2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/aio.h>
36 static struct vfsmount *shm_mnt;
40 * This virtual memory filesystem is heavily based on the ramfs. It
41 * extends ramfs by the ability to use swap and honor resource limits
42 * which makes it a completely usable filesystem.
45 #include <linux/xattr.h>
46 #include <linux/exportfs.h>
47 #include <linux/posix_acl.h>
48 #include <linux/posix_acl_xattr.h>
49 #include <linux/mman.h>
50 #include <linux/string.h>
51 #include <linux/slab.h>
52 #include <linux/backing-dev.h>
53 #include <linux/shmem_fs.h>
54 #include <linux/writeback.h>
55 #include <linux/blkdev.h>
56 #include <linux/pagevec.h>
57 #include <linux/percpu_counter.h>
58 #include <linux/falloc.h>
59 #include <linux/splice.h>
60 #include <linux/security.h>
61 #include <linux/swapops.h>
62 #include <linux/mempolicy.h>
63 #include <linux/namei.h>
64 #include <linux/ctype.h>
65 #include <linux/migrate.h>
66 #include <linux/highmem.h>
67 #include <linux/seq_file.h>
68 #include <linux/magic.h>
70 #include <asm/uaccess.h>
71 #include <asm/pgtable.h>
73 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
74 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
76 /* Pretend that each entry is of this size in directory's i_size */
77 #define BOGO_DIRENT_SIZE 20
79 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
80 #define SHORT_SYMLINK_LEN 128
83 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
84 * inode->i_private (with i_mutex making sure that it has only one user at
85 * a time): we would prefer not to enlarge the shmem inode just for that.
88 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
89 pgoff_t start; /* start of range currently being fallocated */
90 pgoff_t next; /* the next page offset to be fallocated */
91 pgoff_t nr_falloced; /* how many new pages have been fallocated */
92 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
95 /* Flag allocation requirements to shmem_getpage */
97 SGP_READ, /* don't exceed i_size, don't allocate page */
98 SGP_CACHE, /* don't exceed i_size, may allocate page */
99 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
100 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
101 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
105 static unsigned long shmem_default_max_blocks(void)
107 return totalram_pages / 2;
110 static unsigned long shmem_default_max_inodes(void)
112 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
116 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
117 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
118 struct shmem_inode_info *info, pgoff_t index);
119 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
120 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
122 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
123 struct page **pagep, enum sgp_type sgp, int *fault_type)
125 return shmem_getpage_gfp(inode, index, pagep, sgp,
126 mapping_gfp_mask(inode->i_mapping), fault_type);
129 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
131 return sb->s_fs_info;
135 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
136 * for shared memory and for shared anonymous (/dev/zero) mappings
137 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
138 * consistent with the pre-accounting of private mappings ...
140 static inline int shmem_acct_size(unsigned long flags, loff_t size)
142 return (flags & VM_NORESERVE) ?
143 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
146 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
148 if (!(flags & VM_NORESERVE))
149 vm_unacct_memory(VM_ACCT(size));
153 * ... whereas tmpfs objects are accounted incrementally as
154 * pages are allocated, in order to allow huge sparse files.
155 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
156 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
158 static inline int shmem_acct_block(unsigned long flags)
160 return (flags & VM_NORESERVE) ?
161 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
164 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
166 if (flags & VM_NORESERVE)
167 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
170 static const struct super_operations shmem_ops;
171 static const struct address_space_operations shmem_aops;
172 static const struct file_operations shmem_file_operations;
173 static const struct inode_operations shmem_inode_operations;
174 static const struct inode_operations shmem_dir_inode_operations;
175 static const struct inode_operations shmem_special_inode_operations;
176 static const struct vm_operations_struct shmem_vm_ops;
178 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
179 .ra_pages = 0, /* No readahead */
180 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
183 static LIST_HEAD(shmem_swaplist);
184 static DEFINE_MUTEX(shmem_swaplist_mutex);
186 static int shmem_reserve_inode(struct super_block *sb)
188 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
189 if (sbinfo->max_inodes) {
190 spin_lock(&sbinfo->stat_lock);
191 if (!sbinfo->free_inodes) {
192 spin_unlock(&sbinfo->stat_lock);
195 sbinfo->free_inodes--;
196 spin_unlock(&sbinfo->stat_lock);
201 static void shmem_free_inode(struct super_block *sb)
203 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
204 if (sbinfo->max_inodes) {
205 spin_lock(&sbinfo->stat_lock);
206 sbinfo->free_inodes++;
207 spin_unlock(&sbinfo->stat_lock);
212 * shmem_recalc_inode - recalculate the block usage of an inode
213 * @inode: inode to recalc
215 * We have to calculate the free blocks since the mm can drop
216 * undirtied hole pages behind our back.
218 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
219 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
221 * It has to be called with the spinlock held.
223 static void shmem_recalc_inode(struct inode *inode)
225 struct shmem_inode_info *info = SHMEM_I(inode);
228 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
230 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
231 if (sbinfo->max_blocks)
232 percpu_counter_add(&sbinfo->used_blocks, -freed);
233 info->alloced -= freed;
234 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
235 shmem_unacct_blocks(info->flags, freed);
240 * Replace item expected in radix tree by a new item, while holding tree lock.
242 static int shmem_radix_tree_replace(struct address_space *mapping,
243 pgoff_t index, void *expected, void *replacement)
248 VM_BUG_ON(!expected);
249 VM_BUG_ON(!replacement);
250 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
253 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
254 if (item != expected)
256 radix_tree_replace_slot(pslot, replacement);
261 * Sometimes, before we decide whether to proceed or to fail, we must check
262 * that an entry was not already brought back from swap by a racing thread.
264 * Checking page is not enough: by the time a SwapCache page is locked, it
265 * might be reused, and again be SwapCache, using the same swap as before.
267 static bool shmem_confirm_swap(struct address_space *mapping,
268 pgoff_t index, swp_entry_t swap)
273 item = radix_tree_lookup(&mapping->page_tree, index);
275 return item == swp_to_radix_entry(swap);
279 * Like add_to_page_cache_locked, but error if expected item has gone.
281 static int shmem_add_to_page_cache(struct page *page,
282 struct address_space *mapping,
283 pgoff_t index, gfp_t gfp, void *expected)
287 VM_BUG_ON_PAGE(!PageLocked(page), page);
288 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
290 page_cache_get(page);
291 page->mapping = mapping;
294 spin_lock_irq(&mapping->tree_lock);
296 error = radix_tree_insert(&mapping->page_tree, index, page);
298 error = shmem_radix_tree_replace(mapping, index, expected,
302 __inc_zone_page_state(page, NR_FILE_PAGES);
303 __inc_zone_page_state(page, NR_SHMEM);
304 spin_unlock_irq(&mapping->tree_lock);
306 page->mapping = NULL;
307 spin_unlock_irq(&mapping->tree_lock);
308 page_cache_release(page);
314 * Like delete_from_page_cache, but substitutes swap for page.
316 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
318 struct address_space *mapping = page->mapping;
321 spin_lock_irq(&mapping->tree_lock);
322 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
323 page->mapping = NULL;
325 __dec_zone_page_state(page, NR_FILE_PAGES);
326 __dec_zone_page_state(page, NR_SHMEM);
327 spin_unlock_irq(&mapping->tree_lock);
328 page_cache_release(page);
333 * Like find_get_pages, but collecting swap entries as well as pages.
335 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
336 pgoff_t start, unsigned int nr_pages,
337 struct page **pages, pgoff_t *indices)
340 unsigned int ret = 0;
341 struct radix_tree_iter iter;
348 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
351 page = radix_tree_deref_slot(slot);
354 if (radix_tree_exception(page)) {
355 if (radix_tree_deref_retry(page))
358 * Otherwise, we must be storing a swap entry
359 * here as an exceptional entry: so return it
360 * without attempting to raise page count.
364 if (!page_cache_get_speculative(page))
367 /* Has the page moved? */
368 if (unlikely(page != *slot)) {
369 page_cache_release(page);
373 indices[ret] = iter.index;
375 if (++ret == nr_pages)
383 * Remove swap entry from radix tree, free the swap and its page cache.
385 static int shmem_free_swap(struct address_space *mapping,
386 pgoff_t index, void *radswap)
390 spin_lock_irq(&mapping->tree_lock);
391 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
392 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);
536 while (index < end) {
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 all gone or hole-punch or unfalloc, we're done */
543 if (index == start || end != -1)
545 /* But if truncating, restart to make sure all gone */
549 mem_cgroup_uncharge_start();
550 for (i = 0; i < pagevec_count(&pvec); i++) {
551 struct page *page = pvec.pages[i];
557 if (radix_tree_exceptional_entry(page)) {
560 if (shmem_free_swap(mapping, index, page)) {
561 /* Swap was replaced by page: retry */
570 if (!unfalloc || !PageUptodate(page)) {
571 if (page->mapping == mapping) {
572 VM_BUG_ON_PAGE(PageWriteback(page), page);
573 truncate_inode_page(mapping, page);
575 /* Page was replaced by swap: retry */
583 shmem_deswap_pagevec(&pvec);
584 pagevec_release(&pvec);
585 mem_cgroup_uncharge_end();
589 spin_lock(&info->lock);
590 info->swapped -= nr_swaps_freed;
591 shmem_recalc_inode(inode);
592 spin_unlock(&info->lock);
595 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
597 shmem_undo_range(inode, lstart, lend, false);
598 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
600 EXPORT_SYMBOL_GPL(shmem_truncate_range);
602 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
604 struct inode *inode = dentry->d_inode;
607 error = inode_change_ok(inode, attr);
611 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
612 loff_t oldsize = inode->i_size;
613 loff_t newsize = attr->ia_size;
615 if (newsize != oldsize) {
616 i_size_write(inode, newsize);
617 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
619 if (newsize < oldsize) {
620 loff_t holebegin = round_up(newsize, PAGE_SIZE);
621 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
622 shmem_truncate_range(inode, newsize, (loff_t)-1);
623 /* unmap again to remove racily COWed private pages */
624 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
628 setattr_copy(inode, attr);
629 if (attr->ia_valid & ATTR_MODE)
630 error = posix_acl_chmod(inode, inode->i_mode);
634 static void shmem_evict_inode(struct inode *inode)
636 struct shmem_inode_info *info = SHMEM_I(inode);
638 if (inode->i_mapping->a_ops == &shmem_aops) {
639 shmem_unacct_size(info->flags, inode->i_size);
641 shmem_truncate_range(inode, 0, (loff_t)-1);
642 if (!list_empty(&info->swaplist)) {
643 mutex_lock(&shmem_swaplist_mutex);
644 list_del_init(&info->swaplist);
645 mutex_unlock(&shmem_swaplist_mutex);
648 kfree(info->symlink);
650 simple_xattrs_free(&info->xattrs);
651 WARN_ON(inode->i_blocks);
652 shmem_free_inode(inode->i_sb);
657 * If swap found in inode, free it and move page from swapcache to filecache.
659 static int shmem_unuse_inode(struct shmem_inode_info *info,
660 swp_entry_t swap, struct page **pagep)
662 struct address_space *mapping = info->vfs_inode.i_mapping;
668 radswap = swp_to_radix_entry(swap);
669 index = radix_tree_locate_item(&mapping->page_tree, radswap);
674 * Move _head_ to start search for next from here.
675 * But be careful: shmem_evict_inode checks list_empty without taking
676 * mutex, and there's an instant in list_move_tail when info->swaplist
677 * would appear empty, if it were the only one on shmem_swaplist.
679 if (shmem_swaplist.next != &info->swaplist)
680 list_move_tail(&shmem_swaplist, &info->swaplist);
682 gfp = mapping_gfp_mask(mapping);
683 if (shmem_should_replace_page(*pagep, gfp)) {
684 mutex_unlock(&shmem_swaplist_mutex);
685 error = shmem_replace_page(pagep, gfp, info, index);
686 mutex_lock(&shmem_swaplist_mutex);
688 * We needed to drop mutex to make that restrictive page
689 * allocation, but the inode might have been freed while we
690 * dropped it: although a racing shmem_evict_inode() cannot
691 * complete without emptying the radix_tree, our page lock
692 * on this swapcache page is not enough to prevent that -
693 * free_swap_and_cache() of our swap entry will only
694 * trylock_page(), removing swap from radix_tree whatever.
696 * We must not proceed to shmem_add_to_page_cache() if the
697 * inode has been freed, but of course we cannot rely on
698 * inode or mapping or info to check that. However, we can
699 * safely check if our swap entry is still in use (and here
700 * it can't have got reused for another page): if it's still
701 * in use, then the inode cannot have been freed yet, and we
702 * can safely proceed (if it's no longer in use, that tells
703 * nothing about the inode, but we don't need to unuse swap).
705 if (!page_swapcount(*pagep))
710 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
711 * but also to hold up shmem_evict_inode(): so inode cannot be freed
712 * beneath us (pagelock doesn't help until the page is in pagecache).
715 error = shmem_add_to_page_cache(*pagep, mapping, index,
716 GFP_NOWAIT, radswap);
717 if (error != -ENOMEM) {
719 * Truncation and eviction use free_swap_and_cache(), which
720 * only does trylock page: if we raced, best clean up here.
722 delete_from_swap_cache(*pagep);
723 set_page_dirty(*pagep);
725 spin_lock(&info->lock);
727 spin_unlock(&info->lock);
730 error = 1; /* not an error, but entry was found */
736 * Search through swapped inodes to find and replace swap by page.
738 int shmem_unuse(swp_entry_t swap, struct page *page)
740 struct list_head *this, *next;
741 struct shmem_inode_info *info;
746 * There's a faint possibility that swap page was replaced before
747 * caller locked it: caller will come back later with the right page.
749 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
753 * Charge page using GFP_KERNEL while we can wait, before taking
754 * the shmem_swaplist_mutex which might hold up shmem_writepage().
755 * Charged back to the user (not to caller) when swap account is used.
757 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
760 /* No radix_tree_preload: swap entry keeps a place for page in tree */
762 mutex_lock(&shmem_swaplist_mutex);
763 list_for_each_safe(this, next, &shmem_swaplist) {
764 info = list_entry(this, struct shmem_inode_info, swaplist);
766 found = shmem_unuse_inode(info, swap, &page);
768 list_del_init(&info->swaplist);
773 mutex_unlock(&shmem_swaplist_mutex);
779 page_cache_release(page);
784 * Move the page from the page cache to the swap cache.
786 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
788 struct shmem_inode_info *info;
789 struct address_space *mapping;
794 BUG_ON(!PageLocked(page));
795 mapping = page->mapping;
797 inode = mapping->host;
798 info = SHMEM_I(inode);
799 if (info->flags & VM_LOCKED)
801 if (!total_swap_pages)
805 * shmem_backing_dev_info's capabilities prevent regular writeback or
806 * sync from ever calling shmem_writepage; but a stacking filesystem
807 * might use ->writepage of its underlying filesystem, in which case
808 * tmpfs should write out to swap only in response to memory pressure,
809 * and not for the writeback threads or sync.
811 if (!wbc->for_reclaim) {
812 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
817 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
818 * value into swapfile.c, the only way we can correctly account for a
819 * fallocated page arriving here is now to initialize it and write it.
821 * That's okay for a page already fallocated earlier, but if we have
822 * not yet completed the fallocation, then (a) we want to keep track
823 * of this page in case we have to undo it, and (b) it may not be a
824 * good idea to continue anyway, once we're pushing into swap. So
825 * reactivate the page, and let shmem_fallocate() quit when too many.
827 if (!PageUptodate(page)) {
828 if (inode->i_private) {
829 struct shmem_falloc *shmem_falloc;
830 spin_lock(&inode->i_lock);
831 shmem_falloc = inode->i_private;
833 !shmem_falloc->waitq &&
834 index >= shmem_falloc->start &&
835 index < shmem_falloc->next)
836 shmem_falloc->nr_unswapped++;
839 spin_unlock(&inode->i_lock);
843 clear_highpage(page);
844 flush_dcache_page(page);
845 SetPageUptodate(page);
848 swap = get_swap_page();
853 * Add inode to shmem_unuse()'s list of swapped-out inodes,
854 * if it's not already there. Do it now before the page is
855 * moved to swap cache, when its pagelock no longer protects
856 * the inode from eviction. But don't unlock the mutex until
857 * we've incremented swapped, because shmem_unuse_inode() will
858 * prune a !swapped inode from the swaplist under this mutex.
860 mutex_lock(&shmem_swaplist_mutex);
861 if (list_empty(&info->swaplist))
862 list_add_tail(&info->swaplist, &shmem_swaplist);
864 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
865 swap_shmem_alloc(swap);
866 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
868 spin_lock(&info->lock);
870 shmem_recalc_inode(inode);
871 spin_unlock(&info->lock);
873 mutex_unlock(&shmem_swaplist_mutex);
874 BUG_ON(page_mapped(page));
875 swap_writepage(page, wbc);
879 mutex_unlock(&shmem_swaplist_mutex);
880 swapcache_free(swap, NULL);
882 set_page_dirty(page);
883 if (wbc->for_reclaim)
884 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
891 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
895 if (!mpol || mpol->mode == MPOL_DEFAULT)
896 return; /* show nothing */
898 mpol_to_str(buffer, sizeof(buffer), mpol);
900 seq_printf(seq, ",mpol=%s", buffer);
903 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
905 struct mempolicy *mpol = NULL;
907 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
910 spin_unlock(&sbinfo->stat_lock);
914 #endif /* CONFIG_TMPFS */
916 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
917 struct shmem_inode_info *info, pgoff_t index)
919 struct vm_area_struct pvma;
922 /* Create a pseudo vma that just contains the policy */
924 /* Bias interleave by inode number to distribute better across nodes */
925 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
927 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
929 page = swapin_readahead(swap, gfp, &pvma, 0);
931 /* Drop reference taken by mpol_shared_policy_lookup() */
932 mpol_cond_put(pvma.vm_policy);
937 static struct page *shmem_alloc_page(gfp_t gfp,
938 struct shmem_inode_info *info, pgoff_t index)
940 struct vm_area_struct pvma;
943 /* Create a pseudo vma that just contains the policy */
945 /* Bias interleave by inode number to distribute better across nodes */
946 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
948 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
950 page = alloc_page_vma(gfp, &pvma, 0);
952 /* Drop reference taken by mpol_shared_policy_lookup() */
953 mpol_cond_put(pvma.vm_policy);
957 #else /* !CONFIG_NUMA */
959 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
962 #endif /* CONFIG_TMPFS */
964 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
965 struct shmem_inode_info *info, pgoff_t index)
967 return swapin_readahead(swap, gfp, NULL, 0);
970 static inline struct page *shmem_alloc_page(gfp_t gfp,
971 struct shmem_inode_info *info, pgoff_t index)
973 return alloc_page(gfp);
975 #endif /* CONFIG_NUMA */
977 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
978 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
985 * When a page is moved from swapcache to shmem filecache (either by the
986 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
987 * shmem_unuse_inode()), it may have been read in earlier from swap, in
988 * ignorance of the mapping it belongs to. If that mapping has special
989 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
990 * we may need to copy to a suitable page before moving to filecache.
992 * In a future release, this may well be extended to respect cpuset and
993 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
994 * but for now it is a simple matter of zone.
996 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
998 return page_zonenum(page) > gfp_zone(gfp);
1001 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1002 struct shmem_inode_info *info, pgoff_t index)
1004 struct page *oldpage, *newpage;
1005 struct address_space *swap_mapping;
1010 swap_index = page_private(oldpage);
1011 swap_mapping = page_mapping(oldpage);
1014 * We have arrived here because our zones are constrained, so don't
1015 * limit chance of success by further cpuset and node constraints.
1017 gfp &= ~GFP_CONSTRAINT_MASK;
1018 newpage = shmem_alloc_page(gfp, info, index);
1022 page_cache_get(newpage);
1023 copy_highpage(newpage, oldpage);
1024 flush_dcache_page(newpage);
1026 __set_page_locked(newpage);
1027 SetPageUptodate(newpage);
1028 SetPageSwapBacked(newpage);
1029 set_page_private(newpage, swap_index);
1030 SetPageSwapCache(newpage);
1033 * Our caller will very soon move newpage out of swapcache, but it's
1034 * a nice clean interface for us to replace oldpage by newpage there.
1036 spin_lock_irq(&swap_mapping->tree_lock);
1037 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1040 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1041 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1043 spin_unlock_irq(&swap_mapping->tree_lock);
1045 if (unlikely(error)) {
1047 * Is this possible? I think not, now that our callers check
1048 * both PageSwapCache and page_private after getting page lock;
1049 * but be defensive. Reverse old to newpage for clear and free.
1053 mem_cgroup_replace_page_cache(oldpage, newpage);
1054 lru_cache_add_anon(newpage);
1058 ClearPageSwapCache(oldpage);
1059 set_page_private(oldpage, 0);
1061 unlock_page(oldpage);
1062 page_cache_release(oldpage);
1063 page_cache_release(oldpage);
1068 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1070 * If we allocate a new one we do not mark it dirty. That's up to the
1071 * vm. If we swap it in we mark it dirty since we also free the swap
1072 * entry since a page cannot live in both the swap and page cache
1074 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1075 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1077 struct address_space *mapping = inode->i_mapping;
1078 struct shmem_inode_info *info;
1079 struct shmem_sb_info *sbinfo;
1086 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1090 page = find_lock_page(mapping, index);
1091 if (radix_tree_exceptional_entry(page)) {
1092 swap = radix_to_swp_entry(page);
1096 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1097 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1102 /* fallocated page? */
1103 if (page && !PageUptodate(page)) {
1104 if (sgp != SGP_READ)
1107 page_cache_release(page);
1110 if (page || (sgp == SGP_READ && !swap.val)) {
1116 * Fast cache lookup did not find it:
1117 * bring it back from swap or allocate.
1119 info = SHMEM_I(inode);
1120 sbinfo = SHMEM_SB(inode->i_sb);
1123 /* Look it up and read it in.. */
1124 page = lookup_swap_cache(swap);
1126 /* here we actually do the io */
1128 *fault_type |= VM_FAULT_MAJOR;
1129 page = shmem_swapin(swap, gfp, info, index);
1136 /* We have to do this with page locked to prevent races */
1138 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1139 !shmem_confirm_swap(mapping, index, swap)) {
1140 error = -EEXIST; /* try again */
1143 if (!PageUptodate(page)) {
1147 wait_on_page_writeback(page);
1149 if (shmem_should_replace_page(page, gfp)) {
1150 error = shmem_replace_page(&page, gfp, info, index);
1155 error = mem_cgroup_cache_charge(page, current->mm,
1156 gfp & GFP_RECLAIM_MASK);
1158 error = shmem_add_to_page_cache(page, mapping, index,
1159 gfp, swp_to_radix_entry(swap));
1161 * We already confirmed swap under page lock, and make
1162 * no memory allocation here, so usually no possibility
1163 * of error; but free_swap_and_cache() only trylocks a
1164 * page, so it is just possible that the entry has been
1165 * truncated or holepunched since swap was confirmed.
1166 * shmem_undo_range() will have done some of the
1167 * unaccounting, now delete_from_swap_cache() will do
1168 * the rest (including mem_cgroup_uncharge_swapcache).
1169 * Reset swap.val? No, leave it so "failed" goes back to
1170 * "repeat": reading a hole and writing should succeed.
1173 delete_from_swap_cache(page);
1178 spin_lock(&info->lock);
1180 shmem_recalc_inode(inode);
1181 spin_unlock(&info->lock);
1183 delete_from_swap_cache(page);
1184 set_page_dirty(page);
1188 if (shmem_acct_block(info->flags)) {
1192 if (sbinfo->max_blocks) {
1193 if (percpu_counter_compare(&sbinfo->used_blocks,
1194 sbinfo->max_blocks) >= 0) {
1198 percpu_counter_inc(&sbinfo->used_blocks);
1201 page = shmem_alloc_page(gfp, info, index);
1207 SetPageSwapBacked(page);
1208 __set_page_locked(page);
1209 error = mem_cgroup_cache_charge(page, current->mm,
1210 gfp & GFP_RECLAIM_MASK);
1213 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
1215 error = shmem_add_to_page_cache(page, mapping, index,
1217 radix_tree_preload_end();
1220 mem_cgroup_uncharge_cache_page(page);
1223 lru_cache_add_anon(page);
1225 spin_lock(&info->lock);
1227 inode->i_blocks += BLOCKS_PER_PAGE;
1228 shmem_recalc_inode(inode);
1229 spin_unlock(&info->lock);
1233 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1235 if (sgp == SGP_FALLOC)
1239 * Let SGP_WRITE caller clear ends if write does not fill page;
1240 * but SGP_FALLOC on a page fallocated earlier must initialize
1241 * it now, lest undo on failure cancel our earlier guarantee.
1243 if (sgp != SGP_WRITE) {
1244 clear_highpage(page);
1245 flush_dcache_page(page);
1246 SetPageUptodate(page);
1248 if (sgp == SGP_DIRTY)
1249 set_page_dirty(page);
1252 /* Perhaps the file has been truncated since we checked */
1253 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1254 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1268 info = SHMEM_I(inode);
1269 ClearPageDirty(page);
1270 delete_from_page_cache(page);
1271 spin_lock(&info->lock);
1273 inode->i_blocks -= BLOCKS_PER_PAGE;
1274 spin_unlock(&info->lock);
1276 sbinfo = SHMEM_SB(inode->i_sb);
1277 if (sbinfo->max_blocks)
1278 percpu_counter_add(&sbinfo->used_blocks, -1);
1280 shmem_unacct_blocks(info->flags, 1);
1282 if (swap.val && error != -EINVAL &&
1283 !shmem_confirm_swap(mapping, index, swap))
1288 page_cache_release(page);
1290 if (error == -ENOSPC && !once++) {
1291 info = SHMEM_I(inode);
1292 spin_lock(&info->lock);
1293 shmem_recalc_inode(inode);
1294 spin_unlock(&info->lock);
1297 if (error == -EEXIST) /* from above or from radix_tree_insert */
1302 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1304 struct inode *inode = file_inode(vma->vm_file);
1306 int ret = VM_FAULT_LOCKED;
1309 * Trinity finds that probing a hole which tmpfs is punching can
1310 * prevent the hole-punch from ever completing: which in turn
1311 * locks writers out with its hold on i_mutex. So refrain from
1312 * faulting pages into the hole while it's being punched. Although
1313 * shmem_undo_range() does remove the additions, it may be unable to
1314 * keep up, as each new page needs its own unmap_mapping_range() call,
1315 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1317 * It does not matter if we sometimes reach this check just before the
1318 * hole-punch begins, so that one fault then races with the punch:
1319 * we just need to make racing faults a rare case.
1321 * The implementation below would be much simpler if we just used a
1322 * standard mutex or completion: but we cannot take i_mutex in fault,
1323 * and bloating every shmem inode for this unlikely case would be sad.
1325 if (unlikely(inode->i_private)) {
1326 struct shmem_falloc *shmem_falloc;
1328 spin_lock(&inode->i_lock);
1329 shmem_falloc = inode->i_private;
1331 shmem_falloc->waitq &&
1332 vmf->pgoff >= shmem_falloc->start &&
1333 vmf->pgoff < shmem_falloc->next) {
1334 wait_queue_head_t *shmem_falloc_waitq;
1335 DEFINE_WAIT(shmem_fault_wait);
1337 ret = VM_FAULT_NOPAGE;
1338 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1339 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1340 /* It's polite to up mmap_sem if we can */
1341 up_read(&vma->vm_mm->mmap_sem);
1342 ret = VM_FAULT_RETRY;
1345 shmem_falloc_waitq = shmem_falloc->waitq;
1346 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1347 TASK_UNINTERRUPTIBLE);
1348 spin_unlock(&inode->i_lock);
1352 * shmem_falloc_waitq points into the shmem_fallocate()
1353 * stack of the hole-punching task: shmem_falloc_waitq
1354 * is usually invalid by the time we reach here, but
1355 * finish_wait() does not dereference it in that case;
1356 * though i_lock needed lest racing with wake_up_all().
1358 spin_lock(&inode->i_lock);
1359 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1360 spin_unlock(&inode->i_lock);
1363 spin_unlock(&inode->i_lock);
1366 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1368 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1370 if (ret & VM_FAULT_MAJOR) {
1371 count_vm_event(PGMAJFAULT);
1372 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1378 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1380 struct inode *inode = file_inode(vma->vm_file);
1381 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1384 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1387 struct inode *inode = file_inode(vma->vm_file);
1390 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1391 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1395 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1397 struct inode *inode = file_inode(file);
1398 struct shmem_inode_info *info = SHMEM_I(inode);
1399 int retval = -ENOMEM;
1401 spin_lock(&info->lock);
1402 if (lock && !(info->flags & VM_LOCKED)) {
1403 if (!user_shm_lock(inode->i_size, user))
1405 info->flags |= VM_LOCKED;
1406 mapping_set_unevictable(file->f_mapping);
1408 if (!lock && (info->flags & VM_LOCKED) && user) {
1409 user_shm_unlock(inode->i_size, user);
1410 info->flags &= ~VM_LOCKED;
1411 mapping_clear_unevictable(file->f_mapping);
1416 spin_unlock(&info->lock);
1420 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1422 file_accessed(file);
1423 vma->vm_ops = &shmem_vm_ops;
1427 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1428 umode_t mode, dev_t dev, unsigned long flags)
1430 struct inode *inode;
1431 struct shmem_inode_info *info;
1432 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1434 if (shmem_reserve_inode(sb))
1437 inode = new_inode(sb);
1439 inode->i_ino = get_next_ino();
1440 inode_init_owner(inode, dir, mode);
1441 inode->i_blocks = 0;
1442 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1443 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1444 inode->i_generation = get_seconds();
1445 info = SHMEM_I(inode);
1446 memset(info, 0, (char *)inode - (char *)info);
1447 spin_lock_init(&info->lock);
1448 info->flags = flags & VM_NORESERVE;
1449 INIT_LIST_HEAD(&info->swaplist);
1450 simple_xattrs_init(&info->xattrs);
1451 cache_no_acl(inode);
1453 switch (mode & S_IFMT) {
1455 inode->i_op = &shmem_special_inode_operations;
1456 init_special_inode(inode, mode, dev);
1459 inode->i_mapping->a_ops = &shmem_aops;
1460 inode->i_op = &shmem_inode_operations;
1461 inode->i_fop = &shmem_file_operations;
1462 mpol_shared_policy_init(&info->policy,
1463 shmem_get_sbmpol(sbinfo));
1467 /* Some things misbehave if size == 0 on a directory */
1468 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1469 inode->i_op = &shmem_dir_inode_operations;
1470 inode->i_fop = &simple_dir_operations;
1474 * Must not load anything in the rbtree,
1475 * mpol_free_shared_policy will not be called.
1477 mpol_shared_policy_init(&info->policy, NULL);
1481 shmem_free_inode(sb);
1486 static const struct inode_operations shmem_symlink_inode_operations;
1487 static const struct inode_operations shmem_short_symlink_operations;
1489 #ifdef CONFIG_TMPFS_XATTR
1490 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1492 #define shmem_initxattrs NULL
1496 shmem_write_begin(struct file *file, struct address_space *mapping,
1497 loff_t pos, unsigned len, unsigned flags,
1498 struct page **pagep, void **fsdata)
1500 struct inode *inode = mapping->host;
1501 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1502 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1506 shmem_write_end(struct file *file, struct address_space *mapping,
1507 loff_t pos, unsigned len, unsigned copied,
1508 struct page *page, void *fsdata)
1510 struct inode *inode = mapping->host;
1512 if (pos + copied > inode->i_size)
1513 i_size_write(inode, pos + copied);
1515 if (!PageUptodate(page)) {
1516 if (copied < PAGE_CACHE_SIZE) {
1517 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1518 zero_user_segments(page, 0, from,
1519 from + copied, PAGE_CACHE_SIZE);
1521 SetPageUptodate(page);
1523 set_page_dirty(page);
1525 page_cache_release(page);
1530 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1532 struct inode *inode = file_inode(filp);
1533 struct address_space *mapping = inode->i_mapping;
1535 unsigned long offset;
1536 enum sgp_type sgp = SGP_READ;
1539 * Might this read be for a stacking filesystem? Then when reading
1540 * holes of a sparse file, we actually need to allocate those pages,
1541 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1543 if (segment_eq(get_fs(), KERNEL_DS))
1546 index = *ppos >> PAGE_CACHE_SHIFT;
1547 offset = *ppos & ~PAGE_CACHE_MASK;
1550 struct page *page = NULL;
1552 unsigned long nr, ret;
1553 loff_t i_size = i_size_read(inode);
1555 end_index = i_size >> PAGE_CACHE_SHIFT;
1556 if (index > end_index)
1558 if (index == end_index) {
1559 nr = i_size & ~PAGE_CACHE_MASK;
1564 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1566 if (desc->error == -EINVAL)
1574 * We must evaluate after, since reads (unlike writes)
1575 * are called without i_mutex protection against truncate
1577 nr = PAGE_CACHE_SIZE;
1578 i_size = i_size_read(inode);
1579 end_index = i_size >> PAGE_CACHE_SHIFT;
1580 if (index == end_index) {
1581 nr = i_size & ~PAGE_CACHE_MASK;
1584 page_cache_release(page);
1592 * If users can be writing to this page using arbitrary
1593 * virtual addresses, take care about potential aliasing
1594 * before reading the page on the kernel side.
1596 if (mapping_writably_mapped(mapping))
1597 flush_dcache_page(page);
1599 * Mark the page accessed if we read the beginning.
1602 mark_page_accessed(page);
1604 page = ZERO_PAGE(0);
1605 page_cache_get(page);
1609 * Ok, we have the page, and it's up-to-date, so
1610 * now we can copy it to user space...
1612 * The actor routine returns how many bytes were actually used..
1613 * NOTE! This may not be the same as how much of a user buffer
1614 * we filled up (we may be padding etc), so we can only update
1615 * "pos" here (the actor routine has to update the user buffer
1616 * pointers and the remaining count).
1618 ret = actor(desc, page, offset, nr);
1620 index += offset >> PAGE_CACHE_SHIFT;
1621 offset &= ~PAGE_CACHE_MASK;
1623 page_cache_release(page);
1624 if (ret != nr || !desc->count)
1630 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1631 file_accessed(filp);
1634 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1635 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1637 struct file *filp = iocb->ki_filp;
1641 loff_t *ppos = &iocb->ki_pos;
1643 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1647 for (seg = 0; seg < nr_segs; seg++) {
1648 read_descriptor_t desc;
1651 desc.arg.buf = iov[seg].iov_base;
1652 desc.count = iov[seg].iov_len;
1653 if (desc.count == 0)
1656 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1657 retval += desc.written;
1659 retval = retval ?: desc.error;
1668 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1669 struct pipe_inode_info *pipe, size_t len,
1672 struct address_space *mapping = in->f_mapping;
1673 struct inode *inode = mapping->host;
1674 unsigned int loff, nr_pages, req_pages;
1675 struct page *pages[PIPE_DEF_BUFFERS];
1676 struct partial_page partial[PIPE_DEF_BUFFERS];
1678 pgoff_t index, end_index;
1681 struct splice_pipe_desc spd = {
1684 .nr_pages_max = PIPE_DEF_BUFFERS,
1686 .ops = &page_cache_pipe_buf_ops,
1687 .spd_release = spd_release_page,
1690 isize = i_size_read(inode);
1691 if (unlikely(*ppos >= isize))
1694 left = isize - *ppos;
1695 if (unlikely(left < len))
1698 if (splice_grow_spd(pipe, &spd))
1701 index = *ppos >> PAGE_CACHE_SHIFT;
1702 loff = *ppos & ~PAGE_CACHE_MASK;
1703 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1704 nr_pages = min(req_pages, pipe->buffers);
1706 spd.nr_pages = find_get_pages_contig(mapping, index,
1707 nr_pages, spd.pages);
1708 index += spd.nr_pages;
1711 while (spd.nr_pages < nr_pages) {
1712 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1716 spd.pages[spd.nr_pages++] = page;
1720 index = *ppos >> PAGE_CACHE_SHIFT;
1721 nr_pages = spd.nr_pages;
1724 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1725 unsigned int this_len;
1730 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1731 page = spd.pages[page_nr];
1733 if (!PageUptodate(page) || page->mapping != mapping) {
1734 error = shmem_getpage(inode, index, &page,
1739 page_cache_release(spd.pages[page_nr]);
1740 spd.pages[page_nr] = page;
1743 isize = i_size_read(inode);
1744 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1745 if (unlikely(!isize || index > end_index))
1748 if (end_index == index) {
1751 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1755 this_len = min(this_len, plen - loff);
1759 spd.partial[page_nr].offset = loff;
1760 spd.partial[page_nr].len = this_len;
1767 while (page_nr < nr_pages)
1768 page_cache_release(spd.pages[page_nr++]);
1771 error = splice_to_pipe(pipe, &spd);
1773 splice_shrink_spd(&spd);
1783 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1785 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1786 pgoff_t index, pgoff_t end, int whence)
1789 struct pagevec pvec;
1790 pgoff_t indices[PAGEVEC_SIZE];
1794 pagevec_init(&pvec, 0);
1795 pvec.nr = 1; /* start small: we may be there already */
1797 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
1798 pvec.nr, pvec.pages, indices);
1800 if (whence == SEEK_DATA)
1804 for (i = 0; i < pvec.nr; i++, index++) {
1805 if (index < indices[i]) {
1806 if (whence == SEEK_HOLE) {
1812 page = pvec.pages[i];
1813 if (page && !radix_tree_exceptional_entry(page)) {
1814 if (!PageUptodate(page))
1818 (page && whence == SEEK_DATA) ||
1819 (!page && whence == SEEK_HOLE)) {
1824 shmem_deswap_pagevec(&pvec);
1825 pagevec_release(&pvec);
1826 pvec.nr = PAGEVEC_SIZE;
1832 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
1834 struct address_space *mapping = file->f_mapping;
1835 struct inode *inode = mapping->host;
1839 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1840 return generic_file_llseek_size(file, offset, whence,
1841 MAX_LFS_FILESIZE, i_size_read(inode));
1842 mutex_lock(&inode->i_mutex);
1843 /* We're holding i_mutex so we can access i_size directly */
1847 else if (offset >= inode->i_size)
1850 start = offset >> PAGE_CACHE_SHIFT;
1851 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1852 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
1853 new_offset <<= PAGE_CACHE_SHIFT;
1854 if (new_offset > offset) {
1855 if (new_offset < inode->i_size)
1856 offset = new_offset;
1857 else if (whence == SEEK_DATA)
1860 offset = inode->i_size;
1865 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
1866 mutex_unlock(&inode->i_mutex);
1870 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1873 struct inode *inode = file_inode(file);
1874 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1875 struct shmem_falloc shmem_falloc;
1876 pgoff_t start, index, end;
1879 mutex_lock(&inode->i_mutex);
1881 if (mode & FALLOC_FL_PUNCH_HOLE) {
1882 struct address_space *mapping = file->f_mapping;
1883 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1884 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1885 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
1887 shmem_falloc.waitq = &shmem_falloc_waitq;
1888 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
1889 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
1890 spin_lock(&inode->i_lock);
1891 inode->i_private = &shmem_falloc;
1892 spin_unlock(&inode->i_lock);
1894 if ((u64)unmap_end > (u64)unmap_start)
1895 unmap_mapping_range(mapping, unmap_start,
1896 1 + unmap_end - unmap_start, 0);
1897 shmem_truncate_range(inode, offset, offset + len - 1);
1898 /* No need to unmap again: hole-punching leaves COWed pages */
1900 spin_lock(&inode->i_lock);
1901 inode->i_private = NULL;
1902 wake_up_all(&shmem_falloc_waitq);
1903 spin_unlock(&inode->i_lock);
1908 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1909 error = inode_newsize_ok(inode, offset + len);
1913 start = offset >> PAGE_CACHE_SHIFT;
1914 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1915 /* Try to avoid a swapstorm if len is impossible to satisfy */
1916 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1921 shmem_falloc.waitq = NULL;
1922 shmem_falloc.start = start;
1923 shmem_falloc.next = start;
1924 shmem_falloc.nr_falloced = 0;
1925 shmem_falloc.nr_unswapped = 0;
1926 spin_lock(&inode->i_lock);
1927 inode->i_private = &shmem_falloc;
1928 spin_unlock(&inode->i_lock);
1930 for (index = start; index < end; index++) {
1934 * Good, the fallocate(2) manpage permits EINTR: we may have
1935 * been interrupted because we are using up too much memory.
1937 if (signal_pending(current))
1939 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1942 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1945 /* Remove the !PageUptodate pages we added */
1946 shmem_undo_range(inode,
1947 (loff_t)start << PAGE_CACHE_SHIFT,
1948 (loff_t)index << PAGE_CACHE_SHIFT, true);
1953 * Inform shmem_writepage() how far we have reached.
1954 * No need for lock or barrier: we have the page lock.
1956 shmem_falloc.next++;
1957 if (!PageUptodate(page))
1958 shmem_falloc.nr_falloced++;
1961 * If !PageUptodate, leave it that way so that freeable pages
1962 * can be recognized if we need to rollback on error later.
1963 * But set_page_dirty so that memory pressure will swap rather
1964 * than free the pages we are allocating (and SGP_CACHE pages
1965 * might still be clean: we now need to mark those dirty too).
1967 set_page_dirty(page);
1969 page_cache_release(page);
1973 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1974 i_size_write(inode, offset + len);
1975 inode->i_ctime = CURRENT_TIME;
1977 spin_lock(&inode->i_lock);
1978 inode->i_private = NULL;
1979 spin_unlock(&inode->i_lock);
1981 mutex_unlock(&inode->i_mutex);
1985 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1987 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1989 buf->f_type = TMPFS_MAGIC;
1990 buf->f_bsize = PAGE_CACHE_SIZE;
1991 buf->f_namelen = NAME_MAX;
1992 if (sbinfo->max_blocks) {
1993 buf->f_blocks = sbinfo->max_blocks;
1995 buf->f_bfree = sbinfo->max_blocks -
1996 percpu_counter_sum(&sbinfo->used_blocks);
1998 if (sbinfo->max_inodes) {
1999 buf->f_files = sbinfo->max_inodes;
2000 buf->f_ffree = sbinfo->free_inodes;
2002 /* else leave those fields 0 like simple_statfs */
2007 * File creation. Allocate an inode, and we're done..
2010 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2012 struct inode *inode;
2013 int error = -ENOSPC;
2015 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2017 error = simple_acl_create(dir, inode);
2020 error = security_inode_init_security(inode, dir,
2022 shmem_initxattrs, NULL);
2023 if (error && error != -EOPNOTSUPP)
2027 dir->i_size += BOGO_DIRENT_SIZE;
2028 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2029 d_instantiate(dentry, inode);
2030 dget(dentry); /* Extra count - pin the dentry in core */
2039 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2041 struct inode *inode;
2042 int error = -ENOSPC;
2044 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2046 error = security_inode_init_security(inode, dir,
2048 shmem_initxattrs, NULL);
2049 if (error && error != -EOPNOTSUPP)
2051 error = simple_acl_create(dir, inode);
2054 d_tmpfile(dentry, inode);
2062 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2066 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2072 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2075 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2081 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2083 struct inode *inode = old_dentry->d_inode;
2087 * No ordinary (disk based) filesystem counts links as inodes;
2088 * but each new link needs a new dentry, pinning lowmem, and
2089 * tmpfs dentries cannot be pruned until they are unlinked.
2091 ret = shmem_reserve_inode(inode->i_sb);
2095 dir->i_size += BOGO_DIRENT_SIZE;
2096 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2098 ihold(inode); /* New dentry reference */
2099 dget(dentry); /* Extra pinning count for the created dentry */
2100 d_instantiate(dentry, inode);
2105 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2107 struct inode *inode = dentry->d_inode;
2109 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2110 shmem_free_inode(inode->i_sb);
2112 dir->i_size -= BOGO_DIRENT_SIZE;
2113 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2115 dput(dentry); /* Undo the count from "create" - this does all the work */
2119 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2121 if (!simple_empty(dentry))
2124 drop_nlink(dentry->d_inode);
2126 return shmem_unlink(dir, dentry);
2130 * The VFS layer already does all the dentry stuff for rename,
2131 * we just have to decrement the usage count for the target if
2132 * it exists so that the VFS layer correctly free's it when it
2135 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2137 struct inode *inode = old_dentry->d_inode;
2138 int they_are_dirs = S_ISDIR(inode->i_mode);
2140 if (!simple_empty(new_dentry))
2143 if (new_dentry->d_inode) {
2144 (void) shmem_unlink(new_dir, new_dentry);
2145 if (they_are_dirs) {
2146 drop_nlink(new_dentry->d_inode);
2147 drop_nlink(old_dir);
2149 } else if (they_are_dirs) {
2150 drop_nlink(old_dir);
2154 old_dir->i_size -= BOGO_DIRENT_SIZE;
2155 new_dir->i_size += BOGO_DIRENT_SIZE;
2156 old_dir->i_ctime = old_dir->i_mtime =
2157 new_dir->i_ctime = new_dir->i_mtime =
2158 inode->i_ctime = CURRENT_TIME;
2162 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2166 struct inode *inode;
2169 struct shmem_inode_info *info;
2171 len = strlen(symname) + 1;
2172 if (len > PAGE_CACHE_SIZE)
2173 return -ENAMETOOLONG;
2175 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2179 error = security_inode_init_security(inode, dir, &dentry->d_name,
2180 shmem_initxattrs, NULL);
2182 if (error != -EOPNOTSUPP) {
2189 info = SHMEM_I(inode);
2190 inode->i_size = len-1;
2191 if (len <= SHORT_SYMLINK_LEN) {
2192 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2193 if (!info->symlink) {
2197 inode->i_op = &shmem_short_symlink_operations;
2199 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2204 inode->i_mapping->a_ops = &shmem_aops;
2205 inode->i_op = &shmem_symlink_inode_operations;
2206 kaddr = kmap_atomic(page);
2207 memcpy(kaddr, symname, len);
2208 kunmap_atomic(kaddr);
2209 SetPageUptodate(page);
2210 set_page_dirty(page);
2212 page_cache_release(page);
2214 dir->i_size += BOGO_DIRENT_SIZE;
2215 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2216 d_instantiate(dentry, inode);
2221 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2223 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2227 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2229 struct page *page = NULL;
2230 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2231 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2237 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2239 if (!IS_ERR(nd_get_link(nd))) {
2240 struct page *page = cookie;
2242 mark_page_accessed(page);
2243 page_cache_release(page);
2247 #ifdef CONFIG_TMPFS_XATTR
2249 * Superblocks without xattr inode operations may get some security.* xattr
2250 * support from the LSM "for free". As soon as we have any other xattrs
2251 * like ACLs, we also need to implement the security.* handlers at
2252 * filesystem level, though.
2256 * Callback for security_inode_init_security() for acquiring xattrs.
2258 static int shmem_initxattrs(struct inode *inode,
2259 const struct xattr *xattr_array,
2262 struct shmem_inode_info *info = SHMEM_I(inode);
2263 const struct xattr *xattr;
2264 struct simple_xattr *new_xattr;
2267 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2268 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
2272 len = strlen(xattr->name) + 1;
2273 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2275 if (!new_xattr->name) {
2280 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2281 XATTR_SECURITY_PREFIX_LEN);
2282 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2285 simple_xattr_list_add(&info->xattrs, new_xattr);
2291 static const struct xattr_handler *shmem_xattr_handlers[] = {
2292 #ifdef CONFIG_TMPFS_POSIX_ACL
2293 &posix_acl_access_xattr_handler,
2294 &posix_acl_default_xattr_handler,
2299 static int shmem_xattr_validate(const char *name)
2301 struct { const char *prefix; size_t len; } arr[] = {
2302 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2303 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2307 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2308 size_t preflen = arr[i].len;
2309 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2318 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2319 void *buffer, size_t size)
2321 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2325 * If this is a request for a synthetic attribute in the system.*
2326 * namespace use the generic infrastructure to resolve a handler
2327 * for it via sb->s_xattr.
2329 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2330 return generic_getxattr(dentry, name, buffer, size);
2332 err = shmem_xattr_validate(name);
2336 return simple_xattr_get(&info->xattrs, name, buffer, size);
2339 static int shmem_setxattr(struct dentry *dentry, const char *name,
2340 const void *value, size_t size, int flags)
2342 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2346 * If this is a request for a synthetic attribute in the system.*
2347 * namespace use the generic infrastructure to resolve a handler
2348 * for it via sb->s_xattr.
2350 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2351 return generic_setxattr(dentry, name, value, size, flags);
2353 err = shmem_xattr_validate(name);
2357 return simple_xattr_set(&info->xattrs, name, value, size, flags);
2360 static int shmem_removexattr(struct dentry *dentry, const char *name)
2362 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2366 * If this is a request for a synthetic attribute in the system.*
2367 * namespace use the generic infrastructure to resolve a handler
2368 * for it via sb->s_xattr.
2370 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2371 return generic_removexattr(dentry, name);
2373 err = shmem_xattr_validate(name);
2377 return simple_xattr_remove(&info->xattrs, name);
2380 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2382 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2383 return simple_xattr_list(&info->xattrs, buffer, size);
2385 #endif /* CONFIG_TMPFS_XATTR */
2387 static const struct inode_operations shmem_short_symlink_operations = {
2388 .readlink = generic_readlink,
2389 .follow_link = shmem_follow_short_symlink,
2390 #ifdef CONFIG_TMPFS_XATTR
2391 .setxattr = shmem_setxattr,
2392 .getxattr = shmem_getxattr,
2393 .listxattr = shmem_listxattr,
2394 .removexattr = shmem_removexattr,
2398 static const struct inode_operations shmem_symlink_inode_operations = {
2399 .readlink = generic_readlink,
2400 .follow_link = shmem_follow_link,
2401 .put_link = shmem_put_link,
2402 #ifdef CONFIG_TMPFS_XATTR
2403 .setxattr = shmem_setxattr,
2404 .getxattr = shmem_getxattr,
2405 .listxattr = shmem_listxattr,
2406 .removexattr = shmem_removexattr,
2410 static struct dentry *shmem_get_parent(struct dentry *child)
2412 return ERR_PTR(-ESTALE);
2415 static int shmem_match(struct inode *ino, void *vfh)
2419 inum = (inum << 32) | fh[1];
2420 return ino->i_ino == inum && fh[0] == ino->i_generation;
2423 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2424 struct fid *fid, int fh_len, int fh_type)
2426 struct inode *inode;
2427 struct dentry *dentry = NULL;
2434 inum = (inum << 32) | fid->raw[1];
2436 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2437 shmem_match, fid->raw);
2439 dentry = d_find_alias(inode);
2446 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2447 struct inode *parent)
2451 return FILEID_INVALID;
2454 if (inode_unhashed(inode)) {
2455 /* Unfortunately insert_inode_hash is not idempotent,
2456 * so as we hash inodes here rather than at creation
2457 * time, we need a lock to ensure we only try
2460 static DEFINE_SPINLOCK(lock);
2462 if (inode_unhashed(inode))
2463 __insert_inode_hash(inode,
2464 inode->i_ino + inode->i_generation);
2468 fh[0] = inode->i_generation;
2469 fh[1] = inode->i_ino;
2470 fh[2] = ((__u64)inode->i_ino) >> 32;
2476 static const struct export_operations shmem_export_ops = {
2477 .get_parent = shmem_get_parent,
2478 .encode_fh = shmem_encode_fh,
2479 .fh_to_dentry = shmem_fh_to_dentry,
2482 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2485 char *this_char, *value, *rest;
2486 struct mempolicy *mpol = NULL;
2490 while (options != NULL) {
2491 this_char = options;
2494 * NUL-terminate this option: unfortunately,
2495 * mount options form a comma-separated list,
2496 * but mpol's nodelist may also contain commas.
2498 options = strchr(options, ',');
2499 if (options == NULL)
2502 if (!isdigit(*options)) {
2509 if ((value = strchr(this_char,'=')) != NULL) {
2513 "tmpfs: No value for mount option '%s'\n",
2518 if (!strcmp(this_char,"size")) {
2519 unsigned long long size;
2520 size = memparse(value,&rest);
2522 size <<= PAGE_SHIFT;
2523 size *= totalram_pages;
2529 sbinfo->max_blocks =
2530 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2531 } else if (!strcmp(this_char,"nr_blocks")) {
2532 sbinfo->max_blocks = memparse(value, &rest);
2535 } else if (!strcmp(this_char,"nr_inodes")) {
2536 sbinfo->max_inodes = memparse(value, &rest);
2539 } else if (!strcmp(this_char,"mode")) {
2542 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2545 } else if (!strcmp(this_char,"uid")) {
2548 uid = simple_strtoul(value, &rest, 0);
2551 sbinfo->uid = make_kuid(current_user_ns(), uid);
2552 if (!uid_valid(sbinfo->uid))
2554 } else if (!strcmp(this_char,"gid")) {
2557 gid = simple_strtoul(value, &rest, 0);
2560 sbinfo->gid = make_kgid(current_user_ns(), gid);
2561 if (!gid_valid(sbinfo->gid))
2563 } else if (!strcmp(this_char,"mpol")) {
2566 if (mpol_parse_str(value, &mpol))
2569 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2574 sbinfo->mpol = mpol;
2578 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2586 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2588 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2589 struct shmem_sb_info config = *sbinfo;
2590 unsigned long inodes;
2591 int error = -EINVAL;
2594 if (shmem_parse_options(data, &config, true))
2597 spin_lock(&sbinfo->stat_lock);
2598 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2599 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2601 if (config.max_inodes < inodes)
2604 * Those tests disallow limited->unlimited while any are in use;
2605 * but we must separately disallow unlimited->limited, because
2606 * in that case we have no record of how much is already in use.
2608 if (config.max_blocks && !sbinfo->max_blocks)
2610 if (config.max_inodes && !sbinfo->max_inodes)
2614 sbinfo->max_blocks = config.max_blocks;
2615 sbinfo->max_inodes = config.max_inodes;
2616 sbinfo->free_inodes = config.max_inodes - inodes;
2619 * Preserve previous mempolicy unless mpol remount option was specified.
2622 mpol_put(sbinfo->mpol);
2623 sbinfo->mpol = config.mpol; /* transfers initial ref */
2626 spin_unlock(&sbinfo->stat_lock);
2630 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2632 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2634 if (sbinfo->max_blocks != shmem_default_max_blocks())
2635 seq_printf(seq, ",size=%luk",
2636 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2637 if (sbinfo->max_inodes != shmem_default_max_inodes())
2638 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2639 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2640 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2641 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2642 seq_printf(seq, ",uid=%u",
2643 from_kuid_munged(&init_user_ns, sbinfo->uid));
2644 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2645 seq_printf(seq, ",gid=%u",
2646 from_kgid_munged(&init_user_ns, sbinfo->gid));
2647 shmem_show_mpol(seq, sbinfo->mpol);
2650 #endif /* CONFIG_TMPFS */
2652 static void shmem_put_super(struct super_block *sb)
2654 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2656 percpu_counter_destroy(&sbinfo->used_blocks);
2657 mpol_put(sbinfo->mpol);
2659 sb->s_fs_info = NULL;
2662 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2664 struct inode *inode;
2665 struct shmem_sb_info *sbinfo;
2668 /* Round up to L1_CACHE_BYTES to resist false sharing */
2669 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2670 L1_CACHE_BYTES), GFP_KERNEL);
2674 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2675 sbinfo->uid = current_fsuid();
2676 sbinfo->gid = current_fsgid();
2677 sb->s_fs_info = sbinfo;
2681 * Per default we only allow half of the physical ram per
2682 * tmpfs instance, limiting inodes to one per page of lowmem;
2683 * but the internal instance is left unlimited.
2685 if (!(sb->s_flags & MS_KERNMOUNT)) {
2686 sbinfo->max_blocks = shmem_default_max_blocks();
2687 sbinfo->max_inodes = shmem_default_max_inodes();
2688 if (shmem_parse_options(data, sbinfo, false)) {
2693 sb->s_flags |= MS_NOUSER;
2695 sb->s_export_op = &shmem_export_ops;
2696 sb->s_flags |= MS_NOSEC;
2698 sb->s_flags |= MS_NOUSER;
2701 spin_lock_init(&sbinfo->stat_lock);
2702 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2704 sbinfo->free_inodes = sbinfo->max_inodes;
2706 sb->s_maxbytes = MAX_LFS_FILESIZE;
2707 sb->s_blocksize = PAGE_CACHE_SIZE;
2708 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2709 sb->s_magic = TMPFS_MAGIC;
2710 sb->s_op = &shmem_ops;
2711 sb->s_time_gran = 1;
2712 #ifdef CONFIG_TMPFS_XATTR
2713 sb->s_xattr = shmem_xattr_handlers;
2715 #ifdef CONFIG_TMPFS_POSIX_ACL
2716 sb->s_flags |= MS_POSIXACL;
2719 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2722 inode->i_uid = sbinfo->uid;
2723 inode->i_gid = sbinfo->gid;
2724 sb->s_root = d_make_root(inode);
2730 shmem_put_super(sb);
2734 static struct kmem_cache *shmem_inode_cachep;
2736 static struct inode *shmem_alloc_inode(struct super_block *sb)
2738 struct shmem_inode_info *info;
2739 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2742 return &info->vfs_inode;
2745 static void shmem_destroy_callback(struct rcu_head *head)
2747 struct inode *inode = container_of(head, struct inode, i_rcu);
2748 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2751 static void shmem_destroy_inode(struct inode *inode)
2753 if (S_ISREG(inode->i_mode))
2754 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2755 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2758 static void shmem_init_inode(void *foo)
2760 struct shmem_inode_info *info = foo;
2761 inode_init_once(&info->vfs_inode);
2764 static int shmem_init_inodecache(void)
2766 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2767 sizeof(struct shmem_inode_info),
2768 0, SLAB_PANIC, shmem_init_inode);
2772 static void shmem_destroy_inodecache(void)
2774 kmem_cache_destroy(shmem_inode_cachep);
2777 static const struct address_space_operations shmem_aops = {
2778 .writepage = shmem_writepage,
2779 .set_page_dirty = __set_page_dirty_no_writeback,
2781 .write_begin = shmem_write_begin,
2782 .write_end = shmem_write_end,
2784 .migratepage = migrate_page,
2785 .error_remove_page = generic_error_remove_page,
2788 static const struct file_operations shmem_file_operations = {
2791 .llseek = shmem_file_llseek,
2792 .read = do_sync_read,
2793 .write = do_sync_write,
2794 .aio_read = shmem_file_aio_read,
2795 .aio_write = generic_file_aio_write,
2796 .fsync = noop_fsync,
2797 .splice_read = shmem_file_splice_read,
2798 .splice_write = generic_file_splice_write,
2799 .fallocate = shmem_fallocate,
2803 static const struct inode_operations shmem_inode_operations = {
2804 .setattr = shmem_setattr,
2805 #ifdef CONFIG_TMPFS_XATTR
2806 .setxattr = shmem_setxattr,
2807 .getxattr = shmem_getxattr,
2808 .listxattr = shmem_listxattr,
2809 .removexattr = shmem_removexattr,
2810 .set_acl = simple_set_acl,
2814 static const struct inode_operations shmem_dir_inode_operations = {
2816 .create = shmem_create,
2817 .lookup = simple_lookup,
2819 .unlink = shmem_unlink,
2820 .symlink = shmem_symlink,
2821 .mkdir = shmem_mkdir,
2822 .rmdir = shmem_rmdir,
2823 .mknod = shmem_mknod,
2824 .rename = shmem_rename,
2825 .tmpfile = shmem_tmpfile,
2827 #ifdef CONFIG_TMPFS_XATTR
2828 .setxattr = shmem_setxattr,
2829 .getxattr = shmem_getxattr,
2830 .listxattr = shmem_listxattr,
2831 .removexattr = shmem_removexattr,
2833 #ifdef CONFIG_TMPFS_POSIX_ACL
2834 .setattr = shmem_setattr,
2835 .set_acl = simple_set_acl,
2839 static const struct inode_operations shmem_special_inode_operations = {
2840 #ifdef CONFIG_TMPFS_XATTR
2841 .setxattr = shmem_setxattr,
2842 .getxattr = shmem_getxattr,
2843 .listxattr = shmem_listxattr,
2844 .removexattr = shmem_removexattr,
2846 #ifdef CONFIG_TMPFS_POSIX_ACL
2847 .setattr = shmem_setattr,
2848 .set_acl = simple_set_acl,
2852 static const struct super_operations shmem_ops = {
2853 .alloc_inode = shmem_alloc_inode,
2854 .destroy_inode = shmem_destroy_inode,
2856 .statfs = shmem_statfs,
2857 .remount_fs = shmem_remount_fs,
2858 .show_options = shmem_show_options,
2860 .evict_inode = shmem_evict_inode,
2861 .drop_inode = generic_delete_inode,
2862 .put_super = shmem_put_super,
2865 static const struct vm_operations_struct shmem_vm_ops = {
2866 .fault = shmem_fault,
2868 .set_policy = shmem_set_policy,
2869 .get_policy = shmem_get_policy,
2871 .remap_pages = generic_file_remap_pages,
2874 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2875 int flags, const char *dev_name, void *data)
2877 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2880 static struct file_system_type shmem_fs_type = {
2881 .owner = THIS_MODULE,
2883 .mount = shmem_mount,
2884 .kill_sb = kill_litter_super,
2885 .fs_flags = FS_USERNS_MOUNT,
2888 int __init shmem_init(void)
2892 /* If rootfs called this, don't re-init */
2893 if (shmem_inode_cachep)
2896 error = bdi_init(&shmem_backing_dev_info);
2900 error = shmem_init_inodecache();
2904 error = register_filesystem(&shmem_fs_type);
2906 printk(KERN_ERR "Could not register tmpfs\n");
2910 shm_mnt = kern_mount(&shmem_fs_type);
2911 if (IS_ERR(shm_mnt)) {
2912 error = PTR_ERR(shm_mnt);
2913 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2919 unregister_filesystem(&shmem_fs_type);
2921 shmem_destroy_inodecache();
2923 bdi_destroy(&shmem_backing_dev_info);
2925 shm_mnt = ERR_PTR(error);
2929 #else /* !CONFIG_SHMEM */
2932 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2934 * This is intended for small system where the benefits of the full
2935 * shmem code (swap-backed and resource-limited) are outweighed by
2936 * their complexity. On systems without swap this code should be
2937 * effectively equivalent, but much lighter weight.
2940 static struct file_system_type shmem_fs_type = {
2942 .mount = ramfs_mount,
2943 .kill_sb = kill_litter_super,
2944 .fs_flags = FS_USERNS_MOUNT,
2947 int __init shmem_init(void)
2949 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2951 shm_mnt = kern_mount(&shmem_fs_type);
2952 BUG_ON(IS_ERR(shm_mnt));
2957 int shmem_unuse(swp_entry_t swap, struct page *page)
2962 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2967 void shmem_unlock_mapping(struct address_space *mapping)
2971 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2973 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2975 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2977 #define shmem_vm_ops generic_file_vm_ops
2978 #define shmem_file_operations ramfs_file_operations
2979 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2980 #define shmem_acct_size(flags, size) 0
2981 #define shmem_unacct_size(flags, size) do {} while (0)
2983 #endif /* CONFIG_SHMEM */
2987 static struct dentry_operations anon_ops = {
2988 .d_dname = simple_dname
2991 static struct file *__shmem_file_setup(const char *name, loff_t size,
2992 unsigned long flags, unsigned int i_flags)
2995 struct inode *inode;
2997 struct super_block *sb;
3000 if (IS_ERR(shm_mnt))
3001 return ERR_CAST(shm_mnt);
3003 if (size < 0 || size > MAX_LFS_FILESIZE)
3004 return ERR_PTR(-EINVAL);
3006 if (shmem_acct_size(flags, size))
3007 return ERR_PTR(-ENOMEM);
3009 res = ERR_PTR(-ENOMEM);
3011 this.len = strlen(name);
3012 this.hash = 0; /* will go */
3013 sb = shm_mnt->mnt_sb;
3014 path.dentry = d_alloc_pseudo(sb, &this);
3017 d_set_d_op(path.dentry, &anon_ops);
3018 path.mnt = mntget(shm_mnt);
3020 res = ERR_PTR(-ENOSPC);
3021 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
3025 inode->i_flags |= i_flags;
3026 d_instantiate(path.dentry, inode);
3027 inode->i_size = size;
3028 clear_nlink(inode); /* It is unlinked */
3029 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
3033 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
3034 &shmem_file_operations);
3043 shmem_unacct_size(flags, size);
3048 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3049 * kernel internal. There will be NO LSM permission checks against the
3050 * underlying inode. So users of this interface must do LSM checks at a
3051 * higher layer. The one user is the big_key implementation. LSM checks
3052 * are provided at the key level rather than the inode level.
3053 * @name: name for dentry (to be seen in /proc/<pid>/maps
3054 * @size: size to be set for the file
3055 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3057 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
3059 return __shmem_file_setup(name, size, flags, S_PRIVATE);
3063 * shmem_file_setup - get an unlinked file living in tmpfs
3064 * @name: name for dentry (to be seen in /proc/<pid>/maps
3065 * @size: size to be set for the file
3066 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3068 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3070 return __shmem_file_setup(name, size, flags, 0);
3072 EXPORT_SYMBOL_GPL(shmem_file_setup);
3075 * shmem_zero_setup - setup a shared anonymous mapping
3076 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3078 int shmem_zero_setup(struct vm_area_struct *vma)
3081 loff_t size = vma->vm_end - vma->vm_start;
3083 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3085 return PTR_ERR(file);
3089 vma->vm_file = file;
3090 vma->vm_ops = &shmem_vm_ops;
3095 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3096 * @mapping: the page's address_space
3097 * @index: the page index
3098 * @gfp: the page allocator flags to use if allocating
3100 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3101 * with any new page allocations done using the specified allocation flags.
3102 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3103 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3104 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3106 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3107 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3109 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3110 pgoff_t index, gfp_t gfp)
3113 struct inode *inode = mapping->host;
3117 BUG_ON(mapping->a_ops != &shmem_aops);
3118 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3120 page = ERR_PTR(error);
3126 * The tiny !SHMEM case uses ramfs without swap
3128 return read_cache_page_gfp(mapping, index, gfp);
3131 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);