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/generic_acl.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 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
251 item = radix_tree_deref_slot_protected(pslot,
252 &mapping->tree_lock);
253 if (item != expected)
256 radix_tree_replace_slot(pslot, replacement);
258 radix_tree_delete(&mapping->page_tree, index);
263 * Sometimes, before we decide whether to proceed or to fail, we must check
264 * that an entry was not already brought back from swap by a racing thread.
266 * Checking page is not enough: by the time a SwapCache page is locked, it
267 * might be reused, and again be SwapCache, using the same swap as before.
269 static bool shmem_confirm_swap(struct address_space *mapping,
270 pgoff_t index, swp_entry_t swap)
275 item = radix_tree_lookup(&mapping->page_tree, index);
277 return item == swp_to_radix_entry(swap);
281 * Like add_to_page_cache_locked, but error if expected item has gone.
283 static int shmem_add_to_page_cache(struct page *page,
284 struct address_space *mapping,
285 pgoff_t index, gfp_t gfp, void *expected)
289 VM_BUG_ON(!PageLocked(page));
290 VM_BUG_ON(!PageSwapBacked(page));
292 page_cache_get(page);
293 page->mapping = mapping;
296 spin_lock_irq(&mapping->tree_lock);
298 error = radix_tree_insert(&mapping->page_tree, index, page);
300 error = shmem_radix_tree_replace(mapping, index, expected,
304 __inc_zone_page_state(page, NR_FILE_PAGES);
305 __inc_zone_page_state(page, NR_SHMEM);
306 spin_unlock_irq(&mapping->tree_lock);
308 page->mapping = NULL;
309 spin_unlock_irq(&mapping->tree_lock);
310 page_cache_release(page);
316 * Like delete_from_page_cache, but substitutes swap for page.
318 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
320 struct address_space *mapping = page->mapping;
323 spin_lock_irq(&mapping->tree_lock);
324 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
325 page->mapping = NULL;
327 __dec_zone_page_state(page, NR_FILE_PAGES);
328 __dec_zone_page_state(page, NR_SHMEM);
329 spin_unlock_irq(&mapping->tree_lock);
330 page_cache_release(page);
335 * Like find_get_pages, but collecting swap entries as well as pages.
337 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
338 pgoff_t start, unsigned int nr_pages,
339 struct page **pages, pgoff_t *indices)
342 unsigned int ret = 0;
343 struct radix_tree_iter iter;
350 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
353 page = radix_tree_deref_slot(slot);
356 if (radix_tree_exception(page)) {
357 if (radix_tree_deref_retry(page))
360 * Otherwise, we must be storing a swap entry
361 * here as an exceptional entry: so return it
362 * without attempting to raise page count.
366 if (!page_cache_get_speculative(page))
369 /* Has the page moved? */
370 if (unlikely(page != *slot)) {
371 page_cache_release(page);
375 indices[ret] = iter.index;
377 if (++ret == nr_pages)
385 * Remove swap entry from radix tree, free the swap and its page cache.
387 static int shmem_free_swap(struct address_space *mapping,
388 pgoff_t index, void *radswap)
392 spin_lock_irq(&mapping->tree_lock);
393 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
394 spin_unlock_irq(&mapping->tree_lock);
396 free_swap_and_cache(radix_to_swp_entry(radswap));
401 * Pagevec may contain swap entries, so shuffle up pages before releasing.
403 static void shmem_deswap_pagevec(struct pagevec *pvec)
407 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
408 struct page *page = pvec->pages[i];
409 if (!radix_tree_exceptional_entry(page))
410 pvec->pages[j++] = page;
416 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
418 void shmem_unlock_mapping(struct address_space *mapping)
421 pgoff_t indices[PAGEVEC_SIZE];
424 pagevec_init(&pvec, 0);
426 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
428 while (!mapping_unevictable(mapping)) {
430 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
431 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
433 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
434 PAGEVEC_SIZE, pvec.pages, indices);
437 index = indices[pvec.nr - 1] + 1;
438 shmem_deswap_pagevec(&pvec);
439 check_move_unevictable_pages(pvec.pages, pvec.nr);
440 pagevec_release(&pvec);
446 * Remove range of pages and swap entries from radix tree, and free them.
447 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
449 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
452 struct address_space *mapping = inode->i_mapping;
453 struct shmem_inode_info *info = SHMEM_I(inode);
454 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
455 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
456 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
457 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
459 pgoff_t indices[PAGEVEC_SIZE];
460 long nr_swaps_freed = 0;
465 end = -1; /* unsigned, so actually very big */
467 pagevec_init(&pvec, 0);
469 while (index < end) {
470 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
471 min(end - index, (pgoff_t)PAGEVEC_SIZE),
472 pvec.pages, indices);
475 mem_cgroup_uncharge_start();
476 for (i = 0; i < pagevec_count(&pvec); i++) {
477 struct page *page = pvec.pages[i];
483 if (radix_tree_exceptional_entry(page)) {
486 nr_swaps_freed += !shmem_free_swap(mapping,
491 if (!trylock_page(page))
493 if (!unfalloc || !PageUptodate(page)) {
494 if (page->mapping == mapping) {
495 VM_BUG_ON(PageWriteback(page));
496 truncate_inode_page(mapping, page);
501 shmem_deswap_pagevec(&pvec);
502 pagevec_release(&pvec);
503 mem_cgroup_uncharge_end();
509 struct page *page = NULL;
510 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
512 unsigned int top = PAGE_CACHE_SIZE;
517 zero_user_segment(page, partial_start, top);
518 set_page_dirty(page);
520 page_cache_release(page);
524 struct page *page = NULL;
525 shmem_getpage(inode, end, &page, SGP_READ, NULL);
527 zero_user_segment(page, 0, partial_end);
528 set_page_dirty(page);
530 page_cache_release(page);
537 while (index < end) {
539 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
540 min(end - index, (pgoff_t)PAGEVEC_SIZE),
541 pvec.pages, indices);
543 /* If all gone or hole-punch or unfalloc, we're done */
544 if (index == start || end != -1)
546 /* But if truncating, restart to make sure all gone */
550 mem_cgroup_uncharge_start();
551 for (i = 0; i < pagevec_count(&pvec); i++) {
552 struct page *page = pvec.pages[i];
558 if (radix_tree_exceptional_entry(page)) {
561 if (shmem_free_swap(mapping, index, page)) {
562 /* Swap was replaced by page: retry */
571 if (!unfalloc || !PageUptodate(page)) {
572 if (page->mapping == mapping) {
573 VM_BUG_ON(PageWriteback(page));
574 truncate_inode_page(mapping, page);
576 /* Page was replaced by swap: retry */
584 shmem_deswap_pagevec(&pvec);
585 pagevec_release(&pvec);
586 mem_cgroup_uncharge_end();
590 spin_lock(&info->lock);
591 info->swapped -= nr_swaps_freed;
592 shmem_recalc_inode(inode);
593 spin_unlock(&info->lock);
596 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
598 shmem_undo_range(inode, lstart, lend, false);
599 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
601 EXPORT_SYMBOL_GPL(shmem_truncate_range);
603 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
605 struct inode *inode = dentry->d_inode;
608 error = inode_change_ok(inode, attr);
612 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
613 loff_t oldsize = inode->i_size;
614 loff_t newsize = attr->ia_size;
616 if (newsize != oldsize) {
617 i_size_write(inode, newsize);
618 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
620 if (newsize < oldsize) {
621 loff_t holebegin = round_up(newsize, PAGE_SIZE);
622 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
623 shmem_truncate_range(inode, newsize, (loff_t)-1);
624 /* unmap again to remove racily COWed private pages */
625 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
629 setattr_copy(inode, attr);
630 #ifdef CONFIG_TMPFS_POSIX_ACL
631 if (attr->ia_valid & ATTR_MODE)
632 error = generic_acl_chmod(inode);
637 static void shmem_evict_inode(struct inode *inode)
639 struct shmem_inode_info *info = SHMEM_I(inode);
641 if (inode->i_mapping->a_ops == &shmem_aops) {
642 shmem_unacct_size(info->flags, inode->i_size);
644 shmem_truncate_range(inode, 0, (loff_t)-1);
645 if (!list_empty(&info->swaplist)) {
646 mutex_lock(&shmem_swaplist_mutex);
647 list_del_init(&info->swaplist);
648 mutex_unlock(&shmem_swaplist_mutex);
651 kfree(info->symlink);
653 simple_xattrs_free(&info->xattrs);
654 WARN_ON(inode->i_blocks);
655 shmem_free_inode(inode->i_sb);
660 * If swap found in inode, free it and move page from swapcache to filecache.
662 static int shmem_unuse_inode(struct shmem_inode_info *info,
663 swp_entry_t swap, struct page **pagep)
665 struct address_space *mapping = info->vfs_inode.i_mapping;
671 radswap = swp_to_radix_entry(swap);
672 index = radix_tree_locate_item(&mapping->page_tree, radswap);
677 * Move _head_ to start search for next from here.
678 * But be careful: shmem_evict_inode checks list_empty without taking
679 * mutex, and there's an instant in list_move_tail when info->swaplist
680 * would appear empty, if it were the only one on shmem_swaplist.
682 if (shmem_swaplist.next != &info->swaplist)
683 list_move_tail(&shmem_swaplist, &info->swaplist);
685 gfp = mapping_gfp_mask(mapping);
686 if (shmem_should_replace_page(*pagep, gfp)) {
687 mutex_unlock(&shmem_swaplist_mutex);
688 error = shmem_replace_page(pagep, gfp, info, index);
689 mutex_lock(&shmem_swaplist_mutex);
691 * We needed to drop mutex to make that restrictive page
692 * allocation, but the inode might have been freed while we
693 * dropped it: although a racing shmem_evict_inode() cannot
694 * complete without emptying the radix_tree, our page lock
695 * on this swapcache page is not enough to prevent that -
696 * free_swap_and_cache() of our swap entry will only
697 * trylock_page(), removing swap from radix_tree whatever.
699 * We must not proceed to shmem_add_to_page_cache() if the
700 * inode has been freed, but of course we cannot rely on
701 * inode or mapping or info to check that. However, we can
702 * safely check if our swap entry is still in use (and here
703 * it can't have got reused for another page): if it's still
704 * in use, then the inode cannot have been freed yet, and we
705 * can safely proceed (if it's no longer in use, that tells
706 * nothing about the inode, but we don't need to unuse swap).
708 if (!page_swapcount(*pagep))
713 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
714 * but also to hold up shmem_evict_inode(): so inode cannot be freed
715 * beneath us (pagelock doesn't help until the page is in pagecache).
718 error = shmem_add_to_page_cache(*pagep, mapping, index,
719 GFP_NOWAIT, radswap);
720 if (error != -ENOMEM) {
722 * Truncation and eviction use free_swap_and_cache(), which
723 * only does trylock page: if we raced, best clean up here.
725 delete_from_swap_cache(*pagep);
726 set_page_dirty(*pagep);
728 spin_lock(&info->lock);
730 spin_unlock(&info->lock);
733 error = 1; /* not an error, but entry was found */
739 * Search through swapped inodes to find and replace swap by page.
741 int shmem_unuse(swp_entry_t swap, struct page *page)
743 struct list_head *this, *next;
744 struct shmem_inode_info *info;
749 * There's a faint possibility that swap page was replaced before
750 * caller locked it: caller will come back later with the right page.
752 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
756 * Charge page using GFP_KERNEL while we can wait, before taking
757 * the shmem_swaplist_mutex which might hold up shmem_writepage().
758 * Charged back to the user (not to caller) when swap account is used.
760 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
763 /* No radix_tree_preload: swap entry keeps a place for page in tree */
765 mutex_lock(&shmem_swaplist_mutex);
766 list_for_each_safe(this, next, &shmem_swaplist) {
767 info = list_entry(this, struct shmem_inode_info, swaplist);
769 found = shmem_unuse_inode(info, swap, &page);
771 list_del_init(&info->swaplist);
776 mutex_unlock(&shmem_swaplist_mutex);
782 page_cache_release(page);
787 * Move the page from the page cache to the swap cache.
789 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
791 struct shmem_inode_info *info;
792 struct address_space *mapping;
797 BUG_ON(!PageLocked(page));
798 mapping = page->mapping;
800 inode = mapping->host;
801 info = SHMEM_I(inode);
802 if (info->flags & VM_LOCKED)
804 if (!total_swap_pages)
808 * shmem_backing_dev_info's capabilities prevent regular writeback or
809 * sync from ever calling shmem_writepage; but a stacking filesystem
810 * might use ->writepage of its underlying filesystem, in which case
811 * tmpfs should write out to swap only in response to memory pressure,
812 * and not for the writeback threads or sync.
814 if (!wbc->for_reclaim) {
815 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
820 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
821 * value into swapfile.c, the only way we can correctly account for a
822 * fallocated page arriving here is now to initialize it and write it.
824 * That's okay for a page already fallocated earlier, but if we have
825 * not yet completed the fallocation, then (a) we want to keep track
826 * of this page in case we have to undo it, and (b) it may not be a
827 * good idea to continue anyway, once we're pushing into swap. So
828 * reactivate the page, and let shmem_fallocate() quit when too many.
830 if (!PageUptodate(page)) {
831 if (inode->i_private) {
832 struct shmem_falloc *shmem_falloc;
833 spin_lock(&inode->i_lock);
834 shmem_falloc = inode->i_private;
836 !shmem_falloc->waitq &&
837 index >= shmem_falloc->start &&
838 index < shmem_falloc->next)
839 shmem_falloc->nr_unswapped++;
842 spin_unlock(&inode->i_lock);
846 clear_highpage(page);
847 flush_dcache_page(page);
848 SetPageUptodate(page);
851 swap = get_swap_page();
856 * Add inode to shmem_unuse()'s list of swapped-out inodes,
857 * if it's not already there. Do it now before the page is
858 * moved to swap cache, when its pagelock no longer protects
859 * the inode from eviction. But don't unlock the mutex until
860 * we've incremented swapped, because shmem_unuse_inode() will
861 * prune a !swapped inode from the swaplist under this mutex.
863 mutex_lock(&shmem_swaplist_mutex);
864 if (list_empty(&info->swaplist))
865 list_add_tail(&info->swaplist, &shmem_swaplist);
867 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
868 swap_shmem_alloc(swap);
869 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
871 spin_lock(&info->lock);
873 shmem_recalc_inode(inode);
874 spin_unlock(&info->lock);
876 mutex_unlock(&shmem_swaplist_mutex);
877 BUG_ON(page_mapped(page));
878 swap_writepage(page, wbc);
882 mutex_unlock(&shmem_swaplist_mutex);
883 swapcache_free(swap, NULL);
885 set_page_dirty(page);
886 if (wbc->for_reclaim)
887 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
894 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
898 if (!mpol || mpol->mode == MPOL_DEFAULT)
899 return; /* show nothing */
901 mpol_to_str(buffer, sizeof(buffer), mpol);
903 seq_printf(seq, ",mpol=%s", buffer);
906 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
908 struct mempolicy *mpol = NULL;
910 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
913 spin_unlock(&sbinfo->stat_lock);
917 #endif /* CONFIG_TMPFS */
919 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
920 struct shmem_inode_info *info, pgoff_t index)
922 struct vm_area_struct pvma;
925 /* Create a pseudo vma that just contains the policy */
927 /* Bias interleave by inode number to distribute better across nodes */
928 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
930 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
932 page = swapin_readahead(swap, gfp, &pvma, 0);
934 /* Drop reference taken by mpol_shared_policy_lookup() */
935 mpol_cond_put(pvma.vm_policy);
940 static struct page *shmem_alloc_page(gfp_t gfp,
941 struct shmem_inode_info *info, pgoff_t index)
943 struct vm_area_struct pvma;
946 /* Create a pseudo vma that just contains the policy */
948 /* Bias interleave by inode number to distribute better across nodes */
949 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
951 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
953 page = alloc_page_vma(gfp, &pvma, 0);
955 /* Drop reference taken by mpol_shared_policy_lookup() */
956 mpol_cond_put(pvma.vm_policy);
960 #else /* !CONFIG_NUMA */
962 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
965 #endif /* CONFIG_TMPFS */
967 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
968 struct shmem_inode_info *info, pgoff_t index)
970 return swapin_readahead(swap, gfp, NULL, 0);
973 static inline struct page *shmem_alloc_page(gfp_t gfp,
974 struct shmem_inode_info *info, pgoff_t index)
976 return alloc_page(gfp);
978 #endif /* CONFIG_NUMA */
980 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
981 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
988 * When a page is moved from swapcache to shmem filecache (either by the
989 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
990 * shmem_unuse_inode()), it may have been read in earlier from swap, in
991 * ignorance of the mapping it belongs to. If that mapping has special
992 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
993 * we may need to copy to a suitable page before moving to filecache.
995 * In a future release, this may well be extended to respect cpuset and
996 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
997 * but for now it is a simple matter of zone.
999 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1001 return page_zonenum(page) > gfp_zone(gfp);
1004 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1005 struct shmem_inode_info *info, pgoff_t index)
1007 struct page *oldpage, *newpage;
1008 struct address_space *swap_mapping;
1013 swap_index = page_private(oldpage);
1014 swap_mapping = page_mapping(oldpage);
1017 * We have arrived here because our zones are constrained, so don't
1018 * limit chance of success by further cpuset and node constraints.
1020 gfp &= ~GFP_CONSTRAINT_MASK;
1021 newpage = shmem_alloc_page(gfp, info, index);
1025 page_cache_get(newpage);
1026 copy_highpage(newpage, oldpage);
1027 flush_dcache_page(newpage);
1029 __set_page_locked(newpage);
1030 SetPageUptodate(newpage);
1031 SetPageSwapBacked(newpage);
1032 set_page_private(newpage, swap_index);
1033 SetPageSwapCache(newpage);
1036 * Our caller will very soon move newpage out of swapcache, but it's
1037 * a nice clean interface for us to replace oldpage by newpage there.
1039 spin_lock_irq(&swap_mapping->tree_lock);
1040 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1043 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1044 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1046 spin_unlock_irq(&swap_mapping->tree_lock);
1048 if (unlikely(error)) {
1050 * Is this possible? I think not, now that our callers check
1051 * both PageSwapCache and page_private after getting page lock;
1052 * but be defensive. Reverse old to newpage for clear and free.
1056 mem_cgroup_replace_page_cache(oldpage, newpage);
1057 lru_cache_add_anon(newpage);
1061 ClearPageSwapCache(oldpage);
1062 set_page_private(oldpage, 0);
1064 unlock_page(oldpage);
1065 page_cache_release(oldpage);
1066 page_cache_release(oldpage);
1071 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1073 * If we allocate a new one we do not mark it dirty. That's up to the
1074 * vm. If we swap it in we mark it dirty since we also free the swap
1075 * entry since a page cannot live in both the swap and page cache
1077 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1078 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1080 struct address_space *mapping = inode->i_mapping;
1081 struct shmem_inode_info *info;
1082 struct shmem_sb_info *sbinfo;
1089 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1093 page = find_lock_page(mapping, index);
1094 if (radix_tree_exceptional_entry(page)) {
1095 swap = radix_to_swp_entry(page);
1099 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1100 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1105 /* fallocated page? */
1106 if (page && !PageUptodate(page)) {
1107 if (sgp != SGP_READ)
1110 page_cache_release(page);
1113 if (page || (sgp == SGP_READ && !swap.val)) {
1119 * Fast cache lookup did not find it:
1120 * bring it back from swap or allocate.
1122 info = SHMEM_I(inode);
1123 sbinfo = SHMEM_SB(inode->i_sb);
1126 /* Look it up and read it in.. */
1127 page = lookup_swap_cache(swap);
1129 /* here we actually do the io */
1131 *fault_type |= VM_FAULT_MAJOR;
1132 page = shmem_swapin(swap, gfp, info, index);
1139 /* We have to do this with page locked to prevent races */
1141 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1142 !shmem_confirm_swap(mapping, index, swap)) {
1143 error = -EEXIST; /* try again */
1146 if (!PageUptodate(page)) {
1150 wait_on_page_writeback(page);
1152 if (shmem_should_replace_page(page, gfp)) {
1153 error = shmem_replace_page(&page, gfp, info, index);
1158 error = mem_cgroup_cache_charge(page, current->mm,
1159 gfp & GFP_RECLAIM_MASK);
1161 error = shmem_add_to_page_cache(page, mapping, index,
1162 gfp, swp_to_radix_entry(swap));
1164 * We already confirmed swap under page lock, and make
1165 * no memory allocation here, so usually no possibility
1166 * of error; but free_swap_and_cache() only trylocks a
1167 * page, so it is just possible that the entry has been
1168 * truncated or holepunched since swap was confirmed.
1169 * shmem_undo_range() will have done some of the
1170 * unaccounting, now delete_from_swap_cache() will do
1171 * the rest (including mem_cgroup_uncharge_swapcache).
1172 * Reset swap.val? No, leave it so "failed" goes back to
1173 * "repeat": reading a hole and writing should succeed.
1176 delete_from_swap_cache(page);
1181 spin_lock(&info->lock);
1183 shmem_recalc_inode(inode);
1184 spin_unlock(&info->lock);
1186 delete_from_swap_cache(page);
1187 set_page_dirty(page);
1191 if (shmem_acct_block(info->flags)) {
1195 if (sbinfo->max_blocks) {
1196 if (percpu_counter_compare(&sbinfo->used_blocks,
1197 sbinfo->max_blocks) >= 0) {
1201 percpu_counter_inc(&sbinfo->used_blocks);
1204 page = shmem_alloc_page(gfp, info, index);
1210 SetPageSwapBacked(page);
1211 __set_page_locked(page);
1212 error = mem_cgroup_cache_charge(page, current->mm,
1213 gfp & GFP_RECLAIM_MASK);
1216 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
1218 error = shmem_add_to_page_cache(page, mapping, index,
1220 radix_tree_preload_end();
1223 mem_cgroup_uncharge_cache_page(page);
1226 lru_cache_add_anon(page);
1228 spin_lock(&info->lock);
1230 inode->i_blocks += BLOCKS_PER_PAGE;
1231 shmem_recalc_inode(inode);
1232 spin_unlock(&info->lock);
1236 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1238 if (sgp == SGP_FALLOC)
1242 * Let SGP_WRITE caller clear ends if write does not fill page;
1243 * but SGP_FALLOC on a page fallocated earlier must initialize
1244 * it now, lest undo on failure cancel our earlier guarantee.
1246 if (sgp != SGP_WRITE) {
1247 clear_highpage(page);
1248 flush_dcache_page(page);
1249 SetPageUptodate(page);
1251 if (sgp == SGP_DIRTY)
1252 set_page_dirty(page);
1255 /* Perhaps the file has been truncated since we checked */
1256 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1257 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1271 info = SHMEM_I(inode);
1272 ClearPageDirty(page);
1273 delete_from_page_cache(page);
1274 spin_lock(&info->lock);
1276 inode->i_blocks -= BLOCKS_PER_PAGE;
1277 spin_unlock(&info->lock);
1279 sbinfo = SHMEM_SB(inode->i_sb);
1280 if (sbinfo->max_blocks)
1281 percpu_counter_add(&sbinfo->used_blocks, -1);
1283 shmem_unacct_blocks(info->flags, 1);
1285 if (swap.val && error != -EINVAL &&
1286 !shmem_confirm_swap(mapping, index, swap))
1291 page_cache_release(page);
1293 if (error == -ENOSPC && !once++) {
1294 info = SHMEM_I(inode);
1295 spin_lock(&info->lock);
1296 shmem_recalc_inode(inode);
1297 spin_unlock(&info->lock);
1300 if (error == -EEXIST) /* from above or from radix_tree_insert */
1305 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1307 struct inode *inode = file_inode(vma->vm_file);
1309 int ret = VM_FAULT_LOCKED;
1312 * Trinity finds that probing a hole which tmpfs is punching can
1313 * prevent the hole-punch from ever completing: which in turn
1314 * locks writers out with its hold on i_mutex. So refrain from
1315 * faulting pages into the hole while it's being punched. Although
1316 * shmem_undo_range() does remove the additions, it may be unable to
1317 * keep up, as each new page needs its own unmap_mapping_range() call,
1318 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1320 * It does not matter if we sometimes reach this check just before the
1321 * hole-punch begins, so that one fault then races with the punch:
1322 * we just need to make racing faults a rare case.
1324 * The implementation below would be much simpler if we just used a
1325 * standard mutex or completion: but we cannot take i_mutex in fault,
1326 * and bloating every shmem inode for this unlikely case would be sad.
1328 if (unlikely(inode->i_private)) {
1329 struct shmem_falloc *shmem_falloc;
1331 spin_lock(&inode->i_lock);
1332 shmem_falloc = inode->i_private;
1334 shmem_falloc->waitq &&
1335 vmf->pgoff >= shmem_falloc->start &&
1336 vmf->pgoff < shmem_falloc->next) {
1337 wait_queue_head_t *shmem_falloc_waitq;
1338 DEFINE_WAIT(shmem_fault_wait);
1340 ret = VM_FAULT_NOPAGE;
1341 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1342 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1343 /* It's polite to up mmap_sem if we can */
1344 up_read(&vma->vm_mm->mmap_sem);
1345 ret = VM_FAULT_RETRY;
1348 shmem_falloc_waitq = shmem_falloc->waitq;
1349 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1350 TASK_UNINTERRUPTIBLE);
1351 spin_unlock(&inode->i_lock);
1355 * shmem_falloc_waitq points into the shmem_fallocate()
1356 * stack of the hole-punching task: shmem_falloc_waitq
1357 * is usually invalid by the time we reach here, but
1358 * finish_wait() does not dereference it in that case;
1359 * though i_lock needed lest racing with wake_up_all().
1361 spin_lock(&inode->i_lock);
1362 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1363 spin_unlock(&inode->i_lock);
1366 spin_unlock(&inode->i_lock);
1369 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1371 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1373 if (ret & VM_FAULT_MAJOR) {
1374 count_vm_event(PGMAJFAULT);
1375 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1381 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1383 struct inode *inode = file_inode(vma->vm_file);
1384 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1387 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1390 struct inode *inode = file_inode(vma->vm_file);
1393 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1394 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1398 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1400 struct inode *inode = file_inode(file);
1401 struct shmem_inode_info *info = SHMEM_I(inode);
1402 int retval = -ENOMEM;
1404 spin_lock(&info->lock);
1405 if (lock && !(info->flags & VM_LOCKED)) {
1406 if (!user_shm_lock(inode->i_size, user))
1408 info->flags |= VM_LOCKED;
1409 mapping_set_unevictable(file->f_mapping);
1411 if (!lock && (info->flags & VM_LOCKED) && user) {
1412 user_shm_unlock(inode->i_size, user);
1413 info->flags &= ~VM_LOCKED;
1414 mapping_clear_unevictable(file->f_mapping);
1419 spin_unlock(&info->lock);
1423 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1425 file_accessed(file);
1426 vma->vm_ops = &shmem_vm_ops;
1430 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1431 umode_t mode, dev_t dev, unsigned long flags)
1433 struct inode *inode;
1434 struct shmem_inode_info *info;
1435 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1437 if (shmem_reserve_inode(sb))
1440 inode = new_inode(sb);
1442 inode->i_ino = get_next_ino();
1443 inode_init_owner(inode, dir, mode);
1444 inode->i_blocks = 0;
1445 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1446 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1447 inode->i_generation = get_seconds();
1448 info = SHMEM_I(inode);
1449 memset(info, 0, (char *)inode - (char *)info);
1450 spin_lock_init(&info->lock);
1451 info->flags = flags & VM_NORESERVE;
1452 INIT_LIST_HEAD(&info->swaplist);
1453 simple_xattrs_init(&info->xattrs);
1454 cache_no_acl(inode);
1456 switch (mode & S_IFMT) {
1458 inode->i_op = &shmem_special_inode_operations;
1459 init_special_inode(inode, mode, dev);
1462 inode->i_mapping->a_ops = &shmem_aops;
1463 inode->i_op = &shmem_inode_operations;
1464 inode->i_fop = &shmem_file_operations;
1465 mpol_shared_policy_init(&info->policy,
1466 shmem_get_sbmpol(sbinfo));
1470 /* Some things misbehave if size == 0 on a directory */
1471 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1472 inode->i_op = &shmem_dir_inode_operations;
1473 inode->i_fop = &simple_dir_operations;
1477 * Must not load anything in the rbtree,
1478 * mpol_free_shared_policy will not be called.
1480 mpol_shared_policy_init(&info->policy, NULL);
1484 shmem_free_inode(sb);
1489 static const struct inode_operations shmem_symlink_inode_operations;
1490 static const struct inode_operations shmem_short_symlink_operations;
1492 #ifdef CONFIG_TMPFS_XATTR
1493 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1495 #define shmem_initxattrs NULL
1499 shmem_write_begin(struct file *file, struct address_space *mapping,
1500 loff_t pos, unsigned len, unsigned flags,
1501 struct page **pagep, void **fsdata)
1503 struct inode *inode = mapping->host;
1504 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1505 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1509 shmem_write_end(struct file *file, struct address_space *mapping,
1510 loff_t pos, unsigned len, unsigned copied,
1511 struct page *page, void *fsdata)
1513 struct inode *inode = mapping->host;
1515 if (pos + copied > inode->i_size)
1516 i_size_write(inode, pos + copied);
1518 if (!PageUptodate(page)) {
1519 if (copied < PAGE_CACHE_SIZE) {
1520 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1521 zero_user_segments(page, 0, from,
1522 from + copied, PAGE_CACHE_SIZE);
1524 SetPageUptodate(page);
1526 set_page_dirty(page);
1528 page_cache_release(page);
1533 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1535 struct inode *inode = file_inode(filp);
1536 struct address_space *mapping = inode->i_mapping;
1538 unsigned long offset;
1539 enum sgp_type sgp = SGP_READ;
1542 * Might this read be for a stacking filesystem? Then when reading
1543 * holes of a sparse file, we actually need to allocate those pages,
1544 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1546 if (segment_eq(get_fs(), KERNEL_DS))
1549 index = *ppos >> PAGE_CACHE_SHIFT;
1550 offset = *ppos & ~PAGE_CACHE_MASK;
1553 struct page *page = NULL;
1555 unsigned long nr, ret;
1556 loff_t i_size = i_size_read(inode);
1558 end_index = i_size >> PAGE_CACHE_SHIFT;
1559 if (index > end_index)
1561 if (index == end_index) {
1562 nr = i_size & ~PAGE_CACHE_MASK;
1567 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1569 if (desc->error == -EINVAL)
1577 * We must evaluate after, since reads (unlike writes)
1578 * are called without i_mutex protection against truncate
1580 nr = PAGE_CACHE_SIZE;
1581 i_size = i_size_read(inode);
1582 end_index = i_size >> PAGE_CACHE_SHIFT;
1583 if (index == end_index) {
1584 nr = i_size & ~PAGE_CACHE_MASK;
1587 page_cache_release(page);
1595 * If users can be writing to this page using arbitrary
1596 * virtual addresses, take care about potential aliasing
1597 * before reading the page on the kernel side.
1599 if (mapping_writably_mapped(mapping))
1600 flush_dcache_page(page);
1602 * Mark the page accessed if we read the beginning.
1605 mark_page_accessed(page);
1607 page = ZERO_PAGE(0);
1608 page_cache_get(page);
1612 * Ok, we have the page, and it's up-to-date, so
1613 * now we can copy it to user space...
1615 * The actor routine returns how many bytes were actually used..
1616 * NOTE! This may not be the same as how much of a user buffer
1617 * we filled up (we may be padding etc), so we can only update
1618 * "pos" here (the actor routine has to update the user buffer
1619 * pointers and the remaining count).
1621 ret = actor(desc, page, offset, nr);
1623 index += offset >> PAGE_CACHE_SHIFT;
1624 offset &= ~PAGE_CACHE_MASK;
1626 page_cache_release(page);
1627 if (ret != nr || !desc->count)
1633 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1634 file_accessed(filp);
1637 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1638 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1640 struct file *filp = iocb->ki_filp;
1644 loff_t *ppos = &iocb->ki_pos;
1646 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1650 for (seg = 0; seg < nr_segs; seg++) {
1651 read_descriptor_t desc;
1654 desc.arg.buf = iov[seg].iov_base;
1655 desc.count = iov[seg].iov_len;
1656 if (desc.count == 0)
1659 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1660 retval += desc.written;
1662 retval = retval ?: desc.error;
1671 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1672 struct pipe_inode_info *pipe, size_t len,
1675 struct address_space *mapping = in->f_mapping;
1676 struct inode *inode = mapping->host;
1677 unsigned int loff, nr_pages, req_pages;
1678 struct page *pages[PIPE_DEF_BUFFERS];
1679 struct partial_page partial[PIPE_DEF_BUFFERS];
1681 pgoff_t index, end_index;
1684 struct splice_pipe_desc spd = {
1687 .nr_pages_max = PIPE_DEF_BUFFERS,
1689 .ops = &page_cache_pipe_buf_ops,
1690 .spd_release = spd_release_page,
1693 isize = i_size_read(inode);
1694 if (unlikely(*ppos >= isize))
1697 left = isize - *ppos;
1698 if (unlikely(left < len))
1701 if (splice_grow_spd(pipe, &spd))
1704 index = *ppos >> PAGE_CACHE_SHIFT;
1705 loff = *ppos & ~PAGE_CACHE_MASK;
1706 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1707 nr_pages = min(req_pages, pipe->buffers);
1709 spd.nr_pages = find_get_pages_contig(mapping, index,
1710 nr_pages, spd.pages);
1711 index += spd.nr_pages;
1714 while (spd.nr_pages < nr_pages) {
1715 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1719 spd.pages[spd.nr_pages++] = page;
1723 index = *ppos >> PAGE_CACHE_SHIFT;
1724 nr_pages = spd.nr_pages;
1727 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1728 unsigned int this_len;
1733 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1734 page = spd.pages[page_nr];
1736 if (!PageUptodate(page) || page->mapping != mapping) {
1737 error = shmem_getpage(inode, index, &page,
1742 page_cache_release(spd.pages[page_nr]);
1743 spd.pages[page_nr] = page;
1746 isize = i_size_read(inode);
1747 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1748 if (unlikely(!isize || index > end_index))
1751 if (end_index == index) {
1754 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1758 this_len = min(this_len, plen - loff);
1762 spd.partial[page_nr].offset = loff;
1763 spd.partial[page_nr].len = this_len;
1770 while (page_nr < nr_pages)
1771 page_cache_release(spd.pages[page_nr++]);
1774 error = splice_to_pipe(pipe, &spd);
1776 splice_shrink_spd(&spd);
1786 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1788 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1789 pgoff_t index, pgoff_t end, int whence)
1792 struct pagevec pvec;
1793 pgoff_t indices[PAGEVEC_SIZE];
1797 pagevec_init(&pvec, 0);
1798 pvec.nr = 1; /* start small: we may be there already */
1800 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
1801 pvec.nr, pvec.pages, indices);
1803 if (whence == SEEK_DATA)
1807 for (i = 0; i < pvec.nr; i++, index++) {
1808 if (index < indices[i]) {
1809 if (whence == SEEK_HOLE) {
1815 page = pvec.pages[i];
1816 if (page && !radix_tree_exceptional_entry(page)) {
1817 if (!PageUptodate(page))
1821 (page && whence == SEEK_DATA) ||
1822 (!page && whence == SEEK_HOLE)) {
1827 shmem_deswap_pagevec(&pvec);
1828 pagevec_release(&pvec);
1829 pvec.nr = PAGEVEC_SIZE;
1835 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
1837 struct address_space *mapping = file->f_mapping;
1838 struct inode *inode = mapping->host;
1842 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1843 return generic_file_llseek_size(file, offset, whence,
1844 MAX_LFS_FILESIZE, i_size_read(inode));
1845 mutex_lock(&inode->i_mutex);
1846 /* We're holding i_mutex so we can access i_size directly */
1850 else if (offset >= inode->i_size)
1853 start = offset >> PAGE_CACHE_SHIFT;
1854 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1855 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
1856 new_offset <<= PAGE_CACHE_SHIFT;
1857 if (new_offset > offset) {
1858 if (new_offset < inode->i_size)
1859 offset = new_offset;
1860 else if (whence == SEEK_DATA)
1863 offset = inode->i_size;
1867 if (offset >= 0 && offset != file->f_pos) {
1868 file->f_pos = offset;
1869 file->f_version = 0;
1871 mutex_unlock(&inode->i_mutex);
1875 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1878 struct inode *inode = file_inode(file);
1879 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1880 struct shmem_falloc shmem_falloc;
1881 pgoff_t start, index, end;
1884 mutex_lock(&inode->i_mutex);
1886 if (mode & FALLOC_FL_PUNCH_HOLE) {
1887 struct address_space *mapping = file->f_mapping;
1888 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1889 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1890 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
1892 shmem_falloc.waitq = &shmem_falloc_waitq;
1893 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
1894 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
1895 spin_lock(&inode->i_lock);
1896 inode->i_private = &shmem_falloc;
1897 spin_unlock(&inode->i_lock);
1899 if ((u64)unmap_end > (u64)unmap_start)
1900 unmap_mapping_range(mapping, unmap_start,
1901 1 + unmap_end - unmap_start, 0);
1902 shmem_truncate_range(inode, offset, offset + len - 1);
1903 /* No need to unmap again: hole-punching leaves COWed pages */
1905 spin_lock(&inode->i_lock);
1906 inode->i_private = NULL;
1907 wake_up_all(&shmem_falloc_waitq);
1908 spin_unlock(&inode->i_lock);
1913 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1914 error = inode_newsize_ok(inode, offset + len);
1918 start = offset >> PAGE_CACHE_SHIFT;
1919 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1920 /* Try to avoid a swapstorm if len is impossible to satisfy */
1921 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1926 shmem_falloc.waitq = NULL;
1927 shmem_falloc.start = start;
1928 shmem_falloc.next = start;
1929 shmem_falloc.nr_falloced = 0;
1930 shmem_falloc.nr_unswapped = 0;
1931 spin_lock(&inode->i_lock);
1932 inode->i_private = &shmem_falloc;
1933 spin_unlock(&inode->i_lock);
1935 for (index = start; index < end; index++) {
1939 * Good, the fallocate(2) manpage permits EINTR: we may have
1940 * been interrupted because we are using up too much memory.
1942 if (signal_pending(current))
1944 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1947 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1950 /* Remove the !PageUptodate pages we added */
1951 shmem_undo_range(inode,
1952 (loff_t)start << PAGE_CACHE_SHIFT,
1953 (loff_t)index << PAGE_CACHE_SHIFT, true);
1958 * Inform shmem_writepage() how far we have reached.
1959 * No need for lock or barrier: we have the page lock.
1961 shmem_falloc.next++;
1962 if (!PageUptodate(page))
1963 shmem_falloc.nr_falloced++;
1966 * If !PageUptodate, leave it that way so that freeable pages
1967 * can be recognized if we need to rollback on error later.
1968 * But set_page_dirty so that memory pressure will swap rather
1969 * than free the pages we are allocating (and SGP_CACHE pages
1970 * might still be clean: we now need to mark those dirty too).
1972 set_page_dirty(page);
1974 page_cache_release(page);
1978 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1979 i_size_write(inode, offset + len);
1980 inode->i_ctime = CURRENT_TIME;
1982 spin_lock(&inode->i_lock);
1983 inode->i_private = NULL;
1984 spin_unlock(&inode->i_lock);
1986 mutex_unlock(&inode->i_mutex);
1990 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1992 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1994 buf->f_type = TMPFS_MAGIC;
1995 buf->f_bsize = PAGE_CACHE_SIZE;
1996 buf->f_namelen = NAME_MAX;
1997 if (sbinfo->max_blocks) {
1998 buf->f_blocks = sbinfo->max_blocks;
2000 buf->f_bfree = sbinfo->max_blocks -
2001 percpu_counter_sum(&sbinfo->used_blocks);
2003 if (sbinfo->max_inodes) {
2004 buf->f_files = sbinfo->max_inodes;
2005 buf->f_ffree = sbinfo->free_inodes;
2007 /* else leave those fields 0 like simple_statfs */
2012 * File creation. Allocate an inode, and we're done..
2015 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2017 struct inode *inode;
2018 int error = -ENOSPC;
2020 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2022 error = security_inode_init_security(inode, dir,
2024 shmem_initxattrs, NULL);
2026 if (error != -EOPNOTSUPP) {
2031 #ifdef CONFIG_TMPFS_POSIX_ACL
2032 error = generic_acl_init(inode, dir);
2040 dir->i_size += BOGO_DIRENT_SIZE;
2041 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2042 d_instantiate(dentry, inode);
2043 dget(dentry); /* Extra count - pin the dentry in core */
2048 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2052 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2058 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2061 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2067 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2069 struct inode *inode = old_dentry->d_inode;
2073 * No ordinary (disk based) filesystem counts links as inodes;
2074 * but each new link needs a new dentry, pinning lowmem, and
2075 * tmpfs dentries cannot be pruned until they are unlinked.
2077 ret = shmem_reserve_inode(inode->i_sb);
2081 dir->i_size += BOGO_DIRENT_SIZE;
2082 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2084 ihold(inode); /* New dentry reference */
2085 dget(dentry); /* Extra pinning count for the created dentry */
2086 d_instantiate(dentry, inode);
2091 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2093 struct inode *inode = dentry->d_inode;
2095 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2096 shmem_free_inode(inode->i_sb);
2098 dir->i_size -= BOGO_DIRENT_SIZE;
2099 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2101 dput(dentry); /* Undo the count from "create" - this does all the work */
2105 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2107 if (!simple_empty(dentry))
2110 drop_nlink(dentry->d_inode);
2112 return shmem_unlink(dir, dentry);
2116 * The VFS layer already does all the dentry stuff for rename,
2117 * we just have to decrement the usage count for the target if
2118 * it exists so that the VFS layer correctly free's it when it
2121 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2123 struct inode *inode = old_dentry->d_inode;
2124 int they_are_dirs = S_ISDIR(inode->i_mode);
2126 if (!simple_empty(new_dentry))
2129 if (new_dentry->d_inode) {
2130 (void) shmem_unlink(new_dir, new_dentry);
2131 if (they_are_dirs) {
2132 drop_nlink(new_dentry->d_inode);
2133 drop_nlink(old_dir);
2135 } else if (they_are_dirs) {
2136 drop_nlink(old_dir);
2140 old_dir->i_size -= BOGO_DIRENT_SIZE;
2141 new_dir->i_size += BOGO_DIRENT_SIZE;
2142 old_dir->i_ctime = old_dir->i_mtime =
2143 new_dir->i_ctime = new_dir->i_mtime =
2144 inode->i_ctime = CURRENT_TIME;
2148 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2152 struct inode *inode;
2155 struct shmem_inode_info *info;
2157 len = strlen(symname) + 1;
2158 if (len > PAGE_CACHE_SIZE)
2159 return -ENAMETOOLONG;
2161 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2165 error = security_inode_init_security(inode, dir, &dentry->d_name,
2166 shmem_initxattrs, NULL);
2168 if (error != -EOPNOTSUPP) {
2175 info = SHMEM_I(inode);
2176 inode->i_size = len-1;
2177 if (len <= SHORT_SYMLINK_LEN) {
2178 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2179 if (!info->symlink) {
2183 inode->i_op = &shmem_short_symlink_operations;
2185 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2190 inode->i_mapping->a_ops = &shmem_aops;
2191 inode->i_op = &shmem_symlink_inode_operations;
2192 kaddr = kmap_atomic(page);
2193 memcpy(kaddr, symname, len);
2194 kunmap_atomic(kaddr);
2195 SetPageUptodate(page);
2196 set_page_dirty(page);
2198 page_cache_release(page);
2200 dir->i_size += BOGO_DIRENT_SIZE;
2201 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2202 d_instantiate(dentry, inode);
2207 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2209 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2213 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2215 struct page *page = NULL;
2216 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2217 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2223 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2225 if (!IS_ERR(nd_get_link(nd))) {
2226 struct page *page = cookie;
2228 mark_page_accessed(page);
2229 page_cache_release(page);
2233 #ifdef CONFIG_TMPFS_XATTR
2235 * Superblocks without xattr inode operations may get some security.* xattr
2236 * support from the LSM "for free". As soon as we have any other xattrs
2237 * like ACLs, we also need to implement the security.* handlers at
2238 * filesystem level, though.
2242 * Callback for security_inode_init_security() for acquiring xattrs.
2244 static int shmem_initxattrs(struct inode *inode,
2245 const struct xattr *xattr_array,
2248 struct shmem_inode_info *info = SHMEM_I(inode);
2249 const struct xattr *xattr;
2250 struct simple_xattr *new_xattr;
2253 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2254 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
2258 len = strlen(xattr->name) + 1;
2259 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2261 if (!new_xattr->name) {
2266 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2267 XATTR_SECURITY_PREFIX_LEN);
2268 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2271 simple_xattr_list_add(&info->xattrs, new_xattr);
2277 static const struct xattr_handler *shmem_xattr_handlers[] = {
2278 #ifdef CONFIG_TMPFS_POSIX_ACL
2279 &generic_acl_access_handler,
2280 &generic_acl_default_handler,
2285 static int shmem_xattr_validate(const char *name)
2287 struct { const char *prefix; size_t len; } arr[] = {
2288 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2289 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2293 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2294 size_t preflen = arr[i].len;
2295 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2304 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2305 void *buffer, size_t size)
2307 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2311 * If this is a request for a synthetic attribute in the system.*
2312 * namespace use the generic infrastructure to resolve a handler
2313 * for it via sb->s_xattr.
2315 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2316 return generic_getxattr(dentry, name, buffer, size);
2318 err = shmem_xattr_validate(name);
2322 return simple_xattr_get(&info->xattrs, name, buffer, size);
2325 static int shmem_setxattr(struct dentry *dentry, const char *name,
2326 const void *value, size_t size, int flags)
2328 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2332 * If this is a request for a synthetic attribute in the system.*
2333 * namespace use the generic infrastructure to resolve a handler
2334 * for it via sb->s_xattr.
2336 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2337 return generic_setxattr(dentry, name, value, size, flags);
2339 err = shmem_xattr_validate(name);
2343 return simple_xattr_set(&info->xattrs, name, value, size, flags);
2346 static int shmem_removexattr(struct dentry *dentry, const char *name)
2348 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2352 * If this is a request for a synthetic attribute in the system.*
2353 * namespace use the generic infrastructure to resolve a handler
2354 * for it via sb->s_xattr.
2356 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2357 return generic_removexattr(dentry, name);
2359 err = shmem_xattr_validate(name);
2363 return simple_xattr_remove(&info->xattrs, name);
2366 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2368 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2369 return simple_xattr_list(&info->xattrs, buffer, size);
2371 #endif /* CONFIG_TMPFS_XATTR */
2373 static const struct inode_operations shmem_short_symlink_operations = {
2374 .readlink = generic_readlink,
2375 .follow_link = shmem_follow_short_symlink,
2376 #ifdef CONFIG_TMPFS_XATTR
2377 .setxattr = shmem_setxattr,
2378 .getxattr = shmem_getxattr,
2379 .listxattr = shmem_listxattr,
2380 .removexattr = shmem_removexattr,
2384 static const struct inode_operations shmem_symlink_inode_operations = {
2385 .readlink = generic_readlink,
2386 .follow_link = shmem_follow_link,
2387 .put_link = shmem_put_link,
2388 #ifdef CONFIG_TMPFS_XATTR
2389 .setxattr = shmem_setxattr,
2390 .getxattr = shmem_getxattr,
2391 .listxattr = shmem_listxattr,
2392 .removexattr = shmem_removexattr,
2396 static struct dentry *shmem_get_parent(struct dentry *child)
2398 return ERR_PTR(-ESTALE);
2401 static int shmem_match(struct inode *ino, void *vfh)
2405 inum = (inum << 32) | fh[1];
2406 return ino->i_ino == inum && fh[0] == ino->i_generation;
2409 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2410 struct fid *fid, int fh_len, int fh_type)
2412 struct inode *inode;
2413 struct dentry *dentry = NULL;
2420 inum = (inum << 32) | fid->raw[1];
2422 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2423 shmem_match, fid->raw);
2425 dentry = d_find_alias(inode);
2432 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2433 struct inode *parent)
2437 return FILEID_INVALID;
2440 if (inode_unhashed(inode)) {
2441 /* Unfortunately insert_inode_hash is not idempotent,
2442 * so as we hash inodes here rather than at creation
2443 * time, we need a lock to ensure we only try
2446 static DEFINE_SPINLOCK(lock);
2448 if (inode_unhashed(inode))
2449 __insert_inode_hash(inode,
2450 inode->i_ino + inode->i_generation);
2454 fh[0] = inode->i_generation;
2455 fh[1] = inode->i_ino;
2456 fh[2] = ((__u64)inode->i_ino) >> 32;
2462 static const struct export_operations shmem_export_ops = {
2463 .get_parent = shmem_get_parent,
2464 .encode_fh = shmem_encode_fh,
2465 .fh_to_dentry = shmem_fh_to_dentry,
2468 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2471 char *this_char, *value, *rest;
2472 struct mempolicy *mpol = NULL;
2476 while (options != NULL) {
2477 this_char = options;
2480 * NUL-terminate this option: unfortunately,
2481 * mount options form a comma-separated list,
2482 * but mpol's nodelist may also contain commas.
2484 options = strchr(options, ',');
2485 if (options == NULL)
2488 if (!isdigit(*options)) {
2495 if ((value = strchr(this_char,'=')) != NULL) {
2499 "tmpfs: No value for mount option '%s'\n",
2504 if (!strcmp(this_char,"size")) {
2505 unsigned long long size;
2506 size = memparse(value,&rest);
2508 size <<= PAGE_SHIFT;
2509 size *= totalram_pages;
2515 sbinfo->max_blocks =
2516 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2517 } else if (!strcmp(this_char,"nr_blocks")) {
2518 sbinfo->max_blocks = memparse(value, &rest);
2521 } else if (!strcmp(this_char,"nr_inodes")) {
2522 sbinfo->max_inodes = memparse(value, &rest);
2525 } else if (!strcmp(this_char,"mode")) {
2528 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2531 } else if (!strcmp(this_char,"uid")) {
2534 uid = simple_strtoul(value, &rest, 0);
2537 sbinfo->uid = make_kuid(current_user_ns(), uid);
2538 if (!uid_valid(sbinfo->uid))
2540 } else if (!strcmp(this_char,"gid")) {
2543 gid = simple_strtoul(value, &rest, 0);
2546 sbinfo->gid = make_kgid(current_user_ns(), gid);
2547 if (!gid_valid(sbinfo->gid))
2549 } else if (!strcmp(this_char,"mpol")) {
2552 if (mpol_parse_str(value, &mpol))
2555 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2560 sbinfo->mpol = mpol;
2564 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2572 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2574 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2575 struct shmem_sb_info config = *sbinfo;
2576 unsigned long inodes;
2577 int error = -EINVAL;
2580 if (shmem_parse_options(data, &config, true))
2583 spin_lock(&sbinfo->stat_lock);
2584 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2585 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2587 if (config.max_inodes < inodes)
2590 * Those tests disallow limited->unlimited while any are in use;
2591 * but we must separately disallow unlimited->limited, because
2592 * in that case we have no record of how much is already in use.
2594 if (config.max_blocks && !sbinfo->max_blocks)
2596 if (config.max_inodes && !sbinfo->max_inodes)
2600 sbinfo->max_blocks = config.max_blocks;
2601 sbinfo->max_inodes = config.max_inodes;
2602 sbinfo->free_inodes = config.max_inodes - inodes;
2605 * Preserve previous mempolicy unless mpol remount option was specified.
2608 mpol_put(sbinfo->mpol);
2609 sbinfo->mpol = config.mpol; /* transfers initial ref */
2612 spin_unlock(&sbinfo->stat_lock);
2616 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2618 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2620 if (sbinfo->max_blocks != shmem_default_max_blocks())
2621 seq_printf(seq, ",size=%luk",
2622 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2623 if (sbinfo->max_inodes != shmem_default_max_inodes())
2624 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2625 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2626 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2627 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2628 seq_printf(seq, ",uid=%u",
2629 from_kuid_munged(&init_user_ns, sbinfo->uid));
2630 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2631 seq_printf(seq, ",gid=%u",
2632 from_kgid_munged(&init_user_ns, sbinfo->gid));
2633 shmem_show_mpol(seq, sbinfo->mpol);
2636 #endif /* CONFIG_TMPFS */
2638 static void shmem_put_super(struct super_block *sb)
2640 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2642 percpu_counter_destroy(&sbinfo->used_blocks);
2643 mpol_put(sbinfo->mpol);
2645 sb->s_fs_info = NULL;
2648 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2650 struct inode *inode;
2651 struct shmem_sb_info *sbinfo;
2654 /* Round up to L1_CACHE_BYTES to resist false sharing */
2655 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2656 L1_CACHE_BYTES), GFP_KERNEL);
2660 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2661 sbinfo->uid = current_fsuid();
2662 sbinfo->gid = current_fsgid();
2663 sb->s_fs_info = sbinfo;
2667 * Per default we only allow half of the physical ram per
2668 * tmpfs instance, limiting inodes to one per page of lowmem;
2669 * but the internal instance is left unlimited.
2671 if (!(sb->s_flags & MS_NOUSER)) {
2672 sbinfo->max_blocks = shmem_default_max_blocks();
2673 sbinfo->max_inodes = shmem_default_max_inodes();
2674 if (shmem_parse_options(data, sbinfo, false)) {
2679 sb->s_export_op = &shmem_export_ops;
2680 sb->s_flags |= MS_NOSEC;
2682 sb->s_flags |= MS_NOUSER;
2685 spin_lock_init(&sbinfo->stat_lock);
2686 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2688 sbinfo->free_inodes = sbinfo->max_inodes;
2690 sb->s_maxbytes = MAX_LFS_FILESIZE;
2691 sb->s_blocksize = PAGE_CACHE_SIZE;
2692 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2693 sb->s_magic = TMPFS_MAGIC;
2694 sb->s_op = &shmem_ops;
2695 sb->s_time_gran = 1;
2696 #ifdef CONFIG_TMPFS_XATTR
2697 sb->s_xattr = shmem_xattr_handlers;
2699 #ifdef CONFIG_TMPFS_POSIX_ACL
2700 sb->s_flags |= MS_POSIXACL;
2703 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2706 inode->i_uid = sbinfo->uid;
2707 inode->i_gid = sbinfo->gid;
2708 sb->s_root = d_make_root(inode);
2714 shmem_put_super(sb);
2718 static struct kmem_cache *shmem_inode_cachep;
2720 static struct inode *shmem_alloc_inode(struct super_block *sb)
2722 struct shmem_inode_info *info;
2723 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2726 return &info->vfs_inode;
2729 static void shmem_destroy_callback(struct rcu_head *head)
2731 struct inode *inode = container_of(head, struct inode, i_rcu);
2732 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2735 static void shmem_destroy_inode(struct inode *inode)
2737 if (S_ISREG(inode->i_mode))
2738 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2739 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2742 static void shmem_init_inode(void *foo)
2744 struct shmem_inode_info *info = foo;
2745 inode_init_once(&info->vfs_inode);
2748 static int shmem_init_inodecache(void)
2750 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2751 sizeof(struct shmem_inode_info),
2752 0, SLAB_PANIC, shmem_init_inode);
2756 static void shmem_destroy_inodecache(void)
2758 kmem_cache_destroy(shmem_inode_cachep);
2761 static const struct address_space_operations shmem_aops = {
2762 .writepage = shmem_writepage,
2763 .set_page_dirty = __set_page_dirty_no_writeback,
2765 .write_begin = shmem_write_begin,
2766 .write_end = shmem_write_end,
2768 .migratepage = migrate_page,
2769 .error_remove_page = generic_error_remove_page,
2772 static const struct file_operations shmem_file_operations = {
2775 .llseek = shmem_file_llseek,
2776 .read = do_sync_read,
2777 .write = do_sync_write,
2778 .aio_read = shmem_file_aio_read,
2779 .aio_write = generic_file_aio_write,
2780 .fsync = noop_fsync,
2781 .splice_read = shmem_file_splice_read,
2782 .splice_write = generic_file_splice_write,
2783 .fallocate = shmem_fallocate,
2787 static const struct inode_operations shmem_inode_operations = {
2788 .setattr = shmem_setattr,
2789 #ifdef CONFIG_TMPFS_XATTR
2790 .setxattr = shmem_setxattr,
2791 .getxattr = shmem_getxattr,
2792 .listxattr = shmem_listxattr,
2793 .removexattr = shmem_removexattr,
2797 static const struct inode_operations shmem_dir_inode_operations = {
2799 .create = shmem_create,
2800 .lookup = simple_lookup,
2802 .unlink = shmem_unlink,
2803 .symlink = shmem_symlink,
2804 .mkdir = shmem_mkdir,
2805 .rmdir = shmem_rmdir,
2806 .mknod = shmem_mknod,
2807 .rename = shmem_rename,
2809 #ifdef CONFIG_TMPFS_XATTR
2810 .setxattr = shmem_setxattr,
2811 .getxattr = shmem_getxattr,
2812 .listxattr = shmem_listxattr,
2813 .removexattr = shmem_removexattr,
2815 #ifdef CONFIG_TMPFS_POSIX_ACL
2816 .setattr = shmem_setattr,
2820 static const struct inode_operations shmem_special_inode_operations = {
2821 #ifdef CONFIG_TMPFS_XATTR
2822 .setxattr = shmem_setxattr,
2823 .getxattr = shmem_getxattr,
2824 .listxattr = shmem_listxattr,
2825 .removexattr = shmem_removexattr,
2827 #ifdef CONFIG_TMPFS_POSIX_ACL
2828 .setattr = shmem_setattr,
2832 static const struct super_operations shmem_ops = {
2833 .alloc_inode = shmem_alloc_inode,
2834 .destroy_inode = shmem_destroy_inode,
2836 .statfs = shmem_statfs,
2837 .remount_fs = shmem_remount_fs,
2838 .show_options = shmem_show_options,
2840 .evict_inode = shmem_evict_inode,
2841 .drop_inode = generic_delete_inode,
2842 .put_super = shmem_put_super,
2845 static const struct vm_operations_struct shmem_vm_ops = {
2846 .fault = shmem_fault,
2848 .set_policy = shmem_set_policy,
2849 .get_policy = shmem_get_policy,
2851 .remap_pages = generic_file_remap_pages,
2854 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2855 int flags, const char *dev_name, void *data)
2857 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2860 static struct file_system_type shmem_fs_type = {
2861 .owner = THIS_MODULE,
2863 .mount = shmem_mount,
2864 .kill_sb = kill_litter_super,
2865 .fs_flags = FS_USERNS_MOUNT,
2868 int __init shmem_init(void)
2872 error = bdi_init(&shmem_backing_dev_info);
2876 error = shmem_init_inodecache();
2880 error = register_filesystem(&shmem_fs_type);
2882 printk(KERN_ERR "Could not register tmpfs\n");
2886 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2887 shmem_fs_type.name, NULL);
2888 if (IS_ERR(shm_mnt)) {
2889 error = PTR_ERR(shm_mnt);
2890 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2896 unregister_filesystem(&shmem_fs_type);
2898 shmem_destroy_inodecache();
2900 bdi_destroy(&shmem_backing_dev_info);
2902 shm_mnt = ERR_PTR(error);
2906 #else /* !CONFIG_SHMEM */
2909 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2911 * This is intended for small system where the benefits of the full
2912 * shmem code (swap-backed and resource-limited) are outweighed by
2913 * their complexity. On systems without swap this code should be
2914 * effectively equivalent, but much lighter weight.
2917 static struct file_system_type shmem_fs_type = {
2919 .mount = ramfs_mount,
2920 .kill_sb = kill_litter_super,
2921 .fs_flags = FS_USERNS_MOUNT,
2924 int __init shmem_init(void)
2926 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2928 shm_mnt = kern_mount(&shmem_fs_type);
2929 BUG_ON(IS_ERR(shm_mnt));
2934 int shmem_unuse(swp_entry_t swap, struct page *page)
2939 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2944 void shmem_unlock_mapping(struct address_space *mapping)
2948 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2950 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2952 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2954 #define shmem_vm_ops generic_file_vm_ops
2955 #define shmem_file_operations ramfs_file_operations
2956 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2957 #define shmem_acct_size(flags, size) 0
2958 #define shmem_unacct_size(flags, size) do {} while (0)
2960 #endif /* CONFIG_SHMEM */
2964 static struct dentry_operations anon_ops = {
2965 .d_dname = simple_dname
2969 * shmem_file_setup - get an unlinked file living in tmpfs
2970 * @name: name for dentry (to be seen in /proc/<pid>/maps
2971 * @size: size to be set for the file
2972 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2974 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2977 struct inode *inode;
2979 struct super_block *sb;
2982 if (IS_ERR(shm_mnt))
2983 return ERR_CAST(shm_mnt);
2985 if (size < 0 || size > MAX_LFS_FILESIZE)
2986 return ERR_PTR(-EINVAL);
2988 if (shmem_acct_size(flags, size))
2989 return ERR_PTR(-ENOMEM);
2991 res = ERR_PTR(-ENOMEM);
2993 this.len = strlen(name);
2994 this.hash = 0; /* will go */
2995 sb = shm_mnt->mnt_sb;
2996 path.dentry = d_alloc_pseudo(sb, &this);
2999 d_set_d_op(path.dentry, &anon_ops);
3000 path.mnt = mntget(shm_mnt);
3002 res = ERR_PTR(-ENOSPC);
3003 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
3007 d_instantiate(path.dentry, inode);
3008 inode->i_size = size;
3009 clear_nlink(inode); /* It is unlinked */
3010 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
3014 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
3015 &shmem_file_operations);
3024 shmem_unacct_size(flags, size);
3027 EXPORT_SYMBOL_GPL(shmem_file_setup);
3030 * shmem_zero_setup - setup a shared anonymous mapping
3031 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3033 int shmem_zero_setup(struct vm_area_struct *vma)
3036 loff_t size = vma->vm_end - vma->vm_start;
3038 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3040 return PTR_ERR(file);
3044 vma->vm_file = file;
3045 vma->vm_ops = &shmem_vm_ops;
3050 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3051 * @mapping: the page's address_space
3052 * @index: the page index
3053 * @gfp: the page allocator flags to use if allocating
3055 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3056 * with any new page allocations done using the specified allocation flags.
3057 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3058 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3059 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3061 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3062 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3064 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3065 pgoff_t index, gfp_t gfp)
3068 struct inode *inode = mapping->host;
3072 BUG_ON(mapping->a_ops != &shmem_aops);
3073 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3075 page = ERR_PTR(error);
3081 * The tiny !SHMEM case uses ramfs without swap
3083 return read_cache_page_gfp(mapping, index, gfp);
3086 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);