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)
814 struct mm_struct *mm = vma->vm_mm;
819 * Call hugetlb_follow_page_mask for hugetlb vmas as it will use
820 * special hugetlb page table walking code. This eliminates the
821 * need to check for hugetlb entries in the general walking code.
823 if (is_vm_hugetlb_page(vma))
824 return hugetlb_follow_page_mask(vma, address, flags,
827 pgd = pgd_offset(mm, address);
829 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
830 return no_page_table(vma, flags);
832 return follow_p4d_mask(vma, address, pgd, flags, ctx);
835 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
836 unsigned int foll_flags)
838 struct follow_page_context ctx = { NULL };
841 if (vma_is_secretmem(vma))
844 if (WARN_ON_ONCE(foll_flags & FOLL_PIN))
847 page = follow_page_mask(vma, address, foll_flags, &ctx);
849 put_dev_pagemap(ctx.pgmap);
853 static int get_gate_page(struct mm_struct *mm, unsigned long address,
854 unsigned int gup_flags, struct vm_area_struct **vma,
865 /* user gate pages are read-only */
866 if (gup_flags & FOLL_WRITE)
868 if (address > TASK_SIZE)
869 pgd = pgd_offset_k(address);
871 pgd = pgd_offset_gate(mm, address);
874 p4d = p4d_offset(pgd, address);
877 pud = pud_offset(p4d, address);
880 pmd = pmd_offset(pud, address);
881 if (!pmd_present(*pmd))
883 pte = pte_offset_map(pmd, address);
886 entry = ptep_get(pte);
889 *vma = get_gate_vma(mm);
892 *page = vm_normal_page(*vma, address, entry);
894 if ((gup_flags & FOLL_DUMP) || !is_zero_pfn(pte_pfn(entry)))
896 *page = pte_page(entry);
898 ret = try_grab_page(*page, gup_flags);
909 * mmap_lock must be held on entry. If @flags has FOLL_UNLOCKABLE but not
910 * FOLL_NOWAIT, the mmap_lock may be released. If it is, *@locked will be set
911 * to 0 and -EBUSY returned.
913 static int faultin_page(struct vm_area_struct *vma,
914 unsigned long address, unsigned int *flags, bool unshare,
917 unsigned int fault_flags = 0;
920 if (*flags & FOLL_NOFAULT)
922 if (*flags & FOLL_WRITE)
923 fault_flags |= FAULT_FLAG_WRITE;
924 if (*flags & FOLL_REMOTE)
925 fault_flags |= FAULT_FLAG_REMOTE;
926 if (*flags & FOLL_UNLOCKABLE) {
927 fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
929 * FAULT_FLAG_INTERRUPTIBLE is opt-in. GUP callers must set
930 * FOLL_INTERRUPTIBLE to enable FAULT_FLAG_INTERRUPTIBLE.
931 * That's because some callers may not be prepared to
932 * handle early exits caused by non-fatal signals.
934 if (*flags & FOLL_INTERRUPTIBLE)
935 fault_flags |= FAULT_FLAG_INTERRUPTIBLE;
937 if (*flags & FOLL_NOWAIT)
938 fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT;
939 if (*flags & FOLL_TRIED) {
941 * Note: FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_TRIED
944 fault_flags |= FAULT_FLAG_TRIED;
947 fault_flags |= FAULT_FLAG_UNSHARE;
948 /* FAULT_FLAG_WRITE and FAULT_FLAG_UNSHARE are incompatible */
949 VM_BUG_ON(fault_flags & FAULT_FLAG_WRITE);
952 ret = handle_mm_fault(vma, address, fault_flags, NULL);
954 if (ret & VM_FAULT_COMPLETED) {
956 * With FAULT_FLAG_RETRY_NOWAIT we'll never release the
957 * mmap lock in the page fault handler. Sanity check this.
959 WARN_ON_ONCE(fault_flags & FAULT_FLAG_RETRY_NOWAIT);
963 * We should do the same as VM_FAULT_RETRY, but let's not
964 * return -EBUSY since that's not reflecting the reality of
965 * what has happened - we've just fully completed a page
966 * fault, with the mmap lock released. Use -EAGAIN to show
967 * that we want to take the mmap lock _again_.
972 if (ret & VM_FAULT_ERROR) {
973 int err = vm_fault_to_errno(ret, *flags);
980 if (ret & VM_FAULT_RETRY) {
981 if (!(fault_flags & FAULT_FLAG_RETRY_NOWAIT))
990 * Writing to file-backed mappings which require folio dirty tracking using GUP
991 * is a fundamentally broken operation, as kernel write access to GUP mappings
992 * do not adhere to the semantics expected by a file system.
994 * Consider the following scenario:-
996 * 1. A folio is written to via GUP which write-faults the memory, notifying
997 * the file system and dirtying the folio.
998 * 2. Later, writeback is triggered, resulting in the folio being cleaned and
999 * the PTE being marked read-only.
1000 * 3. The GUP caller writes to the folio, as it is mapped read/write via the
1002 * 4. The GUP caller, now done with the page, unpins it and sets it dirty
1003 * (though it does not have to).
1005 * This results in both data being written to a folio without writenotify, and
1006 * the folio being dirtied unexpectedly (if the caller decides to do so).
1008 static bool writable_file_mapping_allowed(struct vm_area_struct *vma,
1009 unsigned long gup_flags)
1012 * If we aren't pinning then no problematic write can occur. A long term
1013 * pin is the most egregious case so this is the case we disallow.
1015 if ((gup_flags & (FOLL_PIN | FOLL_LONGTERM)) !=
1016 (FOLL_PIN | FOLL_LONGTERM))
1020 * If the VMA does not require dirty tracking then no problematic write
1023 return !vma_needs_dirty_tracking(vma);
1026 static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags)
1028 vm_flags_t vm_flags = vma->vm_flags;
1029 int write = (gup_flags & FOLL_WRITE);
1030 int foreign = (gup_flags & FOLL_REMOTE);
1031 bool vma_anon = vma_is_anonymous(vma);
1033 if (vm_flags & (VM_IO | VM_PFNMAP))
1036 if ((gup_flags & FOLL_ANON) && !vma_anon)
1039 if ((gup_flags & FOLL_LONGTERM) && vma_is_fsdax(vma))
1042 if (vma_is_secretmem(vma))
1047 !writable_file_mapping_allowed(vma, gup_flags))
1050 if (!(vm_flags & VM_WRITE)) {
1051 if (!(gup_flags & FOLL_FORCE))
1053 /* hugetlb does not support FOLL_FORCE|FOLL_WRITE. */
1054 if (is_vm_hugetlb_page(vma))
1057 * We used to let the write,force case do COW in a
1058 * VM_MAYWRITE VM_SHARED !VM_WRITE vma, so ptrace could
1059 * set a breakpoint in a read-only mapping of an
1060 * executable, without corrupting the file (yet only
1061 * when that file had been opened for writing!).
1062 * Anon pages in shared mappings are surprising: now
1065 if (!is_cow_mapping(vm_flags))
1068 } else if (!(vm_flags & VM_READ)) {
1069 if (!(gup_flags & FOLL_FORCE))
1072 * Is there actually any vma we can reach here which does not
1073 * have VM_MAYREAD set?
1075 if (!(vm_flags & VM_MAYREAD))
1079 * gups are always data accesses, not instruction
1080 * fetches, so execute=false here
1082 if (!arch_vma_access_permitted(vma, write, false, foreign))
1088 * This is "vma_lookup()", but with a warning if we would have
1089 * historically expanded the stack in the GUP code.
1091 static struct vm_area_struct *gup_vma_lookup(struct mm_struct *mm,
1094 #ifdef CONFIG_STACK_GROWSUP
1095 return vma_lookup(mm, addr);
1097 static volatile unsigned long next_warn;
1098 struct vm_area_struct *vma;
1099 unsigned long now, next;
1101 vma = find_vma(mm, addr);
1102 if (!vma || (addr >= vma->vm_start))
1105 /* Only warn for half-way relevant accesses */
1106 if (!(vma->vm_flags & VM_GROWSDOWN))
1108 if (vma->vm_start - addr > 65536)
1111 /* Let's not warn more than once an hour.. */
1112 now = jiffies; next = next_warn;
1113 if (next && time_before(now, next))
1115 next_warn = now + 60*60*HZ;
1117 /* Let people know things may have changed. */
1118 pr_warn("GUP no longer grows the stack in %s (%d): %lx-%lx (%lx)\n",
1119 current->comm, task_pid_nr(current),
1120 vma->vm_start, vma->vm_end, addr);
1127 * __get_user_pages() - pin user pages in memory
1128 * @mm: mm_struct of target mm
1129 * @start: starting user address
1130 * @nr_pages: number of pages from start to pin
1131 * @gup_flags: flags modifying pin behaviour
1132 * @pages: array that receives pointers to the pages pinned.
1133 * Should be at least nr_pages long. Or NULL, if caller
1134 * only intends to ensure the pages are faulted in.
1135 * @locked: whether we're still with the mmap_lock held
1137 * Returns either number of pages pinned (which may be less than the
1138 * number requested), or an error. Details about the return value:
1140 * -- If nr_pages is 0, returns 0.
1141 * -- If nr_pages is >0, but no pages were pinned, returns -errno.
1142 * -- If nr_pages is >0, and some pages were pinned, returns the number of
1143 * pages pinned. Again, this may be less than nr_pages.
1144 * -- 0 return value is possible when the fault would need to be retried.
1146 * The caller is responsible for releasing returned @pages, via put_page().
1148 * Must be called with mmap_lock held. It may be released. See below.
1150 * __get_user_pages walks a process's page tables and takes a reference to
1151 * each struct page that each user address corresponds to at a given
1152 * instant. That is, it takes the page that would be accessed if a user
1153 * thread accesses the given user virtual address at that instant.
1155 * This does not guarantee that the page exists in the user mappings when
1156 * __get_user_pages returns, and there may even be a completely different
1157 * page there in some cases (eg. if mmapped pagecache has been invalidated
1158 * and subsequently re-faulted). However it does guarantee that the page
1159 * won't be freed completely. And mostly callers simply care that the page
1160 * contains data that was valid *at some point in time*. Typically, an IO
1161 * or similar operation cannot guarantee anything stronger anyway because
1162 * locks can't be held over the syscall boundary.
1164 * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If
1165 * the page is written to, set_page_dirty (or set_page_dirty_lock, as
1166 * appropriate) must be called after the page is finished with, and
1167 * before put_page is called.
1169 * If FOLL_UNLOCKABLE is set without FOLL_NOWAIT then the mmap_lock may
1170 * be released. If this happens *@locked will be set to 0 on return.
1172 * A caller using such a combination of @gup_flags must therefore hold the
1173 * mmap_lock for reading only, and recognize when it's been released. Otherwise,
1174 * it must be held for either reading or writing and will not be released.
1176 * In most cases, get_user_pages or get_user_pages_fast should be used
1177 * instead of __get_user_pages. __get_user_pages should be used only if
1178 * you need some special @gup_flags.
1180 static long __get_user_pages(struct mm_struct *mm,
1181 unsigned long start, unsigned long nr_pages,
1182 unsigned int gup_flags, struct page **pages,
1185 long ret = 0, i = 0;
1186 struct vm_area_struct *vma = NULL;
1187 struct follow_page_context ctx = { NULL };
1192 start = untagged_addr_remote(mm, start);
1194 VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN)));
1198 unsigned int foll_flags = gup_flags;
1199 unsigned int page_increm;
1201 /* first iteration or cross vma bound */
1202 if (!vma || start >= vma->vm_end) {
1203 vma = gup_vma_lookup(mm, start);
1204 if (!vma && in_gate_area(mm, start)) {
1205 ret = get_gate_page(mm, start & PAGE_MASK,
1207 pages ? &page : NULL);
1218 ret = check_vma_flags(vma, gup_flags);
1224 * If we have a pending SIGKILL, don't keep faulting pages and
1225 * potentially allocating memory.
1227 if (fatal_signal_pending(current)) {
1233 page = follow_page_mask(vma, start, foll_flags, &ctx);
1234 if (!page || PTR_ERR(page) == -EMLINK) {
1235 ret = faultin_page(vma, start, &foll_flags,
1236 PTR_ERR(page) == -EMLINK, locked);
1250 } else if (PTR_ERR(page) == -EEXIST) {
1252 * Proper page table entry exists, but no corresponding
1253 * struct page. If the caller expects **pages to be
1254 * filled in, bail out now, because that can't be done
1258 ret = PTR_ERR(page);
1261 } else if (IS_ERR(page)) {
1262 ret = PTR_ERR(page);
1266 page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask);
1267 if (page_increm > nr_pages)
1268 page_increm = nr_pages;
1271 struct page *subpage;
1275 * This must be a large folio (and doesn't need to
1276 * be the whole folio; it can be part of it), do
1277 * the refcount work for all the subpages too.
1279 * NOTE: here the page may not be the head page
1280 * e.g. when start addr is not thp-size aligned.
1281 * try_grab_folio() should have taken care of tail
1284 if (page_increm > 1) {
1285 struct folio *folio;
1288 * Since we already hold refcount on the
1289 * large folio, this should never fail.
1291 folio = try_grab_folio(page, page_increm - 1,
1293 if (WARN_ON_ONCE(!folio)) {
1295 * Release the 1st page ref if the
1296 * folio is problematic, fail hard.
1298 gup_put_folio(page_folio(page), 1,
1305 for (j = 0; j < page_increm; j++) {
1306 subpage = nth_page(page, j);
1307 pages[i + j] = subpage;
1308 flush_anon_page(vma, subpage, start + j * PAGE_SIZE);
1309 flush_dcache_page(subpage);
1314 start += page_increm * PAGE_SIZE;
1315 nr_pages -= page_increm;
1319 put_dev_pagemap(ctx.pgmap);
1323 static bool vma_permits_fault(struct vm_area_struct *vma,
1324 unsigned int fault_flags)
1326 bool write = !!(fault_flags & FAULT_FLAG_WRITE);
1327 bool foreign = !!(fault_flags & FAULT_FLAG_REMOTE);
1328 vm_flags_t vm_flags = write ? VM_WRITE : VM_READ;
1330 if (!(vm_flags & vma->vm_flags))
1334 * The architecture might have a hardware protection
1335 * mechanism other than read/write that can deny access.
1337 * gup always represents data access, not instruction
1338 * fetches, so execute=false here:
1340 if (!arch_vma_access_permitted(vma, write, false, foreign))
1347 * fixup_user_fault() - manually resolve a user page fault
1348 * @mm: mm_struct of target mm
1349 * @address: user address
1350 * @fault_flags:flags to pass down to handle_mm_fault()
1351 * @unlocked: did we unlock the mmap_lock while retrying, maybe NULL if caller
1352 * does not allow retry. If NULL, the caller must guarantee
1353 * that fault_flags does not contain FAULT_FLAG_ALLOW_RETRY.
1355 * This is meant to be called in the specific scenario where for locking reasons
1356 * we try to access user memory in atomic context (within a pagefault_disable()
1357 * section), this returns -EFAULT, and we want to resolve the user fault before
1360 * Typically this is meant to be used by the futex code.
1362 * The main difference with get_user_pages() is that this function will
1363 * unconditionally call handle_mm_fault() which will in turn perform all the
1364 * necessary SW fixup of the dirty and young bits in the PTE, while
1365 * get_user_pages() only guarantees to update these in the struct page.
1367 * This is important for some architectures where those bits also gate the
1368 * access permission to the page because they are maintained in software. On
1369 * such architectures, gup() will not be enough to make a subsequent access
1372 * This function will not return with an unlocked mmap_lock. So it has not the
1373 * same semantics wrt the @mm->mmap_lock as does filemap_fault().
1375 int fixup_user_fault(struct mm_struct *mm,
1376 unsigned long address, unsigned int fault_flags,
1379 struct vm_area_struct *vma;
1382 address = untagged_addr_remote(mm, address);
1385 fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
1388 vma = gup_vma_lookup(mm, address);
1392 if (!vma_permits_fault(vma, fault_flags))
1395 if ((fault_flags & FAULT_FLAG_KILLABLE) &&
1396 fatal_signal_pending(current))
1399 ret = handle_mm_fault(vma, address, fault_flags, NULL);
1401 if (ret & VM_FAULT_COMPLETED) {
1403 * NOTE: it's a pity that we need to retake the lock here
1404 * to pair with the unlock() in the callers. Ideally we
1405 * could tell the callers so they do not need to unlock.
1412 if (ret & VM_FAULT_ERROR) {
1413 int err = vm_fault_to_errno(ret, 0);
1420 if (ret & VM_FAULT_RETRY) {
1423 fault_flags |= FAULT_FLAG_TRIED;
1429 EXPORT_SYMBOL_GPL(fixup_user_fault);
1432 * GUP always responds to fatal signals. When FOLL_INTERRUPTIBLE is
1433 * specified, it'll also respond to generic signals. The caller of GUP
1434 * that has FOLL_INTERRUPTIBLE should take care of the GUP interruption.
1436 static bool gup_signal_pending(unsigned int flags)
1438 if (fatal_signal_pending(current))
1441 if (!(flags & FOLL_INTERRUPTIBLE))
1444 return signal_pending(current);
1448 * Locking: (*locked == 1) means that the mmap_lock has already been acquired by
1449 * the caller. This function may drop the mmap_lock. If it does so, then it will
1450 * set (*locked = 0).
1452 * (*locked == 0) means that the caller expects this function to acquire and
1453 * drop the mmap_lock. Therefore, the value of *locked will still be zero when
1454 * the function returns, even though it may have changed temporarily during
1455 * function execution.
1457 * Please note that this function, unlike __get_user_pages(), will not return 0
1458 * for nr_pages > 0, unless FOLL_NOWAIT is used.
1460 static __always_inline long __get_user_pages_locked(struct mm_struct *mm,
1461 unsigned long start,
1462 unsigned long nr_pages,
1463 struct page **pages,
1467 long ret, pages_done;
1468 bool must_unlock = false;
1471 * The internal caller expects GUP to manage the lock internally and the
1472 * lock must be released when this returns.
1475 if (mmap_read_lock_killable(mm))
1481 mmap_assert_locked(mm);
1483 if (flags & FOLL_PIN)
1484 mm_set_has_pinned_flag(&mm->flags);
1487 * FOLL_PIN and FOLL_GET are mutually exclusive. Traditional behavior
1488 * is to set FOLL_GET if the caller wants pages[] filled in (but has
1489 * carelessly failed to specify FOLL_GET), so keep doing that, but only
1490 * for FOLL_GET, not for the newer FOLL_PIN.
1492 * FOLL_PIN always expects pages to be non-null, but no need to assert
1493 * that here, as any failures will be obvious enough.
1495 if (pages && !(flags & FOLL_PIN))
1500 ret = __get_user_pages(mm, start, nr_pages, flags, pages,
1502 if (!(flags & FOLL_UNLOCKABLE)) {
1503 /* VM_FAULT_RETRY couldn't trigger, bypass */
1508 /* VM_FAULT_RETRY or VM_FAULT_COMPLETED cannot return errors */
1511 BUG_ON(ret >= nr_pages);
1522 * VM_FAULT_RETRY didn't trigger or it was a
1530 * VM_FAULT_RETRY triggered, so seek to the faulting offset.
1531 * For the prefault case (!pages) we only update counts.
1535 start += ret << PAGE_SHIFT;
1537 /* The lock was temporarily dropped, so we must unlock later */
1542 * Repeat on the address that fired VM_FAULT_RETRY
1543 * with both FAULT_FLAG_ALLOW_RETRY and
1544 * FAULT_FLAG_TRIED. Note that GUP can be interrupted
1545 * by fatal signals of even common signals, depending on
1546 * the caller's request. So we need to check it before we
1547 * start trying again otherwise it can loop forever.
1549 if (gup_signal_pending(flags)) {
1551 pages_done = -EINTR;
1555 ret = mmap_read_lock_killable(mm);
1564 ret = __get_user_pages(mm, start, 1, flags | FOLL_TRIED,
1567 /* Continue to retry until we succeeded */
1585 if (must_unlock && *locked) {
1587 * We either temporarily dropped the lock, or the caller
1588 * requested that we both acquire and drop the lock. Either way,
1589 * we must now unlock, and notify the caller of that state.
1591 mmap_read_unlock(mm);
1598 * populate_vma_page_range() - populate a range of pages in the vma.
1600 * @start: start address
1602 * @locked: whether the mmap_lock is still held
1604 * This takes care of mlocking the pages too if VM_LOCKED is set.
1606 * Return either number of pages pinned in the vma, or a negative error
1609 * vma->vm_mm->mmap_lock must be held.
1611 * If @locked is NULL, it may be held for read or write and will
1614 * If @locked is non-NULL, it must held for read only and may be
1615 * released. If it's released, *@locked will be set to 0.
1617 long populate_vma_page_range(struct vm_area_struct *vma,
1618 unsigned long start, unsigned long end, int *locked)
1620 struct mm_struct *mm = vma->vm_mm;
1621 unsigned long nr_pages = (end - start) / PAGE_SIZE;
1622 int local_locked = 1;
1626 VM_BUG_ON(!PAGE_ALIGNED(start));
1627 VM_BUG_ON(!PAGE_ALIGNED(end));
1628 VM_BUG_ON_VMA(start < vma->vm_start, vma);
1629 VM_BUG_ON_VMA(end > vma->vm_end, vma);
1630 mmap_assert_locked(mm);
1633 * Rightly or wrongly, the VM_LOCKONFAULT case has never used
1634 * faultin_page() to break COW, so it has no work to do here.
1636 if (vma->vm_flags & VM_LOCKONFAULT)
1639 gup_flags = FOLL_TOUCH;
1641 * We want to touch writable mappings with a write fault in order
1642 * to break COW, except for shared mappings because these don't COW
1643 * and we would not want to dirty them for nothing.
1645 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
1646 gup_flags |= FOLL_WRITE;
1649 * We want mlock to succeed for regions that have any permissions
1650 * other than PROT_NONE.
1652 if (vma_is_accessible(vma))
1653 gup_flags |= FOLL_FORCE;
1656 gup_flags |= FOLL_UNLOCKABLE;
1659 * We made sure addr is within a VMA, so the following will
1660 * not result in a stack expansion that recurses back here.
1662 ret = __get_user_pages(mm, start, nr_pages, gup_flags,
1663 NULL, locked ? locked : &local_locked);
1669 * faultin_vma_page_range() - populate (prefault) page tables inside the
1670 * given VMA range readable/writable
1672 * This takes care of mlocking the pages, too, if VM_LOCKED is set.
1675 * @start: start address
1677 * @write: whether to prefault readable or writable
1678 * @locked: whether the mmap_lock is still held
1680 * Returns either number of processed pages in the vma, or a negative error
1681 * code on error (see __get_user_pages()).
1683 * vma->vm_mm->mmap_lock must be held. The range must be page-aligned and
1684 * covered by the VMA. If it's released, *@locked will be set to 0.
1686 long faultin_vma_page_range(struct vm_area_struct *vma, unsigned long start,
1687 unsigned long end, bool write, int *locked)
1689 struct mm_struct *mm = vma->vm_mm;
1690 unsigned long nr_pages = (end - start) / PAGE_SIZE;
1694 VM_BUG_ON(!PAGE_ALIGNED(start));
1695 VM_BUG_ON(!PAGE_ALIGNED(end));
1696 VM_BUG_ON_VMA(start < vma->vm_start, vma);
1697 VM_BUG_ON_VMA(end > vma->vm_end, vma);
1698 mmap_assert_locked(mm);
1701 * FOLL_TOUCH: Mark page accessed and thereby young; will also mark
1702 * the page dirty with FOLL_WRITE -- which doesn't make a
1703 * difference with !FOLL_FORCE, because the page is writable
1704 * in the page table.
1705 * FOLL_HWPOISON: Return -EHWPOISON instead of -EFAULT when we hit
1707 * !FOLL_FORCE: Require proper access permissions.
1709 gup_flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_UNLOCKABLE;
1711 gup_flags |= FOLL_WRITE;
1714 * We want to report -EINVAL instead of -EFAULT for any permission
1715 * problems or incompatible mappings.
1717 if (check_vma_flags(vma, gup_flags))
1720 ret = __get_user_pages(mm, start, nr_pages, gup_flags,
1727 * __mm_populate - populate and/or mlock pages within a range of address space.
1729 * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
1730 * flags. VMAs must be already marked with the desired vm_flags, and
1731 * mmap_lock must not be held.
1733 int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
1735 struct mm_struct *mm = current->mm;
1736 unsigned long end, nstart, nend;
1737 struct vm_area_struct *vma = NULL;
1743 for (nstart = start; nstart < end; nstart = nend) {
1745 * We want to fault in pages for [nstart; end) address range.
1746 * Find first corresponding VMA.
1751 vma = find_vma_intersection(mm, nstart, end);
1752 } else if (nstart >= vma->vm_end)
1753 vma = find_vma_intersection(mm, vma->vm_end, end);
1758 * Set [nstart; nend) to intersection of desired address
1759 * range with the first VMA. Also, skip undesirable VMA types.
1761 nend = min(end, vma->vm_end);
1762 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
1764 if (nstart < vma->vm_start)
1765 nstart = vma->vm_start;
1767 * Now fault in a range of pages. populate_vma_page_range()
1768 * double checks the vma flags, so that it won't mlock pages
1769 * if the vma was already munlocked.
1771 ret = populate_vma_page_range(vma, nstart, nend, &locked);
1773 if (ignore_errors) {
1775 continue; /* continue at next VMA */
1779 nend = nstart + ret * PAGE_SIZE;
1783 mmap_read_unlock(mm);
1784 return ret; /* 0 or negative error code */
1786 #else /* CONFIG_MMU */
1787 static long __get_user_pages_locked(struct mm_struct *mm, unsigned long start,
1788 unsigned long nr_pages, struct page **pages,
1789 int *locked, unsigned int foll_flags)
1791 struct vm_area_struct *vma;
1792 bool must_unlock = false;
1793 unsigned long vm_flags;
1800 * The internal caller expects GUP to manage the lock internally and the
1801 * lock must be released when this returns.
1804 if (mmap_read_lock_killable(mm))
1810 /* calculate required read or write permissions.
1811 * If FOLL_FORCE is set, we only require the "MAY" flags.
1813 vm_flags = (foll_flags & FOLL_WRITE) ?
1814 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1815 vm_flags &= (foll_flags & FOLL_FORCE) ?
1816 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1818 for (i = 0; i < nr_pages; i++) {
1819 vma = find_vma(mm, start);
1823 /* protect what we can, including chardevs */
1824 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1825 !(vm_flags & vma->vm_flags))
1829 pages[i] = virt_to_page((void *)start);
1834 start = (start + PAGE_SIZE) & PAGE_MASK;
1837 if (must_unlock && *locked) {
1838 mmap_read_unlock(mm);
1842 return i ? : -EFAULT;
1844 #endif /* !CONFIG_MMU */
1847 * fault_in_writeable - fault in userspace address range for writing
1848 * @uaddr: start of address range
1849 * @size: size of address range
1851 * Returns the number of bytes not faulted in (like copy_to_user() and
1852 * copy_from_user()).
1854 size_t fault_in_writeable(char __user *uaddr, size_t size)
1856 char __user *start = uaddr, *end;
1858 if (unlikely(size == 0))
1860 if (!user_write_access_begin(uaddr, size))
1862 if (!PAGE_ALIGNED(uaddr)) {
1863 unsafe_put_user(0, uaddr, out);
1864 uaddr = (char __user *)PAGE_ALIGN((unsigned long)uaddr);
1866 end = (char __user *)PAGE_ALIGN((unsigned long)start + size);
1867 if (unlikely(end < start))
1869 while (uaddr != end) {
1870 unsafe_put_user(0, uaddr, out);
1875 user_write_access_end();
1876 if (size > uaddr - start)
1877 return size - (uaddr - start);
1880 EXPORT_SYMBOL(fault_in_writeable);
1883 * fault_in_subpage_writeable - fault in an address range for writing
1884 * @uaddr: start of address range
1885 * @size: size of address range
1887 * Fault in a user address range for writing while checking for permissions at
1888 * sub-page granularity (e.g. arm64 MTE). This function should be used when
1889 * the caller cannot guarantee forward progress of a copy_to_user() loop.
1891 * Returns the number of bytes not faulted in (like copy_to_user() and
1892 * copy_from_user()).
1894 size_t fault_in_subpage_writeable(char __user *uaddr, size_t size)
1899 * Attempt faulting in at page granularity first for page table
1900 * permission checking. The arch-specific probe_subpage_writeable()
1901 * functions may not check for this.
1903 faulted_in = size - fault_in_writeable(uaddr, size);
1905 faulted_in -= probe_subpage_writeable(uaddr, faulted_in);
1907 return size - faulted_in;
1909 EXPORT_SYMBOL(fault_in_subpage_writeable);
1912 * fault_in_safe_writeable - fault in an address range for writing
1913 * @uaddr: start of address range
1914 * @size: length of address range
1916 * Faults in an address range for writing. This is primarily useful when we
1917 * already know that some or all of the pages in the address range aren't in
1920 * Unlike fault_in_writeable(), this function is non-destructive.
1922 * Note that we don't pin or otherwise hold the pages referenced that we fault
1923 * in. There's no guarantee that they'll stay in memory for any duration of
1926 * Returns the number of bytes not faulted in, like copy_to_user() and
1929 size_t fault_in_safe_writeable(const char __user *uaddr, size_t size)
1931 unsigned long start = (unsigned long)uaddr, end;
1932 struct mm_struct *mm = current->mm;
1933 bool unlocked = false;
1935 if (unlikely(size == 0))
1937 end = PAGE_ALIGN(start + size);
1943 if (fixup_user_fault(mm, start, FAULT_FLAG_WRITE, &unlocked))
1945 start = (start + PAGE_SIZE) & PAGE_MASK;
1946 } while (start != end);
1947 mmap_read_unlock(mm);
1949 if (size > (unsigned long)uaddr - start)
1950 return size - ((unsigned long)uaddr - start);
1953 EXPORT_SYMBOL(fault_in_safe_writeable);
1956 * fault_in_readable - fault in userspace address range for reading
1957 * @uaddr: start of user address range
1958 * @size: size of user address range
1960 * Returns the number of bytes not faulted in (like copy_to_user() and
1961 * copy_from_user()).
1963 size_t fault_in_readable(const char __user *uaddr, size_t size)
1965 const char __user *start = uaddr, *end;
1968 if (unlikely(size == 0))
1970 if (!user_read_access_begin(uaddr, size))
1972 if (!PAGE_ALIGNED(uaddr)) {
1973 unsafe_get_user(c, uaddr, out);
1974 uaddr = (const char __user *)PAGE_ALIGN((unsigned long)uaddr);
1976 end = (const char __user *)PAGE_ALIGN((unsigned long)start + size);
1977 if (unlikely(end < start))
1979 while (uaddr != end) {
1980 unsafe_get_user(c, uaddr, out);
1985 user_read_access_end();
1987 if (size > uaddr - start)
1988 return size - (uaddr - start);
1991 EXPORT_SYMBOL(fault_in_readable);
1994 * get_dump_page() - pin user page in memory while writing it to core dump
1995 * @addr: user address
1997 * Returns struct page pointer of user page pinned for dump,
1998 * to be freed afterwards by put_page().
2000 * Returns NULL on any kind of failure - a hole must then be inserted into
2001 * the corefile, to preserve alignment with its headers; and also returns
2002 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
2003 * allowing a hole to be left in the corefile to save disk space.
2005 * Called without mmap_lock (takes and releases the mmap_lock by itself).
2007 #ifdef CONFIG_ELF_CORE
2008 struct page *get_dump_page(unsigned long addr)
2014 ret = __get_user_pages_locked(current->mm, addr, 1, &page, &locked,
2015 FOLL_FORCE | FOLL_DUMP | FOLL_GET);
2016 return (ret == 1) ? page : NULL;
2018 #endif /* CONFIG_ELF_CORE */
2020 #ifdef CONFIG_MIGRATION
2022 * Returns the number of collected pages. Return value is always >= 0.
2024 static unsigned long collect_longterm_unpinnable_pages(
2025 struct list_head *movable_page_list,
2026 unsigned long nr_pages,
2027 struct page **pages)
2029 unsigned long i, collected = 0;
2030 struct folio *prev_folio = NULL;
2031 bool drain_allow = true;
2033 for (i = 0; i < nr_pages; i++) {
2034 struct folio *folio = page_folio(pages[i]);
2036 if (folio == prev_folio)
2040 if (folio_is_longterm_pinnable(folio))
2045 if (folio_is_device_coherent(folio))
2048 if (folio_test_hugetlb(folio)) {
2049 isolate_hugetlb(folio, movable_page_list);
2053 if (!folio_test_lru(folio) && drain_allow) {
2054 lru_add_drain_all();
2055 drain_allow = false;
2058 if (!folio_isolate_lru(folio))
2061 list_add_tail(&folio->lru, movable_page_list);
2062 node_stat_mod_folio(folio,
2063 NR_ISOLATED_ANON + folio_is_file_lru(folio),
2064 folio_nr_pages(folio));
2071 * Unpins all pages and migrates device coherent pages and movable_page_list.
2072 * Returns -EAGAIN if all pages were successfully migrated or -errno for failure
2073 * (or partial success).
2075 static int migrate_longterm_unpinnable_pages(
2076 struct list_head *movable_page_list,
2077 unsigned long nr_pages,
2078 struct page **pages)
2083 for (i = 0; i < nr_pages; i++) {
2084 struct folio *folio = page_folio(pages[i]);
2086 if (folio_is_device_coherent(folio)) {
2088 * Migration will fail if the page is pinned, so convert
2089 * the pin on the source page to a normal reference.
2093 gup_put_folio(folio, 1, FOLL_PIN);
2095 if (migrate_device_coherent_page(&folio->page)) {
2104 * We can't migrate pages with unexpected references, so drop
2105 * the reference obtained by __get_user_pages_locked().
2106 * Migrating pages have been added to movable_page_list after
2107 * calling folio_isolate_lru() which takes a reference so the
2108 * page won't be freed if it's migrating.
2110 unpin_user_page(pages[i]);
2114 if (!list_empty(movable_page_list)) {
2115 struct migration_target_control mtc = {
2116 .nid = NUMA_NO_NODE,
2117 .gfp_mask = GFP_USER | __GFP_NOWARN,
2120 if (migrate_pages(movable_page_list, alloc_migration_target,
2121 NULL, (unsigned long)&mtc, MIGRATE_SYNC,
2122 MR_LONGTERM_PIN, NULL)) {
2128 putback_movable_pages(movable_page_list);
2133 for (i = 0; i < nr_pages; i++)
2135 unpin_user_page(pages[i]);
2136 putback_movable_pages(movable_page_list);
2142 * Check whether all pages are *allowed* to be pinned. Rather confusingly, all
2143 * pages in the range are required to be pinned via FOLL_PIN, before calling
2146 * If any pages in the range are not allowed to be pinned, then this routine
2147 * will migrate those pages away, unpin all the pages in the range and return
2148 * -EAGAIN. The caller should re-pin the entire range with FOLL_PIN and then
2149 * call this routine again.
2151 * If an error other than -EAGAIN occurs, this indicates a migration failure.
2152 * The caller should give up, and propagate the error back up the call stack.
2154 * If everything is OK and all pages in the range are allowed to be pinned, then
2155 * this routine leaves all pages pinned and returns zero for success.
2157 static long check_and_migrate_movable_pages(unsigned long nr_pages,
2158 struct page **pages)
2160 unsigned long collected;
2161 LIST_HEAD(movable_page_list);
2163 collected = collect_longterm_unpinnable_pages(&movable_page_list,
2168 return migrate_longterm_unpinnable_pages(&movable_page_list, nr_pages,
2172 static long check_and_migrate_movable_pages(unsigned long nr_pages,
2173 struct page **pages)
2177 #endif /* CONFIG_MIGRATION */
2180 * __gup_longterm_locked() is a wrapper for __get_user_pages_locked which
2181 * allows us to process the FOLL_LONGTERM flag.
2183 static long __gup_longterm_locked(struct mm_struct *mm,
2184 unsigned long start,
2185 unsigned long nr_pages,
2186 struct page **pages,
2188 unsigned int gup_flags)
2191 long rc, nr_pinned_pages;
2193 if (!(gup_flags & FOLL_LONGTERM))
2194 return __get_user_pages_locked(mm, start, nr_pages, pages,
2197 flags = memalloc_pin_save();
2199 nr_pinned_pages = __get_user_pages_locked(mm, start, nr_pages,
2202 if (nr_pinned_pages <= 0) {
2203 rc = nr_pinned_pages;
2207 /* FOLL_LONGTERM implies FOLL_PIN */
2208 rc = check_and_migrate_movable_pages(nr_pinned_pages, pages);
2209 } while (rc == -EAGAIN);
2210 memalloc_pin_restore(flags);
2211 return rc ? rc : nr_pinned_pages;
2215 * Check that the given flags are valid for the exported gup/pup interface, and
2216 * update them with the required flags that the caller must have set.
2218 static bool is_valid_gup_args(struct page **pages, int *locked,
2219 unsigned int *gup_flags_p, unsigned int to_set)
2221 unsigned int gup_flags = *gup_flags_p;
2224 * These flags not allowed to be specified externally to the gup
2226 * - FOLL_PIN/FOLL_TRIED/FOLL_FAST_ONLY are internal only
2227 * - FOLL_REMOTE is internal only and used on follow_page()
2228 * - FOLL_UNLOCKABLE is internal only and used if locked is !NULL
2230 if (WARN_ON_ONCE(gup_flags & (FOLL_PIN | FOLL_TRIED | FOLL_UNLOCKABLE |
2231 FOLL_REMOTE | FOLL_FAST_ONLY)))
2234 gup_flags |= to_set;
2236 /* At the external interface locked must be set */
2237 if (WARN_ON_ONCE(*locked != 1))
2240 gup_flags |= FOLL_UNLOCKABLE;
2243 /* FOLL_GET and FOLL_PIN are mutually exclusive. */
2244 if (WARN_ON_ONCE((gup_flags & (FOLL_PIN | FOLL_GET)) ==
2245 (FOLL_PIN | FOLL_GET)))
2248 /* LONGTERM can only be specified when pinning */
2249 if (WARN_ON_ONCE(!(gup_flags & FOLL_PIN) && (gup_flags & FOLL_LONGTERM)))
2252 /* Pages input must be given if using GET/PIN */
2253 if (WARN_ON_ONCE((gup_flags & (FOLL_GET | FOLL_PIN)) && !pages))
2256 /* We want to allow the pgmap to be hot-unplugged at all times */
2257 if (WARN_ON_ONCE((gup_flags & FOLL_LONGTERM) &&
2258 (gup_flags & FOLL_PCI_P2PDMA)))
2261 *gup_flags_p = gup_flags;
2267 * get_user_pages_remote() - pin user pages in memory
2268 * @mm: mm_struct of target mm
2269 * @start: starting user address
2270 * @nr_pages: number of pages from start to pin
2271 * @gup_flags: flags modifying lookup behaviour
2272 * @pages: array that receives pointers to the pages pinned.
2273 * Should be at least nr_pages long. Or NULL, if caller
2274 * only intends to ensure the pages are faulted in.
2275 * @locked: pointer to lock flag indicating whether lock is held and
2276 * subsequently whether VM_FAULT_RETRY functionality can be
2277 * utilised. Lock must initially be held.
2279 * Returns either number of pages pinned (which may be less than the
2280 * number requested), or an error. Details about the return value:
2282 * -- If nr_pages is 0, returns 0.
2283 * -- If nr_pages is >0, but no pages were pinned, returns -errno.
2284 * -- If nr_pages is >0, and some pages were pinned, returns the number of
2285 * pages pinned. Again, this may be less than nr_pages.
2287 * The caller is responsible for releasing returned @pages, via put_page().
2289 * Must be called with mmap_lock held for read or write.
2291 * get_user_pages_remote walks a process's page tables and takes a reference
2292 * to each struct page that each user address corresponds to at a given
2293 * instant. That is, it takes the page that would be accessed if a user
2294 * thread accesses the given user virtual address at that instant.
2296 * This does not guarantee that the page exists in the user mappings when
2297 * get_user_pages_remote returns, and there may even be a completely different
2298 * page there in some cases (eg. if mmapped pagecache has been invalidated
2299 * and subsequently re-faulted). However it does guarantee that the page
2300 * won't be freed completely. And mostly callers simply care that the page
2301 * contains data that was valid *at some point in time*. Typically, an IO
2302 * or similar operation cannot guarantee anything stronger anyway because
2303 * locks can't be held over the syscall boundary.
2305 * If gup_flags & FOLL_WRITE == 0, the page must not be written to. If the page
2306 * is written to, set_page_dirty (or set_page_dirty_lock, as appropriate) must
2307 * be called after the page is finished with, and before put_page is called.
2309 * get_user_pages_remote is typically used for fewer-copy IO operations,
2310 * to get a handle on the memory by some means other than accesses
2311 * via the user virtual addresses. The pages may be submitted for
2312 * DMA to devices or accessed via their kernel linear mapping (via the
2313 * kmap APIs). Care should be taken to use the correct cache flushing APIs.
2315 * See also get_user_pages_fast, for performance critical applications.
2317 * get_user_pages_remote should be phased out in favor of
2318 * get_user_pages_locked|unlocked or get_user_pages_fast. Nothing
2319 * should use get_user_pages_remote because it cannot pass
2320 * FAULT_FLAG_ALLOW_RETRY to handle_mm_fault.
2322 long get_user_pages_remote(struct mm_struct *mm,
2323 unsigned long start, unsigned long nr_pages,
2324 unsigned int gup_flags, struct page **pages,
2327 int local_locked = 1;
2329 if (!is_valid_gup_args(pages, locked, &gup_flags,
2330 FOLL_TOUCH | FOLL_REMOTE))
2333 return __get_user_pages_locked(mm, start, nr_pages, pages,
2334 locked ? locked : &local_locked,
2337 EXPORT_SYMBOL(get_user_pages_remote);
2339 #else /* CONFIG_MMU */
2340 long get_user_pages_remote(struct mm_struct *mm,
2341 unsigned long start, unsigned long nr_pages,
2342 unsigned int gup_flags, struct page **pages,
2347 #endif /* !CONFIG_MMU */
2350 * get_user_pages() - pin user pages in memory
2351 * @start: starting user address
2352 * @nr_pages: number of pages from start to pin
2353 * @gup_flags: flags modifying lookup behaviour
2354 * @pages: array that receives pointers to the pages pinned.
2355 * Should be at least nr_pages long. Or NULL, if caller
2356 * only intends to ensure the pages are faulted in.
2358 * This is the same as get_user_pages_remote(), just with a less-flexible
2359 * calling convention where we assume that the mm being operated on belongs to
2360 * the current task, and doesn't allow passing of a locked parameter. We also
2361 * obviously don't pass FOLL_REMOTE in here.
2363 long get_user_pages(unsigned long start, unsigned long nr_pages,
2364 unsigned int gup_flags, struct page **pages)
2368 if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_TOUCH))
2371 return __get_user_pages_locked(current->mm, start, nr_pages, pages,
2372 &locked, gup_flags);
2374 EXPORT_SYMBOL(get_user_pages);
2377 * get_user_pages_unlocked() is suitable to replace the form:
2379 * mmap_read_lock(mm);
2380 * get_user_pages(mm, ..., pages, NULL);
2381 * mmap_read_unlock(mm);
2385 * get_user_pages_unlocked(mm, ..., pages);
2387 * It is functionally equivalent to get_user_pages_fast so
2388 * get_user_pages_fast should be used instead if specific gup_flags
2389 * (e.g. FOLL_FORCE) are not required.
2391 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
2392 struct page **pages, unsigned int gup_flags)
2396 if (!is_valid_gup_args(pages, NULL, &gup_flags,
2397 FOLL_TOUCH | FOLL_UNLOCKABLE))
2400 return __get_user_pages_locked(current->mm, start, nr_pages, pages,
2401 &locked, gup_flags);
2403 EXPORT_SYMBOL(get_user_pages_unlocked);
2408 * get_user_pages_fast attempts to pin user pages by walking the page
2409 * tables directly and avoids taking locks. Thus the walker needs to be
2410 * protected from page table pages being freed from under it, and should
2411 * block any THP splits.
2413 * One way to achieve this is to have the walker disable interrupts, and
2414 * rely on IPIs from the TLB flushing code blocking before the page table
2415 * pages are freed. This is unsuitable for architectures that do not need
2416 * to broadcast an IPI when invalidating TLBs.
2418 * Another way to achieve this is to batch up page table containing pages
2419 * belonging to more than one mm_user, then rcu_sched a callback to free those
2420 * pages. Disabling interrupts will allow the fast_gup walker to both block
2421 * the rcu_sched callback, and an IPI that we broadcast for splitting THPs
2422 * (which is a relatively rare event). The code below adopts this strategy.
2424 * Before activating this code, please be aware that the following assumptions
2425 * are currently made:
2427 * *) Either MMU_GATHER_RCU_TABLE_FREE is enabled, and tlb_remove_table() is used to
2428 * free pages containing page tables or TLB flushing requires IPI broadcast.
2430 * *) ptes can be read atomically by the architecture.
2432 * *) access_ok is sufficient to validate userspace address ranges.
2434 * The last two assumptions can be relaxed by the addition of helper functions.
2436 * This code is based heavily on the PowerPC implementation by Nick Piggin.
2438 #ifdef CONFIG_HAVE_FAST_GUP
2441 * Used in the GUP-fast path to determine whether a pin is permitted for a
2444 * This call assumes the caller has pinned the folio, that the lowest page table
2445 * level still points to this folio, and that interrupts have been disabled.
2447 * Writing to pinned file-backed dirty tracked folios is inherently problematic
2448 * (see comment describing the writable_file_mapping_allowed() function). We
2449 * therefore try to avoid the most egregious case of a long-term mapping doing
2452 * This function cannot be as thorough as that one as the VMA is not available
2453 * in the fast path, so instead we whitelist known good cases and if in doubt,
2454 * fall back to the slow path.
2456 static bool folio_fast_pin_allowed(struct folio *folio, unsigned int flags)
2458 struct address_space *mapping;
2459 unsigned long mapping_flags;
2462 * If we aren't pinning then no problematic write can occur. A long term
2463 * pin is the most egregious case so this is the one we disallow.
2465 if ((flags & (FOLL_PIN | FOLL_LONGTERM | FOLL_WRITE)) !=
2466 (FOLL_PIN | FOLL_LONGTERM | FOLL_WRITE))
2469 /* The folio is pinned, so we can safely access folio fields. */
2471 if (WARN_ON_ONCE(folio_test_slab(folio)))
2474 /* hugetlb mappings do not require dirty-tracking. */
2475 if (folio_test_hugetlb(folio))
2479 * GUP-fast disables IRQs. When IRQS are disabled, RCU grace periods
2480 * cannot proceed, which means no actions performed under RCU can
2483 * inodes and thus their mappings are freed under RCU, which means the
2484 * mapping cannot be freed beneath us and thus we can safely dereference
2487 lockdep_assert_irqs_disabled();
2490 * However, there may be operations which _alter_ the mapping, so ensure
2491 * we read it once and only once.
2493 mapping = READ_ONCE(folio->mapping);
2496 * The mapping may have been truncated, in any case we cannot determine
2497 * if this mapping is safe - fall back to slow path to determine how to
2503 /* Anonymous folios pose no problem. */
2504 mapping_flags = (unsigned long)mapping & PAGE_MAPPING_FLAGS;
2506 return mapping_flags & PAGE_MAPPING_ANON;
2509 * At this point, we know the mapping is non-null and points to an
2510 * address_space object. The only remaining whitelisted file system is
2513 return shmem_mapping(mapping);
2516 static void __maybe_unused undo_dev_pagemap(int *nr, int nr_start,
2518 struct page **pages)
2520 while ((*nr) - nr_start) {
2521 struct page *page = pages[--(*nr)];
2523 ClearPageReferenced(page);
2524 if (flags & FOLL_PIN)
2525 unpin_user_page(page);
2531 #ifdef CONFIG_ARCH_HAS_PTE_SPECIAL
2533 * Fast-gup relies on pte change detection to avoid concurrent pgtable
2536 * To pin the page, fast-gup needs to do below in order:
2537 * (1) pin the page (by prefetching pte), then (2) check pte not changed.
2539 * For the rest of pgtable operations where pgtable updates can be racy
2540 * with fast-gup, we need to do (1) clear pte, then (2) check whether page
2543 * Above will work for all pte-level operations, including THP split.
2545 * For THP collapse, it's a bit more complicated because fast-gup may be
2546 * walking a pgtable page that is being freed (pte is still valid but pmd
2547 * can be cleared already). To avoid race in such condition, we need to
2548 * also check pmd here to make sure pmd doesn't change (corresponds to
2549 * pmdp_collapse_flush() in the THP collapse code path).
2551 static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr,
2552 unsigned long end, unsigned int flags,
2553 struct page **pages, int *nr)
2555 struct dev_pagemap *pgmap = NULL;
2556 int nr_start = *nr, ret = 0;
2559 ptem = ptep = pte_offset_map(&pmd, addr);
2563 pte_t pte = ptep_get_lockless(ptep);
2565 struct folio *folio;
2567 if (pte_protnone(pte) && !gup_can_follow_protnone(flags))
2570 if (!pte_access_permitted(pte, flags & FOLL_WRITE))
2573 if (pte_devmap(pte)) {
2574 if (unlikely(flags & FOLL_LONGTERM))
2577 pgmap = get_dev_pagemap(pte_pfn(pte), pgmap);
2578 if (unlikely(!pgmap)) {
2579 undo_dev_pagemap(nr, nr_start, flags, pages);
2582 } else if (pte_special(pte))
2585 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
2586 page = pte_page(pte);
2588 folio = try_grab_folio(page, 1, flags);
2592 if (unlikely(page_is_secretmem(page))) {
2593 gup_put_folio(folio, 1, flags);
2597 if (unlikely(pmd_val(pmd) != pmd_val(*pmdp)) ||
2598 unlikely(pte_val(pte) != pte_val(ptep_get(ptep)))) {
2599 gup_put_folio(folio, 1, flags);
2603 if (!folio_fast_pin_allowed(folio, flags)) {
2604 gup_put_folio(folio, 1, flags);
2608 if (!pte_write(pte) && gup_must_unshare(NULL, flags, page)) {
2609 gup_put_folio(folio, 1, flags);
2614 * We need to make the page accessible if and only if we are
2615 * going to access its content (the FOLL_PIN case). Please
2616 * see Documentation/core-api/pin_user_pages.rst for
2619 if (flags & FOLL_PIN) {
2620 ret = arch_make_page_accessible(page);
2622 gup_put_folio(folio, 1, flags);
2626 folio_set_referenced(folio);
2629 } while (ptep++, addr += PAGE_SIZE, addr != end);
2635 put_dev_pagemap(pgmap);
2642 * If we can't determine whether or not a pte is special, then fail immediately
2643 * for ptes. Note, we can still pin HugeTLB and THP as these are guaranteed not
2646 * For a futex to be placed on a THP tail page, get_futex_key requires a
2647 * get_user_pages_fast_only implementation that can pin pages. Thus it's still
2648 * useful to have gup_huge_pmd even if we can't operate on ptes.
2650 static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr,
2651 unsigned long end, unsigned int flags,
2652 struct page **pages, int *nr)
2656 #endif /* CONFIG_ARCH_HAS_PTE_SPECIAL */
2658 #if defined(CONFIG_ARCH_HAS_PTE_DEVMAP) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
2659 static int __gup_device_huge(unsigned long pfn, unsigned long addr,
2660 unsigned long end, unsigned int flags,
2661 struct page **pages, int *nr)
2664 struct dev_pagemap *pgmap = NULL;
2667 struct page *page = pfn_to_page(pfn);
2669 pgmap = get_dev_pagemap(pfn, pgmap);
2670 if (unlikely(!pgmap)) {
2671 undo_dev_pagemap(nr, nr_start, flags, pages);
2675 if (!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page)) {
2676 undo_dev_pagemap(nr, nr_start, flags, pages);
2680 SetPageReferenced(page);
2682 if (unlikely(try_grab_page(page, flags))) {
2683 undo_dev_pagemap(nr, nr_start, flags, pages);
2688 } while (addr += PAGE_SIZE, addr != end);
2690 put_dev_pagemap(pgmap);
2694 static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
2695 unsigned long end, unsigned int flags,
2696 struct page **pages, int *nr)
2698 unsigned long fault_pfn;
2701 fault_pfn = pmd_pfn(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
2702 if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr))
2705 if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) {
2706 undo_dev_pagemap(nr, nr_start, flags, pages);
2712 static int __gup_device_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr,
2713 unsigned long end, unsigned int flags,
2714 struct page **pages, int *nr)
2716 unsigned long fault_pfn;
2719 fault_pfn = pud_pfn(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
2720 if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr))
2723 if (unlikely(pud_val(orig) != pud_val(*pudp))) {
2724 undo_dev_pagemap(nr, nr_start, flags, pages);
2730 static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
2731 unsigned long end, unsigned int flags,
2732 struct page **pages, int *nr)
2738 static int __gup_device_huge_pud(pud_t pud, pud_t *pudp, unsigned long addr,
2739 unsigned long end, unsigned int flags,
2740 struct page **pages, int *nr)
2747 static int record_subpages(struct page *page, unsigned long addr,
2748 unsigned long end, struct page **pages)
2752 for (nr = 0; addr != end; nr++, addr += PAGE_SIZE)
2753 pages[nr] = nth_page(page, nr);
2758 #ifdef CONFIG_ARCH_HAS_HUGEPD
2759 static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end,
2762 unsigned long __boundary = (addr + sz) & ~(sz-1);
2763 return (__boundary - 1 < end - 1) ? __boundary : end;
2766 static int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr,
2767 unsigned long end, unsigned int flags,
2768 struct page **pages, int *nr)
2770 unsigned long pte_end;
2772 struct folio *folio;
2776 pte_end = (addr + sz) & ~(sz-1);
2780 pte = huge_ptep_get(ptep);
2782 if (!pte_access_permitted(pte, flags & FOLL_WRITE))
2785 /* hugepages are never "special" */
2786 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
2788 page = nth_page(pte_page(pte), (addr & (sz - 1)) >> PAGE_SHIFT);
2789 refs = record_subpages(page, addr, end, pages + *nr);
2791 folio = try_grab_folio(page, refs, flags);
2795 if (unlikely(pte_val(pte) != pte_val(ptep_get(ptep)))) {
2796 gup_put_folio(folio, refs, flags);
2800 if (!folio_fast_pin_allowed(folio, flags)) {
2801 gup_put_folio(folio, refs, flags);
2805 if (!pte_write(pte) && gup_must_unshare(NULL, flags, &folio->page)) {
2806 gup_put_folio(folio, refs, flags);
2811 folio_set_referenced(folio);
2815 static int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
2816 unsigned int pdshift, unsigned long end, unsigned int flags,
2817 struct page **pages, int *nr)
2820 unsigned long sz = 1UL << hugepd_shift(hugepd);
2823 ptep = hugepte_offset(hugepd, addr, pdshift);
2825 next = hugepte_addr_end(addr, end, sz);
2826 if (!gup_hugepte(ptep, sz, addr, end, flags, pages, nr))
2828 } while (ptep++, addr = next, addr != end);
2833 static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
2834 unsigned int pdshift, unsigned long end, unsigned int flags,
2835 struct page **pages, int *nr)
2839 #endif /* CONFIG_ARCH_HAS_HUGEPD */
2841 static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
2842 unsigned long end, unsigned int flags,
2843 struct page **pages, int *nr)
2846 struct folio *folio;
2849 if (!pmd_access_permitted(orig, flags & FOLL_WRITE))
2852 if (pmd_devmap(orig)) {
2853 if (unlikely(flags & FOLL_LONGTERM))
2855 return __gup_device_huge_pmd(orig, pmdp, addr, end, flags,
2859 page = nth_page(pmd_page(orig), (addr & ~PMD_MASK) >> PAGE_SHIFT);
2860 refs = record_subpages(page, addr, end, pages + *nr);
2862 folio = try_grab_folio(page, refs, flags);
2866 if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) {
2867 gup_put_folio(folio, refs, flags);
2871 if (!folio_fast_pin_allowed(folio, flags)) {
2872 gup_put_folio(folio, refs, flags);
2875 if (!pmd_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) {
2876 gup_put_folio(folio, refs, flags);
2881 folio_set_referenced(folio);
2885 static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr,
2886 unsigned long end, unsigned int flags,
2887 struct page **pages, int *nr)
2890 struct folio *folio;
2893 if (!pud_access_permitted(orig, flags & FOLL_WRITE))
2896 if (pud_devmap(orig)) {
2897 if (unlikely(flags & FOLL_LONGTERM))
2899 return __gup_device_huge_pud(orig, pudp, addr, end, flags,
2903 page = nth_page(pud_page(orig), (addr & ~PUD_MASK) >> PAGE_SHIFT);
2904 refs = record_subpages(page, addr, end, pages + *nr);
2906 folio = try_grab_folio(page, refs, flags);
2910 if (unlikely(pud_val(orig) != pud_val(*pudp))) {
2911 gup_put_folio(folio, refs, flags);
2915 if (!folio_fast_pin_allowed(folio, flags)) {
2916 gup_put_folio(folio, refs, flags);
2920 if (!pud_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) {
2921 gup_put_folio(folio, refs, flags);
2926 folio_set_referenced(folio);
2930 static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr,
2931 unsigned long end, unsigned int flags,
2932 struct page **pages, int *nr)
2936 struct folio *folio;
2938 if (!pgd_access_permitted(orig, flags & FOLL_WRITE))
2941 BUILD_BUG_ON(pgd_devmap(orig));
2943 page = nth_page(pgd_page(orig), (addr & ~PGDIR_MASK) >> PAGE_SHIFT);
2944 refs = record_subpages(page, addr, end, pages + *nr);
2946 folio = try_grab_folio(page, refs, flags);
2950 if (unlikely(pgd_val(orig) != pgd_val(*pgdp))) {
2951 gup_put_folio(folio, refs, flags);
2955 if (!pgd_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) {
2956 gup_put_folio(folio, refs, flags);
2960 if (!folio_fast_pin_allowed(folio, flags)) {
2961 gup_put_folio(folio, refs, flags);
2966 folio_set_referenced(folio);
2970 static int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr, unsigned long end,
2971 unsigned int flags, struct page **pages, int *nr)
2976 pmdp = pmd_offset_lockless(pudp, pud, addr);
2978 pmd_t pmd = pmdp_get_lockless(pmdp);
2980 next = pmd_addr_end(addr, end);
2981 if (!pmd_present(pmd))
2984 if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) ||
2986 if (pmd_protnone(pmd) &&
2987 !gup_can_follow_protnone(flags))
2990 if (!gup_huge_pmd(pmd, pmdp, addr, next, flags,
2994 } else if (unlikely(is_hugepd(__hugepd(pmd_val(pmd))))) {
2996 * architecture have different format for hugetlbfs
2997 * pmd format and THP pmd format
2999 if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr,
3000 PMD_SHIFT, next, flags, pages, nr))
3002 } else if (!gup_pte_range(pmd, pmdp, addr, next, flags, pages, nr))
3004 } while (pmdp++, addr = next, addr != end);
3009 static int gup_pud_range(p4d_t *p4dp, p4d_t p4d, unsigned long addr, unsigned long end,
3010 unsigned int flags, struct page **pages, int *nr)
3015 pudp = pud_offset_lockless(p4dp, p4d, addr);
3017 pud_t pud = READ_ONCE(*pudp);
3019 next = pud_addr_end(addr, end);
3020 if (unlikely(!pud_present(pud)))
3022 if (unlikely(pud_huge(pud) || pud_devmap(pud))) {
3023 if (!gup_huge_pud(pud, pudp, addr, next, flags,
3026 } else if (unlikely(is_hugepd(__hugepd(pud_val(pud))))) {
3027 if (!gup_huge_pd(__hugepd(pud_val(pud)), addr,
3028 PUD_SHIFT, next, flags, pages, nr))
3030 } else if (!gup_pmd_range(pudp, pud, addr, next, flags, pages, nr))
3032 } while (pudp++, addr = next, addr != end);
3037 static int gup_p4d_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr, unsigned long end,
3038 unsigned int flags, struct page **pages, int *nr)
3043 p4dp = p4d_offset_lockless(pgdp, pgd, addr);
3045 p4d_t p4d = READ_ONCE(*p4dp);
3047 next = p4d_addr_end(addr, end);
3050 BUILD_BUG_ON(p4d_huge(p4d));
3051 if (unlikely(is_hugepd(__hugepd(p4d_val(p4d))))) {
3052 if (!gup_huge_pd(__hugepd(p4d_val(p4d)), addr,
3053 P4D_SHIFT, next, flags, pages, nr))
3055 } else if (!gup_pud_range(p4dp, p4d, addr, next, flags, pages, nr))
3057 } while (p4dp++, addr = next, addr != end);
3062 static void gup_pgd_range(unsigned long addr, unsigned long end,
3063 unsigned int flags, struct page **pages, int *nr)
3068 pgdp = pgd_offset(current->mm, addr);
3070 pgd_t pgd = READ_ONCE(*pgdp);
3072 next = pgd_addr_end(addr, end);
3075 if (unlikely(pgd_huge(pgd))) {
3076 if (!gup_huge_pgd(pgd, pgdp, addr, next, flags,
3079 } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) {
3080 if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr,
3081 PGDIR_SHIFT, next, flags, pages, nr))
3083 } else if (!gup_p4d_range(pgdp, pgd, addr, next, flags, pages, nr))
3085 } while (pgdp++, addr = next, addr != end);
3088 static inline void gup_pgd_range(unsigned long addr, unsigned long end,
3089 unsigned int flags, struct page **pages, int *nr)
3092 #endif /* CONFIG_HAVE_FAST_GUP */
3094 #ifndef gup_fast_permitted
3096 * Check if it's allowed to use get_user_pages_fast_only() for the range, or
3097 * we need to fall back to the slow version:
3099 static bool gup_fast_permitted(unsigned long start, unsigned long end)
3105 static unsigned long lockless_pages_from_mm(unsigned long start,
3107 unsigned int gup_flags,
3108 struct page **pages)
3110 unsigned long flags;
3114 if (!IS_ENABLED(CONFIG_HAVE_FAST_GUP) ||
3115 !gup_fast_permitted(start, end))
3118 if (gup_flags & FOLL_PIN) {
3119 seq = raw_read_seqcount(¤t->mm->write_protect_seq);
3125 * Disable interrupts. The nested form is used, in order to allow full,
3126 * general purpose use of this routine.
3128 * With interrupts disabled, we block page table pages from being freed
3129 * from under us. See struct mmu_table_batch comments in
3130 * include/asm-generic/tlb.h for more details.
3132 * We do not adopt an rcu_read_lock() here as we also want to block IPIs
3133 * that come from THPs splitting.
3135 local_irq_save(flags);
3136 gup_pgd_range(start, end, gup_flags, pages, &nr_pinned);
3137 local_irq_restore(flags);
3140 * When pinning pages for DMA there could be a concurrent write protect
3141 * from fork() via copy_page_range(), in this case always fail fast GUP.
3143 if (gup_flags & FOLL_PIN) {
3144 if (read_seqcount_retry(¤t->mm->write_protect_seq, seq)) {
3145 unpin_user_pages_lockless(pages, nr_pinned);
3148 sanity_check_pinned_pages(pages, nr_pinned);
3154 static int internal_get_user_pages_fast(unsigned long start,
3155 unsigned long nr_pages,
3156 unsigned int gup_flags,
3157 struct page **pages)
3159 unsigned long len, end;
3160 unsigned long nr_pinned;
3164 if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM |
3165 FOLL_FORCE | FOLL_PIN | FOLL_GET |
3166 FOLL_FAST_ONLY | FOLL_NOFAULT |
3170 if (gup_flags & FOLL_PIN)
3171 mm_set_has_pinned_flag(¤t->mm->flags);
3173 if (!(gup_flags & FOLL_FAST_ONLY))
3174 might_lock_read(¤t->mm->mmap_lock);
3176 start = untagged_addr(start) & PAGE_MASK;
3177 len = nr_pages << PAGE_SHIFT;
3178 if (check_add_overflow(start, len, &end))
3180 if (end > TASK_SIZE_MAX)
3182 if (unlikely(!access_ok((void __user *)start, len)))
3185 nr_pinned = lockless_pages_from_mm(start, end, gup_flags, pages);
3186 if (nr_pinned == nr_pages || gup_flags & FOLL_FAST_ONLY)
3189 /* Slow path: try to get the remaining pages with get_user_pages */
3190 start += nr_pinned << PAGE_SHIFT;
3192 ret = __gup_longterm_locked(current->mm, start, nr_pages - nr_pinned,
3194 gup_flags | FOLL_TOUCH | FOLL_UNLOCKABLE);
3197 * The caller has to unpin the pages we already pinned so
3198 * returning -errno is not an option
3204 return ret + nr_pinned;
3208 * get_user_pages_fast_only() - pin user pages in memory
3209 * @start: starting user address
3210 * @nr_pages: number of pages from start to pin
3211 * @gup_flags: flags modifying pin behaviour
3212 * @pages: array that receives pointers to the pages pinned.
3213 * Should be at least nr_pages long.
3215 * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to
3218 * If the architecture does not support this function, simply return with no
3221 * Careful, careful! COW breaking can go either way, so a non-write
3222 * access can get ambiguous page results. If you call this function without
3223 * 'write' set, you'd better be sure that you're ok with that ambiguity.
3225 int get_user_pages_fast_only(unsigned long start, int nr_pages,
3226 unsigned int gup_flags, struct page **pages)
3229 * Internally (within mm/gup.c), gup fast variants must set FOLL_GET,
3230 * because gup fast is always a "pin with a +1 page refcount" request.
3232 * FOLL_FAST_ONLY is required in order to match the API description of
3233 * this routine: no fall back to regular ("slow") GUP.
3235 if (!is_valid_gup_args(pages, NULL, &gup_flags,
3236 FOLL_GET | FOLL_FAST_ONLY))
3239 return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages);
3241 EXPORT_SYMBOL_GPL(get_user_pages_fast_only);
3244 * get_user_pages_fast() - pin user pages in memory
3245 * @start: starting user address
3246 * @nr_pages: number of pages from start to pin
3247 * @gup_flags: flags modifying pin behaviour
3248 * @pages: array that receives pointers to the pages pinned.
3249 * Should be at least nr_pages long.
3251 * Attempt to pin user pages in memory without taking mm->mmap_lock.
3252 * If not successful, it will fall back to taking the lock and
3253 * calling get_user_pages().
3255 * Returns number of pages pinned. This may be fewer than the number requested.
3256 * If nr_pages is 0 or negative, returns 0. If no pages were pinned, returns
3259 int get_user_pages_fast(unsigned long start, int nr_pages,
3260 unsigned int gup_flags, struct page **pages)
3263 * The caller may or may not have explicitly set FOLL_GET; either way is
3264 * OK. However, internally (within mm/gup.c), gup fast variants must set
3265 * FOLL_GET, because gup fast is always a "pin with a +1 page refcount"
3268 if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_GET))
3270 return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages);
3272 EXPORT_SYMBOL_GPL(get_user_pages_fast);
3275 * pin_user_pages_fast() - pin user pages in memory without taking locks
3277 * @start: starting user address
3278 * @nr_pages: number of pages from start to pin
3279 * @gup_flags: flags modifying pin behaviour
3280 * @pages: array that receives pointers to the pages pinned.
3281 * Should be at least nr_pages long.
3283 * Nearly the same as get_user_pages_fast(), except that FOLL_PIN is set. See
3284 * get_user_pages_fast() for documentation on the function arguments, because
3285 * the arguments here are identical.
3287 * FOLL_PIN means that the pages must be released via unpin_user_page(). Please
3288 * see Documentation/core-api/pin_user_pages.rst for further details.
3290 * Note that if a zero_page is amongst the returned pages, it will not have
3291 * pins in it and unpin_user_page() will not remove pins from it.
3293 int pin_user_pages_fast(unsigned long start, int nr_pages,
3294 unsigned int gup_flags, struct page **pages)
3296 if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_PIN))
3298 return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages);
3300 EXPORT_SYMBOL_GPL(pin_user_pages_fast);
3303 * pin_user_pages_remote() - pin pages of a remote process
3305 * @mm: mm_struct of target mm
3306 * @start: starting user address
3307 * @nr_pages: number of pages from start to pin
3308 * @gup_flags: flags modifying lookup behaviour
3309 * @pages: array that receives pointers to the pages pinned.
3310 * Should be at least nr_pages long.
3311 * @locked: pointer to lock flag indicating whether lock is held and
3312 * subsequently whether VM_FAULT_RETRY functionality can be
3313 * utilised. Lock must initially be held.
3315 * Nearly the same as get_user_pages_remote(), except that FOLL_PIN is set. See
3316 * get_user_pages_remote() for documentation on the function arguments, because
3317 * the arguments here are identical.
3319 * FOLL_PIN means that the pages must be released via unpin_user_page(). Please
3320 * see Documentation/core-api/pin_user_pages.rst for details.
3322 * Note that if a zero_page is amongst the returned pages, it will not have
3323 * pins in it and unpin_user_page*() will not remove pins from it.
3325 long pin_user_pages_remote(struct mm_struct *mm,
3326 unsigned long start, unsigned long nr_pages,
3327 unsigned int gup_flags, struct page **pages,
3330 int local_locked = 1;
3332 if (!is_valid_gup_args(pages, locked, &gup_flags,
3333 FOLL_PIN | FOLL_TOUCH | FOLL_REMOTE))
3335 return __gup_longterm_locked(mm, start, nr_pages, pages,
3336 locked ? locked : &local_locked,
3339 EXPORT_SYMBOL(pin_user_pages_remote);
3342 * pin_user_pages() - pin user pages in memory for use by other devices
3344 * @start: starting user address
3345 * @nr_pages: number of pages from start to pin
3346 * @gup_flags: flags modifying lookup behaviour
3347 * @pages: array that receives pointers to the pages pinned.
3348 * Should be at least nr_pages long.
3350 * Nearly the same as get_user_pages(), except that FOLL_TOUCH is not set, and
3353 * FOLL_PIN means that the pages must be released via unpin_user_page(). Please
3354 * see Documentation/core-api/pin_user_pages.rst for details.
3356 * Note that if a zero_page is amongst the returned pages, it will not have
3357 * pins in it and unpin_user_page*() will not remove pins from it.
3359 long pin_user_pages(unsigned long start, unsigned long nr_pages,
3360 unsigned int gup_flags, struct page **pages)
3364 if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_PIN))
3366 return __gup_longterm_locked(current->mm, start, nr_pages,
3367 pages, &locked, gup_flags);
3369 EXPORT_SYMBOL(pin_user_pages);
3372 * pin_user_pages_unlocked() is the FOLL_PIN variant of
3373 * get_user_pages_unlocked(). Behavior is the same, except that this one sets
3374 * FOLL_PIN and rejects FOLL_GET.
3376 * Note that if a zero_page is amongst the returned pages, it will not have
3377 * pins in it and unpin_user_page*() will not remove pins from it.
3379 long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
3380 struct page **pages, unsigned int gup_flags)
3384 if (!is_valid_gup_args(pages, NULL, &gup_flags,
3385 FOLL_PIN | FOLL_TOUCH | FOLL_UNLOCKABLE))
3388 return __gup_longterm_locked(current->mm, start, nr_pages, pages,
3389 &locked, gup_flags);
3391 EXPORT_SYMBOL(pin_user_pages_unlocked);