
681 : out:
682 0 : pte_unmap_unlock(ptep, ptl);
683 : return page;
684 : no_page:
685 0 : pte_unmap_unlock(ptep, ptl);
686 0 : if (!pte_none(pte))
687 : return NULL;
688 : return no_page_table(vma, flags);
689 : }
690 :
691 0 : static struct page *follow_pmd_mask(struct vm_area_struct *vma,
692 : unsigned long address, pud_t *pudp,
693 : unsigned int flags,
694 : struct follow_page_context *ctx)
695 : {
696 : pmd_t *pmd, pmdval;
697 : spinlock_t *ptl;
698 : struct page *page;
699 0 : struct mm_struct *mm = vma->vm_mm;
700 :
701 0 : pmd = pmd_offset(pudp, address);
702 0 : pmdval = pmdp_get_lockless(pmd);
703 0 : if (pmd_none(pmdval))
704 : return no_page_table(vma, flags);
705 0 : if (!pmd_present(pmdval))
706 : return no_page_table(vma, flags);
707 0 : if (pmd_devmap(pmdval)) {
708 : ptl = pmd_lock(mm, pmd);
709 : page = follow_devmap_pmd(vma, address, pmd, flags, &ctx->pgmap);
710 : spin_unlock(ptl);
711 : if (page)
712 : return page;
713 : }
714 0 : if (likely(!pmd_trans_huge(pmdval)))
715 0 : return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
716 :
717 : if (pmd_protnone(pmdval) && !gup_can_follow_protnone(flags))
718 : return no_page_table(vma, flags);
719 :
720 : ptl = pmd_lock(mm, pmd);
721 : if (unlikely(!pmd_present(*pmd))) {
722 : spin_unlock(ptl);
723 : return no_page_table(vma, flags);
724 : }
725 : if (unlikely(!pmd_trans_huge(*pmd))) {
726 : spin_unlock(ptl);
727 : return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
728 : }
729 : if (flags & FOLL_SPLIT_PMD) {
730 : spin_unlock(ptl);
731 : split_huge_pmd(vma, pmd, address);
732 : /* If pmd was left empty, stuff a page table in there quickly */
733 : return pte_alloc(mm, pmd) ? ERR_PTR(-ENOMEM) :
734 : follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
735 : }
736 : page = follow_trans_huge_pmd(vma, address, pmd, flags);
737 : spin_unlock(ptl);
738 : ctx->page_mask = HPAGE_PMD_NR - 1;
739 : return page;
740 : }
741 :
742 0 : static struct page *follow_pud_mask(struct vm_area_struct *vma,
743 : unsigned long address, p4d_t *p4dp,
744 : unsigned int flags,
745 : struct follow_page_context *ctx)
746 : {
747 : pud_t *pud;
748 : spinlock_t *ptl;
749 : struct page *page;
750 0 : struct mm_struct *mm = vma->vm_mm;
751 :
752 0 : pud = pud_offset(p4dp, address);
753 0 : if (pud_none(*pud))
754 : return no_page_table(vma, flags);
755 : if (pud_devmap(*pud)) {
756 : ptl = pud_lock(mm, pud);
757 : page = follow_devmap_pud(vma, address, pud, flags, &ctx->pgmap);
758 : spin_unlock(ptl);
759 : if (page)
760 : return page;
761 : }
762 0 : if (unlikely(pud_bad(*pud)))
763 : return no_page_table(vma, flags);
764 :
765 0 : return follow_pmd_mask(vma, address, pud, flags, ctx);
766 : }
767 :
768 : static struct page *follow_p4d_mask(struct vm_area_struct *vma,
769 : unsigned long address, pgd_t *pgdp,
770 : unsigned int flags,
771 : struct follow_page_context *ctx)
772 : {
773 : p4d_t *p4d;
774 :
775 0 : p4d = p4d_offset(pgdp, address);
776 : if (p4d_none(*p4d))
777 : return no_page_table(vma, flags);
778 : BUILD_BUG_ON(p4d_huge(*p4d));
779 : if (unlikely(p4d_bad(*p4d)))
780 : return no_page_table(vma, flags);
781 :
782 0 : return follow_pud_mask(vma, address, p4d, flags, ctx);
783 : }
784 :
785 : /**
786 : * follow_page_mask - look up a page descriptor from a user-virtual address
787 : * @vma: vm_area_struct mapping @address
788 : * @address: virtual address to look up
789 : * @flags: flags modifying lookup behaviour
790 : * @ctx: contains dev_pagemap for %ZONE_DEVICE memory pinning and a
791 : * pointer to output page_mask
792 : *
793 : * @flags can have FOLL_ flags set, defined in <linux/mm.h>
794 : *
795 : * When getting pages from ZONE_DEVICE memory, the @ctx->pgmap caches
796 : * the device's dev_pagemap metadata to avoid repeating expensive lookups.
797 : *
798 : * When getting an anonymous page and the caller has to trigger unsharing
799 : * of a shared anonymous page first, -EMLINK is returned. The caller should
800 : * trigger a fault with FAULT_FLAG_UNSHARE set. Note that unsharing is only
801 : * relevant with FOLL_PIN and !FOLL_WRITE.
802 : *
803 : * On output, the @ctx->page_mask is set according to the size of the page.
804 : *
805 : * Return: the mapped (struct page *), %NULL if no mapping exists, or
806 : * an error pointer if there is a mapping to something not represented
807 : * by a page descriptor (see also vm_normal_page()).
808 : */
809 : static struct page *follow_page_mask(struct vm_area_struct *vma,
810 : unsigned long address, unsigned int flags,
811 : struct follow_page_context *ctx)
812 : {
813 : pgd_t *pgd;
814 : struct page *page;
815 0 : struct mm_struct *mm = vma->vm_mm;
816 :
817 0 : ctx->page_mask = 0;
818 :
819 : /*
820 : * Call hugetlb_follow_page_mask for hugetlb vmas as it will use
821 : * special hugetlb page table walking code. This eliminates the
822 : * need to check for hugetlb entries in the general walking code.
823 : *
824 : * hugetlb_follow_page_mask is only for follow_page() handling here.
825 : * Ordinary GUP uses follow_hugetlb_page for hugetlb processing.
826 : */
827 0 : if (is_vm_hugetlb_page(vma)) {
828 : page = hugetlb_follow_page_mask(vma, address, flags);
829 : if (!page)
830 : page = no_page_table(vma, flags);
831 : return page;
832 : }
833 :
834 0 : pgd = pgd_offset(mm, address);
835 :
836 : if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
837 : return no_page_table(vma, flags);
838 :
839 0 : return follow_p4d_mask(vma, address, pgd, flags, ctx);
840 : }
841 :
842 0 : struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
843 : unsigned int foll_flags)
844 : {
845 0 : struct follow_page_context ctx = { NULL };
846 : struct page *page;
847 :
848 0 : if (vma_is_secretmem(vma))
849 : return NULL;
850 :
851 0 : if (WARN_ON_ONCE(foll_flags & FOLL_PIN))
852 : return NULL;
853 :
854 0 : page = follow_page_mask(vma, address, foll_flags, &ctx);
855 : if (ctx.pgmap)
856 : put_dev_pagemap(ctx.pgmap);
857 0 : return page;
858 : }
859 :
860 : static int get_gate_page(struct mm_struct *mm, unsigned long address,
861 : unsigned int gup_flags, struct vm_area_struct **vma,
862 : struct page **page)
863 : {
864 : pgd_t *pgd;
865 : p4d_t *p4d;
866 : pud_t *pud;
867 : pmd_t *pmd;
868 : pte_t *pte;
869 : pte_t entry;
870 : int ret = -EFAULT;
871 :
872 : /* user gate pages are read-only */
873 : if (gup_flags & FOLL_WRITE)
874 : return -EFAULT;
875 : if (address > TASK_SIZE)
876 : pgd = pgd_offset_k(address);
877 : else
878 : pgd = pgd_offset_gate(mm, address);
879 : if (pgd_none(*pgd))
880 : return -EFAULT;
881 : p4d = p4d_offset(pgd, address);
882 : if (p4d_none(*p4d))
883 : return -EFAULT;
884 : pud = pud_offset(p4d, address);
885 : if (pud_none(*pud))
886 : return -EFAULT;
887 : pmd = pmd_offset(pud, address);
888 : if (!pmd_present(*pmd))
889 : return -EFAULT;
890 : pte = pte_offset_map(pmd, address);
891 : if (!pte)
892 : return -EFAULT;
893 : entry = ptep_get(pte);
894 : if (pte_none(entry))
895 : goto unmap;
896 : *vma = get_gate_vma(mm);
897 : if (!page)
898 : goto out;
899 : *page = vm_normal_page(*vma, address, entry);
900 : if (!*page) {
901 : if ((gup_flags & FOLL_DUMP) || !is_zero_pfn(pte_pfn(entry)))
902 : goto unmap;
903 : *page = pte_page(entry);
904 : }
905 : ret = try_grab_page(*page, gup_flags);
906 : if (unlikely(ret))
907 : goto unmap;
908 : out:
909 : ret = 0;
910 : unmap:
911 : pte_unmap(pte);
912 : return ret;
913 : }
914 :
915 : /*
916 : * mmap_lock must be held on entry. If @flags has FOLL_UNLOCKABLE but not
917 : * FOLL_NOWAIT, the mmap_lock may be released. If it is, *@locked will be set
918 : * to 0 and -EBUSY returned.
919 : */
920 0 : static int faultin_page(struct vm_area_struct *vma,
921 : unsigned long address, unsigned int *flags, bool unshare,
922 : int *locked)
923 : {
924 0 : unsigned int fault_flags = 0;
925 : vm_fault_t ret;
926 :
927 0 : if (*flags & FOLL_NOFAULT)
928 : return -EFAULT;
929 0 : if (*flags & FOLL_WRITE)
930 0 : fault_flags |= FAULT_FLAG_WRITE;
931 0 : if (*flags & FOLL_REMOTE)
932 0 : fault_flags |= FAULT_FLAG_REMOTE;
933 0 : if (*flags & FOLL_UNLOCKABLE) {
934 0 : fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
935 : /*
936 : * FAULT_FLAG_INTERRUPTIBLE is opt-in. GUP callers must set
937 : * FOLL_INTERRUPTIBLE to enable FAULT_FLAG_INTERRUPTIBLE.
938 : * That's because some callers may not be prepared to
939 : * handle early exits caused by non-fatal signals.
940 : */
941 0 : if (*flags & FOLL_INTERRUPTIBLE)
942 0 : fault_flags |= FAULT_FLAG_INTERRUPTIBLE;
943 : }
944 0 : if (*flags & FOLL_NOWAIT)
945 0 : fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT;
946 0 : if (*flags & FOLL_TRIED) {
947 : /*
948 : * Note: FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_TRIED
949 : * can co-exist
950 : */
951 0 : fault_flags |= FAULT_FLAG_TRIED;
952 : }
953 0 : if (unshare) {
954 0 : fault_flags |= FAULT_FLAG_UNSHARE;
955 : /* FAULT_FLAG_WRITE and FAULT_FLAG_UNSHARE are incompatible */
956 : VM_BUG_ON(fault_flags & FAULT_FLAG_WRITE);
957 : }
958 :
959 0 : ret = handle_mm_fault(vma, address, fault_flags, NULL);
960 :
961 0 : if (ret & VM_FAULT_COMPLETED) {
962 : /*
963 : * With FAULT_FLAG_RETRY_NOWAIT we'll never release the
964 : * mmap lock in the page fault handler. Sanity check this.
965 : */
966 0 : WARN_ON_ONCE(fault_flags & FAULT_FLAG_RETRY_NOWAIT);
967 0 : *locked = 0;
968 :
969 : /*
970 : * We should do the same as VM_FAULT_RETRY, but let's not
971 : * return -EBUSY since that's not reflecting the reality of
972 : * what has happened - we've just fully completed a page
973 : * fault, with the mmap lock released. Use -EAGAIN to show
974 : * that we want to take the mmap lock _again_.
975 : */
976 0 : return -EAGAIN;
977 : }
978 :
979 0 : if (ret & VM_FAULT_ERROR) {
980 0 : int err = vm_fault_to_errno(ret, *flags);
981 :
982 0 : if (err)
983 : return err;
984 0 : BUG();
985 : }
986 :
987 0 : if (ret & VM_FAULT_RETRY) {
988 0 : if (!(fault_flags & FAULT_FLAG_RETRY_NOWAIT))
989 0 : *locked = 0;
990 : return -EBUSY;
991 : }
992 :
993 : return 0;
994 : }
995 :
996 : /*
997 : * Writing to file-backed mappings which require folio dirty tracking using GUP
998 : * is a fundamentally broken operation, as kernel write access to GUP mappings
999 : * do not adhere to the semantics expected by a file system.
1000 : *
1001 : * Consider the following scenario:-
1002 : *
1003 : * 1. A folio is written to via GUP which write-faults the memory, notifying
1004 : * the file system and dirtying the folio.
1005 : * 2. Later, writeback is triggered, resulting in the folio being cleaned and
1006 : * the PTE being marked read-only.
1007 : * 3. The GUP caller writes to the folio, as it is mapped read/write via the
1008 : * direct mapping.
1009 : * 4. The GUP caller, now done with the page, unpins it and sets it dirty
1010 : * (though it does not have to).
1011 : *
1012 : * This results in both data being written to a folio without writenotify, and
1013 : * the folio being dirtied unexpectedly (if the caller decides to do so).
1014 : */
1015 : static bool writable_file_mapping_allowed(struct vm_area_struct *vma,
1016 : unsigned long gup_flags)
1017 : {
1018 : /*
1019 : * If we aren't pinning then no problematic write can occur. A long term
1020 : * pin is the most egregious case so this is the case we disallow.
1021 : */
1022 0 : if ((gup_flags & (FOLL_PIN | FOLL_LONGTERM)) !=
1023 : (FOLL_PIN | FOLL_LONGTERM))
1024 : return true;
1025 :
1026 : /*
1027 : * If the VMA does not require dirty tracking then no problematic write
1028 : * can occur either.
1029 : */
1030 0 : return !vma_needs_dirty_tracking(vma);
1031 : }
1032 :
1033 0 : static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags)
1034 : {
1035 0 : vm_flags_t vm_flags = vma->vm_flags;
1036 0 : int write = (gup_flags & FOLL_WRITE);
1037 0 : int foreign = (gup_flags & FOLL_REMOTE);
1038 0 : bool vma_anon = vma_is_anonymous(vma);
1039 :
1040 0 : if (vm_flags & (VM_IO | VM_PFNMAP))
1041 : return -EFAULT;
1042 :
1043 0 : if ((gup_flags & FOLL_ANON) && !vma_anon)
1044 : return -EFAULT;
1045 :
1046 : if ((gup_flags & FOLL_LONGTERM) && vma_is_fsdax(vma))
1047 : return -EOPNOTSUPP;
1048 :
1049 0 : if (vma_is_secretmem(vma))
1050 : return -EFAULT;
1051 :
1052 0 : if (write) {
1053 0 : if (!vma_anon &&
1054 0 : !writable_file_mapping_allowed(vma, gup_flags))
1055 : return -EFAULT;
1056 :
1057 0 : if (!(vm_flags & VM_WRITE)) {
1058 0 : if (!(gup_flags & FOLL_FORCE))
1059 : return -EFAULT;
1060 : /* hugetlb does not support FOLL_FORCE|FOLL_WRITE. */
1061 0 : if (is_vm_hugetlb_page(vma))
1062 : return -EFAULT;
1063 : /*
1064 : * We used to let the write,force case do COW in a
1065 : * VM_MAYWRITE VM_SHARED !VM_WRITE vma, so ptrace could
1066 : * set a breakpoint in a read-only mapping of an
1067 : * executable, without corrupting the file (yet only
1068 : * when that file had been opened for writing!).
1069 : * Anon pages in shared mappings are surprising: now
1070 : * just reject it.
1071 : */
1072 0 : if (!is_cow_mapping(vm_flags))
1073 : return -EFAULT;
1074 : }
1075 0 : } else if (!(vm_flags & VM_READ)) {
1076 0 : if (!(gup_flags & FOLL_FORCE))
1077 : return -EFAULT;
1078 : /*
1079 : * Is there actually any vma we can reach here which does not
1080 : * have VM_MAYREAD set?
1081 : */
1082 0 : if (!(vm_flags & VM_MAYREAD))
1083 : return -EFAULT;
1084 : }
1085 : /*
1086 : * gups are always data accesses, not instruction
1087 : * fetches, so execute=false here
1088 : */
1089 0 : if (!arch_vma_access_permitted(vma, write, false, foreign))
1090 : return -EFAULT;
1091 0 : return 0;
1092 : }
1093 :
1094 : /*
1095 : * This is "vma_lookup()", but with a warning if we would have
1096 : * historically expanded the stack in the GUP code.
1097 : */
1098 0 : static struct vm_area_struct *gup_vma_lookup(struct mm_struct *mm,
1099 : unsigned long addr)
1100 : {
1101 : #ifdef CONFIG_STACK_GROWSUP
1102 : return vma_lookup(mm, addr);
1103 : #else
1104 : static volatile unsigned long next_warn;
1105 : struct vm_area_struct *vma;
1106 : unsigned long now, next;
1107 :
1108 0 : vma = find_vma(mm, addr);
1109 0 : if (!vma || (addr >= vma->vm_start))
1110 : return vma;
1111 :
1112 : /* Only warn for half-way relevant accesses */
1113 0 : if (!(vma->vm_flags & VM_GROWSDOWN))
1114 : return NULL;
1115 0 : if (vma->vm_start - addr > 65536)
1116 : return NULL;
1117 :
1118 : /* Let's not warn more than once an hour.. */
1119 0 : now = jiffies; next = next_warn;
1120 0 : if (next && time_before(now, next))
1121 : return NULL;
1122 0 : next_warn = now + 60*60*HZ;
1123 :
1124 : /* Let people know things may have changed. */
1125 0 : pr_warn("GUP no longer grows the stack in %s (%d): %lx-%lx (%lx)\n",
1126 : current->comm, task_pid_nr(current),
1127 : vma->vm_start, vma->vm_end, addr);
1128 0 : dump_stack();
1129 0 : return NULL;
1130 : #endif
1131 : }
1132 :
1133 : /**
1134 : * __get_user_pages() - pin user pages in memory
1135 : * @mm: mm_struct of target mm
1136 : * @start: starting user address
1137 : * @nr_pages: number of pages from start to pin
1138 : * @gup_flags: flags modifying pin behaviour
1139 : * @pages: array that receives pointers to the pages pinned.
1140 : * Should be at least nr_pages long. Or NULL, if caller
1141 : * only intends to ensure the pages are faulted in.
1142 : * @locked: whether we're still with the mmap_lock held
1143 : *
1144 : * Returns either number of pages pinned (which may be less than the
1145 : * number requested), or an error. Details about the return value:
1146 : *
1147 : * -- If nr_pages is 0, returns 0.
1148 : * -- If nr_pages is >0, but no pages were pinned, returns -errno.
1149 : * -- If nr_pages is >0, and some pages were pinned, returns the number of
1150 : * pages pinned. Again, this may be less than nr_pages.
1151 : * -- 0 return value is possible when the fault would need to be retried.
1152 : *
1153 : * The caller is responsible for releasing returned @pages, via put_page().
1154 : *
1155 : * Must be called with mmap_lock held. It may be released. See below.
1156 : *
1157 : * __get_user_pages walks a process's page tables and takes a reference to
1158 : * each struct page that each user address corresponds to at a given
1159 : * instant. That is, it takes the page that would be accessed if a user
1160 : * thread accesses the given user virtual address at that instant.
1161 : *
1162 : * This does not guarantee that the page exists in the user mappings when
1163 : * __get_user_pages returns, and there may even be a completely different
1164 : * page there in some cases (eg. if mmapped pagecache has been invalidated
1165 : * and subsequently re-faulted). However it does guarantee that the page
1166 : * won't be freed completely. And mostly callers simply care that the page
1167 : * contains data that was valid *at some point in time*. Typically, an IO
1168 : * or similar operation cannot guarantee anything stronger anyway because
1169 : * locks can't be held over the syscall boundary.
1170 : *
1171 : * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If
1172 : * the page is written to, set_page_dirty (or set_page_dirty_lock, as
1173 : * appropriate) must be called after the page is finished with, and
1174 : * before put_page is called.
1175 : *
1176 : * If FOLL_UNLOCKABLE is set without FOLL_NOWAIT then the mmap_lock may
1177 : * be released. If this happens *@locked will be set to 0 on return.
1178 : *
1179 : * A caller using such a combination of @gup_flags must therefore hold the
1180 : * mmap_lock for reading only, and recognize when it's been released. Otherwise,
1181 : * it must be held for either reading or writing and will not be released.
1182 : *
1183 : * In most cases, get_user_pages or get_user_pages_fast should be used
1184 : * instead of __get_user_pages. __get_user_pages should be used only if
1185 : * you need some special @gup_flags.
1186 : */
1187 0 : static long __get_user_pages(struct mm_struct *mm,
1188 : unsigned long start, unsigned long nr_pages,
1189 : unsigned int gup_flags, struct page **pages,
1190 : int *locked)
1191 : {
1192 0 : long ret = 0, i = 0;
1193 0 : struct vm_area_struct *vma = NULL;
1194 0 : struct follow_page_context ctx = { NULL };
1195 :
1196 0 : if (!nr_pages)
1197 : return 0;
1198 :
1199 : start = untagged_addr_remote(mm, start);
1200 :
1201 : VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN)));
1202 :
1203 : do {
1204 : struct page *page;
1205 0 : unsigned int foll_flags = gup_flags;
1206 : unsigned int page_increm;
1207 :
1208 : /* first iteration or cross vma bound */
1209 0 : if (!vma || start >= vma->vm_end) {
1210 0 : vma = gup_vma_lookup(mm, start);
1211 : if (!vma && in_gate_area(mm, start)) {
1212 : ret = get_gate_page(mm, start & PAGE_MASK,
1213 : gup_flags, &vma,
1214 : pages ? &pages[i] : NULL);
1215 : if (ret)
1216 : goto out;
1217 : ctx.page_mask = 0;
1218 : goto next_page;
1219 : }
1220 :
1221 0 : if (!vma) {
1222 : ret = -EFAULT;
1223 : goto out;
1224 : }
1225 0 : ret = check_vma_flags(vma, gup_flags);
1226 0 : if (ret)
1227 : goto out;
1228 :
1229 : if (is_vm_hugetlb_page(vma)) {
1230 : i = follow_hugetlb_page(mm, vma, pages,
1231 : &start, &nr_pages, i,
1232 : gup_flags, locked);
1233 : if (!*locked) {
1234 : /*
1235 : * We've got a VM_FAULT_RETRY
1236 : * and we've lost mmap_lock.
1237 : * We must stop here.
1238 : */
1239 : BUG_ON(gup_flags & FOLL_NOWAIT);
1240 : goto out;
1241 : }
1242 : continue;
1243 : }
1244 : }
1245 : retry:
1246 : /*
1247 : * If we have a pending SIGKILL, don't keep faulting pages and
1248 : * potentially allocating memory.
1249 : */
1250 0 : if (fatal_signal_pending(current)) {
1251 : ret = -EINTR;
1252 : goto out;
1253 : }
1254 0 : cond_resched();
1255 :
1256 0 : page = follow_page_mask(vma, start, foll_flags, &ctx);
1257 0 : if (!page || PTR_ERR(page) == -EMLINK) {
1258 0 : ret = faultin_page(vma, start, &foll_flags,
1259 0 : PTR_ERR(page) == -EMLINK, locked);
1260 0 : switch (ret) {
1261 : case 0:
1262 : goto retry;
1263 : case -EBUSY:
1264 : case -EAGAIN:
1265 0 : ret = 0;
1266 : fallthrough;
1267 : case -EFAULT:
1268 : case -ENOMEM:
1269 : case -EHWPOISON:
1270 : goto out;
1271 : }
1272 0 : BUG();
1273 0 : } else if (PTR_ERR(page) == -EEXIST) {
1274 : /*
1275 : * Proper page table entry exists, but no corresponding
1276 : * struct page. If the caller expects **pages to be
1277 : * filled in, bail out now, because that can't be done
1278 : * for this page.
1279 : */
1280 0 : if (pages) {
1281 : ret = PTR_ERR(page);
1282 : goto out;
1283 : }
1284 :
1285 : goto next_page;
1286 0 : } else if (IS_ERR(page)) {
1287 : ret = PTR_ERR(page);
1288 : goto out;
1289 : }
1290 0 : if (pages) {
1291 0 : pages[i] = page;
1292 0 : flush_anon_page(vma, page, start);
1293 : flush_dcache_page(page);
1294 0 : ctx.page_mask = 0;
1295 : }
1296 : next_page:
1297 0 : page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask);
1298 0 : if (page_increm > nr_pages)
1299 0 : page_increm = nr_pages;
1300 0 : i += page_increm;
1301 0 : start += page_increm * PAGE_SIZE;
1302 0 : nr_pages -= page_increm;
1303 0 : } while (nr_pages);
1304 : out:
1305 : if (ctx.pgmap)
1306 : put_dev_pagemap(ctx.pgmap);
1307 0 : return i ? i : ret;
1308 : }
1309 :
1310 : static bool vma_permits_fault(struct vm_area_struct *vma,
1311 : unsigned int fault_flags)
1312 : {
1313 0 : bool write = !!(fault_flags & FAULT_FLAG_WRITE);
1314 0 : bool foreign = !!(fault_flags & FAULT_FLAG_REMOTE);
1315 0 : vm_flags_t vm_flags = write ? VM_WRITE : VM_READ;
1316 :
1317 0 : if (!(vm_flags & vma->vm_flags))
1318 : return false;
1319 :
1320 : /*
1321 : * The architecture might have a hardware protection
1322 : * mechanism other than read/write that can deny access.
1323 : *
1324 : * gup always represents data access, not instruction
1325 : * fetches, so execute=false here:
1326 : */
1327 0 : if (!arch_vma_access_permitted(vma, write, false, foreign))
1328 : return false;
1329 :
1330 : return true;
1331 : }
1332 :
1333 : /**
1334 : * fixup_user_fault() - manually resolve a user page fault
1335 : * @mm: mm_struct of target mm
1336 : * @address: user address
1337 : * @fault_flags:flags to pass down to handle_mm_fault()
1338 : * @unlocked: did we unlock the mmap_lock while retrying, maybe NULL if caller
1339 : * does not allow retry. If NULL, the caller must guarantee
1340 : * that fault_flags does not contain FAULT_FLAG_ALLOW_RETRY.
1341 : *
1342 : * This is meant to be called in the specific scenario where for locking reasons
1343 : * we try to access user memory in atomic context (within a pagefault_disable()
1344 : * section), this returns -EFAULT, and we want to resolve the user fault before
1345 : * trying again.
1346 : *
1347 : * Typically this is meant to be used by the futex code.
1348 : *
1349 : * The main difference with get_user_pages() is that this function will
1350 : * unconditionally call handle_mm_fault() which will in turn perform all the
1351 : * necessary SW fixup of the dirty and young bits in the PTE, while
1352 : * get_user_pages() only guarantees to update these in the struct page.
1353 : *
1354 : * This is important for some architectures where those bits also gate the
1355 : * access permission to the page because they are maintained in software. On
1356 : * such architectures, gup() will not be enough to make a subsequent access
1357 : * succeed.
1358 : *
1359 : * This function will not return with an unlocked mmap_lock. So it has not the
1360 : * same semantics wrt the @mm->mmap_lock as does filemap_fault().
1361 : */
1362 0 : int fixup_user_fault(struct mm_struct *mm,
1363 : unsigned long address, unsigned int fault_flags,
1364 : bool *unlocked)
1365 : {
1366 : struct vm_area_struct *vma;
1367 : vm_fault_t ret;
1368 :
1369 0 : address = untagged_addr_remote(mm, address);
1370 :
1371 0 : if (unlocked)
1372 0 : fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
1373 :
1374 : retry:
1375 0 : vma = gup_vma_lookup(mm, address);
1376 0 : if (!vma)
1377 : return -EFAULT;
1378 :
1379 0 : if (!vma_permits_fault(vma, fault_flags))
1380 : return -EFAULT;
1381 :
1382 0 : if ((fault_flags & FAULT_FLAG_KILLABLE) &&
1383 0 : fatal_signal_pending(current))
1384 : return -EINTR;
1385 :
1386 0 : ret = handle_mm_fault(vma, address, fault_flags, NULL);
1387 :
1388 0 : if (ret & VM_FAULT_COMPLETED) {
1389 : /*
1390 : * NOTE: it's a pity that we need to retake the lock here
1391 : * to pair with the unlock() in the callers. Ideally we
1392 : * could tell the callers so they do not need to unlock.
1393 : */
1394 0 : mmap_read_lock(mm);
1395 0 : *unlocked = true;
1396 0 : return 0;
1397 : }
1398 :
1399 0 : if (ret & VM_FAULT_ERROR) {
1400 0 : int err = vm_fault_to_errno(ret, 0);
1401 :
1402 0 : if (err)
1403 : return err;
1404 0 : BUG();
1405 : }
1406 :
1407 0 : if (ret & VM_FAULT_RETRY) {
1408 0 : mmap_read_lock(mm);
1409 0 : *unlocked = true;
1410 0 : fault_flags |= FAULT_FLAG_TRIED;
1411 0 : goto retry;
1412 : }
1413 :
1414 : return 0;
1415 : }
1416 : EXPORT_SYMBOL_GPL(fixup_user_fault);
1417 :
1418 : /*
1419 : * GUP always responds to fatal signals. When FOLL_INTERRUPTIBLE is
1420 : * specified, it'll also respond to generic signals. The caller of GUP
1421 : * that has FOLL_INTERRUPTIBLE should take care of the GUP interruption.
1422 : */
1423 0 : static bool gup_signal_pending(unsigned int flags)
1424 : {
1425 0 : if (fatal_signal_pending(current))
1426 : return true;
1427 :
1428 0 : if (!(flags & FOLL_INTERRUPTIBLE))
1429 : return false;
1430 :
1431 0 : return signal_pending(current);
1432 : }
1433 :
1434 : /*
1435 : * Locking: (*locked == 1) means that the mmap_lock has already been acquired by
1436 : * the caller. This function may drop the mmap_lock. If it does so, then it will
1437 : * set (*locked = 0).
1438 : *
1439 : * (*locked == 0) means that the caller expects this function to acquire and
1440 : * drop the mmap_lock. Therefore, the value of *locked will still be zero when
1441 : * the function returns, even though it may have changed temporarily during
1442 : * function execution.
1443 : *
1444 : * Please note that this function, unlike __get_user_pages(), will not return 0
1445 : * for nr_pages > 0, unless FOLL_NOWAIT is used.
1446 : */
1447 : static __always_inline long __get_user_pages_locked(struct mm_struct *mm,
1448 : unsigned long start,
1449 : unsigned long nr_pages,
1450 : struct page **pages,
1451 : int *locked,
1452 : unsigned int flags)
1453 : {
1454 : long ret, pages_done;
1455 0 : bool must_unlock = false;
1456 :
1457 : /*
1458 : * The internal caller expects GUP to manage the lock internally and the
1459 : * lock must be released when this returns.
1460 : */
1461 0 : if (!*locked) {
1462 0 : if (mmap_read_lock_killable(mm))
1463 : return -EAGAIN;
1464 0 : must_unlock = true;
1465 0 : *locked = 1;
1466 : }
1467 : else
1468 : mmap_assert_locked(mm);
1469 :
1470 0 : if (flags & FOLL_PIN)
1471 0 : mm_set_has_pinned_flag(&mm->flags);
1472 :
1473 : /*
1474 : * FOLL_PIN and FOLL_GET are mutually exclusive. Traditional behavior
1475 : * is to set FOLL_GET if the caller wants pages[] filled in (but has
1476 : * carelessly failed to specify FOLL_GET), so keep doing that, but only
1477 : * for FOLL_GET, not for the newer FOLL_PIN.
1478 : *
1479 : * FOLL_PIN always expects pages to be non-null, but no need to assert
1480 : * that here, as any failures will be obvious enough.
1481 : */
1482 0 : if (pages && !(flags & FOLL_PIN))
1483 0 : flags |= FOLL_GET;
1484 :
1485 0 : pages_done = 0;
1486 : for (;;) {
1487 0 : ret = __get_user_pages(mm, start, nr_pages, flags, pages,
1488 : locked);
1489 0 : if (!(flags & FOLL_UNLOCKABLE)) {
1490 : /* VM_FAULT_RETRY couldn't trigger, bypass */
1491 : pages_done = ret;
1492 : break;
1493 : }
1494 :
1495 : /* VM_FAULT_RETRY or VM_FAULT_COMPLETED cannot return errors */
1496 0 : if (!*locked) {
1497 0 : BUG_ON(ret < 0);
1498 0 : BUG_ON(ret >= nr_pages);
1499 : }
1500 :
1501 0 : if (ret > 0) {
1502 0 : nr_pages -= ret;
1503 0 : pages_done += ret;
1504 0 : if (!nr_pages)
1505 : break;
1506 : }
1507 0 : if (*locked) {
1508 : /*
1509 : * VM_FAULT_RETRY didn't trigger or it was a
1510 : * FOLL_NOWAIT.
1511 : */
1512 0 : if (!pages_done)
1513 0 : pages_done = ret;
1514 : break;
1515 : }
1516 : /*
1517 : * VM_FAULT_RETRY triggered, so seek to the faulting offset.
1518 : * For the prefault case (!pages) we only update counts.
1519 : */
1520 0 : if (likely(pages))
1521 0 : pages += ret;
1522 0 : start += ret << PAGE_SHIFT;
1523 :
1524 : /* The lock was temporarily dropped, so we must unlock later */
1525 0 : must_unlock = true;
1526 :
1527 : retry:
1528 : /*
1529 : * Repeat on the address that fired VM_FAULT_RETRY
1530 : * with both FAULT_FLAG_ALLOW_RETRY and
1531 : * FAULT_FLAG_TRIED. Note that GUP can be interrupted
1532 : * by fatal signals of even common signals, depending on
1533 : * the caller's request. So we need to check it before we
1534 : * start trying again otherwise it can loop forever.
1535 : */
1536 0 : if (gup_signal_pending(flags)) {
1537 0 : if (!pages_done)
1538 0 : pages_done = -EINTR;
1539 : break;
1540 : }
1541 :
1542 0 : ret = mmap_read_lock_killable(mm);
1543 0 : if (ret) {
1544 0 : BUG_ON(ret > 0);
1545 0 : if (!pages_done)
1546 0 : pages_done = ret;
1547 : break;
1548 : }
1549 :
1550 0 : *locked = 1;
1551 0 : ret = __get_user_pages(mm, start, 1, flags | FOLL_TRIED,
1552 : pages, locked);
1553 0 : if (!*locked) {
1554 : /* Continue to retry until we succeeded */
1555 0 : BUG_ON(ret != 0);
1556 : goto retry;
1557 : }
1558 0 : if (ret != 1) {
1559 0 : BUG_ON(ret > 1);
1560 0 : if (!pages_done)
1561 0 : pages_done = ret;
1562 : break;
1563 : }
1564 0 : nr_pages--;
1565 0 : pages_done++;
1566 0 : if (!nr_pages)
1567 : break;
1568 0 : if (likely(pages))
1569 0 : pages++;
1570 0 : start += PAGE_SIZE;
1571 : }
1572 0 : if (must_unlock && *locked) {
1573 : /*
1574 : * We either temporarily dropped the lock, or the caller
1575 : * requested that we both acquire and drop the lock. Either way,
1576 : * we must now unlock, and notify the caller of that state.
1577 : */
1578 0 : mmap_read_unlock(mm);
1579 0 : *locked = 0;
1580 : }
1581 : return pages_done;
1582 : }
1583 :
1584 : /**
1585 : * populate_vma_page_range() - populate a range of pages in the vma.
1586 : * @vma: target vma
1587 : * @start: start address
1588 : * @end: end address
1589 : * @locked: whether the mmap_lock is still held
1590 : *
1591 : * This takes care of mlocking the pages too if VM_LOCKED is set.
1592 : *
1593 : * Return either number of pages pinned in the vma, or a negative error
1594 : * code on error.
1595 : *
1596 : * vma->vm_mm->mmap_lock must be held.
1597 : *
1598 : * If @locked is NULL, it may be held for read or write and will
1599 : * be unperturbed.
1600 : *
1601 : * If @locked is non-NULL, it must held for read only and may be
1602 : * released. If it's released, *@locked will be set to 0.
1603 : */
1604 0 : long populate_vma_page_range(struct vm_area_struct *vma,
1605 : unsigned long start, unsigned long end, int *locked)
1606 : {
1607 0 : struct mm_struct *mm = vma->vm_mm;
1608 0 : unsigned long nr_pages = (end - start) / PAGE_SIZE;
1609 0 : int local_locked = 1;
1610 : int gup_flags;
1611 : long ret;
1612 :
1613 : VM_BUG_ON(!PAGE_ALIGNED(start));
1614 : VM_BUG_ON(!PAGE_ALIGNED(end));
1615 : VM_BUG_ON_VMA(start < vma->vm_start, vma);
1616 : VM_BUG_ON_VMA(end > vma->vm_end, vma);
1617 0 : mmap_assert_locked(mm);
1618 :
1619 : /*
1620 : * Rightly or wrongly, the VM_LOCKONFAULT case has never used
1621 : * faultin_page() to break COW, so it has no work to do here.
1622 : */
1623 0 : if (vma->vm_flags & VM_LOCKONFAULT)
1624 0 : return nr_pages;
1625 :
1626 0 : gup_flags = FOLL_TOUCH;
1627 : /*
1628 : * We want to touch writable mappings with a write fault in order
1629 : * to break COW, except for shared mappings because these don't COW
1630 : * and we would not want to dirty them for nothing.
1631 : */
1632 0 : if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
1633 0 : gup_flags |= FOLL_WRITE;
1634 :
1635 : /*
1636 : * We want mlock to succeed for regions that have any permissions
1637 : * other than PROT_NONE.
1638 : */
1639 0 : if (vma_is_accessible(vma))
1640 0 : gup_flags |= FOLL_FORCE;
1641 :
1642 0 : if (locked)
1643 0 : gup_flags |= FOLL_UNLOCKABLE;
1644 :
1645 : /*
1646 : * We made sure addr is within a VMA, so the following will
1647 : * not result in a stack expansion that recurses back here.
1648 : */
1649 0 : ret = __get_user_pages(mm, start, nr_pages, gup_flags,
1650 : NULL, locked ? locked : &local_locked);
1651 0 : lru_add_drain();
1652 0 : return ret;
1653 : }
1654 :
1655 : /*
1656 : * faultin_vma_page_range() - populate (prefault) page tables inside the
1657 : * given VMA range readable/writable
1658 : *
1659 : * This takes care of mlocking the pages, too, if VM_LOCKED is set.
1660 : *
1661 : * @vma: target vma
1662 : * @start: start address
1663 : * @end: end address
1664 : * @write: whether to prefault readable or writable
1665 : * @locked: whether the mmap_lock is still held
1666 : *
1667 : * Returns either number of processed pages in the vma, or a negative error
1668 : * code on error (see __get_user_pages()).
1669 : *
1670 : * vma->vm_mm->mmap_lock must be held. The range must be page-aligned and
1671 : * covered by the VMA. If it's released, *@locked will be set to 0.
1672 : */
1673 0 : long faultin_vma_page_range(struct vm_area_struct *vma, unsigned long start,
1674 : unsigned long end, bool write, int *locked)
1675 : {
1676 0 : struct mm_struct *mm = vma->vm_mm;
1677 0 : unsigned long nr_pages = (end - start) / PAGE_SIZE;
1678 : int gup_flags;
1679 : long ret;
1680 :
1681 : VM_BUG_ON(!PAGE_ALIGNED(start));
1682 : VM_BUG_ON(!PAGE_ALIGNED(end));
1683 : VM_BUG_ON_VMA(start < vma->vm_start, vma);
1684 : VM_BUG_ON_VMA(end > vma->vm_end, vma);
1685 0 : mmap_assert_locked(mm);
1686 :
1687 : /*
1688 : * FOLL_TOUCH: Mark page accessed and thereby young; will also mark
1689 : * the page dirty with FOLL_WRITE -- which doesn't make a
1690 : * difference with !FOLL_FORCE, because the page is writable
1691 : * in the page table.
1692 : * FOLL_HWPOISON: Return -EHWPOISON instead of -EFAULT when we hit
1693 : * a poisoned page.
1694 : * !FOLL_FORCE: Require proper access permissions.
1695 : */
1696 0 : gup_flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_UNLOCKABLE;
1697 0 : if (write)
1698 0 : gup_flags |= FOLL_WRITE;
1699 :
1700 : /*
1701 : * We want to report -EINVAL instead of -EFAULT for any permission
1702 : * problems or incompatible mappings.
1703 : */
1704 0 : if (check_vma_flags(vma, gup_flags))
1705 : return -EINVAL;
1706 :
1707 0 : ret = __get_user_pages(mm, start, nr_pages, gup_flags,
1708 : NULL, locked);
1709 0 : lru_add_drain();
1710 0 : return ret;
1711 : }
1712 :
1713 : /*
1714 : * __mm_populate - populate and/or mlock pages within a range of address space.
1715 : *
1716 : * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
1717 : * flags. VMAs must be already marked with the desired vm_flags, and
1718 : * mmap_lock must not be held.
1719 : */
1720 0 : int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
1721 : {
1722 0 : struct mm_struct *mm = current->mm;
1723 : unsigned long end, nstart, nend;
1724 0 : struct vm_area_struct *vma = NULL;
1725 0 : int locked = 0;
1726 0 : long ret = 0;
1727 :
1728 0 : end = start + len;
1729 :
1730 0 : for (nstart = start; nstart < end; nstart = nend) {
1731 : /*
1732 : * We want to fault in pages for [nstart; end) address range.
1733 : * Find first corresponding VMA.
1734 : */
1735 0 : if (!locked) {
1736 0 : locked = 1;
1737 0 : mmap_read_lock(mm);
1738 0 : vma = find_vma_intersection(mm, nstart, end);
1739 0 : } else if (nstart >= vma->vm_end)
1740 0 : vma = find_vma_intersection(mm, vma->vm_end, end);
1741 :
1742 0 : if (!vma)
1743 : break;
1744 : /*
1745 : * Set [nstart; nend) to intersection of desired address
1746 : * range with the first VMA. Also, skip undesirable VMA types.
1747 : */
1748 0 : nend = min(end, vma->vm_end);
1749 0 : if (vma->vm_flags & (VM_IO | VM_PFNMAP))
1750 0 : continue;
1751 0 : if (nstart < vma->vm_start)
1752 0 : nstart = vma->vm_start;
1753 : /*
1754 : * Now fault in a range of pages. populate_vma_page_range()
1755 : * double checks the vma flags, so that it won't mlock pages
1756 : * if the vma was already munlocked.
1757 : */
1758 0 : ret = populate_vma_page_range(vma, nstart, nend, &locked);
1759 0 : if (ret < 0) {
1760 0 : if (ignore_errors) {
1761 0 : ret = 0;
1762 0 : continue; /* continue at next VMA */
1763 : }
1764 : break;
1765 : }
1766 0 : nend = nstart + ret * PAGE_SIZE;
1767 0 : ret = 0;
1768 : }
1769 0 : if (locked)
1770 : mmap_read_unlock(mm);
1771 0 : return ret; /* 0 or negative error code */
1772 : }
1773 : #else /* CONFIG_MMU */
1774 : static long __get_user_pages_locked(struct mm_struct *mm, unsigned long start,
1775 : unsigned long nr_pages, struct page **pages,
1776 : int *locked, unsigned int foll_flags)
1777 : {
1778 : struct vm_area_struct *vma;
1779 : bool must_unlock = false;
1780 : unsigned long vm_flags;
1781 : long i;
1782 :
1783 : if (!nr_pages)
1784 : return 0;
1785 :
1786 : /*
1787 : * The internal caller expects GUP to manage the lock internally and the
1788 : * lock must be released when this returns.
1789 : */
1790 : if (!*locked) {
1791 : if (mmap_read_lock_killable(mm))
1792 : return -EAGAIN;
1793 : must_unlock = true;
1794 : *locked = 1;
1795 : }
1796 :
1797 : /* calculate required read or write permissions.
1798 : * If FOLL_FORCE is set, we only require the "MAY" flags.
1799 : */
1800 : vm_flags = (foll_flags & FOLL_WRITE) ?
1801 : (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1802 : vm_flags &= (foll_flags & FOLL_FORCE) ?
1803 : (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1804 :
1805 : for (i = 0; i < nr_pages; i++) {
1806 : vma = find_vma(mm, start);
1807 : if (!vma)
1808 : break;
1809 :
1810 : /* protect what we can, including chardevs */
1811 : if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1812 : !(vm_flags & vma->vm_flags))
1813 : break;
1814 :
1815 : if (pages) {
1816 : pages[i] = virt_to_page((void *)start);
1817 : if (pages[i])
1818 : get_page(pages[i]);
1819 : }
1820 :
1821 : start = (start + PAGE_SIZE) & PAGE_MASK;
1822 : }
1823 :
1824 : if (must_unlock && *locked) {
1825 : mmap_read_unlock(mm);
1826 : *locked = 0;
1827 : }
1828 :
1829 : return i ? : -EFAULT;
1830 : }
1831 : #endif /* !CONFIG_MMU */
1832 :
1833 : /**
1834 : * fault_in_writeable - fault in userspace address range for writing
1835 : * @uaddr: start of address range
1836 : * @size: size of address range
1837 : *
1838 : * Returns the number of bytes not faulted in (like copy_to_user() and
1839 : * copy_from_user()).
1840 : */
1841 0 : size_t fault_in_writeable(char __user *uaddr, size_t size)
1842 : {
1843 0 : char __user *start = uaddr, *end;
1844 :
1845 0 : if (unlikely(size == 0))
1846 : return 0;
1847 0 : if (!user_write_access_begin(uaddr, size))
1848 : return size;
1849 0 : if (!PAGE_ALIGNED(uaddr)) {
1850 0 : unsafe_put_user(0, uaddr, out);
1851 0 : uaddr = (char __user *)PAGE_ALIGN((unsigned long)uaddr);
1852 : }
1853 0 : end = (char __user *)PAGE_ALIGN((unsigned long)start + size);
1854 0 : if (unlikely(end < start))
1855 0 : end = NULL;
1856 0 : while (uaddr != end) {
1857 0 : unsafe_put_user(0, uaddr, out);
1858 0 : uaddr += PAGE_SIZE;
1859 : }
1860 :
1861 : out:
1862 : user_write_access_end();
1863 0 : if (size > uaddr - start)
1864 0 : return size - (uaddr - start);
1865 : return 0;
1866 : }
1867 : EXPORT_SYMBOL(fault_in_writeable);
1868 :
1869 : /**
1870 : * fault_in_subpage_writeable - fault in an address range for writing
1871 : * @uaddr: start of address range
1872 : * @size: size of address range
1873 : *
1874 : * Fault in a user address range for writing while checking for permissions at
1875 : * sub-page granularity (e.g. arm64 MTE). This function should be used when
1876 : * the caller cannot guarantee forward progress of a copy_to_user() loop.
1877 : *
1878 : * Returns the number of bytes not faulted in (like copy_to_user() and
1879 : * copy_from_user()).
1880 : */
1881 0 : size_t fault_in_subpage_writeable(char __user *uaddr, size_t size)
1882 : {
1883 : size_t faulted_in;
1884 :
1885 : /*
1886 : * Attempt faulting in at page granularity first for page table
1887 : * permission checking. The arch-specific probe_subpage_writeable()
1888 : * functions may not check for this.
1889 : */
1890 0 : faulted_in = size - fault_in_writeable(uaddr, size);
1891 : if (faulted_in)
1892 : faulted_in -= probe_subpage_writeable(uaddr, faulted_in);
1893 :
1894 0 : return size - faulted_in;
1895 : }
1896 : EXPORT_SYMBOL(fault_in_subpage_writeable);
1897 :
1898 : /*
1899 : * fault_in_safe_writeable - fault in an address range for writing
1900 : * @uaddr: start of address range
1901 : * @size: length of address range
1902 : *
1903 : * Faults in an address range for writing. This is primarily useful when we
1904 : * already know that some or all of the pages in the address range aren't in
1905 : * memory.
1906 : *
1907 : * Unlike fault_in_writeable(), this function is non-destructive.
1908 : *
1909 : * Note that we don't pin or otherwise hold the pages referenced that we fault
1910 : * in. There's no guarantee that they'll stay in memory for any duration of
1911 : * time.
1912 : *
1913 : * Returns the number of bytes not faulted in, like copy_to_user() and
1914 : * copy_from_user().
1915 : */
1916 0 : size_t fault_in_safe_writeable(const char __user *uaddr, size_t size)
1917 : {
1918 0 : unsigned long start = (unsigned long)uaddr, end;
1919 0 : struct mm_struct *mm = current->mm;
1920 0 : bool unlocked = false;
1921 :
1922 0 : if (unlikely(size == 0))
1923 : return 0;
1924 0 : end = PAGE_ALIGN(start + size);
1925 0 : if (end < start)
1926 0 : end = 0;
1927 :
1928 : mmap_read_lock(mm);
1929 : do {
1930 0 : if (fixup_user_fault(mm, start, FAULT_FLAG_WRITE, &unlocked))
1931 : break;
1932 0 : start = (start + PAGE_SIZE) & PAGE_MASK;
1933 0 : } while (start != end);
1934 0 : mmap_read_unlock(mm);
1935 :
1936 0 : if (size > (unsigned long)uaddr - start)
1937 0 : return size - ((unsigned long)uaddr - start);
1938 : return 0;
1939 : }
1940 : EXPORT_SYMBOL(fault_in_safe_writeable);
1941 :
1942 : /**
1943 : * fault_in_readable - fault in userspace address range for reading
1944 : * @uaddr: start of user address range
1945 : * @size: size of user address range
1946 : *
1947 : * Returns the number of bytes not faulted in (like copy_to_user() and
1948 : * copy_from_user()).
1949 : */
1950 0 : size_t fault_in_readable(const char __user *uaddr, size_t size)
1951 : {
1952 0 : const char __user *start = uaddr, *end;
1953 : volatile char c;
1954 :
1955 0 : if (unlikely(size == 0))
1956 : return 0;
1957 0 : if (!user_read_access_begin(uaddr, size))
1958 : return size;
1959 0 : if (!PAGE_ALIGNED(uaddr)) {
1960 0 : unsafe_get_user(c, uaddr, out);
1961 0 : uaddr = (const char __user *)PAGE_ALIGN((unsigned long)uaddr);
1962 : }
1963 0 : end = (const char __user *)PAGE_ALIGN((unsigned long)start + size);
1964 0 : if (unlikely(end < start))
1965 0 : end = NULL;
1966 0 : while (uaddr != end) {
1967 0 : unsafe_get_user(c, uaddr, out);
1968 0 : uaddr += PAGE_SIZE;
1969 : }
1970 :
1971 : out:
1972 : user_read_access_end();
1973 0 : (void)c;
1974 0 : if (size > uaddr - start)
1975 0 : return size - (uaddr - start);
1976 : return 0;
1977 : }
1978 : EXPORT_SYMBOL(fault_in_readable);
1979 :
1980 : /**
1981 : * get_dump_page() - pin user page in memory while writing it to core dump
1982 : * @addr: user address
1983 : *
1984 : * Returns struct page pointer of user page pinned for dump,
1985 : * to be freed afterwards by put_page().
1986 : *
1987 : * Returns NULL on any kind of failure - a hole must then be inserted into
1988 : * the corefile, to preserve alignment with its headers; and also returns
1989 : * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1990 : * allowing a hole to be left in the corefile to save disk space.
1991 : *
1992 : * Called without mmap_lock (takes and releases the mmap_lock by itself).
1993 : */
1994 : #ifdef CONFIG_ELF_CORE
1995 0 : struct page *get_dump_page(unsigned long addr)
1996 : {
1997 : struct page *page;
1998 0 : int locked = 0;
1999 : int ret;
2000 :
2001 0 : ret = __get_user_pages_locked(current->mm, addr, 1, &page, &locked,
2002 : FOLL_FORCE | FOLL_DUMP | FOLL_GET);
2003 0 : return (ret == 1) ? page : NULL;
2004 : }
2005 : #endif /* CONFIG_ELF_CORE */
2006 :
2007 : #ifdef CONFIG_MIGRATION
2008 : /*
2009 : * Returns the number of collected pages. Return value is always >= 0.
2010 : */
2011 0 : static unsigned long collect_longterm_unpinnable_pages(
2012 : struct list_head *movable_page_list,
2013 : unsigned long nr_pages,
2014 : struct page **pages)
2015 : {
2016 0 : unsigned long i, collected = 0;
2017 0 : struct folio *prev_folio = NULL;
2018 0 : bool drain_allow = true;
2019 :
2020 0 : for (i = 0; i < nr_pages; i++) {
2021 0 : struct folio *folio = page_folio(pages[i]);
2022 :
2023 0 : if (folio == prev_folio)
2024 0 : continue;
2025 0 : prev_folio = folio;
2026 :
2027 0 : if (folio_is_longterm_pinnable(folio))
2028 0 : continue;
2029 :
2030 0 : collected++;
2031 :
2032 0 : if (folio_is_device_coherent(folio))
2033 : continue;
2034 :
2035 0 : if (folio_test_hugetlb(folio)) {
2036 : isolate_hugetlb(folio, movable_page_list);
2037 : continue;
2038 : }
2039 :
2040 0 : if (!folio_test_lru(folio) && drain_allow) {
2041 0 : lru_add_drain_all();
2042 0 : drain_allow = false;
2043 : }
2044 :
2045 0 : if (!folio_isolate_lru(folio))
2046 0 : continue;
2047 :
2048 0 : list_add_tail(&folio->lru, movable_page_list);
2049 0 : node_stat_mod_folio(folio,
2050 0 : NR_ISOLATED_ANON + folio_is_file_lru(folio),
2051 : folio_nr_pages(folio));
2052 : }
2053 :
2054 0 : return collected;
2055 : }
2056 :
2057 : /*
2058 : * Unpins all pages and migrates device coherent pages and movable_page_list.
2059 : * Returns -EAGAIN if all pages were successfully migrated or -errno for failure
2060 : * (or partial success).
2061 : */
2062 0 : static int migrate_longterm_unpinnable_pages(
2063 : struct list_head *movable_page_list,
2064 : unsigned long nr_pages,
2065 : struct page **pages)
2066 : {
2067 : int ret;
2068 : unsigned long i;
2069 :
2070 0 : for (i = 0; i < nr_pages; i++) {
2071 0 : struct folio *folio = page_folio(pages[i]);
2072 :
2073 0 : if (folio_is_device_coherent(folio)) {
2074 : /*
2075 : * Migration will fail if the page is pinned, so convert
2076 : * the pin on the source page to a normal reference.
2077 : */
2078 : pages[i] = NULL;
2079 : folio_get(folio);
2080 : gup_put_folio(folio, 1, FOLL_PIN);
2081 :
2082 : if (migrate_device_coherent_page(&folio->page)) {
2083 : ret = -EBUSY;
2084 : goto err;
2085 : }
2086 :
2087 : continue;
2088 : }
2089 :
2090 : /*
2091 : * We can't migrate pages with unexpected references, so drop
2092 : * the reference obtained by __get_user_pages_locked().
2093 : * Migrating pages have been added to movable_page_list after
2094 : * calling folio_isolate_lru() which takes a reference so the
2095 : * page won't be freed if it's migrating.
2096 : */
2097 0 : unpin_user_page(pages[i]);
2098 0 : pages[i] = NULL;
2099 : }
2100 :
2101 0 : if (!list_empty(movable_page_list)) {
2102 0 : struct migration_target_control mtc = {
2103 : .nid = NUMA_NO_NODE,
2104 : .gfp_mask = GFP_USER | __GFP_NOWARN,
2105 : };
2106 :
2107 0 : if (migrate_pages(movable_page_list, alloc_migration_target,
2108 : NULL, (unsigned long)&mtc, MIGRATE_SYNC,
2109 : MR_LONGTERM_PIN, NULL)) {
2110 0 : ret = -ENOMEM;
2111 0 : goto err;
2112 : }
2113 : }
2114 :
2115 0 : putback_movable_pages(movable_page_list);
2116 :
2117 0 : return -EAGAIN;
2118 :
2119 : err:
2120 0 : for (i = 0; i < nr_pages; i++)
2121 0 : if (pages[i])
2122 0 : unpin_user_page(pages[i]);
2123 0 : putback_movable_pages(movable_page_list);
2124 :
2125 0 : return ret;
2126 : }
2127 :
2128 : /*
2129 : * Check whether all pages are *allowed* to be pinned. Rather confusingly, all
2130 : * pages in the range are required to be pinned via FOLL_PIN, before calling
2131 : * this routine.
2132 : *
2133 : * If any pages in the range are not allowed to be pinned, then this routine
2134 : * will migrate those pages away, unpin all the pages in the range and return
2135 : * -EAGAIN. The caller should re-pin the entire range with FOLL_PIN and then
2136 : * call this routine again.
2137 : *
2138 : * If an error other than -EAGAIN occurs, this indicates a migration failure.
2139 : * The caller should give up, and propagate the error back up the call stack.
2140 : *
2141 : * If everything is OK and all pages in the range are allowed to be pinned, then
2142 : * this routine leaves all pages pinned and returns zero for success.
2143 : */
2144 0 : static long check_and_migrate_movable_pages(unsigned long nr_pages,
2145 : struct page **pages)
2146 : {
2147 : unsigned long collected;
2148 0 : LIST_HEAD(movable_page_list);
2149 :
2150 0 : collected = collect_longterm_unpinnable_pages(&movable_page_list,
2151 : nr_pages, pages);
2152 0 : if (!collected)
2153 : return 0;
2154 :
2155 0 : return migrate_longterm_unpinnable_pages(&movable_page_list, nr_pages,
2156 : pages);
2157 : }
2158 : #else
2159 : static long check_and_migrate_movable_pages(unsigned long nr_pages,
2160 : struct page **pages)
2161 : {
2162 : return 0;
2163 : }
2164 : #endif /* CONFIG_MIGRATION */
2165 :
2166 : /*
2167 : * __gup_longterm_locked() is a wrapper for __get_user_pages_locked which
2168 : * allows us to process the FOLL_LONGTERM flag.
2169 : */
2170 0 : static long __gup_longterm_locked(struct mm_struct *mm,
2171 : unsigned long start,
2172 : unsigned long nr_pages,
2173 : struct page **pages,
2174 : int *locked,
2175 : unsigned int gup_flags)
2176 : {
2177 : unsigned int flags;
2178 : long rc, nr_pinned_pages;
2179 :
2180 0 : if (!(gup_flags & FOLL_LONGTERM))
2181 : return __get_user_pages_locked(mm, start, nr_pages, pages,
2182 : locked, gup_flags);
2183 :
2184 0 : flags = memalloc_pin_save();
2185 : do {
2186 0 : nr_pinned_pages = __get_user_pages_locked(mm, start, nr_pages,
2187 : pages, locked,
2188 : gup_flags);
2189 0 : if (nr_pinned_pages <= 0) {
2190 : rc = nr_pinned_pages;
2191 : break;
2192 : }
2193 :
2194 : /* FOLL_LONGTERM implies FOLL_PIN */
2195 0 : rc = check_and_migrate_movable_pages(nr_pinned_pages, pages);
2196 0 : } while (rc == -EAGAIN);
2197 0 : memalloc_pin_restore(flags);
2198 0 : return rc ? rc : nr_pinned_pages;
2199 : }
2200 :
2201 : /*
2202 : * Check that the given flags are valid for the exported gup/pup interface, and
2203 : * update them with the required flags that the caller must have set.
2204 : */
2205 0 : static bool is_valid_gup_args(struct page **pages, int *locked,
2206 : unsigned int *gup_flags_p, unsigned int to_set)
2207 : {
2208 0 : unsigned int gup_flags = *gup_flags_p;
2209 :
2210 : /*
2211 : * These flags not allowed to be specified externally to the gup
2212 : * interfaces:
2213 : * - FOLL_PIN/FOLL_TRIED/FOLL_FAST_ONLY are internal only
2214 : * - FOLL_REMOTE is internal only and used on follow_page()
2215 : * - FOLL_UNLOCKABLE is internal only and used if locked is !NULL
2216 : */
2217 0 : if (WARN_ON_ONCE(gup_flags & (FOLL_PIN | FOLL_TRIED | FOLL_UNLOCKABLE |
2218 : FOLL_REMOTE | FOLL_FAST_ONLY)))
2219 : return false;
2220 :
2221 0 : gup_flags |= to_set;
2222 0 : if (locked) {
2223 : /* At the external interface locked must be set */
2224 0 : if (WARN_ON_ONCE(*locked != 1))
2225 : return false;
2226 :
2227 0 : gup_flags |= FOLL_UNLOCKABLE;
2228 : }
2229 :
2230 : /* FOLL_GET and FOLL_PIN are mutually exclusive. */
2231 0 : if (WARN_ON_ONCE((gup_flags & (FOLL_PIN | FOLL_GET)) ==
2232 : (FOLL_PIN | FOLL_GET)))
2233 : return false;
2234 :
2235 : /* LONGTERM can only be specified when pinning */
2236 0 : if (WARN_ON_ONCE(!(gup_flags & FOLL_PIN) && (gup_flags & FOLL_LONGTERM)))
2237 : return false;
2238 :
2239 : /* Pages input must be given if using GET/PIN */
2240 0 : if (WARN_ON_ONCE((gup_flags & (FOLL_GET | FOLL_PIN)) && !pages))
2241 : return false;
2242 :
2243 : /* We want to allow the pgmap to be hot-unplugged at all times */
2244 0 : if (WARN_ON_ONCE((gup_flags & FOLL_LONGTERM) &&
2245 : (gup_flags & FOLL_PCI_P2PDMA)))
2246 : return false;
2247 :
2248 0 : *gup_flags_p = gup_flags;
2249 0 : return true;
2250 : }
2251 :
2252 : #ifdef CONFIG_MMU
2253 : /**
2254 : * get_user_pages_remote() - pin user pages in memory
2255 : * @mm: mm_struct of target mm
2256 : * @start: starting user address
2257 : * @nr_pages: number of pages from start to pin
2258 : * @gup_flags: flags modifying lookup behaviour
2259 : * @pages: array that receives pointers to the pages pinned.
2260 : * Should be at least nr_pages long. Or NULL, if caller
2261 : * only intends to ensure the pages are faulted in.
2262 : * @locked: pointer to lock flag indicating whether lock is held and
2263 : * subsequently whether VM_FAULT_RETRY functionality can be
2264 : * utilised. Lock must initially be held.
2265 : *
2266 : * Returns either number of pages pinned (which may be less than the
2267 : * number requested), or an error. Details about the return value:
2268 : *
2269 : * -- If nr_pages is 0, returns 0.
2270 : * -- If nr_pages is >0, but no pages were pinned, returns -errno.
2271 : * -- If nr_pages is >0, and some pages were pinned, returns the number of
2272 : * pages pinned. Again, this may be less than nr_pages.
2273 : *
2274 : * The caller is responsible for releasing returned @pages, via put_page().
2275 : *
2276 : * Must be called with mmap_lock held for read or write.
2277 : *
2278 : * get_user_pages_remote walks a process's page tables and takes a reference
2279 : * to each struct page that each user address corresponds to at a given
2280 : * instant. That is, it takes the page that would be accessed if a user
2281 : * thread accesses the given user virtual address at that instant.
2282 : *
2283 : * This does not guarantee that the page exists in the user mappings when
2284 : * get_user_pages_remote returns, and there may even be a completely different
2285 : * page there in some cases (eg. if mmapped pagecache has been invalidated
2286 : * and subsequently re-faulted). However it does guarantee that the page
2287 : * won't be freed completely. And mostly callers simply care that the page
2288 : * contains data that was valid *at some point in time*. Typically, an IO
2289 : * or similar operation cannot guarantee anything stronger anyway because
2290 : * locks can't be held over the syscall boundary.
2291 : *
2292 : * If gup_flags & FOLL_WRITE == 0, the page must not be written to. If the page
2293 : * is written to, set_page_dirty (or set_page_dirty_lock, as appropriate) must
2294 : * be called after the page is finished with, and before put_page is called.
2295 : *
2296 : * get_user_pages_remote is typically used for fewer-copy IO operations,
2297 : * to get a handle on the memory by some means other than accesses
2298 : * via the user virtual addresses. The pages may be submitted for
2299 : * DMA to devices or accessed via their kernel linear mapping (via the
2300 : * kmap APIs). Care should be taken to use the correct cache flushing APIs.
2301 : *
2302 : * See also get_user_pages_fast, for performance critical applications.
2303 : *
2304 : * get_user_pages_remote should be phased out in favor of
2305 : * get_user_pages_locked|unlocked or get_user_pages_fast. Nothing
2306 : * should use get_user_pages_remote because it cannot pass
2307 : * FAULT_FLAG_ALLOW_RETRY to handle_mm_fault.
2308 : */
2309 0 : long get_user_pages_remote(struct mm_struct *mm,
2310 : unsigned long start, unsigned long nr_pages,
2311 : unsigned int gup_flags, struct page **pages,
2312 : int *locked)
2313 : {
2314 0 : int local_locked = 1;
2315 :
2316 0 : if (!is_valid_gup_args(pages, locked, &gup_flags,
2317 : FOLL_TOUCH | FOLL_REMOTE))
2318 : return -EINVAL;
2319 :
2320 0 : return __get_user_pages_locked(mm, start, nr_pages, pages,
2321 : locked ? locked : &local_locked,
2322 : gup_flags);
2323 : }
2324 : EXPORT_SYMBOL(get_user_pages_remote);
2325 :
2326 : #else /* CONFIG_MMU */
2327 : long get_user_pages_remote(struct mm_struct *mm,
2328 : unsigned long start, unsigned long nr_pages,
2329 : unsigned int gup_flags, struct page **pages,
2330 : int *locked)
2331 : {
2332 : return 0;
2333 : }
2334 : #endif /* !CONFIG_MMU */
2335 :
2336 : /**
2337 : * get_user_pages() - pin user pages in memory
2338 : * @start: starting user address
2339 : * @nr_pages: number of pages from start to pin
2340 : * @gup_flags: flags modifying lookup behaviour
2341 : * @pages: array that receives pointers to the pages pinned.
2342 : * Should be at least nr_pages long. Or NULL, if caller
2343 : * only intends to ensure the pages are faulted in.
2344 : *
2345 : * This is the same as get_user_pages_remote(), just with a less-flexible
2346 : * calling convention where we assume that the mm being operated on belongs to
2347 : * the current task, and doesn't allow passing of a locked parameter. We also
2348 : * obviously don't pass FOLL_REMOTE in here.
2349 : */
2350 0 : long get_user_pages(unsigned long start, unsigned long nr_pages,
2351 : unsigned int gup_flags, struct page **pages)
2352 : {
2353 0 : int locked = 1;
2354 :
2355 0 : if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_TOUCH))
2356 : return -EINVAL;
2357 :
2358 0 : return __get_user_pages_locked(current->mm, start, nr_pages, pages,
2359 : &locked, gup_flags);
2360 : }
2361 : EXPORT_SYMBOL(get_user_pages);
2362 :
2363 : /*
2364 : * get_user_pages_unlocked() is suitable to replace the form:
2365 : *
2366 : * mmap_read_lock(mm);
2367 : * get_user_pages(mm, ..., pages, NULL);
2368 : * mmap_read_unlock(mm);
2369 : *
2370 : * with:
2371 : *
2372 : * get_user_pages_unlocked(mm, ..., pages);
2373 : *
2374 : * It is functionally equivalent to get_user_pages_fast so
2375 : * get_user_pages_fast should be used instead if specific gup_flags
2376 : * (e.g. FOLL_FORCE) are not required.
2377 : */
2378 0 : long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
2379 : struct page **pages, unsigned int gup_flags)
2380 : {
2381 0 : int locked = 0;
2382 :
2383 0 : if (!is_valid_gup_args(pages, NULL, &gup_flags,
2384 : FOLL_TOUCH | FOLL_UNLOCKABLE))
2385 : return -EINVAL;
2386 :
2387 0 : return __get_user_pages_locked(current->mm, start, nr_pages, pages,
2388 : &locked, gup_flags);
2389 : }
2390 : EXPORT_SYMBOL(get_user_pages_unlocked);
2391 :
2392 : /*
2393 : * Fast GUP
2394 : *
2395 : * get_user_pages_fast attempts to pin user pages by walking the page
2396 : * tables directly and avoids taking locks. Thus the walker needs to be
2397 : * protected from page table pages being freed from under it, and should
2398 : * block any THP splits.
2399 : *
2400 : * One way to achieve this is to have the walker disable interrupts, and
2401 : * rely on IPIs from the TLB flushing code blocking before the page table
2402 : * pages are freed. This is unsuitable for architectures that do not need
2403 : * to broadcast an IPI when invalidating TLBs.
2404 : *
2405 : * Another way to achieve this is to batch up page table containing pages
2406 : * belonging to more than one mm_user, then rcu_sched a callback to free those
2407 : * pages. Disabling interrupts will allow the fast_gup walker to both block
2408 : * the rcu_sched callback, and an IPI that we broadcast for splitting THPs
2409 : * (which is a relatively rare event). The code below adopts this strategy.
2410 : *
2411 : * Before activating this code, please be aware that the following assumptions
2412 : * are currently made:
2413 : *
2414 : * *) Either MMU_GATHER_RCU_TABLE_FREE is enabled, and tlb_remove_table() is used to
2415 : * free pages containing page tables or TLB flushing requires IPI broadcast.
2416 : *
2417 : * *) ptes can be read atomically by the architecture.
2418 : *
2419 : * *) access_ok is sufficient to validate userspace address ranges.
2420 : *
2421 : * The last two assumptions can be relaxed by the addition of helper functions.
2422 : *
2423 : * This code is based heavily on the PowerPC implementation by Nick Piggin.
2424 : */
2425 : #ifdef CONFIG_HAVE_FAST_GUP
2426 :
2427 : /*
2428 : * Used in the GUP-fast path to determine whether a pin is permitted for a
2429 : * specific folio.
2430 : *
2431 : * This call assumes the caller has pinned the folio, that the lowest page table
2432 : * level still points to this folio, and that interrupts have been disabled.
2433 : *
2434 : * Writing to pinned file-backed dirty tracked folios is inherently problematic
2435 : * (see comment describing the writable_file_mapping_allowed() function). We
2436 : * therefore try to avoid the most egregious case of a long-term mapping doing
2437 : * so.
2438 : *
2439 : * This function cannot be as thorough as that one as the VMA is not available
2440 : * in the fast path, so instead we whitelist known good cases and if in doubt,
2441 : * fall back to the slow path.
2442 : */
2443 : static bool folio_fast_pin_allowed(struct folio *folio, unsigned int flags)
2444 : {
2445 : struct address_space *mapping;
2446 : unsigned long mapping_flags;
2447 :
2448 : /*
2449 : * If we aren't pinning then no problematic write can occur. A long term
2450 : * pin is the most egregious case so this is the one we disallow.
2451 : */
2452 : if ((flags & (FOLL_PIN | FOLL_LONGTERM | FOLL_WRITE)) !=
2453 : (FOLL_PIN | FOLL_LONGTERM | FOLL_WRITE))
2454 : return true;
2455 :
2456 : /* The folio is pinned, so we can safely access folio fields. */
2457 :
2458 : if (WARN_ON_ONCE(folio_test_slab(folio)))
2459 : return false;
2460 :
2461 : /* hugetlb mappings do not require dirty-tracking. */
2462 : if (folio_test_hugetlb(folio))
2463 : return true;
2464 :
2465 : /*
2466 : * GUP-fast disables IRQs. When IRQS are disabled, RCU grace periods
2467 : * cannot proceed, which means no actions performed under RCU can
2468 : * proceed either.
2469 : *
2470 : * inodes and thus their mappings are freed under RCU, which means the
2471 : * mapping cannot be freed beneath us and thus we can safely dereference
2472 : * it.
2473 : */
2474 : lockdep_assert_irqs_disabled();
2475 :
2476 : /*
2477 : * However, there may be operations which _alter_ the mapping, so ensure
2478 : * we read it once and only once.
2479 : */
2480 : mapping = READ_ONCE(folio->mapping);
2481 :
2482 : /*
2483 : * The mapping may have been truncated, in any case we cannot determine
2484 : * if this mapping is safe - fall back to slow path to determine how to
2485 : * proceed.
2486 : */
2487 : if (!mapping)
2488 : return false;
2489 :
2490 : /* Anonymous folios pose no problem. */
2491 : mapping_flags = (unsigned long)mapping & PAGE_MAPPING_FLAGS;
2492 : if (mapping_flags)
2493 : return mapping_flags & PAGE_MAPPING_ANON;
2494 :
2495 : /*
2496 : * At this point, we know the mapping is non-null and points to an
2497 : * address_space object. The only remaining whitelisted file system is
2498 : * shmem.
2499 : */
2500 : return shmem_mapping(mapping);
2501 : }
2502 :
2503 : static void __maybe_unused undo_dev_pagemap(int *nr, int nr_start,
2504 : unsigned int flags,
2505 : struct page **pages)
2506 : {
2507 : while ((*nr) - nr_start) {
2508 : struct page *page = pages[--(*nr)];
2509 :
2510 : ClearPageReferenced(page);
2511 : if (flags & FOLL_PIN)
2512 : unpin_user_page(page);
2513 : else
2514 : put_page(page);
2515 : }
2516 : }
2517 :
2518 : #ifdef CONFIG_ARCH_HAS_PTE_SPECIAL
2519 : /*
2520 : * Fast-gup relies on pte change detection to avoid concurrent pgtable
2521 : * operations.
2522 : *
2523 : * To pin the page, fast-gup needs to do below in order:
2524 : * (1) pin the page (by prefetching pte), then (2) check pte not changed.
2525 : *
2526 : * For the rest of pgtable operations where pgtable updates can be racy
2527 : * with fast-gup, we need to do (1) clear pte, then (2) check whether page
2528 : * is pinned.
2529 : *
2530 : * Above will work for all pte-level operations, including THP split.
2531 : *
2532 : * For THP collapse, it's a bit more complicated because fast-gup may be
2533 : * walking a pgtable page that is being freed (pte is still valid but pmd
2534 : * can be cleared already). To avoid race in such condition, we need to
2535 : * also check pmd here to make sure pmd doesn't change (corresponds to
2536 : * pmdp_collapse_flush() in the THP collapse code path).
2537 : */
2538 : static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr,
2539 : unsigned long end, unsigned int flags,
2540 : struct page **pages, int *nr)
2541 : {
2542 : struct dev_pagemap *pgmap = NULL;
2543 : int nr_start = *nr, ret = 0;
2544 : pte_t *ptep, *ptem;
2545 :
2546 : ptem = ptep = pte_offset_map(&pmd, addr);
2547 : if (!ptep)
2548 : return 0;
2549 : do {
2550 : pte_t pte = ptep_get_lockless(ptep);
2551 : struct page *page;
2552 : struct folio *folio;
2553 :
2554 : if (pte_protnone(pte) && !gup_can_follow_protnone(flags))
2555 : goto pte_unmap;
2556 :
2557 : if (!pte_access_permitted(pte, flags & FOLL_WRITE))
2558 : goto pte_unmap;
2559 :
2560 : if (pte_devmap(pte)) {
2561 : if (unlikely(flags & FOLL_LONGTERM))
2562 : goto pte_unmap;
2563 :
2564 : pgmap = get_dev_pagemap(pte_pfn(pte), pgmap);
2565 : if (unlikely(!pgmap)) {
2566 : undo_dev_pagemap(nr, nr_start, flags, pages);
2567 : goto pte_unmap;
2568 : }
2569 : } else if (pte_special(pte))
2570 : goto pte_unmap;
2571 :
2572 : VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
2573 : page = pte_page(pte);
2574 :
2575 : folio = try_grab_folio(page, 1, flags);
2576 : if (!folio)
2577 : goto pte_unmap;
2578 :
2579 : if (unlikely(page_is_secretmem(page))) {
2580 : gup_put_folio(folio, 1, flags);
2581 : goto pte_unmap;
2582 : }
2583 :
2584 : if (unlikely(pmd_val(pmd) != pmd_val(*pmdp)) ||
2585 : unlikely(pte_val(pte) != pte_val(ptep_get(ptep)))) {
2586 : gup_put_folio(folio, 1, flags);
2587 : goto pte_unmap;
2588 : }
2589 :
2590 : if (!folio_fast_pin_allowed(folio, flags)) {
2591 : gup_put_folio(folio, 1, flags);
2592 : goto pte_unmap;
2593 : }
2594 :
2595 : if (!pte_write(pte) && gup_must_unshare(NULL, flags, page)) {
2596 : gup_put_folio(folio, 1, flags);
2597 : goto pte_unmap;
2598 : }
2599 :
2600 : /*
2601 : * We need to make the page accessible if and only if we are
2602 : * going to access its content (the FOLL_PIN case). Please
2603 : * see Documentation/core-api/pin_user_pages.rst for
2604 : * details.
2605 : */
2606 : if (flags & FOLL_PIN) {
2607 : ret = arch_make_page_accessible(page);
2608 : if (ret) {
2609 : gup_put_folio(folio, 1, flags);
2610 : goto pte_unmap;
2611 : }
2612 : }
2613 : folio_set_referenced(folio);
2614 : pages[*nr] = page;
2615 : (*nr)++;
2616 : } while (ptep++, addr += PAGE_SIZE, addr != end);
2617 :
2618 : ret = 1;
2619 :
2620 : pte_unmap:
2621 : if (pgmap)
2622 : put_dev_pagemap(pgmap);
2623 : pte_unmap(ptem);
2624 : return ret;
2625 : }
2626 : #else
2627 :
2628 : /*
2629 : * If we can't determine whether or not a pte is special, then fail immediately
2630 : * for ptes. Note, we can still pin HugeTLB and THP as these are guaranteed not
2631 : * to be special.
2632 : *
2633 : * For a futex to be placed on a THP tail page, get_futex_key requires a
2634 : * get_user_pages_fast_only implementation that can pin pages. Thus it's still
2635 : * useful to have gup_huge_pmd even if we can't operate on ptes.
2636 : */
2637 : static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr,
2638 : unsigned long end, unsigned int flags,
2639 : struct page **pages, int *nr)
2640 : {
2641 : return 0;
2642 : }
2643 : #endif /* CONFIG_ARCH_HAS_PTE_SPECIAL */
2644 :
2645 : #if defined(CONFIG_ARCH_HAS_PTE_DEVMAP) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
2646 : static int __gup_device_huge(unsigned long pfn, unsigned long addr,
2647 : unsigned long end, unsigned int flags,
2648 : struct page **pages, int *nr)
2649 : {
2650 : int nr_start = *nr;
2651 : struct dev_pagemap *pgmap = NULL;
2652 :
2653 : do {
2654 : struct page *page = pfn_to_page(pfn);
2655 :
2656 : pgmap = get_dev_pagemap(pfn, pgmap);
2657 : if (unlikely(!pgmap)) {
2658 : undo_dev_pagemap(nr, nr_start, flags, pages);
2659 : break;
2660 : }
2661 :
2662 : if (!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page)) {
2663 : undo_dev_pagemap(nr, nr_start, flags, pages);
2664 : break;
2665 : }
2666 :
2667 : SetPageReferenced(page);
2668 : pages[*nr] = page;
2669 : if (unlikely(try_grab_page(page, flags))) {
2670 : undo_dev_pagemap(nr, nr_start, flags, pages);
2671 : break;
2672 : }
2673 : (*nr)++;
2674 : pfn++;
2675 : } while (addr += PAGE_SIZE, addr != end);
2676 :
2677 : put_dev_pagemap(pgmap);
2678 : return addr == end;
2679 : }
2680 :
2681 : static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
2682 : unsigned long end, unsigned int flags,
2683 : struct page **pages, int *nr)
2684 : {
2685 : unsigned long fault_pfn;
2686 : int nr_start = *nr;
2687 :
2688 : fault_pfn = pmd_pfn(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
2689 : if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr))
2690 : return 0;
2691 :
2692 : if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) {
2693 : undo_dev_pagemap(nr, nr_start, flags, pages);
2694 : return 0;
2695 : }
2696 : return 1;
2697 : }
2698 :
2699 : static int __gup_device_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr,
2700 : unsigned long end, unsigned int flags,
2701 : struct page **pages, int *nr)
2702 : {
2703 : unsigned long fault_pfn;
2704 : int nr_start = *nr;
2705 :
2706 : fault_pfn = pud_pfn(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
2707 : if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr))
2708 : return 0;
2709 :
2710 : if (unlikely(pud_val(orig) != pud_val(*pudp))) {
2711 : undo_dev_pagemap(nr, nr_start, flags, pages);
2712 : return 0;
2713 : }
2714 : return 1;
2715 : }
2716 : #else
2717 : static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
2718 : unsigned long end, unsigned int flags,
2719 : struct page **pages, int *nr)
2720 : {
2721 : BUILD_BUG();
2722 : return 0;
2723 : }
2724 :
2725 : static int __gup_device_huge_pud(pud_t pud, pud_t *pudp, unsigned long addr,
2726 : unsigned long end, unsigned int flags,
2727 : struct page **pages, int *nr)
2728 : {
2729 : BUILD_BUG();
2730 : return 0;
2731 : }
2732 : #endif
2733 :
2734 : static int record_subpages(struct page *page, unsigned long addr,
2735 : unsigned long end, struct page **pages)
2736 : {
2737 : int nr;
2738 :
2739 : for (nr = 0; addr != end; nr++, addr += PAGE_SIZE)
2740 : pages[nr] = nth_page(page, nr);
2741 :
2742 : return nr;
2743 : }
2744 :
2745 : #ifdef CONFIG_ARCH_HAS_HUGEPD
2746 : static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end,
2747 : unsigned long sz)
2748 : {
2749 : unsigned long __boundary = (addr + sz) & ~(sz-1);
2750 : return (__boundary - 1 < end - 1) ? __boundary : end;
2751 : }
2752 :
2753 : static int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr,
2754 : unsigned long end, unsigned int flags,
2755 : struct page **pages, int *nr)
2756 : {
2757 : unsigned long pte_end;
2758 : struct page *page;
2759 : struct folio *folio;
2760 : pte_t pte;
2761 : int refs;
2762 :
2763 : pte_end = (addr + sz) & ~(sz-1);
2764 : if (pte_end < end)
2765 : end = pte_end;
2766 :
2767 : pte = huge_ptep_get(ptep);
2768 :
2769 : if (!pte_access_permitted(pte, flags & FOLL_WRITE))
2770 : return 0;
2771 :
2772 : /* hugepages are never "special" */
2773 : VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
2774 :
2775 : page = nth_page(pte_page(pte), (addr & (sz - 1)) >> PAGE_SHIFT);
2776 : refs = record_subpages(page, addr, end, pages + *nr);
2777 :
2778 : folio = try_grab_folio(page, refs, flags);
2779 : if (!folio)
2780 : return 0;
2781 :
2782 : if (unlikely(pte_val(pte) != pte_val(ptep_get(ptep)))) {
2783 : gup_put_folio(folio, refs, flags);
2784 : return 0;
2785 : }
2786 :
2787 : if (!folio_fast_pin_allowed(folio, flags)) {
2788 : gup_put_folio(folio, refs, flags);
2789 : return 0;
2790 : }
2791 :
2792 : if (!pte_write(pte) && gup_must_unshare(NULL, flags, &folio->page)) {
2793 : gup_put_folio(folio, refs, flags);
2794 : return 0;
2795 : }
2796 :
2797 : *nr += refs;
2798 : folio_set_referenced(folio);
2799 : return 1;
2800 : }
2801 :
2802 : static int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
2803 : unsigned int pdshift, unsigned long end, unsigned int flags,
2804 : struct page **pages, int *nr)
2805 : {
2806 : pte_t *ptep;
2807 : unsigned long sz = 1UL << hugepd_shift(hugepd);
2808 : unsigned long next;
2809 :
2810 : ptep = hugepte_offset(hugepd, addr, pdshift);
2811 : do {
2812 : next = hugepte_addr_end(addr, end, sz);
2813 : if (!gup_hugepte(ptep, sz, addr, end, flags, pages, nr))
2814 : return 0;
2815 : } while (ptep++, addr = next, addr != end);
2816 :
2817 : return 1;
2818 : }
2819 : #else
2820 : static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
2821 : unsigned int pdshift, unsigned long end, unsigned int flags,
2822 : struct page **pages, int *nr)
2823 : {
2824 : return 0;
2825 : }
2826 : #endif /* CONFIG_ARCH_HAS_HUGEPD */
2827 :
2828 : static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
2829 : unsigned long end, unsigned int flags,
2830 : struct page **pages, int *nr)
2831 : {
2832 : struct page *page;
2833 : struct folio *folio;
2834 : int refs;
2835 :
2836 : if (!pmd_access_permitted(orig, flags & FOLL_WRITE))
2837 : return 0;
2838 :
2839 : if (pmd_devmap(orig)) {
2840 : if (unlikely(flags & FOLL_LONGTERM))
2841 : return 0;
2842 : return __gup_device_huge_pmd(orig, pmdp, addr, end, flags,
2843 : pages, nr);
2844 : }
2845 :
2846 : page = nth_page(pmd_page(orig), (addr & ~PMD_MASK) >> PAGE_SHIFT);
2847 : refs = record_subpages(page, addr, end, pages + *nr);
2848 :
2849 : folio = try_grab_folio(page, refs, flags);
2850 : if (!folio)
2851 : return 0;
2852 :
2853 : if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) {
2854 : gup_put_folio(folio, refs, flags);
2855 : return 0;
2856 : }
2857 :
2858 : if (!folio_fast_pin_allowed(folio, flags)) {
2859 : gup_put_folio(folio, refs, flags);
2860 : return 0;
2861 : }
2862 : if (!pmd_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) {
2863 : gup_put_folio(folio, refs, flags);
2864 : return 0;
2865 : }
2866 :
2867 : *nr += refs;
2868 : folio_set_referenced(folio);
2869 : return 1;
2870 : }
2871 :
2872 : static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr,
2873 : unsigned long end, unsigned int flags,
2874 : struct page **pages, int *nr)
2875 : {
2876 : struct page *page;
2877 : struct folio *folio;
2878 : int refs;
2879 :
2880 : if (!pud_access_permitted(orig, flags & FOLL_WRITE))
2881 : return 0;
2882 :
2883 : if (pud_devmap(orig)) {
2884 : if (unlikely(flags & FOLL_LONGTERM))
2885 : return 0;
2886 : return __gup_device_huge_pud(orig, pudp, addr, end, flags,
2887 : pages, nr);
2888 : }
2889 :
2890 : page = nth_page(pud_page(orig), (addr & ~PUD_MASK) >> PAGE_SHIFT);
2891 : refs = record_subpages(page, addr, end, pages + *nr);
2892 :
2893 : folio = try_grab_folio(page, refs, flags);
2894 : if (!folio)
2895 : return 0;
2896 :
2897 : if (unlikely(pud_val(orig) != pud_val(*pudp))) {
2898 : gup_put_folio(folio, refs, flags);
2899 : return 0;
2900 : }
2901 :
2902 : if (!folio_fast_pin_allowed(folio, flags)) {
2903 : gup_put_folio(folio, refs, flags);
2904 : return 0;
2905 : }
2906 :
2907 : if (!pud_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) {
2908 : gup_put_folio(folio, refs, flags);
2909 : return 0;
2910 : }
2911 :
2912 : *nr += refs;
2913 : folio_set_referenced(folio);
2914 : return 1;
2915 : }
2916 :
2917 : static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr,
2918 : unsigned long end, unsigned int flags,
2919 : struct page **pages, int *nr)
2920 : {
2921 : int refs;
2922 : struct page *page;
2923 : struct folio *folio;
2924 :
2925 : if (!pgd_access_permitted(orig, flags & FOLL_WRITE))
2926 : return 0;
2927 :
2928 : BUILD_BUG_ON(pgd_devmap(orig));
2929 :
2930 : page = nth_page(pgd_page(orig), (addr & ~PGDIR_MASK) >> PAGE_SHIFT);
2931 : refs = record_subpages(page, addr, end, pages + *nr);
2932 :
2933 : folio = try_grab_folio(page, refs, flags);
2934 : if (!folio)
2935 : return 0;
2936 :
2937 : if (unlikely(pgd_val(orig) != pgd_val(*pgdp))) {
2938 : gup_put_folio(folio, refs, flags);
2939 : return 0;
2940 : }
2941 :
2942 : if (!pgd_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) {
2943 : gup_put_folio(folio, refs, flags);
2944 : return 0;
2945 : }
2946 :
2947 : if (!folio_fast_pin_allowed(folio, flags)) {
2948 : gup_put_folio(folio, refs, flags);
2949 : return 0;
2950 : }
2951 :
2952 : *nr += refs;
2953 : folio_set_referenced(folio);
2954 : return 1;
2955 : }
2956 :
2957 : static int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr, unsigned long end,
2958 : unsigned int flags, struct page **pages, int *nr)
2959 : {
2960 : unsigned long next;
2961 : pmd_t *pmdp;
2962 :
2963 : pmdp = pmd_offset_lockless(pudp, pud, addr);
2964 : do {
2965 : pmd_t pmd = pmdp_get_lockless(pmdp);
2966 :
2967 : next = pmd_addr_end(addr, end);
2968 : if (!pmd_present(pmd))
2969 : return 0;
2970 :
2971 : if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) ||
2972 : pmd_devmap(pmd))) {
2973 : if (pmd_protnone(pmd) &&
2974 : !gup_can_follow_protnone(flags))
2975 : return 0;
2976 :
2977 : if (!gup_huge_pmd(pmd, pmdp, addr, next, flags,
2978 : pages, nr))
2979 : return 0;
2980 :
2981 : } else if (unlikely(is_hugepd(__hugepd(pmd_val(pmd))))) {
2982 : /*
2983 : * architecture have different format for hugetlbfs
2984 : * pmd format and THP pmd format
2985 : */
2986 : if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr,
2987 : PMD_SHIFT, next, flags, pages, nr))
2988 : return 0;
2989 : } else if (!gup_pte_range(pmd, pmdp, addr, next, flags, pages, nr))
2990 : return 0;
2991 : } while (pmdp++, addr = next, addr != end);
2992 :
2993 : return 1;
2994 : }
2995 :
2996 : static int gup_pud_range(p4d_t *p4dp, p4d_t p4d, unsigned long addr, unsigned long end,
2997 : unsigned int flags, struct page **pages, int *nr)
2998 : {
2999 : unsigned long next;
3000 : pud_t *pudp;
3001 :
3002 : pudp = pud_offset_lockless(p4dp, p4d, addr);
3003 : do {
3004 : pud_t pud = READ_ONCE(*pudp);
3005 :
3006 : next = pud_addr_end(addr, end);
3007 : if (unlikely(!pud_present(pud)))
3008 : return 0;
3009 : if (unlikely(pud_huge(pud) || pud_devmap(pud))) {
3010 : if (!gup_huge_pud(pud, pudp, addr, next, flags,
3011 : pages, nr))
3012 : return 0;
3013 : } else if (unlikely(is_hugepd(__hugepd(pud_val(pud))))) {
3014 : if (!gup_huge_pd(__hugepd(pud_val(pud)), addr,
3015 : PUD_SHIFT, next, flags, pages, nr))
3016 : return 0;
3017 : } else if (!gup_pmd_range(pudp, pud, addr, next, flags, pages, nr))
3018 : return 0;
3019 : } while (pudp++, addr = next, addr != end);
3020 :
3021 : return 1;
3022 : }
3023 :
3024 : static int gup_p4d_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr, unsigned long end,
3025 : unsigned int flags, struct page **pages, int *nr)
3026 : {
3027 : unsigned long next;
3028 : p4d_t *p4dp;
3029 :
3030 : p4dp = p4d_offset_lockless(pgdp, pgd, addr);
3031 : do {
3032 : p4d_t p4d = READ_ONCE(*p4dp);
3033 :
3034 : next = p4d_addr_end(addr, end);
3035 : if (p4d_none(p4d))
3036 : return 0;
3037 : BUILD_BUG_ON(p4d_huge(p4d));
3038 : if (unlikely(is_hugepd(__hugepd(p4d_val(p4d))))) {
3039 : if (!gup_huge_pd(__hugepd(p4d_val(p4d)), addr,
3040 : P4D_SHIFT, next, flags, pages, nr))
3041 : return 0;
3042 : } else if (!gup_pud_range(p4dp, p4d, addr, next, flags, pages, nr))
3043 : return 0;
3044 : } while (p4dp++, addr = next, addr != end);
3045 :
3046 : return 1;
3047 : }
3048 :
3049 : static void gup_pgd_range(unsigned long addr, unsigned long end,
3050 : unsigned int flags, struct page **pages, int *nr)
3051 : {
3052 : unsigned long next;
3053 : pgd_t *pgdp;
3054 :
3055 : pgdp = pgd_offset(current->mm, addr);
3056 : do {
3057 : pgd_t pgd = READ_ONCE(*pgdp);
3058 :
3059 : next = pgd_addr_end(addr, end);
3060 : if (pgd_none(pgd))
3061 : return;
3062 : if (unlikely(pgd_huge(pgd))) {
3063 : if (!gup_huge_pgd(pgd, pgdp, addr, next, flags,
3064 : pages, nr))
3065 : return;
3066 : } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) {
3067 : if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr,
3068 : PGDIR_SHIFT, next, flags, pages, nr))
3069 : return;
3070 : } else if (!gup_p4d_range(pgdp, pgd, addr, next, flags, pages, nr))
3071 : return;
3072 : } while (pgdp++, addr = next, addr != end);
3073 : }
3074 : #else
3075 : static inline void gup_pgd_range(unsigned long addr, unsigned long end,
3076 : unsigned int flags, struct page **pages, int *nr)
3077 : {
3078 : }
3079 : #endif /* CONFIG_HAVE_FAST_GUP */
3080 :
3081 : #ifndef gup_fast_permitted
3082 : /*
3083 : * Check if it's allowed to use get_user_pages_fast_only() for the range, or
3084 : * we need to fall back to the slow version:
3085 : */
3086 : static bool gup_fast_permitted(unsigned long start, unsigned long end)
3087 : {
3088 : return true;
3089 : }
3090 : #endif
3091 :
3092 : static unsigned long lockless_pages_from_mm(unsigned long start,
3093 : unsigned long end,
3094 : unsigned int gup_flags,
3095 : struct page **pages)
3096 : {
3097 : unsigned long flags;
3098 0 : int nr_pinned = 0;
3099 : unsigned seq;
3100 :
3101 : if (!IS_ENABLED(CONFIG_HAVE_FAST_GUP) ||
3102 : !gup_fast_permitted(start, end))
3103 : return 0;
3104 :
3105 : if (gup_flags & FOLL_PIN) {
3106 : seq = raw_read_seqcount(¤t->mm->write_protect_seq);
3107 : if (seq & 1)
3108 : return 0;
3109 : }
3110 :
3111 : /*
3112 : * Disable interrupts. The nested form is used, in order to allow full,
3113 : * general purpose use of this routine.
3114 : *
3115 : * With interrupts disabled, we block page table pages from being freed
3116 : * from under us. See struct mmu_table_batch comments in
3117 : * include/asm-generic/tlb.h for more details.
3118 : *
3119 : * We do not adopt an rcu_read_lock() here as we also want to block IPIs
3120 : * that come from THPs splitting.
3121 : */
3122 : local_irq_save(flags);
3123 : gup_pgd_range(start, end, gup_flags, pages, &nr_pinned);
3124 : local_irq_restore(flags);
3125 :
3126 : /*
3127 : * When pinning pages for DMA there could be a concurrent write protect
3128 : * from fork() via copy_page_range(), in this case always fail fast GUP.
3129 : */
3130 : if (gup_flags & FOLL_PIN) {
3131 : if (read_seqcount_retry(¤t->mm->write_protect_seq, seq)) {
3132 : unpin_user_pages_lockless(pages, nr_pinned);
3133 : return 0;
3134 : } else {
3135 : sanity_check_pinned_pages(pages, nr_pinned);
3136 : }
3137 : }
3138 : return nr_pinned;
3139 : }
3140 :
3141 0 : static int internal_get_user_pages_fast(unsigned long start,
3142 : unsigned long nr_pages,
3143 : unsigned int gup_flags,
3144 : struct page **pages)
3145 : {
3146 : unsigned long len, end;
3147 : unsigned long nr_pinned;
3148 0 : int locked = 0;
3149 : int ret;
3150 :
3151 0 : if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM |
3152 : FOLL_FORCE | FOLL_PIN | FOLL_GET |
3153 : FOLL_FAST_ONLY | FOLL_NOFAULT |
3154 : FOLL_PCI_P2PDMA)))
3155 : return -EINVAL;
3156 :
3157 0 : if (gup_flags & FOLL_PIN)
3158 0 : mm_set_has_pinned_flag(¤t->mm->flags);
3159 :
3160 0 : if (!(gup_flags & FOLL_FAST_ONLY))
3161 : might_lock_read(¤t->mm->mmap_lock);
3162 :
3163 0 : start = untagged_addr(start) & PAGE_MASK;
3164 0 : len = nr_pages << PAGE_SHIFT;
3165 0 : if (check_add_overflow(start, len, &end))
3166 : return -EOVERFLOW;
3167 0 : if (end > TASK_SIZE_MAX)
3168 : return -EFAULT;
3169 0 : if (unlikely(!access_ok((void __user *)start, len)))
3170 : return -EFAULT;
3171 :
3172 0 : nr_pinned = lockless_pages_from_mm(start, end, gup_flags, pages);
3173 0 : if (nr_pinned == nr_pages || gup_flags & FOLL_FAST_ONLY)
3174 : return nr_pinned;
3175 :
3176 : /* Slow path: try to get the remaining pages with get_user_pages */
3177 0 : start += nr_pinned << PAGE_SHIFT;
3178 0 : pages += nr_pinned;
3179 0 : ret = __gup_longterm_locked(current->mm, start, nr_pages - nr_pinned,
3180 : pages, &locked,
3181 : gup_flags | FOLL_TOUCH | FOLL_UNLOCKABLE);
3182 : if (ret < 0) {
3183 : /*
3184 : * The caller has to unpin the pages we already pinned so
3185 : * returning -errno is not an option
3186 : */
3187 : if (nr_pinned)
3188 : return nr_pinned;
3189 : return ret;
3190 : }
3191 : return ret + nr_pinned;
3192 : }
3193 :
3194 : /**
3195 : * get_user_pages_fast_only() - pin user pages in memory
3196 : * @start: starting user address
3197 : * @nr_pages: number of pages from start to pin
3198 : * @gup_flags: flags modifying pin behaviour
3199 : * @pages: array that receives pointers to the pages pinned.
3200 : * Should be at least nr_pages long.
3201 : *
3202 : * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to
3203 : * the regular GUP.
3204 : *
3205 : * If the architecture does not support this function, simply return with no
3206 : * pages pinned.
3207 : *
3208 : * Careful, careful! COW breaking can go either way, so a non-write
3209 : * access can get ambiguous page results. If you call this function without
3210 : * 'write' set, you'd better be sure that you're ok with that ambiguity.
3211 : */
3212 0 : int get_user_pages_fast_only(unsigned long start, int nr_pages,
3213 : unsigned int gup_flags, struct page **pages)
3214 : {
3215 : /*
3216 : * Internally (within mm/gup.c), gup fast variants must set FOLL_GET,
3217 : * because gup fast is always a "pin with a +1 page refcount" request.
3218 : *
3219 : * FOLL_FAST_ONLY is required in order to match the API description of
3220 : * this routine: no fall back to regular ("slow") GUP.
3221 : */
3222 0 : if (!is_valid_gup_args(pages, NULL, &gup_flags,
3223 : FOLL_GET | FOLL_FAST_ONLY))
3224 : return -EINVAL;
3225 :
3226 0 : return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages);
3227 : }
3228 : EXPORT_SYMBOL_GPL(get_user_pages_fast_only);
3229 :
3230 : /**
3231 : * get_user_pages_fast() - pin user pages in memory
3232 : * @start: starting user address
3233 : * @nr_pages: number of pages from start to pin
3234 : * @gup_flags: flags modifying pin behaviour
3235 : * @pages: array that receives pointers to the pages pinned.
3236 : * Should be at least nr_pages long.
3237 : *
3238 : * Attempt to pin user pages in memory without taking mm->mmap_lock.
3239 : * If not successful, it will fall back to taking the lock and
3240 : * calling get_user_pages().
3241 : *
3242 : * Returns number of pages pinned. This may be fewer than the number requested.
3243 : * If nr_pages is 0 or negative, returns 0. If no pages were pinned, returns
3244 : * -errno.
3245 : */
3246 0 : int get_user_pages_fast(unsigned long start, int nr_pages,
3247 : unsigned int gup_flags, struct page **pages)
3248 : {
3249 : /*
3250 : * The caller may or may not have explicitly set FOLL_GET; either way is
3251 : * OK. However, internally (within mm/gup.c), gup fast variants must set
3252 : * FOLL_GET, because gup fast is always a "pin with a +1 page refcount"
3253 : * request.
3254 : */
3255 0 : if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_GET))
3256 : return -EINVAL;
3257 0 : return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages);
3258 : }
3259 : EXPORT_SYMBOL_GPL(get_user_pages_fast);
3260 :
3261 : /**
3262 : * pin_user_pages_fast() - pin user pages in memory without taking locks
3263 : *
3264 : * @start: starting user address
3265 : * @nr_pages: number of pages from start to pin
3266 : * @gup_flags: flags modifying pin behaviour
3267 : * @pages: array that receives pointers to the pages pinned.
3268 : * Should be at least nr_pages long.
3269 : *
3270 : * Nearly the same as get_user_pages_fast(), except that FOLL_PIN is set. See
3271 : * get_user_pages_fast() for documentation on the function arguments, because
3272 : * the arguments here are identical.
3273 : *
3274 : * FOLL_PIN means that the pages must be released via unpin_user_page(). Please
3275 : * see Documentation/core-api/pin_user_pages.rst for further details.
3276 : *
3277 : * Note that if a zero_page is amongst the returned pages, it will not have
3278 : * pins in it and unpin_user_page() will not remove pins from it.
3279 : */
3280 0 : int pin_user_pages_fast(unsigned long start, int nr_pages,
3281 : unsigned int gup_flags, struct page **pages)
3282 : {
3283 0 : if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_PIN))
3284 : return -EINVAL;
3285 0 : return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages);
3286 : }
3287 : EXPORT_SYMBOL_GPL(pin_user_pages_fast);
3288 :
3289 : /**
3290 : * pin_user_pages_remote() - pin pages of a remote process
3291 : *
3292 : * @mm: mm_struct of target mm
3293 : * @start: starting user address
3294 : * @nr_pages: number of pages from start to pin
3295 : * @gup_flags: flags modifying lookup behaviour
3296 : * @pages: array that receives pointers to the pages pinned.
3297 : * Should be at least nr_pages long.
3298 : * @locked: pointer to lock flag indicating whether lock is held and
3299 : * subsequently whether VM_FAULT_RETRY functionality can be
3300 : * utilised. Lock must initially be held.
3301 : *
3302 : * Nearly the same as get_user_pages_remote(), except that FOLL_PIN is set. See
3303 : * get_user_pages_remote() for documentation on the function arguments, because
3304 : * the arguments here are identical.
3305 : *
3306 : * FOLL_PIN means that the pages must be released via unpin_user_page(). Please
3307 : * see Documentation/core-api/pin_user_pages.rst for details.
3308 : *
3309 : * Note that if a zero_page is amongst the returned pages, it will not have
3310 : * pins in it and unpin_user_page*() will not remove pins from it.
3311 : */
3312 0 : long pin_user_pages_remote(struct mm_struct *mm,
3313 : unsigned long start, unsigned long nr_pages,
3314 : unsigned int gup_flags, struct page **pages,
3315 : int *locked)
3316 : {
3317 0 : int local_locked = 1;
3318 :
3319 0 : if (!is_valid_gup_args(pages, locked, &gup_flags,
3320 : FOLL_PIN | FOLL_TOUCH | FOLL_REMOTE))
3321 : return 0;
3322 0 : return __gup_longterm_locked(mm, start, nr_pages, pages,
3323 : locked ? locked : &local_locked,
3324 : gup_flags);
3325 : }
3326 : EXPORT_SYMBOL(pin_user_pages_remote);
3327 :
3328 : /**
3329 : * pin_user_pages() - pin user pages in memory for use by other devices
3330 : *
3331 : * @start: starting user address
3332 : * @nr_pages: number of pages from start to pin
3333 : * @gup_flags: flags modifying lookup behaviour
3334 : * @pages: array that receives pointers to the pages pinned.
3335 : * Should be at least nr_pages long.
3336 : *
3337 : * Nearly the same as get_user_pages(), except that FOLL_TOUCH is not set, and
3338 : * FOLL_PIN is set.
3339 : *
3340 : * FOLL_PIN means that the pages must be released via unpin_user_page(). Please
3341 : * see Documentation/core-api/pin_user_pages.rst for details.
3342 : *
3343 : * Note that if a zero_page is amongst the returned pages, it will not have
3344 : * pins in it and unpin_user_page*() will not remove pins from it.
3345 : */
3346 0 : long pin_user_pages(unsigned long start, unsigned long nr_pages,
3347 : unsigned int gup_flags, struct page **pages)
3348 : {
3349 0 : int locked = 1;
3350 :
3351 0 : if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_PIN))
3352 : return 0;
3353 0 : return __gup_longterm_locked(current->mm, start, nr_pages,
3354 : pages, &locked, gup_flags);
3355 : }
3356 : EXPORT_SYMBOL(pin_user_pages);
3357 :
3358 : /*
3359 : * pin_user_pages_unlocked() is the FOLL_PIN variant of
3360 : * get_user_pages_unlocked(). Behavior is the same, except that this one sets
3361 : * FOLL_PIN and rejects FOLL_GET.
3362 : *
3363 : * Note that if a zero_page is amongst the returned pages, it will not have
3364 : * pins in it and unpin_user_page*() will not remove pins from it.
3365 : */
3366 0 : long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
3367 : struct page **pages, unsigned int gup_flags)
3368 : {
3369 0 : int locked = 0;
3370 :
3371 0 : if (!is_valid_gup_args(pages, NULL, &gup_flags,
3372 : FOLL_PIN | FOLL_TOUCH | FOLL_UNLOCKABLE))
3373 : return 0;
3374 :
3375 0 : return __gup_longterm_locked(current->mm, start, nr_pages, pages,
3376 : &locked, gup_flags);
3377 : }
3378 : EXPORT_SYMBOL(pin_user_pages_unlocked);
|
