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 VM_BUG_ON(!replacement);
249 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
252 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
253 if (item != expected)
255 radix_tree_replace_slot(pslot, replacement);
260 * Sometimes, before we decide whether to proceed or to fail, we must check
261 * that an entry was not already brought back from swap by a racing thread.
263 * Checking page is not enough: by the time a SwapCache page is locked, it
264 * might be reused, and again be SwapCache, using the same swap as before.
266 static bool shmem_confirm_swap(struct address_space *mapping,
267 pgoff_t index, swp_entry_t swap)
272 item = radix_tree_lookup(&mapping->page_tree, index);
274 return item == swp_to_radix_entry(swap);
278 * Like add_to_page_cache_locked, but error if expected item has gone.
280 static int shmem_add_to_page_cache(struct page *page,
281 struct address_space *mapping,
282 pgoff_t index, gfp_t gfp, void *expected)
286 VM_BUG_ON_PAGE(!PageLocked(page), page);
287 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
289 page_cache_get(page);
290 page->mapping = mapping;
293 spin_lock_irq(&mapping->tree_lock);
295 error = radix_tree_insert(&mapping->page_tree, index, page);
297 error = shmem_radix_tree_replace(mapping, index, expected,
301 __inc_zone_page_state(page, NR_FILE_PAGES);
302 __inc_zone_page_state(page, NR_SHMEM);
303 spin_unlock_irq(&mapping->tree_lock);
305 page->mapping = NULL;
306 spin_unlock_irq(&mapping->tree_lock);
307 page_cache_release(page);
313 * Like delete_from_page_cache, but substitutes swap for page.
315 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
317 struct address_space *mapping = page->mapping;
320 spin_lock_irq(&mapping->tree_lock);
321 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
322 page->mapping = NULL;
324 __dec_zone_page_state(page, NR_FILE_PAGES);
325 __dec_zone_page_state(page, NR_SHMEM);
326 spin_unlock_irq(&mapping->tree_lock);
327 page_cache_release(page);
332 * Remove swap entry from radix tree, free the swap and its page cache.
334 static int shmem_free_swap(struct address_space *mapping,
335 pgoff_t index, void *radswap)
339 spin_lock_irq(&mapping->tree_lock);
340 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
341 spin_unlock_irq(&mapping->tree_lock);
344 free_swap_and_cache(radix_to_swp_entry(radswap));
349 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
351 void shmem_unlock_mapping(struct address_space *mapping)
354 pgoff_t indices[PAGEVEC_SIZE];
357 pagevec_init(&pvec, 0);
359 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
361 while (!mapping_unevictable(mapping)) {
363 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
364 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
366 pvec.nr = find_get_entries(mapping, index,
367 PAGEVEC_SIZE, pvec.pages, indices);
370 index = indices[pvec.nr - 1] + 1;
371 pagevec_remove_exceptionals(&pvec);
372 check_move_unevictable_pages(pvec.pages, pvec.nr);
373 pagevec_release(&pvec);
379 * Remove range of pages and swap entries from radix tree, and free them.
380 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
382 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
385 struct address_space *mapping = inode->i_mapping;
386 struct shmem_inode_info *info = SHMEM_I(inode);
387 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
388 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
389 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
390 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
392 pgoff_t indices[PAGEVEC_SIZE];
393 long nr_swaps_freed = 0;
398 end = -1; /* unsigned, so actually very big */
400 pagevec_init(&pvec, 0);
402 while (index < end) {
403 pvec.nr = find_get_entries(mapping, index,
404 min(end - index, (pgoff_t)PAGEVEC_SIZE),
405 pvec.pages, indices);
408 mem_cgroup_uncharge_start();
409 for (i = 0; i < pagevec_count(&pvec); i++) {
410 struct page *page = pvec.pages[i];
416 if (radix_tree_exceptional_entry(page)) {
419 nr_swaps_freed += !shmem_free_swap(mapping,
424 if (!trylock_page(page))
426 if (!unfalloc || !PageUptodate(page)) {
427 if (page->mapping == mapping) {
428 VM_BUG_ON_PAGE(PageWriteback(page), page);
429 truncate_inode_page(mapping, page);
434 pagevec_remove_exceptionals(&pvec);
435 pagevec_release(&pvec);
436 mem_cgroup_uncharge_end();
442 struct page *page = NULL;
443 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
445 unsigned int top = PAGE_CACHE_SIZE;
450 zero_user_segment(page, partial_start, top);
451 set_page_dirty(page);
453 page_cache_release(page);
457 struct page *page = NULL;
458 shmem_getpage(inode, end, &page, SGP_READ, NULL);
460 zero_user_segment(page, 0, partial_end);
461 set_page_dirty(page);
463 page_cache_release(page);
473 pvec.nr = find_get_entries(mapping, index,
474 min(end - index, (pgoff_t)PAGEVEC_SIZE),
475 pvec.pages, indices);
477 if (index == start || unfalloc)
482 if ((index == start || unfalloc) && indices[0] >= end) {
483 pagevec_remove_exceptionals(&pvec);
484 pagevec_release(&pvec);
487 mem_cgroup_uncharge_start();
488 for (i = 0; i < pagevec_count(&pvec); i++) {
489 struct page *page = pvec.pages[i];
495 if (radix_tree_exceptional_entry(page)) {
498 nr_swaps_freed += !shmem_free_swap(mapping,
504 if (!unfalloc || !PageUptodate(page)) {
505 if (page->mapping == mapping) {
506 VM_BUG_ON_PAGE(PageWriteback(page), page);
507 truncate_inode_page(mapping, page);
512 pagevec_remove_exceptionals(&pvec);
513 pagevec_release(&pvec);
514 mem_cgroup_uncharge_end();
518 spin_lock(&info->lock);
519 info->swapped -= nr_swaps_freed;
520 shmem_recalc_inode(inode);
521 spin_unlock(&info->lock);
524 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
526 shmem_undo_range(inode, lstart, lend, false);
527 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
529 EXPORT_SYMBOL_GPL(shmem_truncate_range);
531 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
533 struct inode *inode = dentry->d_inode;
536 error = inode_change_ok(inode, attr);
540 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
541 loff_t oldsize = inode->i_size;
542 loff_t newsize = attr->ia_size;
544 if (newsize != oldsize) {
545 i_size_write(inode, newsize);
546 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
548 if (newsize < oldsize) {
549 loff_t holebegin = round_up(newsize, PAGE_SIZE);
550 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
551 shmem_truncate_range(inode, newsize, (loff_t)-1);
552 /* unmap again to remove racily COWed private pages */
553 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
557 setattr_copy(inode, attr);
558 if (attr->ia_valid & ATTR_MODE)
559 error = posix_acl_chmod(inode, inode->i_mode);
563 static void shmem_evict_inode(struct inode *inode)
565 struct shmem_inode_info *info = SHMEM_I(inode);
567 if (inode->i_mapping->a_ops == &shmem_aops) {
568 shmem_unacct_size(info->flags, inode->i_size);
570 shmem_truncate_range(inode, 0, (loff_t)-1);
571 if (!list_empty(&info->swaplist)) {
572 mutex_lock(&shmem_swaplist_mutex);
573 list_del_init(&info->swaplist);
574 mutex_unlock(&shmem_swaplist_mutex);
577 kfree(info->symlink);
579 simple_xattrs_free(&info->xattrs);
580 WARN_ON(inode->i_blocks);
581 shmem_free_inode(inode->i_sb);
586 * If swap found in inode, free it and move page from swapcache to filecache.
588 static int shmem_unuse_inode(struct shmem_inode_info *info,
589 swp_entry_t swap, struct page **pagep)
591 struct address_space *mapping = info->vfs_inode.i_mapping;
597 radswap = swp_to_radix_entry(swap);
598 index = radix_tree_locate_item(&mapping->page_tree, radswap);
603 * Move _head_ to start search for next from here.
604 * But be careful: shmem_evict_inode checks list_empty without taking
605 * mutex, and there's an instant in list_move_tail when info->swaplist
606 * would appear empty, if it were the only one on shmem_swaplist.
608 if (shmem_swaplist.next != &info->swaplist)
609 list_move_tail(&shmem_swaplist, &info->swaplist);
611 gfp = mapping_gfp_mask(mapping);
612 if (shmem_should_replace_page(*pagep, gfp)) {
613 mutex_unlock(&shmem_swaplist_mutex);
614 error = shmem_replace_page(pagep, gfp, info, index);
615 mutex_lock(&shmem_swaplist_mutex);
617 * We needed to drop mutex to make that restrictive page
618 * allocation, but the inode might have been freed while we
619 * dropped it: although a racing shmem_evict_inode() cannot
620 * complete without emptying the radix_tree, our page lock
621 * on this swapcache page is not enough to prevent that -
622 * free_swap_and_cache() of our swap entry will only
623 * trylock_page(), removing swap from radix_tree whatever.
625 * We must not proceed to shmem_add_to_page_cache() if the
626 * inode has been freed, but of course we cannot rely on
627 * inode or mapping or info to check that. However, we can
628 * safely check if our swap entry is still in use (and here
629 * it can't have got reused for another page): if it's still
630 * in use, then the inode cannot have been freed yet, and we
631 * can safely proceed (if it's no longer in use, that tells
632 * nothing about the inode, but we don't need to unuse swap).
634 if (!page_swapcount(*pagep))
639 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
640 * but also to hold up shmem_evict_inode(): so inode cannot be freed
641 * beneath us (pagelock doesn't help until the page is in pagecache).
644 error = shmem_add_to_page_cache(*pagep, mapping, index,
645 GFP_NOWAIT, radswap);
646 if (error != -ENOMEM) {
648 * Truncation and eviction use free_swap_and_cache(), which
649 * only does trylock page: if we raced, best clean up here.
651 delete_from_swap_cache(*pagep);
652 set_page_dirty(*pagep);
654 spin_lock(&info->lock);
656 spin_unlock(&info->lock);
659 error = 1; /* not an error, but entry was found */
665 * Search through swapped inodes to find and replace swap by page.
667 int shmem_unuse(swp_entry_t swap, struct page *page)
669 struct list_head *this, *next;
670 struct shmem_inode_info *info;
675 * There's a faint possibility that swap page was replaced before
676 * caller locked it: caller will come back later with the right page.
678 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
682 * Charge page using GFP_KERNEL while we can wait, before taking
683 * the shmem_swaplist_mutex which might hold up shmem_writepage().
684 * Charged back to the user (not to caller) when swap account is used.
686 error = mem_cgroup_charge_file(page, current->mm, GFP_KERNEL);
689 /* No radix_tree_preload: swap entry keeps a place for page in tree */
691 mutex_lock(&shmem_swaplist_mutex);
692 list_for_each_safe(this, next, &shmem_swaplist) {
693 info = list_entry(this, struct shmem_inode_info, swaplist);
695 found = shmem_unuse_inode(info, swap, &page);
697 list_del_init(&info->swaplist);
702 mutex_unlock(&shmem_swaplist_mutex);
708 page_cache_release(page);
713 * Move the page from the page cache to the swap cache.
715 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
717 struct shmem_inode_info *info;
718 struct address_space *mapping;
723 BUG_ON(!PageLocked(page));
724 mapping = page->mapping;
726 inode = mapping->host;
727 info = SHMEM_I(inode);
728 if (info->flags & VM_LOCKED)
730 if (!total_swap_pages)
734 * shmem_backing_dev_info's capabilities prevent regular writeback or
735 * sync from ever calling shmem_writepage; but a stacking filesystem
736 * might use ->writepage of its underlying filesystem, in which case
737 * tmpfs should write out to swap only in response to memory pressure,
738 * and not for the writeback threads or sync.
740 if (!wbc->for_reclaim) {
741 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
746 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
747 * value into swapfile.c, the only way we can correctly account for a
748 * fallocated page arriving here is now to initialize it and write it.
750 * That's okay for a page already fallocated earlier, but if we have
751 * not yet completed the fallocation, then (a) we want to keep track
752 * of this page in case we have to undo it, and (b) it may not be a
753 * good idea to continue anyway, once we're pushing into swap. So
754 * reactivate the page, and let shmem_fallocate() quit when too many.
756 if (!PageUptodate(page)) {
757 if (inode->i_private) {
758 struct shmem_falloc *shmem_falloc;
759 spin_lock(&inode->i_lock);
760 shmem_falloc = inode->i_private;
762 index >= shmem_falloc->start &&
763 index < shmem_falloc->next)
764 shmem_falloc->nr_unswapped++;
767 spin_unlock(&inode->i_lock);
771 clear_highpage(page);
772 flush_dcache_page(page);
773 SetPageUptodate(page);
776 swap = get_swap_page();
781 * Add inode to shmem_unuse()'s list of swapped-out inodes,
782 * if it's not already there. Do it now before the page is
783 * moved to swap cache, when its pagelock no longer protects
784 * the inode from eviction. But don't unlock the mutex until
785 * we've incremented swapped, because shmem_unuse_inode() will
786 * prune a !swapped inode from the swaplist under this mutex.
788 mutex_lock(&shmem_swaplist_mutex);
789 if (list_empty(&info->swaplist))
790 list_add_tail(&info->swaplist, &shmem_swaplist);
792 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
793 swap_shmem_alloc(swap);
794 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
796 spin_lock(&info->lock);
798 shmem_recalc_inode(inode);
799 spin_unlock(&info->lock);
801 mutex_unlock(&shmem_swaplist_mutex);
802 BUG_ON(page_mapped(page));
803 swap_writepage(page, wbc);
807 mutex_unlock(&shmem_swaplist_mutex);
808 swapcache_free(swap, NULL);
810 set_page_dirty(page);
811 if (wbc->for_reclaim)
812 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
819 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
823 if (!mpol || mpol->mode == MPOL_DEFAULT)
824 return; /* show nothing */
826 mpol_to_str(buffer, sizeof(buffer), mpol);
828 seq_printf(seq, ",mpol=%s", buffer);
831 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
833 struct mempolicy *mpol = NULL;
835 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
838 spin_unlock(&sbinfo->stat_lock);
842 #endif /* CONFIG_TMPFS */
844 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
845 struct shmem_inode_info *info, pgoff_t index)
847 struct vm_area_struct pvma;
850 /* Create a pseudo vma that just contains the policy */
852 /* Bias interleave by inode number to distribute better across nodes */
853 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
855 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
857 page = swapin_readahead(swap, gfp, &pvma, 0);
859 /* Drop reference taken by mpol_shared_policy_lookup() */
860 mpol_cond_put(pvma.vm_policy);
865 static struct page *shmem_alloc_page(gfp_t gfp,
866 struct shmem_inode_info *info, pgoff_t index)
868 struct vm_area_struct pvma;
871 /* Create a pseudo vma that just contains the policy */
873 /* Bias interleave by inode number to distribute better across nodes */
874 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
876 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
878 page = alloc_page_vma(gfp, &pvma, 0);
880 /* Drop reference taken by mpol_shared_policy_lookup() */
881 mpol_cond_put(pvma.vm_policy);
885 #else /* !CONFIG_NUMA */
887 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
890 #endif /* CONFIG_TMPFS */
892 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
893 struct shmem_inode_info *info, pgoff_t index)
895 return swapin_readahead(swap, gfp, NULL, 0);
898 static inline struct page *shmem_alloc_page(gfp_t gfp,
899 struct shmem_inode_info *info, pgoff_t index)
901 return alloc_page(gfp);
903 #endif /* CONFIG_NUMA */
905 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
906 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
913 * When a page is moved from swapcache to shmem filecache (either by the
914 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
915 * shmem_unuse_inode()), it may have been read in earlier from swap, in
916 * ignorance of the mapping it belongs to. If that mapping has special
917 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
918 * we may need to copy to a suitable page before moving to filecache.
920 * In a future release, this may well be extended to respect cpuset and
921 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
922 * but for now it is a simple matter of zone.
924 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
926 return page_zonenum(page) > gfp_zone(gfp);
929 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
930 struct shmem_inode_info *info, pgoff_t index)
932 struct page *oldpage, *newpage;
933 struct address_space *swap_mapping;
938 swap_index = page_private(oldpage);
939 swap_mapping = page_mapping(oldpage);
942 * We have arrived here because our zones are constrained, so don't
943 * limit chance of success by further cpuset and node constraints.
945 gfp &= ~GFP_CONSTRAINT_MASK;
946 newpage = shmem_alloc_page(gfp, info, index);
950 page_cache_get(newpage);
951 copy_highpage(newpage, oldpage);
952 flush_dcache_page(newpage);
954 __set_page_locked(newpage);
955 SetPageUptodate(newpage);
956 SetPageSwapBacked(newpage);
957 set_page_private(newpage, swap_index);
958 SetPageSwapCache(newpage);
961 * Our caller will very soon move newpage out of swapcache, but it's
962 * a nice clean interface for us to replace oldpage by newpage there.
964 spin_lock_irq(&swap_mapping->tree_lock);
965 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
968 __inc_zone_page_state(newpage, NR_FILE_PAGES);
969 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
971 spin_unlock_irq(&swap_mapping->tree_lock);
973 if (unlikely(error)) {
975 * Is this possible? I think not, now that our callers check
976 * both PageSwapCache and page_private after getting page lock;
977 * but be defensive. Reverse old to newpage for clear and free.
981 mem_cgroup_replace_page_cache(oldpage, newpage);
982 lru_cache_add_anon(newpage);
986 ClearPageSwapCache(oldpage);
987 set_page_private(oldpage, 0);
989 unlock_page(oldpage);
990 page_cache_release(oldpage);
991 page_cache_release(oldpage);
996 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
998 * If we allocate a new one we do not mark it dirty. That's up to the
999 * vm. If we swap it in we mark it dirty since we also free the swap
1000 * entry since a page cannot live in both the swap and page cache
1002 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1003 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1005 struct address_space *mapping = inode->i_mapping;
1006 struct shmem_inode_info *info;
1007 struct shmem_sb_info *sbinfo;
1014 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1018 page = find_lock_entry(mapping, index);
1019 if (radix_tree_exceptional_entry(page)) {
1020 swap = radix_to_swp_entry(page);
1024 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1025 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1030 /* fallocated page? */
1031 if (page && !PageUptodate(page)) {
1032 if (sgp != SGP_READ)
1035 page_cache_release(page);
1038 if (page || (sgp == SGP_READ && !swap.val)) {
1044 * Fast cache lookup did not find it:
1045 * bring it back from swap or allocate.
1047 info = SHMEM_I(inode);
1048 sbinfo = SHMEM_SB(inode->i_sb);
1051 /* Look it up and read it in.. */
1052 page = lookup_swap_cache(swap);
1054 /* here we actually do the io */
1056 *fault_type |= VM_FAULT_MAJOR;
1057 page = shmem_swapin(swap, gfp, info, index);
1064 /* We have to do this with page locked to prevent races */
1066 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1067 !shmem_confirm_swap(mapping, index, swap)) {
1068 error = -EEXIST; /* try again */
1071 if (!PageUptodate(page)) {
1075 wait_on_page_writeback(page);
1077 if (shmem_should_replace_page(page, gfp)) {
1078 error = shmem_replace_page(&page, gfp, info, index);
1083 error = mem_cgroup_charge_file(page, current->mm,
1084 gfp & GFP_RECLAIM_MASK);
1086 error = shmem_add_to_page_cache(page, mapping, index,
1087 gfp, swp_to_radix_entry(swap));
1089 * We already confirmed swap under page lock, and make
1090 * no memory allocation here, so usually no possibility
1091 * of error; but free_swap_and_cache() only trylocks a
1092 * page, so it is just possible that the entry has been
1093 * truncated or holepunched since swap was confirmed.
1094 * shmem_undo_range() will have done some of the
1095 * unaccounting, now delete_from_swap_cache() will do
1096 * the rest (including mem_cgroup_uncharge_swapcache).
1097 * Reset swap.val? No, leave it so "failed" goes back to
1098 * "repeat": reading a hole and writing should succeed.
1101 delete_from_swap_cache(page);
1106 spin_lock(&info->lock);
1108 shmem_recalc_inode(inode);
1109 spin_unlock(&info->lock);
1111 delete_from_swap_cache(page);
1112 set_page_dirty(page);
1116 if (shmem_acct_block(info->flags)) {
1120 if (sbinfo->max_blocks) {
1121 if (percpu_counter_compare(&sbinfo->used_blocks,
1122 sbinfo->max_blocks) >= 0) {
1126 percpu_counter_inc(&sbinfo->used_blocks);
1129 page = shmem_alloc_page(gfp, info, index);
1135 SetPageSwapBacked(page);
1136 __set_page_locked(page);
1137 error = mem_cgroup_charge_file(page, current->mm,
1138 gfp & GFP_RECLAIM_MASK);
1141 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
1143 error = shmem_add_to_page_cache(page, mapping, index,
1145 radix_tree_preload_end();
1148 mem_cgroup_uncharge_cache_page(page);
1151 lru_cache_add_anon(page);
1153 spin_lock(&info->lock);
1155 inode->i_blocks += BLOCKS_PER_PAGE;
1156 shmem_recalc_inode(inode);
1157 spin_unlock(&info->lock);
1161 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1163 if (sgp == SGP_FALLOC)
1167 * Let SGP_WRITE caller clear ends if write does not fill page;
1168 * but SGP_FALLOC on a page fallocated earlier must initialize
1169 * it now, lest undo on failure cancel our earlier guarantee.
1171 if (sgp != SGP_WRITE) {
1172 clear_highpage(page);
1173 flush_dcache_page(page);
1174 SetPageUptodate(page);
1176 if (sgp == SGP_DIRTY)
1177 set_page_dirty(page);
1180 /* Perhaps the file has been truncated since we checked */
1181 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1182 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1196 info = SHMEM_I(inode);
1197 ClearPageDirty(page);
1198 delete_from_page_cache(page);
1199 spin_lock(&info->lock);
1201 inode->i_blocks -= BLOCKS_PER_PAGE;
1202 spin_unlock(&info->lock);
1204 sbinfo = SHMEM_SB(inode->i_sb);
1205 if (sbinfo->max_blocks)
1206 percpu_counter_add(&sbinfo->used_blocks, -1);
1208 shmem_unacct_blocks(info->flags, 1);
1210 if (swap.val && error != -EINVAL &&
1211 !shmem_confirm_swap(mapping, index, swap))
1216 page_cache_release(page);
1218 if (error == -ENOSPC && !once++) {
1219 info = SHMEM_I(inode);
1220 spin_lock(&info->lock);
1221 shmem_recalc_inode(inode);
1222 spin_unlock(&info->lock);
1225 if (error == -EEXIST) /* from above or from radix_tree_insert */
1230 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1232 struct inode *inode = file_inode(vma->vm_file);
1234 int ret = VM_FAULT_LOCKED;
1236 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1238 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1240 if (ret & VM_FAULT_MAJOR) {
1241 count_vm_event(PGMAJFAULT);
1242 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1248 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1250 struct inode *inode = file_inode(vma->vm_file);
1251 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1254 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1257 struct inode *inode = file_inode(vma->vm_file);
1260 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1261 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1265 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1267 struct inode *inode = file_inode(file);
1268 struct shmem_inode_info *info = SHMEM_I(inode);
1269 int retval = -ENOMEM;
1271 spin_lock(&info->lock);
1272 if (lock && !(info->flags & VM_LOCKED)) {
1273 if (!user_shm_lock(inode->i_size, user))
1275 info->flags |= VM_LOCKED;
1276 mapping_set_unevictable(file->f_mapping);
1278 if (!lock && (info->flags & VM_LOCKED) && user) {
1279 user_shm_unlock(inode->i_size, user);
1280 info->flags &= ~VM_LOCKED;
1281 mapping_clear_unevictable(file->f_mapping);
1286 spin_unlock(&info->lock);
1290 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1292 file_accessed(file);
1293 vma->vm_ops = &shmem_vm_ops;
1297 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1298 umode_t mode, dev_t dev, unsigned long flags)
1300 struct inode *inode;
1301 struct shmem_inode_info *info;
1302 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1304 if (shmem_reserve_inode(sb))
1307 inode = new_inode(sb);
1309 inode->i_ino = get_next_ino();
1310 inode_init_owner(inode, dir, mode);
1311 inode->i_blocks = 0;
1312 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1313 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1314 inode->i_generation = get_seconds();
1315 info = SHMEM_I(inode);
1316 memset(info, 0, (char *)inode - (char *)info);
1317 spin_lock_init(&info->lock);
1318 info->flags = flags & VM_NORESERVE;
1319 INIT_LIST_HEAD(&info->swaplist);
1320 simple_xattrs_init(&info->xattrs);
1321 cache_no_acl(inode);
1323 switch (mode & S_IFMT) {
1325 inode->i_op = &shmem_special_inode_operations;
1326 init_special_inode(inode, mode, dev);
1329 inode->i_mapping->a_ops = &shmem_aops;
1330 inode->i_op = &shmem_inode_operations;
1331 inode->i_fop = &shmem_file_operations;
1332 mpol_shared_policy_init(&info->policy,
1333 shmem_get_sbmpol(sbinfo));
1337 /* Some things misbehave if size == 0 on a directory */
1338 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1339 inode->i_op = &shmem_dir_inode_operations;
1340 inode->i_fop = &simple_dir_operations;
1344 * Must not load anything in the rbtree,
1345 * mpol_free_shared_policy will not be called.
1347 mpol_shared_policy_init(&info->policy, NULL);
1351 shmem_free_inode(sb);
1355 bool shmem_mapping(struct address_space *mapping)
1357 return mapping->backing_dev_info == &shmem_backing_dev_info;
1361 static const struct inode_operations shmem_symlink_inode_operations;
1362 static const struct inode_operations shmem_short_symlink_operations;
1364 #ifdef CONFIG_TMPFS_XATTR
1365 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1367 #define shmem_initxattrs NULL
1371 shmem_write_begin(struct file *file, struct address_space *mapping,
1372 loff_t pos, unsigned len, unsigned flags,
1373 struct page **pagep, void **fsdata)
1375 struct inode *inode = mapping->host;
1376 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1377 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1381 shmem_write_end(struct file *file, struct address_space *mapping,
1382 loff_t pos, unsigned len, unsigned copied,
1383 struct page *page, void *fsdata)
1385 struct inode *inode = mapping->host;
1387 if (pos + copied > inode->i_size)
1388 i_size_write(inode, pos + copied);
1390 if (!PageUptodate(page)) {
1391 if (copied < PAGE_CACHE_SIZE) {
1392 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1393 zero_user_segments(page, 0, from,
1394 from + copied, PAGE_CACHE_SIZE);
1396 SetPageUptodate(page);
1398 set_page_dirty(page);
1400 page_cache_release(page);
1405 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1406 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1408 struct file *file = iocb->ki_filp;
1409 struct inode *inode = file_inode(file);
1410 struct address_space *mapping = inode->i_mapping;
1412 unsigned long offset;
1413 enum sgp_type sgp = SGP_READ;
1416 size_t count = iov_length(iov, nr_segs);
1417 loff_t *ppos = &iocb->ki_pos;
1418 struct iov_iter iter;
1420 iov_iter_init(&iter, READ, iov, nr_segs, count);
1423 * Might this read be for a stacking filesystem? Then when reading
1424 * holes of a sparse file, we actually need to allocate those pages,
1425 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1427 if (segment_eq(get_fs(), KERNEL_DS))
1430 index = *ppos >> PAGE_CACHE_SHIFT;
1431 offset = *ppos & ~PAGE_CACHE_MASK;
1434 struct page *page = NULL;
1436 unsigned long nr, ret;
1437 loff_t i_size = i_size_read(inode);
1439 end_index = i_size >> PAGE_CACHE_SHIFT;
1440 if (index > end_index)
1442 if (index == end_index) {
1443 nr = i_size & ~PAGE_CACHE_MASK;
1448 error = shmem_getpage(inode, index, &page, sgp, NULL);
1450 if (error == -EINVAL)
1458 * We must evaluate after, since reads (unlike writes)
1459 * are called without i_mutex protection against truncate
1461 nr = PAGE_CACHE_SIZE;
1462 i_size = i_size_read(inode);
1463 end_index = i_size >> PAGE_CACHE_SHIFT;
1464 if (index == end_index) {
1465 nr = i_size & ~PAGE_CACHE_MASK;
1468 page_cache_release(page);
1476 * If users can be writing to this page using arbitrary
1477 * virtual addresses, take care about potential aliasing
1478 * before reading the page on the kernel side.
1480 if (mapping_writably_mapped(mapping))
1481 flush_dcache_page(page);
1483 * Mark the page accessed if we read the beginning.
1486 mark_page_accessed(page);
1488 page = ZERO_PAGE(0);
1489 page_cache_get(page);
1493 * Ok, we have the page, and it's up-to-date, so
1494 * now we can copy it to user space...
1496 ret = copy_page_to_iter(page, offset, nr, &iter);
1499 index += offset >> PAGE_CACHE_SHIFT;
1500 offset &= ~PAGE_CACHE_MASK;
1502 page_cache_release(page);
1503 if (!iov_iter_count(&iter))
1512 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1513 file_accessed(file);
1514 return retval ? retval : error;
1517 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1518 struct pipe_inode_info *pipe, size_t len,
1521 struct address_space *mapping = in->f_mapping;
1522 struct inode *inode = mapping->host;
1523 unsigned int loff, nr_pages, req_pages;
1524 struct page *pages[PIPE_DEF_BUFFERS];
1525 struct partial_page partial[PIPE_DEF_BUFFERS];
1527 pgoff_t index, end_index;
1530 struct splice_pipe_desc spd = {
1533 .nr_pages_max = PIPE_DEF_BUFFERS,
1535 .ops = &page_cache_pipe_buf_ops,
1536 .spd_release = spd_release_page,
1539 isize = i_size_read(inode);
1540 if (unlikely(*ppos >= isize))
1543 left = isize - *ppos;
1544 if (unlikely(left < len))
1547 if (splice_grow_spd(pipe, &spd))
1550 index = *ppos >> PAGE_CACHE_SHIFT;
1551 loff = *ppos & ~PAGE_CACHE_MASK;
1552 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1553 nr_pages = min(req_pages, spd.nr_pages_max);
1555 spd.nr_pages = find_get_pages_contig(mapping, index,
1556 nr_pages, spd.pages);
1557 index += spd.nr_pages;
1560 while (spd.nr_pages < nr_pages) {
1561 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1565 spd.pages[spd.nr_pages++] = page;
1569 index = *ppos >> PAGE_CACHE_SHIFT;
1570 nr_pages = spd.nr_pages;
1573 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1574 unsigned int this_len;
1579 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1580 page = spd.pages[page_nr];
1582 if (!PageUptodate(page) || page->mapping != mapping) {
1583 error = shmem_getpage(inode, index, &page,
1588 page_cache_release(spd.pages[page_nr]);
1589 spd.pages[page_nr] = page;
1592 isize = i_size_read(inode);
1593 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1594 if (unlikely(!isize || index > end_index))
1597 if (end_index == index) {
1600 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1604 this_len = min(this_len, plen - loff);
1608 spd.partial[page_nr].offset = loff;
1609 spd.partial[page_nr].len = this_len;
1616 while (page_nr < nr_pages)
1617 page_cache_release(spd.pages[page_nr++]);
1620 error = splice_to_pipe(pipe, &spd);
1622 splice_shrink_spd(&spd);
1632 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1634 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1635 pgoff_t index, pgoff_t end, int whence)
1638 struct pagevec pvec;
1639 pgoff_t indices[PAGEVEC_SIZE];
1643 pagevec_init(&pvec, 0);
1644 pvec.nr = 1; /* start small: we may be there already */
1646 pvec.nr = find_get_entries(mapping, index,
1647 pvec.nr, pvec.pages, indices);
1649 if (whence == SEEK_DATA)
1653 for (i = 0; i < pvec.nr; i++, index++) {
1654 if (index < indices[i]) {
1655 if (whence == SEEK_HOLE) {
1661 page = pvec.pages[i];
1662 if (page && !radix_tree_exceptional_entry(page)) {
1663 if (!PageUptodate(page))
1667 (page && whence == SEEK_DATA) ||
1668 (!page && whence == SEEK_HOLE)) {
1673 pagevec_remove_exceptionals(&pvec);
1674 pagevec_release(&pvec);
1675 pvec.nr = PAGEVEC_SIZE;
1681 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
1683 struct address_space *mapping = file->f_mapping;
1684 struct inode *inode = mapping->host;
1688 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1689 return generic_file_llseek_size(file, offset, whence,
1690 MAX_LFS_FILESIZE, i_size_read(inode));
1691 mutex_lock(&inode->i_mutex);
1692 /* We're holding i_mutex so we can access i_size directly */
1696 else if (offset >= inode->i_size)
1699 start = offset >> PAGE_CACHE_SHIFT;
1700 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1701 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
1702 new_offset <<= PAGE_CACHE_SHIFT;
1703 if (new_offset > offset) {
1704 if (new_offset < inode->i_size)
1705 offset = new_offset;
1706 else if (whence == SEEK_DATA)
1709 offset = inode->i_size;
1714 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
1715 mutex_unlock(&inode->i_mutex);
1719 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1722 struct inode *inode = file_inode(file);
1723 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1724 struct shmem_falloc shmem_falloc;
1725 pgoff_t start, index, end;
1728 mutex_lock(&inode->i_mutex);
1730 if (mode & FALLOC_FL_PUNCH_HOLE) {
1731 struct address_space *mapping = file->f_mapping;
1732 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1733 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1735 if ((u64)unmap_end > (u64)unmap_start)
1736 unmap_mapping_range(mapping, unmap_start,
1737 1 + unmap_end - unmap_start, 0);
1738 shmem_truncate_range(inode, offset, offset + len - 1);
1739 /* No need to unmap again: hole-punching leaves COWed pages */
1744 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1745 error = inode_newsize_ok(inode, offset + len);
1749 start = offset >> PAGE_CACHE_SHIFT;
1750 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1751 /* Try to avoid a swapstorm if len is impossible to satisfy */
1752 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1757 shmem_falloc.start = start;
1758 shmem_falloc.next = start;
1759 shmem_falloc.nr_falloced = 0;
1760 shmem_falloc.nr_unswapped = 0;
1761 spin_lock(&inode->i_lock);
1762 inode->i_private = &shmem_falloc;
1763 spin_unlock(&inode->i_lock);
1765 for (index = start; index < end; index++) {
1769 * Good, the fallocate(2) manpage permits EINTR: we may have
1770 * been interrupted because we are using up too much memory.
1772 if (signal_pending(current))
1774 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1777 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1780 /* Remove the !PageUptodate pages we added */
1781 shmem_undo_range(inode,
1782 (loff_t)start << PAGE_CACHE_SHIFT,
1783 (loff_t)index << PAGE_CACHE_SHIFT, true);
1788 * Inform shmem_writepage() how far we have reached.
1789 * No need for lock or barrier: we have the page lock.
1791 shmem_falloc.next++;
1792 if (!PageUptodate(page))
1793 shmem_falloc.nr_falloced++;
1796 * If !PageUptodate, leave it that way so that freeable pages
1797 * can be recognized if we need to rollback on error later.
1798 * But set_page_dirty so that memory pressure will swap rather
1799 * than free the pages we are allocating (and SGP_CACHE pages
1800 * might still be clean: we now need to mark those dirty too).
1802 set_page_dirty(page);
1804 page_cache_release(page);
1808 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1809 i_size_write(inode, offset + len);
1810 inode->i_ctime = CURRENT_TIME;
1812 spin_lock(&inode->i_lock);
1813 inode->i_private = NULL;
1814 spin_unlock(&inode->i_lock);
1816 mutex_unlock(&inode->i_mutex);
1820 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1822 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1824 buf->f_type = TMPFS_MAGIC;
1825 buf->f_bsize = PAGE_CACHE_SIZE;
1826 buf->f_namelen = NAME_MAX;
1827 if (sbinfo->max_blocks) {
1828 buf->f_blocks = sbinfo->max_blocks;
1830 buf->f_bfree = sbinfo->max_blocks -
1831 percpu_counter_sum(&sbinfo->used_blocks);
1833 if (sbinfo->max_inodes) {
1834 buf->f_files = sbinfo->max_inodes;
1835 buf->f_ffree = sbinfo->free_inodes;
1837 /* else leave those fields 0 like simple_statfs */
1842 * File creation. Allocate an inode, and we're done..
1845 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1847 struct inode *inode;
1848 int error = -ENOSPC;
1850 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1852 error = simple_acl_create(dir, inode);
1855 error = security_inode_init_security(inode, dir,
1857 shmem_initxattrs, NULL);
1858 if (error && error != -EOPNOTSUPP)
1862 dir->i_size += BOGO_DIRENT_SIZE;
1863 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1864 d_instantiate(dentry, inode);
1865 dget(dentry); /* Extra count - pin the dentry in core */
1874 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
1876 struct inode *inode;
1877 int error = -ENOSPC;
1879 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
1881 error = security_inode_init_security(inode, dir,
1883 shmem_initxattrs, NULL);
1884 if (error && error != -EOPNOTSUPP)
1886 error = simple_acl_create(dir, inode);
1889 d_tmpfile(dentry, inode);
1897 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1901 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1907 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1910 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1916 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1918 struct inode *inode = old_dentry->d_inode;
1922 * No ordinary (disk based) filesystem counts links as inodes;
1923 * but each new link needs a new dentry, pinning lowmem, and
1924 * tmpfs dentries cannot be pruned until they are unlinked.
1926 ret = shmem_reserve_inode(inode->i_sb);
1930 dir->i_size += BOGO_DIRENT_SIZE;
1931 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1933 ihold(inode); /* New dentry reference */
1934 dget(dentry); /* Extra pinning count for the created dentry */
1935 d_instantiate(dentry, inode);
1940 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1942 struct inode *inode = dentry->d_inode;
1944 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1945 shmem_free_inode(inode->i_sb);
1947 dir->i_size -= BOGO_DIRENT_SIZE;
1948 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1950 dput(dentry); /* Undo the count from "create" - this does all the work */
1954 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1956 if (!simple_empty(dentry))
1959 drop_nlink(dentry->d_inode);
1961 return shmem_unlink(dir, dentry);
1965 * The VFS layer already does all the dentry stuff for rename,
1966 * we just have to decrement the usage count for the target if
1967 * it exists so that the VFS layer correctly free's it when it
1970 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1972 struct inode *inode = old_dentry->d_inode;
1973 int they_are_dirs = S_ISDIR(inode->i_mode);
1975 if (!simple_empty(new_dentry))
1978 if (new_dentry->d_inode) {
1979 (void) shmem_unlink(new_dir, new_dentry);
1981 drop_nlink(old_dir);
1982 } else if (they_are_dirs) {
1983 drop_nlink(old_dir);
1987 old_dir->i_size -= BOGO_DIRENT_SIZE;
1988 new_dir->i_size += BOGO_DIRENT_SIZE;
1989 old_dir->i_ctime = old_dir->i_mtime =
1990 new_dir->i_ctime = new_dir->i_mtime =
1991 inode->i_ctime = CURRENT_TIME;
1995 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1999 struct inode *inode;
2002 struct shmem_inode_info *info;
2004 len = strlen(symname) + 1;
2005 if (len > PAGE_CACHE_SIZE)
2006 return -ENAMETOOLONG;
2008 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2012 error = security_inode_init_security(inode, dir, &dentry->d_name,
2013 shmem_initxattrs, NULL);
2015 if (error != -EOPNOTSUPP) {
2022 info = SHMEM_I(inode);
2023 inode->i_size = len-1;
2024 if (len <= SHORT_SYMLINK_LEN) {
2025 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2026 if (!info->symlink) {
2030 inode->i_op = &shmem_short_symlink_operations;
2032 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2037 inode->i_mapping->a_ops = &shmem_aops;
2038 inode->i_op = &shmem_symlink_inode_operations;
2039 kaddr = kmap_atomic(page);
2040 memcpy(kaddr, symname, len);
2041 kunmap_atomic(kaddr);
2042 SetPageUptodate(page);
2043 set_page_dirty(page);
2045 page_cache_release(page);
2047 dir->i_size += BOGO_DIRENT_SIZE;
2048 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2049 d_instantiate(dentry, inode);
2054 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2056 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2060 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2062 struct page *page = NULL;
2063 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2064 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2070 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2072 if (!IS_ERR(nd_get_link(nd))) {
2073 struct page *page = cookie;
2075 mark_page_accessed(page);
2076 page_cache_release(page);
2080 #ifdef CONFIG_TMPFS_XATTR
2082 * Superblocks without xattr inode operations may get some security.* xattr
2083 * support from the LSM "for free". As soon as we have any other xattrs
2084 * like ACLs, we also need to implement the security.* handlers at
2085 * filesystem level, though.
2089 * Callback for security_inode_init_security() for acquiring xattrs.
2091 static int shmem_initxattrs(struct inode *inode,
2092 const struct xattr *xattr_array,
2095 struct shmem_inode_info *info = SHMEM_I(inode);
2096 const struct xattr *xattr;
2097 struct simple_xattr *new_xattr;
2100 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2101 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
2105 len = strlen(xattr->name) + 1;
2106 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2108 if (!new_xattr->name) {
2113 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2114 XATTR_SECURITY_PREFIX_LEN);
2115 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2118 simple_xattr_list_add(&info->xattrs, new_xattr);
2124 static const struct xattr_handler *shmem_xattr_handlers[] = {
2125 #ifdef CONFIG_TMPFS_POSIX_ACL
2126 &posix_acl_access_xattr_handler,
2127 &posix_acl_default_xattr_handler,
2132 static int shmem_xattr_validate(const char *name)
2134 struct { const char *prefix; size_t len; } arr[] = {
2135 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2136 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2140 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2141 size_t preflen = arr[i].len;
2142 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2151 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2152 void *buffer, size_t size)
2154 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2158 * If this is a request for a synthetic attribute in the system.*
2159 * namespace use the generic infrastructure to resolve a handler
2160 * for it via sb->s_xattr.
2162 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2163 return generic_getxattr(dentry, name, buffer, size);
2165 err = shmem_xattr_validate(name);
2169 return simple_xattr_get(&info->xattrs, name, buffer, size);
2172 static int shmem_setxattr(struct dentry *dentry, const char *name,
2173 const void *value, size_t size, int flags)
2175 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2179 * If this is a request for a synthetic attribute in the system.*
2180 * namespace use the generic infrastructure to resolve a handler
2181 * for it via sb->s_xattr.
2183 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2184 return generic_setxattr(dentry, name, value, size, flags);
2186 err = shmem_xattr_validate(name);
2190 return simple_xattr_set(&info->xattrs, name, value, size, flags);
2193 static int shmem_removexattr(struct dentry *dentry, const char *name)
2195 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2199 * If this is a request for a synthetic attribute in the system.*
2200 * namespace use the generic infrastructure to resolve a handler
2201 * for it via sb->s_xattr.
2203 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2204 return generic_removexattr(dentry, name);
2206 err = shmem_xattr_validate(name);
2210 return simple_xattr_remove(&info->xattrs, name);
2213 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2215 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2216 return simple_xattr_list(&info->xattrs, buffer, size);
2218 #endif /* CONFIG_TMPFS_XATTR */
2220 static const struct inode_operations shmem_short_symlink_operations = {
2221 .readlink = generic_readlink,
2222 .follow_link = shmem_follow_short_symlink,
2223 #ifdef CONFIG_TMPFS_XATTR
2224 .setxattr = shmem_setxattr,
2225 .getxattr = shmem_getxattr,
2226 .listxattr = shmem_listxattr,
2227 .removexattr = shmem_removexattr,
2231 static const struct inode_operations shmem_symlink_inode_operations = {
2232 .readlink = generic_readlink,
2233 .follow_link = shmem_follow_link,
2234 .put_link = shmem_put_link,
2235 #ifdef CONFIG_TMPFS_XATTR
2236 .setxattr = shmem_setxattr,
2237 .getxattr = shmem_getxattr,
2238 .listxattr = shmem_listxattr,
2239 .removexattr = shmem_removexattr,
2243 static struct dentry *shmem_get_parent(struct dentry *child)
2245 return ERR_PTR(-ESTALE);
2248 static int shmem_match(struct inode *ino, void *vfh)
2252 inum = (inum << 32) | fh[1];
2253 return ino->i_ino == inum && fh[0] == ino->i_generation;
2256 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2257 struct fid *fid, int fh_len, int fh_type)
2259 struct inode *inode;
2260 struct dentry *dentry = NULL;
2267 inum = (inum << 32) | fid->raw[1];
2269 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2270 shmem_match, fid->raw);
2272 dentry = d_find_alias(inode);
2279 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2280 struct inode *parent)
2284 return FILEID_INVALID;
2287 if (inode_unhashed(inode)) {
2288 /* Unfortunately insert_inode_hash is not idempotent,
2289 * so as we hash inodes here rather than at creation
2290 * time, we need a lock to ensure we only try
2293 static DEFINE_SPINLOCK(lock);
2295 if (inode_unhashed(inode))
2296 __insert_inode_hash(inode,
2297 inode->i_ino + inode->i_generation);
2301 fh[0] = inode->i_generation;
2302 fh[1] = inode->i_ino;
2303 fh[2] = ((__u64)inode->i_ino) >> 32;
2309 static const struct export_operations shmem_export_ops = {
2310 .get_parent = shmem_get_parent,
2311 .encode_fh = shmem_encode_fh,
2312 .fh_to_dentry = shmem_fh_to_dentry,
2315 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2318 char *this_char, *value, *rest;
2319 struct mempolicy *mpol = NULL;
2323 while (options != NULL) {
2324 this_char = options;
2327 * NUL-terminate this option: unfortunately,
2328 * mount options form a comma-separated list,
2329 * but mpol's nodelist may also contain commas.
2331 options = strchr(options, ',');
2332 if (options == NULL)
2335 if (!isdigit(*options)) {
2342 if ((value = strchr(this_char,'=')) != NULL) {
2346 "tmpfs: No value for mount option '%s'\n",
2351 if (!strcmp(this_char,"size")) {
2352 unsigned long long size;
2353 size = memparse(value,&rest);
2355 size <<= PAGE_SHIFT;
2356 size *= totalram_pages;
2362 sbinfo->max_blocks =
2363 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2364 } else if (!strcmp(this_char,"nr_blocks")) {
2365 sbinfo->max_blocks = memparse(value, &rest);
2368 } else if (!strcmp(this_char,"nr_inodes")) {
2369 sbinfo->max_inodes = memparse(value, &rest);
2372 } else if (!strcmp(this_char,"mode")) {
2375 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2378 } else if (!strcmp(this_char,"uid")) {
2381 uid = simple_strtoul(value, &rest, 0);
2384 sbinfo->uid = make_kuid(current_user_ns(), uid);
2385 if (!uid_valid(sbinfo->uid))
2387 } else if (!strcmp(this_char,"gid")) {
2390 gid = simple_strtoul(value, &rest, 0);
2393 sbinfo->gid = make_kgid(current_user_ns(), gid);
2394 if (!gid_valid(sbinfo->gid))
2396 } else if (!strcmp(this_char,"mpol")) {
2399 if (mpol_parse_str(value, &mpol))
2402 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2407 sbinfo->mpol = mpol;
2411 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2419 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2421 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2422 struct shmem_sb_info config = *sbinfo;
2423 unsigned long inodes;
2424 int error = -EINVAL;
2427 if (shmem_parse_options(data, &config, true))
2430 spin_lock(&sbinfo->stat_lock);
2431 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2432 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2434 if (config.max_inodes < inodes)
2437 * Those tests disallow limited->unlimited while any are in use;
2438 * but we must separately disallow unlimited->limited, because
2439 * in that case we have no record of how much is already in use.
2441 if (config.max_blocks && !sbinfo->max_blocks)
2443 if (config.max_inodes && !sbinfo->max_inodes)
2447 sbinfo->max_blocks = config.max_blocks;
2448 sbinfo->max_inodes = config.max_inodes;
2449 sbinfo->free_inodes = config.max_inodes - inodes;
2452 * Preserve previous mempolicy unless mpol remount option was specified.
2455 mpol_put(sbinfo->mpol);
2456 sbinfo->mpol = config.mpol; /* transfers initial ref */
2459 spin_unlock(&sbinfo->stat_lock);
2463 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2465 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2467 if (sbinfo->max_blocks != shmem_default_max_blocks())
2468 seq_printf(seq, ",size=%luk",
2469 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2470 if (sbinfo->max_inodes != shmem_default_max_inodes())
2471 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2472 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2473 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2474 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2475 seq_printf(seq, ",uid=%u",
2476 from_kuid_munged(&init_user_ns, sbinfo->uid));
2477 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2478 seq_printf(seq, ",gid=%u",
2479 from_kgid_munged(&init_user_ns, sbinfo->gid));
2480 shmem_show_mpol(seq, sbinfo->mpol);
2483 #endif /* CONFIG_TMPFS */
2485 static void shmem_put_super(struct super_block *sb)
2487 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2489 percpu_counter_destroy(&sbinfo->used_blocks);
2490 mpol_put(sbinfo->mpol);
2492 sb->s_fs_info = NULL;
2495 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2497 struct inode *inode;
2498 struct shmem_sb_info *sbinfo;
2501 /* Round up to L1_CACHE_BYTES to resist false sharing */
2502 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2503 L1_CACHE_BYTES), GFP_KERNEL);
2507 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2508 sbinfo->uid = current_fsuid();
2509 sbinfo->gid = current_fsgid();
2510 sb->s_fs_info = sbinfo;
2514 * Per default we only allow half of the physical ram per
2515 * tmpfs instance, limiting inodes to one per page of lowmem;
2516 * but the internal instance is left unlimited.
2518 if (!(sb->s_flags & MS_KERNMOUNT)) {
2519 sbinfo->max_blocks = shmem_default_max_blocks();
2520 sbinfo->max_inodes = shmem_default_max_inodes();
2521 if (shmem_parse_options(data, sbinfo, false)) {
2526 sb->s_flags |= MS_NOUSER;
2528 sb->s_export_op = &shmem_export_ops;
2529 sb->s_flags |= MS_NOSEC;
2531 sb->s_flags |= MS_NOUSER;
2534 spin_lock_init(&sbinfo->stat_lock);
2535 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2537 sbinfo->free_inodes = sbinfo->max_inodes;
2539 sb->s_maxbytes = MAX_LFS_FILESIZE;
2540 sb->s_blocksize = PAGE_CACHE_SIZE;
2541 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2542 sb->s_magic = TMPFS_MAGIC;
2543 sb->s_op = &shmem_ops;
2544 sb->s_time_gran = 1;
2545 #ifdef CONFIG_TMPFS_XATTR
2546 sb->s_xattr = shmem_xattr_handlers;
2548 #ifdef CONFIG_TMPFS_POSIX_ACL
2549 sb->s_flags |= MS_POSIXACL;
2552 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2555 inode->i_uid = sbinfo->uid;
2556 inode->i_gid = sbinfo->gid;
2557 sb->s_root = d_make_root(inode);
2563 shmem_put_super(sb);
2567 static struct kmem_cache *shmem_inode_cachep;
2569 static struct inode *shmem_alloc_inode(struct super_block *sb)
2571 struct shmem_inode_info *info;
2572 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2575 return &info->vfs_inode;
2578 static void shmem_destroy_callback(struct rcu_head *head)
2580 struct inode *inode = container_of(head, struct inode, i_rcu);
2581 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2584 static void shmem_destroy_inode(struct inode *inode)
2586 if (S_ISREG(inode->i_mode))
2587 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2588 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2591 static void shmem_init_inode(void *foo)
2593 struct shmem_inode_info *info = foo;
2594 inode_init_once(&info->vfs_inode);
2597 static int shmem_init_inodecache(void)
2599 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2600 sizeof(struct shmem_inode_info),
2601 0, SLAB_PANIC, shmem_init_inode);
2605 static void shmem_destroy_inodecache(void)
2607 kmem_cache_destroy(shmem_inode_cachep);
2610 static const struct address_space_operations shmem_aops = {
2611 .writepage = shmem_writepage,
2612 .set_page_dirty = __set_page_dirty_no_writeback,
2614 .write_begin = shmem_write_begin,
2615 .write_end = shmem_write_end,
2617 .migratepage = migrate_page,
2618 .error_remove_page = generic_error_remove_page,
2621 static const struct file_operations shmem_file_operations = {
2624 .llseek = shmem_file_llseek,
2625 .read = do_sync_read,
2626 .write = do_sync_write,
2627 .aio_read = shmem_file_aio_read,
2628 .aio_write = generic_file_aio_write,
2629 .fsync = noop_fsync,
2630 .splice_read = shmem_file_splice_read,
2631 .splice_write = generic_file_splice_write,
2632 .fallocate = shmem_fallocate,
2636 static const struct inode_operations shmem_inode_operations = {
2637 .setattr = shmem_setattr,
2638 #ifdef CONFIG_TMPFS_XATTR
2639 .setxattr = shmem_setxattr,
2640 .getxattr = shmem_getxattr,
2641 .listxattr = shmem_listxattr,
2642 .removexattr = shmem_removexattr,
2643 .set_acl = simple_set_acl,
2647 static const struct inode_operations shmem_dir_inode_operations = {
2649 .create = shmem_create,
2650 .lookup = simple_lookup,
2652 .unlink = shmem_unlink,
2653 .symlink = shmem_symlink,
2654 .mkdir = shmem_mkdir,
2655 .rmdir = shmem_rmdir,
2656 .mknod = shmem_mknod,
2657 .rename = shmem_rename,
2658 .tmpfile = shmem_tmpfile,
2660 #ifdef CONFIG_TMPFS_XATTR
2661 .setxattr = shmem_setxattr,
2662 .getxattr = shmem_getxattr,
2663 .listxattr = shmem_listxattr,
2664 .removexattr = shmem_removexattr,
2666 #ifdef CONFIG_TMPFS_POSIX_ACL
2667 .setattr = shmem_setattr,
2668 .set_acl = simple_set_acl,
2672 static const struct inode_operations shmem_special_inode_operations = {
2673 #ifdef CONFIG_TMPFS_XATTR
2674 .setxattr = shmem_setxattr,
2675 .getxattr = shmem_getxattr,
2676 .listxattr = shmem_listxattr,
2677 .removexattr = shmem_removexattr,
2679 #ifdef CONFIG_TMPFS_POSIX_ACL
2680 .setattr = shmem_setattr,
2681 .set_acl = simple_set_acl,
2685 static const struct super_operations shmem_ops = {
2686 .alloc_inode = shmem_alloc_inode,
2687 .destroy_inode = shmem_destroy_inode,
2689 .statfs = shmem_statfs,
2690 .remount_fs = shmem_remount_fs,
2691 .show_options = shmem_show_options,
2693 .evict_inode = shmem_evict_inode,
2694 .drop_inode = generic_delete_inode,
2695 .put_super = shmem_put_super,
2698 static const struct vm_operations_struct shmem_vm_ops = {
2699 .fault = shmem_fault,
2700 .map_pages = filemap_map_pages,
2702 .set_policy = shmem_set_policy,
2703 .get_policy = shmem_get_policy,
2705 .remap_pages = generic_file_remap_pages,
2708 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2709 int flags, const char *dev_name, void *data)
2711 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2714 static struct file_system_type shmem_fs_type = {
2715 .owner = THIS_MODULE,
2717 .mount = shmem_mount,
2718 .kill_sb = kill_litter_super,
2719 .fs_flags = FS_USERNS_MOUNT,
2722 int __init shmem_init(void)
2726 /* If rootfs called this, don't re-init */
2727 if (shmem_inode_cachep)
2730 error = bdi_init(&shmem_backing_dev_info);
2734 error = shmem_init_inodecache();
2738 error = register_filesystem(&shmem_fs_type);
2740 printk(KERN_ERR "Could not register tmpfs\n");
2744 shm_mnt = kern_mount(&shmem_fs_type);
2745 if (IS_ERR(shm_mnt)) {
2746 error = PTR_ERR(shm_mnt);
2747 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2753 unregister_filesystem(&shmem_fs_type);
2755 shmem_destroy_inodecache();
2757 bdi_destroy(&shmem_backing_dev_info);
2759 shm_mnt = ERR_PTR(error);
2763 #else /* !CONFIG_SHMEM */
2766 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2768 * This is intended for small system where the benefits of the full
2769 * shmem code (swap-backed and resource-limited) are outweighed by
2770 * their complexity. On systems without swap this code should be
2771 * effectively equivalent, but much lighter weight.
2774 static struct file_system_type shmem_fs_type = {
2776 .mount = ramfs_mount,
2777 .kill_sb = kill_litter_super,
2778 .fs_flags = FS_USERNS_MOUNT,
2781 int __init shmem_init(void)
2783 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2785 shm_mnt = kern_mount(&shmem_fs_type);
2786 BUG_ON(IS_ERR(shm_mnt));
2791 int shmem_unuse(swp_entry_t swap, struct page *page)
2796 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2801 void shmem_unlock_mapping(struct address_space *mapping)
2805 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2807 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2809 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2811 #define shmem_vm_ops generic_file_vm_ops
2812 #define shmem_file_operations ramfs_file_operations
2813 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2814 #define shmem_acct_size(flags, size) 0
2815 #define shmem_unacct_size(flags, size) do {} while (0)
2817 #endif /* CONFIG_SHMEM */
2821 static struct dentry_operations anon_ops = {
2822 .d_dname = simple_dname
2825 static struct file *__shmem_file_setup(const char *name, loff_t size,
2826 unsigned long flags, unsigned int i_flags)
2829 struct inode *inode;
2831 struct super_block *sb;
2834 if (IS_ERR(shm_mnt))
2835 return ERR_CAST(shm_mnt);
2837 if (size < 0 || size > MAX_LFS_FILESIZE)
2838 return ERR_PTR(-EINVAL);
2840 if (shmem_acct_size(flags, size))
2841 return ERR_PTR(-ENOMEM);
2843 res = ERR_PTR(-ENOMEM);
2845 this.len = strlen(name);
2846 this.hash = 0; /* will go */
2847 sb = shm_mnt->mnt_sb;
2848 path.dentry = d_alloc_pseudo(sb, &this);
2851 d_set_d_op(path.dentry, &anon_ops);
2852 path.mnt = mntget(shm_mnt);
2854 res = ERR_PTR(-ENOSPC);
2855 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2859 inode->i_flags |= i_flags;
2860 d_instantiate(path.dentry, inode);
2861 inode->i_size = size;
2862 clear_nlink(inode); /* It is unlinked */
2863 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
2867 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2868 &shmem_file_operations);
2877 shmem_unacct_size(flags, size);
2882 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
2883 * kernel internal. There will be NO LSM permission checks against the
2884 * underlying inode. So users of this interface must do LSM checks at a
2885 * higher layer. The one user is the big_key implementation. LSM checks
2886 * are provided at the key level rather than the inode level.
2887 * @name: name for dentry (to be seen in /proc/<pid>/maps
2888 * @size: size to be set for the file
2889 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2891 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
2893 return __shmem_file_setup(name, size, flags, S_PRIVATE);
2897 * shmem_file_setup - get an unlinked file living in tmpfs
2898 * @name: name for dentry (to be seen in /proc/<pid>/maps
2899 * @size: size to be set for the file
2900 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2902 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2904 return __shmem_file_setup(name, size, flags, 0);
2906 EXPORT_SYMBOL_GPL(shmem_file_setup);
2909 * shmem_zero_setup - setup a shared anonymous mapping
2910 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2912 int shmem_zero_setup(struct vm_area_struct *vma)
2915 loff_t size = vma->vm_end - vma->vm_start;
2917 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2919 return PTR_ERR(file);
2923 vma->vm_file = file;
2924 vma->vm_ops = &shmem_vm_ops;
2929 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2930 * @mapping: the page's address_space
2931 * @index: the page index
2932 * @gfp: the page allocator flags to use if allocating
2934 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2935 * with any new page allocations done using the specified allocation flags.
2936 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2937 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2938 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2940 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2941 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2943 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2944 pgoff_t index, gfp_t gfp)
2947 struct inode *inode = mapping->host;
2951 BUG_ON(mapping->a_ops != &shmem_aops);
2952 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2954 page = ERR_PTR(error);
2960 * The tiny !SHMEM case uses ramfs without swap
2962 return read_cache_page_gfp(mapping, index, gfp);
2965 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);