2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2005 Hugh Dickins.
10 * Copyright (C) 2002-2005 VERITAS Software Corporation.
11 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 * Extended attribute support for tmpfs:
14 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
15 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
18 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 * This file is released under the GPL.
24 #include <linux/init.h>
25 #include <linux/vfs.h>
26 #include <linux/mount.h>
27 #include <linux/pagemap.h>
28 #include <linux/file.h>
30 #include <linux/module.h>
31 #include <linux/swap.h>
33 static struct vfsmount *shm_mnt;
37 * This virtual memory filesystem is heavily based on the ramfs. It
38 * extends ramfs by the ability to use swap and honor resource limits
39 * which makes it a completely usable filesystem.
42 #include <linux/xattr.h>
43 #include <linux/exportfs.h>
44 #include <linux/posix_acl.h>
45 #include <linux/generic_acl.h>
46 #include <linux/mman.h>
47 #include <linux/string.h>
48 #include <linux/slab.h>
49 #include <linux/backing-dev.h>
50 #include <linux/shmem_fs.h>
51 #include <linux/writeback.h>
52 #include <linux/blkdev.h>
53 #include <linux/pagevec.h>
54 #include <linux/percpu_counter.h>
55 #include <linux/splice.h>
56 #include <linux/security.h>
57 #include <linux/swapops.h>
58 #include <linux/mempolicy.h>
59 #include <linux/namei.h>
60 #include <linux/ctype.h>
61 #include <linux/migrate.h>
62 #include <linux/highmem.h>
63 #include <linux/seq_file.h>
64 #include <linux/magic.h>
66 #include <asm/uaccess.h>
67 #include <asm/pgtable.h>
69 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
70 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
72 /* Pretend that each entry is of this size in directory's i_size */
73 #define BOGO_DIRENT_SIZE 20
76 struct list_head list; /* anchored by shmem_inode_info->xattr_list */
77 char *name; /* xattr name */
82 /* Flag allocation requirements to shmem_getpage */
84 SGP_READ, /* don't exceed i_size, don't allocate page */
85 SGP_CACHE, /* don't exceed i_size, may allocate page */
86 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
87 SGP_WRITE, /* may exceed i_size, may allocate page */
91 static unsigned long shmem_default_max_blocks(void)
93 return totalram_pages / 2;
96 static unsigned long shmem_default_max_inodes(void)
98 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
102 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
103 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
105 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
106 struct page **pagep, enum sgp_type sgp, int *fault_type)
108 return shmem_getpage_gfp(inode, index, pagep, sgp,
109 mapping_gfp_mask(inode->i_mapping), fault_type);
112 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
114 return sb->s_fs_info;
118 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
119 * for shared memory and for shared anonymous (/dev/zero) mappings
120 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
121 * consistent with the pre-accounting of private mappings ...
123 static inline int shmem_acct_size(unsigned long flags, loff_t size)
125 return (flags & VM_NORESERVE) ?
126 0 : security_vm_enough_memory_kern(VM_ACCT(size));
129 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
131 if (!(flags & VM_NORESERVE))
132 vm_unacct_memory(VM_ACCT(size));
136 * ... whereas tmpfs objects are accounted incrementally as
137 * pages are allocated, in order to allow huge sparse files.
138 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
139 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
141 static inline int shmem_acct_block(unsigned long flags)
143 return (flags & VM_NORESERVE) ?
144 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
147 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
149 if (flags & VM_NORESERVE)
150 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
153 static const struct super_operations shmem_ops;
154 static const struct address_space_operations shmem_aops;
155 static const struct file_operations shmem_file_operations;
156 static const struct inode_operations shmem_inode_operations;
157 static const struct inode_operations shmem_dir_inode_operations;
158 static const struct inode_operations shmem_special_inode_operations;
159 static const struct vm_operations_struct shmem_vm_ops;
161 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
162 .ra_pages = 0, /* No readahead */
163 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
166 static LIST_HEAD(shmem_swaplist);
167 static DEFINE_MUTEX(shmem_swaplist_mutex);
169 static void shmem_free_blocks(struct inode *inode, long pages)
171 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
172 if (sbinfo->max_blocks) {
173 percpu_counter_add(&sbinfo->used_blocks, -pages);
174 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
178 static int shmem_reserve_inode(struct super_block *sb)
180 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
181 if (sbinfo->max_inodes) {
182 spin_lock(&sbinfo->stat_lock);
183 if (!sbinfo->free_inodes) {
184 spin_unlock(&sbinfo->stat_lock);
187 sbinfo->free_inodes--;
188 spin_unlock(&sbinfo->stat_lock);
193 static void shmem_free_inode(struct super_block *sb)
195 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
196 if (sbinfo->max_inodes) {
197 spin_lock(&sbinfo->stat_lock);
198 sbinfo->free_inodes++;
199 spin_unlock(&sbinfo->stat_lock);
204 * shmem_recalc_inode - recalculate the block usage of an inode
205 * @inode: inode to recalc
207 * We have to calculate the free blocks since the mm can drop
208 * undirtied hole pages behind our back.
210 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
211 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
213 * It has to be called with the spinlock held.
215 static void shmem_recalc_inode(struct inode *inode)
217 struct shmem_inode_info *info = SHMEM_I(inode);
220 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
222 info->alloced -= freed;
223 shmem_unacct_blocks(info->flags, freed);
224 shmem_free_blocks(inode, freed);
228 static void shmem_put_swap(struct shmem_inode_info *info, pgoff_t index,
231 if (index < SHMEM_NR_DIRECT)
232 info->i_direct[index] = swap;
235 static swp_entry_t shmem_get_swap(struct shmem_inode_info *info, pgoff_t index)
237 return (index < SHMEM_NR_DIRECT) ?
238 info->i_direct[index] : (swp_entry_t){0};
241 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
243 struct address_space *mapping = inode->i_mapping;
244 struct shmem_inode_info *info = SHMEM_I(inode);
245 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
246 unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
247 pgoff_t end = (lend >> PAGE_CACHE_SHIFT);
253 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
255 pagevec_init(&pvec, 0);
257 while (index <= end && pagevec_lookup(&pvec, mapping, index,
258 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
259 mem_cgroup_uncharge_start();
260 for (i = 0; i < pagevec_count(&pvec); i++) {
261 struct page *page = pvec.pages[i];
263 /* We rely upon deletion not changing page->index */
268 if (!trylock_page(page))
270 WARN_ON(page->index != index);
271 if (PageWriteback(page)) {
275 truncate_inode_page(mapping, page);
278 pagevec_release(&pvec);
279 mem_cgroup_uncharge_end();
285 struct page *page = NULL;
286 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
288 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
289 set_page_dirty(page);
291 page_cache_release(page);
298 if (!pagevec_lookup(&pvec, mapping, index,
299 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
305 if (index == start && pvec.pages[0]->index > end) {
306 pagevec_release(&pvec);
309 mem_cgroup_uncharge_start();
310 for (i = 0; i < pagevec_count(&pvec); i++) {
311 struct page *page = pvec.pages[i];
313 /* We rely upon deletion not changing page->index */
319 WARN_ON(page->index != index);
320 wait_on_page_writeback(page);
321 truncate_inode_page(mapping, page);
324 pagevec_release(&pvec);
325 mem_cgroup_uncharge_end();
329 if (end > SHMEM_NR_DIRECT)
330 end = SHMEM_NR_DIRECT;
332 spin_lock(&info->lock);
333 for (index = start; index < end; index++) {
334 swap = shmem_get_swap(info, index);
336 free_swap_and_cache(swap);
337 shmem_put_swap(info, index, (swp_entry_t){0});
342 if (mapping->nrpages) {
343 spin_unlock(&info->lock);
345 * A page may have meanwhile sneaked in from swap.
347 truncate_inode_pages_range(mapping, lstart, lend);
348 spin_lock(&info->lock);
351 shmem_recalc_inode(inode);
352 spin_unlock(&info->lock);
354 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
356 EXPORT_SYMBOL_GPL(shmem_truncate_range);
358 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
360 struct inode *inode = dentry->d_inode;
363 error = inode_change_ok(inode, attr);
367 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
368 loff_t oldsize = inode->i_size;
369 loff_t newsize = attr->ia_size;
371 if (newsize != oldsize) {
372 i_size_write(inode, newsize);
373 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
375 if (newsize < oldsize) {
376 loff_t holebegin = round_up(newsize, PAGE_SIZE);
377 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
378 shmem_truncate_range(inode, newsize, (loff_t)-1);
379 /* unmap again to remove racily COWed private pages */
380 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
384 setattr_copy(inode, attr);
385 #ifdef CONFIG_TMPFS_POSIX_ACL
386 if (attr->ia_valid & ATTR_MODE)
387 error = generic_acl_chmod(inode);
392 static void shmem_evict_inode(struct inode *inode)
394 struct shmem_inode_info *info = SHMEM_I(inode);
395 struct shmem_xattr *xattr, *nxattr;
397 if (inode->i_mapping->a_ops == &shmem_aops) {
398 shmem_unacct_size(info->flags, inode->i_size);
400 shmem_truncate_range(inode, 0, (loff_t)-1);
401 if (!list_empty(&info->swaplist)) {
402 mutex_lock(&shmem_swaplist_mutex);
403 list_del_init(&info->swaplist);
404 mutex_unlock(&shmem_swaplist_mutex);
408 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
412 BUG_ON(inode->i_blocks);
413 shmem_free_inode(inode->i_sb);
414 end_writeback(inode);
417 static int shmem_unuse_inode(struct shmem_inode_info *info,
418 swp_entry_t swap, struct page *page)
420 struct address_space *mapping = info->vfs_inode.i_mapping;
424 for (index = 0; index < SHMEM_NR_DIRECT; index++)
425 if (shmem_get_swap(info, index).val == swap.val)
429 spin_lock(&info->lock);
430 if (shmem_get_swap(info, index).val != swap.val) {
431 spin_unlock(&info->lock);
436 * Move _head_ to start search for next from here.
437 * But be careful: shmem_evict_inode checks list_empty without taking
438 * mutex, and there's an instant in list_move_tail when info->swaplist
439 * would appear empty, if it were the only one on shmem_swaplist.
441 if (shmem_swaplist.next != &info->swaplist)
442 list_move_tail(&shmem_swaplist, &info->swaplist);
445 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
446 * but also to hold up shmem_evict_inode(): so inode cannot be freed
447 * beneath us (pagelock doesn't help until the page is in pagecache).
449 error = add_to_page_cache_locked(page, mapping, index, GFP_NOWAIT);
450 /* which does mem_cgroup_uncharge_cache_page on error */
452 if (error != -ENOMEM) {
453 delete_from_swap_cache(page);
454 set_page_dirty(page);
455 shmem_put_swap(info, index, (swp_entry_t){0});
458 error = 1; /* not an error, but entry was found */
460 spin_unlock(&info->lock);
465 * shmem_unuse() search for an eventually swapped out shmem page.
467 int shmem_unuse(swp_entry_t swap, struct page *page)
469 struct list_head *this, *next;
470 struct shmem_inode_info *info;
475 * Charge page using GFP_KERNEL while we can wait, before taking
476 * the shmem_swaplist_mutex which might hold up shmem_writepage().
477 * Charged back to the user (not to caller) when swap account is used.
478 * add_to_page_cache() will be called with GFP_NOWAIT.
480 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
484 * Try to preload while we can wait, to not make a habit of
485 * draining atomic reserves; but don't latch on to this cpu,
486 * it's okay if sometimes we get rescheduled after this.
488 error = radix_tree_preload(GFP_KERNEL);
491 radix_tree_preload_end();
493 mutex_lock(&shmem_swaplist_mutex);
494 list_for_each_safe(this, next, &shmem_swaplist) {
495 info = list_entry(this, struct shmem_inode_info, swaplist);
496 if (!info->swapped) {
497 spin_lock(&info->lock);
499 list_del_init(&info->swaplist);
500 spin_unlock(&info->lock);
503 found = shmem_unuse_inode(info, swap, page);
508 mutex_unlock(&shmem_swaplist_mutex);
512 mem_cgroup_uncharge_cache_page(page);
517 page_cache_release(page);
522 * Move the page from the page cache to the swap cache.
524 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
526 struct shmem_inode_info *info;
527 swp_entry_t swap, oswap;
528 struct address_space *mapping;
532 BUG_ON(!PageLocked(page));
533 mapping = page->mapping;
535 inode = mapping->host;
536 info = SHMEM_I(inode);
537 if (info->flags & VM_LOCKED)
539 if (!total_swap_pages)
543 * shmem_backing_dev_info's capabilities prevent regular writeback or
544 * sync from ever calling shmem_writepage; but a stacking filesystem
545 * might use ->writepage of its underlying filesystem, in which case
546 * tmpfs should write out to swap only in response to memory pressure,
547 * and not for the writeback threads or sync.
549 if (!wbc->for_reclaim) {
550 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
555 * Just for this patch, we have a toy implementation,
556 * which can swap out only the first SHMEM_NR_DIRECT pages:
557 * for simple demonstration of where we need to think about swap.
559 if (index >= SHMEM_NR_DIRECT)
562 swap = get_swap_page();
567 * Add inode to shmem_unuse()'s list of swapped-out inodes,
568 * if it's not already there. Do it now because we cannot take
569 * mutex while holding spinlock, and must do so before the page
570 * is moved to swap cache, when its pagelock no longer protects
571 * the inode from eviction. But don't unlock the mutex until
572 * we've taken the spinlock, because shmem_unuse_inode() will
573 * prune a !swapped inode from the swaplist under both locks.
575 mutex_lock(&shmem_swaplist_mutex);
576 if (list_empty(&info->swaplist))
577 list_add_tail(&info->swaplist, &shmem_swaplist);
579 spin_lock(&info->lock);
580 mutex_unlock(&shmem_swaplist_mutex);
582 oswap = shmem_get_swap(info, index);
584 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
585 free_swap_and_cache(oswap);
586 shmem_put_swap(info, index, (swp_entry_t){0});
589 shmem_recalc_inode(inode);
591 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
592 delete_from_page_cache(page);
593 shmem_put_swap(info, index, swap);
595 swap_shmem_alloc(swap);
596 spin_unlock(&info->lock);
597 BUG_ON(page_mapped(page));
598 swap_writepage(page, wbc);
602 spin_unlock(&info->lock);
603 swapcache_free(swap, NULL);
605 set_page_dirty(page);
606 if (wbc->for_reclaim)
607 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
614 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
618 if (!mpol || mpol->mode == MPOL_DEFAULT)
619 return; /* show nothing */
621 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
623 seq_printf(seq, ",mpol=%s", buffer);
626 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
628 struct mempolicy *mpol = NULL;
630 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
633 spin_unlock(&sbinfo->stat_lock);
637 #endif /* CONFIG_TMPFS */
639 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
640 struct shmem_inode_info *info, pgoff_t index)
642 struct mempolicy mpol, *spol;
643 struct vm_area_struct pvma;
645 spol = mpol_cond_copy(&mpol,
646 mpol_shared_policy_lookup(&info->policy, index));
648 /* Create a pseudo vma that just contains the policy */
650 pvma.vm_pgoff = index;
652 pvma.vm_policy = spol;
653 return swapin_readahead(swap, gfp, &pvma, 0);
656 static struct page *shmem_alloc_page(gfp_t gfp,
657 struct shmem_inode_info *info, pgoff_t index)
659 struct vm_area_struct pvma;
661 /* Create a pseudo vma that just contains the policy */
663 pvma.vm_pgoff = index;
665 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
668 * alloc_page_vma() will drop the shared policy reference
670 return alloc_page_vma(gfp, &pvma, 0);
672 #else /* !CONFIG_NUMA */
674 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
677 #endif /* CONFIG_TMPFS */
679 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
680 struct shmem_inode_info *info, pgoff_t index)
682 return swapin_readahead(swap, gfp, NULL, 0);
685 static inline struct page *shmem_alloc_page(gfp_t gfp,
686 struct shmem_inode_info *info, pgoff_t index)
688 return alloc_page(gfp);
690 #endif /* CONFIG_NUMA */
692 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
693 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
700 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
702 * If we allocate a new one we do not mark it dirty. That's up to the
703 * vm. If we swap it in we mark it dirty since we also free the swap
704 * entry since a page cannot live in both the swap and page cache
706 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
707 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
709 struct address_space *mapping = inode->i_mapping;
710 struct shmem_inode_info *info = SHMEM_I(inode);
711 struct shmem_sb_info *sbinfo;
713 struct page *prealloc_page = NULL;
717 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
720 page = find_lock_page(mapping, index);
723 * Once we can get the page lock, it must be uptodate:
724 * if there were an error in reading back from swap,
725 * the page would not be inserted into the filecache.
727 BUG_ON(!PageUptodate(page));
732 * Try to preload while we can wait, to not make a habit of
733 * draining atomic reserves; but don't latch on to this cpu.
735 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
738 radix_tree_preload_end();
740 if (sgp != SGP_READ && !prealloc_page) {
741 prealloc_page = shmem_alloc_page(gfp, info, index);
743 SetPageSwapBacked(prealloc_page);
744 if (mem_cgroup_cache_charge(prealloc_page,
745 current->mm, GFP_KERNEL)) {
746 page_cache_release(prealloc_page);
747 prealloc_page = NULL;
752 spin_lock(&info->lock);
753 shmem_recalc_inode(inode);
754 swap = shmem_get_swap(info, index);
756 /* Look it up and read it in.. */
757 page = lookup_swap_cache(swap);
759 spin_unlock(&info->lock);
760 /* here we actually do the io */
762 *fault_type |= VM_FAULT_MAJOR;
763 page = shmem_swapin(swap, gfp, info, index);
765 swp_entry_t nswap = shmem_get_swap(info, index);
766 if (nswap.val == swap.val) {
772 wait_on_page_locked(page);
773 page_cache_release(page);
777 /* We have to do this with page locked to prevent races */
778 if (!trylock_page(page)) {
779 spin_unlock(&info->lock);
780 wait_on_page_locked(page);
781 page_cache_release(page);
784 if (PageWriteback(page)) {
785 spin_unlock(&info->lock);
786 wait_on_page_writeback(page);
788 page_cache_release(page);
791 if (!PageUptodate(page)) {
792 spin_unlock(&info->lock);
794 page_cache_release(page);
799 error = add_to_page_cache_locked(page, mapping,
802 spin_unlock(&info->lock);
803 if (error == -ENOMEM) {
805 * reclaim from proper memory cgroup and
806 * call memcg's OOM if needed.
808 error = mem_cgroup_shmem_charge_fallback(
809 page, current->mm, gfp);
812 page_cache_release(page);
817 page_cache_release(page);
821 delete_from_swap_cache(page);
822 shmem_put_swap(info, index, (swp_entry_t){0});
824 spin_unlock(&info->lock);
825 set_page_dirty(page);
828 } else if (sgp == SGP_READ) {
829 page = find_get_page(mapping, index);
830 if (page && !trylock_page(page)) {
831 spin_unlock(&info->lock);
832 wait_on_page_locked(page);
833 page_cache_release(page);
836 spin_unlock(&info->lock);
838 } else if (prealloc_page) {
839 sbinfo = SHMEM_SB(inode->i_sb);
840 if (sbinfo->max_blocks) {
841 if (percpu_counter_compare(&sbinfo->used_blocks,
842 sbinfo->max_blocks) >= 0 ||
843 shmem_acct_block(info->flags))
845 percpu_counter_inc(&sbinfo->used_blocks);
846 inode->i_blocks += BLOCKS_PER_PAGE;
847 } else if (shmem_acct_block(info->flags))
850 page = prealloc_page;
851 prealloc_page = NULL;
853 swap = shmem_get_swap(info, index);
855 mem_cgroup_uncharge_cache_page(page);
857 error = add_to_page_cache_lru(page, mapping,
860 * At add_to_page_cache_lru() failure,
861 * uncharge will be done automatically.
863 if (swap.val || error) {
864 shmem_unacct_blocks(info->flags, 1);
865 shmem_free_blocks(inode, 1);
866 spin_unlock(&info->lock);
867 page_cache_release(page);
872 spin_unlock(&info->lock);
873 clear_highpage(page);
874 flush_dcache_page(page);
875 SetPageUptodate(page);
876 if (sgp == SGP_DIRTY)
877 set_page_dirty(page);
880 spin_unlock(&info->lock);
889 mem_cgroup_uncharge_cache_page(prealloc_page);
890 page_cache_release(prealloc_page);
896 * Perhaps the page was brought in from swap between find_lock_page
897 * and taking info->lock? We allow for that at add_to_page_cache_lru,
898 * but must also avoid reporting a spurious ENOSPC while working on a
901 page = find_get_page(mapping, index);
902 spin_unlock(&info->lock);
904 page_cache_release(page);
911 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
913 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
915 int ret = VM_FAULT_LOCKED;
917 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
918 return VM_FAULT_SIGBUS;
920 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
922 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
924 if (ret & VM_FAULT_MAJOR) {
925 count_vm_event(PGMAJFAULT);
926 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
932 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
934 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
935 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
938 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
941 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
944 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
945 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
949 int shmem_lock(struct file *file, int lock, struct user_struct *user)
951 struct inode *inode = file->f_path.dentry->d_inode;
952 struct shmem_inode_info *info = SHMEM_I(inode);
953 int retval = -ENOMEM;
955 spin_lock(&info->lock);
956 if (lock && !(info->flags & VM_LOCKED)) {
957 if (!user_shm_lock(inode->i_size, user))
959 info->flags |= VM_LOCKED;
960 mapping_set_unevictable(file->f_mapping);
962 if (!lock && (info->flags & VM_LOCKED) && user) {
963 user_shm_unlock(inode->i_size, user);
964 info->flags &= ~VM_LOCKED;
965 mapping_clear_unevictable(file->f_mapping);
966 scan_mapping_unevictable_pages(file->f_mapping);
971 spin_unlock(&info->lock);
975 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
978 vma->vm_ops = &shmem_vm_ops;
979 vma->vm_flags |= VM_CAN_NONLINEAR;
983 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
984 int mode, dev_t dev, unsigned long flags)
987 struct shmem_inode_info *info;
988 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
990 if (shmem_reserve_inode(sb))
993 inode = new_inode(sb);
995 inode->i_ino = get_next_ino();
996 inode_init_owner(inode, dir, mode);
998 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
999 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1000 inode->i_generation = get_seconds();
1001 info = SHMEM_I(inode);
1002 memset(info, 0, (char *)inode - (char *)info);
1003 spin_lock_init(&info->lock);
1004 info->flags = flags & VM_NORESERVE;
1005 INIT_LIST_HEAD(&info->swaplist);
1006 INIT_LIST_HEAD(&info->xattr_list);
1007 cache_no_acl(inode);
1009 switch (mode & S_IFMT) {
1011 inode->i_op = &shmem_special_inode_operations;
1012 init_special_inode(inode, mode, dev);
1015 inode->i_mapping->a_ops = &shmem_aops;
1016 inode->i_op = &shmem_inode_operations;
1017 inode->i_fop = &shmem_file_operations;
1018 mpol_shared_policy_init(&info->policy,
1019 shmem_get_sbmpol(sbinfo));
1023 /* Some things misbehave if size == 0 on a directory */
1024 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1025 inode->i_op = &shmem_dir_inode_operations;
1026 inode->i_fop = &simple_dir_operations;
1030 * Must not load anything in the rbtree,
1031 * mpol_free_shared_policy will not be called.
1033 mpol_shared_policy_init(&info->policy, NULL);
1037 shmem_free_inode(sb);
1042 static const struct inode_operations shmem_symlink_inode_operations;
1043 static const struct inode_operations shmem_symlink_inline_operations;
1046 shmem_write_begin(struct file *file, struct address_space *mapping,
1047 loff_t pos, unsigned len, unsigned flags,
1048 struct page **pagep, void **fsdata)
1050 struct inode *inode = mapping->host;
1051 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1052 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1056 shmem_write_end(struct file *file, struct address_space *mapping,
1057 loff_t pos, unsigned len, unsigned copied,
1058 struct page *page, void *fsdata)
1060 struct inode *inode = mapping->host;
1062 if (pos + copied > inode->i_size)
1063 i_size_write(inode, pos + copied);
1065 set_page_dirty(page);
1067 page_cache_release(page);
1072 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1074 struct inode *inode = filp->f_path.dentry->d_inode;
1075 struct address_space *mapping = inode->i_mapping;
1077 unsigned long offset;
1078 enum sgp_type sgp = SGP_READ;
1081 * Might this read be for a stacking filesystem? Then when reading
1082 * holes of a sparse file, we actually need to allocate those pages,
1083 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1085 if (segment_eq(get_fs(), KERNEL_DS))
1088 index = *ppos >> PAGE_CACHE_SHIFT;
1089 offset = *ppos & ~PAGE_CACHE_MASK;
1092 struct page *page = NULL;
1094 unsigned long nr, ret;
1095 loff_t i_size = i_size_read(inode);
1097 end_index = i_size >> PAGE_CACHE_SHIFT;
1098 if (index > end_index)
1100 if (index == end_index) {
1101 nr = i_size & ~PAGE_CACHE_MASK;
1106 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1108 if (desc->error == -EINVAL)
1116 * We must evaluate after, since reads (unlike writes)
1117 * are called without i_mutex protection against truncate
1119 nr = PAGE_CACHE_SIZE;
1120 i_size = i_size_read(inode);
1121 end_index = i_size >> PAGE_CACHE_SHIFT;
1122 if (index == end_index) {
1123 nr = i_size & ~PAGE_CACHE_MASK;
1126 page_cache_release(page);
1134 * If users can be writing to this page using arbitrary
1135 * virtual addresses, take care about potential aliasing
1136 * before reading the page on the kernel side.
1138 if (mapping_writably_mapped(mapping))
1139 flush_dcache_page(page);
1141 * Mark the page accessed if we read the beginning.
1144 mark_page_accessed(page);
1146 page = ZERO_PAGE(0);
1147 page_cache_get(page);
1151 * Ok, we have the page, and it's up-to-date, so
1152 * now we can copy it to user space...
1154 * The actor routine returns how many bytes were actually used..
1155 * NOTE! This may not be the same as how much of a user buffer
1156 * we filled up (we may be padding etc), so we can only update
1157 * "pos" here (the actor routine has to update the user buffer
1158 * pointers and the remaining count).
1160 ret = actor(desc, page, offset, nr);
1162 index += offset >> PAGE_CACHE_SHIFT;
1163 offset &= ~PAGE_CACHE_MASK;
1165 page_cache_release(page);
1166 if (ret != nr || !desc->count)
1172 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1173 file_accessed(filp);
1176 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1177 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1179 struct file *filp = iocb->ki_filp;
1183 loff_t *ppos = &iocb->ki_pos;
1185 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1189 for (seg = 0; seg < nr_segs; seg++) {
1190 read_descriptor_t desc;
1193 desc.arg.buf = iov[seg].iov_base;
1194 desc.count = iov[seg].iov_len;
1195 if (desc.count == 0)
1198 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1199 retval += desc.written;
1201 retval = retval ?: desc.error;
1210 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1211 struct pipe_inode_info *pipe, size_t len,
1214 struct address_space *mapping = in->f_mapping;
1215 struct inode *inode = mapping->host;
1216 unsigned int loff, nr_pages, req_pages;
1217 struct page *pages[PIPE_DEF_BUFFERS];
1218 struct partial_page partial[PIPE_DEF_BUFFERS];
1220 pgoff_t index, end_index;
1223 struct splice_pipe_desc spd = {
1227 .ops = &page_cache_pipe_buf_ops,
1228 .spd_release = spd_release_page,
1231 isize = i_size_read(inode);
1232 if (unlikely(*ppos >= isize))
1235 left = isize - *ppos;
1236 if (unlikely(left < len))
1239 if (splice_grow_spd(pipe, &spd))
1242 index = *ppos >> PAGE_CACHE_SHIFT;
1243 loff = *ppos & ~PAGE_CACHE_MASK;
1244 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1245 nr_pages = min(req_pages, pipe->buffers);
1247 spd.nr_pages = find_get_pages_contig(mapping, index,
1248 nr_pages, spd.pages);
1249 index += spd.nr_pages;
1252 while (spd.nr_pages < nr_pages) {
1253 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1257 spd.pages[spd.nr_pages++] = page;
1261 index = *ppos >> PAGE_CACHE_SHIFT;
1262 nr_pages = spd.nr_pages;
1265 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1266 unsigned int this_len;
1271 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1272 page = spd.pages[page_nr];
1274 if (!PageUptodate(page) || page->mapping != mapping) {
1275 error = shmem_getpage(inode, index, &page,
1280 page_cache_release(spd.pages[page_nr]);
1281 spd.pages[page_nr] = page;
1284 isize = i_size_read(inode);
1285 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1286 if (unlikely(!isize || index > end_index))
1289 if (end_index == index) {
1292 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1296 this_len = min(this_len, plen - loff);
1300 spd.partial[page_nr].offset = loff;
1301 spd.partial[page_nr].len = this_len;
1308 while (page_nr < nr_pages)
1309 page_cache_release(spd.pages[page_nr++]);
1312 error = splice_to_pipe(pipe, &spd);
1314 splice_shrink_spd(pipe, &spd);
1323 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1325 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1327 buf->f_type = TMPFS_MAGIC;
1328 buf->f_bsize = PAGE_CACHE_SIZE;
1329 buf->f_namelen = NAME_MAX;
1330 if (sbinfo->max_blocks) {
1331 buf->f_blocks = sbinfo->max_blocks;
1333 buf->f_bfree = sbinfo->max_blocks -
1334 percpu_counter_sum(&sbinfo->used_blocks);
1336 if (sbinfo->max_inodes) {
1337 buf->f_files = sbinfo->max_inodes;
1338 buf->f_ffree = sbinfo->free_inodes;
1340 /* else leave those fields 0 like simple_statfs */
1345 * File creation. Allocate an inode, and we're done..
1348 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1350 struct inode *inode;
1351 int error = -ENOSPC;
1353 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1355 error = security_inode_init_security(inode, dir,
1356 &dentry->d_name, NULL,
1359 if (error != -EOPNOTSUPP) {
1364 #ifdef CONFIG_TMPFS_POSIX_ACL
1365 error = generic_acl_init(inode, dir);
1373 dir->i_size += BOGO_DIRENT_SIZE;
1374 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1375 d_instantiate(dentry, inode);
1376 dget(dentry); /* Extra count - pin the dentry in core */
1381 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1385 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1391 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1392 struct nameidata *nd)
1394 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1400 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1402 struct inode *inode = old_dentry->d_inode;
1406 * No ordinary (disk based) filesystem counts links as inodes;
1407 * but each new link needs a new dentry, pinning lowmem, and
1408 * tmpfs dentries cannot be pruned until they are unlinked.
1410 ret = shmem_reserve_inode(inode->i_sb);
1414 dir->i_size += BOGO_DIRENT_SIZE;
1415 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1417 ihold(inode); /* New dentry reference */
1418 dget(dentry); /* Extra pinning count for the created dentry */
1419 d_instantiate(dentry, inode);
1424 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1426 struct inode *inode = dentry->d_inode;
1428 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1429 shmem_free_inode(inode->i_sb);
1431 dir->i_size -= BOGO_DIRENT_SIZE;
1432 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1434 dput(dentry); /* Undo the count from "create" - this does all the work */
1438 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1440 if (!simple_empty(dentry))
1443 drop_nlink(dentry->d_inode);
1445 return shmem_unlink(dir, dentry);
1449 * The VFS layer already does all the dentry stuff for rename,
1450 * we just have to decrement the usage count for the target if
1451 * it exists so that the VFS layer correctly free's it when it
1454 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1456 struct inode *inode = old_dentry->d_inode;
1457 int they_are_dirs = S_ISDIR(inode->i_mode);
1459 if (!simple_empty(new_dentry))
1462 if (new_dentry->d_inode) {
1463 (void) shmem_unlink(new_dir, new_dentry);
1465 drop_nlink(old_dir);
1466 } else if (they_are_dirs) {
1467 drop_nlink(old_dir);
1471 old_dir->i_size -= BOGO_DIRENT_SIZE;
1472 new_dir->i_size += BOGO_DIRENT_SIZE;
1473 old_dir->i_ctime = old_dir->i_mtime =
1474 new_dir->i_ctime = new_dir->i_mtime =
1475 inode->i_ctime = CURRENT_TIME;
1479 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1483 struct inode *inode;
1486 struct shmem_inode_info *info;
1488 len = strlen(symname) + 1;
1489 if (len > PAGE_CACHE_SIZE)
1490 return -ENAMETOOLONG;
1492 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1496 error = security_inode_init_security(inode, dir, &dentry->d_name, NULL,
1499 if (error != -EOPNOTSUPP) {
1506 info = SHMEM_I(inode);
1507 inode->i_size = len-1;
1508 if (len <= SHMEM_SYMLINK_INLINE_LEN) {
1510 memcpy(info->inline_symlink, symname, len);
1511 inode->i_op = &shmem_symlink_inline_operations;
1513 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1518 inode->i_mapping->a_ops = &shmem_aops;
1519 inode->i_op = &shmem_symlink_inode_operations;
1520 kaddr = kmap_atomic(page, KM_USER0);
1521 memcpy(kaddr, symname, len);
1522 kunmap_atomic(kaddr, KM_USER0);
1523 set_page_dirty(page);
1525 page_cache_release(page);
1527 dir->i_size += BOGO_DIRENT_SIZE;
1528 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1529 d_instantiate(dentry, inode);
1534 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1536 nd_set_link(nd, SHMEM_I(dentry->d_inode)->inline_symlink);
1540 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1542 struct page *page = NULL;
1543 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1544 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
1550 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1552 if (!IS_ERR(nd_get_link(nd))) {
1553 struct page *page = cookie;
1555 mark_page_accessed(page);
1556 page_cache_release(page);
1560 #ifdef CONFIG_TMPFS_XATTR
1562 * Superblocks without xattr inode operations may get some security.* xattr
1563 * support from the LSM "for free". As soon as we have any other xattrs
1564 * like ACLs, we also need to implement the security.* handlers at
1565 * filesystem level, though.
1568 static int shmem_xattr_get(struct dentry *dentry, const char *name,
1569 void *buffer, size_t size)
1571 struct shmem_inode_info *info;
1572 struct shmem_xattr *xattr;
1575 info = SHMEM_I(dentry->d_inode);
1577 spin_lock(&info->lock);
1578 list_for_each_entry(xattr, &info->xattr_list, list) {
1579 if (strcmp(name, xattr->name))
1584 if (size < xattr->size)
1587 memcpy(buffer, xattr->value, xattr->size);
1591 spin_unlock(&info->lock);
1595 static int shmem_xattr_set(struct dentry *dentry, const char *name,
1596 const void *value, size_t size, int flags)
1598 struct inode *inode = dentry->d_inode;
1599 struct shmem_inode_info *info = SHMEM_I(inode);
1600 struct shmem_xattr *xattr;
1601 struct shmem_xattr *new_xattr = NULL;
1605 /* value == NULL means remove */
1608 len = sizeof(*new_xattr) + size;
1609 if (len <= sizeof(*new_xattr))
1612 new_xattr = kmalloc(len, GFP_KERNEL);
1616 new_xattr->name = kstrdup(name, GFP_KERNEL);
1617 if (!new_xattr->name) {
1622 new_xattr->size = size;
1623 memcpy(new_xattr->value, value, size);
1626 spin_lock(&info->lock);
1627 list_for_each_entry(xattr, &info->xattr_list, list) {
1628 if (!strcmp(name, xattr->name)) {
1629 if (flags & XATTR_CREATE) {
1632 } else if (new_xattr) {
1633 list_replace(&xattr->list, &new_xattr->list);
1635 list_del(&xattr->list);
1640 if (flags & XATTR_REPLACE) {
1644 list_add(&new_xattr->list, &info->xattr_list);
1648 spin_unlock(&info->lock);
1655 static const struct xattr_handler *shmem_xattr_handlers[] = {
1656 #ifdef CONFIG_TMPFS_POSIX_ACL
1657 &generic_acl_access_handler,
1658 &generic_acl_default_handler,
1663 static int shmem_xattr_validate(const char *name)
1665 struct { const char *prefix; size_t len; } arr[] = {
1666 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
1667 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
1671 for (i = 0; i < ARRAY_SIZE(arr); i++) {
1672 size_t preflen = arr[i].len;
1673 if (strncmp(name, arr[i].prefix, preflen) == 0) {
1682 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
1683 void *buffer, size_t size)
1688 * If this is a request for a synthetic attribute in the system.*
1689 * namespace use the generic infrastructure to resolve a handler
1690 * for it via sb->s_xattr.
1692 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1693 return generic_getxattr(dentry, name, buffer, size);
1695 err = shmem_xattr_validate(name);
1699 return shmem_xattr_get(dentry, name, buffer, size);
1702 static int shmem_setxattr(struct dentry *dentry, const char *name,
1703 const void *value, size_t size, int flags)
1708 * If this is a request for a synthetic attribute in the system.*
1709 * namespace use the generic infrastructure to resolve a handler
1710 * for it via sb->s_xattr.
1712 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1713 return generic_setxattr(dentry, name, value, size, flags);
1715 err = shmem_xattr_validate(name);
1720 value = ""; /* empty EA, do not remove */
1722 return shmem_xattr_set(dentry, name, value, size, flags);
1726 static int shmem_removexattr(struct dentry *dentry, const char *name)
1731 * If this is a request for a synthetic attribute in the system.*
1732 * namespace use the generic infrastructure to resolve a handler
1733 * for it via sb->s_xattr.
1735 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1736 return generic_removexattr(dentry, name);
1738 err = shmem_xattr_validate(name);
1742 return shmem_xattr_set(dentry, name, NULL, 0, XATTR_REPLACE);
1745 static bool xattr_is_trusted(const char *name)
1747 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
1750 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
1752 bool trusted = capable(CAP_SYS_ADMIN);
1753 struct shmem_xattr *xattr;
1754 struct shmem_inode_info *info;
1757 info = SHMEM_I(dentry->d_inode);
1759 spin_lock(&info->lock);
1760 list_for_each_entry(xattr, &info->xattr_list, list) {
1763 /* skip "trusted." attributes for unprivileged callers */
1764 if (!trusted && xattr_is_trusted(xattr->name))
1767 len = strlen(xattr->name) + 1;
1774 memcpy(buffer, xattr->name, len);
1778 spin_unlock(&info->lock);
1782 #endif /* CONFIG_TMPFS_XATTR */
1784 static const struct inode_operations shmem_symlink_inline_operations = {
1785 .readlink = generic_readlink,
1786 .follow_link = shmem_follow_link_inline,
1787 #ifdef CONFIG_TMPFS_XATTR
1788 .setxattr = shmem_setxattr,
1789 .getxattr = shmem_getxattr,
1790 .listxattr = shmem_listxattr,
1791 .removexattr = shmem_removexattr,
1795 static const struct inode_operations shmem_symlink_inode_operations = {
1796 .readlink = generic_readlink,
1797 .follow_link = shmem_follow_link,
1798 .put_link = shmem_put_link,
1799 #ifdef CONFIG_TMPFS_XATTR
1800 .setxattr = shmem_setxattr,
1801 .getxattr = shmem_getxattr,
1802 .listxattr = shmem_listxattr,
1803 .removexattr = shmem_removexattr,
1807 static struct dentry *shmem_get_parent(struct dentry *child)
1809 return ERR_PTR(-ESTALE);
1812 static int shmem_match(struct inode *ino, void *vfh)
1816 inum = (inum << 32) | fh[1];
1817 return ino->i_ino == inum && fh[0] == ino->i_generation;
1820 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
1821 struct fid *fid, int fh_len, int fh_type)
1823 struct inode *inode;
1824 struct dentry *dentry = NULL;
1825 u64 inum = fid->raw[2];
1826 inum = (inum << 32) | fid->raw[1];
1831 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
1832 shmem_match, fid->raw);
1834 dentry = d_find_alias(inode);
1841 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
1844 struct inode *inode = dentry->d_inode;
1851 if (inode_unhashed(inode)) {
1852 /* Unfortunately insert_inode_hash is not idempotent,
1853 * so as we hash inodes here rather than at creation
1854 * time, we need a lock to ensure we only try
1857 static DEFINE_SPINLOCK(lock);
1859 if (inode_unhashed(inode))
1860 __insert_inode_hash(inode,
1861 inode->i_ino + inode->i_generation);
1865 fh[0] = inode->i_generation;
1866 fh[1] = inode->i_ino;
1867 fh[2] = ((__u64)inode->i_ino) >> 32;
1873 static const struct export_operations shmem_export_ops = {
1874 .get_parent = shmem_get_parent,
1875 .encode_fh = shmem_encode_fh,
1876 .fh_to_dentry = shmem_fh_to_dentry,
1879 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
1882 char *this_char, *value, *rest;
1884 while (options != NULL) {
1885 this_char = options;
1888 * NUL-terminate this option: unfortunately,
1889 * mount options form a comma-separated list,
1890 * but mpol's nodelist may also contain commas.
1892 options = strchr(options, ',');
1893 if (options == NULL)
1896 if (!isdigit(*options)) {
1903 if ((value = strchr(this_char,'=')) != NULL) {
1907 "tmpfs: No value for mount option '%s'\n",
1912 if (!strcmp(this_char,"size")) {
1913 unsigned long long size;
1914 size = memparse(value,&rest);
1916 size <<= PAGE_SHIFT;
1917 size *= totalram_pages;
1923 sbinfo->max_blocks =
1924 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
1925 } else if (!strcmp(this_char,"nr_blocks")) {
1926 sbinfo->max_blocks = memparse(value, &rest);
1929 } else if (!strcmp(this_char,"nr_inodes")) {
1930 sbinfo->max_inodes = memparse(value, &rest);
1933 } else if (!strcmp(this_char,"mode")) {
1936 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
1939 } else if (!strcmp(this_char,"uid")) {
1942 sbinfo->uid = simple_strtoul(value, &rest, 0);
1945 } else if (!strcmp(this_char,"gid")) {
1948 sbinfo->gid = simple_strtoul(value, &rest, 0);
1951 } else if (!strcmp(this_char,"mpol")) {
1952 if (mpol_parse_str(value, &sbinfo->mpol, 1))
1955 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1963 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1969 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1971 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1972 struct shmem_sb_info config = *sbinfo;
1973 unsigned long inodes;
1974 int error = -EINVAL;
1976 if (shmem_parse_options(data, &config, true))
1979 spin_lock(&sbinfo->stat_lock);
1980 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
1981 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
1983 if (config.max_inodes < inodes)
1986 * Those tests also disallow limited->unlimited while any are in
1987 * use, so i_blocks will always be zero when max_blocks is zero;
1988 * but we must separately disallow unlimited->limited, because
1989 * in that case we have no record of how much is already in use.
1991 if (config.max_blocks && !sbinfo->max_blocks)
1993 if (config.max_inodes && !sbinfo->max_inodes)
1997 sbinfo->max_blocks = config.max_blocks;
1998 sbinfo->max_inodes = config.max_inodes;
1999 sbinfo->free_inodes = config.max_inodes - inodes;
2001 mpol_put(sbinfo->mpol);
2002 sbinfo->mpol = config.mpol; /* transfers initial ref */
2004 spin_unlock(&sbinfo->stat_lock);
2008 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2010 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2012 if (sbinfo->max_blocks != shmem_default_max_blocks())
2013 seq_printf(seq, ",size=%luk",
2014 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2015 if (sbinfo->max_inodes != shmem_default_max_inodes())
2016 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2017 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2018 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2019 if (sbinfo->uid != 0)
2020 seq_printf(seq, ",uid=%u", sbinfo->uid);
2021 if (sbinfo->gid != 0)
2022 seq_printf(seq, ",gid=%u", sbinfo->gid);
2023 shmem_show_mpol(seq, sbinfo->mpol);
2026 #endif /* CONFIG_TMPFS */
2028 static void shmem_put_super(struct super_block *sb)
2030 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2032 percpu_counter_destroy(&sbinfo->used_blocks);
2034 sb->s_fs_info = NULL;
2037 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2039 struct inode *inode;
2040 struct dentry *root;
2041 struct shmem_sb_info *sbinfo;
2044 /* Round up to L1_CACHE_BYTES to resist false sharing */
2045 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2046 L1_CACHE_BYTES), GFP_KERNEL);
2050 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2051 sbinfo->uid = current_fsuid();
2052 sbinfo->gid = current_fsgid();
2053 sb->s_fs_info = sbinfo;
2057 * Per default we only allow half of the physical ram per
2058 * tmpfs instance, limiting inodes to one per page of lowmem;
2059 * but the internal instance is left unlimited.
2061 if (!(sb->s_flags & MS_NOUSER)) {
2062 sbinfo->max_blocks = shmem_default_max_blocks();
2063 sbinfo->max_inodes = shmem_default_max_inodes();
2064 if (shmem_parse_options(data, sbinfo, false)) {
2069 sb->s_export_op = &shmem_export_ops;
2071 sb->s_flags |= MS_NOUSER;
2074 spin_lock_init(&sbinfo->stat_lock);
2075 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2077 sbinfo->free_inodes = sbinfo->max_inodes;
2079 sb->s_maxbytes = MAX_LFS_FILESIZE;
2080 sb->s_blocksize = PAGE_CACHE_SIZE;
2081 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2082 sb->s_magic = TMPFS_MAGIC;
2083 sb->s_op = &shmem_ops;
2084 sb->s_time_gran = 1;
2085 #ifdef CONFIG_TMPFS_XATTR
2086 sb->s_xattr = shmem_xattr_handlers;
2088 #ifdef CONFIG_TMPFS_POSIX_ACL
2089 sb->s_flags |= MS_POSIXACL;
2092 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2095 inode->i_uid = sbinfo->uid;
2096 inode->i_gid = sbinfo->gid;
2097 root = d_alloc_root(inode);
2106 shmem_put_super(sb);
2110 static struct kmem_cache *shmem_inode_cachep;
2112 static struct inode *shmem_alloc_inode(struct super_block *sb)
2114 struct shmem_inode_info *info;
2115 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2118 return &info->vfs_inode;
2121 static void shmem_destroy_callback(struct rcu_head *head)
2123 struct inode *inode = container_of(head, struct inode, i_rcu);
2124 INIT_LIST_HEAD(&inode->i_dentry);
2125 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2128 static void shmem_destroy_inode(struct inode *inode)
2130 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2131 /* only struct inode is valid if it's an inline symlink */
2132 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2134 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2137 static void shmem_init_inode(void *foo)
2139 struct shmem_inode_info *info = foo;
2140 inode_init_once(&info->vfs_inode);
2143 static int shmem_init_inodecache(void)
2145 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2146 sizeof(struct shmem_inode_info),
2147 0, SLAB_PANIC, shmem_init_inode);
2151 static void shmem_destroy_inodecache(void)
2153 kmem_cache_destroy(shmem_inode_cachep);
2156 static const struct address_space_operations shmem_aops = {
2157 .writepage = shmem_writepage,
2158 .set_page_dirty = __set_page_dirty_no_writeback,
2160 .write_begin = shmem_write_begin,
2161 .write_end = shmem_write_end,
2163 .migratepage = migrate_page,
2164 .error_remove_page = generic_error_remove_page,
2167 static const struct file_operations shmem_file_operations = {
2170 .llseek = generic_file_llseek,
2171 .read = do_sync_read,
2172 .write = do_sync_write,
2173 .aio_read = shmem_file_aio_read,
2174 .aio_write = generic_file_aio_write,
2175 .fsync = noop_fsync,
2176 .splice_read = shmem_file_splice_read,
2177 .splice_write = generic_file_splice_write,
2181 static const struct inode_operations shmem_inode_operations = {
2182 .setattr = shmem_setattr,
2183 .truncate_range = shmem_truncate_range,
2184 #ifdef CONFIG_TMPFS_XATTR
2185 .setxattr = shmem_setxattr,
2186 .getxattr = shmem_getxattr,
2187 .listxattr = shmem_listxattr,
2188 .removexattr = shmem_removexattr,
2192 static const struct inode_operations shmem_dir_inode_operations = {
2194 .create = shmem_create,
2195 .lookup = simple_lookup,
2197 .unlink = shmem_unlink,
2198 .symlink = shmem_symlink,
2199 .mkdir = shmem_mkdir,
2200 .rmdir = shmem_rmdir,
2201 .mknod = shmem_mknod,
2202 .rename = shmem_rename,
2204 #ifdef CONFIG_TMPFS_XATTR
2205 .setxattr = shmem_setxattr,
2206 .getxattr = shmem_getxattr,
2207 .listxattr = shmem_listxattr,
2208 .removexattr = shmem_removexattr,
2210 #ifdef CONFIG_TMPFS_POSIX_ACL
2211 .setattr = shmem_setattr,
2215 static const struct inode_operations shmem_special_inode_operations = {
2216 #ifdef CONFIG_TMPFS_XATTR
2217 .setxattr = shmem_setxattr,
2218 .getxattr = shmem_getxattr,
2219 .listxattr = shmem_listxattr,
2220 .removexattr = shmem_removexattr,
2222 #ifdef CONFIG_TMPFS_POSIX_ACL
2223 .setattr = shmem_setattr,
2227 static const struct super_operations shmem_ops = {
2228 .alloc_inode = shmem_alloc_inode,
2229 .destroy_inode = shmem_destroy_inode,
2231 .statfs = shmem_statfs,
2232 .remount_fs = shmem_remount_fs,
2233 .show_options = shmem_show_options,
2235 .evict_inode = shmem_evict_inode,
2236 .drop_inode = generic_delete_inode,
2237 .put_super = shmem_put_super,
2240 static const struct vm_operations_struct shmem_vm_ops = {
2241 .fault = shmem_fault,
2243 .set_policy = shmem_set_policy,
2244 .get_policy = shmem_get_policy,
2248 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2249 int flags, const char *dev_name, void *data)
2251 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2254 static struct file_system_type shmem_fs_type = {
2255 .owner = THIS_MODULE,
2257 .mount = shmem_mount,
2258 .kill_sb = kill_litter_super,
2261 int __init shmem_init(void)
2265 error = bdi_init(&shmem_backing_dev_info);
2269 error = shmem_init_inodecache();
2273 error = register_filesystem(&shmem_fs_type);
2275 printk(KERN_ERR "Could not register tmpfs\n");
2279 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2280 shmem_fs_type.name, NULL);
2281 if (IS_ERR(shm_mnt)) {
2282 error = PTR_ERR(shm_mnt);
2283 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2289 unregister_filesystem(&shmem_fs_type);
2291 shmem_destroy_inodecache();
2293 bdi_destroy(&shmem_backing_dev_info);
2295 shm_mnt = ERR_PTR(error);
2299 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2301 * mem_cgroup_get_shmem_target - find page or swap assigned to the shmem file
2302 * @inode: the inode to be searched
2303 * @index: the page offset to be searched
2304 * @pagep: the pointer for the found page to be stored
2305 * @swapp: the pointer for the found swap entry to be stored
2307 * If a page is found, refcount of it is incremented. Callers should handle
2310 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t index,
2311 struct page **pagep, swp_entry_t *swapp)
2313 struct shmem_inode_info *info = SHMEM_I(inode);
2314 struct page *page = NULL;
2315 swp_entry_t swap = {0};
2317 if ((index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2320 spin_lock(&info->lock);
2322 swap = shmem_get_swap(info, index);
2324 page = find_get_page(&swapper_space, swap.val);
2327 page = find_get_page(inode->i_mapping, index);
2328 spin_unlock(&info->lock);
2335 #else /* !CONFIG_SHMEM */
2338 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2340 * This is intended for small system where the benefits of the full
2341 * shmem code (swap-backed and resource-limited) are outweighed by
2342 * their complexity. On systems without swap this code should be
2343 * effectively equivalent, but much lighter weight.
2346 #include <linux/ramfs.h>
2348 static struct file_system_type shmem_fs_type = {
2350 .mount = ramfs_mount,
2351 .kill_sb = kill_litter_super,
2354 int __init shmem_init(void)
2356 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2358 shm_mnt = kern_mount(&shmem_fs_type);
2359 BUG_ON(IS_ERR(shm_mnt));
2364 int shmem_unuse(swp_entry_t swap, struct page *page)
2369 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2374 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2376 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2378 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2380 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2382 * mem_cgroup_get_shmem_target - find page or swap assigned to the shmem file
2383 * @inode: the inode to be searched
2384 * @index: the page offset to be searched
2385 * @pagep: the pointer for the found page to be stored
2386 * @swapp: the pointer for the found swap entry to be stored
2388 * If a page is found, refcount of it is incremented. Callers should handle
2391 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t index,
2392 struct page **pagep, swp_entry_t *swapp)
2394 struct page *page = NULL;
2396 if ((index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2398 page = find_get_page(inode->i_mapping, index);
2401 *swapp = (swp_entry_t){0};
2405 #define shmem_vm_ops generic_file_vm_ops
2406 #define shmem_file_operations ramfs_file_operations
2407 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2408 #define shmem_acct_size(flags, size) 0
2409 #define shmem_unacct_size(flags, size) do {} while (0)
2411 #endif /* CONFIG_SHMEM */
2416 * shmem_file_setup - get an unlinked file living in tmpfs
2417 * @name: name for dentry (to be seen in /proc/<pid>/maps
2418 * @size: size to be set for the file
2419 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2421 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2425 struct inode *inode;
2427 struct dentry *root;
2430 if (IS_ERR(shm_mnt))
2431 return (void *)shm_mnt;
2433 if (size < 0 || size > MAX_LFS_FILESIZE)
2434 return ERR_PTR(-EINVAL);
2436 if (shmem_acct_size(flags, size))
2437 return ERR_PTR(-ENOMEM);
2441 this.len = strlen(name);
2442 this.hash = 0; /* will go */
2443 root = shm_mnt->mnt_root;
2444 path.dentry = d_alloc(root, &this);
2447 path.mnt = mntget(shm_mnt);
2450 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2454 d_instantiate(path.dentry, inode);
2455 inode->i_size = size;
2456 inode->i_nlink = 0; /* It is unlinked */
2458 error = ramfs_nommu_expand_for_mapping(inode, size);
2464 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2465 &shmem_file_operations);
2474 shmem_unacct_size(flags, size);
2475 return ERR_PTR(error);
2477 EXPORT_SYMBOL_GPL(shmem_file_setup);
2480 * shmem_zero_setup - setup a shared anonymous mapping
2481 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2483 int shmem_zero_setup(struct vm_area_struct *vma)
2486 loff_t size = vma->vm_end - vma->vm_start;
2488 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2490 return PTR_ERR(file);
2494 vma->vm_file = file;
2495 vma->vm_ops = &shmem_vm_ops;
2496 vma->vm_flags |= VM_CAN_NONLINEAR;
2501 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2502 * @mapping: the page's address_space
2503 * @index: the page index
2504 * @gfp: the page allocator flags to use if allocating
2506 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2507 * with any new page allocations done using the specified allocation flags.
2508 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2509 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2510 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2512 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2513 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2515 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2516 pgoff_t index, gfp_t gfp)
2519 struct inode *inode = mapping->host;
2523 BUG_ON(mapping->a_ops != &shmem_aops);
2524 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2526 page = ERR_PTR(error);
2532 * The tiny !SHMEM case uses ramfs without swap
2534 return read_cache_page_gfp(mapping, index, gfp);
2537 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);