2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/aio.h>
36 static struct vfsmount *shm_mnt;
40 * This virtual memory filesystem is heavily based on the ramfs. It
41 * extends ramfs by the ability to use swap and honor resource limits
42 * which makes it a completely usable filesystem.
45 #include <linux/xattr.h>
46 #include <linux/exportfs.h>
47 #include <linux/posix_acl.h>
48 #include <linux/posix_acl_xattr.h>
49 #include <linux/mman.h>
50 #include <linux/string.h>
51 #include <linux/slab.h>
52 #include <linux/backing-dev.h>
53 #include <linux/shmem_fs.h>
54 #include <linux/writeback.h>
55 #include <linux/blkdev.h>
56 #include <linux/pagevec.h>
57 #include <linux/percpu_counter.h>
58 #include <linux/falloc.h>
59 #include <linux/splice.h>
60 #include <linux/security.h>
61 #include <linux/swapops.h>
62 #include <linux/mempolicy.h>
63 #include <linux/namei.h>
64 #include <linux/ctype.h>
65 #include <linux/migrate.h>
66 #include <linux/highmem.h>
67 #include <linux/seq_file.h>
68 #include <linux/magic.h>
70 #include <asm/uaccess.h>
71 #include <asm/pgtable.h>
73 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
74 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
76 /* Pretend that each entry is of this size in directory's i_size */
77 #define BOGO_DIRENT_SIZE 20
79 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
80 #define SHORT_SYMLINK_LEN 128
83 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
84 * inode->i_private (with i_mutex making sure that it has only one user at
85 * a time): we would prefer not to enlarge the shmem inode just for that.
88 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
89 pgoff_t start; /* start of range currently being fallocated */
90 pgoff_t next; /* the next page offset to be fallocated */
91 pgoff_t nr_falloced; /* how many new pages have been fallocated */
92 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
95 /* Flag allocation requirements to shmem_getpage */
97 SGP_READ, /* don't exceed i_size, don't allocate page */
98 SGP_CACHE, /* don't exceed i_size, may allocate page */
99 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
100 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
101 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
105 static unsigned long shmem_default_max_blocks(void)
107 return totalram_pages / 2;
110 static unsigned long shmem_default_max_inodes(void)
112 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
116 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
117 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
118 struct shmem_inode_info *info, pgoff_t index);
119 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
120 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
122 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
123 struct page **pagep, enum sgp_type sgp, int *fault_type)
125 return shmem_getpage_gfp(inode, index, pagep, sgp,
126 mapping_gfp_mask(inode->i_mapping), fault_type);
129 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
131 return sb->s_fs_info;
135 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
136 * for shared memory and for shared anonymous (/dev/zero) mappings
137 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
138 * consistent with the pre-accounting of private mappings ...
140 static inline int shmem_acct_size(unsigned long flags, loff_t size)
142 return (flags & VM_NORESERVE) ?
143 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
146 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
148 if (!(flags & VM_NORESERVE))
149 vm_unacct_memory(VM_ACCT(size));
152 static inline int shmem_reacct_size(unsigned long flags,
153 loff_t oldsize, loff_t newsize)
155 if (!(flags & VM_NORESERVE)) {
156 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
157 return security_vm_enough_memory_mm(current->mm,
158 VM_ACCT(newsize) - VM_ACCT(oldsize));
159 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
160 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
166 * ... whereas tmpfs objects are accounted incrementally as
167 * pages are allocated, in order to allow huge sparse files.
168 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
169 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
171 static inline int shmem_acct_block(unsigned long flags)
173 return (flags & VM_NORESERVE) ?
174 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
177 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
179 if (flags & VM_NORESERVE)
180 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
183 static const struct super_operations shmem_ops;
184 static const struct address_space_operations shmem_aops;
185 static const struct file_operations shmem_file_operations;
186 static const struct inode_operations shmem_inode_operations;
187 static const struct inode_operations shmem_dir_inode_operations;
188 static const struct inode_operations shmem_special_inode_operations;
189 static const struct vm_operations_struct shmem_vm_ops;
191 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
192 .ra_pages = 0, /* No readahead */
193 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
196 static LIST_HEAD(shmem_swaplist);
197 static DEFINE_MUTEX(shmem_swaplist_mutex);
199 static int shmem_reserve_inode(struct super_block *sb)
201 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
202 if (sbinfo->max_inodes) {
203 spin_lock(&sbinfo->stat_lock);
204 if (!sbinfo->free_inodes) {
205 spin_unlock(&sbinfo->stat_lock);
208 sbinfo->free_inodes--;
209 spin_unlock(&sbinfo->stat_lock);
214 static void shmem_free_inode(struct super_block *sb)
216 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
217 if (sbinfo->max_inodes) {
218 spin_lock(&sbinfo->stat_lock);
219 sbinfo->free_inodes++;
220 spin_unlock(&sbinfo->stat_lock);
225 * shmem_recalc_inode - recalculate the block usage of an inode
226 * @inode: inode to recalc
228 * We have to calculate the free blocks since the mm can drop
229 * undirtied hole pages behind our back.
231 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
232 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
234 * It has to be called with the spinlock held.
236 static void shmem_recalc_inode(struct inode *inode)
238 struct shmem_inode_info *info = SHMEM_I(inode);
241 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
243 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
244 if (sbinfo->max_blocks)
245 percpu_counter_add(&sbinfo->used_blocks, -freed);
246 info->alloced -= freed;
247 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
248 shmem_unacct_blocks(info->flags, freed);
253 * Replace item expected in radix tree by a new item, while holding tree lock.
255 static int shmem_radix_tree_replace(struct address_space *mapping,
256 pgoff_t index, void *expected, void *replacement)
261 VM_BUG_ON(!expected);
262 VM_BUG_ON(!replacement);
263 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
266 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
267 if (item != expected)
269 radix_tree_replace_slot(pslot, replacement);
274 * Sometimes, before we decide whether to proceed or to fail, we must check
275 * that an entry was not already brought back from swap by a racing thread.
277 * Checking page is not enough: by the time a SwapCache page is locked, it
278 * might be reused, and again be SwapCache, using the same swap as before.
280 static bool shmem_confirm_swap(struct address_space *mapping,
281 pgoff_t index, swp_entry_t swap)
286 item = radix_tree_lookup(&mapping->page_tree, index);
288 return item == swp_to_radix_entry(swap);
292 * Like add_to_page_cache_locked, but error if expected item has gone.
294 static int shmem_add_to_page_cache(struct page *page,
295 struct address_space *mapping,
296 pgoff_t index, void *expected)
300 VM_BUG_ON_PAGE(!PageLocked(page), page);
301 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
303 page_cache_get(page);
304 page->mapping = mapping;
307 spin_lock_irq(&mapping->tree_lock);
309 error = radix_tree_insert(&mapping->page_tree, index, page);
311 error = shmem_radix_tree_replace(mapping, index, expected,
315 __inc_zone_page_state(page, NR_FILE_PAGES);
316 __inc_zone_page_state(page, NR_SHMEM);
317 spin_unlock_irq(&mapping->tree_lock);
319 page->mapping = NULL;
320 spin_unlock_irq(&mapping->tree_lock);
321 page_cache_release(page);
327 * Like delete_from_page_cache, but substitutes swap for page.
329 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
331 struct address_space *mapping = page->mapping;
334 spin_lock_irq(&mapping->tree_lock);
335 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
336 page->mapping = NULL;
338 __dec_zone_page_state(page, NR_FILE_PAGES);
339 __dec_zone_page_state(page, NR_SHMEM);
340 spin_unlock_irq(&mapping->tree_lock);
341 page_cache_release(page);
346 * Remove swap entry from radix tree, free the swap and its page cache.
348 static int shmem_free_swap(struct address_space *mapping,
349 pgoff_t index, void *radswap)
353 spin_lock_irq(&mapping->tree_lock);
354 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
355 spin_unlock_irq(&mapping->tree_lock);
358 free_swap_and_cache(radix_to_swp_entry(radswap));
363 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
365 void shmem_unlock_mapping(struct address_space *mapping)
368 pgoff_t indices[PAGEVEC_SIZE];
371 pagevec_init(&pvec, 0);
373 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
375 while (!mapping_unevictable(mapping)) {
377 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
378 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
380 pvec.nr = find_get_entries(mapping, index,
381 PAGEVEC_SIZE, pvec.pages, indices);
384 index = indices[pvec.nr - 1] + 1;
385 pagevec_remove_exceptionals(&pvec);
386 check_move_unevictable_pages(pvec.pages, pvec.nr);
387 pagevec_release(&pvec);
393 * Remove range of pages and swap entries from radix tree, and free them.
394 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
396 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
399 struct address_space *mapping = inode->i_mapping;
400 struct shmem_inode_info *info = SHMEM_I(inode);
401 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
402 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
403 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
404 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
406 pgoff_t indices[PAGEVEC_SIZE];
407 long nr_swaps_freed = 0;
412 end = -1; /* unsigned, so actually very big */
414 pagevec_init(&pvec, 0);
416 while (index < end) {
417 pvec.nr = find_get_entries(mapping, index,
418 min(end - index, (pgoff_t)PAGEVEC_SIZE),
419 pvec.pages, indices);
422 mem_cgroup_uncharge_start();
423 for (i = 0; i < pagevec_count(&pvec); i++) {
424 struct page *page = pvec.pages[i];
430 if (radix_tree_exceptional_entry(page)) {
433 nr_swaps_freed += !shmem_free_swap(mapping,
438 if (!trylock_page(page))
440 if (!unfalloc || !PageUptodate(page)) {
441 if (page->mapping == mapping) {
442 VM_BUG_ON_PAGE(PageWriteback(page), page);
443 truncate_inode_page(mapping, page);
448 pagevec_remove_exceptionals(&pvec);
449 pagevec_release(&pvec);
450 mem_cgroup_uncharge_end();
456 struct page *page = NULL;
457 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
459 unsigned int top = PAGE_CACHE_SIZE;
464 zero_user_segment(page, partial_start, top);
465 set_page_dirty(page);
467 page_cache_release(page);
471 struct page *page = NULL;
472 shmem_getpage(inode, end, &page, SGP_READ, NULL);
474 zero_user_segment(page, 0, partial_end);
475 set_page_dirty(page);
477 page_cache_release(page);
484 while (index < end) {
487 pvec.nr = find_get_entries(mapping, index,
488 min(end - index, (pgoff_t)PAGEVEC_SIZE),
489 pvec.pages, indices);
491 /* If all gone or hole-punch or unfalloc, we're done */
492 if (index == start || end != -1)
494 /* But if truncating, restart to make sure all gone */
498 mem_cgroup_uncharge_start();
499 for (i = 0; i < pagevec_count(&pvec); i++) {
500 struct page *page = pvec.pages[i];
506 if (radix_tree_exceptional_entry(page)) {
509 if (shmem_free_swap(mapping, index, page)) {
510 /* Swap was replaced by page: retry */
519 if (!unfalloc || !PageUptodate(page)) {
520 if (page->mapping == mapping) {
521 VM_BUG_ON_PAGE(PageWriteback(page), page);
522 truncate_inode_page(mapping, page);
524 /* Page was replaced by swap: retry */
532 pagevec_remove_exceptionals(&pvec);
533 pagevec_release(&pvec);
534 mem_cgroup_uncharge_end();
538 spin_lock(&info->lock);
539 info->swapped -= nr_swaps_freed;
540 shmem_recalc_inode(inode);
541 spin_unlock(&info->lock);
544 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
546 shmem_undo_range(inode, lstart, lend, false);
547 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
549 EXPORT_SYMBOL_GPL(shmem_truncate_range);
551 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
553 struct inode *inode = dentry->d_inode;
556 error = inode_change_ok(inode, attr);
560 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
561 loff_t oldsize = inode->i_size;
562 loff_t newsize = attr->ia_size;
564 if (newsize != oldsize) {
565 error = shmem_reacct_size(SHMEM_I(inode)->flags,
569 i_size_write(inode, newsize);
570 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
572 if (newsize < oldsize) {
573 loff_t holebegin = round_up(newsize, PAGE_SIZE);
574 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
575 shmem_truncate_range(inode, newsize, (loff_t)-1);
576 /* unmap again to remove racily COWed private pages */
577 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
581 setattr_copy(inode, attr);
582 if (attr->ia_valid & ATTR_MODE)
583 error = posix_acl_chmod(inode, inode->i_mode);
587 static void shmem_evict_inode(struct inode *inode)
589 struct shmem_inode_info *info = SHMEM_I(inode);
591 if (inode->i_mapping->a_ops == &shmem_aops) {
592 shmem_unacct_size(info->flags, inode->i_size);
594 shmem_truncate_range(inode, 0, (loff_t)-1);
595 if (!list_empty(&info->swaplist)) {
596 mutex_lock(&shmem_swaplist_mutex);
597 list_del_init(&info->swaplist);
598 mutex_unlock(&shmem_swaplist_mutex);
601 kfree(info->symlink);
603 simple_xattrs_free(&info->xattrs);
604 WARN_ON(inode->i_blocks);
605 shmem_free_inode(inode->i_sb);
610 * If swap found in inode, free it and move page from swapcache to filecache.
612 static int shmem_unuse_inode(struct shmem_inode_info *info,
613 swp_entry_t swap, struct page **pagep)
615 struct address_space *mapping = info->vfs_inode.i_mapping;
621 radswap = swp_to_radix_entry(swap);
622 index = radix_tree_locate_item(&mapping->page_tree, radswap);
624 return -EAGAIN; /* tell shmem_unuse we found nothing */
627 * Move _head_ to start search for next from here.
628 * But be careful: shmem_evict_inode checks list_empty without taking
629 * mutex, and there's an instant in list_move_tail when info->swaplist
630 * would appear empty, if it were the only one on shmem_swaplist.
632 if (shmem_swaplist.next != &info->swaplist)
633 list_move_tail(&shmem_swaplist, &info->swaplist);
635 gfp = mapping_gfp_mask(mapping);
636 if (shmem_should_replace_page(*pagep, gfp)) {
637 mutex_unlock(&shmem_swaplist_mutex);
638 error = shmem_replace_page(pagep, gfp, info, index);
639 mutex_lock(&shmem_swaplist_mutex);
641 * We needed to drop mutex to make that restrictive page
642 * allocation, but the inode might have been freed while we
643 * dropped it: although a racing shmem_evict_inode() cannot
644 * complete without emptying the radix_tree, our page lock
645 * on this swapcache page is not enough to prevent that -
646 * free_swap_and_cache() of our swap entry will only
647 * trylock_page(), removing swap from radix_tree whatever.
649 * We must not proceed to shmem_add_to_page_cache() if the
650 * inode has been freed, but of course we cannot rely on
651 * inode or mapping or info to check that. However, we can
652 * safely check if our swap entry is still in use (and here
653 * it can't have got reused for another page): if it's still
654 * in use, then the inode cannot have been freed yet, and we
655 * can safely proceed (if it's no longer in use, that tells
656 * nothing about the inode, but we don't need to unuse swap).
658 if (!page_swapcount(*pagep))
663 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
664 * but also to hold up shmem_evict_inode(): so inode cannot be freed
665 * beneath us (pagelock doesn't help until the page is in pagecache).
668 error = shmem_add_to_page_cache(*pagep, mapping, index,
670 if (error != -ENOMEM) {
672 * Truncation and eviction use free_swap_and_cache(), which
673 * only does trylock page: if we raced, best clean up here.
675 delete_from_swap_cache(*pagep);
676 set_page_dirty(*pagep);
678 spin_lock(&info->lock);
680 spin_unlock(&info->lock);
688 * Search through swapped inodes to find and replace swap by page.
690 int shmem_unuse(swp_entry_t swap, struct page *page)
692 struct list_head *this, *next;
693 struct shmem_inode_info *info;
694 struct mem_cgroup *memcg;
698 * There's a faint possibility that swap page was replaced before
699 * caller locked it: caller will come back later with the right page.
701 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
705 * Charge page using GFP_KERNEL while we can wait, before taking
706 * the shmem_swaplist_mutex which might hold up shmem_writepage().
707 * Charged back to the user (not to caller) when swap account is used.
709 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg);
712 /* No radix_tree_preload: swap entry keeps a place for page in tree */
715 mutex_lock(&shmem_swaplist_mutex);
716 list_for_each_safe(this, next, &shmem_swaplist) {
717 info = list_entry(this, struct shmem_inode_info, swaplist);
719 error = shmem_unuse_inode(info, swap, &page);
721 list_del_init(&info->swaplist);
723 if (error != -EAGAIN)
725 /* found nothing in this: move on to search the next */
727 mutex_unlock(&shmem_swaplist_mutex);
730 if (error != -ENOMEM)
732 mem_cgroup_cancel_charge(page, memcg);
734 mem_cgroup_commit_charge(page, memcg, true);
737 page_cache_release(page);
742 * Move the page from the page cache to the swap cache.
744 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
746 struct shmem_inode_info *info;
747 struct address_space *mapping;
752 BUG_ON(!PageLocked(page));
753 mapping = page->mapping;
755 inode = mapping->host;
756 info = SHMEM_I(inode);
757 if (info->flags & VM_LOCKED)
759 if (!total_swap_pages)
763 * shmem_backing_dev_info's capabilities prevent regular writeback or
764 * sync from ever calling shmem_writepage; but a stacking filesystem
765 * might use ->writepage of its underlying filesystem, in which case
766 * tmpfs should write out to swap only in response to memory pressure,
767 * and not for the writeback threads or sync.
769 if (!wbc->for_reclaim) {
770 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
775 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
776 * value into swapfile.c, the only way we can correctly account for a
777 * fallocated page arriving here is now to initialize it and write it.
779 * That's okay for a page already fallocated earlier, but if we have
780 * not yet completed the fallocation, then (a) we want to keep track
781 * of this page in case we have to undo it, and (b) it may not be a
782 * good idea to continue anyway, once we're pushing into swap. So
783 * reactivate the page, and let shmem_fallocate() quit when too many.
785 if (!PageUptodate(page)) {
786 if (inode->i_private) {
787 struct shmem_falloc *shmem_falloc;
788 spin_lock(&inode->i_lock);
789 shmem_falloc = inode->i_private;
791 !shmem_falloc->waitq &&
792 index >= shmem_falloc->start &&
793 index < shmem_falloc->next)
794 shmem_falloc->nr_unswapped++;
797 spin_unlock(&inode->i_lock);
801 clear_highpage(page);
802 flush_dcache_page(page);
803 SetPageUptodate(page);
806 swap = get_swap_page();
811 * Add inode to shmem_unuse()'s list of swapped-out inodes,
812 * if it's not already there. Do it now before the page is
813 * moved to swap cache, when its pagelock no longer protects
814 * the inode from eviction. But don't unlock the mutex until
815 * we've incremented swapped, because shmem_unuse_inode() will
816 * prune a !swapped inode from the swaplist under this mutex.
818 mutex_lock(&shmem_swaplist_mutex);
819 if (list_empty(&info->swaplist))
820 list_add_tail(&info->swaplist, &shmem_swaplist);
822 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
823 swap_shmem_alloc(swap);
824 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
826 spin_lock(&info->lock);
828 shmem_recalc_inode(inode);
829 spin_unlock(&info->lock);
831 mutex_unlock(&shmem_swaplist_mutex);
832 BUG_ON(page_mapped(page));
833 swap_writepage(page, wbc);
837 mutex_unlock(&shmem_swaplist_mutex);
838 swapcache_free(swap, NULL);
840 set_page_dirty(page);
841 if (wbc->for_reclaim)
842 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
849 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
853 if (!mpol || mpol->mode == MPOL_DEFAULT)
854 return; /* show nothing */
856 mpol_to_str(buffer, sizeof(buffer), mpol);
858 seq_printf(seq, ",mpol=%s", buffer);
861 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
863 struct mempolicy *mpol = NULL;
865 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
868 spin_unlock(&sbinfo->stat_lock);
872 #endif /* CONFIG_TMPFS */
874 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
875 struct shmem_inode_info *info, pgoff_t index)
877 struct vm_area_struct pvma;
880 /* Create a pseudo vma that just contains the policy */
882 /* Bias interleave by inode number to distribute better across nodes */
883 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
885 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
887 page = swapin_readahead(swap, gfp, &pvma, 0);
889 /* Drop reference taken by mpol_shared_policy_lookup() */
890 mpol_cond_put(pvma.vm_policy);
895 static struct page *shmem_alloc_page(gfp_t gfp,
896 struct shmem_inode_info *info, pgoff_t index)
898 struct vm_area_struct pvma;
901 /* Create a pseudo vma that just contains the policy */
903 /* Bias interleave by inode number to distribute better across nodes */
904 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
906 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
908 page = alloc_page_vma(gfp, &pvma, 0);
910 /* Drop reference taken by mpol_shared_policy_lookup() */
911 mpol_cond_put(pvma.vm_policy);
915 #else /* !CONFIG_NUMA */
917 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
920 #endif /* CONFIG_TMPFS */
922 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
923 struct shmem_inode_info *info, pgoff_t index)
925 return swapin_readahead(swap, gfp, NULL, 0);
928 static inline struct page *shmem_alloc_page(gfp_t gfp,
929 struct shmem_inode_info *info, pgoff_t index)
931 return alloc_page(gfp);
933 #endif /* CONFIG_NUMA */
935 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
936 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
943 * When a page is moved from swapcache to shmem filecache (either by the
944 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
945 * shmem_unuse_inode()), it may have been read in earlier from swap, in
946 * ignorance of the mapping it belongs to. If that mapping has special
947 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
948 * we may need to copy to a suitable page before moving to filecache.
950 * In a future release, this may well be extended to respect cpuset and
951 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
952 * but for now it is a simple matter of zone.
954 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
956 return page_zonenum(page) > gfp_zone(gfp);
959 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
960 struct shmem_inode_info *info, pgoff_t index)
962 struct page *oldpage, *newpage;
963 struct address_space *swap_mapping;
968 swap_index = page_private(oldpage);
969 swap_mapping = page_mapping(oldpage);
972 * We have arrived here because our zones are constrained, so don't
973 * limit chance of success by further cpuset and node constraints.
975 gfp &= ~GFP_CONSTRAINT_MASK;
976 newpage = shmem_alloc_page(gfp, info, index);
980 page_cache_get(newpage);
981 copy_highpage(newpage, oldpage);
982 flush_dcache_page(newpage);
984 __set_page_locked(newpage);
985 SetPageUptodate(newpage);
986 SetPageSwapBacked(newpage);
987 set_page_private(newpage, swap_index);
988 SetPageSwapCache(newpage);
991 * Our caller will very soon move newpage out of swapcache, but it's
992 * a nice clean interface for us to replace oldpage by newpage there.
994 spin_lock_irq(&swap_mapping->tree_lock);
995 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
998 __inc_zone_page_state(newpage, NR_FILE_PAGES);
999 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1001 spin_unlock_irq(&swap_mapping->tree_lock);
1003 if (unlikely(error)) {
1005 * Is this possible? I think not, now that our callers check
1006 * both PageSwapCache and page_private after getting page lock;
1007 * but be defensive. Reverse old to newpage for clear and free.
1011 mem_cgroup_replace_page_cache(oldpage, newpage);
1012 lru_cache_add_anon(newpage);
1016 ClearPageSwapCache(oldpage);
1017 set_page_private(oldpage, 0);
1019 unlock_page(oldpage);
1020 page_cache_release(oldpage);
1021 page_cache_release(oldpage);
1026 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1028 * If we allocate a new one we do not mark it dirty. That's up to the
1029 * vm. If we swap it in we mark it dirty since we also free the swap
1030 * entry since a page cannot live in both the swap and page cache
1032 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1033 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1035 struct address_space *mapping = inode->i_mapping;
1036 struct shmem_inode_info *info;
1037 struct shmem_sb_info *sbinfo;
1038 struct mem_cgroup *memcg;
1045 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1049 page = find_lock_entry(mapping, index);
1050 if (radix_tree_exceptional_entry(page)) {
1051 swap = radix_to_swp_entry(page);
1055 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1056 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1061 if (page && sgp == SGP_WRITE)
1062 mark_page_accessed(page);
1064 /* fallocated page? */
1065 if (page && !PageUptodate(page)) {
1066 if (sgp != SGP_READ)
1069 page_cache_release(page);
1072 if (page || (sgp == SGP_READ && !swap.val)) {
1078 * Fast cache lookup did not find it:
1079 * bring it back from swap or allocate.
1081 info = SHMEM_I(inode);
1082 sbinfo = SHMEM_SB(inode->i_sb);
1085 /* Look it up and read it in.. */
1086 page = lookup_swap_cache(swap);
1088 /* here we actually do the io */
1090 *fault_type |= VM_FAULT_MAJOR;
1091 page = shmem_swapin(swap, gfp, info, index);
1098 /* We have to do this with page locked to prevent races */
1100 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1101 !shmem_confirm_swap(mapping, index, swap)) {
1102 error = -EEXIST; /* try again */
1105 if (!PageUptodate(page)) {
1109 wait_on_page_writeback(page);
1111 if (shmem_should_replace_page(page, gfp)) {
1112 error = shmem_replace_page(&page, gfp, info, index);
1117 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg);
1119 error = shmem_add_to_page_cache(page, mapping, index,
1120 swp_to_radix_entry(swap));
1122 * We already confirmed swap under page lock, and make
1123 * no memory allocation here, so usually no possibility
1124 * of error; but free_swap_and_cache() only trylocks a
1125 * page, so it is just possible that the entry has been
1126 * truncated or holepunched since swap was confirmed.
1127 * shmem_undo_range() will have done some of the
1128 * unaccounting, now delete_from_swap_cache() will do
1129 * the rest (including mem_cgroup_uncharge_swapcache).
1130 * Reset swap.val? No, leave it so "failed" goes back to
1131 * "repeat": reading a hole and writing should succeed.
1134 mem_cgroup_cancel_charge(page, memcg);
1135 delete_from_swap_cache(page);
1141 mem_cgroup_commit_charge(page, memcg, true);
1143 spin_lock(&info->lock);
1145 shmem_recalc_inode(inode);
1146 spin_unlock(&info->lock);
1148 if (sgp == SGP_WRITE)
1149 mark_page_accessed(page);
1151 delete_from_swap_cache(page);
1152 set_page_dirty(page);
1156 if (shmem_acct_block(info->flags)) {
1160 if (sbinfo->max_blocks) {
1161 if (percpu_counter_compare(&sbinfo->used_blocks,
1162 sbinfo->max_blocks) >= 0) {
1166 percpu_counter_inc(&sbinfo->used_blocks);
1169 page = shmem_alloc_page(gfp, info, index);
1175 __SetPageSwapBacked(page);
1176 __set_page_locked(page);
1177 if (sgp == SGP_WRITE)
1178 __SetPageReferenced(page);
1180 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg);
1183 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
1185 error = shmem_add_to_page_cache(page, mapping, index,
1187 radix_tree_preload_end();
1190 mem_cgroup_cancel_charge(page, memcg);
1193 mem_cgroup_commit_charge(page, memcg, false);
1194 lru_cache_add_anon(page);
1196 spin_lock(&info->lock);
1198 inode->i_blocks += BLOCKS_PER_PAGE;
1199 shmem_recalc_inode(inode);
1200 spin_unlock(&info->lock);
1204 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1206 if (sgp == SGP_FALLOC)
1210 * Let SGP_WRITE caller clear ends if write does not fill page;
1211 * but SGP_FALLOC on a page fallocated earlier must initialize
1212 * it now, lest undo on failure cancel our earlier guarantee.
1214 if (sgp != SGP_WRITE) {
1215 clear_highpage(page);
1216 flush_dcache_page(page);
1217 SetPageUptodate(page);
1219 if (sgp == SGP_DIRTY)
1220 set_page_dirty(page);
1223 /* Perhaps the file has been truncated since we checked */
1224 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1225 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1239 info = SHMEM_I(inode);
1240 ClearPageDirty(page);
1241 delete_from_page_cache(page);
1242 spin_lock(&info->lock);
1244 inode->i_blocks -= BLOCKS_PER_PAGE;
1245 spin_unlock(&info->lock);
1247 sbinfo = SHMEM_SB(inode->i_sb);
1248 if (sbinfo->max_blocks)
1249 percpu_counter_add(&sbinfo->used_blocks, -1);
1251 shmem_unacct_blocks(info->flags, 1);
1253 if (swap.val && error != -EINVAL &&
1254 !shmem_confirm_swap(mapping, index, swap))
1259 page_cache_release(page);
1261 if (error == -ENOSPC && !once++) {
1262 info = SHMEM_I(inode);
1263 spin_lock(&info->lock);
1264 shmem_recalc_inode(inode);
1265 spin_unlock(&info->lock);
1268 if (error == -EEXIST) /* from above or from radix_tree_insert */
1273 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1275 struct inode *inode = file_inode(vma->vm_file);
1277 int ret = VM_FAULT_LOCKED;
1280 * Trinity finds that probing a hole which tmpfs is punching can
1281 * prevent the hole-punch from ever completing: which in turn
1282 * locks writers out with its hold on i_mutex. So refrain from
1283 * faulting pages into the hole while it's being punched. Although
1284 * shmem_undo_range() does remove the additions, it may be unable to
1285 * keep up, as each new page needs its own unmap_mapping_range() call,
1286 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1288 * It does not matter if we sometimes reach this check just before the
1289 * hole-punch begins, so that one fault then races with the punch:
1290 * we just need to make racing faults a rare case.
1292 * The implementation below would be much simpler if we just used a
1293 * standard mutex or completion: but we cannot take i_mutex in fault,
1294 * and bloating every shmem inode for this unlikely case would be sad.
1296 if (unlikely(inode->i_private)) {
1297 struct shmem_falloc *shmem_falloc;
1299 spin_lock(&inode->i_lock);
1300 shmem_falloc = inode->i_private;
1302 shmem_falloc->waitq &&
1303 vmf->pgoff >= shmem_falloc->start &&
1304 vmf->pgoff < shmem_falloc->next) {
1305 wait_queue_head_t *shmem_falloc_waitq;
1306 DEFINE_WAIT(shmem_fault_wait);
1308 ret = VM_FAULT_NOPAGE;
1309 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1310 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1311 /* It's polite to up mmap_sem if we can */
1312 up_read(&vma->vm_mm->mmap_sem);
1313 ret = VM_FAULT_RETRY;
1316 shmem_falloc_waitq = shmem_falloc->waitq;
1317 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1318 TASK_UNINTERRUPTIBLE);
1319 spin_unlock(&inode->i_lock);
1323 * shmem_falloc_waitq points into the shmem_fallocate()
1324 * stack of the hole-punching task: shmem_falloc_waitq
1325 * is usually invalid by the time we reach here, but
1326 * finish_wait() does not dereference it in that case;
1327 * though i_lock needed lest racing with wake_up_all().
1329 spin_lock(&inode->i_lock);
1330 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1331 spin_unlock(&inode->i_lock);
1334 spin_unlock(&inode->i_lock);
1337 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1339 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1341 if (ret & VM_FAULT_MAJOR) {
1342 count_vm_event(PGMAJFAULT);
1343 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1349 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1351 struct inode *inode = file_inode(vma->vm_file);
1352 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1355 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1358 struct inode *inode = file_inode(vma->vm_file);
1361 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1362 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1366 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1368 struct inode *inode = file_inode(file);
1369 struct shmem_inode_info *info = SHMEM_I(inode);
1370 int retval = -ENOMEM;
1372 spin_lock(&info->lock);
1373 if (lock && !(info->flags & VM_LOCKED)) {
1374 if (!user_shm_lock(inode->i_size, user))
1376 info->flags |= VM_LOCKED;
1377 mapping_set_unevictable(file->f_mapping);
1379 if (!lock && (info->flags & VM_LOCKED) && user) {
1380 user_shm_unlock(inode->i_size, user);
1381 info->flags &= ~VM_LOCKED;
1382 mapping_clear_unevictable(file->f_mapping);
1387 spin_unlock(&info->lock);
1391 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1393 file_accessed(file);
1394 vma->vm_ops = &shmem_vm_ops;
1398 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1399 umode_t mode, dev_t dev, unsigned long flags)
1401 struct inode *inode;
1402 struct shmem_inode_info *info;
1403 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1405 if (shmem_reserve_inode(sb))
1408 inode = new_inode(sb);
1410 inode->i_ino = get_next_ino();
1411 inode_init_owner(inode, dir, mode);
1412 inode->i_blocks = 0;
1413 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1414 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1415 inode->i_generation = get_seconds();
1416 info = SHMEM_I(inode);
1417 memset(info, 0, (char *)inode - (char *)info);
1418 spin_lock_init(&info->lock);
1419 info->flags = flags & VM_NORESERVE;
1420 INIT_LIST_HEAD(&info->swaplist);
1421 simple_xattrs_init(&info->xattrs);
1422 cache_no_acl(inode);
1424 switch (mode & S_IFMT) {
1426 inode->i_op = &shmem_special_inode_operations;
1427 init_special_inode(inode, mode, dev);
1430 inode->i_mapping->a_ops = &shmem_aops;
1431 inode->i_op = &shmem_inode_operations;
1432 inode->i_fop = &shmem_file_operations;
1433 mpol_shared_policy_init(&info->policy,
1434 shmem_get_sbmpol(sbinfo));
1438 /* Some things misbehave if size == 0 on a directory */
1439 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1440 inode->i_op = &shmem_dir_inode_operations;
1441 inode->i_fop = &simple_dir_operations;
1445 * Must not load anything in the rbtree,
1446 * mpol_free_shared_policy will not be called.
1448 mpol_shared_policy_init(&info->policy, NULL);
1452 shmem_free_inode(sb);
1456 bool shmem_mapping(struct address_space *mapping)
1458 return mapping->backing_dev_info == &shmem_backing_dev_info;
1462 static const struct inode_operations shmem_symlink_inode_operations;
1463 static const struct inode_operations shmem_short_symlink_operations;
1465 #ifdef CONFIG_TMPFS_XATTR
1466 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1468 #define shmem_initxattrs NULL
1472 shmem_write_begin(struct file *file, struct address_space *mapping,
1473 loff_t pos, unsigned len, unsigned flags,
1474 struct page **pagep, void **fsdata)
1476 struct inode *inode = mapping->host;
1477 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1478 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1482 shmem_write_end(struct file *file, struct address_space *mapping,
1483 loff_t pos, unsigned len, unsigned copied,
1484 struct page *page, void *fsdata)
1486 struct inode *inode = mapping->host;
1488 if (pos + copied > inode->i_size)
1489 i_size_write(inode, pos + copied);
1491 if (!PageUptodate(page)) {
1492 if (copied < PAGE_CACHE_SIZE) {
1493 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1494 zero_user_segments(page, 0, from,
1495 from + copied, PAGE_CACHE_SIZE);
1497 SetPageUptodate(page);
1499 set_page_dirty(page);
1501 page_cache_release(page);
1506 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1508 struct file *file = iocb->ki_filp;
1509 struct inode *inode = file_inode(file);
1510 struct address_space *mapping = inode->i_mapping;
1512 unsigned long offset;
1513 enum sgp_type sgp = SGP_READ;
1516 loff_t *ppos = &iocb->ki_pos;
1519 * Might this read be for a stacking filesystem? Then when reading
1520 * holes of a sparse file, we actually need to allocate those pages,
1521 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1523 if (segment_eq(get_fs(), KERNEL_DS))
1526 index = *ppos >> PAGE_CACHE_SHIFT;
1527 offset = *ppos & ~PAGE_CACHE_MASK;
1530 struct page *page = NULL;
1532 unsigned long nr, ret;
1533 loff_t i_size = i_size_read(inode);
1535 end_index = i_size >> PAGE_CACHE_SHIFT;
1536 if (index > end_index)
1538 if (index == end_index) {
1539 nr = i_size & ~PAGE_CACHE_MASK;
1544 error = shmem_getpage(inode, index, &page, sgp, NULL);
1546 if (error == -EINVAL)
1554 * We must evaluate after, since reads (unlike writes)
1555 * are called without i_mutex protection against truncate
1557 nr = PAGE_CACHE_SIZE;
1558 i_size = i_size_read(inode);
1559 end_index = i_size >> PAGE_CACHE_SHIFT;
1560 if (index == end_index) {
1561 nr = i_size & ~PAGE_CACHE_MASK;
1564 page_cache_release(page);
1572 * If users can be writing to this page using arbitrary
1573 * virtual addresses, take care about potential aliasing
1574 * before reading the page on the kernel side.
1576 if (mapping_writably_mapped(mapping))
1577 flush_dcache_page(page);
1579 * Mark the page accessed if we read the beginning.
1582 mark_page_accessed(page);
1584 page = ZERO_PAGE(0);
1585 page_cache_get(page);
1589 * Ok, we have the page, and it's up-to-date, so
1590 * now we can copy it to user space...
1592 ret = copy_page_to_iter(page, offset, nr, to);
1595 index += offset >> PAGE_CACHE_SHIFT;
1596 offset &= ~PAGE_CACHE_MASK;
1598 page_cache_release(page);
1599 if (!iov_iter_count(to))
1608 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1609 file_accessed(file);
1610 return retval ? retval : error;
1613 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1614 struct pipe_inode_info *pipe, size_t len,
1617 struct address_space *mapping = in->f_mapping;
1618 struct inode *inode = mapping->host;
1619 unsigned int loff, nr_pages, req_pages;
1620 struct page *pages[PIPE_DEF_BUFFERS];
1621 struct partial_page partial[PIPE_DEF_BUFFERS];
1623 pgoff_t index, end_index;
1626 struct splice_pipe_desc spd = {
1629 .nr_pages_max = PIPE_DEF_BUFFERS,
1631 .ops = &page_cache_pipe_buf_ops,
1632 .spd_release = spd_release_page,
1635 isize = i_size_read(inode);
1636 if (unlikely(*ppos >= isize))
1639 left = isize - *ppos;
1640 if (unlikely(left < len))
1643 if (splice_grow_spd(pipe, &spd))
1646 index = *ppos >> PAGE_CACHE_SHIFT;
1647 loff = *ppos & ~PAGE_CACHE_MASK;
1648 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1649 nr_pages = min(req_pages, spd.nr_pages_max);
1651 spd.nr_pages = find_get_pages_contig(mapping, index,
1652 nr_pages, spd.pages);
1653 index += spd.nr_pages;
1656 while (spd.nr_pages < nr_pages) {
1657 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1661 spd.pages[spd.nr_pages++] = page;
1665 index = *ppos >> PAGE_CACHE_SHIFT;
1666 nr_pages = spd.nr_pages;
1669 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1670 unsigned int this_len;
1675 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1676 page = spd.pages[page_nr];
1678 if (!PageUptodate(page) || page->mapping != mapping) {
1679 error = shmem_getpage(inode, index, &page,
1684 page_cache_release(spd.pages[page_nr]);
1685 spd.pages[page_nr] = page;
1688 isize = i_size_read(inode);
1689 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1690 if (unlikely(!isize || index > end_index))
1693 if (end_index == index) {
1696 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1700 this_len = min(this_len, plen - loff);
1704 spd.partial[page_nr].offset = loff;
1705 spd.partial[page_nr].len = this_len;
1712 while (page_nr < nr_pages)
1713 page_cache_release(spd.pages[page_nr++]);
1716 error = splice_to_pipe(pipe, &spd);
1718 splice_shrink_spd(&spd);
1728 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1730 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1731 pgoff_t index, pgoff_t end, int whence)
1734 struct pagevec pvec;
1735 pgoff_t indices[PAGEVEC_SIZE];
1739 pagevec_init(&pvec, 0);
1740 pvec.nr = 1; /* start small: we may be there already */
1742 pvec.nr = find_get_entries(mapping, index,
1743 pvec.nr, pvec.pages, indices);
1745 if (whence == SEEK_DATA)
1749 for (i = 0; i < pvec.nr; i++, index++) {
1750 if (index < indices[i]) {
1751 if (whence == SEEK_HOLE) {
1757 page = pvec.pages[i];
1758 if (page && !radix_tree_exceptional_entry(page)) {
1759 if (!PageUptodate(page))
1763 (page && whence == SEEK_DATA) ||
1764 (!page && whence == SEEK_HOLE)) {
1769 pagevec_remove_exceptionals(&pvec);
1770 pagevec_release(&pvec);
1771 pvec.nr = PAGEVEC_SIZE;
1777 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
1779 struct address_space *mapping = file->f_mapping;
1780 struct inode *inode = mapping->host;
1784 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1785 return generic_file_llseek_size(file, offset, whence,
1786 MAX_LFS_FILESIZE, i_size_read(inode));
1787 mutex_lock(&inode->i_mutex);
1788 /* We're holding i_mutex so we can access i_size directly */
1792 else if (offset >= inode->i_size)
1795 start = offset >> PAGE_CACHE_SHIFT;
1796 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1797 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
1798 new_offset <<= PAGE_CACHE_SHIFT;
1799 if (new_offset > offset) {
1800 if (new_offset < inode->i_size)
1801 offset = new_offset;
1802 else if (whence == SEEK_DATA)
1805 offset = inode->i_size;
1810 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
1811 mutex_unlock(&inode->i_mutex);
1815 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1818 struct inode *inode = file_inode(file);
1819 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1820 struct shmem_falloc shmem_falloc;
1821 pgoff_t start, index, end;
1824 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1827 mutex_lock(&inode->i_mutex);
1829 if (mode & FALLOC_FL_PUNCH_HOLE) {
1830 struct address_space *mapping = file->f_mapping;
1831 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1832 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1833 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
1835 shmem_falloc.waitq = &shmem_falloc_waitq;
1836 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
1837 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
1838 spin_lock(&inode->i_lock);
1839 inode->i_private = &shmem_falloc;
1840 spin_unlock(&inode->i_lock);
1842 if ((u64)unmap_end > (u64)unmap_start)
1843 unmap_mapping_range(mapping, unmap_start,
1844 1 + unmap_end - unmap_start, 0);
1845 shmem_truncate_range(inode, offset, offset + len - 1);
1846 /* No need to unmap again: hole-punching leaves COWed pages */
1848 spin_lock(&inode->i_lock);
1849 inode->i_private = NULL;
1850 wake_up_all(&shmem_falloc_waitq);
1851 spin_unlock(&inode->i_lock);
1856 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1857 error = inode_newsize_ok(inode, offset + len);
1861 start = offset >> PAGE_CACHE_SHIFT;
1862 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1863 /* Try to avoid a swapstorm if len is impossible to satisfy */
1864 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1869 shmem_falloc.waitq = NULL;
1870 shmem_falloc.start = start;
1871 shmem_falloc.next = start;
1872 shmem_falloc.nr_falloced = 0;
1873 shmem_falloc.nr_unswapped = 0;
1874 spin_lock(&inode->i_lock);
1875 inode->i_private = &shmem_falloc;
1876 spin_unlock(&inode->i_lock);
1878 for (index = start; index < end; index++) {
1882 * Good, the fallocate(2) manpage permits EINTR: we may have
1883 * been interrupted because we are using up too much memory.
1885 if (signal_pending(current))
1887 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1890 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1893 /* Remove the !PageUptodate pages we added */
1894 shmem_undo_range(inode,
1895 (loff_t)start << PAGE_CACHE_SHIFT,
1896 (loff_t)index << PAGE_CACHE_SHIFT, true);
1901 * Inform shmem_writepage() how far we have reached.
1902 * No need for lock or barrier: we have the page lock.
1904 shmem_falloc.next++;
1905 if (!PageUptodate(page))
1906 shmem_falloc.nr_falloced++;
1909 * If !PageUptodate, leave it that way so that freeable pages
1910 * can be recognized if we need to rollback on error later.
1911 * But set_page_dirty so that memory pressure will swap rather
1912 * than free the pages we are allocating (and SGP_CACHE pages
1913 * might still be clean: we now need to mark those dirty too).
1915 set_page_dirty(page);
1917 page_cache_release(page);
1921 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1922 i_size_write(inode, offset + len);
1923 inode->i_ctime = CURRENT_TIME;
1925 spin_lock(&inode->i_lock);
1926 inode->i_private = NULL;
1927 spin_unlock(&inode->i_lock);
1929 mutex_unlock(&inode->i_mutex);
1933 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1935 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1937 buf->f_type = TMPFS_MAGIC;
1938 buf->f_bsize = PAGE_CACHE_SIZE;
1939 buf->f_namelen = NAME_MAX;
1940 if (sbinfo->max_blocks) {
1941 buf->f_blocks = sbinfo->max_blocks;
1943 buf->f_bfree = sbinfo->max_blocks -
1944 percpu_counter_sum(&sbinfo->used_blocks);
1946 if (sbinfo->max_inodes) {
1947 buf->f_files = sbinfo->max_inodes;
1948 buf->f_ffree = sbinfo->free_inodes;
1950 /* else leave those fields 0 like simple_statfs */
1955 * File creation. Allocate an inode, and we're done..
1958 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1960 struct inode *inode;
1961 int error = -ENOSPC;
1963 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1965 error = simple_acl_create(dir, inode);
1968 error = security_inode_init_security(inode, dir,
1970 shmem_initxattrs, NULL);
1971 if (error && error != -EOPNOTSUPP)
1975 dir->i_size += BOGO_DIRENT_SIZE;
1976 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1977 d_instantiate(dentry, inode);
1978 dget(dentry); /* Extra count - pin the dentry in core */
1987 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
1989 struct inode *inode;
1990 int error = -ENOSPC;
1992 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
1994 error = security_inode_init_security(inode, dir,
1996 shmem_initxattrs, NULL);
1997 if (error && error != -EOPNOTSUPP)
1999 error = simple_acl_create(dir, inode);
2002 d_tmpfile(dentry, inode);
2010 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2014 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2020 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2023 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2029 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2031 struct inode *inode = old_dentry->d_inode;
2035 * No ordinary (disk based) filesystem counts links as inodes;
2036 * but each new link needs a new dentry, pinning lowmem, and
2037 * tmpfs dentries cannot be pruned until they are unlinked.
2039 ret = shmem_reserve_inode(inode->i_sb);
2043 dir->i_size += BOGO_DIRENT_SIZE;
2044 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2046 ihold(inode); /* New dentry reference */
2047 dget(dentry); /* Extra pinning count for the created dentry */
2048 d_instantiate(dentry, inode);
2053 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2055 struct inode *inode = dentry->d_inode;
2057 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2058 shmem_free_inode(inode->i_sb);
2060 dir->i_size -= BOGO_DIRENT_SIZE;
2061 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2063 dput(dentry); /* Undo the count from "create" - this does all the work */
2067 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2069 if (!simple_empty(dentry))
2072 drop_nlink(dentry->d_inode);
2074 return shmem_unlink(dir, dentry);
2078 * The VFS layer already does all the dentry stuff for rename,
2079 * we just have to decrement the usage count for the target if
2080 * it exists so that the VFS layer correctly free's it when it
2083 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2085 struct inode *inode = old_dentry->d_inode;
2086 int they_are_dirs = S_ISDIR(inode->i_mode);
2088 if (!simple_empty(new_dentry))
2091 if (new_dentry->d_inode) {
2092 (void) shmem_unlink(new_dir, new_dentry);
2094 drop_nlink(old_dir);
2095 } else if (they_are_dirs) {
2096 drop_nlink(old_dir);
2100 old_dir->i_size -= BOGO_DIRENT_SIZE;
2101 new_dir->i_size += BOGO_DIRENT_SIZE;
2102 old_dir->i_ctime = old_dir->i_mtime =
2103 new_dir->i_ctime = new_dir->i_mtime =
2104 inode->i_ctime = CURRENT_TIME;
2108 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2112 struct inode *inode;
2115 struct shmem_inode_info *info;
2117 len = strlen(symname) + 1;
2118 if (len > PAGE_CACHE_SIZE)
2119 return -ENAMETOOLONG;
2121 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2125 error = security_inode_init_security(inode, dir, &dentry->d_name,
2126 shmem_initxattrs, NULL);
2128 if (error != -EOPNOTSUPP) {
2135 info = SHMEM_I(inode);
2136 inode->i_size = len-1;
2137 if (len <= SHORT_SYMLINK_LEN) {
2138 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2139 if (!info->symlink) {
2143 inode->i_op = &shmem_short_symlink_operations;
2145 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2150 inode->i_mapping->a_ops = &shmem_aops;
2151 inode->i_op = &shmem_symlink_inode_operations;
2152 kaddr = kmap_atomic(page);
2153 memcpy(kaddr, symname, len);
2154 kunmap_atomic(kaddr);
2155 SetPageUptodate(page);
2156 set_page_dirty(page);
2158 page_cache_release(page);
2160 dir->i_size += BOGO_DIRENT_SIZE;
2161 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2162 d_instantiate(dentry, inode);
2167 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2169 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2173 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2175 struct page *page = NULL;
2176 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2177 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2183 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2185 if (!IS_ERR(nd_get_link(nd))) {
2186 struct page *page = cookie;
2188 mark_page_accessed(page);
2189 page_cache_release(page);
2193 #ifdef CONFIG_TMPFS_XATTR
2195 * Superblocks without xattr inode operations may get some security.* xattr
2196 * support from the LSM "for free". As soon as we have any other xattrs
2197 * like ACLs, we also need to implement the security.* handlers at
2198 * filesystem level, though.
2202 * Callback for security_inode_init_security() for acquiring xattrs.
2204 static int shmem_initxattrs(struct inode *inode,
2205 const struct xattr *xattr_array,
2208 struct shmem_inode_info *info = SHMEM_I(inode);
2209 const struct xattr *xattr;
2210 struct simple_xattr *new_xattr;
2213 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2214 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
2218 len = strlen(xattr->name) + 1;
2219 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2221 if (!new_xattr->name) {
2226 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2227 XATTR_SECURITY_PREFIX_LEN);
2228 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2231 simple_xattr_list_add(&info->xattrs, new_xattr);
2237 static const struct xattr_handler *shmem_xattr_handlers[] = {
2238 #ifdef CONFIG_TMPFS_POSIX_ACL
2239 &posix_acl_access_xattr_handler,
2240 &posix_acl_default_xattr_handler,
2245 static int shmem_xattr_validate(const char *name)
2247 struct { const char *prefix; size_t len; } arr[] = {
2248 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2249 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2253 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2254 size_t preflen = arr[i].len;
2255 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2264 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2265 void *buffer, size_t size)
2267 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2271 * If this is a request for a synthetic attribute in the system.*
2272 * namespace use the generic infrastructure to resolve a handler
2273 * for it via sb->s_xattr.
2275 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2276 return generic_getxattr(dentry, name, buffer, size);
2278 err = shmem_xattr_validate(name);
2282 return simple_xattr_get(&info->xattrs, name, buffer, size);
2285 static int shmem_setxattr(struct dentry *dentry, const char *name,
2286 const void *value, size_t size, int flags)
2288 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2292 * If this is a request for a synthetic attribute in the system.*
2293 * namespace use the generic infrastructure to resolve a handler
2294 * for it via sb->s_xattr.
2296 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2297 return generic_setxattr(dentry, name, value, size, flags);
2299 err = shmem_xattr_validate(name);
2303 return simple_xattr_set(&info->xattrs, name, value, size, flags);
2306 static int shmem_removexattr(struct dentry *dentry, const char *name)
2308 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2312 * If this is a request for a synthetic attribute in the system.*
2313 * namespace use the generic infrastructure to resolve a handler
2314 * for it via sb->s_xattr.
2316 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2317 return generic_removexattr(dentry, name);
2319 err = shmem_xattr_validate(name);
2323 return simple_xattr_remove(&info->xattrs, name);
2326 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2328 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2329 return simple_xattr_list(&info->xattrs, buffer, size);
2331 #endif /* CONFIG_TMPFS_XATTR */
2333 static const struct inode_operations shmem_short_symlink_operations = {
2334 .readlink = generic_readlink,
2335 .follow_link = shmem_follow_short_symlink,
2336 #ifdef CONFIG_TMPFS_XATTR
2337 .setxattr = shmem_setxattr,
2338 .getxattr = shmem_getxattr,
2339 .listxattr = shmem_listxattr,
2340 .removexattr = shmem_removexattr,
2344 static const struct inode_operations shmem_symlink_inode_operations = {
2345 .readlink = generic_readlink,
2346 .follow_link = shmem_follow_link,
2347 .put_link = shmem_put_link,
2348 #ifdef CONFIG_TMPFS_XATTR
2349 .setxattr = shmem_setxattr,
2350 .getxattr = shmem_getxattr,
2351 .listxattr = shmem_listxattr,
2352 .removexattr = shmem_removexattr,
2356 static struct dentry *shmem_get_parent(struct dentry *child)
2358 return ERR_PTR(-ESTALE);
2361 static int shmem_match(struct inode *ino, void *vfh)
2365 inum = (inum << 32) | fh[1];
2366 return ino->i_ino == inum && fh[0] == ino->i_generation;
2369 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2370 struct fid *fid, int fh_len, int fh_type)
2372 struct inode *inode;
2373 struct dentry *dentry = NULL;
2380 inum = (inum << 32) | fid->raw[1];
2382 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2383 shmem_match, fid->raw);
2385 dentry = d_find_alias(inode);
2392 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2393 struct inode *parent)
2397 return FILEID_INVALID;
2400 if (inode_unhashed(inode)) {
2401 /* Unfortunately insert_inode_hash is not idempotent,
2402 * so as we hash inodes here rather than at creation
2403 * time, we need a lock to ensure we only try
2406 static DEFINE_SPINLOCK(lock);
2408 if (inode_unhashed(inode))
2409 __insert_inode_hash(inode,
2410 inode->i_ino + inode->i_generation);
2414 fh[0] = inode->i_generation;
2415 fh[1] = inode->i_ino;
2416 fh[2] = ((__u64)inode->i_ino) >> 32;
2422 static const struct export_operations shmem_export_ops = {
2423 .get_parent = shmem_get_parent,
2424 .encode_fh = shmem_encode_fh,
2425 .fh_to_dentry = shmem_fh_to_dentry,
2428 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2431 char *this_char, *value, *rest;
2432 struct mempolicy *mpol = NULL;
2436 while (options != NULL) {
2437 this_char = options;
2440 * NUL-terminate this option: unfortunately,
2441 * mount options form a comma-separated list,
2442 * but mpol's nodelist may also contain commas.
2444 options = strchr(options, ',');
2445 if (options == NULL)
2448 if (!isdigit(*options)) {
2455 if ((value = strchr(this_char,'=')) != NULL) {
2459 "tmpfs: No value for mount option '%s'\n",
2464 if (!strcmp(this_char,"size")) {
2465 unsigned long long size;
2466 size = memparse(value,&rest);
2468 size <<= PAGE_SHIFT;
2469 size *= totalram_pages;
2475 sbinfo->max_blocks =
2476 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2477 } else if (!strcmp(this_char,"nr_blocks")) {
2478 sbinfo->max_blocks = memparse(value, &rest);
2481 } else if (!strcmp(this_char,"nr_inodes")) {
2482 sbinfo->max_inodes = memparse(value, &rest);
2485 } else if (!strcmp(this_char,"mode")) {
2488 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2491 } else if (!strcmp(this_char,"uid")) {
2494 uid = simple_strtoul(value, &rest, 0);
2497 sbinfo->uid = make_kuid(current_user_ns(), uid);
2498 if (!uid_valid(sbinfo->uid))
2500 } else if (!strcmp(this_char,"gid")) {
2503 gid = simple_strtoul(value, &rest, 0);
2506 sbinfo->gid = make_kgid(current_user_ns(), gid);
2507 if (!gid_valid(sbinfo->gid))
2509 } else if (!strcmp(this_char,"mpol")) {
2512 if (mpol_parse_str(value, &mpol))
2515 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2520 sbinfo->mpol = mpol;
2524 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2532 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2534 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2535 struct shmem_sb_info config = *sbinfo;
2536 unsigned long inodes;
2537 int error = -EINVAL;
2540 if (shmem_parse_options(data, &config, true))
2543 spin_lock(&sbinfo->stat_lock);
2544 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2545 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2547 if (config.max_inodes < inodes)
2550 * Those tests disallow limited->unlimited while any are in use;
2551 * but we must separately disallow unlimited->limited, because
2552 * in that case we have no record of how much is already in use.
2554 if (config.max_blocks && !sbinfo->max_blocks)
2556 if (config.max_inodes && !sbinfo->max_inodes)
2560 sbinfo->max_blocks = config.max_blocks;
2561 sbinfo->max_inodes = config.max_inodes;
2562 sbinfo->free_inodes = config.max_inodes - inodes;
2565 * Preserve previous mempolicy unless mpol remount option was specified.
2568 mpol_put(sbinfo->mpol);
2569 sbinfo->mpol = config.mpol; /* transfers initial ref */
2572 spin_unlock(&sbinfo->stat_lock);
2576 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2578 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2580 if (sbinfo->max_blocks != shmem_default_max_blocks())
2581 seq_printf(seq, ",size=%luk",
2582 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2583 if (sbinfo->max_inodes != shmem_default_max_inodes())
2584 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2585 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2586 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2587 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2588 seq_printf(seq, ",uid=%u",
2589 from_kuid_munged(&init_user_ns, sbinfo->uid));
2590 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2591 seq_printf(seq, ",gid=%u",
2592 from_kgid_munged(&init_user_ns, sbinfo->gid));
2593 shmem_show_mpol(seq, sbinfo->mpol);
2596 #endif /* CONFIG_TMPFS */
2598 static void shmem_put_super(struct super_block *sb)
2600 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2602 percpu_counter_destroy(&sbinfo->used_blocks);
2603 mpol_put(sbinfo->mpol);
2605 sb->s_fs_info = NULL;
2608 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2610 struct inode *inode;
2611 struct shmem_sb_info *sbinfo;
2614 /* Round up to L1_CACHE_BYTES to resist false sharing */
2615 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2616 L1_CACHE_BYTES), GFP_KERNEL);
2620 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2621 sbinfo->uid = current_fsuid();
2622 sbinfo->gid = current_fsgid();
2623 sb->s_fs_info = sbinfo;
2627 * Per default we only allow half of the physical ram per
2628 * tmpfs instance, limiting inodes to one per page of lowmem;
2629 * but the internal instance is left unlimited.
2631 if (!(sb->s_flags & MS_KERNMOUNT)) {
2632 sbinfo->max_blocks = shmem_default_max_blocks();
2633 sbinfo->max_inodes = shmem_default_max_inodes();
2634 if (shmem_parse_options(data, sbinfo, false)) {
2639 sb->s_flags |= MS_NOUSER;
2641 sb->s_export_op = &shmem_export_ops;
2642 sb->s_flags |= MS_NOSEC;
2644 sb->s_flags |= MS_NOUSER;
2647 spin_lock_init(&sbinfo->stat_lock);
2648 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2650 sbinfo->free_inodes = sbinfo->max_inodes;
2652 sb->s_maxbytes = MAX_LFS_FILESIZE;
2653 sb->s_blocksize = PAGE_CACHE_SIZE;
2654 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2655 sb->s_magic = TMPFS_MAGIC;
2656 sb->s_op = &shmem_ops;
2657 sb->s_time_gran = 1;
2658 #ifdef CONFIG_TMPFS_XATTR
2659 sb->s_xattr = shmem_xattr_handlers;
2661 #ifdef CONFIG_TMPFS_POSIX_ACL
2662 sb->s_flags |= MS_POSIXACL;
2665 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2668 inode->i_uid = sbinfo->uid;
2669 inode->i_gid = sbinfo->gid;
2670 sb->s_root = d_make_root(inode);
2676 shmem_put_super(sb);
2680 static struct kmem_cache *shmem_inode_cachep;
2682 static struct inode *shmem_alloc_inode(struct super_block *sb)
2684 struct shmem_inode_info *info;
2685 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2688 return &info->vfs_inode;
2691 static void shmem_destroy_callback(struct rcu_head *head)
2693 struct inode *inode = container_of(head, struct inode, i_rcu);
2694 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2697 static void shmem_destroy_inode(struct inode *inode)
2699 if (S_ISREG(inode->i_mode))
2700 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2701 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2704 static void shmem_init_inode(void *foo)
2706 struct shmem_inode_info *info = foo;
2707 inode_init_once(&info->vfs_inode);
2710 static int shmem_init_inodecache(void)
2712 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2713 sizeof(struct shmem_inode_info),
2714 0, SLAB_PANIC, shmem_init_inode);
2718 static void shmem_destroy_inodecache(void)
2720 kmem_cache_destroy(shmem_inode_cachep);
2723 static const struct address_space_operations shmem_aops = {
2724 .writepage = shmem_writepage,
2725 .set_page_dirty = __set_page_dirty_no_writeback,
2727 .write_begin = shmem_write_begin,
2728 .write_end = shmem_write_end,
2730 .migratepage = migrate_page,
2731 .error_remove_page = generic_error_remove_page,
2734 static const struct file_operations shmem_file_operations = {
2737 .llseek = shmem_file_llseek,
2738 .read = new_sync_read,
2739 .write = new_sync_write,
2740 .read_iter = shmem_file_read_iter,
2741 .write_iter = generic_file_write_iter,
2742 .fsync = noop_fsync,
2743 .splice_read = shmem_file_splice_read,
2744 .splice_write = iter_file_splice_write,
2745 .fallocate = shmem_fallocate,
2749 static const struct inode_operations shmem_inode_operations = {
2750 .setattr = shmem_setattr,
2751 #ifdef CONFIG_TMPFS_XATTR
2752 .setxattr = shmem_setxattr,
2753 .getxattr = shmem_getxattr,
2754 .listxattr = shmem_listxattr,
2755 .removexattr = shmem_removexattr,
2756 .set_acl = simple_set_acl,
2760 static const struct inode_operations shmem_dir_inode_operations = {
2762 .create = shmem_create,
2763 .lookup = simple_lookup,
2765 .unlink = shmem_unlink,
2766 .symlink = shmem_symlink,
2767 .mkdir = shmem_mkdir,
2768 .rmdir = shmem_rmdir,
2769 .mknod = shmem_mknod,
2770 .rename = shmem_rename,
2771 .tmpfile = shmem_tmpfile,
2773 #ifdef CONFIG_TMPFS_XATTR
2774 .setxattr = shmem_setxattr,
2775 .getxattr = shmem_getxattr,
2776 .listxattr = shmem_listxattr,
2777 .removexattr = shmem_removexattr,
2779 #ifdef CONFIG_TMPFS_POSIX_ACL
2780 .setattr = shmem_setattr,
2781 .set_acl = simple_set_acl,
2785 static const struct inode_operations shmem_special_inode_operations = {
2786 #ifdef CONFIG_TMPFS_XATTR
2787 .setxattr = shmem_setxattr,
2788 .getxattr = shmem_getxattr,
2789 .listxattr = shmem_listxattr,
2790 .removexattr = shmem_removexattr,
2792 #ifdef CONFIG_TMPFS_POSIX_ACL
2793 .setattr = shmem_setattr,
2794 .set_acl = simple_set_acl,
2798 static const struct super_operations shmem_ops = {
2799 .alloc_inode = shmem_alloc_inode,
2800 .destroy_inode = shmem_destroy_inode,
2802 .statfs = shmem_statfs,
2803 .remount_fs = shmem_remount_fs,
2804 .show_options = shmem_show_options,
2806 .evict_inode = shmem_evict_inode,
2807 .drop_inode = generic_delete_inode,
2808 .put_super = shmem_put_super,
2811 static const struct vm_operations_struct shmem_vm_ops = {
2812 .fault = shmem_fault,
2813 .map_pages = filemap_map_pages,
2815 .set_policy = shmem_set_policy,
2816 .get_policy = shmem_get_policy,
2818 .remap_pages = generic_file_remap_pages,
2821 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2822 int flags, const char *dev_name, void *data)
2824 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2827 static struct file_system_type shmem_fs_type = {
2828 .owner = THIS_MODULE,
2830 .mount = shmem_mount,
2831 .kill_sb = kill_litter_super,
2832 .fs_flags = FS_USERNS_MOUNT,
2835 int __init shmem_init(void)
2839 /* If rootfs called this, don't re-init */
2840 if (shmem_inode_cachep)
2843 error = bdi_init(&shmem_backing_dev_info);
2847 error = shmem_init_inodecache();
2851 error = register_filesystem(&shmem_fs_type);
2853 printk(KERN_ERR "Could not register tmpfs\n");
2857 shm_mnt = kern_mount(&shmem_fs_type);
2858 if (IS_ERR(shm_mnt)) {
2859 error = PTR_ERR(shm_mnt);
2860 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2866 unregister_filesystem(&shmem_fs_type);
2868 shmem_destroy_inodecache();
2870 bdi_destroy(&shmem_backing_dev_info);
2872 shm_mnt = ERR_PTR(error);
2876 #else /* !CONFIG_SHMEM */
2879 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2881 * This is intended for small system where the benefits of the full
2882 * shmem code (swap-backed and resource-limited) are outweighed by
2883 * their complexity. On systems without swap this code should be
2884 * effectively equivalent, but much lighter weight.
2887 static struct file_system_type shmem_fs_type = {
2889 .mount = ramfs_mount,
2890 .kill_sb = kill_litter_super,
2891 .fs_flags = FS_USERNS_MOUNT,
2894 int __init shmem_init(void)
2896 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2898 shm_mnt = kern_mount(&shmem_fs_type);
2899 BUG_ON(IS_ERR(shm_mnt));
2904 int shmem_unuse(swp_entry_t swap, struct page *page)
2909 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2914 void shmem_unlock_mapping(struct address_space *mapping)
2918 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2920 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2922 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2924 #define shmem_vm_ops generic_file_vm_ops
2925 #define shmem_file_operations ramfs_file_operations
2926 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2927 #define shmem_acct_size(flags, size) 0
2928 #define shmem_unacct_size(flags, size) do {} while (0)
2930 #endif /* CONFIG_SHMEM */
2934 static struct dentry_operations anon_ops = {
2935 .d_dname = simple_dname
2938 static struct file *__shmem_file_setup(const char *name, loff_t size,
2939 unsigned long flags, unsigned int i_flags)
2942 struct inode *inode;
2944 struct super_block *sb;
2947 if (IS_ERR(shm_mnt))
2948 return ERR_CAST(shm_mnt);
2950 if (size < 0 || size > MAX_LFS_FILESIZE)
2951 return ERR_PTR(-EINVAL);
2953 if (shmem_acct_size(flags, size))
2954 return ERR_PTR(-ENOMEM);
2956 res = ERR_PTR(-ENOMEM);
2958 this.len = strlen(name);
2959 this.hash = 0; /* will go */
2960 sb = shm_mnt->mnt_sb;
2961 path.mnt = mntget(shm_mnt);
2962 path.dentry = d_alloc_pseudo(sb, &this);
2965 d_set_d_op(path.dentry, &anon_ops);
2967 res = ERR_PTR(-ENOSPC);
2968 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2972 inode->i_flags |= i_flags;
2973 d_instantiate(path.dentry, inode);
2974 inode->i_size = size;
2975 clear_nlink(inode); /* It is unlinked */
2976 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
2980 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2981 &shmem_file_operations);
2988 shmem_unacct_size(flags, size);
2995 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
2996 * kernel internal. There will be NO LSM permission checks against the
2997 * underlying inode. So users of this interface must do LSM checks at a
2998 * higher layer. The one user is the big_key implementation. LSM checks
2999 * are provided at the key level rather than the inode level.
3000 * @name: name for dentry (to be seen in /proc/<pid>/maps
3001 * @size: size to be set for the file
3002 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3004 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
3006 return __shmem_file_setup(name, size, flags, S_PRIVATE);
3010 * shmem_file_setup - get an unlinked file living in tmpfs
3011 * @name: name for dentry (to be seen in /proc/<pid>/maps
3012 * @size: size to be set for the file
3013 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3015 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3017 return __shmem_file_setup(name, size, flags, 0);
3019 EXPORT_SYMBOL_GPL(shmem_file_setup);
3022 * shmem_zero_setup - setup a shared anonymous mapping
3023 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3025 int shmem_zero_setup(struct vm_area_struct *vma)
3028 loff_t size = vma->vm_end - vma->vm_start;
3030 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3032 return PTR_ERR(file);
3036 vma->vm_file = file;
3037 vma->vm_ops = &shmem_vm_ops;
3042 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3043 * @mapping: the page's address_space
3044 * @index: the page index
3045 * @gfp: the page allocator flags to use if allocating
3047 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3048 * with any new page allocations done using the specified allocation flags.
3049 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3050 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3051 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3053 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3054 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3056 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3057 pgoff_t index, gfp_t gfp)
3060 struct inode *inode = mapping->host;
3064 BUG_ON(mapping->a_ops != &shmem_aops);
3065 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3067 page = ERR_PTR(error);
3073 * The tiny !SHMEM case uses ramfs without swap
3075 return read_cache_page_gfp(mapping, index, gfp);
3078 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);