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 and shmem_writepage communicate via inode->i_private
84 * (with i_mutex making sure that it has only one user at a time):
85 * we would prefer not to enlarge the shmem inode just for that.
88 pgoff_t start; /* start of range currently being fallocated */
89 pgoff_t next; /* the next page offset to be fallocated */
90 pgoff_t nr_falloced; /* how many new pages have been fallocated */
91 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
94 /* Flag allocation requirements to shmem_getpage */
96 SGP_READ, /* don't exceed i_size, don't allocate page */
97 SGP_CACHE, /* don't exceed i_size, may allocate page */
98 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
99 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
100 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
104 static unsigned long shmem_default_max_blocks(void)
106 return totalram_pages / 2;
109 static unsigned long shmem_default_max_inodes(void)
111 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
115 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
116 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
117 struct shmem_inode_info *info, pgoff_t index);
118 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
119 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
121 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
122 struct page **pagep, enum sgp_type sgp, int *fault_type)
124 return shmem_getpage_gfp(inode, index, pagep, sgp,
125 mapping_gfp_mask(inode->i_mapping), fault_type);
128 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
130 return sb->s_fs_info;
134 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
135 * for shared memory and for shared anonymous (/dev/zero) mappings
136 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
137 * consistent with the pre-accounting of private mappings ...
139 static inline int shmem_acct_size(unsigned long flags, loff_t size)
141 return (flags & VM_NORESERVE) ?
142 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
145 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
147 if (!(flags & VM_NORESERVE))
148 vm_unacct_memory(VM_ACCT(size));
152 * ... whereas tmpfs objects are accounted incrementally as
153 * pages are allocated, in order to allow huge sparse files.
154 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
155 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
157 static inline int shmem_acct_block(unsigned long flags)
159 return (flags & VM_NORESERVE) ?
160 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
163 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
165 if (flags & VM_NORESERVE)
166 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
169 static const struct super_operations shmem_ops;
170 static const struct address_space_operations shmem_aops;
171 static const struct file_operations shmem_file_operations;
172 static const struct inode_operations shmem_inode_operations;
173 static const struct inode_operations shmem_dir_inode_operations;
174 static const struct inode_operations shmem_special_inode_operations;
175 static const struct vm_operations_struct shmem_vm_ops;
177 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
178 .ra_pages = 0, /* No readahead */
179 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
182 static LIST_HEAD(shmem_swaplist);
183 static DEFINE_MUTEX(shmem_swaplist_mutex);
185 static int shmem_reserve_inode(struct super_block *sb)
187 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
188 if (sbinfo->max_inodes) {
189 spin_lock(&sbinfo->stat_lock);
190 if (!sbinfo->free_inodes) {
191 spin_unlock(&sbinfo->stat_lock);
194 sbinfo->free_inodes--;
195 spin_unlock(&sbinfo->stat_lock);
200 static void shmem_free_inode(struct super_block *sb)
202 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
203 if (sbinfo->max_inodes) {
204 spin_lock(&sbinfo->stat_lock);
205 sbinfo->free_inodes++;
206 spin_unlock(&sbinfo->stat_lock);
211 * shmem_recalc_inode - recalculate the block usage of an inode
212 * @inode: inode to recalc
214 * We have to calculate the free blocks since the mm can drop
215 * undirtied hole pages behind our back.
217 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
218 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
220 * It has to be called with the spinlock held.
222 static void shmem_recalc_inode(struct inode *inode)
224 struct shmem_inode_info *info = SHMEM_I(inode);
227 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
229 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
230 if (sbinfo->max_blocks)
231 percpu_counter_add(&sbinfo->used_blocks, -freed);
232 info->alloced -= freed;
233 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
234 shmem_unacct_blocks(info->flags, freed);
239 * Replace item expected in radix tree by a new item, while holding tree lock.
241 static int shmem_radix_tree_replace(struct address_space *mapping,
242 pgoff_t index, void *expected, void *replacement)
247 VM_BUG_ON(!expected);
248 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
250 item = radix_tree_deref_slot_protected(pslot,
251 &mapping->tree_lock);
252 if (item != expected)
255 radix_tree_replace_slot(pslot, replacement);
257 radix_tree_delete(&mapping->page_tree, index);
262 * Sometimes, before we decide whether to proceed or to fail, we must check
263 * that an entry was not already brought back from swap by a racing thread.
265 * Checking page is not enough: by the time a SwapCache page is locked, it
266 * might be reused, and again be SwapCache, using the same swap as before.
268 static bool shmem_confirm_swap(struct address_space *mapping,
269 pgoff_t index, swp_entry_t swap)
274 item = radix_tree_lookup(&mapping->page_tree, index);
276 return item == swp_to_radix_entry(swap);
280 * Like add_to_page_cache_locked, but error if expected item has gone.
282 static int shmem_add_to_page_cache(struct page *page,
283 struct address_space *mapping,
284 pgoff_t index, gfp_t gfp, void *expected)
288 VM_BUG_ON_PAGE(!PageLocked(page), page);
289 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
291 page_cache_get(page);
292 page->mapping = mapping;
295 spin_lock_irq(&mapping->tree_lock);
297 error = radix_tree_insert(&mapping->page_tree, index, page);
299 error = shmem_radix_tree_replace(mapping, index, expected,
303 __inc_zone_page_state(page, NR_FILE_PAGES);
304 __inc_zone_page_state(page, NR_SHMEM);
305 spin_unlock_irq(&mapping->tree_lock);
307 page->mapping = NULL;
308 spin_unlock_irq(&mapping->tree_lock);
309 page_cache_release(page);
315 * Like delete_from_page_cache, but substitutes swap for page.
317 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
319 struct address_space *mapping = page->mapping;
322 spin_lock_irq(&mapping->tree_lock);
323 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
324 page->mapping = NULL;
326 __dec_zone_page_state(page, NR_FILE_PAGES);
327 __dec_zone_page_state(page, NR_SHMEM);
328 spin_unlock_irq(&mapping->tree_lock);
329 page_cache_release(page);
334 * Like find_get_pages, but collecting swap entries as well as pages.
336 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
337 pgoff_t start, unsigned int nr_pages,
338 struct page **pages, pgoff_t *indices)
341 unsigned int ret = 0;
342 struct radix_tree_iter iter;
349 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
352 page = radix_tree_deref_slot(slot);
355 if (radix_tree_exception(page)) {
356 if (radix_tree_deref_retry(page))
359 * Otherwise, we must be storing a swap entry
360 * here as an exceptional entry: so return it
361 * without attempting to raise page count.
365 if (!page_cache_get_speculative(page))
368 /* Has the page moved? */
369 if (unlikely(page != *slot)) {
370 page_cache_release(page);
374 indices[ret] = iter.index;
376 if (++ret == nr_pages)
384 * Remove swap entry from radix tree, free the swap and its page cache.
386 static int shmem_free_swap(struct address_space *mapping,
387 pgoff_t index, void *radswap)
391 spin_lock_irq(&mapping->tree_lock);
392 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
393 spin_unlock_irq(&mapping->tree_lock);
395 free_swap_and_cache(radix_to_swp_entry(radswap));
400 * Pagevec may contain swap entries, so shuffle up pages before releasing.
402 static void shmem_deswap_pagevec(struct pagevec *pvec)
406 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
407 struct page *page = pvec->pages[i];
408 if (!radix_tree_exceptional_entry(page))
409 pvec->pages[j++] = page;
415 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
417 void shmem_unlock_mapping(struct address_space *mapping)
420 pgoff_t indices[PAGEVEC_SIZE];
423 pagevec_init(&pvec, 0);
425 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
427 while (!mapping_unevictable(mapping)) {
429 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
430 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
432 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
433 PAGEVEC_SIZE, pvec.pages, indices);
436 index = indices[pvec.nr - 1] + 1;
437 shmem_deswap_pagevec(&pvec);
438 check_move_unevictable_pages(pvec.pages, pvec.nr);
439 pagevec_release(&pvec);
445 * Remove range of pages and swap entries from radix tree, and free them.
446 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
448 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
451 struct address_space *mapping = inode->i_mapping;
452 struct shmem_inode_info *info = SHMEM_I(inode);
453 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
454 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
455 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
456 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
458 pgoff_t indices[PAGEVEC_SIZE];
459 long nr_swaps_freed = 0;
464 end = -1; /* unsigned, so actually very big */
466 pagevec_init(&pvec, 0);
468 while (index < end) {
469 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
470 min(end - index, (pgoff_t)PAGEVEC_SIZE),
471 pvec.pages, indices);
474 mem_cgroup_uncharge_start();
475 for (i = 0; i < pagevec_count(&pvec); i++) {
476 struct page *page = pvec.pages[i];
482 if (radix_tree_exceptional_entry(page)) {
485 nr_swaps_freed += !shmem_free_swap(mapping,
490 if (!trylock_page(page))
492 if (!unfalloc || !PageUptodate(page)) {
493 if (page->mapping == mapping) {
494 VM_BUG_ON_PAGE(PageWriteback(page), page);
495 truncate_inode_page(mapping, page);
500 shmem_deswap_pagevec(&pvec);
501 pagevec_release(&pvec);
502 mem_cgroup_uncharge_end();
508 struct page *page = NULL;
509 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
511 unsigned int top = PAGE_CACHE_SIZE;
516 zero_user_segment(page, partial_start, top);
517 set_page_dirty(page);
519 page_cache_release(page);
523 struct page *page = NULL;
524 shmem_getpage(inode, end, &page, SGP_READ, NULL);
526 zero_user_segment(page, 0, partial_end);
527 set_page_dirty(page);
529 page_cache_release(page);
538 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
539 min(end - index, (pgoff_t)PAGEVEC_SIZE),
540 pvec.pages, indices);
542 if (index == start || unfalloc)
547 if ((index == start || unfalloc) && indices[0] >= end) {
548 shmem_deswap_pagevec(&pvec);
549 pagevec_release(&pvec);
552 mem_cgroup_uncharge_start();
553 for (i = 0; i < pagevec_count(&pvec); i++) {
554 struct page *page = pvec.pages[i];
560 if (radix_tree_exceptional_entry(page)) {
563 nr_swaps_freed += !shmem_free_swap(mapping,
569 if (!unfalloc || !PageUptodate(page)) {
570 if (page->mapping == mapping) {
571 VM_BUG_ON_PAGE(PageWriteback(page), page);
572 truncate_inode_page(mapping, page);
577 shmem_deswap_pagevec(&pvec);
578 pagevec_release(&pvec);
579 mem_cgroup_uncharge_end();
583 spin_lock(&info->lock);
584 info->swapped -= nr_swaps_freed;
585 shmem_recalc_inode(inode);
586 spin_unlock(&info->lock);
589 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
591 shmem_undo_range(inode, lstart, lend, false);
592 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
594 EXPORT_SYMBOL_GPL(shmem_truncate_range);
596 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
598 struct inode *inode = dentry->d_inode;
601 error = inode_change_ok(inode, attr);
605 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
606 loff_t oldsize = inode->i_size;
607 loff_t newsize = attr->ia_size;
609 if (newsize != oldsize) {
610 i_size_write(inode, newsize);
611 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
613 if (newsize < oldsize) {
614 loff_t holebegin = round_up(newsize, PAGE_SIZE);
615 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
616 shmem_truncate_range(inode, newsize, (loff_t)-1);
617 /* unmap again to remove racily COWed private pages */
618 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
622 setattr_copy(inode, attr);
623 if (attr->ia_valid & ATTR_MODE)
624 error = posix_acl_chmod(inode, inode->i_mode);
628 static void shmem_evict_inode(struct inode *inode)
630 struct shmem_inode_info *info = SHMEM_I(inode);
632 if (inode->i_mapping->a_ops == &shmem_aops) {
633 shmem_unacct_size(info->flags, inode->i_size);
635 shmem_truncate_range(inode, 0, (loff_t)-1);
636 if (!list_empty(&info->swaplist)) {
637 mutex_lock(&shmem_swaplist_mutex);
638 list_del_init(&info->swaplist);
639 mutex_unlock(&shmem_swaplist_mutex);
642 kfree(info->symlink);
644 simple_xattrs_free(&info->xattrs);
645 WARN_ON(inode->i_blocks);
646 shmem_free_inode(inode->i_sb);
651 * If swap found in inode, free it and move page from swapcache to filecache.
653 static int shmem_unuse_inode(struct shmem_inode_info *info,
654 swp_entry_t swap, struct page **pagep)
656 struct address_space *mapping = info->vfs_inode.i_mapping;
662 radswap = swp_to_radix_entry(swap);
663 index = radix_tree_locate_item(&mapping->page_tree, radswap);
668 * Move _head_ to start search for next from here.
669 * But be careful: shmem_evict_inode checks list_empty without taking
670 * mutex, and there's an instant in list_move_tail when info->swaplist
671 * would appear empty, if it were the only one on shmem_swaplist.
673 if (shmem_swaplist.next != &info->swaplist)
674 list_move_tail(&shmem_swaplist, &info->swaplist);
676 gfp = mapping_gfp_mask(mapping);
677 if (shmem_should_replace_page(*pagep, gfp)) {
678 mutex_unlock(&shmem_swaplist_mutex);
679 error = shmem_replace_page(pagep, gfp, info, index);
680 mutex_lock(&shmem_swaplist_mutex);
682 * We needed to drop mutex to make that restrictive page
683 * allocation, but the inode might have been freed while we
684 * dropped it: although a racing shmem_evict_inode() cannot
685 * complete without emptying the radix_tree, our page lock
686 * on this swapcache page is not enough to prevent that -
687 * free_swap_and_cache() of our swap entry will only
688 * trylock_page(), removing swap from radix_tree whatever.
690 * We must not proceed to shmem_add_to_page_cache() if the
691 * inode has been freed, but of course we cannot rely on
692 * inode or mapping or info to check that. However, we can
693 * safely check if our swap entry is still in use (and here
694 * it can't have got reused for another page): if it's still
695 * in use, then the inode cannot have been freed yet, and we
696 * can safely proceed (if it's no longer in use, that tells
697 * nothing about the inode, but we don't need to unuse swap).
699 if (!page_swapcount(*pagep))
704 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
705 * but also to hold up shmem_evict_inode(): so inode cannot be freed
706 * beneath us (pagelock doesn't help until the page is in pagecache).
709 error = shmem_add_to_page_cache(*pagep, mapping, index,
710 GFP_NOWAIT, radswap);
711 if (error != -ENOMEM) {
713 * Truncation and eviction use free_swap_and_cache(), which
714 * only does trylock page: if we raced, best clean up here.
716 delete_from_swap_cache(*pagep);
717 set_page_dirty(*pagep);
719 spin_lock(&info->lock);
721 spin_unlock(&info->lock);
724 error = 1; /* not an error, but entry was found */
730 * Search through swapped inodes to find and replace swap by page.
732 int shmem_unuse(swp_entry_t swap, struct page *page)
734 struct list_head *this, *next;
735 struct shmem_inode_info *info;
740 * There's a faint possibility that swap page was replaced before
741 * caller locked it: caller will come back later with the right page.
743 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
747 * Charge page using GFP_KERNEL while we can wait, before taking
748 * the shmem_swaplist_mutex which might hold up shmem_writepage().
749 * Charged back to the user (not to caller) when swap account is used.
751 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
754 /* No radix_tree_preload: swap entry keeps a place for page in tree */
756 mutex_lock(&shmem_swaplist_mutex);
757 list_for_each_safe(this, next, &shmem_swaplist) {
758 info = list_entry(this, struct shmem_inode_info, swaplist);
760 found = shmem_unuse_inode(info, swap, &page);
762 list_del_init(&info->swaplist);
767 mutex_unlock(&shmem_swaplist_mutex);
773 page_cache_release(page);
778 * Move the page from the page cache to the swap cache.
780 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
782 struct shmem_inode_info *info;
783 struct address_space *mapping;
788 BUG_ON(!PageLocked(page));
789 mapping = page->mapping;
791 inode = mapping->host;
792 info = SHMEM_I(inode);
793 if (info->flags & VM_LOCKED)
795 if (!total_swap_pages)
799 * shmem_backing_dev_info's capabilities prevent regular writeback or
800 * sync from ever calling shmem_writepage; but a stacking filesystem
801 * might use ->writepage of its underlying filesystem, in which case
802 * tmpfs should write out to swap only in response to memory pressure,
803 * and not for the writeback threads or sync.
805 if (!wbc->for_reclaim) {
806 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
811 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
812 * value into swapfile.c, the only way we can correctly account for a
813 * fallocated page arriving here is now to initialize it and write it.
815 * That's okay for a page already fallocated earlier, but if we have
816 * not yet completed the fallocation, then (a) we want to keep track
817 * of this page in case we have to undo it, and (b) it may not be a
818 * good idea to continue anyway, once we're pushing into swap. So
819 * reactivate the page, and let shmem_fallocate() quit when too many.
821 if (!PageUptodate(page)) {
822 if (inode->i_private) {
823 struct shmem_falloc *shmem_falloc;
824 spin_lock(&inode->i_lock);
825 shmem_falloc = inode->i_private;
827 index >= shmem_falloc->start &&
828 index < shmem_falloc->next)
829 shmem_falloc->nr_unswapped++;
832 spin_unlock(&inode->i_lock);
836 clear_highpage(page);
837 flush_dcache_page(page);
838 SetPageUptodate(page);
841 swap = get_swap_page();
846 * Add inode to shmem_unuse()'s list of swapped-out inodes,
847 * if it's not already there. Do it now before the page is
848 * moved to swap cache, when its pagelock no longer protects
849 * the inode from eviction. But don't unlock the mutex until
850 * we've incremented swapped, because shmem_unuse_inode() will
851 * prune a !swapped inode from the swaplist under this mutex.
853 mutex_lock(&shmem_swaplist_mutex);
854 if (list_empty(&info->swaplist))
855 list_add_tail(&info->swaplist, &shmem_swaplist);
857 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
858 swap_shmem_alloc(swap);
859 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
861 spin_lock(&info->lock);
863 shmem_recalc_inode(inode);
864 spin_unlock(&info->lock);
866 mutex_unlock(&shmem_swaplist_mutex);
867 BUG_ON(page_mapped(page));
868 swap_writepage(page, wbc);
872 mutex_unlock(&shmem_swaplist_mutex);
873 swapcache_free(swap, NULL);
875 set_page_dirty(page);
876 if (wbc->for_reclaim)
877 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
884 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
888 if (!mpol || mpol->mode == MPOL_DEFAULT)
889 return; /* show nothing */
891 mpol_to_str(buffer, sizeof(buffer), mpol);
893 seq_printf(seq, ",mpol=%s", buffer);
896 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
898 struct mempolicy *mpol = NULL;
900 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
903 spin_unlock(&sbinfo->stat_lock);
907 #endif /* CONFIG_TMPFS */
909 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
910 struct shmem_inode_info *info, pgoff_t index)
912 struct vm_area_struct pvma;
915 /* Create a pseudo vma that just contains the policy */
917 /* Bias interleave by inode number to distribute better across nodes */
918 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
920 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
922 page = swapin_readahead(swap, gfp, &pvma, 0);
924 /* Drop reference taken by mpol_shared_policy_lookup() */
925 mpol_cond_put(pvma.vm_policy);
930 static struct page *shmem_alloc_page(gfp_t gfp,
931 struct shmem_inode_info *info, pgoff_t index)
933 struct vm_area_struct pvma;
936 /* Create a pseudo vma that just contains the policy */
938 /* Bias interleave by inode number to distribute better across nodes */
939 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
941 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
943 page = alloc_page_vma(gfp, &pvma, 0);
945 /* Drop reference taken by mpol_shared_policy_lookup() */
946 mpol_cond_put(pvma.vm_policy);
950 #else /* !CONFIG_NUMA */
952 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
955 #endif /* CONFIG_TMPFS */
957 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
958 struct shmem_inode_info *info, pgoff_t index)
960 return swapin_readahead(swap, gfp, NULL, 0);
963 static inline struct page *shmem_alloc_page(gfp_t gfp,
964 struct shmem_inode_info *info, pgoff_t index)
966 return alloc_page(gfp);
968 #endif /* CONFIG_NUMA */
970 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
971 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
978 * When a page is moved from swapcache to shmem filecache (either by the
979 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
980 * shmem_unuse_inode()), it may have been read in earlier from swap, in
981 * ignorance of the mapping it belongs to. If that mapping has special
982 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
983 * we may need to copy to a suitable page before moving to filecache.
985 * In a future release, this may well be extended to respect cpuset and
986 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
987 * but for now it is a simple matter of zone.
989 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
991 return page_zonenum(page) > gfp_zone(gfp);
994 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
995 struct shmem_inode_info *info, pgoff_t index)
997 struct page *oldpage, *newpage;
998 struct address_space *swap_mapping;
1003 swap_index = page_private(oldpage);
1004 swap_mapping = page_mapping(oldpage);
1007 * We have arrived here because our zones are constrained, so don't
1008 * limit chance of success by further cpuset and node constraints.
1010 gfp &= ~GFP_CONSTRAINT_MASK;
1011 newpage = shmem_alloc_page(gfp, info, index);
1015 page_cache_get(newpage);
1016 copy_highpage(newpage, oldpage);
1017 flush_dcache_page(newpage);
1019 __set_page_locked(newpage);
1020 SetPageUptodate(newpage);
1021 SetPageSwapBacked(newpage);
1022 set_page_private(newpage, swap_index);
1023 SetPageSwapCache(newpage);
1026 * Our caller will very soon move newpage out of swapcache, but it's
1027 * a nice clean interface for us to replace oldpage by newpage there.
1029 spin_lock_irq(&swap_mapping->tree_lock);
1030 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1033 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1034 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1036 spin_unlock_irq(&swap_mapping->tree_lock);
1038 if (unlikely(error)) {
1040 * Is this possible? I think not, now that our callers check
1041 * both PageSwapCache and page_private after getting page lock;
1042 * but be defensive. Reverse old to newpage for clear and free.
1046 mem_cgroup_replace_page_cache(oldpage, newpage);
1047 lru_cache_add_anon(newpage);
1051 ClearPageSwapCache(oldpage);
1052 set_page_private(oldpage, 0);
1054 unlock_page(oldpage);
1055 page_cache_release(oldpage);
1056 page_cache_release(oldpage);
1061 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1063 * If we allocate a new one we do not mark it dirty. That's up to the
1064 * vm. If we swap it in we mark it dirty since we also free the swap
1065 * entry since a page cannot live in both the swap and page cache
1067 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1068 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1070 struct address_space *mapping = inode->i_mapping;
1071 struct shmem_inode_info *info;
1072 struct shmem_sb_info *sbinfo;
1079 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1083 page = find_lock_page(mapping, index);
1084 if (radix_tree_exceptional_entry(page)) {
1085 swap = radix_to_swp_entry(page);
1089 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1090 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1095 /* fallocated page? */
1096 if (page && !PageUptodate(page)) {
1097 if (sgp != SGP_READ)
1100 page_cache_release(page);
1103 if (page || (sgp == SGP_READ && !swap.val)) {
1109 * Fast cache lookup did not find it:
1110 * bring it back from swap or allocate.
1112 info = SHMEM_I(inode);
1113 sbinfo = SHMEM_SB(inode->i_sb);
1116 /* Look it up and read it in.. */
1117 page = lookup_swap_cache(swap);
1119 /* here we actually do the io */
1121 *fault_type |= VM_FAULT_MAJOR;
1122 page = shmem_swapin(swap, gfp, info, index);
1129 /* We have to do this with page locked to prevent races */
1131 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1132 !shmem_confirm_swap(mapping, index, swap)) {
1133 error = -EEXIST; /* try again */
1136 if (!PageUptodate(page)) {
1140 wait_on_page_writeback(page);
1142 if (shmem_should_replace_page(page, gfp)) {
1143 error = shmem_replace_page(&page, gfp, info, index);
1148 error = mem_cgroup_cache_charge(page, current->mm,
1149 gfp & GFP_RECLAIM_MASK);
1151 error = shmem_add_to_page_cache(page, mapping, index,
1152 gfp, swp_to_radix_entry(swap));
1154 * We already confirmed swap under page lock, and make
1155 * no memory allocation here, so usually no possibility
1156 * of error; but free_swap_and_cache() only trylocks a
1157 * page, so it is just possible that the entry has been
1158 * truncated or holepunched since swap was confirmed.
1159 * shmem_undo_range() will have done some of the
1160 * unaccounting, now delete_from_swap_cache() will do
1161 * the rest (including mem_cgroup_uncharge_swapcache).
1162 * Reset swap.val? No, leave it so "failed" goes back to
1163 * "repeat": reading a hole and writing should succeed.
1166 delete_from_swap_cache(page);
1171 spin_lock(&info->lock);
1173 shmem_recalc_inode(inode);
1174 spin_unlock(&info->lock);
1176 delete_from_swap_cache(page);
1177 set_page_dirty(page);
1181 if (shmem_acct_block(info->flags)) {
1185 if (sbinfo->max_blocks) {
1186 if (percpu_counter_compare(&sbinfo->used_blocks,
1187 sbinfo->max_blocks) >= 0) {
1191 percpu_counter_inc(&sbinfo->used_blocks);
1194 page = shmem_alloc_page(gfp, info, index);
1200 SetPageSwapBacked(page);
1201 __set_page_locked(page);
1202 error = mem_cgroup_cache_charge(page, current->mm,
1203 gfp & GFP_RECLAIM_MASK);
1206 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
1208 error = shmem_add_to_page_cache(page, mapping, index,
1210 radix_tree_preload_end();
1213 mem_cgroup_uncharge_cache_page(page);
1216 lru_cache_add_anon(page);
1218 spin_lock(&info->lock);
1220 inode->i_blocks += BLOCKS_PER_PAGE;
1221 shmem_recalc_inode(inode);
1222 spin_unlock(&info->lock);
1226 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1228 if (sgp == SGP_FALLOC)
1232 * Let SGP_WRITE caller clear ends if write does not fill page;
1233 * but SGP_FALLOC on a page fallocated earlier must initialize
1234 * it now, lest undo on failure cancel our earlier guarantee.
1236 if (sgp != SGP_WRITE) {
1237 clear_highpage(page);
1238 flush_dcache_page(page);
1239 SetPageUptodate(page);
1241 if (sgp == SGP_DIRTY)
1242 set_page_dirty(page);
1245 /* Perhaps the file has been truncated since we checked */
1246 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1247 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1261 info = SHMEM_I(inode);
1262 ClearPageDirty(page);
1263 delete_from_page_cache(page);
1264 spin_lock(&info->lock);
1266 inode->i_blocks -= BLOCKS_PER_PAGE;
1267 spin_unlock(&info->lock);
1269 sbinfo = SHMEM_SB(inode->i_sb);
1270 if (sbinfo->max_blocks)
1271 percpu_counter_add(&sbinfo->used_blocks, -1);
1273 shmem_unacct_blocks(info->flags, 1);
1275 if (swap.val && error != -EINVAL &&
1276 !shmem_confirm_swap(mapping, index, swap))
1281 page_cache_release(page);
1283 if (error == -ENOSPC && !once++) {
1284 info = SHMEM_I(inode);
1285 spin_lock(&info->lock);
1286 shmem_recalc_inode(inode);
1287 spin_unlock(&info->lock);
1290 if (error == -EEXIST) /* from above or from radix_tree_insert */
1295 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1297 struct inode *inode = file_inode(vma->vm_file);
1299 int ret = VM_FAULT_LOCKED;
1301 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1303 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1305 if (ret & VM_FAULT_MAJOR) {
1306 count_vm_event(PGMAJFAULT);
1307 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1313 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1315 struct inode *inode = file_inode(vma->vm_file);
1316 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1319 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1322 struct inode *inode = file_inode(vma->vm_file);
1325 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1326 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1330 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1332 struct inode *inode = file_inode(file);
1333 struct shmem_inode_info *info = SHMEM_I(inode);
1334 int retval = -ENOMEM;
1336 spin_lock(&info->lock);
1337 if (lock && !(info->flags & VM_LOCKED)) {
1338 if (!user_shm_lock(inode->i_size, user))
1340 info->flags |= VM_LOCKED;
1341 mapping_set_unevictable(file->f_mapping);
1343 if (!lock && (info->flags & VM_LOCKED) && user) {
1344 user_shm_unlock(inode->i_size, user);
1345 info->flags &= ~VM_LOCKED;
1346 mapping_clear_unevictable(file->f_mapping);
1351 spin_unlock(&info->lock);
1355 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1357 file_accessed(file);
1358 vma->vm_ops = &shmem_vm_ops;
1362 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1363 umode_t mode, dev_t dev, unsigned long flags)
1365 struct inode *inode;
1366 struct shmem_inode_info *info;
1367 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1369 if (shmem_reserve_inode(sb))
1372 inode = new_inode(sb);
1374 inode->i_ino = get_next_ino();
1375 inode_init_owner(inode, dir, mode);
1376 inode->i_blocks = 0;
1377 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1378 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1379 inode->i_generation = get_seconds();
1380 info = SHMEM_I(inode);
1381 memset(info, 0, (char *)inode - (char *)info);
1382 spin_lock_init(&info->lock);
1383 info->flags = flags & VM_NORESERVE;
1384 INIT_LIST_HEAD(&info->swaplist);
1385 simple_xattrs_init(&info->xattrs);
1386 cache_no_acl(inode);
1388 switch (mode & S_IFMT) {
1390 inode->i_op = &shmem_special_inode_operations;
1391 init_special_inode(inode, mode, dev);
1394 inode->i_mapping->a_ops = &shmem_aops;
1395 inode->i_op = &shmem_inode_operations;
1396 inode->i_fop = &shmem_file_operations;
1397 mpol_shared_policy_init(&info->policy,
1398 shmem_get_sbmpol(sbinfo));
1402 /* Some things misbehave if size == 0 on a directory */
1403 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1404 inode->i_op = &shmem_dir_inode_operations;
1405 inode->i_fop = &simple_dir_operations;
1409 * Must not load anything in the rbtree,
1410 * mpol_free_shared_policy will not be called.
1412 mpol_shared_policy_init(&info->policy, NULL);
1416 shmem_free_inode(sb);
1421 static const struct inode_operations shmem_symlink_inode_operations;
1422 static const struct inode_operations shmem_short_symlink_operations;
1424 #ifdef CONFIG_TMPFS_XATTR
1425 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1427 #define shmem_initxattrs NULL
1431 shmem_write_begin(struct file *file, struct address_space *mapping,
1432 loff_t pos, unsigned len, unsigned flags,
1433 struct page **pagep, void **fsdata)
1435 struct inode *inode = mapping->host;
1436 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1437 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1441 shmem_write_end(struct file *file, struct address_space *mapping,
1442 loff_t pos, unsigned len, unsigned copied,
1443 struct page *page, void *fsdata)
1445 struct inode *inode = mapping->host;
1447 if (pos + copied > inode->i_size)
1448 i_size_write(inode, pos + copied);
1450 if (!PageUptodate(page)) {
1451 if (copied < PAGE_CACHE_SIZE) {
1452 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1453 zero_user_segments(page, 0, from,
1454 from + copied, PAGE_CACHE_SIZE);
1456 SetPageUptodate(page);
1458 set_page_dirty(page);
1460 page_cache_release(page);
1465 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1466 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1468 struct file *file = iocb->ki_filp;
1469 struct inode *inode = file_inode(file);
1470 struct address_space *mapping = inode->i_mapping;
1472 unsigned long offset;
1473 enum sgp_type sgp = SGP_READ;
1477 loff_t *ppos = &iocb->ki_pos;
1478 struct iov_iter iter;
1480 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1483 iov_iter_init(&iter, iov, nr_segs, count, 0);
1486 * Might this read be for a stacking filesystem? Then when reading
1487 * holes of a sparse file, we actually need to allocate those pages,
1488 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1490 if (segment_eq(get_fs(), KERNEL_DS))
1493 index = *ppos >> PAGE_CACHE_SHIFT;
1494 offset = *ppos & ~PAGE_CACHE_MASK;
1497 struct page *page = NULL;
1499 unsigned long nr, ret;
1500 loff_t i_size = i_size_read(inode);
1502 end_index = i_size >> PAGE_CACHE_SHIFT;
1503 if (index > end_index)
1505 if (index == end_index) {
1506 nr = i_size & ~PAGE_CACHE_MASK;
1511 error = shmem_getpage(inode, index, &page, sgp, NULL);
1513 if (error == -EINVAL)
1521 * We must evaluate after, since reads (unlike writes)
1522 * are called without i_mutex protection against truncate
1524 nr = PAGE_CACHE_SIZE;
1525 i_size = i_size_read(inode);
1526 end_index = i_size >> PAGE_CACHE_SHIFT;
1527 if (index == end_index) {
1528 nr = i_size & ~PAGE_CACHE_MASK;
1531 page_cache_release(page);
1539 * If users can be writing to this page using arbitrary
1540 * virtual addresses, take care about potential aliasing
1541 * before reading the page on the kernel side.
1543 if (mapping_writably_mapped(mapping))
1544 flush_dcache_page(page);
1546 * Mark the page accessed if we read the beginning.
1549 mark_page_accessed(page);
1551 page = ZERO_PAGE(0);
1552 page_cache_get(page);
1556 * Ok, we have the page, and it's up-to-date, so
1557 * now we can copy it to user space...
1559 ret = copy_page_to_iter(page, offset, nr, &iter);
1562 index += offset >> PAGE_CACHE_SHIFT;
1563 offset &= ~PAGE_CACHE_MASK;
1565 page_cache_release(page);
1566 if (!iov_iter_count(&iter))
1575 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1576 file_accessed(file);
1577 return retval ? retval : error;
1580 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1581 struct pipe_inode_info *pipe, size_t len,
1584 struct address_space *mapping = in->f_mapping;
1585 struct inode *inode = mapping->host;
1586 unsigned int loff, nr_pages, req_pages;
1587 struct page *pages[PIPE_DEF_BUFFERS];
1588 struct partial_page partial[PIPE_DEF_BUFFERS];
1590 pgoff_t index, end_index;
1593 struct splice_pipe_desc spd = {
1596 .nr_pages_max = PIPE_DEF_BUFFERS,
1598 .ops = &page_cache_pipe_buf_ops,
1599 .spd_release = spd_release_page,
1602 isize = i_size_read(inode);
1603 if (unlikely(*ppos >= isize))
1606 left = isize - *ppos;
1607 if (unlikely(left < len))
1610 if (splice_grow_spd(pipe, &spd))
1613 index = *ppos >> PAGE_CACHE_SHIFT;
1614 loff = *ppos & ~PAGE_CACHE_MASK;
1615 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1616 nr_pages = min(req_pages, spd.nr_pages_max);
1618 spd.nr_pages = find_get_pages_contig(mapping, index,
1619 nr_pages, spd.pages);
1620 index += spd.nr_pages;
1623 while (spd.nr_pages < nr_pages) {
1624 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1628 spd.pages[spd.nr_pages++] = page;
1632 index = *ppos >> PAGE_CACHE_SHIFT;
1633 nr_pages = spd.nr_pages;
1636 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1637 unsigned int this_len;
1642 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1643 page = spd.pages[page_nr];
1645 if (!PageUptodate(page) || page->mapping != mapping) {
1646 error = shmem_getpage(inode, index, &page,
1651 page_cache_release(spd.pages[page_nr]);
1652 spd.pages[page_nr] = page;
1655 isize = i_size_read(inode);
1656 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1657 if (unlikely(!isize || index > end_index))
1660 if (end_index == index) {
1663 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1667 this_len = min(this_len, plen - loff);
1671 spd.partial[page_nr].offset = loff;
1672 spd.partial[page_nr].len = this_len;
1679 while (page_nr < nr_pages)
1680 page_cache_release(spd.pages[page_nr++]);
1683 error = splice_to_pipe(pipe, &spd);
1685 splice_shrink_spd(&spd);
1695 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1697 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1698 pgoff_t index, pgoff_t end, int whence)
1701 struct pagevec pvec;
1702 pgoff_t indices[PAGEVEC_SIZE];
1706 pagevec_init(&pvec, 0);
1707 pvec.nr = 1; /* start small: we may be there already */
1709 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
1710 pvec.nr, pvec.pages, indices);
1712 if (whence == SEEK_DATA)
1716 for (i = 0; i < pvec.nr; i++, index++) {
1717 if (index < indices[i]) {
1718 if (whence == SEEK_HOLE) {
1724 page = pvec.pages[i];
1725 if (page && !radix_tree_exceptional_entry(page)) {
1726 if (!PageUptodate(page))
1730 (page && whence == SEEK_DATA) ||
1731 (!page && whence == SEEK_HOLE)) {
1736 shmem_deswap_pagevec(&pvec);
1737 pagevec_release(&pvec);
1738 pvec.nr = PAGEVEC_SIZE;
1744 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
1746 struct address_space *mapping = file->f_mapping;
1747 struct inode *inode = mapping->host;
1751 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1752 return generic_file_llseek_size(file, offset, whence,
1753 MAX_LFS_FILESIZE, i_size_read(inode));
1754 mutex_lock(&inode->i_mutex);
1755 /* We're holding i_mutex so we can access i_size directly */
1759 else if (offset >= inode->i_size)
1762 start = offset >> PAGE_CACHE_SHIFT;
1763 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1764 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
1765 new_offset <<= PAGE_CACHE_SHIFT;
1766 if (new_offset > offset) {
1767 if (new_offset < inode->i_size)
1768 offset = new_offset;
1769 else if (whence == SEEK_DATA)
1772 offset = inode->i_size;
1777 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
1778 mutex_unlock(&inode->i_mutex);
1782 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1785 struct inode *inode = file_inode(file);
1786 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1787 struct shmem_falloc shmem_falloc;
1788 pgoff_t start, index, end;
1791 mutex_lock(&inode->i_mutex);
1793 if (mode & FALLOC_FL_PUNCH_HOLE) {
1794 struct address_space *mapping = file->f_mapping;
1795 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1796 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1798 if ((u64)unmap_end > (u64)unmap_start)
1799 unmap_mapping_range(mapping, unmap_start,
1800 1 + unmap_end - unmap_start, 0);
1801 shmem_truncate_range(inode, offset, offset + len - 1);
1802 /* No need to unmap again: hole-punching leaves COWed pages */
1807 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1808 error = inode_newsize_ok(inode, offset + len);
1812 start = offset >> PAGE_CACHE_SHIFT;
1813 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1814 /* Try to avoid a swapstorm if len is impossible to satisfy */
1815 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1820 shmem_falloc.start = start;
1821 shmem_falloc.next = start;
1822 shmem_falloc.nr_falloced = 0;
1823 shmem_falloc.nr_unswapped = 0;
1824 spin_lock(&inode->i_lock);
1825 inode->i_private = &shmem_falloc;
1826 spin_unlock(&inode->i_lock);
1828 for (index = start; index < end; index++) {
1832 * Good, the fallocate(2) manpage permits EINTR: we may have
1833 * been interrupted because we are using up too much memory.
1835 if (signal_pending(current))
1837 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1840 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1843 /* Remove the !PageUptodate pages we added */
1844 shmem_undo_range(inode,
1845 (loff_t)start << PAGE_CACHE_SHIFT,
1846 (loff_t)index << PAGE_CACHE_SHIFT, true);
1851 * Inform shmem_writepage() how far we have reached.
1852 * No need for lock or barrier: we have the page lock.
1854 shmem_falloc.next++;
1855 if (!PageUptodate(page))
1856 shmem_falloc.nr_falloced++;
1859 * If !PageUptodate, leave it that way so that freeable pages
1860 * can be recognized if we need to rollback on error later.
1861 * But set_page_dirty so that memory pressure will swap rather
1862 * than free the pages we are allocating (and SGP_CACHE pages
1863 * might still be clean: we now need to mark those dirty too).
1865 set_page_dirty(page);
1867 page_cache_release(page);
1871 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1872 i_size_write(inode, offset + len);
1873 inode->i_ctime = CURRENT_TIME;
1875 spin_lock(&inode->i_lock);
1876 inode->i_private = NULL;
1877 spin_unlock(&inode->i_lock);
1879 mutex_unlock(&inode->i_mutex);
1883 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1885 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1887 buf->f_type = TMPFS_MAGIC;
1888 buf->f_bsize = PAGE_CACHE_SIZE;
1889 buf->f_namelen = NAME_MAX;
1890 if (sbinfo->max_blocks) {
1891 buf->f_blocks = sbinfo->max_blocks;
1893 buf->f_bfree = sbinfo->max_blocks -
1894 percpu_counter_sum(&sbinfo->used_blocks);
1896 if (sbinfo->max_inodes) {
1897 buf->f_files = sbinfo->max_inodes;
1898 buf->f_ffree = sbinfo->free_inodes;
1900 /* else leave those fields 0 like simple_statfs */
1905 * File creation. Allocate an inode, and we're done..
1908 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1910 struct inode *inode;
1911 int error = -ENOSPC;
1913 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1915 error = simple_acl_create(dir, inode);
1918 error = security_inode_init_security(inode, dir,
1920 shmem_initxattrs, NULL);
1921 if (error && error != -EOPNOTSUPP)
1925 dir->i_size += BOGO_DIRENT_SIZE;
1926 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1927 d_instantiate(dentry, inode);
1928 dget(dentry); /* Extra count - pin the dentry in core */
1937 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
1939 struct inode *inode;
1940 int error = -ENOSPC;
1942 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
1944 error = security_inode_init_security(inode, dir,
1946 shmem_initxattrs, NULL);
1947 if (error && error != -EOPNOTSUPP)
1949 error = simple_acl_create(dir, inode);
1952 d_tmpfile(dentry, inode);
1960 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1964 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1970 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1973 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1979 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1981 struct inode *inode = old_dentry->d_inode;
1985 * No ordinary (disk based) filesystem counts links as inodes;
1986 * but each new link needs a new dentry, pinning lowmem, and
1987 * tmpfs dentries cannot be pruned until they are unlinked.
1989 ret = shmem_reserve_inode(inode->i_sb);
1993 dir->i_size += BOGO_DIRENT_SIZE;
1994 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1996 ihold(inode); /* New dentry reference */
1997 dget(dentry); /* Extra pinning count for the created dentry */
1998 d_instantiate(dentry, inode);
2003 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2005 struct inode *inode = dentry->d_inode;
2007 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2008 shmem_free_inode(inode->i_sb);
2010 dir->i_size -= BOGO_DIRENT_SIZE;
2011 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2013 dput(dentry); /* Undo the count from "create" - this does all the work */
2017 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2019 if (!simple_empty(dentry))
2022 drop_nlink(dentry->d_inode);
2024 return shmem_unlink(dir, dentry);
2028 * The VFS layer already does all the dentry stuff for rename,
2029 * we just have to decrement the usage count for the target if
2030 * it exists so that the VFS layer correctly free's it when it
2033 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2035 struct inode *inode = old_dentry->d_inode;
2036 int they_are_dirs = S_ISDIR(inode->i_mode);
2038 if (!simple_empty(new_dentry))
2041 if (new_dentry->d_inode) {
2042 (void) shmem_unlink(new_dir, new_dentry);
2044 drop_nlink(old_dir);
2045 } else if (they_are_dirs) {
2046 drop_nlink(old_dir);
2050 old_dir->i_size -= BOGO_DIRENT_SIZE;
2051 new_dir->i_size += BOGO_DIRENT_SIZE;
2052 old_dir->i_ctime = old_dir->i_mtime =
2053 new_dir->i_ctime = new_dir->i_mtime =
2054 inode->i_ctime = CURRENT_TIME;
2058 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2062 struct inode *inode;
2065 struct shmem_inode_info *info;
2067 len = strlen(symname) + 1;
2068 if (len > PAGE_CACHE_SIZE)
2069 return -ENAMETOOLONG;
2071 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2075 error = security_inode_init_security(inode, dir, &dentry->d_name,
2076 shmem_initxattrs, NULL);
2078 if (error != -EOPNOTSUPP) {
2085 info = SHMEM_I(inode);
2086 inode->i_size = len-1;
2087 if (len <= SHORT_SYMLINK_LEN) {
2088 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2089 if (!info->symlink) {
2093 inode->i_op = &shmem_short_symlink_operations;
2095 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2100 inode->i_mapping->a_ops = &shmem_aops;
2101 inode->i_op = &shmem_symlink_inode_operations;
2102 kaddr = kmap_atomic(page);
2103 memcpy(kaddr, symname, len);
2104 kunmap_atomic(kaddr);
2105 SetPageUptodate(page);
2106 set_page_dirty(page);
2108 page_cache_release(page);
2110 dir->i_size += BOGO_DIRENT_SIZE;
2111 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2112 d_instantiate(dentry, inode);
2117 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2119 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2123 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2125 struct page *page = NULL;
2126 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2127 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2133 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2135 if (!IS_ERR(nd_get_link(nd))) {
2136 struct page *page = cookie;
2138 mark_page_accessed(page);
2139 page_cache_release(page);
2143 #ifdef CONFIG_TMPFS_XATTR
2145 * Superblocks without xattr inode operations may get some security.* xattr
2146 * support from the LSM "for free". As soon as we have any other xattrs
2147 * like ACLs, we also need to implement the security.* handlers at
2148 * filesystem level, though.
2152 * Callback for security_inode_init_security() for acquiring xattrs.
2154 static int shmem_initxattrs(struct inode *inode,
2155 const struct xattr *xattr_array,
2158 struct shmem_inode_info *info = SHMEM_I(inode);
2159 const struct xattr *xattr;
2160 struct simple_xattr *new_xattr;
2163 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2164 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
2168 len = strlen(xattr->name) + 1;
2169 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2171 if (!new_xattr->name) {
2176 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2177 XATTR_SECURITY_PREFIX_LEN);
2178 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2181 simple_xattr_list_add(&info->xattrs, new_xattr);
2187 static const struct xattr_handler *shmem_xattr_handlers[] = {
2188 #ifdef CONFIG_TMPFS_POSIX_ACL
2189 &posix_acl_access_xattr_handler,
2190 &posix_acl_default_xattr_handler,
2195 static int shmem_xattr_validate(const char *name)
2197 struct { const char *prefix; size_t len; } arr[] = {
2198 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2199 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2203 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2204 size_t preflen = arr[i].len;
2205 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2214 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2215 void *buffer, size_t size)
2217 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2221 * If this is a request for a synthetic attribute in the system.*
2222 * namespace use the generic infrastructure to resolve a handler
2223 * for it via sb->s_xattr.
2225 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2226 return generic_getxattr(dentry, name, buffer, size);
2228 err = shmem_xattr_validate(name);
2232 return simple_xattr_get(&info->xattrs, name, buffer, size);
2235 static int shmem_setxattr(struct dentry *dentry, const char *name,
2236 const void *value, size_t size, int flags)
2238 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2242 * If this is a request for a synthetic attribute in the system.*
2243 * namespace use the generic infrastructure to resolve a handler
2244 * for it via sb->s_xattr.
2246 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2247 return generic_setxattr(dentry, name, value, size, flags);
2249 err = shmem_xattr_validate(name);
2253 return simple_xattr_set(&info->xattrs, name, value, size, flags);
2256 static int shmem_removexattr(struct dentry *dentry, const char *name)
2258 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2262 * If this is a request for a synthetic attribute in the system.*
2263 * namespace use the generic infrastructure to resolve a handler
2264 * for it via sb->s_xattr.
2266 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2267 return generic_removexattr(dentry, name);
2269 err = shmem_xattr_validate(name);
2273 return simple_xattr_remove(&info->xattrs, name);
2276 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2278 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2279 return simple_xattr_list(&info->xattrs, buffer, size);
2281 #endif /* CONFIG_TMPFS_XATTR */
2283 static const struct inode_operations shmem_short_symlink_operations = {
2284 .readlink = generic_readlink,
2285 .follow_link = shmem_follow_short_symlink,
2286 #ifdef CONFIG_TMPFS_XATTR
2287 .setxattr = shmem_setxattr,
2288 .getxattr = shmem_getxattr,
2289 .listxattr = shmem_listxattr,
2290 .removexattr = shmem_removexattr,
2294 static const struct inode_operations shmem_symlink_inode_operations = {
2295 .readlink = generic_readlink,
2296 .follow_link = shmem_follow_link,
2297 .put_link = shmem_put_link,
2298 #ifdef CONFIG_TMPFS_XATTR
2299 .setxattr = shmem_setxattr,
2300 .getxattr = shmem_getxattr,
2301 .listxattr = shmem_listxattr,
2302 .removexattr = shmem_removexattr,
2306 static struct dentry *shmem_get_parent(struct dentry *child)
2308 return ERR_PTR(-ESTALE);
2311 static int shmem_match(struct inode *ino, void *vfh)
2315 inum = (inum << 32) | fh[1];
2316 return ino->i_ino == inum && fh[0] == ino->i_generation;
2319 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2320 struct fid *fid, int fh_len, int fh_type)
2322 struct inode *inode;
2323 struct dentry *dentry = NULL;
2330 inum = (inum << 32) | fid->raw[1];
2332 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2333 shmem_match, fid->raw);
2335 dentry = d_find_alias(inode);
2342 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2343 struct inode *parent)
2347 return FILEID_INVALID;
2350 if (inode_unhashed(inode)) {
2351 /* Unfortunately insert_inode_hash is not idempotent,
2352 * so as we hash inodes here rather than at creation
2353 * time, we need a lock to ensure we only try
2356 static DEFINE_SPINLOCK(lock);
2358 if (inode_unhashed(inode))
2359 __insert_inode_hash(inode,
2360 inode->i_ino + inode->i_generation);
2364 fh[0] = inode->i_generation;
2365 fh[1] = inode->i_ino;
2366 fh[2] = ((__u64)inode->i_ino) >> 32;
2372 static const struct export_operations shmem_export_ops = {
2373 .get_parent = shmem_get_parent,
2374 .encode_fh = shmem_encode_fh,
2375 .fh_to_dentry = shmem_fh_to_dentry,
2378 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2381 char *this_char, *value, *rest;
2382 struct mempolicy *mpol = NULL;
2386 while (options != NULL) {
2387 this_char = options;
2390 * NUL-terminate this option: unfortunately,
2391 * mount options form a comma-separated list,
2392 * but mpol's nodelist may also contain commas.
2394 options = strchr(options, ',');
2395 if (options == NULL)
2398 if (!isdigit(*options)) {
2405 if ((value = strchr(this_char,'=')) != NULL) {
2409 "tmpfs: No value for mount option '%s'\n",
2414 if (!strcmp(this_char,"size")) {
2415 unsigned long long size;
2416 size = memparse(value,&rest);
2418 size <<= PAGE_SHIFT;
2419 size *= totalram_pages;
2425 sbinfo->max_blocks =
2426 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2427 } else if (!strcmp(this_char,"nr_blocks")) {
2428 sbinfo->max_blocks = memparse(value, &rest);
2431 } else if (!strcmp(this_char,"nr_inodes")) {
2432 sbinfo->max_inodes = memparse(value, &rest);
2435 } else if (!strcmp(this_char,"mode")) {
2438 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2441 } else if (!strcmp(this_char,"uid")) {
2444 uid = simple_strtoul(value, &rest, 0);
2447 sbinfo->uid = make_kuid(current_user_ns(), uid);
2448 if (!uid_valid(sbinfo->uid))
2450 } else if (!strcmp(this_char,"gid")) {
2453 gid = simple_strtoul(value, &rest, 0);
2456 sbinfo->gid = make_kgid(current_user_ns(), gid);
2457 if (!gid_valid(sbinfo->gid))
2459 } else if (!strcmp(this_char,"mpol")) {
2462 if (mpol_parse_str(value, &mpol))
2465 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2470 sbinfo->mpol = mpol;
2474 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2482 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2484 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2485 struct shmem_sb_info config = *sbinfo;
2486 unsigned long inodes;
2487 int error = -EINVAL;
2490 if (shmem_parse_options(data, &config, true))
2493 spin_lock(&sbinfo->stat_lock);
2494 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2495 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2497 if (config.max_inodes < inodes)
2500 * Those tests disallow limited->unlimited while any are in use;
2501 * but we must separately disallow unlimited->limited, because
2502 * in that case we have no record of how much is already in use.
2504 if (config.max_blocks && !sbinfo->max_blocks)
2506 if (config.max_inodes && !sbinfo->max_inodes)
2510 sbinfo->max_blocks = config.max_blocks;
2511 sbinfo->max_inodes = config.max_inodes;
2512 sbinfo->free_inodes = config.max_inodes - inodes;
2515 * Preserve previous mempolicy unless mpol remount option was specified.
2518 mpol_put(sbinfo->mpol);
2519 sbinfo->mpol = config.mpol; /* transfers initial ref */
2522 spin_unlock(&sbinfo->stat_lock);
2526 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2528 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2530 if (sbinfo->max_blocks != shmem_default_max_blocks())
2531 seq_printf(seq, ",size=%luk",
2532 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2533 if (sbinfo->max_inodes != shmem_default_max_inodes())
2534 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2535 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2536 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2537 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2538 seq_printf(seq, ",uid=%u",
2539 from_kuid_munged(&init_user_ns, sbinfo->uid));
2540 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2541 seq_printf(seq, ",gid=%u",
2542 from_kgid_munged(&init_user_ns, sbinfo->gid));
2543 shmem_show_mpol(seq, sbinfo->mpol);
2546 #endif /* CONFIG_TMPFS */
2548 static void shmem_put_super(struct super_block *sb)
2550 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2552 percpu_counter_destroy(&sbinfo->used_blocks);
2553 mpol_put(sbinfo->mpol);
2555 sb->s_fs_info = NULL;
2558 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2560 struct inode *inode;
2561 struct shmem_sb_info *sbinfo;
2564 /* Round up to L1_CACHE_BYTES to resist false sharing */
2565 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2566 L1_CACHE_BYTES), GFP_KERNEL);
2570 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2571 sbinfo->uid = current_fsuid();
2572 sbinfo->gid = current_fsgid();
2573 sb->s_fs_info = sbinfo;
2577 * Per default we only allow half of the physical ram per
2578 * tmpfs instance, limiting inodes to one per page of lowmem;
2579 * but the internal instance is left unlimited.
2581 if (!(sb->s_flags & MS_KERNMOUNT)) {
2582 sbinfo->max_blocks = shmem_default_max_blocks();
2583 sbinfo->max_inodes = shmem_default_max_inodes();
2584 if (shmem_parse_options(data, sbinfo, false)) {
2589 sb->s_flags |= MS_NOUSER;
2591 sb->s_export_op = &shmem_export_ops;
2592 sb->s_flags |= MS_NOSEC;
2594 sb->s_flags |= MS_NOUSER;
2597 spin_lock_init(&sbinfo->stat_lock);
2598 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2600 sbinfo->free_inodes = sbinfo->max_inodes;
2602 sb->s_maxbytes = MAX_LFS_FILESIZE;
2603 sb->s_blocksize = PAGE_CACHE_SIZE;
2604 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2605 sb->s_magic = TMPFS_MAGIC;
2606 sb->s_op = &shmem_ops;
2607 sb->s_time_gran = 1;
2608 #ifdef CONFIG_TMPFS_XATTR
2609 sb->s_xattr = shmem_xattr_handlers;
2611 #ifdef CONFIG_TMPFS_POSIX_ACL
2612 sb->s_flags |= MS_POSIXACL;
2615 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2618 inode->i_uid = sbinfo->uid;
2619 inode->i_gid = sbinfo->gid;
2620 sb->s_root = d_make_root(inode);
2626 shmem_put_super(sb);
2630 static struct kmem_cache *shmem_inode_cachep;
2632 static struct inode *shmem_alloc_inode(struct super_block *sb)
2634 struct shmem_inode_info *info;
2635 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2638 return &info->vfs_inode;
2641 static void shmem_destroy_callback(struct rcu_head *head)
2643 struct inode *inode = container_of(head, struct inode, i_rcu);
2644 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2647 static void shmem_destroy_inode(struct inode *inode)
2649 if (S_ISREG(inode->i_mode))
2650 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2651 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2654 static void shmem_init_inode(void *foo)
2656 struct shmem_inode_info *info = foo;
2657 inode_init_once(&info->vfs_inode);
2660 static int shmem_init_inodecache(void)
2662 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2663 sizeof(struct shmem_inode_info),
2664 0, SLAB_PANIC, shmem_init_inode);
2668 static void shmem_destroy_inodecache(void)
2670 kmem_cache_destroy(shmem_inode_cachep);
2673 static const struct address_space_operations shmem_aops = {
2674 .writepage = shmem_writepage,
2675 .set_page_dirty = __set_page_dirty_no_writeback,
2677 .write_begin = shmem_write_begin,
2678 .write_end = shmem_write_end,
2680 .migratepage = migrate_page,
2681 .error_remove_page = generic_error_remove_page,
2684 static const struct file_operations shmem_file_operations = {
2687 .llseek = shmem_file_llseek,
2688 .read = do_sync_read,
2689 .write = do_sync_write,
2690 .aio_read = shmem_file_aio_read,
2691 .aio_write = generic_file_aio_write,
2692 .fsync = noop_fsync,
2693 .splice_read = shmem_file_splice_read,
2694 .splice_write = generic_file_splice_write,
2695 .fallocate = shmem_fallocate,
2699 static const struct inode_operations shmem_inode_operations = {
2700 .setattr = shmem_setattr,
2701 #ifdef CONFIG_TMPFS_XATTR
2702 .setxattr = shmem_setxattr,
2703 .getxattr = shmem_getxattr,
2704 .listxattr = shmem_listxattr,
2705 .removexattr = shmem_removexattr,
2706 .set_acl = simple_set_acl,
2710 static const struct inode_operations shmem_dir_inode_operations = {
2712 .create = shmem_create,
2713 .lookup = simple_lookup,
2715 .unlink = shmem_unlink,
2716 .symlink = shmem_symlink,
2717 .mkdir = shmem_mkdir,
2718 .rmdir = shmem_rmdir,
2719 .mknod = shmem_mknod,
2720 .rename = shmem_rename,
2721 .tmpfile = shmem_tmpfile,
2723 #ifdef CONFIG_TMPFS_XATTR
2724 .setxattr = shmem_setxattr,
2725 .getxattr = shmem_getxattr,
2726 .listxattr = shmem_listxattr,
2727 .removexattr = shmem_removexattr,
2729 #ifdef CONFIG_TMPFS_POSIX_ACL
2730 .setattr = shmem_setattr,
2731 .set_acl = simple_set_acl,
2735 static const struct inode_operations shmem_special_inode_operations = {
2736 #ifdef CONFIG_TMPFS_XATTR
2737 .setxattr = shmem_setxattr,
2738 .getxattr = shmem_getxattr,
2739 .listxattr = shmem_listxattr,
2740 .removexattr = shmem_removexattr,
2742 #ifdef CONFIG_TMPFS_POSIX_ACL
2743 .setattr = shmem_setattr,
2744 .set_acl = simple_set_acl,
2748 static const struct super_operations shmem_ops = {
2749 .alloc_inode = shmem_alloc_inode,
2750 .destroy_inode = shmem_destroy_inode,
2752 .statfs = shmem_statfs,
2753 .remount_fs = shmem_remount_fs,
2754 .show_options = shmem_show_options,
2756 .evict_inode = shmem_evict_inode,
2757 .drop_inode = generic_delete_inode,
2758 .put_super = shmem_put_super,
2761 static const struct vm_operations_struct shmem_vm_ops = {
2762 .fault = shmem_fault,
2764 .set_policy = shmem_set_policy,
2765 .get_policy = shmem_get_policy,
2767 .remap_pages = generic_file_remap_pages,
2770 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2771 int flags, const char *dev_name, void *data)
2773 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2776 static struct file_system_type shmem_fs_type = {
2777 .owner = THIS_MODULE,
2779 .mount = shmem_mount,
2780 .kill_sb = kill_litter_super,
2781 .fs_flags = FS_USERNS_MOUNT,
2784 int __init shmem_init(void)
2788 /* If rootfs called this, don't re-init */
2789 if (shmem_inode_cachep)
2792 error = bdi_init(&shmem_backing_dev_info);
2796 error = shmem_init_inodecache();
2800 error = register_filesystem(&shmem_fs_type);
2802 printk(KERN_ERR "Could not register tmpfs\n");
2806 shm_mnt = kern_mount(&shmem_fs_type);
2807 if (IS_ERR(shm_mnt)) {
2808 error = PTR_ERR(shm_mnt);
2809 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2815 unregister_filesystem(&shmem_fs_type);
2817 shmem_destroy_inodecache();
2819 bdi_destroy(&shmem_backing_dev_info);
2821 shm_mnt = ERR_PTR(error);
2825 #else /* !CONFIG_SHMEM */
2828 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2830 * This is intended for small system where the benefits of the full
2831 * shmem code (swap-backed and resource-limited) are outweighed by
2832 * their complexity. On systems without swap this code should be
2833 * effectively equivalent, but much lighter weight.
2836 static struct file_system_type shmem_fs_type = {
2838 .mount = ramfs_mount,
2839 .kill_sb = kill_litter_super,
2840 .fs_flags = FS_USERNS_MOUNT,
2843 int __init shmem_init(void)
2845 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2847 shm_mnt = kern_mount(&shmem_fs_type);
2848 BUG_ON(IS_ERR(shm_mnt));
2853 int shmem_unuse(swp_entry_t swap, struct page *page)
2858 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2863 void shmem_unlock_mapping(struct address_space *mapping)
2867 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2869 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2871 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2873 #define shmem_vm_ops generic_file_vm_ops
2874 #define shmem_file_operations ramfs_file_operations
2875 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2876 #define shmem_acct_size(flags, size) 0
2877 #define shmem_unacct_size(flags, size) do {} while (0)
2879 #endif /* CONFIG_SHMEM */
2883 static struct dentry_operations anon_ops = {
2884 .d_dname = simple_dname
2887 static struct file *__shmem_file_setup(const char *name, loff_t size,
2888 unsigned long flags, unsigned int i_flags)
2891 struct inode *inode;
2893 struct super_block *sb;
2896 if (IS_ERR(shm_mnt))
2897 return ERR_CAST(shm_mnt);
2899 if (size < 0 || size > MAX_LFS_FILESIZE)
2900 return ERR_PTR(-EINVAL);
2902 if (shmem_acct_size(flags, size))
2903 return ERR_PTR(-ENOMEM);
2905 res = ERR_PTR(-ENOMEM);
2907 this.len = strlen(name);
2908 this.hash = 0; /* will go */
2909 sb = shm_mnt->mnt_sb;
2910 path.dentry = d_alloc_pseudo(sb, &this);
2913 d_set_d_op(path.dentry, &anon_ops);
2914 path.mnt = mntget(shm_mnt);
2916 res = ERR_PTR(-ENOSPC);
2917 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2921 inode->i_flags |= i_flags;
2922 d_instantiate(path.dentry, inode);
2923 inode->i_size = size;
2924 clear_nlink(inode); /* It is unlinked */
2925 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
2929 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2930 &shmem_file_operations);
2939 shmem_unacct_size(flags, size);
2944 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
2945 * kernel internal. There will be NO LSM permission checks against the
2946 * underlying inode. So users of this interface must do LSM checks at a
2947 * higher layer. The one user is the big_key implementation. LSM checks
2948 * are provided at the key level rather than the inode level.
2949 * @name: name for dentry (to be seen in /proc/<pid>/maps
2950 * @size: size to be set for the file
2951 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2953 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
2955 return __shmem_file_setup(name, size, flags, S_PRIVATE);
2959 * shmem_file_setup - get an unlinked file living in tmpfs
2960 * @name: name for dentry (to be seen in /proc/<pid>/maps
2961 * @size: size to be set for the file
2962 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2964 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2966 return __shmem_file_setup(name, size, flags, 0);
2968 EXPORT_SYMBOL_GPL(shmem_file_setup);
2971 * shmem_zero_setup - setup a shared anonymous mapping
2972 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2974 int shmem_zero_setup(struct vm_area_struct *vma)
2977 loff_t size = vma->vm_end - vma->vm_start;
2979 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2981 return PTR_ERR(file);
2985 vma->vm_file = file;
2986 vma->vm_ops = &shmem_vm_ops;
2991 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2992 * @mapping: the page's address_space
2993 * @index: the page index
2994 * @gfp: the page allocator flags to use if allocating
2996 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2997 * with any new page allocations done using the specified allocation flags.
2998 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2999 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3000 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3002 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3003 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3005 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3006 pgoff_t index, gfp_t gfp)
3009 struct inode *inode = mapping->host;
3013 BUG_ON(mapping->a_ops != &shmem_aops);
3014 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3016 page = ERR_PTR(error);
3022 * The tiny !SHMEM case uses ramfs without swap
3024 return read_cache_page_gfp(mapping, index, gfp);
3027 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);