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/pagemap.h>
29 #include <linux/file.h>
31 #include <linux/export.h>
32 #include <linux/swap.h>
34 static struct vfsmount *shm_mnt;
38 * This virtual memory filesystem is heavily based on the ramfs. It
39 * extends ramfs by the ability to use swap and honor resource limits
40 * which makes it a completely usable filesystem.
43 #include <linux/xattr.h>
44 #include <linux/exportfs.h>
45 #include <linux/posix_acl.h>
46 #include <linux/generic_acl.h>
47 #include <linux/mman.h>
48 #include <linux/string.h>
49 #include <linux/slab.h>
50 #include <linux/backing-dev.h>
51 #include <linux/shmem_fs.h>
52 #include <linux/writeback.h>
53 #include <linux/blkdev.h>
54 #include <linux/pagevec.h>
55 #include <linux/percpu_counter.h>
56 #include <linux/falloc.h>
57 #include <linux/splice.h>
58 #include <linux/security.h>
59 #include <linux/swapops.h>
60 #include <linux/mempolicy.h>
61 #include <linux/namei.h>
62 #include <linux/ctype.h>
63 #include <linux/migrate.h>
64 #include <linux/highmem.h>
65 #include <linux/seq_file.h>
66 #include <linux/magic.h>
68 #include <asm/uaccess.h>
69 #include <asm/pgtable.h>
71 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
72 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
74 /* Pretend that each entry is of this size in directory's i_size */
75 #define BOGO_DIRENT_SIZE 20
77 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
78 #define SHORT_SYMLINK_LEN 128
81 struct list_head list; /* anchored by shmem_inode_info->xattr_list */
82 char *name; /* xattr name */
88 * shmem_fallocate and shmem_writepage communicate via inode->i_private
89 * (with i_mutex making sure that it has only one user at a time):
90 * we would prefer not to enlarge the shmem inode just for that.
93 pgoff_t start; /* start of range currently being fallocated */
94 pgoff_t next; /* the next page offset to be fallocated */
95 pgoff_t nr_falloced; /* how many new pages have been fallocated */
96 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
99 /* Flag allocation requirements to shmem_getpage */
101 SGP_READ, /* don't exceed i_size, don't allocate page */
102 SGP_CACHE, /* don't exceed i_size, may allocate page */
103 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
104 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
105 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
109 static unsigned long shmem_default_max_blocks(void)
111 return totalram_pages / 2;
114 static unsigned long shmem_default_max_inodes(void)
116 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
120 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
121 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
122 struct shmem_inode_info *info, pgoff_t index);
123 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
124 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
126 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
127 struct page **pagep, enum sgp_type sgp, int *fault_type)
129 return shmem_getpage_gfp(inode, index, pagep, sgp,
130 mapping_gfp_mask(inode->i_mapping), fault_type);
133 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
135 return sb->s_fs_info;
139 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
140 * for shared memory and for shared anonymous (/dev/zero) mappings
141 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
142 * consistent with the pre-accounting of private mappings ...
144 static inline int shmem_acct_size(unsigned long flags, loff_t size)
146 return (flags & VM_NORESERVE) ?
147 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
150 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
152 if (!(flags & VM_NORESERVE))
153 vm_unacct_memory(VM_ACCT(size));
157 * ... whereas tmpfs objects are accounted incrementally as
158 * pages are allocated, in order to allow huge sparse files.
159 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
160 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
162 static inline int shmem_acct_block(unsigned long flags)
164 return (flags & VM_NORESERVE) ?
165 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
168 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
170 if (flags & VM_NORESERVE)
171 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
174 static const struct super_operations shmem_ops;
175 static const struct address_space_operations shmem_aops;
176 static const struct file_operations shmem_file_operations;
177 static const struct inode_operations shmem_inode_operations;
178 static const struct inode_operations shmem_dir_inode_operations;
179 static const struct inode_operations shmem_special_inode_operations;
180 static const struct vm_operations_struct shmem_vm_ops;
182 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
183 .ra_pages = 0, /* No readahead */
184 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
187 static LIST_HEAD(shmem_swaplist);
188 static DEFINE_MUTEX(shmem_swaplist_mutex);
190 static int shmem_reserve_inode(struct super_block *sb)
192 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
193 if (sbinfo->max_inodes) {
194 spin_lock(&sbinfo->stat_lock);
195 if (!sbinfo->free_inodes) {
196 spin_unlock(&sbinfo->stat_lock);
199 sbinfo->free_inodes--;
200 spin_unlock(&sbinfo->stat_lock);
205 static void shmem_free_inode(struct super_block *sb)
207 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
208 if (sbinfo->max_inodes) {
209 spin_lock(&sbinfo->stat_lock);
210 sbinfo->free_inodes++;
211 spin_unlock(&sbinfo->stat_lock);
216 * shmem_recalc_inode - recalculate the block usage of an inode
217 * @inode: inode to recalc
219 * We have to calculate the free blocks since the mm can drop
220 * undirtied hole pages behind our back.
222 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
223 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
225 * It has to be called with the spinlock held.
227 static void shmem_recalc_inode(struct inode *inode)
229 struct shmem_inode_info *info = SHMEM_I(inode);
232 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
234 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
235 if (sbinfo->max_blocks)
236 percpu_counter_add(&sbinfo->used_blocks, -freed);
237 info->alloced -= freed;
238 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
239 shmem_unacct_blocks(info->flags, freed);
244 * Replace item expected in radix tree by a new item, while holding tree lock.
246 static int shmem_radix_tree_replace(struct address_space *mapping,
247 pgoff_t index, void *expected, void *replacement)
252 VM_BUG_ON(!expected);
253 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
255 item = radix_tree_deref_slot_protected(pslot,
256 &mapping->tree_lock);
257 if (item != expected)
260 radix_tree_replace_slot(pslot, replacement);
262 radix_tree_delete(&mapping->page_tree, index);
267 * Like add_to_page_cache_locked, but error if expected item has gone.
269 static int shmem_add_to_page_cache(struct page *page,
270 struct address_space *mapping,
271 pgoff_t index, gfp_t gfp, void *expected)
275 VM_BUG_ON(!PageLocked(page));
276 VM_BUG_ON(!PageSwapBacked(page));
279 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
281 page_cache_get(page);
282 page->mapping = mapping;
285 spin_lock_irq(&mapping->tree_lock);
287 error = radix_tree_insert(&mapping->page_tree,
290 error = shmem_radix_tree_replace(mapping, index,
294 __inc_zone_page_state(page, NR_FILE_PAGES);
295 __inc_zone_page_state(page, NR_SHMEM);
296 spin_unlock_irq(&mapping->tree_lock);
298 page->mapping = NULL;
299 spin_unlock_irq(&mapping->tree_lock);
300 page_cache_release(page);
303 radix_tree_preload_end();
306 mem_cgroup_uncharge_cache_page(page);
311 * Like delete_from_page_cache, but substitutes swap for page.
313 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
315 struct address_space *mapping = page->mapping;
318 spin_lock_irq(&mapping->tree_lock);
319 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
320 page->mapping = NULL;
322 __dec_zone_page_state(page, NR_FILE_PAGES);
323 __dec_zone_page_state(page, NR_SHMEM);
324 spin_unlock_irq(&mapping->tree_lock);
325 page_cache_release(page);
330 * Like find_get_pages, but collecting swap entries as well as pages.
332 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
333 pgoff_t start, unsigned int nr_pages,
334 struct page **pages, pgoff_t *indices)
338 unsigned int nr_found;
342 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
343 (void ***)pages, indices, start, nr_pages);
345 for (i = 0; i < nr_found; i++) {
348 page = radix_tree_deref_slot((void **)pages[i]);
351 if (radix_tree_exception(page)) {
352 if (radix_tree_deref_retry(page))
355 * Otherwise, we must be storing a swap entry
356 * here as an exceptional entry: so return it
357 * without attempting to raise page count.
361 if (!page_cache_get_speculative(page))
364 /* Has the page moved? */
365 if (unlikely(page != *((void **)pages[i]))) {
366 page_cache_release(page);
370 indices[ret] = indices[i];
374 if (unlikely(!ret && nr_found))
381 * Remove swap entry from radix tree, free the swap and its page cache.
383 static int shmem_free_swap(struct address_space *mapping,
384 pgoff_t index, void *radswap)
388 spin_lock_irq(&mapping->tree_lock);
389 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
390 spin_unlock_irq(&mapping->tree_lock);
392 free_swap_and_cache(radix_to_swp_entry(radswap));
397 * Pagevec may contain swap entries, so shuffle up pages before releasing.
399 static void shmem_deswap_pagevec(struct pagevec *pvec)
403 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
404 struct page *page = pvec->pages[i];
405 if (!radix_tree_exceptional_entry(page))
406 pvec->pages[j++] = page;
412 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
414 void shmem_unlock_mapping(struct address_space *mapping)
417 pgoff_t indices[PAGEVEC_SIZE];
420 pagevec_init(&pvec, 0);
422 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
424 while (!mapping_unevictable(mapping)) {
426 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
427 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
429 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
430 PAGEVEC_SIZE, pvec.pages, indices);
433 index = indices[pvec.nr - 1] + 1;
434 shmem_deswap_pagevec(&pvec);
435 check_move_unevictable_pages(pvec.pages, pvec.nr);
436 pagevec_release(&pvec);
442 * Remove range of pages and swap entries from radix tree, and free them.
443 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
445 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
448 struct address_space *mapping = inode->i_mapping;
449 struct shmem_inode_info *info = SHMEM_I(inode);
450 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
451 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
452 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
453 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
455 pgoff_t indices[PAGEVEC_SIZE];
456 long nr_swaps_freed = 0;
461 end = -1; /* unsigned, so actually very big */
463 pagevec_init(&pvec, 0);
465 while (index < end) {
466 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
467 min(end - index, (pgoff_t)PAGEVEC_SIZE),
468 pvec.pages, indices);
471 mem_cgroup_uncharge_start();
472 for (i = 0; i < pagevec_count(&pvec); i++) {
473 struct page *page = pvec.pages[i];
479 if (radix_tree_exceptional_entry(page)) {
482 nr_swaps_freed += !shmem_free_swap(mapping,
487 if (!trylock_page(page))
489 if (!unfalloc || !PageUptodate(page)) {
490 if (page->mapping == mapping) {
491 VM_BUG_ON(PageWriteback(page));
492 truncate_inode_page(mapping, page);
497 shmem_deswap_pagevec(&pvec);
498 pagevec_release(&pvec);
499 mem_cgroup_uncharge_end();
505 struct page *page = NULL;
506 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
508 unsigned int top = PAGE_CACHE_SIZE;
513 zero_user_segment(page, partial_start, top);
514 set_page_dirty(page);
516 page_cache_release(page);
520 struct page *page = NULL;
521 shmem_getpage(inode, end, &page, SGP_READ, NULL);
523 zero_user_segment(page, 0, partial_end);
524 set_page_dirty(page);
526 page_cache_release(page);
535 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
536 min(end - index, (pgoff_t)PAGEVEC_SIZE),
537 pvec.pages, indices);
539 if (index == start || unfalloc)
544 if ((index == start || unfalloc) && indices[0] >= end) {
545 shmem_deswap_pagevec(&pvec);
546 pagevec_release(&pvec);
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 nr_swaps_freed += !shmem_free_swap(mapping,
566 if (!unfalloc || !PageUptodate(page)) {
567 if (page->mapping == mapping) {
568 VM_BUG_ON(PageWriteback(page));
569 truncate_inode_page(mapping, page);
574 shmem_deswap_pagevec(&pvec);
575 pagevec_release(&pvec);
576 mem_cgroup_uncharge_end();
580 spin_lock(&info->lock);
581 info->swapped -= nr_swaps_freed;
582 shmem_recalc_inode(inode);
583 spin_unlock(&info->lock);
586 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
588 shmem_undo_range(inode, lstart, lend, false);
589 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
591 EXPORT_SYMBOL_GPL(shmem_truncate_range);
593 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
595 struct inode *inode = dentry->d_inode;
598 error = inode_change_ok(inode, attr);
602 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
603 loff_t oldsize = inode->i_size;
604 loff_t newsize = attr->ia_size;
606 if (newsize != oldsize) {
607 i_size_write(inode, newsize);
608 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
610 if (newsize < oldsize) {
611 loff_t holebegin = round_up(newsize, PAGE_SIZE);
612 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
613 shmem_truncate_range(inode, newsize, (loff_t)-1);
614 /* unmap again to remove racily COWed private pages */
615 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
619 setattr_copy(inode, attr);
620 #ifdef CONFIG_TMPFS_POSIX_ACL
621 if (attr->ia_valid & ATTR_MODE)
622 error = generic_acl_chmod(inode);
627 static void shmem_evict_inode(struct inode *inode)
629 struct shmem_inode_info *info = SHMEM_I(inode);
630 struct shmem_xattr *xattr, *nxattr;
632 if (inode->i_mapping->a_ops == &shmem_aops) {
633 shmem_unacct_size(info->flags, inode->i_size);
635 shmem_truncate_range(inode, 0, (loff_t)-1);
636 if (!list_empty(&info->swaplist)) {
637 mutex_lock(&shmem_swaplist_mutex);
638 list_del_init(&info->swaplist);
639 mutex_unlock(&shmem_swaplist_mutex);
642 kfree(info->symlink);
644 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
648 BUG_ON(inode->i_blocks);
649 shmem_free_inode(inode->i_sb);
654 * If swap found in inode, free it and move page from swapcache to filecache.
656 static int shmem_unuse_inode(struct shmem_inode_info *info,
657 swp_entry_t swap, struct page **pagep)
659 struct address_space *mapping = info->vfs_inode.i_mapping;
665 radswap = swp_to_radix_entry(swap);
666 index = radix_tree_locate_item(&mapping->page_tree, radswap);
671 * Move _head_ to start search for next from here.
672 * But be careful: shmem_evict_inode checks list_empty without taking
673 * mutex, and there's an instant in list_move_tail when info->swaplist
674 * would appear empty, if it were the only one on shmem_swaplist.
676 if (shmem_swaplist.next != &info->swaplist)
677 list_move_tail(&shmem_swaplist, &info->swaplist);
679 gfp = mapping_gfp_mask(mapping);
680 if (shmem_should_replace_page(*pagep, gfp)) {
681 mutex_unlock(&shmem_swaplist_mutex);
682 error = shmem_replace_page(pagep, gfp, info, index);
683 mutex_lock(&shmem_swaplist_mutex);
685 * We needed to drop mutex to make that restrictive page
686 * allocation, but the inode might have been freed while we
687 * dropped it: although a racing shmem_evict_inode() cannot
688 * complete without emptying the radix_tree, our page lock
689 * on this swapcache page is not enough to prevent that -
690 * free_swap_and_cache() of our swap entry will only
691 * trylock_page(), removing swap from radix_tree whatever.
693 * We must not proceed to shmem_add_to_page_cache() if the
694 * inode has been freed, but of course we cannot rely on
695 * inode or mapping or info to check that. However, we can
696 * safely check if our swap entry is still in use (and here
697 * it can't have got reused for another page): if it's still
698 * in use, then the inode cannot have been freed yet, and we
699 * can safely proceed (if it's no longer in use, that tells
700 * nothing about the inode, but we don't need to unuse swap).
702 if (!page_swapcount(*pagep))
707 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
708 * but also to hold up shmem_evict_inode(): so inode cannot be freed
709 * beneath us (pagelock doesn't help until the page is in pagecache).
712 error = shmem_add_to_page_cache(*pagep, mapping, index,
713 GFP_NOWAIT, radswap);
714 if (error != -ENOMEM) {
716 * Truncation and eviction use free_swap_and_cache(), which
717 * only does trylock page: if we raced, best clean up here.
719 delete_from_swap_cache(*pagep);
720 set_page_dirty(*pagep);
722 spin_lock(&info->lock);
724 spin_unlock(&info->lock);
727 error = 1; /* not an error, but entry was found */
733 * Search through swapped inodes to find and replace swap by page.
735 int shmem_unuse(swp_entry_t swap, struct page *page)
737 struct list_head *this, *next;
738 struct shmem_inode_info *info;
743 * There's a faint possibility that swap page was replaced before
744 * caller locked it: caller will come back later with the right page.
746 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
750 * Charge page using GFP_KERNEL while we can wait, before taking
751 * the shmem_swaplist_mutex which might hold up shmem_writepage().
752 * Charged back to the user (not to caller) when swap account is used.
754 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
757 /* No radix_tree_preload: swap entry keeps a place for page in tree */
759 mutex_lock(&shmem_swaplist_mutex);
760 list_for_each_safe(this, next, &shmem_swaplist) {
761 info = list_entry(this, struct shmem_inode_info, swaplist);
763 found = shmem_unuse_inode(info, swap, &page);
765 list_del_init(&info->swaplist);
770 mutex_unlock(&shmem_swaplist_mutex);
776 page_cache_release(page);
781 * Move the page from the page cache to the swap cache.
783 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
785 struct shmem_inode_info *info;
786 struct address_space *mapping;
791 BUG_ON(!PageLocked(page));
792 mapping = page->mapping;
794 inode = mapping->host;
795 info = SHMEM_I(inode);
796 if (info->flags & VM_LOCKED)
798 if (!total_swap_pages)
802 * shmem_backing_dev_info's capabilities prevent regular writeback or
803 * sync from ever calling shmem_writepage; but a stacking filesystem
804 * might use ->writepage of its underlying filesystem, in which case
805 * tmpfs should write out to swap only in response to memory pressure,
806 * and not for the writeback threads or sync.
808 if (!wbc->for_reclaim) {
809 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
814 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
815 * value into swapfile.c, the only way we can correctly account for a
816 * fallocated page arriving here is now to initialize it and write it.
818 * That's okay for a page already fallocated earlier, but if we have
819 * not yet completed the fallocation, then (a) we want to keep track
820 * of this page in case we have to undo it, and (b) it may not be a
821 * good idea to continue anyway, once we're pushing into swap. So
822 * reactivate the page, and let shmem_fallocate() quit when too many.
824 if (!PageUptodate(page)) {
825 if (inode->i_private) {
826 struct shmem_falloc *shmem_falloc;
827 spin_lock(&inode->i_lock);
828 shmem_falloc = inode->i_private;
830 index >= shmem_falloc->start &&
831 index < shmem_falloc->next)
832 shmem_falloc->nr_unswapped++;
835 spin_unlock(&inode->i_lock);
839 clear_highpage(page);
840 flush_dcache_page(page);
841 SetPageUptodate(page);
844 swap = get_swap_page();
849 * Add inode to shmem_unuse()'s list of swapped-out inodes,
850 * if it's not already there. Do it now before the page is
851 * moved to swap cache, when its pagelock no longer protects
852 * the inode from eviction. But don't unlock the mutex until
853 * we've incremented swapped, because shmem_unuse_inode() will
854 * prune a !swapped inode from the swaplist under this mutex.
856 mutex_lock(&shmem_swaplist_mutex);
857 if (list_empty(&info->swaplist))
858 list_add_tail(&info->swaplist, &shmem_swaplist);
860 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
861 swap_shmem_alloc(swap);
862 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
864 spin_lock(&info->lock);
866 shmem_recalc_inode(inode);
867 spin_unlock(&info->lock);
869 mutex_unlock(&shmem_swaplist_mutex);
870 BUG_ON(page_mapped(page));
871 swap_writepage(page, wbc);
875 mutex_unlock(&shmem_swaplist_mutex);
876 swapcache_free(swap, NULL);
878 set_page_dirty(page);
879 if (wbc->for_reclaim)
880 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
887 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
891 if (!mpol || mpol->mode == MPOL_DEFAULT)
892 return; /* show nothing */
894 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
896 seq_printf(seq, ",mpol=%s", buffer);
899 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
901 struct mempolicy *mpol = NULL;
903 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
906 spin_unlock(&sbinfo->stat_lock);
910 #endif /* CONFIG_TMPFS */
912 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
913 struct shmem_inode_info *info, pgoff_t index)
915 struct mempolicy mpol, *spol;
916 struct vm_area_struct pvma;
918 spol = mpol_cond_copy(&mpol,
919 mpol_shared_policy_lookup(&info->policy, index));
921 /* Create a pseudo vma that just contains the policy */
923 pvma.vm_pgoff = index;
925 pvma.vm_policy = spol;
926 return swapin_readahead(swap, gfp, &pvma, 0);
929 static struct page *shmem_alloc_page(gfp_t gfp,
930 struct shmem_inode_info *info, pgoff_t index)
932 struct vm_area_struct pvma;
934 /* Create a pseudo vma that just contains the policy */
936 pvma.vm_pgoff = index;
938 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
941 * alloc_page_vma() will drop the shared policy reference
943 return alloc_page_vma(gfp, &pvma, 0);
945 #else /* !CONFIG_NUMA */
947 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
950 #endif /* CONFIG_TMPFS */
952 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
953 struct shmem_inode_info *info, pgoff_t index)
955 return swapin_readahead(swap, gfp, NULL, 0);
958 static inline struct page *shmem_alloc_page(gfp_t gfp,
959 struct shmem_inode_info *info, pgoff_t index)
961 return alloc_page(gfp);
963 #endif /* CONFIG_NUMA */
965 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
966 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
973 * When a page is moved from swapcache to shmem filecache (either by the
974 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
975 * shmem_unuse_inode()), it may have been read in earlier from swap, in
976 * ignorance of the mapping it belongs to. If that mapping has special
977 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
978 * we may need to copy to a suitable page before moving to filecache.
980 * In a future release, this may well be extended to respect cpuset and
981 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
982 * but for now it is a simple matter of zone.
984 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
986 return page_zonenum(page) > gfp_zone(gfp);
989 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
990 struct shmem_inode_info *info, pgoff_t index)
992 struct page *oldpage, *newpage;
993 struct address_space *swap_mapping;
998 swap_index = page_private(oldpage);
999 swap_mapping = page_mapping(oldpage);
1002 * We have arrived here because our zones are constrained, so don't
1003 * limit chance of success by further cpuset and node constraints.
1005 gfp &= ~GFP_CONSTRAINT_MASK;
1006 newpage = shmem_alloc_page(gfp, info, index);
1010 page_cache_get(newpage);
1011 copy_highpage(newpage, oldpage);
1012 flush_dcache_page(newpage);
1014 __set_page_locked(newpage);
1015 SetPageUptodate(newpage);
1016 SetPageSwapBacked(newpage);
1017 set_page_private(newpage, swap_index);
1018 SetPageSwapCache(newpage);
1021 * Our caller will very soon move newpage out of swapcache, but it's
1022 * a nice clean interface for us to replace oldpage by newpage there.
1024 spin_lock_irq(&swap_mapping->tree_lock);
1025 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1028 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1029 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1031 spin_unlock_irq(&swap_mapping->tree_lock);
1033 if (unlikely(error)) {
1035 * Is this possible? I think not, now that our callers check
1036 * both PageSwapCache and page_private after getting page lock;
1037 * but be defensive. Reverse old to newpage for clear and free.
1041 mem_cgroup_replace_page_cache(oldpage, newpage);
1042 lru_cache_add_anon(newpage);
1046 ClearPageSwapCache(oldpage);
1047 set_page_private(oldpage, 0);
1049 unlock_page(oldpage);
1050 page_cache_release(oldpage);
1051 page_cache_release(oldpage);
1056 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1058 * If we allocate a new one we do not mark it dirty. That's up to the
1059 * vm. If we swap it in we mark it dirty since we also free the swap
1060 * entry since a page cannot live in both the swap and page cache
1062 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1063 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1065 struct address_space *mapping = inode->i_mapping;
1066 struct shmem_inode_info *info;
1067 struct shmem_sb_info *sbinfo;
1074 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1078 page = find_lock_page(mapping, index);
1079 if (radix_tree_exceptional_entry(page)) {
1080 swap = radix_to_swp_entry(page);
1084 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1085 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1090 /* fallocated page? */
1091 if (page && !PageUptodate(page)) {
1092 if (sgp != SGP_READ)
1095 page_cache_release(page);
1098 if (page || (sgp == SGP_READ && !swap.val)) {
1104 * Fast cache lookup did not find it:
1105 * bring it back from swap or allocate.
1107 info = SHMEM_I(inode);
1108 sbinfo = SHMEM_SB(inode->i_sb);
1111 /* Look it up and read it in.. */
1112 page = lookup_swap_cache(swap);
1114 /* here we actually do the io */
1116 *fault_type |= VM_FAULT_MAJOR;
1117 page = shmem_swapin(swap, gfp, info, index);
1124 /* We have to do this with page locked to prevent races */
1126 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1128 error = -EEXIST; /* try again */
1131 if (!PageUptodate(page)) {
1135 wait_on_page_writeback(page);
1137 if (shmem_should_replace_page(page, gfp)) {
1138 error = shmem_replace_page(&page, gfp, info, index);
1143 error = mem_cgroup_cache_charge(page, current->mm,
1144 gfp & GFP_RECLAIM_MASK);
1146 error = shmem_add_to_page_cache(page, mapping, index,
1147 gfp, swp_to_radix_entry(swap));
1151 spin_lock(&info->lock);
1153 shmem_recalc_inode(inode);
1154 spin_unlock(&info->lock);
1156 delete_from_swap_cache(page);
1157 set_page_dirty(page);
1161 if (shmem_acct_block(info->flags)) {
1165 if (sbinfo->max_blocks) {
1166 if (percpu_counter_compare(&sbinfo->used_blocks,
1167 sbinfo->max_blocks) >= 0) {
1171 percpu_counter_inc(&sbinfo->used_blocks);
1174 page = shmem_alloc_page(gfp, info, index);
1180 SetPageSwapBacked(page);
1181 __set_page_locked(page);
1182 error = mem_cgroup_cache_charge(page, current->mm,
1183 gfp & GFP_RECLAIM_MASK);
1185 error = shmem_add_to_page_cache(page, mapping, index,
1189 lru_cache_add_anon(page);
1191 spin_lock(&info->lock);
1193 inode->i_blocks += BLOCKS_PER_PAGE;
1194 shmem_recalc_inode(inode);
1195 spin_unlock(&info->lock);
1199 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1201 if (sgp == SGP_FALLOC)
1205 * Let SGP_WRITE caller clear ends if write does not fill page;
1206 * but SGP_FALLOC on a page fallocated earlier must initialize
1207 * it now, lest undo on failure cancel our earlier guarantee.
1209 if (sgp != SGP_WRITE) {
1210 clear_highpage(page);
1211 flush_dcache_page(page);
1212 SetPageUptodate(page);
1214 if (sgp == SGP_DIRTY)
1215 set_page_dirty(page);
1218 /* Perhaps the file has been truncated since we checked */
1219 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1220 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1234 info = SHMEM_I(inode);
1235 ClearPageDirty(page);
1236 delete_from_page_cache(page);
1237 spin_lock(&info->lock);
1239 inode->i_blocks -= BLOCKS_PER_PAGE;
1240 spin_unlock(&info->lock);
1242 sbinfo = SHMEM_SB(inode->i_sb);
1243 if (sbinfo->max_blocks)
1244 percpu_counter_add(&sbinfo->used_blocks, -1);
1246 shmem_unacct_blocks(info->flags, 1);
1248 if (swap.val && error != -EINVAL) {
1249 struct page *test = find_get_page(mapping, index);
1250 if (test && !radix_tree_exceptional_entry(test))
1251 page_cache_release(test);
1252 /* Have another try if the entry has changed */
1253 if (test != swp_to_radix_entry(swap))
1258 page_cache_release(page);
1260 if (error == -ENOSPC && !once++) {
1261 info = SHMEM_I(inode);
1262 spin_lock(&info->lock);
1263 shmem_recalc_inode(inode);
1264 spin_unlock(&info->lock);
1267 if (error == -EEXIST)
1272 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1274 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1276 int ret = VM_FAULT_LOCKED;
1278 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1280 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1282 if (ret & VM_FAULT_MAJOR) {
1283 count_vm_event(PGMAJFAULT);
1284 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1290 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1292 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1293 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1296 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1299 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1302 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1303 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1307 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1309 struct inode *inode = file->f_path.dentry->d_inode;
1310 struct shmem_inode_info *info = SHMEM_I(inode);
1311 int retval = -ENOMEM;
1313 spin_lock(&info->lock);
1314 if (lock && !(info->flags & VM_LOCKED)) {
1315 if (!user_shm_lock(inode->i_size, user))
1317 info->flags |= VM_LOCKED;
1318 mapping_set_unevictable(file->f_mapping);
1320 if (!lock && (info->flags & VM_LOCKED) && user) {
1321 user_shm_unlock(inode->i_size, user);
1322 info->flags &= ~VM_LOCKED;
1323 mapping_clear_unevictable(file->f_mapping);
1328 spin_unlock(&info->lock);
1332 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1334 file_accessed(file);
1335 vma->vm_ops = &shmem_vm_ops;
1336 vma->vm_flags |= VM_CAN_NONLINEAR;
1340 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1341 umode_t mode, dev_t dev, unsigned long flags)
1343 struct inode *inode;
1344 struct shmem_inode_info *info;
1345 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1347 if (shmem_reserve_inode(sb))
1350 inode = new_inode(sb);
1352 inode->i_ino = get_next_ino();
1353 inode_init_owner(inode, dir, mode);
1354 inode->i_blocks = 0;
1355 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1356 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1357 inode->i_generation = get_seconds();
1358 info = SHMEM_I(inode);
1359 memset(info, 0, (char *)inode - (char *)info);
1360 spin_lock_init(&info->lock);
1361 info->flags = flags & VM_NORESERVE;
1362 INIT_LIST_HEAD(&info->swaplist);
1363 INIT_LIST_HEAD(&info->xattr_list);
1364 cache_no_acl(inode);
1366 switch (mode & S_IFMT) {
1368 inode->i_op = &shmem_special_inode_operations;
1369 init_special_inode(inode, mode, dev);
1372 inode->i_mapping->a_ops = &shmem_aops;
1373 inode->i_op = &shmem_inode_operations;
1374 inode->i_fop = &shmem_file_operations;
1375 mpol_shared_policy_init(&info->policy,
1376 shmem_get_sbmpol(sbinfo));
1380 /* Some things misbehave if size == 0 on a directory */
1381 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1382 inode->i_op = &shmem_dir_inode_operations;
1383 inode->i_fop = &simple_dir_operations;
1387 * Must not load anything in the rbtree,
1388 * mpol_free_shared_policy will not be called.
1390 mpol_shared_policy_init(&info->policy, NULL);
1394 shmem_free_inode(sb);
1399 static const struct inode_operations shmem_symlink_inode_operations;
1400 static const struct inode_operations shmem_short_symlink_operations;
1402 #ifdef CONFIG_TMPFS_XATTR
1403 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1405 #define shmem_initxattrs NULL
1409 shmem_write_begin(struct file *file, struct address_space *mapping,
1410 loff_t pos, unsigned len, unsigned flags,
1411 struct page **pagep, void **fsdata)
1413 struct inode *inode = mapping->host;
1414 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1415 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1419 shmem_write_end(struct file *file, struct address_space *mapping,
1420 loff_t pos, unsigned len, unsigned copied,
1421 struct page *page, void *fsdata)
1423 struct inode *inode = mapping->host;
1425 if (pos + copied > inode->i_size)
1426 i_size_write(inode, pos + copied);
1428 if (!PageUptodate(page)) {
1429 if (copied < PAGE_CACHE_SIZE) {
1430 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1431 zero_user_segments(page, 0, from,
1432 from + copied, PAGE_CACHE_SIZE);
1434 SetPageUptodate(page);
1436 set_page_dirty(page);
1438 page_cache_release(page);
1443 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1445 struct inode *inode = filp->f_path.dentry->d_inode;
1446 struct address_space *mapping = inode->i_mapping;
1448 unsigned long offset;
1449 enum sgp_type sgp = SGP_READ;
1452 * Might this read be for a stacking filesystem? Then when reading
1453 * holes of a sparse file, we actually need to allocate those pages,
1454 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1456 if (segment_eq(get_fs(), KERNEL_DS))
1459 index = *ppos >> PAGE_CACHE_SHIFT;
1460 offset = *ppos & ~PAGE_CACHE_MASK;
1463 struct page *page = NULL;
1465 unsigned long nr, ret;
1466 loff_t i_size = i_size_read(inode);
1468 end_index = i_size >> PAGE_CACHE_SHIFT;
1469 if (index > end_index)
1471 if (index == end_index) {
1472 nr = i_size & ~PAGE_CACHE_MASK;
1477 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1479 if (desc->error == -EINVAL)
1487 * We must evaluate after, since reads (unlike writes)
1488 * are called without i_mutex protection against truncate
1490 nr = PAGE_CACHE_SIZE;
1491 i_size = i_size_read(inode);
1492 end_index = i_size >> PAGE_CACHE_SHIFT;
1493 if (index == end_index) {
1494 nr = i_size & ~PAGE_CACHE_MASK;
1497 page_cache_release(page);
1505 * If users can be writing to this page using arbitrary
1506 * virtual addresses, take care about potential aliasing
1507 * before reading the page on the kernel side.
1509 if (mapping_writably_mapped(mapping))
1510 flush_dcache_page(page);
1512 * Mark the page accessed if we read the beginning.
1515 mark_page_accessed(page);
1517 page = ZERO_PAGE(0);
1518 page_cache_get(page);
1522 * Ok, we have the page, and it's up-to-date, so
1523 * now we can copy it to user space...
1525 * The actor routine returns how many bytes were actually used..
1526 * NOTE! This may not be the same as how much of a user buffer
1527 * we filled up (we may be padding etc), so we can only update
1528 * "pos" here (the actor routine has to update the user buffer
1529 * pointers and the remaining count).
1531 ret = actor(desc, page, offset, nr);
1533 index += offset >> PAGE_CACHE_SHIFT;
1534 offset &= ~PAGE_CACHE_MASK;
1536 page_cache_release(page);
1537 if (ret != nr || !desc->count)
1543 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1544 file_accessed(filp);
1547 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1548 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1550 struct file *filp = iocb->ki_filp;
1554 loff_t *ppos = &iocb->ki_pos;
1556 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1560 for (seg = 0; seg < nr_segs; seg++) {
1561 read_descriptor_t desc;
1564 desc.arg.buf = iov[seg].iov_base;
1565 desc.count = iov[seg].iov_len;
1566 if (desc.count == 0)
1569 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1570 retval += desc.written;
1572 retval = retval ?: desc.error;
1581 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1582 struct pipe_inode_info *pipe, size_t len,
1585 struct address_space *mapping = in->f_mapping;
1586 struct inode *inode = mapping->host;
1587 unsigned int loff, nr_pages, req_pages;
1588 struct page *pages[PIPE_DEF_BUFFERS];
1589 struct partial_page partial[PIPE_DEF_BUFFERS];
1591 pgoff_t index, end_index;
1594 struct splice_pipe_desc spd = {
1597 .nr_pages_max = PIPE_DEF_BUFFERS,
1599 .ops = &page_cache_pipe_buf_ops,
1600 .spd_release = spd_release_page,
1603 isize = i_size_read(inode);
1604 if (unlikely(*ppos >= isize))
1607 left = isize - *ppos;
1608 if (unlikely(left < len))
1611 if (splice_grow_spd(pipe, &spd))
1614 index = *ppos >> PAGE_CACHE_SHIFT;
1615 loff = *ppos & ~PAGE_CACHE_MASK;
1616 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1617 nr_pages = min(req_pages, pipe->buffers);
1619 spd.nr_pages = find_get_pages_contig(mapping, index,
1620 nr_pages, spd.pages);
1621 index += spd.nr_pages;
1624 while (spd.nr_pages < nr_pages) {
1625 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1629 spd.pages[spd.nr_pages++] = page;
1633 index = *ppos >> PAGE_CACHE_SHIFT;
1634 nr_pages = spd.nr_pages;
1637 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1638 unsigned int this_len;
1643 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1644 page = spd.pages[page_nr];
1646 if (!PageUptodate(page) || page->mapping != mapping) {
1647 error = shmem_getpage(inode, index, &page,
1652 page_cache_release(spd.pages[page_nr]);
1653 spd.pages[page_nr] = page;
1656 isize = i_size_read(inode);
1657 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1658 if (unlikely(!isize || index > end_index))
1661 if (end_index == index) {
1664 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1668 this_len = min(this_len, plen - loff);
1672 spd.partial[page_nr].offset = loff;
1673 spd.partial[page_nr].len = this_len;
1680 while (page_nr < nr_pages)
1681 page_cache_release(spd.pages[page_nr++]);
1684 error = splice_to_pipe(pipe, &spd);
1686 splice_shrink_spd(&spd);
1695 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1698 struct inode *inode = file->f_path.dentry->d_inode;
1699 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1700 struct shmem_falloc shmem_falloc;
1701 pgoff_t start, index, end;
1704 mutex_lock(&inode->i_mutex);
1706 if (mode & FALLOC_FL_PUNCH_HOLE) {
1707 struct address_space *mapping = file->f_mapping;
1708 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1709 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1711 if ((u64)unmap_end > (u64)unmap_start)
1712 unmap_mapping_range(mapping, unmap_start,
1713 1 + unmap_end - unmap_start, 0);
1714 shmem_truncate_range(inode, offset, offset + len - 1);
1715 /* No need to unmap again: hole-punching leaves COWed pages */
1720 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1721 error = inode_newsize_ok(inode, offset + len);
1725 start = offset >> PAGE_CACHE_SHIFT;
1726 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1727 /* Try to avoid a swapstorm if len is impossible to satisfy */
1728 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1733 shmem_falloc.start = start;
1734 shmem_falloc.next = start;
1735 shmem_falloc.nr_falloced = 0;
1736 shmem_falloc.nr_unswapped = 0;
1737 spin_lock(&inode->i_lock);
1738 inode->i_private = &shmem_falloc;
1739 spin_unlock(&inode->i_lock);
1741 for (index = start; index < end; index++) {
1745 * Good, the fallocate(2) manpage permits EINTR: we may have
1746 * been interrupted because we are using up too much memory.
1748 if (signal_pending(current))
1750 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1753 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1756 /* Remove the !PageUptodate pages we added */
1757 shmem_undo_range(inode,
1758 (loff_t)start << PAGE_CACHE_SHIFT,
1759 (loff_t)index << PAGE_CACHE_SHIFT, true);
1764 * Inform shmem_writepage() how far we have reached.
1765 * No need for lock or barrier: we have the page lock.
1767 shmem_falloc.next++;
1768 if (!PageUptodate(page))
1769 shmem_falloc.nr_falloced++;
1772 * If !PageUptodate, leave it that way so that freeable pages
1773 * can be recognized if we need to rollback on error later.
1774 * But set_page_dirty so that memory pressure will swap rather
1775 * than free the pages we are allocating (and SGP_CACHE pages
1776 * might still be clean: we now need to mark those dirty too).
1778 set_page_dirty(page);
1780 page_cache_release(page);
1784 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1785 i_size_write(inode, offset + len);
1786 inode->i_ctime = CURRENT_TIME;
1788 spin_lock(&inode->i_lock);
1789 inode->i_private = NULL;
1790 spin_unlock(&inode->i_lock);
1792 mutex_unlock(&inode->i_mutex);
1796 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1798 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1800 buf->f_type = TMPFS_MAGIC;
1801 buf->f_bsize = PAGE_CACHE_SIZE;
1802 buf->f_namelen = NAME_MAX;
1803 if (sbinfo->max_blocks) {
1804 buf->f_blocks = sbinfo->max_blocks;
1806 buf->f_bfree = sbinfo->max_blocks -
1807 percpu_counter_sum(&sbinfo->used_blocks);
1809 if (sbinfo->max_inodes) {
1810 buf->f_files = sbinfo->max_inodes;
1811 buf->f_ffree = sbinfo->free_inodes;
1813 /* else leave those fields 0 like simple_statfs */
1818 * File creation. Allocate an inode, and we're done..
1821 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1823 struct inode *inode;
1824 int error = -ENOSPC;
1826 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1828 error = security_inode_init_security(inode, dir,
1830 shmem_initxattrs, NULL);
1832 if (error != -EOPNOTSUPP) {
1837 #ifdef CONFIG_TMPFS_POSIX_ACL
1838 error = generic_acl_init(inode, dir);
1846 dir->i_size += BOGO_DIRENT_SIZE;
1847 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1848 d_instantiate(dentry, inode);
1849 dget(dentry); /* Extra count - pin the dentry in core */
1854 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1858 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1864 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1865 struct nameidata *nd)
1867 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1873 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1875 struct inode *inode = old_dentry->d_inode;
1879 * No ordinary (disk based) filesystem counts links as inodes;
1880 * but each new link needs a new dentry, pinning lowmem, and
1881 * tmpfs dentries cannot be pruned until they are unlinked.
1883 ret = shmem_reserve_inode(inode->i_sb);
1887 dir->i_size += BOGO_DIRENT_SIZE;
1888 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1890 ihold(inode); /* New dentry reference */
1891 dget(dentry); /* Extra pinning count for the created dentry */
1892 d_instantiate(dentry, inode);
1897 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1899 struct inode *inode = dentry->d_inode;
1901 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1902 shmem_free_inode(inode->i_sb);
1904 dir->i_size -= BOGO_DIRENT_SIZE;
1905 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1907 dput(dentry); /* Undo the count from "create" - this does all the work */
1911 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1913 if (!simple_empty(dentry))
1916 drop_nlink(dentry->d_inode);
1918 return shmem_unlink(dir, dentry);
1922 * The VFS layer already does all the dentry stuff for rename,
1923 * we just have to decrement the usage count for the target if
1924 * it exists so that the VFS layer correctly free's it when it
1927 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1929 struct inode *inode = old_dentry->d_inode;
1930 int they_are_dirs = S_ISDIR(inode->i_mode);
1932 if (!simple_empty(new_dentry))
1935 if (new_dentry->d_inode) {
1936 (void) shmem_unlink(new_dir, new_dentry);
1938 drop_nlink(old_dir);
1939 } else if (they_are_dirs) {
1940 drop_nlink(old_dir);
1944 old_dir->i_size -= BOGO_DIRENT_SIZE;
1945 new_dir->i_size += BOGO_DIRENT_SIZE;
1946 old_dir->i_ctime = old_dir->i_mtime =
1947 new_dir->i_ctime = new_dir->i_mtime =
1948 inode->i_ctime = CURRENT_TIME;
1952 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1956 struct inode *inode;
1959 struct shmem_inode_info *info;
1961 len = strlen(symname) + 1;
1962 if (len > PAGE_CACHE_SIZE)
1963 return -ENAMETOOLONG;
1965 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1969 error = security_inode_init_security(inode, dir, &dentry->d_name,
1970 shmem_initxattrs, NULL);
1972 if (error != -EOPNOTSUPP) {
1979 info = SHMEM_I(inode);
1980 inode->i_size = len-1;
1981 if (len <= SHORT_SYMLINK_LEN) {
1982 info->symlink = kmemdup(symname, len, GFP_KERNEL);
1983 if (!info->symlink) {
1987 inode->i_op = &shmem_short_symlink_operations;
1989 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1994 inode->i_mapping->a_ops = &shmem_aops;
1995 inode->i_op = &shmem_symlink_inode_operations;
1996 kaddr = kmap_atomic(page);
1997 memcpy(kaddr, symname, len);
1998 kunmap_atomic(kaddr);
1999 SetPageUptodate(page);
2000 set_page_dirty(page);
2002 page_cache_release(page);
2004 dir->i_size += BOGO_DIRENT_SIZE;
2005 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2006 d_instantiate(dentry, inode);
2011 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2013 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2017 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2019 struct page *page = NULL;
2020 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2021 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2027 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2029 if (!IS_ERR(nd_get_link(nd))) {
2030 struct page *page = cookie;
2032 mark_page_accessed(page);
2033 page_cache_release(page);
2037 #ifdef CONFIG_TMPFS_XATTR
2039 * Superblocks without xattr inode operations may get some security.* xattr
2040 * support from the LSM "for free". As soon as we have any other xattrs
2041 * like ACLs, we also need to implement the security.* handlers at
2042 * filesystem level, though.
2046 * Allocate new xattr and copy in the value; but leave the name to callers.
2048 static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size)
2050 struct shmem_xattr *new_xattr;
2054 len = sizeof(*new_xattr) + size;
2055 if (len <= sizeof(*new_xattr))
2058 new_xattr = kmalloc(len, GFP_KERNEL);
2062 new_xattr->size = size;
2063 memcpy(new_xattr->value, value, size);
2068 * Callback for security_inode_init_security() for acquiring xattrs.
2070 static int shmem_initxattrs(struct inode *inode,
2071 const struct xattr *xattr_array,
2074 struct shmem_inode_info *info = SHMEM_I(inode);
2075 const struct xattr *xattr;
2076 struct shmem_xattr *new_xattr;
2079 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2080 new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len);
2084 len = strlen(xattr->name) + 1;
2085 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2087 if (!new_xattr->name) {
2092 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2093 XATTR_SECURITY_PREFIX_LEN);
2094 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2097 spin_lock(&info->lock);
2098 list_add(&new_xattr->list, &info->xattr_list);
2099 spin_unlock(&info->lock);
2105 static int shmem_xattr_get(struct dentry *dentry, const char *name,
2106 void *buffer, size_t size)
2108 struct shmem_inode_info *info;
2109 struct shmem_xattr *xattr;
2112 info = SHMEM_I(dentry->d_inode);
2114 spin_lock(&info->lock);
2115 list_for_each_entry(xattr, &info->xattr_list, list) {
2116 if (strcmp(name, xattr->name))
2121 if (size < xattr->size)
2124 memcpy(buffer, xattr->value, xattr->size);
2128 spin_unlock(&info->lock);
2132 static int shmem_xattr_set(struct inode *inode, const char *name,
2133 const void *value, size_t size, int flags)
2135 struct shmem_inode_info *info = SHMEM_I(inode);
2136 struct shmem_xattr *xattr;
2137 struct shmem_xattr *new_xattr = NULL;
2140 /* value == NULL means remove */
2142 new_xattr = shmem_xattr_alloc(value, size);
2146 new_xattr->name = kstrdup(name, GFP_KERNEL);
2147 if (!new_xattr->name) {
2153 spin_lock(&info->lock);
2154 list_for_each_entry(xattr, &info->xattr_list, list) {
2155 if (!strcmp(name, xattr->name)) {
2156 if (flags & XATTR_CREATE) {
2159 } else if (new_xattr) {
2160 list_replace(&xattr->list, &new_xattr->list);
2162 list_del(&xattr->list);
2167 if (flags & XATTR_REPLACE) {
2171 list_add(&new_xattr->list, &info->xattr_list);
2175 spin_unlock(&info->lock);
2182 static const struct xattr_handler *shmem_xattr_handlers[] = {
2183 #ifdef CONFIG_TMPFS_POSIX_ACL
2184 &generic_acl_access_handler,
2185 &generic_acl_default_handler,
2190 static int shmem_xattr_validate(const char *name)
2192 struct { const char *prefix; size_t len; } arr[] = {
2193 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2194 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2198 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2199 size_t preflen = arr[i].len;
2200 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2209 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2210 void *buffer, size_t size)
2215 * If this is a request for a synthetic attribute in the system.*
2216 * namespace use the generic infrastructure to resolve a handler
2217 * for it via sb->s_xattr.
2219 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2220 return generic_getxattr(dentry, name, buffer, size);
2222 err = shmem_xattr_validate(name);
2226 return shmem_xattr_get(dentry, name, buffer, size);
2229 static int shmem_setxattr(struct dentry *dentry, const char *name,
2230 const void *value, size_t size, int flags)
2235 * If this is a request for a synthetic attribute in the system.*
2236 * namespace use the generic infrastructure to resolve a handler
2237 * for it via sb->s_xattr.
2239 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2240 return generic_setxattr(dentry, name, value, size, flags);
2242 err = shmem_xattr_validate(name);
2247 value = ""; /* empty EA, do not remove */
2249 return shmem_xattr_set(dentry->d_inode, name, value, size, flags);
2253 static int shmem_removexattr(struct dentry *dentry, const char *name)
2258 * If this is a request for a synthetic attribute in the system.*
2259 * namespace use the generic infrastructure to resolve a handler
2260 * for it via sb->s_xattr.
2262 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2263 return generic_removexattr(dentry, name);
2265 err = shmem_xattr_validate(name);
2269 return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
2272 static bool xattr_is_trusted(const char *name)
2274 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
2277 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2279 bool trusted = capable(CAP_SYS_ADMIN);
2280 struct shmem_xattr *xattr;
2281 struct shmem_inode_info *info;
2284 info = SHMEM_I(dentry->d_inode);
2286 spin_lock(&info->lock);
2287 list_for_each_entry(xattr, &info->xattr_list, list) {
2290 /* skip "trusted." attributes for unprivileged callers */
2291 if (!trusted && xattr_is_trusted(xattr->name))
2294 len = strlen(xattr->name) + 1;
2301 memcpy(buffer, xattr->name, len);
2305 spin_unlock(&info->lock);
2309 #endif /* CONFIG_TMPFS_XATTR */
2311 static const struct inode_operations shmem_short_symlink_operations = {
2312 .readlink = generic_readlink,
2313 .follow_link = shmem_follow_short_symlink,
2314 #ifdef CONFIG_TMPFS_XATTR
2315 .setxattr = shmem_setxattr,
2316 .getxattr = shmem_getxattr,
2317 .listxattr = shmem_listxattr,
2318 .removexattr = shmem_removexattr,
2322 static const struct inode_operations shmem_symlink_inode_operations = {
2323 .readlink = generic_readlink,
2324 .follow_link = shmem_follow_link,
2325 .put_link = shmem_put_link,
2326 #ifdef CONFIG_TMPFS_XATTR
2327 .setxattr = shmem_setxattr,
2328 .getxattr = shmem_getxattr,
2329 .listxattr = shmem_listxattr,
2330 .removexattr = shmem_removexattr,
2334 static struct dentry *shmem_get_parent(struct dentry *child)
2336 return ERR_PTR(-ESTALE);
2339 static int shmem_match(struct inode *ino, void *vfh)
2343 inum = (inum << 32) | fh[1];
2344 return ino->i_ino == inum && fh[0] == ino->i_generation;
2347 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2348 struct fid *fid, int fh_len, int fh_type)
2350 struct inode *inode;
2351 struct dentry *dentry = NULL;
2352 u64 inum = fid->raw[2];
2353 inum = (inum << 32) | fid->raw[1];
2358 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2359 shmem_match, fid->raw);
2361 dentry = d_find_alias(inode);
2368 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2369 struct inode *parent)
2376 if (inode_unhashed(inode)) {
2377 /* Unfortunately insert_inode_hash is not idempotent,
2378 * so as we hash inodes here rather than at creation
2379 * time, we need a lock to ensure we only try
2382 static DEFINE_SPINLOCK(lock);
2384 if (inode_unhashed(inode))
2385 __insert_inode_hash(inode,
2386 inode->i_ino + inode->i_generation);
2390 fh[0] = inode->i_generation;
2391 fh[1] = inode->i_ino;
2392 fh[2] = ((__u64)inode->i_ino) >> 32;
2398 static const struct export_operations shmem_export_ops = {
2399 .get_parent = shmem_get_parent,
2400 .encode_fh = shmem_encode_fh,
2401 .fh_to_dentry = shmem_fh_to_dentry,
2404 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2407 char *this_char, *value, *rest;
2411 while (options != NULL) {
2412 this_char = options;
2415 * NUL-terminate this option: unfortunately,
2416 * mount options form a comma-separated list,
2417 * but mpol's nodelist may also contain commas.
2419 options = strchr(options, ',');
2420 if (options == NULL)
2423 if (!isdigit(*options)) {
2430 if ((value = strchr(this_char,'=')) != NULL) {
2434 "tmpfs: No value for mount option '%s'\n",
2439 if (!strcmp(this_char,"size")) {
2440 unsigned long long size;
2441 size = memparse(value,&rest);
2443 size <<= PAGE_SHIFT;
2444 size *= totalram_pages;
2450 sbinfo->max_blocks =
2451 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2452 } else if (!strcmp(this_char,"nr_blocks")) {
2453 sbinfo->max_blocks = memparse(value, &rest);
2456 } else if (!strcmp(this_char,"nr_inodes")) {
2457 sbinfo->max_inodes = memparse(value, &rest);
2460 } else if (!strcmp(this_char,"mode")) {
2463 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2466 } else if (!strcmp(this_char,"uid")) {
2469 uid = simple_strtoul(value, &rest, 0);
2472 sbinfo->uid = make_kuid(current_user_ns(), uid);
2473 if (!uid_valid(sbinfo->uid))
2475 } else if (!strcmp(this_char,"gid")) {
2478 gid = simple_strtoul(value, &rest, 0);
2481 sbinfo->gid = make_kgid(current_user_ns(), gid);
2482 if (!gid_valid(sbinfo->gid))
2484 } else if (!strcmp(this_char,"mpol")) {
2485 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2488 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2496 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2502 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2504 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2505 struct shmem_sb_info config = *sbinfo;
2506 unsigned long inodes;
2507 int error = -EINVAL;
2509 if (shmem_parse_options(data, &config, true))
2512 spin_lock(&sbinfo->stat_lock);
2513 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2514 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2516 if (config.max_inodes < inodes)
2519 * Those tests disallow limited->unlimited while any are in use;
2520 * but we must separately disallow unlimited->limited, because
2521 * in that case we have no record of how much is already in use.
2523 if (config.max_blocks && !sbinfo->max_blocks)
2525 if (config.max_inodes && !sbinfo->max_inodes)
2529 sbinfo->max_blocks = config.max_blocks;
2530 sbinfo->max_inodes = config.max_inodes;
2531 sbinfo->free_inodes = config.max_inodes - inodes;
2533 mpol_put(sbinfo->mpol);
2534 sbinfo->mpol = config.mpol; /* transfers initial ref */
2536 spin_unlock(&sbinfo->stat_lock);
2540 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2542 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2544 if (sbinfo->max_blocks != shmem_default_max_blocks())
2545 seq_printf(seq, ",size=%luk",
2546 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2547 if (sbinfo->max_inodes != shmem_default_max_inodes())
2548 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2549 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2550 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2551 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2552 seq_printf(seq, ",uid=%u",
2553 from_kuid_munged(&init_user_ns, sbinfo->uid));
2554 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2555 seq_printf(seq, ",gid=%u",
2556 from_kgid_munged(&init_user_ns, sbinfo->gid));
2557 shmem_show_mpol(seq, sbinfo->mpol);
2560 #endif /* CONFIG_TMPFS */
2562 static void shmem_put_super(struct super_block *sb)
2564 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2566 percpu_counter_destroy(&sbinfo->used_blocks);
2568 sb->s_fs_info = NULL;
2571 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2573 struct inode *inode;
2574 struct shmem_sb_info *sbinfo;
2577 /* Round up to L1_CACHE_BYTES to resist false sharing */
2578 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2579 L1_CACHE_BYTES), GFP_KERNEL);
2583 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2584 sbinfo->uid = current_fsuid();
2585 sbinfo->gid = current_fsgid();
2586 sb->s_fs_info = sbinfo;
2590 * Per default we only allow half of the physical ram per
2591 * tmpfs instance, limiting inodes to one per page of lowmem;
2592 * but the internal instance is left unlimited.
2594 if (!(sb->s_flags & MS_NOUSER)) {
2595 sbinfo->max_blocks = shmem_default_max_blocks();
2596 sbinfo->max_inodes = shmem_default_max_inodes();
2597 if (shmem_parse_options(data, sbinfo, false)) {
2602 sb->s_export_op = &shmem_export_ops;
2603 sb->s_flags |= MS_NOSEC;
2605 sb->s_flags |= MS_NOUSER;
2608 spin_lock_init(&sbinfo->stat_lock);
2609 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2611 sbinfo->free_inodes = sbinfo->max_inodes;
2613 sb->s_maxbytes = MAX_LFS_FILESIZE;
2614 sb->s_blocksize = PAGE_CACHE_SIZE;
2615 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2616 sb->s_magic = TMPFS_MAGIC;
2617 sb->s_op = &shmem_ops;
2618 sb->s_time_gran = 1;
2619 #ifdef CONFIG_TMPFS_XATTR
2620 sb->s_xattr = shmem_xattr_handlers;
2622 #ifdef CONFIG_TMPFS_POSIX_ACL
2623 sb->s_flags |= MS_POSIXACL;
2626 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2629 inode->i_uid = sbinfo->uid;
2630 inode->i_gid = sbinfo->gid;
2631 sb->s_root = d_make_root(inode);
2637 shmem_put_super(sb);
2641 static struct kmem_cache *shmem_inode_cachep;
2643 static struct inode *shmem_alloc_inode(struct super_block *sb)
2645 struct shmem_inode_info *info;
2646 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2649 return &info->vfs_inode;
2652 static void shmem_destroy_callback(struct rcu_head *head)
2654 struct inode *inode = container_of(head, struct inode, i_rcu);
2655 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2658 static void shmem_destroy_inode(struct inode *inode)
2660 if (S_ISREG(inode->i_mode))
2661 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2662 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2665 static void shmem_init_inode(void *foo)
2667 struct shmem_inode_info *info = foo;
2668 inode_init_once(&info->vfs_inode);
2671 static int shmem_init_inodecache(void)
2673 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2674 sizeof(struct shmem_inode_info),
2675 0, SLAB_PANIC, shmem_init_inode);
2679 static void shmem_destroy_inodecache(void)
2681 kmem_cache_destroy(shmem_inode_cachep);
2684 static const struct address_space_operations shmem_aops = {
2685 .writepage = shmem_writepage,
2686 .set_page_dirty = __set_page_dirty_no_writeback,
2688 .write_begin = shmem_write_begin,
2689 .write_end = shmem_write_end,
2691 .migratepage = migrate_page,
2692 .error_remove_page = generic_error_remove_page,
2695 static const struct file_operations shmem_file_operations = {
2698 .llseek = generic_file_llseek,
2699 .read = do_sync_read,
2700 .write = do_sync_write,
2701 .aio_read = shmem_file_aio_read,
2702 .aio_write = generic_file_aio_write,
2703 .fsync = noop_fsync,
2704 .splice_read = shmem_file_splice_read,
2705 .splice_write = generic_file_splice_write,
2706 .fallocate = shmem_fallocate,
2710 static const struct inode_operations shmem_inode_operations = {
2711 .setattr = shmem_setattr,
2712 #ifdef CONFIG_TMPFS_XATTR
2713 .setxattr = shmem_setxattr,
2714 .getxattr = shmem_getxattr,
2715 .listxattr = shmem_listxattr,
2716 .removexattr = shmem_removexattr,
2720 static const struct inode_operations shmem_dir_inode_operations = {
2722 .create = shmem_create,
2723 .lookup = simple_lookup,
2725 .unlink = shmem_unlink,
2726 .symlink = shmem_symlink,
2727 .mkdir = shmem_mkdir,
2728 .rmdir = shmem_rmdir,
2729 .mknod = shmem_mknod,
2730 .rename = shmem_rename,
2732 #ifdef CONFIG_TMPFS_XATTR
2733 .setxattr = shmem_setxattr,
2734 .getxattr = shmem_getxattr,
2735 .listxattr = shmem_listxattr,
2736 .removexattr = shmem_removexattr,
2738 #ifdef CONFIG_TMPFS_POSIX_ACL
2739 .setattr = shmem_setattr,
2743 static const struct inode_operations shmem_special_inode_operations = {
2744 #ifdef CONFIG_TMPFS_XATTR
2745 .setxattr = shmem_setxattr,
2746 .getxattr = shmem_getxattr,
2747 .listxattr = shmem_listxattr,
2748 .removexattr = shmem_removexattr,
2750 #ifdef CONFIG_TMPFS_POSIX_ACL
2751 .setattr = shmem_setattr,
2755 static const struct super_operations shmem_ops = {
2756 .alloc_inode = shmem_alloc_inode,
2757 .destroy_inode = shmem_destroy_inode,
2759 .statfs = shmem_statfs,
2760 .remount_fs = shmem_remount_fs,
2761 .show_options = shmem_show_options,
2763 .evict_inode = shmem_evict_inode,
2764 .drop_inode = generic_delete_inode,
2765 .put_super = shmem_put_super,
2768 static const struct vm_operations_struct shmem_vm_ops = {
2769 .fault = shmem_fault,
2771 .set_policy = shmem_set_policy,
2772 .get_policy = shmem_get_policy,
2776 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2777 int flags, const char *dev_name, void *data)
2779 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2782 static struct file_system_type shmem_fs_type = {
2783 .owner = THIS_MODULE,
2785 .mount = shmem_mount,
2786 .kill_sb = kill_litter_super,
2789 int __init shmem_init(void)
2793 error = bdi_init(&shmem_backing_dev_info);
2797 error = shmem_init_inodecache();
2801 error = register_filesystem(&shmem_fs_type);
2803 printk(KERN_ERR "Could not register tmpfs\n");
2807 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2808 shmem_fs_type.name, NULL);
2809 if (IS_ERR(shm_mnt)) {
2810 error = PTR_ERR(shm_mnt);
2811 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2817 unregister_filesystem(&shmem_fs_type);
2819 shmem_destroy_inodecache();
2821 bdi_destroy(&shmem_backing_dev_info);
2823 shm_mnt = ERR_PTR(error);
2827 #else /* !CONFIG_SHMEM */
2830 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2832 * This is intended for small system where the benefits of the full
2833 * shmem code (swap-backed and resource-limited) are outweighed by
2834 * their complexity. On systems without swap this code should be
2835 * effectively equivalent, but much lighter weight.
2838 #include <linux/ramfs.h>
2840 static struct file_system_type shmem_fs_type = {
2842 .mount = ramfs_mount,
2843 .kill_sb = kill_litter_super,
2846 int __init shmem_init(void)
2848 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2850 shm_mnt = kern_mount(&shmem_fs_type);
2851 BUG_ON(IS_ERR(shm_mnt));
2856 int shmem_unuse(swp_entry_t swap, struct page *page)
2861 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2866 void shmem_unlock_mapping(struct address_space *mapping)
2870 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2872 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2874 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2876 #define shmem_vm_ops generic_file_vm_ops
2877 #define shmem_file_operations ramfs_file_operations
2878 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2879 #define shmem_acct_size(flags, size) 0
2880 #define shmem_unacct_size(flags, size) do {} while (0)
2882 #endif /* CONFIG_SHMEM */
2887 * shmem_file_setup - get an unlinked file living in tmpfs
2888 * @name: name for dentry (to be seen in /proc/<pid>/maps
2889 * @size: size to be set for the file
2890 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2892 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2896 struct inode *inode;
2898 struct dentry *root;
2901 if (IS_ERR(shm_mnt))
2902 return (void *)shm_mnt;
2904 if (size < 0 || size > MAX_LFS_FILESIZE)
2905 return ERR_PTR(-EINVAL);
2907 if (shmem_acct_size(flags, size))
2908 return ERR_PTR(-ENOMEM);
2912 this.len = strlen(name);
2913 this.hash = 0; /* will go */
2914 root = shm_mnt->mnt_root;
2915 path.dentry = d_alloc(root, &this);
2918 path.mnt = mntget(shm_mnt);
2921 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2925 d_instantiate(path.dentry, inode);
2926 inode->i_size = size;
2927 clear_nlink(inode); /* It is unlinked */
2929 error = ramfs_nommu_expand_for_mapping(inode, size);
2935 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2936 &shmem_file_operations);
2945 shmem_unacct_size(flags, size);
2946 return ERR_PTR(error);
2948 EXPORT_SYMBOL_GPL(shmem_file_setup);
2951 * shmem_zero_setup - setup a shared anonymous mapping
2952 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2954 int shmem_zero_setup(struct vm_area_struct *vma)
2957 loff_t size = vma->vm_end - vma->vm_start;
2959 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2961 return PTR_ERR(file);
2965 vma->vm_file = file;
2966 vma->vm_ops = &shmem_vm_ops;
2967 vma->vm_flags |= VM_CAN_NONLINEAR;
2972 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2973 * @mapping: the page's address_space
2974 * @index: the page index
2975 * @gfp: the page allocator flags to use if allocating
2977 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2978 * with any new page allocations done using the specified allocation flags.
2979 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2980 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2981 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2983 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2984 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2986 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2987 pgoff_t index, gfp_t gfp)
2990 struct inode *inode = mapping->host;
2994 BUG_ON(mapping->a_ops != &shmem_aops);
2995 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2997 page = ERR_PTR(error);
3003 * The tiny !SHMEM case uses ramfs without swap
3005 return read_cache_page_gfp(mapping, index, gfp);
3008 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);