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/splice.h>
57 #include <linux/security.h>
58 #include <linux/swapops.h>
59 #include <linux/mempolicy.h>
60 #include <linux/namei.h>
61 #include <linux/ctype.h>
62 #include <linux/migrate.h>
63 #include <linux/highmem.h>
64 #include <linux/seq_file.h>
65 #include <linux/magic.h>
67 #include <asm/uaccess.h>
68 #include <asm/pgtable.h>
70 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
71 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
73 /* Pretend that each entry is of this size in directory's i_size */
74 #define BOGO_DIRENT_SIZE 20
76 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
77 #define SHORT_SYMLINK_LEN 128
80 * vmtruncate_range() communicates with shmem_fault via
81 * inode->i_private (with i_mutex making sure that it has only one user at
82 * a time): we would prefer not to enlarge the shmem inode just for that.
85 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
86 pgoff_t start; /* start of range currently being fallocated */
87 pgoff_t next; /* the next page offset to be fallocated */
91 struct list_head list; /* anchored by shmem_inode_info->xattr_list */
92 char *name; /* xattr name */
97 /* Flag allocation requirements to shmem_getpage */
99 SGP_READ, /* don't exceed i_size, don't allocate page */
100 SGP_CACHE, /* don't exceed i_size, may allocate page */
101 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
102 SGP_WRITE, /* may exceed i_size, may allocate page */
106 static unsigned long shmem_default_max_blocks(void)
108 return totalram_pages / 2;
111 static unsigned long shmem_default_max_inodes(void)
113 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
117 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
118 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
120 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
121 struct page **pagep, enum sgp_type sgp, int *fault_type)
123 return shmem_getpage_gfp(inode, index, pagep, sgp,
124 mapping_gfp_mask(inode->i_mapping), fault_type);
127 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
129 return sb->s_fs_info;
133 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
134 * for shared memory and for shared anonymous (/dev/zero) mappings
135 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
136 * consistent with the pre-accounting of private mappings ...
138 static inline int shmem_acct_size(unsigned long flags, loff_t size)
140 return (flags & VM_NORESERVE) ?
141 0 : security_vm_enough_memory_kern(VM_ACCT(size));
144 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
146 if (!(flags & VM_NORESERVE))
147 vm_unacct_memory(VM_ACCT(size));
151 * ... whereas tmpfs objects are accounted incrementally as
152 * pages are allocated, in order to allow huge sparse files.
153 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
154 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
156 static inline int shmem_acct_block(unsigned long flags)
158 return (flags & VM_NORESERVE) ?
159 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
162 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
164 if (flags & VM_NORESERVE)
165 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
168 static const struct super_operations shmem_ops;
169 static const struct address_space_operations shmem_aops;
170 static const struct file_operations shmem_file_operations;
171 static const struct inode_operations shmem_inode_operations;
172 static const struct inode_operations shmem_dir_inode_operations;
173 static const struct inode_operations shmem_special_inode_operations;
174 static const struct vm_operations_struct shmem_vm_ops;
176 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
177 .ra_pages = 0, /* No readahead */
178 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
181 static LIST_HEAD(shmem_swaplist);
182 static DEFINE_MUTEX(shmem_swaplist_mutex);
184 static int shmem_reserve_inode(struct super_block *sb)
186 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
187 if (sbinfo->max_inodes) {
188 spin_lock(&sbinfo->stat_lock);
189 if (!sbinfo->free_inodes) {
190 spin_unlock(&sbinfo->stat_lock);
193 sbinfo->free_inodes--;
194 spin_unlock(&sbinfo->stat_lock);
199 static void shmem_free_inode(struct super_block *sb)
201 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
202 if (sbinfo->max_inodes) {
203 spin_lock(&sbinfo->stat_lock);
204 sbinfo->free_inodes++;
205 spin_unlock(&sbinfo->stat_lock);
210 * shmem_recalc_inode - recalculate the block usage of an inode
211 * @inode: inode to recalc
213 * We have to calculate the free blocks since the mm can drop
214 * undirtied hole pages behind our back.
216 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
217 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
219 * It has to be called with the spinlock held.
221 static void shmem_recalc_inode(struct inode *inode)
223 struct shmem_inode_info *info = SHMEM_I(inode);
226 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
228 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
229 if (sbinfo->max_blocks)
230 percpu_counter_add(&sbinfo->used_blocks, -freed);
231 info->alloced -= freed;
232 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
233 shmem_unacct_blocks(info->flags, freed);
238 * Replace item expected in radix tree by a new item, while holding tree lock.
240 static int shmem_radix_tree_replace(struct address_space *mapping,
241 pgoff_t index, void *expected, void *replacement)
246 VM_BUG_ON(!expected);
247 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
249 item = radix_tree_deref_slot_protected(pslot,
250 &mapping->tree_lock);
251 if (item != expected)
254 radix_tree_replace_slot(pslot, replacement);
256 radix_tree_delete(&mapping->page_tree, index);
261 * Like add_to_page_cache_locked, but error if expected item has gone.
263 static int shmem_add_to_page_cache(struct page *page,
264 struct address_space *mapping,
265 pgoff_t index, gfp_t gfp, void *expected)
269 VM_BUG_ON(!PageLocked(page));
270 VM_BUG_ON(!PageSwapBacked(page));
273 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
275 page_cache_get(page);
276 page->mapping = mapping;
279 spin_lock_irq(&mapping->tree_lock);
281 error = radix_tree_insert(&mapping->page_tree,
284 error = shmem_radix_tree_replace(mapping, index,
288 __inc_zone_page_state(page, NR_FILE_PAGES);
289 __inc_zone_page_state(page, NR_SHMEM);
290 spin_unlock_irq(&mapping->tree_lock);
292 page->mapping = NULL;
293 spin_unlock_irq(&mapping->tree_lock);
294 page_cache_release(page);
297 radix_tree_preload_end();
300 mem_cgroup_uncharge_cache_page(page);
305 * Like delete_from_page_cache, but substitutes swap for page.
307 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
309 struct address_space *mapping = page->mapping;
312 spin_lock_irq(&mapping->tree_lock);
313 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
314 page->mapping = NULL;
316 __dec_zone_page_state(page, NR_FILE_PAGES);
317 __dec_zone_page_state(page, NR_SHMEM);
318 spin_unlock_irq(&mapping->tree_lock);
319 page_cache_release(page);
324 * Like find_get_pages, but collecting swap entries as well as pages.
326 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
327 pgoff_t start, unsigned int nr_pages,
328 struct page **pages, pgoff_t *indices)
332 unsigned int nr_found;
336 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
337 (void ***)pages, indices, start, nr_pages);
339 for (i = 0; i < nr_found; i++) {
342 page = radix_tree_deref_slot((void **)pages[i]);
345 if (radix_tree_exception(page)) {
346 if (radix_tree_deref_retry(page))
349 * Otherwise, we must be storing a swap entry
350 * here as an exceptional entry: so return it
351 * without attempting to raise page count.
355 if (!page_cache_get_speculative(page))
358 /* Has the page moved? */
359 if (unlikely(page != *((void **)pages[i]))) {
360 page_cache_release(page);
364 indices[ret] = indices[i];
368 if (unlikely(!ret && nr_found))
375 * Remove swap entry from radix tree, free the swap and its page cache.
377 static int shmem_free_swap(struct address_space *mapping,
378 pgoff_t index, void *radswap)
382 spin_lock_irq(&mapping->tree_lock);
383 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
384 spin_unlock_irq(&mapping->tree_lock);
386 free_swap_and_cache(radix_to_swp_entry(radswap));
391 * Pagevec may contain swap entries, so shuffle up pages before releasing.
393 static void shmem_deswap_pagevec(struct pagevec *pvec)
397 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
398 struct page *page = pvec->pages[i];
399 if (!radix_tree_exceptional_entry(page))
400 pvec->pages[j++] = page;
406 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
408 void shmem_unlock_mapping(struct address_space *mapping)
411 pgoff_t indices[PAGEVEC_SIZE];
414 pagevec_init(&pvec, 0);
416 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
418 while (!mapping_unevictable(mapping)) {
420 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
421 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
423 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
424 PAGEVEC_SIZE, pvec.pages, indices);
427 index = indices[pvec.nr - 1] + 1;
428 shmem_deswap_pagevec(&pvec);
429 check_move_unevictable_pages(pvec.pages, pvec.nr);
430 pagevec_release(&pvec);
436 * Remove range of pages and swap entries from radix tree, and free them.
438 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
440 struct address_space *mapping = inode->i_mapping;
441 struct shmem_inode_info *info = SHMEM_I(inode);
442 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
443 unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
444 pgoff_t end = (lend >> PAGE_CACHE_SHIFT);
446 pgoff_t indices[PAGEVEC_SIZE];
447 long nr_swaps_freed = 0;
451 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
453 pagevec_init(&pvec, 0);
455 while (index <= end) {
456 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
457 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
458 pvec.pages, indices);
461 mem_cgroup_uncharge_start();
462 for (i = 0; i < pagevec_count(&pvec); i++) {
463 struct page *page = pvec.pages[i];
469 if (radix_tree_exceptional_entry(page)) {
470 nr_swaps_freed += !shmem_free_swap(mapping,
475 if (!trylock_page(page))
477 if (page->mapping == mapping) {
478 VM_BUG_ON(PageWriteback(page));
479 truncate_inode_page(mapping, page);
483 shmem_deswap_pagevec(&pvec);
484 pagevec_release(&pvec);
485 mem_cgroup_uncharge_end();
491 struct page *page = NULL;
492 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
494 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
495 set_page_dirty(page);
497 page_cache_release(page);
502 while (index <= end) {
504 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
505 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
506 pvec.pages, indices);
508 /* If all gone or hole-punch, we're done */
509 if (index == start || end != -1)
511 /* But if truncating, restart to make sure all gone */
515 mem_cgroup_uncharge_start();
516 for (i = 0; i < pagevec_count(&pvec); i++) {
517 struct page *page = pvec.pages[i];
523 if (radix_tree_exceptional_entry(page)) {
524 if (shmem_free_swap(mapping, index, page)) {
525 /* Swap was replaced by page: retry */
534 if (page->mapping == mapping) {
535 VM_BUG_ON(PageWriteback(page));
536 truncate_inode_page(mapping, page);
538 /* Page was replaced by swap: retry */
545 shmem_deswap_pagevec(&pvec);
546 pagevec_release(&pvec);
547 mem_cgroup_uncharge_end();
551 spin_lock(&info->lock);
552 info->swapped -= nr_swaps_freed;
553 shmem_recalc_inode(inode);
554 spin_unlock(&info->lock);
556 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
558 EXPORT_SYMBOL_GPL(shmem_truncate_range);
560 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
562 struct inode *inode = dentry->d_inode;
565 error = setattr_prepare(dentry, attr);
569 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
570 loff_t oldsize = inode->i_size;
571 loff_t newsize = attr->ia_size;
573 if (newsize != oldsize) {
574 i_size_write(inode, newsize);
575 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
577 if (newsize < oldsize) {
578 loff_t holebegin = round_up(newsize, PAGE_SIZE);
579 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
580 shmem_truncate_range(inode, newsize, (loff_t)-1);
581 /* unmap again to remove racily COWed private pages */
582 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
586 setattr_copy(inode, attr);
587 #ifdef CONFIG_TMPFS_POSIX_ACL
588 if (attr->ia_valid & ATTR_MODE)
589 error = generic_acl_chmod(inode);
594 static void shmem_evict_inode(struct inode *inode)
596 struct shmem_inode_info *info = SHMEM_I(inode);
597 struct shmem_xattr *xattr, *nxattr;
599 if (inode->i_mapping->a_ops == &shmem_aops) {
600 shmem_unacct_size(info->flags, inode->i_size);
602 shmem_truncate_range(inode, 0, (loff_t)-1);
603 if (!list_empty(&info->swaplist)) {
604 mutex_lock(&shmem_swaplist_mutex);
605 list_del_init(&info->swaplist);
606 mutex_unlock(&shmem_swaplist_mutex);
609 kfree(info->symlink);
611 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
615 WARN_ON(inode->i_blocks);
616 shmem_free_inode(inode->i_sb);
617 end_writeback(inode);
621 * If swap found in inode, free it and move page from swapcache to filecache.
623 static int shmem_unuse_inode(struct shmem_inode_info *info,
624 swp_entry_t swap, struct page *page)
626 struct address_space *mapping = info->vfs_inode.i_mapping;
631 radswap = swp_to_radix_entry(swap);
632 index = radix_tree_locate_item(&mapping->page_tree, radswap);
637 * Move _head_ to start search for next from here.
638 * But be careful: shmem_evict_inode checks list_empty without taking
639 * mutex, and there's an instant in list_move_tail when info->swaplist
640 * would appear empty, if it were the only one on shmem_swaplist.
642 if (shmem_swaplist.next != &info->swaplist)
643 list_move_tail(&shmem_swaplist, &info->swaplist);
646 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
647 * but also to hold up shmem_evict_inode(): so inode cannot be freed
648 * beneath us (pagelock doesn't help until the page is in pagecache).
650 error = shmem_add_to_page_cache(page, mapping, index,
651 GFP_NOWAIT, radswap);
652 /* which does mem_cgroup_uncharge_cache_page on error */
654 if (error != -ENOMEM) {
656 * Truncation and eviction use free_swap_and_cache(), which
657 * only does trylock page: if we raced, best clean up here.
659 delete_from_swap_cache(page);
660 set_page_dirty(page);
662 spin_lock(&info->lock);
664 spin_unlock(&info->lock);
667 error = 1; /* not an error, but entry was found */
673 * Search through swapped inodes to find and replace swap by page.
675 int shmem_unuse(swp_entry_t swap, struct page *page)
677 struct list_head *this, *next;
678 struct shmem_inode_info *info;
683 * Charge page using GFP_KERNEL while we can wait, before taking
684 * the shmem_swaplist_mutex which might hold up shmem_writepage().
685 * Charged back to the user (not to caller) when swap account is used.
687 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
690 /* No radix_tree_preload: swap entry keeps a place for page in tree */
692 mutex_lock(&shmem_swaplist_mutex);
693 list_for_each_safe(this, next, &shmem_swaplist) {
694 info = list_entry(this, struct shmem_inode_info, swaplist);
696 found = shmem_unuse_inode(info, swap, page);
698 list_del_init(&info->swaplist);
703 mutex_unlock(&shmem_swaplist_mutex);
706 mem_cgroup_uncharge_cache_page(page);
711 page_cache_release(page);
716 * Move the page from the page cache to the swap cache.
718 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
720 struct shmem_inode_info *info;
721 struct address_space *mapping;
726 BUG_ON(!PageLocked(page));
727 mapping = page->mapping;
729 inode = mapping->host;
730 info = SHMEM_I(inode);
731 if (info->flags & VM_LOCKED)
733 if (!total_swap_pages)
737 * shmem_backing_dev_info's capabilities prevent regular writeback or
738 * sync from ever calling shmem_writepage; but a stacking filesystem
739 * might use ->writepage of its underlying filesystem, in which case
740 * tmpfs should write out to swap only in response to memory pressure,
741 * and not for the writeback threads or sync.
743 if (!wbc->for_reclaim) {
744 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
747 swap = get_swap_page();
752 * Add inode to shmem_unuse()'s list of swapped-out inodes,
753 * if it's not already there. Do it now before the page is
754 * moved to swap cache, when its pagelock no longer protects
755 * the inode from eviction. But don't unlock the mutex until
756 * we've incremented swapped, because shmem_unuse_inode() will
757 * prune a !swapped inode from the swaplist under this mutex.
759 mutex_lock(&shmem_swaplist_mutex);
760 if (list_empty(&info->swaplist))
761 list_add_tail(&info->swaplist, &shmem_swaplist);
763 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
764 swap_shmem_alloc(swap);
765 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
767 spin_lock(&info->lock);
769 shmem_recalc_inode(inode);
770 spin_unlock(&info->lock);
772 mutex_unlock(&shmem_swaplist_mutex);
773 BUG_ON(page_mapped(page));
774 swap_writepage(page, wbc);
778 mutex_unlock(&shmem_swaplist_mutex);
779 swapcache_free(swap, NULL);
781 set_page_dirty(page);
782 if (wbc->for_reclaim)
783 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
790 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
794 if (!mpol || mpol->mode == MPOL_DEFAULT)
795 return; /* show nothing */
797 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
799 seq_printf(seq, ",mpol=%s", buffer);
802 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
804 struct mempolicy *mpol = NULL;
806 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
809 spin_unlock(&sbinfo->stat_lock);
813 #endif /* CONFIG_TMPFS */
815 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
816 struct shmem_inode_info *info, pgoff_t index)
818 struct vm_area_struct pvma;
821 /* Create a pseudo vma that just contains the policy */
823 pvma.vm_pgoff = index;
825 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
827 page = swapin_readahead(swap, gfp, &pvma, 0);
829 /* Drop reference taken by mpol_shared_policy_lookup() */
830 mpol_cond_put(pvma.vm_policy);
835 static struct page *shmem_alloc_page(gfp_t gfp,
836 struct shmem_inode_info *info, pgoff_t index)
838 struct vm_area_struct pvma;
841 /* Create a pseudo vma that just contains the policy */
843 pvma.vm_pgoff = index;
845 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
847 page = alloc_page_vma(gfp, &pvma, 0);
849 /* Drop reference taken by mpol_shared_policy_lookup() */
850 mpol_cond_put(pvma.vm_policy);
854 #else /* !CONFIG_NUMA */
856 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
859 #endif /* CONFIG_TMPFS */
861 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
862 struct shmem_inode_info *info, pgoff_t index)
864 return swapin_readahead(swap, gfp, NULL, 0);
867 static inline struct page *shmem_alloc_page(gfp_t gfp,
868 struct shmem_inode_info *info, pgoff_t index)
870 return alloc_page(gfp);
872 #endif /* CONFIG_NUMA */
874 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
875 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
882 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
884 * If we allocate a new one we do not mark it dirty. That's up to the
885 * vm. If we swap it in we mark it dirty since we also free the swap
886 * entry since a page cannot live in both the swap and page cache
888 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
889 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
891 struct address_space *mapping = inode->i_mapping;
892 struct shmem_inode_info *info;
893 struct shmem_sb_info *sbinfo;
899 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
903 page = find_lock_page(mapping, index);
904 if (radix_tree_exceptional_entry(page)) {
905 swap = radix_to_swp_entry(page);
909 if (sgp != SGP_WRITE &&
910 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
915 if (page || (sgp == SGP_READ && !swap.val)) {
917 * Once we can get the page lock, it must be uptodate:
918 * if there were an error in reading back from swap,
919 * the page would not be inserted into the filecache.
921 BUG_ON(page && !PageUptodate(page));
927 * Fast cache lookup did not find it:
928 * bring it back from swap or allocate.
930 info = SHMEM_I(inode);
931 sbinfo = SHMEM_SB(inode->i_sb);
934 /* Look it up and read it in.. */
935 page = lookup_swap_cache(swap);
937 /* here we actually do the io */
939 *fault_type |= VM_FAULT_MAJOR;
940 page = shmem_swapin(swap, gfp, info, index);
947 /* We have to do this with page locked to prevent races */
949 if (!PageUptodate(page)) {
953 wait_on_page_writeback(page);
955 /* Someone may have already done it for us */
957 if (page->mapping == mapping &&
958 page->index == index)
964 error = mem_cgroup_cache_charge(page, current->mm,
965 gfp & GFP_RECLAIM_MASK);
967 error = shmem_add_to_page_cache(page, mapping, index,
968 gfp, swp_to_radix_entry(swap));
972 spin_lock(&info->lock);
974 shmem_recalc_inode(inode);
975 spin_unlock(&info->lock);
977 delete_from_swap_cache(page);
978 set_page_dirty(page);
982 if (shmem_acct_block(info->flags)) {
986 if (sbinfo->max_blocks) {
987 if (percpu_counter_compare(&sbinfo->used_blocks,
988 sbinfo->max_blocks) >= 0) {
992 percpu_counter_inc(&sbinfo->used_blocks);
995 page = shmem_alloc_page(gfp, info, index);
1001 SetPageSwapBacked(page);
1002 __set_page_locked(page);
1003 error = mem_cgroup_cache_charge(page, current->mm,
1004 gfp & GFP_RECLAIM_MASK);
1006 error = shmem_add_to_page_cache(page, mapping, index,
1010 lru_cache_add_anon(page);
1012 spin_lock(&info->lock);
1014 inode->i_blocks += BLOCKS_PER_PAGE;
1015 shmem_recalc_inode(inode);
1016 spin_unlock(&info->lock);
1018 clear_highpage(page);
1019 flush_dcache_page(page);
1020 SetPageUptodate(page);
1021 if (sgp == SGP_DIRTY)
1022 set_page_dirty(page);
1025 /* Perhaps the file has been truncated since we checked */
1026 if (sgp != SGP_WRITE &&
1027 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1038 ClearPageDirty(page);
1039 delete_from_page_cache(page);
1040 spin_lock(&info->lock);
1042 inode->i_blocks -= BLOCKS_PER_PAGE;
1043 spin_unlock(&info->lock);
1045 if (sbinfo->max_blocks)
1046 percpu_counter_add(&sbinfo->used_blocks, -1);
1048 shmem_unacct_blocks(info->flags, 1);
1050 if (swap.val && error != -EINVAL) {
1051 struct page *test = find_get_page(mapping, index);
1052 if (test && !radix_tree_exceptional_entry(test))
1053 page_cache_release(test);
1054 /* Have another try if the entry has changed */
1055 if (test != swp_to_radix_entry(swap))
1060 page_cache_release(page);
1062 if (error == -ENOSPC && !once++) {
1063 info = SHMEM_I(inode);
1064 spin_lock(&info->lock);
1065 shmem_recalc_inode(inode);
1066 spin_unlock(&info->lock);
1069 if (error == -EEXIST)
1074 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1076 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1078 int ret = VM_FAULT_LOCKED;
1081 * Trinity finds that probing a hole which tmpfs is punching can
1082 * prevent the hole-punch from ever completing: which in turn
1083 * locks writers out with its hold on i_mutex. So refrain from
1084 * faulting pages into the hole while it's being punched. Although
1085 * shmem_truncate_range() does remove the additions, it may be unable to
1086 * keep up, as each new page needs its own unmap_mapping_range() call,
1087 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1089 * It does not matter if we sometimes reach this check just before the
1090 * hole-punch begins, so that one fault then races with the punch:
1091 * we just need to make racing faults a rare case.
1093 * The implementation below would be much simpler if we just used a
1094 * standard mutex or completion: but we cannot take i_mutex in fault,
1095 * and bloating every shmem inode for this unlikely case would be sad.
1097 if (unlikely(inode->i_private)) {
1098 struct shmem_falloc *shmem_falloc;
1100 spin_lock(&inode->i_lock);
1101 shmem_falloc = inode->i_private;
1103 vmf->pgoff >= shmem_falloc->start &&
1104 vmf->pgoff < shmem_falloc->next) {
1105 wait_queue_head_t *shmem_falloc_waitq;
1106 DEFINE_WAIT(shmem_fault_wait);
1108 ret = VM_FAULT_NOPAGE;
1109 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1110 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1111 /* It's polite to up mmap_sem if we can */
1112 up_read(&vma->vm_mm->mmap_sem);
1113 ret = VM_FAULT_RETRY;
1116 shmem_falloc_waitq = shmem_falloc->waitq;
1117 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1118 TASK_UNINTERRUPTIBLE);
1119 spin_unlock(&inode->i_lock);
1123 * shmem_falloc_waitq points into the vmtruncate_range()
1124 * stack of the hole-punching task: shmem_falloc_waitq
1125 * is usually invalid by the time we reach here, but
1126 * finish_wait() does not dereference it in that case;
1127 * though i_lock needed lest racing with wake_up_all().
1129 spin_lock(&inode->i_lock);
1130 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1131 spin_unlock(&inode->i_lock);
1134 spin_unlock(&inode->i_lock);
1137 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1139 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1141 if (ret & VM_FAULT_MAJOR) {
1142 count_vm_event(PGMAJFAULT);
1143 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1148 int vmtruncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1151 * If the underlying filesystem is not going to provide
1152 * a way to truncate a range of blocks (punch a hole) -
1153 * we should return failure right now.
1154 * Only CONFIG_SHMEM shmem.c ever supported i_op->truncate_range().
1156 if (inode->i_op->truncate_range != shmem_truncate_range)
1159 mutex_lock(&inode->i_mutex);
1161 struct shmem_falloc shmem_falloc;
1162 struct address_space *mapping = inode->i_mapping;
1163 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
1164 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
1165 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
1167 shmem_falloc.waitq = &shmem_falloc_waitq;
1168 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
1169 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
1170 spin_lock(&inode->i_lock);
1171 inode->i_private = &shmem_falloc;
1172 spin_unlock(&inode->i_lock);
1174 if ((u64)unmap_end > (u64)unmap_start)
1175 unmap_mapping_range(mapping, unmap_start,
1176 1 + unmap_end - unmap_start, 0);
1177 shmem_truncate_range(inode, lstart, lend);
1178 /* No need to unmap again: hole-punching leaves COWed pages */
1180 spin_lock(&inode->i_lock);
1181 inode->i_private = NULL;
1182 wake_up_all(&shmem_falloc_waitq);
1183 spin_unlock(&inode->i_lock);
1185 mutex_unlock(&inode->i_mutex);
1190 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1192 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1193 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1196 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1199 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1202 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1203 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1207 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1209 struct inode *inode = file->f_path.dentry->d_inode;
1210 struct shmem_inode_info *info = SHMEM_I(inode);
1211 int retval = -ENOMEM;
1213 spin_lock(&info->lock);
1214 if (lock && !(info->flags & VM_LOCKED)) {
1215 if (!user_shm_lock(inode->i_size, user))
1217 info->flags |= VM_LOCKED;
1218 mapping_set_unevictable(file->f_mapping);
1220 if (!lock && (info->flags & VM_LOCKED) && user) {
1221 user_shm_unlock(inode->i_size, user);
1222 info->flags &= ~VM_LOCKED;
1223 mapping_clear_unevictable(file->f_mapping);
1228 spin_unlock(&info->lock);
1232 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1234 file_accessed(file);
1235 vma->vm_ops = &shmem_vm_ops;
1236 vma->vm_flags |= VM_CAN_NONLINEAR;
1240 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1241 int mode, dev_t dev, unsigned long flags)
1243 struct inode *inode;
1244 struct shmem_inode_info *info;
1245 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1247 if (shmem_reserve_inode(sb))
1250 inode = new_inode(sb);
1252 inode->i_ino = get_next_ino();
1253 inode_init_owner(inode, dir, mode);
1254 inode->i_blocks = 0;
1255 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1256 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1257 inode->i_generation = get_seconds();
1258 info = SHMEM_I(inode);
1259 memset(info, 0, (char *)inode - (char *)info);
1260 spin_lock_init(&info->lock);
1261 info->flags = flags & VM_NORESERVE;
1262 INIT_LIST_HEAD(&info->swaplist);
1263 INIT_LIST_HEAD(&info->xattr_list);
1264 cache_no_acl(inode);
1266 switch (mode & S_IFMT) {
1268 inode->i_op = &shmem_special_inode_operations;
1269 init_special_inode(inode, mode, dev);
1272 inode->i_mapping->a_ops = &shmem_aops;
1273 inode->i_op = &shmem_inode_operations;
1274 inode->i_fop = &shmem_file_operations;
1275 mpol_shared_policy_init(&info->policy,
1276 shmem_get_sbmpol(sbinfo));
1280 /* Some things misbehave if size == 0 on a directory */
1281 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1282 inode->i_op = &shmem_dir_inode_operations;
1283 inode->i_fop = &simple_dir_operations;
1287 * Must not load anything in the rbtree,
1288 * mpol_free_shared_policy will not be called.
1290 mpol_shared_policy_init(&info->policy, NULL);
1294 shmem_free_inode(sb);
1299 static const struct inode_operations shmem_symlink_inode_operations;
1300 static const struct inode_operations shmem_short_symlink_operations;
1303 shmem_write_begin(struct file *file, struct address_space *mapping,
1304 loff_t pos, unsigned len, unsigned flags,
1305 struct page **pagep, void **fsdata)
1307 struct inode *inode = mapping->host;
1308 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1309 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1313 shmem_write_end(struct file *file, struct address_space *mapping,
1314 loff_t pos, unsigned len, unsigned copied,
1315 struct page *page, void *fsdata)
1317 struct inode *inode = mapping->host;
1319 if (pos + copied > inode->i_size)
1320 i_size_write(inode, pos + copied);
1322 set_page_dirty(page);
1324 page_cache_release(page);
1329 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1331 struct inode *inode = filp->f_path.dentry->d_inode;
1332 struct address_space *mapping = inode->i_mapping;
1334 unsigned long offset;
1335 enum sgp_type sgp = SGP_READ;
1338 * Might this read be for a stacking filesystem? Then when reading
1339 * holes of a sparse file, we actually need to allocate those pages,
1340 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1342 if (segment_eq(get_fs(), KERNEL_DS))
1345 index = *ppos >> PAGE_CACHE_SHIFT;
1346 offset = *ppos & ~PAGE_CACHE_MASK;
1349 struct page *page = NULL;
1351 unsigned long nr, ret;
1352 loff_t i_size = i_size_read(inode);
1354 end_index = i_size >> PAGE_CACHE_SHIFT;
1355 if (index > end_index)
1357 if (index == end_index) {
1358 nr = i_size & ~PAGE_CACHE_MASK;
1363 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1365 if (desc->error == -EINVAL)
1373 * We must evaluate after, since reads (unlike writes)
1374 * are called without i_mutex protection against truncate
1376 nr = PAGE_CACHE_SIZE;
1377 i_size = i_size_read(inode);
1378 end_index = i_size >> PAGE_CACHE_SHIFT;
1379 if (index == end_index) {
1380 nr = i_size & ~PAGE_CACHE_MASK;
1383 page_cache_release(page);
1391 * If users can be writing to this page using arbitrary
1392 * virtual addresses, take care about potential aliasing
1393 * before reading the page on the kernel side.
1395 if (mapping_writably_mapped(mapping))
1396 flush_dcache_page(page);
1398 * Mark the page accessed if we read the beginning.
1401 mark_page_accessed(page);
1403 page = ZERO_PAGE(0);
1404 page_cache_get(page);
1408 * Ok, we have the page, and it's up-to-date, so
1409 * now we can copy it to user space...
1411 * The actor routine returns how many bytes were actually used..
1412 * NOTE! This may not be the same as how much of a user buffer
1413 * we filled up (we may be padding etc), so we can only update
1414 * "pos" here (the actor routine has to update the user buffer
1415 * pointers and the remaining count).
1417 ret = actor(desc, page, offset, nr);
1419 index += offset >> PAGE_CACHE_SHIFT;
1420 offset &= ~PAGE_CACHE_MASK;
1422 page_cache_release(page);
1423 if (ret != nr || !desc->count)
1429 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1430 file_accessed(filp);
1433 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1434 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1436 struct file *filp = iocb->ki_filp;
1440 loff_t *ppos = &iocb->ki_pos;
1442 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1446 for (seg = 0; seg < nr_segs; seg++) {
1447 read_descriptor_t desc;
1450 desc.arg.buf = iov[seg].iov_base;
1451 desc.count = iov[seg].iov_len;
1452 if (desc.count == 0)
1455 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1456 retval += desc.written;
1458 retval = retval ?: desc.error;
1467 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1468 struct pipe_inode_info *pipe, size_t len,
1471 struct address_space *mapping = in->f_mapping;
1472 struct inode *inode = mapping->host;
1473 unsigned int loff, nr_pages, req_pages;
1474 struct page *pages[PIPE_DEF_BUFFERS];
1475 struct partial_page partial[PIPE_DEF_BUFFERS];
1477 pgoff_t index, end_index;
1480 struct splice_pipe_desc spd = {
1483 .nr_pages_max = PIPE_DEF_BUFFERS,
1485 .ops = &page_cache_pipe_buf_ops,
1486 .spd_release = spd_release_page,
1489 isize = i_size_read(inode);
1490 if (unlikely(*ppos >= isize))
1493 left = isize - *ppos;
1494 if (unlikely(left < len))
1497 if (splice_grow_spd(pipe, &spd))
1500 index = *ppos >> PAGE_CACHE_SHIFT;
1501 loff = *ppos & ~PAGE_CACHE_MASK;
1502 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1503 nr_pages = min(req_pages, pipe->buffers);
1505 spd.nr_pages = find_get_pages_contig(mapping, index,
1506 nr_pages, spd.pages);
1507 index += spd.nr_pages;
1510 while (spd.nr_pages < nr_pages) {
1511 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1515 spd.pages[spd.nr_pages++] = page;
1519 index = *ppos >> PAGE_CACHE_SHIFT;
1520 nr_pages = spd.nr_pages;
1523 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1524 unsigned int this_len;
1529 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1530 page = spd.pages[page_nr];
1532 if (!PageUptodate(page) || page->mapping != mapping) {
1533 error = shmem_getpage(inode, index, &page,
1538 page_cache_release(spd.pages[page_nr]);
1539 spd.pages[page_nr] = page;
1542 isize = i_size_read(inode);
1543 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1544 if (unlikely(!isize || index > end_index))
1547 if (end_index == index) {
1550 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1554 this_len = min(this_len, plen - loff);
1558 spd.partial[page_nr].offset = loff;
1559 spd.partial[page_nr].len = this_len;
1566 while (page_nr < nr_pages)
1567 page_cache_release(spd.pages[page_nr++]);
1570 error = splice_to_pipe(pipe, &spd);
1572 splice_shrink_spd(&spd);
1581 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1583 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1585 buf->f_type = TMPFS_MAGIC;
1586 buf->f_bsize = PAGE_CACHE_SIZE;
1587 buf->f_namelen = NAME_MAX;
1588 if (sbinfo->max_blocks) {
1589 buf->f_blocks = sbinfo->max_blocks;
1591 buf->f_bfree = sbinfo->max_blocks -
1592 percpu_counter_sum(&sbinfo->used_blocks);
1594 if (sbinfo->max_inodes) {
1595 buf->f_files = sbinfo->max_inodes;
1596 buf->f_ffree = sbinfo->free_inodes;
1598 /* else leave those fields 0 like simple_statfs */
1603 * File creation. Allocate an inode, and we're done..
1606 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1608 struct inode *inode;
1609 int error = -ENOSPC;
1611 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1613 error = security_inode_init_security(inode, dir,
1617 if (error != -EOPNOTSUPP) {
1622 #ifdef CONFIG_TMPFS_POSIX_ACL
1623 error = generic_acl_init(inode, dir);
1631 dir->i_size += BOGO_DIRENT_SIZE;
1632 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1633 d_instantiate(dentry, inode);
1634 dget(dentry); /* Extra count - pin the dentry in core */
1639 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1643 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1649 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1650 struct nameidata *nd)
1652 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1658 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1660 struct inode *inode = old_dentry->d_inode;
1664 * No ordinary (disk based) filesystem counts links as inodes;
1665 * but each new link needs a new dentry, pinning lowmem, and
1666 * tmpfs dentries cannot be pruned until they are unlinked.
1668 ret = shmem_reserve_inode(inode->i_sb);
1672 dir->i_size += BOGO_DIRENT_SIZE;
1673 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1675 ihold(inode); /* New dentry reference */
1676 dget(dentry); /* Extra pinning count for the created dentry */
1677 d_instantiate(dentry, inode);
1682 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1684 struct inode *inode = dentry->d_inode;
1686 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1687 shmem_free_inode(inode->i_sb);
1689 dir->i_size -= BOGO_DIRENT_SIZE;
1690 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1692 dput(dentry); /* Undo the count from "create" - this does all the work */
1696 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1698 if (!simple_empty(dentry))
1701 drop_nlink(dentry->d_inode);
1703 return shmem_unlink(dir, dentry);
1707 * The VFS layer already does all the dentry stuff for rename,
1708 * we just have to decrement the usage count for the target if
1709 * it exists so that the VFS layer correctly free's it when it
1712 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1714 struct inode *inode = old_dentry->d_inode;
1715 int they_are_dirs = S_ISDIR(inode->i_mode);
1717 if (!simple_empty(new_dentry))
1720 if (new_dentry->d_inode) {
1721 (void) shmem_unlink(new_dir, new_dentry);
1722 if (they_are_dirs) {
1723 drop_nlink(new_dentry->d_inode);
1724 drop_nlink(old_dir);
1726 } else if (they_are_dirs) {
1727 drop_nlink(old_dir);
1731 old_dir->i_size -= BOGO_DIRENT_SIZE;
1732 new_dir->i_size += BOGO_DIRENT_SIZE;
1733 old_dir->i_ctime = old_dir->i_mtime =
1734 new_dir->i_ctime = new_dir->i_mtime =
1735 inode->i_ctime = CURRENT_TIME;
1739 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1743 struct inode *inode;
1746 struct shmem_inode_info *info;
1748 len = strlen(symname) + 1;
1749 if (len > PAGE_CACHE_SIZE)
1750 return -ENAMETOOLONG;
1752 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1756 error = security_inode_init_security(inode, dir, &dentry->d_name,
1759 if (error != -EOPNOTSUPP) {
1766 info = SHMEM_I(inode);
1767 inode->i_size = len-1;
1768 if (len <= SHORT_SYMLINK_LEN) {
1769 info->symlink = kmemdup(symname, len, GFP_KERNEL);
1770 if (!info->symlink) {
1774 inode->i_op = &shmem_short_symlink_operations;
1776 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1781 inode->i_mapping->a_ops = &shmem_aops;
1782 inode->i_op = &shmem_symlink_inode_operations;
1783 kaddr = kmap_atomic(page, KM_USER0);
1784 memcpy(kaddr, symname, len);
1785 kunmap_atomic(kaddr, KM_USER0);
1786 set_page_dirty(page);
1788 page_cache_release(page);
1790 dir->i_size += BOGO_DIRENT_SIZE;
1791 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1792 d_instantiate(dentry, inode);
1797 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
1799 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
1803 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1805 struct page *page = NULL;
1806 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1807 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
1813 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1815 if (!IS_ERR(nd_get_link(nd))) {
1816 struct page *page = cookie;
1818 mark_page_accessed(page);
1819 page_cache_release(page);
1823 #ifdef CONFIG_TMPFS_XATTR
1825 * Superblocks without xattr inode operations may get some security.* xattr
1826 * support from the LSM "for free". As soon as we have any other xattrs
1827 * like ACLs, we also need to implement the security.* handlers at
1828 * filesystem level, though.
1831 static int shmem_xattr_get(struct dentry *dentry, const char *name,
1832 void *buffer, size_t size)
1834 struct shmem_inode_info *info;
1835 struct shmem_xattr *xattr;
1838 info = SHMEM_I(dentry->d_inode);
1840 spin_lock(&info->lock);
1841 list_for_each_entry(xattr, &info->xattr_list, list) {
1842 if (strcmp(name, xattr->name))
1847 if (size < xattr->size)
1850 memcpy(buffer, xattr->value, xattr->size);
1854 spin_unlock(&info->lock);
1858 static int shmem_xattr_set(struct dentry *dentry, const char *name,
1859 const void *value, size_t size, int flags)
1861 struct inode *inode = dentry->d_inode;
1862 struct shmem_inode_info *info = SHMEM_I(inode);
1863 struct shmem_xattr *xattr;
1864 struct shmem_xattr *new_xattr = NULL;
1868 /* value == NULL means remove */
1871 len = sizeof(*new_xattr) + size;
1872 if (len <= sizeof(*new_xattr))
1875 new_xattr = kmalloc(len, GFP_KERNEL);
1879 new_xattr->name = kstrdup(name, GFP_KERNEL);
1880 if (!new_xattr->name) {
1885 new_xattr->size = size;
1886 memcpy(new_xattr->value, value, size);
1889 spin_lock(&info->lock);
1890 list_for_each_entry(xattr, &info->xattr_list, list) {
1891 if (!strcmp(name, xattr->name)) {
1892 if (flags & XATTR_CREATE) {
1895 } else if (new_xattr) {
1896 list_replace(&xattr->list, &new_xattr->list);
1898 list_del(&xattr->list);
1903 if (flags & XATTR_REPLACE) {
1907 list_add(&new_xattr->list, &info->xattr_list);
1911 spin_unlock(&info->lock);
1918 static const struct xattr_handler *shmem_xattr_handlers[] = {
1919 #ifdef CONFIG_TMPFS_POSIX_ACL
1920 &generic_acl_access_handler,
1921 &generic_acl_default_handler,
1926 static int shmem_xattr_validate(const char *name)
1928 struct { const char *prefix; size_t len; } arr[] = {
1929 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
1930 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
1934 for (i = 0; i < ARRAY_SIZE(arr); i++) {
1935 size_t preflen = arr[i].len;
1936 if (strncmp(name, arr[i].prefix, preflen) == 0) {
1945 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
1946 void *buffer, size_t size)
1951 * If this is a request for a synthetic attribute in the system.*
1952 * namespace use the generic infrastructure to resolve a handler
1953 * for it via sb->s_xattr.
1955 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1956 return generic_getxattr(dentry, name, buffer, size);
1958 err = shmem_xattr_validate(name);
1962 return shmem_xattr_get(dentry, name, buffer, size);
1965 static int shmem_setxattr(struct dentry *dentry, const char *name,
1966 const void *value, size_t size, int flags)
1971 * If this is a request for a synthetic attribute in the system.*
1972 * namespace use the generic infrastructure to resolve a handler
1973 * for it via sb->s_xattr.
1975 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1976 return generic_setxattr(dentry, name, value, size, flags);
1978 err = shmem_xattr_validate(name);
1983 value = ""; /* empty EA, do not remove */
1985 return shmem_xattr_set(dentry, name, value, size, flags);
1989 static int shmem_removexattr(struct dentry *dentry, const char *name)
1994 * If this is a request for a synthetic attribute in the system.*
1995 * namespace use the generic infrastructure to resolve a handler
1996 * for it via sb->s_xattr.
1998 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1999 return generic_removexattr(dentry, name);
2001 err = shmem_xattr_validate(name);
2005 return shmem_xattr_set(dentry, name, NULL, 0, XATTR_REPLACE);
2008 static bool xattr_is_trusted(const char *name)
2010 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
2013 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2015 bool trusted = capable(CAP_SYS_ADMIN);
2016 struct shmem_xattr *xattr;
2017 struct shmem_inode_info *info;
2020 info = SHMEM_I(dentry->d_inode);
2022 spin_lock(&info->lock);
2023 list_for_each_entry(xattr, &info->xattr_list, list) {
2026 /* skip "trusted." attributes for unprivileged callers */
2027 if (!trusted && xattr_is_trusted(xattr->name))
2030 len = strlen(xattr->name) + 1;
2037 memcpy(buffer, xattr->name, len);
2041 spin_unlock(&info->lock);
2045 #endif /* CONFIG_TMPFS_XATTR */
2047 static const struct inode_operations shmem_short_symlink_operations = {
2048 .readlink = generic_readlink,
2049 .follow_link = shmem_follow_short_symlink,
2050 #ifdef CONFIG_TMPFS_XATTR
2051 .setxattr = shmem_setxattr,
2052 .getxattr = shmem_getxattr,
2053 .listxattr = shmem_listxattr,
2054 .removexattr = shmem_removexattr,
2058 static const struct inode_operations shmem_symlink_inode_operations = {
2059 .readlink = generic_readlink,
2060 .follow_link = shmem_follow_link,
2061 .put_link = shmem_put_link,
2062 #ifdef CONFIG_TMPFS_XATTR
2063 .setxattr = shmem_setxattr,
2064 .getxattr = shmem_getxattr,
2065 .listxattr = shmem_listxattr,
2066 .removexattr = shmem_removexattr,
2070 static struct dentry *shmem_get_parent(struct dentry *child)
2072 return ERR_PTR(-ESTALE);
2075 static int shmem_match(struct inode *ino, void *vfh)
2079 inum = (inum << 32) | fh[1];
2080 return ino->i_ino == inum && fh[0] == ino->i_generation;
2083 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2084 struct fid *fid, int fh_len, int fh_type)
2086 struct inode *inode;
2087 struct dentry *dentry = NULL;
2094 inum = (inum << 32) | fid->raw[1];
2096 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2097 shmem_match, fid->raw);
2099 dentry = d_find_alias(inode);
2106 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2109 struct inode *inode = dentry->d_inode;
2116 if (inode_unhashed(inode)) {
2117 /* Unfortunately insert_inode_hash is not idempotent,
2118 * so as we hash inodes here rather than at creation
2119 * time, we need a lock to ensure we only try
2122 static DEFINE_SPINLOCK(lock);
2124 if (inode_unhashed(inode))
2125 __insert_inode_hash(inode,
2126 inode->i_ino + inode->i_generation);
2130 fh[0] = inode->i_generation;
2131 fh[1] = inode->i_ino;
2132 fh[2] = ((__u64)inode->i_ino) >> 32;
2138 static const struct export_operations shmem_export_ops = {
2139 .get_parent = shmem_get_parent,
2140 .encode_fh = shmem_encode_fh,
2141 .fh_to_dentry = shmem_fh_to_dentry,
2144 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2147 char *this_char, *value, *rest;
2149 while (options != NULL) {
2150 this_char = options;
2153 * NUL-terminate this option: unfortunately,
2154 * mount options form a comma-separated list,
2155 * but mpol's nodelist may also contain commas.
2157 options = strchr(options, ',');
2158 if (options == NULL)
2161 if (!isdigit(*options)) {
2168 if ((value = strchr(this_char,'=')) != NULL) {
2172 "tmpfs: No value for mount option '%s'\n",
2177 if (!strcmp(this_char,"size")) {
2178 unsigned long long size;
2179 size = memparse(value,&rest);
2181 size <<= PAGE_SHIFT;
2182 size *= totalram_pages;
2188 sbinfo->max_blocks =
2189 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2190 } else if (!strcmp(this_char,"nr_blocks")) {
2191 sbinfo->max_blocks = memparse(value, &rest);
2194 } else if (!strcmp(this_char,"nr_inodes")) {
2195 sbinfo->max_inodes = memparse(value, &rest);
2198 } else if (!strcmp(this_char,"mode")) {
2201 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2204 } else if (!strcmp(this_char,"uid")) {
2207 sbinfo->uid = simple_strtoul(value, &rest, 0);
2210 } else if (!strcmp(this_char,"gid")) {
2213 sbinfo->gid = simple_strtoul(value, &rest, 0);
2216 } else if (!strcmp(this_char,"mpol")) {
2217 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2220 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2228 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2234 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2236 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2237 struct shmem_sb_info config = *sbinfo;
2238 unsigned long inodes;
2239 int error = -EINVAL;
2242 if (shmem_parse_options(data, &config, true))
2245 spin_lock(&sbinfo->stat_lock);
2246 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2247 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2249 if (config.max_inodes < inodes)
2252 * Those tests disallow limited->unlimited while any are in use;
2253 * but we must separately disallow unlimited->limited, because
2254 * in that case we have no record of how much is already in use.
2256 if (config.max_blocks && !sbinfo->max_blocks)
2258 if (config.max_inodes && !sbinfo->max_inodes)
2262 sbinfo->max_blocks = config.max_blocks;
2263 sbinfo->max_inodes = config.max_inodes;
2264 sbinfo->free_inodes = config.max_inodes - inodes;
2267 * Preserve previous mempolicy unless mpol remount option was specified.
2270 mpol_put(sbinfo->mpol);
2271 sbinfo->mpol = config.mpol; /* transfers initial ref */
2274 spin_unlock(&sbinfo->stat_lock);
2278 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2280 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2282 if (sbinfo->max_blocks != shmem_default_max_blocks())
2283 seq_printf(seq, ",size=%luk",
2284 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2285 if (sbinfo->max_inodes != shmem_default_max_inodes())
2286 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2287 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2288 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2289 if (sbinfo->uid != 0)
2290 seq_printf(seq, ",uid=%u", sbinfo->uid);
2291 if (sbinfo->gid != 0)
2292 seq_printf(seq, ",gid=%u", sbinfo->gid);
2293 shmem_show_mpol(seq, sbinfo->mpol);
2296 #endif /* CONFIG_TMPFS */
2298 static void shmem_put_super(struct super_block *sb)
2300 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2302 percpu_counter_destroy(&sbinfo->used_blocks);
2304 sb->s_fs_info = NULL;
2307 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2309 struct inode *inode;
2310 struct dentry *root;
2311 struct shmem_sb_info *sbinfo;
2314 /* Round up to L1_CACHE_BYTES to resist false sharing */
2315 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2316 L1_CACHE_BYTES), GFP_KERNEL);
2320 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2321 sbinfo->uid = current_fsuid();
2322 sbinfo->gid = current_fsgid();
2323 sb->s_fs_info = sbinfo;
2327 * Per default we only allow half of the physical ram per
2328 * tmpfs instance, limiting inodes to one per page of lowmem;
2329 * but the internal instance is left unlimited.
2331 if (!(sb->s_flags & MS_NOUSER)) {
2332 sbinfo->max_blocks = shmem_default_max_blocks();
2333 sbinfo->max_inodes = shmem_default_max_inodes();
2334 if (shmem_parse_options(data, sbinfo, false)) {
2339 sb->s_export_op = &shmem_export_ops;
2341 sb->s_flags |= MS_NOUSER;
2344 spin_lock_init(&sbinfo->stat_lock);
2345 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2347 sbinfo->free_inodes = sbinfo->max_inodes;
2349 sb->s_maxbytes = MAX_LFS_FILESIZE;
2350 sb->s_blocksize = PAGE_CACHE_SIZE;
2351 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2352 sb->s_magic = TMPFS_MAGIC;
2353 sb->s_op = &shmem_ops;
2354 sb->s_time_gran = 1;
2355 #ifdef CONFIG_TMPFS_XATTR
2356 sb->s_xattr = shmem_xattr_handlers;
2358 #ifdef CONFIG_TMPFS_POSIX_ACL
2359 sb->s_flags |= MS_POSIXACL;
2362 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2365 inode->i_uid = sbinfo->uid;
2366 inode->i_gid = sbinfo->gid;
2367 root = d_alloc_root(inode);
2376 shmem_put_super(sb);
2380 static struct kmem_cache *shmem_inode_cachep;
2382 static struct inode *shmem_alloc_inode(struct super_block *sb)
2384 struct shmem_inode_info *info;
2385 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2388 return &info->vfs_inode;
2391 static void shmem_destroy_callback(struct rcu_head *head)
2393 struct inode *inode = container_of(head, struct inode, i_rcu);
2394 INIT_LIST_HEAD(&inode->i_dentry);
2395 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2398 static void shmem_destroy_inode(struct inode *inode)
2400 if ((inode->i_mode & S_IFMT) == S_IFREG)
2401 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2402 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2405 static void shmem_init_inode(void *foo)
2407 struct shmem_inode_info *info = foo;
2408 inode_init_once(&info->vfs_inode);
2411 static int shmem_init_inodecache(void)
2413 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2414 sizeof(struct shmem_inode_info),
2415 0, SLAB_PANIC, shmem_init_inode);
2419 static void shmem_destroy_inodecache(void)
2421 kmem_cache_destroy(shmem_inode_cachep);
2424 static const struct address_space_operations shmem_aops = {
2425 .writepage = shmem_writepage,
2426 .set_page_dirty = __set_page_dirty_no_writeback,
2428 .write_begin = shmem_write_begin,
2429 .write_end = shmem_write_end,
2431 .migratepage = migrate_page,
2432 .error_remove_page = generic_error_remove_page,
2435 static const struct file_operations shmem_file_operations = {
2438 .llseek = generic_file_llseek,
2439 .read = do_sync_read,
2440 .write = do_sync_write,
2441 .aio_read = shmem_file_aio_read,
2442 .aio_write = generic_file_aio_write,
2443 .fsync = noop_fsync,
2444 .splice_read = shmem_file_splice_read,
2445 .splice_write = generic_file_splice_write,
2449 static const struct inode_operations shmem_inode_operations = {
2450 .setattr = shmem_setattr,
2451 .truncate_range = shmem_truncate_range,
2452 #ifdef CONFIG_TMPFS_XATTR
2453 .setxattr = shmem_setxattr,
2454 .getxattr = shmem_getxattr,
2455 .listxattr = shmem_listxattr,
2456 .removexattr = shmem_removexattr,
2460 static const struct inode_operations shmem_dir_inode_operations = {
2462 .create = shmem_create,
2463 .lookup = simple_lookup,
2465 .unlink = shmem_unlink,
2466 .symlink = shmem_symlink,
2467 .mkdir = shmem_mkdir,
2468 .rmdir = shmem_rmdir,
2469 .mknod = shmem_mknod,
2470 .rename = shmem_rename,
2472 #ifdef CONFIG_TMPFS_XATTR
2473 .setxattr = shmem_setxattr,
2474 .getxattr = shmem_getxattr,
2475 .listxattr = shmem_listxattr,
2476 .removexattr = shmem_removexattr,
2478 #ifdef CONFIG_TMPFS_POSIX_ACL
2479 .setattr = shmem_setattr,
2483 static const struct inode_operations shmem_special_inode_operations = {
2484 #ifdef CONFIG_TMPFS_XATTR
2485 .setxattr = shmem_setxattr,
2486 .getxattr = shmem_getxattr,
2487 .listxattr = shmem_listxattr,
2488 .removexattr = shmem_removexattr,
2490 #ifdef CONFIG_TMPFS_POSIX_ACL
2491 .setattr = shmem_setattr,
2495 static const struct super_operations shmem_ops = {
2496 .alloc_inode = shmem_alloc_inode,
2497 .destroy_inode = shmem_destroy_inode,
2499 .statfs = shmem_statfs,
2500 .remount_fs = shmem_remount_fs,
2501 .show_options = shmem_show_options,
2503 .evict_inode = shmem_evict_inode,
2504 .drop_inode = generic_delete_inode,
2505 .put_super = shmem_put_super,
2508 static const struct vm_operations_struct shmem_vm_ops = {
2509 .fault = shmem_fault,
2511 .set_policy = shmem_set_policy,
2512 .get_policy = shmem_get_policy,
2516 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2517 int flags, const char *dev_name, void *data)
2519 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2522 static struct file_system_type shmem_fs_type = {
2523 .owner = THIS_MODULE,
2525 .mount = shmem_mount,
2526 .kill_sb = kill_litter_super,
2529 int __init shmem_init(void)
2533 error = bdi_init(&shmem_backing_dev_info);
2537 error = shmem_init_inodecache();
2541 error = register_filesystem(&shmem_fs_type);
2543 printk(KERN_ERR "Could not register tmpfs\n");
2547 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2548 shmem_fs_type.name, NULL);
2549 if (IS_ERR(shm_mnt)) {
2550 error = PTR_ERR(shm_mnt);
2551 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2557 unregister_filesystem(&shmem_fs_type);
2559 shmem_destroy_inodecache();
2561 bdi_destroy(&shmem_backing_dev_info);
2563 shm_mnt = ERR_PTR(error);
2567 #else /* !CONFIG_SHMEM */
2570 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2572 * This is intended for small system where the benefits of the full
2573 * shmem code (swap-backed and resource-limited) are outweighed by
2574 * their complexity. On systems without swap this code should be
2575 * effectively equivalent, but much lighter weight.
2578 #include <linux/ramfs.h>
2580 static struct file_system_type shmem_fs_type = {
2582 .mount = ramfs_mount,
2583 .kill_sb = kill_litter_super,
2586 int __init shmem_init(void)
2588 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2590 shm_mnt = kern_mount(&shmem_fs_type);
2591 BUG_ON(IS_ERR(shm_mnt));
2596 int shmem_unuse(swp_entry_t swap, struct page *page)
2601 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2606 void shmem_unlock_mapping(struct address_space *mapping)
2610 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2612 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2614 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2616 int vmtruncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2618 /* Only CONFIG_SHMEM shmem.c ever supported i_op->truncate_range(). */
2622 #define shmem_vm_ops generic_file_vm_ops
2623 #define shmem_file_operations ramfs_file_operations
2624 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2625 #define shmem_acct_size(flags, size) 0
2626 #define shmem_unacct_size(flags, size) do {} while (0)
2628 #endif /* CONFIG_SHMEM */
2633 * shmem_file_setup - get an unlinked file living in tmpfs
2634 * @name: name for dentry (to be seen in /proc/<pid>/maps
2635 * @size: size to be set for the file
2636 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2638 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2642 struct inode *inode;
2644 struct dentry *root;
2647 if (IS_ERR(shm_mnt))
2648 return (void *)shm_mnt;
2650 if (size < 0 || size > MAX_LFS_FILESIZE)
2651 return ERR_PTR(-EINVAL);
2653 if (shmem_acct_size(flags, size))
2654 return ERR_PTR(-ENOMEM);
2658 this.len = strlen(name);
2659 this.hash = 0; /* will go */
2660 root = shm_mnt->mnt_root;
2661 path.dentry = d_alloc(root, &this);
2664 path.mnt = mntget(shm_mnt);
2667 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2671 d_instantiate(path.dentry, inode);
2672 inode->i_size = size;
2673 clear_nlink(inode); /* It is unlinked */
2675 error = ramfs_nommu_expand_for_mapping(inode, size);
2681 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2682 &shmem_file_operations);
2691 shmem_unacct_size(flags, size);
2692 return ERR_PTR(error);
2694 EXPORT_SYMBOL_GPL(shmem_file_setup);
2697 * shmem_zero_setup - setup a shared anonymous mapping
2698 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2700 int shmem_zero_setup(struct vm_area_struct *vma)
2703 loff_t size = vma->vm_end - vma->vm_start;
2705 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2707 return PTR_ERR(file);
2711 vma->vm_file = file;
2712 vma->vm_ops = &shmem_vm_ops;
2713 vma->vm_flags |= VM_CAN_NONLINEAR;
2718 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2719 * @mapping: the page's address_space
2720 * @index: the page index
2721 * @gfp: the page allocator flags to use if allocating
2723 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2724 * with any new page allocations done using the specified allocation flags.
2725 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2726 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2727 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2729 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2730 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2732 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2733 pgoff_t index, gfp_t gfp)
2736 struct inode *inode = mapping->host;
2740 BUG_ON(mapping->a_ops != &shmem_aops);
2741 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2743 page = ERR_PTR(error);
2749 * The tiny !SHMEM case uses ramfs without swap
2751 return read_cache_page_gfp(mapping, index, gfp);
2754 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);