1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/kernel.h>
3 #include <linux/errno.h>
5 #include <linux/spinlock.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>
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>
23 #include <asm/mmu_context.h>
24 #include <asm/tlbflush.h>
28 struct follow_page_context {
29 struct dev_pagemap *pgmap;
30 unsigned int page_mask;
33 static inline void sanity_check_pinned_pages(struct page **pages,
36 if (!IS_ENABLED(CONFIG_DEBUG_VM))
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.
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.
51 for (; npages; npages--, pages++) {
52 struct page *page = *pages;
53 struct folio *folio = page_folio(page);
55 if (is_zero_page(page) ||
56 !folio_test_anon(folio))
58 if (!folio_test_large(folio) || folio_test_hugetlb(folio))
59 VM_BUG_ON_PAGE(!PageAnonExclusive(&folio->page), page);
61 /* Either a PTE-mapped or a PMD-mapped THP. */
62 VM_BUG_ON_PAGE(!PageAnonExclusive(&folio->page) &&
63 !PageAnonExclusive(page), page);
68 * Return the folio with ref appropriately incremented,
69 * or NULL if that failed.
71 static inline struct folio *try_get_folio(struct page *page, int refs)
76 folio = page_folio(page);
77 if (WARN_ON_ONCE(folio_ref_count(folio) < 0))
79 if (unlikely(!folio_ref_try_add_rcu(folio, refs)))
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.
91 if (unlikely(page_folio(page) != folio)) {
92 if (!put_devmap_managed_page_refs(&folio->page, refs))
93 folio_put_refs(folio, refs);
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.
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.
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:
113 * FOLL_GET: folio's refcount will be incremented by @refs.
115 * FOLL_PIN on large folios: folio's refcount will be incremented by
116 * @refs, and its pincount will be incremented by @refs.
118 * FOLL_PIN on single-page folios: folio's refcount will be incremented by
119 * @refs * GUP_PIN_COUNTING_BIAS.
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.
126 struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags)
130 if (WARN_ON_ONCE((flags & (FOLL_GET | FOLL_PIN)) == 0))
133 if (unlikely(!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page)))
136 if (flags & FOLL_GET)
137 return try_get_folio(page, refs);
139 /* FOLL_PIN is set */
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.
145 if (is_zero_page(page))
146 return page_folio(page);
148 folio = try_get_folio(page, refs);
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
157 if (unlikely((flags & FOLL_LONGTERM) &&
158 !folio_is_longterm_pinnable(folio))) {
159 if (!put_devmap_managed_page_refs(&folio->page, refs))
160 folio_put_refs(folio, refs);
165 * When pinning a large folio, use an exact count to track it.
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.
172 if (folio_test_large(folio))
173 atomic_add(refs, &folio->_pincount);
176 refs * (GUP_PIN_COUNTING_BIAS - 1));
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().
182 smp_mb__after_atomic();
184 node_stat_mod_folio(folio, NR_FOLL_PIN_ACQUIRED, refs);
189 static void gup_put_folio(struct folio *folio, int refs, unsigned int flags)
191 if (flags & FOLL_PIN) {
192 if (is_zero_folio(folio))
194 node_stat_mod_folio(folio, NR_FOLL_PIN_RELEASED, refs);
195 if (folio_test_large(folio))
196 atomic_sub(refs, &folio->_pincount);
198 refs *= GUP_PIN_COUNTING_BIAS;
201 if (!put_devmap_managed_page_refs(&folio->page, refs))
202 folio_put_refs(folio, refs);
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.
210 * This might not do anything at all, depending on the flags argument.
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.
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
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:
222 * -ENOMEM FOLL_GET or FOLL_PIN was set, but the page could not
225 int __must_check try_grab_page(struct page *page, unsigned int flags)
227 struct folio *folio = page_folio(page);
229 if (WARN_ON_ONCE(folio_ref_count(folio) <= 0))
232 if (unlikely(!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page)))
235 if (flags & FOLL_GET)
236 folio_ref_inc(folio);
237 else if (flags & FOLL_PIN) {
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.
242 if (is_zero_page(page))
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.
250 if (folio_test_large(folio)) {
251 folio_ref_add(folio, 1);
252 atomic_add(1, &folio->_pincount);
254 folio_ref_add(folio, GUP_PIN_COUNTING_BIAS);
257 node_stat_mod_folio(folio, NR_FOLL_PIN_ACQUIRED, 1);
264 * unpin_user_page() - release a dma-pinned page
265 * @page: pointer to page to be released
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.
272 void unpin_user_page(struct page *page)
274 sanity_check_pinned_pages(&page, 1);
275 gup_put_folio(page_folio(page), 1, FOLL_PIN);
277 EXPORT_SYMBOL(unpin_user_page);
280 * folio_add_pin - Try to get an additional pin on a pinned folio
281 * @folio: The folio to be pinned
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.
286 void folio_add_pin(struct folio *folio)
288 if (is_zero_folio(folio))
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
296 if (folio_test_large(folio)) {
297 WARN_ON_ONCE(atomic_read(&folio->_pincount) < 1);
298 folio_ref_inc(folio);
299 atomic_inc(&folio->_pincount);
301 WARN_ON_ONCE(folio_ref_count(folio) < GUP_PIN_COUNTING_BIAS);
302 folio_ref_add(folio, GUP_PIN_COUNTING_BIAS);
306 static inline struct folio *gup_folio_range_next(struct page *start,
307 unsigned long npages, unsigned long i, unsigned int *ntails)
309 struct page *next = nth_page(start, i);
310 struct folio *folio = page_folio(next);
313 if (folio_test_large(folio))
314 nr = min_t(unsigned int, npages - i,
315 folio_nr_pages(folio) - folio_page_idx(folio, next));
321 static inline struct folio *gup_folio_next(struct page **list,
322 unsigned long npages, unsigned long i, unsigned int *ntails)
324 struct folio *folio = page_folio(list[i]);
327 for (nr = i + 1; nr < npages; nr++) {
328 if (page_folio(list[nr]) != folio)
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
342 * "gup-pinned page" refers to a page that has had one of the get_user_pages()
343 * variants called on that page.
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.
350 * Please see the unpin_user_page() documentation for details.
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().
358 void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages,
366 unpin_user_pages(pages, npages);
370 sanity_check_pinned_pages(pages, npages);
371 for (i = 0; i < npages; i += nr) {
372 folio = gup_folio_next(pages, npages, i, &nr);
374 * Checking PageDirty at this point may race with
375 * clear_page_dirty_for_io(), but that's OK. Two key
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
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.
393 if (!folio_test_dirty(folio)) {
395 folio_mark_dirty(folio);
398 gup_put_folio(folio, nr, FOLL_PIN);
401 EXPORT_SYMBOL(unpin_user_pages_dirty_lock);
404 * unpin_user_page_range_dirty_lock() - release and optionally dirty
405 * gup-pinned page range
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
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.
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.
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().
424 void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages,
431 for (i = 0; i < npages; i += nr) {
432 folio = gup_folio_range_next(page, npages, i, &nr);
433 if (make_dirty && !folio_test_dirty(folio)) {
435 folio_mark_dirty(folio);
438 gup_put_folio(folio, nr, FOLL_PIN);
441 EXPORT_SYMBOL(unpin_user_page_range_dirty_lock);
443 static void unpin_user_pages_lockless(struct page **pages, unsigned long npages)
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.
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);
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.
465 * For each page in the @pages array, release the page using unpin_user_page().
467 * Please see the unpin_user_page() documentation for details.
469 void unpin_user_pages(struct page **pages, unsigned long npages)
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.
480 if (WARN_ON(IS_ERR_VALUE(npages)))
483 sanity_check_pinned_pages(pages, npages);
484 for (i = 0; i < npages; i += nr) {
485 folio = gup_folio_next(pages, npages, i, &nr);
486 gup_put_folio(folio, nr, FOLL_PIN);
489 EXPORT_SYMBOL(unpin_user_pages);
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.
496 static inline void mm_set_has_pinned_flag(unsigned long *mm_flags)
498 if (!test_bit(MMF_HAS_PINNED, mm_flags))
499 set_bit(MMF_HAS_PINNED, mm_flags);
503 static struct page *no_page_table(struct vm_area_struct *vma,
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.
514 if ((flags & FOLL_DUMP) &&
515 (vma_is_anonymous(vma) || !vma->vm_ops->fault))
516 return ERR_PTR(-EFAULT);
520 static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
521 pte_t *pte, unsigned int flags)
523 if (flags & FOLL_TOUCH) {
524 pte_t orig_entry = ptep_get(pte);
525 pte_t entry = orig_entry;
527 if (flags & FOLL_WRITE)
528 entry = pte_mkdirty(entry);
529 entry = pte_mkyoung(entry);
531 if (!pte_same(orig_entry, entry)) {
532 set_pte_at(vma->vm_mm, address, pte, entry);
533 update_mmu_cache(vma, address, pte);
537 /* Proper page table entry exists, but no corresponding struct page */
541 /* FOLL_FORCE can write to even unwritable PTEs in COW mappings. */
542 static inline bool can_follow_write_pte(pte_t pte, struct page *page,
543 struct vm_area_struct *vma,
546 /* If the pte is writable, we can write to the page. */
550 /* Maybe FOLL_FORCE is set to override it? */
551 if (!(flags & FOLL_FORCE))
554 /* But FOLL_FORCE has no effect on shared mappings */
555 if (vma->vm_flags & (VM_MAYSHARE | VM_SHARED))
558 /* ... or read-only private ones */
559 if (!(vma->vm_flags & VM_MAYWRITE))
562 /* ... or already writable ones that just need to take a write fault */
563 if (vma->vm_flags & VM_WRITE)
567 * See can_change_pte_writable(): we broke COW and could map the page
568 * writable if we have an exclusive anonymous page ...
570 if (!page || !PageAnon(page) || !PageAnonExclusive(page))
573 /* ... and a write-fault isn't required for other reasons. */
574 if (vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte))
576 return !userfaultfd_pte_wp(vma, pte);
579 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)
583 struct mm_struct *mm = vma->vm_mm;
589 /* FOLL_GET and FOLL_PIN are mutually exclusive. */
590 if (WARN_ON_ONCE((flags & (FOLL_PIN | FOLL_GET)) ==
591 (FOLL_PIN | FOLL_GET)))
592 return ERR_PTR(-EINVAL);
594 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
596 return no_page_table(vma, flags);
597 pte = ptep_get(ptep);
598 if (!pte_present(pte))
600 if (pte_protnone(pte) && !gup_can_follow_protnone(flags))
603 page = vm_normal_page(vma, address, pte);
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.
609 if ((flags & FOLL_WRITE) &&
610 !can_follow_write_pte(pte, page, vma, flags)) {
615 if (!page && pte_devmap(pte) && (flags & (FOLL_GET | FOLL_PIN))) {
617 * Only return device mapping pages in the FOLL_GET or FOLL_PIN
618 * case since they are only valid while holding the pgmap
621 *pgmap = get_dev_pagemap(pte_pfn(pte), *pgmap);
623 page = pte_page(pte);
626 } else if (unlikely(!page)) {
627 if (flags & FOLL_DUMP) {
628 /* Avoid special (like zero) pages in core dumps */
629 page = ERR_PTR(-EFAULT);
633 if (is_zero_pfn(pte_pfn(pte))) {
634 page = pte_page(pte);
636 ret = follow_pfn_pte(vma, address, ptep, flags);
642 if (!pte_write(pte) && gup_must_unshare(vma, flags, page)) {
643 page = ERR_PTR(-EMLINK);
647 VM_BUG_ON_PAGE((flags & FOLL_PIN) && PageAnon(page) &&
648 !PageAnonExclusive(page), page);
650 /* try_grab_page() does nothing unless FOLL_GET or FOLL_PIN is set. */
651 ret = try_grab_page(page, flags);
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.
662 if (flags & FOLL_PIN) {
663 ret = arch_make_page_accessible(page);
665 unpin_user_page(page);
670 if (flags & FOLL_TOUCH) {
671 if ((flags & FOLL_WRITE) &&
672 !pte_dirty(pte) && !PageDirty(page))
673 set_page_dirty(page);
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().
679 mark_page_accessed(page);
682 pte_unmap_unlock(ptep, ptl);
685 pte_unmap_unlock(ptep, ptl);
688 return no_page_table(vma, flags);
691 static struct page *follow_pmd_mask(struct vm_area_struct *vma,
692 unsigned long address, pud_t *pudp,
694 struct follow_page_context *ctx)
699 struct mm_struct *mm = vma->vm_mm;
701 pmd = pmd_offset(pudp, address);
702 pmdval = pmdp_get_lockless(pmd);
703 if (pmd_none(pmdval))
704 return no_page_table(vma, flags);
705 if (!pmd_present(pmdval))
706 return no_page_table(vma, flags);
707 if (pmd_devmap(pmdval)) {
708 ptl = pmd_lock(mm, pmd);
709 page = follow_devmap_pmd(vma, address, pmd, flags, &ctx->pgmap);
714 if (likely(!pmd_trans_huge(pmdval)))
715 return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
717 if (pmd_protnone(pmdval) && !gup_can_follow_protnone(flags))
718 return no_page_table(vma, flags);
720 ptl = pmd_lock(mm, pmd);
721 if (unlikely(!pmd_present(*pmd))) {
723 return no_page_table(vma, flags);
725 if (unlikely(!pmd_trans_huge(*pmd))) {
727 return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
729 if (flags & FOLL_SPLIT_PMD) {
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);
736 page = follow_trans_huge_pmd(vma, address, pmd, flags);
738 ctx->page_mask = HPAGE_PMD_NR - 1;
742 static struct page *follow_pud_mask(struct vm_area_struct *vma,
743 unsigned long address, p4d_t *p4dp,
745 struct follow_page_context *ctx)
750 struct mm_struct *mm = vma->vm_mm;
752 pud = pud_offset(p4dp, address);
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);
762 if (unlikely(pud_bad(*pud)))
763 return no_page_table(vma, flags);
765 return follow_pmd_mask(vma, address, pud, flags, ctx);
768 static struct page *follow_p4d_mask(struct vm_area_struct *vma,
769 unsigned long address, pgd_t *pgdp,
771 struct follow_page_context *ctx)
775 p4d = p4d_offset(pgdp, address);
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);
782 return follow_pud_mask(vma, address, p4d, flags, ctx);
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
793 * @flags can have FOLL_ flags set, defined in <linux/mm.h>
795 * When getting pages from ZONE_DEVICE memory, the @ctx->pgmap caches
796 * the device's dev_pagemap metadata to avoid repeating expensive lookups.
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.
803 * On output, the @ctx->page_mask is set according to the size of the page.
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()).
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)
815 struct mm_struct *mm = vma->vm_mm;
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.
824 * hugetlb_follow_page_mask is only for follow_page() handling here.
825 * Ordinary GUP uses follow_hugetlb_page for hugetlb processing.
827 if (is_vm_hugetlb_page(vma)) {
828 page = hugetlb_follow_page_mask(vma, address, flags);
830 page = no_page_table(vma, flags);
834 pgd = pgd_offset(mm, address);
836 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
837 return no_page_table(vma, flags);
839 return follow_p4d_mask(vma, address, pgd, flags, ctx);
842 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
843 unsigned int foll_flags)
845 struct follow_page_context ctx = { NULL };
848 if (vma_is_secretmem(vma))
851 if (WARN_ON_ONCE(foll_flags & FOLL_PIN))
854 page = follow_page_mask(vma, address, foll_flags, &ctx);
856 put_dev_pagemap(ctx.pgmap);
860 static int get_gate_page(struct mm_struct *mm, unsigned long address,
861 unsigned int gup_flags, struct vm_area_struct **vma,
872 /* user gate pages are read-only */
873 if (gup_flags & FOLL_WRITE)
875 if (address > TASK_SIZE)
876 pgd = pgd_offset_k(address);
878 pgd = pgd_offset_gate(mm, address);
881 p4d = p4d_offset(pgd, address);
884 pud = pud_offset(p4d, address);
887 pmd = pmd_offset(pud, address);
888 if (!pmd_present(*pmd))
890 pte = pte_offset_map(pmd, address);
893 entry = ptep_get(pte);
896 *vma = get_gate_vma(mm);
899 *page = vm_normal_page(*vma, address, entry);
901 if ((gup_flags & FOLL_DUMP) || !is_zero_pfn(pte_pfn(entry)))
903 *page = pte_page(entry);
905 ret = try_grab_page(*page, gup_flags);
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.
920 static int faultin_page(struct vm_area_struct *vma,
921 unsigned long address, unsigned int *flags, bool unshare,
924 unsigned int fault_flags = 0;
927 if (*flags & FOLL_NOFAULT)
929 if (*flags & FOLL_WRITE)
930 fault_flags |= FAULT_FLAG_WRITE;
931 if (*flags & FOLL_REMOTE)
932 fault_flags |= FAULT_FLAG_REMOTE;
933 if (*flags & FOLL_UNLOCKABLE) {
934 fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
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.
941 if (*flags & FOLL_INTERRUPTIBLE)
942 fault_flags |= FAULT_FLAG_INTERRUPTIBLE;
944 if (*flags & FOLL_NOWAIT)
945 fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT;
946 if (*flags & FOLL_TRIED) {
948 * Note: FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_TRIED
951 fault_flags |= FAULT_FLAG_TRIED;
954 fault_flags |= FAULT_FLAG_UNSHARE;
955 /* FAULT_FLAG_WRITE and FAULT_FLAG_UNSHARE are incompatible */
956 VM_BUG_ON(fault_flags & FAULT_FLAG_WRITE);
959 ret = handle_mm_fault(vma, address, fault_flags, NULL);
961 if (ret & VM_FAULT_COMPLETED) {
963 * With FAULT_FLAG_RETRY_NOWAIT we'll never release the
964 * mmap lock in the page fault handler. Sanity check this.
966 WARN_ON_ONCE(fault_flags & FAULT_FLAG_RETRY_NOWAIT);
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_.
979 if (ret & VM_FAULT_ERROR) {
980 int err = vm_fault_to_errno(ret, *flags);
987 if (ret & VM_FAULT_RETRY) {
988 if (!(fault_flags & FAULT_FLAG_RETRY_NOWAIT))
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.
1001 * Consider the following scenario:-
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
1009 * 4. The GUP caller, now done with the page, unpins it and sets it dirty
1010 * (though it does not have to).
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).
1015 static bool writable_file_mapping_allowed(struct vm_area_struct *vma,
1016 unsigned long gup_flags)
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.
1022 if ((gup_flags & (FOLL_PIN | FOLL_LONGTERM)) !=
1023 (FOLL_PIN | FOLL_LONGTERM))
1027 * If the VMA does not require dirty tracking then no problematic write
1030 return !vma_needs_dirty_tracking(vma);
1033 static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags)
1035 vm_flags_t vm_flags = vma->vm_flags;
1036 int write = (gup_flags & FOLL_WRITE);
1037 int foreign = (gup_flags & FOLL_REMOTE);
1038 bool vma_anon = vma_is_anonymous(vma);
1040 if (vm_flags & (VM_IO | VM_PFNMAP))
1043 if ((gup_flags & FOLL_ANON) && !vma_anon)
1046 if ((gup_flags & FOLL_LONGTERM) && vma_is_fsdax(vma))
1049 if (vma_is_secretmem(vma))
1054 !writable_file_mapping_allowed(vma, gup_flags))
1057 if (!(vm_flags & VM_WRITE)) {
1058 if (!(gup_flags & FOLL_FORCE))
1060 /* hugetlb does not support FOLL_FORCE|FOLL_WRITE. */
1061 if (is_vm_hugetlb_page(vma))
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
1072 if (!is_cow_mapping(vm_flags))
1075 } else if (!(vm_flags & VM_READ)) {
1076 if (!(gup_flags & FOLL_FORCE))
1079 * Is there actually any vma we can reach here which does not
1080 * have VM_MAYREAD set?
1082 if (!(vm_flags & VM_MAYREAD))
1086 * gups are always data accesses, not instruction
1087 * fetches, so execute=false here
1089 if (!arch_vma_access_permitted(vma, write, false, foreign))
1095 * __get_user_pages() - pin user pages in memory
1096 * @mm: mm_struct of target mm
1097 * @start: starting user address
1098 * @nr_pages: number of pages from start to pin
1099 * @gup_flags: flags modifying pin behaviour
1100 * @pages: array that receives pointers to the pages pinned.
1101 * Should be at least nr_pages long. Or NULL, if caller
1102 * only intends to ensure the pages are faulted in.
1103 * @locked: whether we're still with the mmap_lock held
1105 * Returns either number of pages pinned (which may be less than the
1106 * number requested), or an error. Details about the return value:
1108 * -- If nr_pages is 0, returns 0.
1109 * -- If nr_pages is >0, but no pages were pinned, returns -errno.
1110 * -- If nr_pages is >0, and some pages were pinned, returns the number of
1111 * pages pinned. Again, this may be less than nr_pages.
1112 * -- 0 return value is possible when the fault would need to be retried.
1114 * The caller is responsible for releasing returned @pages, via put_page().
1116 * Must be called with mmap_lock held. It may be released. See below.
1118 * __get_user_pages walks a process's page tables and takes a reference to
1119 * each struct page that each user address corresponds to at a given
1120 * instant. That is, it takes the page that would be accessed if a user
1121 * thread accesses the given user virtual address at that instant.
1123 * This does not guarantee that the page exists in the user mappings when
1124 * __get_user_pages returns, and there may even be a completely different
1125 * page there in some cases (eg. if mmapped pagecache has been invalidated
1126 * and subsequently re-faulted). However it does guarantee that the page
1127 * won't be freed completely. And mostly callers simply care that the page
1128 * contains data that was valid *at some point in time*. Typically, an IO
1129 * or similar operation cannot guarantee anything stronger anyway because
1130 * locks can't be held over the syscall boundary.
1132 * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If
1133 * the page is written to, set_page_dirty (or set_page_dirty_lock, as
1134 * appropriate) must be called after the page is finished with, and
1135 * before put_page is called.
1137 * If FOLL_UNLOCKABLE is set without FOLL_NOWAIT then the mmap_lock may
1138 * be released. If this happens *@locked will be set to 0 on return.
1140 * A caller using such a combination of @gup_flags must therefore hold the
1141 * mmap_lock for reading only, and recognize when it's been released. Otherwise,
1142 * it must be held for either reading or writing and will not be released.
1144 * In most cases, get_user_pages or get_user_pages_fast should be used
1145 * instead of __get_user_pages. __get_user_pages should be used only if
1146 * you need some special @gup_flags.
1148 static long __get_user_pages(struct mm_struct *mm,
1149 unsigned long start, unsigned long nr_pages,
1150 unsigned int gup_flags, struct page **pages,
1153 long ret = 0, i = 0;
1154 struct vm_area_struct *vma = NULL;
1155 struct follow_page_context ctx = { NULL };
1160 start = untagged_addr_remote(mm, start);
1162 VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN)));
1166 unsigned int foll_flags = gup_flags;
1167 unsigned int page_increm;
1169 /* first iteration or cross vma bound */
1170 if (!vma || start >= vma->vm_end) {
1171 vma = find_vma(mm, start);
1172 if (vma && (start < vma->vm_start)) {
1173 WARN_ON_ONCE(vma->vm_flags & VM_GROWSDOWN);
1176 if (!vma && in_gate_area(mm, start)) {
1177 ret = get_gate_page(mm, start & PAGE_MASK,
1179 pages ? &pages[i] : NULL);
1190 ret = check_vma_flags(vma, gup_flags);
1194 if (is_vm_hugetlb_page(vma)) {
1195 i = follow_hugetlb_page(mm, vma, pages,
1196 &start, &nr_pages, i,
1200 * We've got a VM_FAULT_RETRY
1201 * and we've lost mmap_lock.
1202 * We must stop here.
1204 BUG_ON(gup_flags & FOLL_NOWAIT);
1212 * If we have a pending SIGKILL, don't keep faulting pages and
1213 * potentially allocating memory.
1215 if (fatal_signal_pending(current)) {
1221 page = follow_page_mask(vma, start, foll_flags, &ctx);
1222 if (!page || PTR_ERR(page) == -EMLINK) {
1223 ret = faultin_page(vma, start, &foll_flags,
1224 PTR_ERR(page) == -EMLINK, locked);
1238 } else if (PTR_ERR(page) == -EEXIST) {
1240 * Proper page table entry exists, but no corresponding
1241 * struct page. If the caller expects **pages to be
1242 * filled in, bail out now, because that can't be done
1246 ret = PTR_ERR(page);
1251 } else if (IS_ERR(page)) {
1252 ret = PTR_ERR(page);
1257 flush_anon_page(vma, page, start);
1258 flush_dcache_page(page);
1262 page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask);
1263 if (page_increm > nr_pages)
1264 page_increm = nr_pages;
1266 start += page_increm * PAGE_SIZE;
1267 nr_pages -= page_increm;
1271 put_dev_pagemap(ctx.pgmap);
1275 static bool vma_permits_fault(struct vm_area_struct *vma,
1276 unsigned int fault_flags)
1278 bool write = !!(fault_flags & FAULT_FLAG_WRITE);
1279 bool foreign = !!(fault_flags & FAULT_FLAG_REMOTE);
1280 vm_flags_t vm_flags = write ? VM_WRITE : VM_READ;
1282 if (!(vm_flags & vma->vm_flags))
1286 * The architecture might have a hardware protection
1287 * mechanism other than read/write that can deny access.
1289 * gup always represents data access, not instruction
1290 * fetches, so execute=false here:
1292 if (!arch_vma_access_permitted(vma, write, false, foreign))
1299 * fixup_user_fault() - manually resolve a user page fault
1300 * @mm: mm_struct of target mm
1301 * @address: user address
1302 * @fault_flags:flags to pass down to handle_mm_fault()
1303 * @unlocked: did we unlock the mmap_lock while retrying, maybe NULL if caller
1304 * does not allow retry. If NULL, the caller must guarantee
1305 * that fault_flags does not contain FAULT_FLAG_ALLOW_RETRY.
1307 * This is meant to be called in the specific scenario where for locking reasons
1308 * we try to access user memory in atomic context (within a pagefault_disable()
1309 * section), this returns -EFAULT, and we want to resolve the user fault before
1312 * Typically this is meant to be used by the futex code.
1314 * The main difference with get_user_pages() is that this function will
1315 * unconditionally call handle_mm_fault() which will in turn perform all the
1316 * necessary SW fixup of the dirty and young bits in the PTE, while
1317 * get_user_pages() only guarantees to update these in the struct page.
1319 * This is important for some architectures where those bits also gate the
1320 * access permission to the page because they are maintained in software. On
1321 * such architectures, gup() will not be enough to make a subsequent access
1324 * This function will not return with an unlocked mmap_lock. So it has not the
1325 * same semantics wrt the @mm->mmap_lock as does filemap_fault().
1327 int fixup_user_fault(struct mm_struct *mm,
1328 unsigned long address, unsigned int fault_flags,
1331 struct vm_area_struct *vma;
1334 address = untagged_addr_remote(mm, address);
1337 fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
1340 vma = find_vma(mm, address);
1343 if (address < vma->vm_start ) {
1344 WARN_ON_ONCE(vma->vm_flags & VM_GROWSDOWN);
1348 if (!vma_permits_fault(vma, fault_flags))
1351 if ((fault_flags & FAULT_FLAG_KILLABLE) &&
1352 fatal_signal_pending(current))
1355 ret = handle_mm_fault(vma, address, fault_flags, NULL);
1357 if (ret & VM_FAULT_COMPLETED) {
1359 * NOTE: it's a pity that we need to retake the lock here
1360 * to pair with the unlock() in the callers. Ideally we
1361 * could tell the callers so they do not need to unlock.
1368 if (ret & VM_FAULT_ERROR) {
1369 int err = vm_fault_to_errno(ret, 0);
1376 if (ret & VM_FAULT_RETRY) {
1379 fault_flags |= FAULT_FLAG_TRIED;
1385 EXPORT_SYMBOL_GPL(fixup_user_fault);
1388 * GUP always responds to fatal signals. When FOLL_INTERRUPTIBLE is
1389 * specified, it'll also respond to generic signals. The caller of GUP
1390 * that has FOLL_INTERRUPTIBLE should take care of the GUP interruption.
1392 static bool gup_signal_pending(unsigned int flags)
1394 if (fatal_signal_pending(current))
1397 if (!(flags & FOLL_INTERRUPTIBLE))
1400 return signal_pending(current);
1404 * Locking: (*locked == 1) means that the mmap_lock has already been acquired by
1405 * the caller. This function may drop the mmap_lock. If it does so, then it will
1406 * set (*locked = 0).
1408 * (*locked == 0) means that the caller expects this function to acquire and
1409 * drop the mmap_lock. Therefore, the value of *locked will still be zero when
1410 * the function returns, even though it may have changed temporarily during
1411 * function execution.
1413 * Please note that this function, unlike __get_user_pages(), will not return 0
1414 * for nr_pages > 0, unless FOLL_NOWAIT is used.
1416 static __always_inline long __get_user_pages_locked(struct mm_struct *mm,
1417 unsigned long start,
1418 unsigned long nr_pages,
1419 struct page **pages,
1423 long ret, pages_done;
1424 bool must_unlock = false;
1427 * The internal caller expects GUP to manage the lock internally and the
1428 * lock must be released when this returns.
1431 if (mmap_read_lock_killable(mm))
1437 mmap_assert_locked(mm);
1439 if (flags & FOLL_PIN)
1440 mm_set_has_pinned_flag(&mm->flags);
1443 * FOLL_PIN and FOLL_GET are mutually exclusive. Traditional behavior
1444 * is to set FOLL_GET if the caller wants pages[] filled in (but has
1445 * carelessly failed to specify FOLL_GET), so keep doing that, but only
1446 * for FOLL_GET, not for the newer FOLL_PIN.
1448 * FOLL_PIN always expects pages to be non-null, but no need to assert
1449 * that here, as any failures will be obvious enough.
1451 if (pages && !(flags & FOLL_PIN))
1456 ret = __get_user_pages(mm, start, nr_pages, flags, pages,
1458 if (!(flags & FOLL_UNLOCKABLE)) {
1459 /* VM_FAULT_RETRY couldn't trigger, bypass */
1464 /* VM_FAULT_RETRY or VM_FAULT_COMPLETED cannot return errors */
1467 BUG_ON(ret >= nr_pages);
1478 * VM_FAULT_RETRY didn't trigger or it was a
1486 * VM_FAULT_RETRY triggered, so seek to the faulting offset.
1487 * For the prefault case (!pages) we only update counts.
1491 start += ret << PAGE_SHIFT;
1493 /* The lock was temporarily dropped, so we must unlock later */
1498 * Repeat on the address that fired VM_FAULT_RETRY
1499 * with both FAULT_FLAG_ALLOW_RETRY and
1500 * FAULT_FLAG_TRIED. Note that GUP can be interrupted
1501 * by fatal signals of even common signals, depending on
1502 * the caller's request. So we need to check it before we
1503 * start trying again otherwise it can loop forever.
1505 if (gup_signal_pending(flags)) {
1507 pages_done = -EINTR;
1511 ret = mmap_read_lock_killable(mm);
1520 ret = __get_user_pages(mm, start, 1, flags | FOLL_TRIED,
1523 /* Continue to retry until we succeeded */
1541 if (must_unlock && *locked) {
1543 * We either temporarily dropped the lock, or the caller
1544 * requested that we both acquire and drop the lock. Either way,
1545 * we must now unlock, and notify the caller of that state.
1547 mmap_read_unlock(mm);
1554 * populate_vma_page_range() - populate a range of pages in the vma.
1556 * @start: start address
1558 * @locked: whether the mmap_lock is still held
1560 * This takes care of mlocking the pages too if VM_LOCKED is set.
1562 * Return either number of pages pinned in the vma, or a negative error
1565 * vma->vm_mm->mmap_lock must be held.
1567 * If @locked is NULL, it may be held for read or write and will
1570 * If @locked is non-NULL, it must held for read only and may be
1571 * released. If it's released, *@locked will be set to 0.
1573 long populate_vma_page_range(struct vm_area_struct *vma,
1574 unsigned long start, unsigned long end, int *locked)
1576 struct mm_struct *mm = vma->vm_mm;
1577 unsigned long nr_pages = (end - start) / PAGE_SIZE;
1578 int local_locked = 1;
1582 VM_BUG_ON(!PAGE_ALIGNED(start));
1583 VM_BUG_ON(!PAGE_ALIGNED(end));
1584 VM_BUG_ON_VMA(start < vma->vm_start, vma);
1585 VM_BUG_ON_VMA(end > vma->vm_end, vma);
1586 mmap_assert_locked(mm);
1589 * Rightly or wrongly, the VM_LOCKONFAULT case has never used
1590 * faultin_page() to break COW, so it has no work to do here.
1592 if (vma->vm_flags & VM_LOCKONFAULT)
1595 gup_flags = FOLL_TOUCH;
1597 * We want to touch writable mappings with a write fault in order
1598 * to break COW, except for shared mappings because these don't COW
1599 * and we would not want to dirty them for nothing.
1601 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
1602 gup_flags |= FOLL_WRITE;
1605 * We want mlock to succeed for regions that have any permissions
1606 * other than PROT_NONE.
1608 if (vma_is_accessible(vma))
1609 gup_flags |= FOLL_FORCE;
1612 gup_flags |= FOLL_UNLOCKABLE;
1615 * We made sure addr is within a VMA, so the following will
1616 * not result in a stack expansion that recurses back here.
1618 ret = __get_user_pages(mm, start, nr_pages, gup_flags,
1619 NULL, locked ? locked : &local_locked);
1625 * faultin_vma_page_range() - populate (prefault) page tables inside the
1626 * given VMA range readable/writable
1628 * This takes care of mlocking the pages, too, if VM_LOCKED is set.
1631 * @start: start address
1633 * @write: whether to prefault readable or writable
1634 * @locked: whether the mmap_lock is still held
1636 * Returns either number of processed pages in the vma, or a negative error
1637 * code on error (see __get_user_pages()).
1639 * vma->vm_mm->mmap_lock must be held. The range must be page-aligned and
1640 * covered by the VMA. If it's released, *@locked will be set to 0.
1642 long faultin_vma_page_range(struct vm_area_struct *vma, unsigned long start,
1643 unsigned long end, bool write, int *locked)
1645 struct mm_struct *mm = vma->vm_mm;
1646 unsigned long nr_pages = (end - start) / PAGE_SIZE;
1650 VM_BUG_ON(!PAGE_ALIGNED(start));
1651 VM_BUG_ON(!PAGE_ALIGNED(end));
1652 VM_BUG_ON_VMA(start < vma->vm_start, vma);
1653 VM_BUG_ON_VMA(end > vma->vm_end, vma);
1654 mmap_assert_locked(mm);
1657 * FOLL_TOUCH: Mark page accessed and thereby young; will also mark
1658 * the page dirty with FOLL_WRITE -- which doesn't make a
1659 * difference with !FOLL_FORCE, because the page is writable
1660 * in the page table.
1661 * FOLL_HWPOISON: Return -EHWPOISON instead of -EFAULT when we hit
1663 * !FOLL_FORCE: Require proper access permissions.
1665 gup_flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_UNLOCKABLE;
1667 gup_flags |= FOLL_WRITE;
1670 * We want to report -EINVAL instead of -EFAULT for any permission
1671 * problems or incompatible mappings.
1673 if (check_vma_flags(vma, gup_flags))
1676 ret = __get_user_pages(mm, start, nr_pages, gup_flags,
1683 * __mm_populate - populate and/or mlock pages within a range of address space.
1685 * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
1686 * flags. VMAs must be already marked with the desired vm_flags, and
1687 * mmap_lock must not be held.
1689 int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
1691 struct mm_struct *mm = current->mm;
1692 unsigned long end, nstart, nend;
1693 struct vm_area_struct *vma = NULL;
1699 for (nstart = start; nstart < end; nstart = nend) {
1701 * We want to fault in pages for [nstart; end) address range.
1702 * Find first corresponding VMA.
1707 vma = find_vma_intersection(mm, nstart, end);
1708 } else if (nstart >= vma->vm_end)
1709 vma = find_vma_intersection(mm, vma->vm_end, end);
1714 * Set [nstart; nend) to intersection of desired address
1715 * range with the first VMA. Also, skip undesirable VMA types.
1717 nend = min(end, vma->vm_end);
1718 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
1720 if (nstart < vma->vm_start)
1721 nstart = vma->vm_start;
1723 * Now fault in a range of pages. populate_vma_page_range()
1724 * double checks the vma flags, so that it won't mlock pages
1725 * if the vma was already munlocked.
1727 ret = populate_vma_page_range(vma, nstart, nend, &locked);
1729 if (ignore_errors) {
1731 continue; /* continue at next VMA */
1735 nend = nstart + ret * PAGE_SIZE;
1739 mmap_read_unlock(mm);
1740 return ret; /* 0 or negative error code */
1742 #else /* CONFIG_MMU */
1743 static long __get_user_pages_locked(struct mm_struct *mm, unsigned long start,
1744 unsigned long nr_pages, struct page **pages,
1745 int *locked, unsigned int foll_flags)
1747 struct vm_area_struct *vma;
1748 bool must_unlock = false;
1749 unsigned long vm_flags;
1756 * The internal caller expects GUP to manage the lock internally and the
1757 * lock must be released when this returns.
1760 if (mmap_read_lock_killable(mm))
1766 /* calculate required read or write permissions.
1767 * If FOLL_FORCE is set, we only require the "MAY" flags.
1769 vm_flags = (foll_flags & FOLL_WRITE) ?
1770 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1771 vm_flags &= (foll_flags & FOLL_FORCE) ?
1772 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1774 for (i = 0; i < nr_pages; i++) {
1775 vma = find_vma(mm, start);
1779 /* protect what we can, including chardevs */
1780 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1781 !(vm_flags & vma->vm_flags))
1785 pages[i] = virt_to_page((void *)start);
1790 start = (start + PAGE_SIZE) & PAGE_MASK;
1793 if (must_unlock && *locked) {
1794 mmap_read_unlock(mm);
1798 return i ? : -EFAULT;
1800 #endif /* !CONFIG_MMU */
1803 * fault_in_writeable - fault in userspace address range for writing
1804 * @uaddr: start of address range
1805 * @size: size of address range
1807 * Returns the number of bytes not faulted in (like copy_to_user() and
1808 * copy_from_user()).
1810 size_t fault_in_writeable(char __user *uaddr, size_t size)
1812 char __user *start = uaddr, *end;
1814 if (unlikely(size == 0))
1816 if (!user_write_access_begin(uaddr, size))
1818 if (!PAGE_ALIGNED(uaddr)) {
1819 unsafe_put_user(0, uaddr, out);
1820 uaddr = (char __user *)PAGE_ALIGN((unsigned long)uaddr);
1822 end = (char __user *)PAGE_ALIGN((unsigned long)start + size);
1823 if (unlikely(end < start))
1825 while (uaddr != end) {
1826 unsafe_put_user(0, uaddr, out);
1831 user_write_access_end();
1832 if (size > uaddr - start)
1833 return size - (uaddr - start);
1836 EXPORT_SYMBOL(fault_in_writeable);
1839 * fault_in_subpage_writeable - fault in an address range for writing
1840 * @uaddr: start of address range
1841 * @size: size of address range
1843 * Fault in a user address range for writing while checking for permissions at
1844 * sub-page granularity (e.g. arm64 MTE). This function should be used when
1845 * the caller cannot guarantee forward progress of a copy_to_user() loop.
1847 * Returns the number of bytes not faulted in (like copy_to_user() and
1848 * copy_from_user()).
1850 size_t fault_in_subpage_writeable(char __user *uaddr, size_t size)
1855 * Attempt faulting in at page granularity first for page table
1856 * permission checking. The arch-specific probe_subpage_writeable()
1857 * functions may not check for this.
1859 faulted_in = size - fault_in_writeable(uaddr, size);
1861 faulted_in -= probe_subpage_writeable(uaddr, faulted_in);
1863 return size - faulted_in;
1865 EXPORT_SYMBOL(fault_in_subpage_writeable);
1868 * fault_in_safe_writeable - fault in an address range for writing
1869 * @uaddr: start of address range
1870 * @size: length of address range
1872 * Faults in an address range for writing. This is primarily useful when we
1873 * already know that some or all of the pages in the address range aren't in
1876 * Unlike fault_in_writeable(), this function is non-destructive.
1878 * Note that we don't pin or otherwise hold the pages referenced that we fault
1879 * in. There's no guarantee that they'll stay in memory for any duration of
1882 * Returns the number of bytes not faulted in, like copy_to_user() and
1885 size_t fault_in_safe_writeable(const char __user *uaddr, size_t size)
1887 unsigned long start = (unsigned long)uaddr, end;
1888 struct mm_struct *mm = current->mm;
1889 bool unlocked = false;
1891 if (unlikely(size == 0))
1893 end = PAGE_ALIGN(start + size);
1899 if (fixup_user_fault(mm, start, FAULT_FLAG_WRITE, &unlocked))
1901 start = (start + PAGE_SIZE) & PAGE_MASK;
1902 } while (start != end);
1903 mmap_read_unlock(mm);
1905 if (size > (unsigned long)uaddr - start)
1906 return size - ((unsigned long)uaddr - start);
1909 EXPORT_SYMBOL(fault_in_safe_writeable);
1912 * fault_in_readable - fault in userspace address range for reading
1913 * @uaddr: start of user address range
1914 * @size: size of user address range
1916 * Returns the number of bytes not faulted in (like copy_to_user() and
1917 * copy_from_user()).
1919 size_t fault_in_readable(const char __user *uaddr, size_t size)
1921 const char __user *start = uaddr, *end;
1924 if (unlikely(size == 0))
1926 if (!user_read_access_begin(uaddr, size))
1928 if (!PAGE_ALIGNED(uaddr)) {
1929 unsafe_get_user(c, uaddr, out);
1930 uaddr = (const char __user *)PAGE_ALIGN((unsigned long)uaddr);
1932 end = (const char __user *)PAGE_ALIGN((unsigned long)start + size);
1933 if (unlikely(end < start))
1935 while (uaddr != end) {
1936 unsafe_get_user(c, uaddr, out);
1941 user_read_access_end();
1943 if (size > uaddr - start)
1944 return size - (uaddr - start);
1947 EXPORT_SYMBOL(fault_in_readable);
1950 * get_dump_page() - pin user page in memory while writing it to core dump
1951 * @addr: user address
1953 * Returns struct page pointer of user page pinned for dump,
1954 * to be freed afterwards by put_page().
1956 * Returns NULL on any kind of failure - a hole must then be inserted into
1957 * the corefile, to preserve alignment with its headers; and also returns
1958 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1959 * allowing a hole to be left in the corefile to save disk space.
1961 * Called without mmap_lock (takes and releases the mmap_lock by itself).
1963 #ifdef CONFIG_ELF_CORE
1964 struct page *get_dump_page(unsigned long addr)
1970 ret = __get_user_pages_locked(current->mm, addr, 1, &page, &locked,
1971 FOLL_FORCE | FOLL_DUMP | FOLL_GET);
1972 return (ret == 1) ? page : NULL;
1974 #endif /* CONFIG_ELF_CORE */
1976 #ifdef CONFIG_MIGRATION
1978 * Returns the number of collected pages. Return value is always >= 0.
1980 static unsigned long collect_longterm_unpinnable_pages(
1981 struct list_head *movable_page_list,
1982 unsigned long nr_pages,
1983 struct page **pages)
1985 unsigned long i, collected = 0;
1986 struct folio *prev_folio = NULL;
1987 bool drain_allow = true;
1989 for (i = 0; i < nr_pages; i++) {
1990 struct folio *folio = page_folio(pages[i]);
1992 if (folio == prev_folio)
1996 if (folio_is_longterm_pinnable(folio))
2001 if (folio_is_device_coherent(folio))
2004 if (folio_test_hugetlb(folio)) {
2005 isolate_hugetlb(folio, movable_page_list);
2009 if (!folio_test_lru(folio) && drain_allow) {
2010 lru_add_drain_all();
2011 drain_allow = false;
2014 if (!folio_isolate_lru(folio))
2017 list_add_tail(&folio->lru, movable_page_list);
2018 node_stat_mod_folio(folio,
2019 NR_ISOLATED_ANON + folio_is_file_lru(folio),
2020 folio_nr_pages(folio));
2027 * Unpins all pages and migrates device coherent pages and movable_page_list.
2028 * Returns -EAGAIN if all pages were successfully migrated or -errno for failure
2029 * (or partial success).
2031 static int migrate_longterm_unpinnable_pages(
2032 struct list_head *movable_page_list,
2033 unsigned long nr_pages,
2034 struct page **pages)
2039 for (i = 0; i < nr_pages; i++) {
2040 struct folio *folio = page_folio(pages[i]);
2042 if (folio_is_device_coherent(folio)) {
2044 * Migration will fail if the page is pinned, so convert
2045 * the pin on the source page to a normal reference.
2049 gup_put_folio(folio, 1, FOLL_PIN);
2051 if (migrate_device_coherent_page(&folio->page)) {
2060 * We can't migrate pages with unexpected references, so drop
2061 * the reference obtained by __get_user_pages_locked().
2062 * Migrating pages have been added to movable_page_list after
2063 * calling folio_isolate_lru() which takes a reference so the
2064 * page won't be freed if it's migrating.
2066 unpin_user_page(pages[i]);
2070 if (!list_empty(movable_page_list)) {
2071 struct migration_target_control mtc = {
2072 .nid = NUMA_NO_NODE,
2073 .gfp_mask = GFP_USER | __GFP_NOWARN,
2076 if (migrate_pages(movable_page_list, alloc_migration_target,
2077 NULL, (unsigned long)&mtc, MIGRATE_SYNC,
2078 MR_LONGTERM_PIN, NULL)) {
2084 putback_movable_pages(movable_page_list);
2089 for (i = 0; i < nr_pages; i++)
2091 unpin_user_page(pages[i]);
2092 putback_movable_pages(movable_page_list);
2098 * Check whether all pages are *allowed* to be pinned. Rather confusingly, all
2099 * pages in the range are required to be pinned via FOLL_PIN, before calling
2102 * If any pages in the range are not allowed to be pinned, then this routine
2103 * will migrate those pages away, unpin all the pages in the range and return
2104 * -EAGAIN. The caller should re-pin the entire range with FOLL_PIN and then
2105 * call this routine again.
2107 * If an error other than -EAGAIN occurs, this indicates a migration failure.
2108 * The caller should give up, and propagate the error back up the call stack.
2110 * If everything is OK and all pages in the range are allowed to be pinned, then
2111 * this routine leaves all pages pinned and returns zero for success.
2113 static long check_and_migrate_movable_pages(unsigned long nr_pages,
2114 struct page **pages)
2116 unsigned long collected;
2117 LIST_HEAD(movable_page_list);
2119 collected = collect_longterm_unpinnable_pages(&movable_page_list,
2124 return migrate_longterm_unpinnable_pages(&movable_page_list, nr_pages,
2128 static long check_and_migrate_movable_pages(unsigned long nr_pages,
2129 struct page **pages)
2133 #endif /* CONFIG_MIGRATION */
2136 * __gup_longterm_locked() is a wrapper for __get_user_pages_locked which
2137 * allows us to process the FOLL_LONGTERM flag.
2139 static long __gup_longterm_locked(struct mm_struct *mm,
2140 unsigned long start,
2141 unsigned long nr_pages,
2142 struct page **pages,
2144 unsigned int gup_flags)
2147 long rc, nr_pinned_pages;
2149 if (!(gup_flags & FOLL_LONGTERM))
2150 return __get_user_pages_locked(mm, start, nr_pages, pages,
2153 flags = memalloc_pin_save();
2155 nr_pinned_pages = __get_user_pages_locked(mm, start, nr_pages,
2158 if (nr_pinned_pages <= 0) {
2159 rc = nr_pinned_pages;
2163 /* FOLL_LONGTERM implies FOLL_PIN */
2164 rc = check_and_migrate_movable_pages(nr_pinned_pages, pages);
2165 } while (rc == -EAGAIN);
2166 memalloc_pin_restore(flags);
2167 return rc ? rc : nr_pinned_pages;
2171 * Check that the given flags are valid for the exported gup/pup interface, and
2172 * update them with the required flags that the caller must have set.
2174 static bool is_valid_gup_args(struct page **pages, int *locked,
2175 unsigned int *gup_flags_p, unsigned int to_set)
2177 unsigned int gup_flags = *gup_flags_p;
2180 * These flags not allowed to be specified externally to the gup
2182 * - FOLL_PIN/FOLL_TRIED/FOLL_FAST_ONLY are internal only
2183 * - FOLL_REMOTE is internal only and used on follow_page()
2184 * - FOLL_UNLOCKABLE is internal only and used if locked is !NULL
2186 if (WARN_ON_ONCE(gup_flags & (FOLL_PIN | FOLL_TRIED | FOLL_UNLOCKABLE |
2187 FOLL_REMOTE | FOLL_FAST_ONLY)))
2190 gup_flags |= to_set;
2192 /* At the external interface locked must be set */
2193 if (WARN_ON_ONCE(*locked != 1))
2196 gup_flags |= FOLL_UNLOCKABLE;
2199 /* FOLL_GET and FOLL_PIN are mutually exclusive. */
2200 if (WARN_ON_ONCE((gup_flags & (FOLL_PIN | FOLL_GET)) ==
2201 (FOLL_PIN | FOLL_GET)))
2204 /* LONGTERM can only be specified when pinning */
2205 if (WARN_ON_ONCE(!(gup_flags & FOLL_PIN) && (gup_flags & FOLL_LONGTERM)))
2208 /* Pages input must be given if using GET/PIN */
2209 if (WARN_ON_ONCE((gup_flags & (FOLL_GET | FOLL_PIN)) && !pages))
2212 /* We want to allow the pgmap to be hot-unplugged at all times */
2213 if (WARN_ON_ONCE((gup_flags & FOLL_LONGTERM) &&
2214 (gup_flags & FOLL_PCI_P2PDMA)))
2217 *gup_flags_p = gup_flags;
2223 * get_user_pages_remote() - pin user pages in memory
2224 * @mm: mm_struct of target mm
2225 * @start: starting user address
2226 * @nr_pages: number of pages from start to pin
2227 * @gup_flags: flags modifying lookup behaviour
2228 * @pages: array that receives pointers to the pages pinned.
2229 * Should be at least nr_pages long. Or NULL, if caller
2230 * only intends to ensure the pages are faulted in.
2231 * @locked: pointer to lock flag indicating whether lock is held and
2232 * subsequently whether VM_FAULT_RETRY functionality can be
2233 * utilised. Lock must initially be held.
2235 * Returns either number of pages pinned (which may be less than the
2236 * number requested), or an error. Details about the return value:
2238 * -- If nr_pages is 0, returns 0.
2239 * -- If nr_pages is >0, but no pages were pinned, returns -errno.
2240 * -- If nr_pages is >0, and some pages were pinned, returns the number of
2241 * pages pinned. Again, this may be less than nr_pages.
2243 * The caller is responsible for releasing returned @pages, via put_page().
2245 * Must be called with mmap_lock held for read or write.
2247 * get_user_pages_remote walks a process's page tables and takes a reference
2248 * to each struct page that each user address corresponds to at a given
2249 * instant. That is, it takes the page that would be accessed if a user
2250 * thread accesses the given user virtual address at that instant.
2252 * This does not guarantee that the page exists in the user mappings when
2253 * get_user_pages_remote returns, and there may even be a completely different
2254 * page there in some cases (eg. if mmapped pagecache has been invalidated
2255 * and subsequently re-faulted). However it does guarantee that the page
2256 * won't be freed completely. And mostly callers simply care that the page
2257 * contains data that was valid *at some point in time*. Typically, an IO
2258 * or similar operation cannot guarantee anything stronger anyway because
2259 * locks can't be held over the syscall boundary.
2261 * If gup_flags & FOLL_WRITE == 0, the page must not be written to. If the page
2262 * is written to, set_page_dirty (or set_page_dirty_lock, as appropriate) must
2263 * be called after the page is finished with, and before put_page is called.
2265 * get_user_pages_remote is typically used for fewer-copy IO operations,
2266 * to get a handle on the memory by some means other than accesses
2267 * via the user virtual addresses. The pages may be submitted for
2268 * DMA to devices or accessed via their kernel linear mapping (via the
2269 * kmap APIs). Care should be taken to use the correct cache flushing APIs.
2271 * See also get_user_pages_fast, for performance critical applications.
2273 * get_user_pages_remote should be phased out in favor of
2274 * get_user_pages_locked|unlocked or get_user_pages_fast. Nothing
2275 * should use get_user_pages_remote because it cannot pass
2276 * FAULT_FLAG_ALLOW_RETRY to handle_mm_fault.
2278 long get_user_pages_remote(struct mm_struct *mm,
2279 unsigned long start, unsigned long nr_pages,
2280 unsigned int gup_flags, struct page **pages,
2283 int local_locked = 1;
2285 if (!is_valid_gup_args(pages, locked, &gup_flags,
2286 FOLL_TOUCH | FOLL_REMOTE))
2289 return __get_user_pages_locked(mm, start, nr_pages, pages,
2290 locked ? locked : &local_locked,
2293 EXPORT_SYMBOL(get_user_pages_remote);
2295 #else /* CONFIG_MMU */
2296 long get_user_pages_remote(struct mm_struct *mm,
2297 unsigned long start, unsigned long nr_pages,
2298 unsigned int gup_flags, struct page **pages,
2303 #endif /* !CONFIG_MMU */
2306 * get_user_pages() - pin user pages in memory
2307 * @start: starting user address
2308 * @nr_pages: number of pages from start to pin
2309 * @gup_flags: flags modifying lookup behaviour
2310 * @pages: array that receives pointers to the pages pinned.
2311 * Should be at least nr_pages long. Or NULL, if caller
2312 * only intends to ensure the pages are faulted in.
2314 * This is the same as get_user_pages_remote(), just with a less-flexible
2315 * calling convention where we assume that the mm being operated on belongs to
2316 * the current task, and doesn't allow passing of a locked parameter. We also
2317 * obviously don't pass FOLL_REMOTE in here.
2319 long get_user_pages(unsigned long start, unsigned long nr_pages,
2320 unsigned int gup_flags, struct page **pages)
2324 if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_TOUCH))
2327 return __get_user_pages_locked(current->mm, start, nr_pages, pages,
2328 &locked, gup_flags);
2330 EXPORT_SYMBOL(get_user_pages);
2333 * get_user_pages_unlocked() is suitable to replace the form:
2335 * mmap_read_lock(mm);
2336 * get_user_pages(mm, ..., pages, NULL);
2337 * mmap_read_unlock(mm);
2341 * get_user_pages_unlocked(mm, ..., pages);
2343 * It is functionally equivalent to get_user_pages_fast so
2344 * get_user_pages_fast should be used instead if specific gup_flags
2345 * (e.g. FOLL_FORCE) are not required.
2347 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
2348 struct page **pages, unsigned int gup_flags)
2352 if (!is_valid_gup_args(pages, NULL, &gup_flags,
2353 FOLL_TOUCH | FOLL_UNLOCKABLE))
2356 return __get_user_pages_locked(current->mm, start, nr_pages, pages,
2357 &locked, gup_flags);
2359 EXPORT_SYMBOL(get_user_pages_unlocked);
2364 * get_user_pages_fast attempts to pin user pages by walking the page
2365 * tables directly and avoids taking locks. Thus the walker needs to be
2366 * protected from page table pages being freed from under it, and should
2367 * block any THP splits.
2369 * One way to achieve this is to have the walker disable interrupts, and
2370 * rely on IPIs from the TLB flushing code blocking before the page table
2371 * pages are freed. This is unsuitable for architectures that do not need
2372 * to broadcast an IPI when invalidating TLBs.
2374 * Another way to achieve this is to batch up page table containing pages
2375 * belonging to more than one mm_user, then rcu_sched a callback to free those
2376 * pages. Disabling interrupts will allow the fast_gup walker to both block
2377 * the rcu_sched callback, and an IPI that we broadcast for splitting THPs
2378 * (which is a relatively rare event). The code below adopts this strategy.
2380 * Before activating this code, please be aware that the following assumptions
2381 * are currently made:
2383 * *) Either MMU_GATHER_RCU_TABLE_FREE is enabled, and tlb_remove_table() is used to
2384 * free pages containing page tables or TLB flushing requires IPI broadcast.
2386 * *) ptes can be read atomically by the architecture.
2388 * *) access_ok is sufficient to validate userspace address ranges.
2390 * The last two assumptions can be relaxed by the addition of helper functions.
2392 * This code is based heavily on the PowerPC implementation by Nick Piggin.
2394 #ifdef CONFIG_HAVE_FAST_GUP
2397 * Used in the GUP-fast path to determine whether a pin is permitted for a
2400 * This call assumes the caller has pinned the folio, that the lowest page table
2401 * level still points to this folio, and that interrupts have been disabled.
2403 * Writing to pinned file-backed dirty tracked folios is inherently problematic
2404 * (see comment describing the writable_file_mapping_allowed() function). We
2405 * therefore try to avoid the most egregious case of a long-term mapping doing
2408 * This function cannot be as thorough as that one as the VMA is not available
2409 * in the fast path, so instead we whitelist known good cases and if in doubt,
2410 * fall back to the slow path.
2412 static bool folio_fast_pin_allowed(struct folio *folio, unsigned int flags)
2414 struct address_space *mapping;
2415 unsigned long mapping_flags;
2418 * If we aren't pinning then no problematic write can occur. A long term
2419 * pin is the most egregious case so this is the one we disallow.
2421 if ((flags & (FOLL_PIN | FOLL_LONGTERM | FOLL_WRITE)) !=
2422 (FOLL_PIN | FOLL_LONGTERM | FOLL_WRITE))
2425 /* The folio is pinned, so we can safely access folio fields. */
2427 if (WARN_ON_ONCE(folio_test_slab(folio)))
2430 /* hugetlb mappings do not require dirty-tracking. */
2431 if (folio_test_hugetlb(folio))
2435 * GUP-fast disables IRQs. When IRQS are disabled, RCU grace periods
2436 * cannot proceed, which means no actions performed under RCU can
2439 * inodes and thus their mappings are freed under RCU, which means the
2440 * mapping cannot be freed beneath us and thus we can safely dereference
2443 lockdep_assert_irqs_disabled();
2446 * However, there may be operations which _alter_ the mapping, so ensure
2447 * we read it once and only once.
2449 mapping = READ_ONCE(folio->mapping);
2452 * The mapping may have been truncated, in any case we cannot determine
2453 * if this mapping is safe - fall back to slow path to determine how to
2459 /* Anonymous folios pose no problem. */
2460 mapping_flags = (unsigned long)mapping & PAGE_MAPPING_FLAGS;
2462 return mapping_flags & PAGE_MAPPING_ANON;
2465 * At this point, we know the mapping is non-null and points to an
2466 * address_space object. The only remaining whitelisted file system is
2469 return shmem_mapping(mapping);
2472 static void __maybe_unused undo_dev_pagemap(int *nr, int nr_start,
2474 struct page **pages)
2476 while ((*nr) - nr_start) {
2477 struct page *page = pages[--(*nr)];
2479 ClearPageReferenced(page);
2480 if (flags & FOLL_PIN)
2481 unpin_user_page(page);
2487 #ifdef CONFIG_ARCH_HAS_PTE_SPECIAL
2489 * Fast-gup relies on pte change detection to avoid concurrent pgtable
2492 * To pin the page, fast-gup needs to do below in order:
2493 * (1) pin the page (by prefetching pte), then (2) check pte not changed.
2495 * For the rest of pgtable operations where pgtable updates can be racy
2496 * with fast-gup, we need to do (1) clear pte, then (2) check whether page
2499 * Above will work for all pte-level operations, including THP split.
2501 * For THP collapse, it's a bit more complicated because fast-gup may be
2502 * walking a pgtable page that is being freed (pte is still valid but pmd
2503 * can be cleared already). To avoid race in such condition, we need to
2504 * also check pmd here to make sure pmd doesn't change (corresponds to
2505 * pmdp_collapse_flush() in the THP collapse code path).
2507 static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr,
2508 unsigned long end, unsigned int flags,
2509 struct page **pages, int *nr)
2511 struct dev_pagemap *pgmap = NULL;
2512 int nr_start = *nr, ret = 0;
2515 ptem = ptep = pte_offset_map(&pmd, addr);
2519 pte_t pte = ptep_get_lockless(ptep);
2521 struct folio *folio;
2523 if (pte_protnone(pte) && !gup_can_follow_protnone(flags))
2526 if (!pte_access_permitted(pte, flags & FOLL_WRITE))
2529 if (pte_devmap(pte)) {
2530 if (unlikely(flags & FOLL_LONGTERM))
2533 pgmap = get_dev_pagemap(pte_pfn(pte), pgmap);
2534 if (unlikely(!pgmap)) {
2535 undo_dev_pagemap(nr, nr_start, flags, pages);
2538 } else if (pte_special(pte))
2541 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
2542 page = pte_page(pte);
2544 folio = try_grab_folio(page, 1, flags);
2548 if (unlikely(page_is_secretmem(page))) {
2549 gup_put_folio(folio, 1, flags);
2553 if (unlikely(pmd_val(pmd) != pmd_val(*pmdp)) ||
2554 unlikely(pte_val(pte) != pte_val(ptep_get(ptep)))) {
2555 gup_put_folio(folio, 1, flags);
2559 if (!folio_fast_pin_allowed(folio, flags)) {
2560 gup_put_folio(folio, 1, flags);
2564 if (!pte_write(pte) && gup_must_unshare(NULL, flags, page)) {
2565 gup_put_folio(folio, 1, flags);
2570 * We need to make the page accessible if and only if we are
2571 * going to access its content (the FOLL_PIN case). Please
2572 * see Documentation/core-api/pin_user_pages.rst for
2575 if (flags & FOLL_PIN) {
2576 ret = arch_make_page_accessible(page);
2578 gup_put_folio(folio, 1, flags);
2582 folio_set_referenced(folio);
2585 } while (ptep++, addr += PAGE_SIZE, addr != end);
2591 put_dev_pagemap(pgmap);
2598 * If we can't determine whether or not a pte is special, then fail immediately
2599 * for ptes. Note, we can still pin HugeTLB and THP as these are guaranteed not
2602 * For a futex to be placed on a THP tail page, get_futex_key requires a
2603 * get_user_pages_fast_only implementation that can pin pages. Thus it's still
2604 * useful to have gup_huge_pmd even if we can't operate on ptes.
2606 static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr,
2607 unsigned long end, unsigned int flags,
2608 struct page **pages, int *nr)
2612 #endif /* CONFIG_ARCH_HAS_PTE_SPECIAL */
2614 #if defined(CONFIG_ARCH_HAS_PTE_DEVMAP) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
2615 static int __gup_device_huge(unsigned long pfn, unsigned long addr,
2616 unsigned long end, unsigned int flags,
2617 struct page **pages, int *nr)
2620 struct dev_pagemap *pgmap = NULL;
2623 struct page *page = pfn_to_page(pfn);
2625 pgmap = get_dev_pagemap(pfn, pgmap);
2626 if (unlikely(!pgmap)) {
2627 undo_dev_pagemap(nr, nr_start, flags, pages);
2631 if (!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page)) {
2632 undo_dev_pagemap(nr, nr_start, flags, pages);
2636 SetPageReferenced(page);
2638 if (unlikely(try_grab_page(page, flags))) {
2639 undo_dev_pagemap(nr, nr_start, flags, pages);
2644 } while (addr += PAGE_SIZE, addr != end);
2646 put_dev_pagemap(pgmap);
2650 static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
2651 unsigned long end, unsigned int flags,
2652 struct page **pages, int *nr)
2654 unsigned long fault_pfn;
2657 fault_pfn = pmd_pfn(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
2658 if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr))
2661 if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) {
2662 undo_dev_pagemap(nr, nr_start, flags, pages);
2668 static int __gup_device_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr,
2669 unsigned long end, unsigned int flags,
2670 struct page **pages, int *nr)
2672 unsigned long fault_pfn;
2675 fault_pfn = pud_pfn(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
2676 if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr))
2679 if (unlikely(pud_val(orig) != pud_val(*pudp))) {
2680 undo_dev_pagemap(nr, nr_start, flags, pages);
2686 static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
2687 unsigned long end, unsigned int flags,
2688 struct page **pages, int *nr)
2694 static int __gup_device_huge_pud(pud_t pud, pud_t *pudp, unsigned long addr,
2695 unsigned long end, unsigned int flags,
2696 struct page **pages, int *nr)
2703 static int record_subpages(struct page *page, unsigned long addr,
2704 unsigned long end, struct page **pages)
2708 for (nr = 0; addr != end; nr++, addr += PAGE_SIZE)
2709 pages[nr] = nth_page(page, nr);
2714 #ifdef CONFIG_ARCH_HAS_HUGEPD
2715 static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end,
2718 unsigned long __boundary = (addr + sz) & ~(sz-1);
2719 return (__boundary - 1 < end - 1) ? __boundary : end;
2722 static int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr,
2723 unsigned long end, unsigned int flags,
2724 struct page **pages, int *nr)
2726 unsigned long pte_end;
2728 struct folio *folio;
2732 pte_end = (addr + sz) & ~(sz-1);
2736 pte = huge_ptep_get(ptep);
2738 if (!pte_access_permitted(pte, flags & FOLL_WRITE))
2741 /* hugepages are never "special" */
2742 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
2744 page = nth_page(pte_page(pte), (addr & (sz - 1)) >> PAGE_SHIFT);
2745 refs = record_subpages(page, addr, end, pages + *nr);
2747 folio = try_grab_folio(page, refs, flags);
2751 if (unlikely(pte_val(pte) != pte_val(ptep_get(ptep)))) {
2752 gup_put_folio(folio, refs, flags);
2756 if (!folio_fast_pin_allowed(folio, flags)) {
2757 gup_put_folio(folio, refs, flags);
2761 if (!pte_write(pte) && gup_must_unshare(NULL, flags, &folio->page)) {
2762 gup_put_folio(folio, refs, flags);
2767 folio_set_referenced(folio);
2771 static int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
2772 unsigned int pdshift, unsigned long end, unsigned int flags,
2773 struct page **pages, int *nr)
2776 unsigned long sz = 1UL << hugepd_shift(hugepd);
2779 ptep = hugepte_offset(hugepd, addr, pdshift);
2781 next = hugepte_addr_end(addr, end, sz);
2782 if (!gup_hugepte(ptep, sz, addr, end, flags, pages, nr))
2784 } while (ptep++, addr = next, addr != end);
2789 static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
2790 unsigned int pdshift, unsigned long end, unsigned int flags,
2791 struct page **pages, int *nr)
2795 #endif /* CONFIG_ARCH_HAS_HUGEPD */
2797 static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
2798 unsigned long end, unsigned int flags,
2799 struct page **pages, int *nr)
2802 struct folio *folio;
2805 if (!pmd_access_permitted(orig, flags & FOLL_WRITE))
2808 if (pmd_devmap(orig)) {
2809 if (unlikely(flags & FOLL_LONGTERM))
2811 return __gup_device_huge_pmd(orig, pmdp, addr, end, flags,
2815 page = nth_page(pmd_page(orig), (addr & ~PMD_MASK) >> PAGE_SHIFT);
2816 refs = record_subpages(page, addr, end, pages + *nr);
2818 folio = try_grab_folio(page, refs, flags);
2822 if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) {
2823 gup_put_folio(folio, refs, flags);
2827 if (!folio_fast_pin_allowed(folio, flags)) {
2828 gup_put_folio(folio, refs, flags);
2831 if (!pmd_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) {
2832 gup_put_folio(folio, refs, flags);
2837 folio_set_referenced(folio);
2841 static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr,
2842 unsigned long end, unsigned int flags,
2843 struct page **pages, int *nr)
2846 struct folio *folio;
2849 if (!pud_access_permitted(orig, flags & FOLL_WRITE))
2852 if (pud_devmap(orig)) {
2853 if (unlikely(flags & FOLL_LONGTERM))
2855 return __gup_device_huge_pud(orig, pudp, addr, end, flags,
2859 page = nth_page(pud_page(orig), (addr & ~PUD_MASK) >> PAGE_SHIFT);
2860 refs = record_subpages(page, addr, end, pages + *nr);
2862 folio = try_grab_folio(page, refs, flags);
2866 if (unlikely(pud_val(orig) != pud_val(*pudp))) {
2867 gup_put_folio(folio, refs, flags);
2871 if (!folio_fast_pin_allowed(folio, flags)) {
2872 gup_put_folio(folio, refs, flags);
2876 if (!pud_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) {
2877 gup_put_folio(folio, refs, flags);
2882 folio_set_referenced(folio);
2886 static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr,
2887 unsigned long end, unsigned int flags,
2888 struct page **pages, int *nr)
2892 struct folio *folio;
2894 if (!pgd_access_permitted(orig, flags & FOLL_WRITE))
2897 BUILD_BUG_ON(pgd_devmap(orig));
2899 page = nth_page(pgd_page(orig), (addr & ~PGDIR_MASK) >> PAGE_SHIFT);
2900 refs = record_subpages(page, addr, end, pages + *nr);
2902 folio = try_grab_folio(page, refs, flags);
2906 if (unlikely(pgd_val(orig) != pgd_val(*pgdp))) {
2907 gup_put_folio(folio, refs, flags);
2911 if (!pgd_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) {
2912 gup_put_folio(folio, refs, flags);
2916 if (!folio_fast_pin_allowed(folio, flags)) {
2917 gup_put_folio(folio, refs, flags);
2922 folio_set_referenced(folio);
2926 static int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr, unsigned long end,
2927 unsigned int flags, struct page **pages, int *nr)
2932 pmdp = pmd_offset_lockless(pudp, pud, addr);
2934 pmd_t pmd = pmdp_get_lockless(pmdp);
2936 next = pmd_addr_end(addr, end);
2937 if (!pmd_present(pmd))
2940 if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) ||
2942 if (pmd_protnone(pmd) &&
2943 !gup_can_follow_protnone(flags))
2946 if (!gup_huge_pmd(pmd, pmdp, addr, next, flags,
2950 } else if (unlikely(is_hugepd(__hugepd(pmd_val(pmd))))) {
2952 * architecture have different format for hugetlbfs
2953 * pmd format and THP pmd format
2955 if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr,
2956 PMD_SHIFT, next, flags, pages, nr))
2958 } else if (!gup_pte_range(pmd, pmdp, addr, next, flags, pages, nr))
2960 } while (pmdp++, addr = next, addr != end);
2965 static int gup_pud_range(p4d_t *p4dp, p4d_t p4d, unsigned long addr, unsigned long end,
2966 unsigned int flags, struct page **pages, int *nr)
2971 pudp = pud_offset_lockless(p4dp, p4d, addr);
2973 pud_t pud = READ_ONCE(*pudp);
2975 next = pud_addr_end(addr, end);
2976 if (unlikely(!pud_present(pud)))
2978 if (unlikely(pud_huge(pud) || pud_devmap(pud))) {
2979 if (!gup_huge_pud(pud, pudp, addr, next, flags,
2982 } else if (unlikely(is_hugepd(__hugepd(pud_val(pud))))) {
2983 if (!gup_huge_pd(__hugepd(pud_val(pud)), addr,
2984 PUD_SHIFT, next, flags, pages, nr))
2986 } else if (!gup_pmd_range(pudp, pud, addr, next, flags, pages, nr))
2988 } while (pudp++, addr = next, addr != end);
2993 static int gup_p4d_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr, unsigned long end,
2994 unsigned int flags, struct page **pages, int *nr)
2999 p4dp = p4d_offset_lockless(pgdp, pgd, addr);
3001 p4d_t p4d = READ_ONCE(*p4dp);
3003 next = p4d_addr_end(addr, end);
3006 BUILD_BUG_ON(p4d_huge(p4d));
3007 if (unlikely(is_hugepd(__hugepd(p4d_val(p4d))))) {
3008 if (!gup_huge_pd(__hugepd(p4d_val(p4d)), addr,
3009 P4D_SHIFT, next, flags, pages, nr))
3011 } else if (!gup_pud_range(p4dp, p4d, addr, next, flags, pages, nr))
3013 } while (p4dp++, addr = next, addr != end);
3018 static void gup_pgd_range(unsigned long addr, unsigned long end,
3019 unsigned int flags, struct page **pages, int *nr)
3024 pgdp = pgd_offset(current->mm, addr);
3026 pgd_t pgd = READ_ONCE(*pgdp);
3028 next = pgd_addr_end(addr, end);
3031 if (unlikely(pgd_huge(pgd))) {
3032 if (!gup_huge_pgd(pgd, pgdp, addr, next, flags,
3035 } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) {
3036 if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr,
3037 PGDIR_SHIFT, next, flags, pages, nr))
3039 } else if (!gup_p4d_range(pgdp, pgd, addr, next, flags, pages, nr))
3041 } while (pgdp++, addr = next, addr != end);
3044 static inline void gup_pgd_range(unsigned long addr, unsigned long end,
3045 unsigned int flags, struct page **pages, int *nr)
3048 #endif /* CONFIG_HAVE_FAST_GUP */
3050 #ifndef gup_fast_permitted
3052 * Check if it's allowed to use get_user_pages_fast_only() for the range, or
3053 * we need to fall back to the slow version:
3055 static bool gup_fast_permitted(unsigned long start, unsigned long end)
3061 static unsigned long lockless_pages_from_mm(unsigned long start,
3063 unsigned int gup_flags,
3064 struct page **pages)
3066 unsigned long flags;
3070 if (!IS_ENABLED(CONFIG_HAVE_FAST_GUP) ||
3071 !gup_fast_permitted(start, end))
3074 if (gup_flags & FOLL_PIN) {
3075 seq = raw_read_seqcount(¤t->mm->write_protect_seq);
3081 * Disable interrupts. The nested form is used, in order to allow full,
3082 * general purpose use of this routine.
3084 * With interrupts disabled, we block page table pages from being freed
3085 * from under us. See struct mmu_table_batch comments in
3086 * include/asm-generic/tlb.h for more details.
3088 * We do not adopt an rcu_read_lock() here as we also want to block IPIs
3089 * that come from THPs splitting.
3091 local_irq_save(flags);
3092 gup_pgd_range(start, end, gup_flags, pages, &nr_pinned);
3093 local_irq_restore(flags);
3096 * When pinning pages for DMA there could be a concurrent write protect
3097 * from fork() via copy_page_range(), in this case always fail fast GUP.
3099 if (gup_flags & FOLL_PIN) {
3100 if (read_seqcount_retry(¤t->mm->write_protect_seq, seq)) {
3101 unpin_user_pages_lockless(pages, nr_pinned);
3104 sanity_check_pinned_pages(pages, nr_pinned);
3110 static int internal_get_user_pages_fast(unsigned long start,
3111 unsigned long nr_pages,
3112 unsigned int gup_flags,
3113 struct page **pages)
3115 unsigned long len, end;
3116 unsigned long nr_pinned;
3120 if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM |
3121 FOLL_FORCE | FOLL_PIN | FOLL_GET |
3122 FOLL_FAST_ONLY | FOLL_NOFAULT |
3126 if (gup_flags & FOLL_PIN)
3127 mm_set_has_pinned_flag(¤t->mm->flags);
3129 if (!(gup_flags & FOLL_FAST_ONLY))
3130 might_lock_read(¤t->mm->mmap_lock);
3132 start = untagged_addr(start) & PAGE_MASK;
3133 len = nr_pages << PAGE_SHIFT;
3134 if (check_add_overflow(start, len, &end))
3136 if (end > TASK_SIZE_MAX)
3138 if (unlikely(!access_ok((void __user *)start, len)))
3141 nr_pinned = lockless_pages_from_mm(start, end, gup_flags, pages);
3142 if (nr_pinned == nr_pages || gup_flags & FOLL_FAST_ONLY)
3145 /* Slow path: try to get the remaining pages with get_user_pages */
3146 start += nr_pinned << PAGE_SHIFT;
3148 ret = __gup_longterm_locked(current->mm, start, nr_pages - nr_pinned,
3150 gup_flags | FOLL_TOUCH | FOLL_UNLOCKABLE);
3153 * The caller has to unpin the pages we already pinned so
3154 * returning -errno is not an option
3160 return ret + nr_pinned;
3164 * get_user_pages_fast_only() - pin user pages in memory
3165 * @start: starting user address
3166 * @nr_pages: number of pages from start to pin
3167 * @gup_flags: flags modifying pin behaviour
3168 * @pages: array that receives pointers to the pages pinned.
3169 * Should be at least nr_pages long.
3171 * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to
3174 * If the architecture does not support this function, simply return with no
3177 * Careful, careful! COW breaking can go either way, so a non-write
3178 * access can get ambiguous page results. If you call this function without
3179 * 'write' set, you'd better be sure that you're ok with that ambiguity.
3181 int get_user_pages_fast_only(unsigned long start, int nr_pages,
3182 unsigned int gup_flags, struct page **pages)
3185 * Internally (within mm/gup.c), gup fast variants must set FOLL_GET,
3186 * because gup fast is always a "pin with a +1 page refcount" request.
3188 * FOLL_FAST_ONLY is required in order to match the API description of
3189 * this routine: no fall back to regular ("slow") GUP.
3191 if (!is_valid_gup_args(pages, NULL, &gup_flags,
3192 FOLL_GET | FOLL_FAST_ONLY))
3195 return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages);
3197 EXPORT_SYMBOL_GPL(get_user_pages_fast_only);
3200 * get_user_pages_fast() - pin user pages in memory
3201 * @start: starting user address
3202 * @nr_pages: number of pages from start to pin
3203 * @gup_flags: flags modifying pin behaviour
3204 * @pages: array that receives pointers to the pages pinned.
3205 * Should be at least nr_pages long.
3207 * Attempt to pin user pages in memory without taking mm->mmap_lock.
3208 * If not successful, it will fall back to taking the lock and
3209 * calling get_user_pages().
3211 * Returns number of pages pinned. This may be fewer than the number requested.
3212 * If nr_pages is 0 or negative, returns 0. If no pages were pinned, returns
3215 int get_user_pages_fast(unsigned long start, int nr_pages,
3216 unsigned int gup_flags, struct page **pages)
3219 * The caller may or may not have explicitly set FOLL_GET; either way is
3220 * OK. However, internally (within mm/gup.c), gup fast variants must set
3221 * FOLL_GET, because gup fast is always a "pin with a +1 page refcount"
3224 if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_GET))
3226 return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages);
3228 EXPORT_SYMBOL_GPL(get_user_pages_fast);
3231 * pin_user_pages_fast() - pin user pages in memory without taking locks
3233 * @start: starting user address
3234 * @nr_pages: number of pages from start to pin
3235 * @gup_flags: flags modifying pin behaviour
3236 * @pages: array that receives pointers to the pages pinned.
3237 * Should be at least nr_pages long.
3239 * Nearly the same as get_user_pages_fast(), except that FOLL_PIN is set. See
3240 * get_user_pages_fast() for documentation on the function arguments, because
3241 * the arguments here are identical.
3243 * FOLL_PIN means that the pages must be released via unpin_user_page(). Please
3244 * see Documentation/core-api/pin_user_pages.rst for further details.
3246 * Note that if a zero_page is amongst the returned pages, it will not have
3247 * pins in it and unpin_user_page() will not remove pins from it.
3249 int pin_user_pages_fast(unsigned long start, int nr_pages,
3250 unsigned int gup_flags, struct page **pages)
3252 if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_PIN))
3254 return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages);
3256 EXPORT_SYMBOL_GPL(pin_user_pages_fast);
3259 * pin_user_pages_remote() - pin pages of a remote process
3261 * @mm: mm_struct of target mm
3262 * @start: starting user address
3263 * @nr_pages: number of pages from start to pin
3264 * @gup_flags: flags modifying lookup behaviour
3265 * @pages: array that receives pointers to the pages pinned.
3266 * Should be at least nr_pages long.
3267 * @locked: pointer to lock flag indicating whether lock is held and
3268 * subsequently whether VM_FAULT_RETRY functionality can be
3269 * utilised. Lock must initially be held.
3271 * Nearly the same as get_user_pages_remote(), except that FOLL_PIN is set. See
3272 * get_user_pages_remote() for documentation on the function arguments, because
3273 * the arguments here are identical.
3275 * FOLL_PIN means that the pages must be released via unpin_user_page(). Please
3276 * see Documentation/core-api/pin_user_pages.rst for details.
3278 * Note that if a zero_page is amongst the returned pages, it will not have
3279 * pins in it and unpin_user_page*() will not remove pins from it.
3281 long pin_user_pages_remote(struct mm_struct *mm,
3282 unsigned long start, unsigned long nr_pages,
3283 unsigned int gup_flags, struct page **pages,
3286 int local_locked = 1;
3288 if (!is_valid_gup_args(pages, locked, &gup_flags,
3289 FOLL_PIN | FOLL_TOUCH | FOLL_REMOTE))
3291 return __gup_longterm_locked(mm, start, nr_pages, pages,
3292 locked ? locked : &local_locked,
3295 EXPORT_SYMBOL(pin_user_pages_remote);
3298 * pin_user_pages() - pin user pages in memory for use by other devices
3300 * @start: starting user address
3301 * @nr_pages: number of pages from start to pin
3302 * @gup_flags: flags modifying lookup behaviour
3303 * @pages: array that receives pointers to the pages pinned.
3304 * Should be at least nr_pages long.
3306 * Nearly the same as get_user_pages(), except that FOLL_TOUCH is not set, and
3309 * FOLL_PIN means that the pages must be released via unpin_user_page(). Please
3310 * see Documentation/core-api/pin_user_pages.rst for details.
3312 * Note that if a zero_page is amongst the returned pages, it will not have
3313 * pins in it and unpin_user_page*() will not remove pins from it.
3315 long pin_user_pages(unsigned long start, unsigned long nr_pages,
3316 unsigned int gup_flags, struct page **pages)
3320 if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_PIN))
3322 return __gup_longterm_locked(current->mm, start, nr_pages,
3323 pages, &locked, gup_flags);
3325 EXPORT_SYMBOL(pin_user_pages);
3328 * pin_user_pages_unlocked() is the FOLL_PIN variant of
3329 * get_user_pages_unlocked(). Behavior is the same, except that this one sets
3330 * FOLL_PIN and rejects FOLL_GET.
3332 * Note that if a zero_page is amongst the returned pages, it will not have
3333 * pins in it and unpin_user_page*() will not remove pins from it.
3335 long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
3336 struct page **pages, unsigned int gup_flags)
3340 if (!is_valid_gup_args(pages, NULL, &gup_flags,
3341 FOLL_PIN | FOLL_TOUCH | FOLL_UNLOCKABLE))
3344 return __gup_longterm_locked(current->mm, start, nr_pages, pages,
3345 &locked, gup_flags);
3347 EXPORT_SYMBOL(pin_user_pages_unlocked);