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 struct list_head list; /* anchored by shmem_inode_info->xattr_list */
81 char *name; /* xattr name */
86 /* Flag allocation requirements to shmem_getpage */
88 SGP_READ, /* don't exceed i_size, don't allocate page */
89 SGP_CACHE, /* don't exceed i_size, may allocate page */
90 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
91 SGP_WRITE, /* may exceed i_size, may allocate page */
95 static unsigned long shmem_default_max_blocks(void)
97 return totalram_pages / 2;
100 static unsigned long shmem_default_max_inodes(void)
102 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
106 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
107 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
108 struct shmem_inode_info *info, pgoff_t index);
109 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
110 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
112 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
113 struct page **pagep, enum sgp_type sgp, int *fault_type)
115 return shmem_getpage_gfp(inode, index, pagep, sgp,
116 mapping_gfp_mask(inode->i_mapping), fault_type);
119 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
121 return sb->s_fs_info;
125 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
126 * for shared memory and for shared anonymous (/dev/zero) mappings
127 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
128 * consistent with the pre-accounting of private mappings ...
130 static inline int shmem_acct_size(unsigned long flags, loff_t size)
132 return (flags & VM_NORESERVE) ?
133 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
136 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
138 if (!(flags & VM_NORESERVE))
139 vm_unacct_memory(VM_ACCT(size));
143 * ... whereas tmpfs objects are accounted incrementally as
144 * pages are allocated, in order to allow huge sparse files.
145 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
146 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
148 static inline int shmem_acct_block(unsigned long flags)
150 return (flags & VM_NORESERVE) ?
151 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
154 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
156 if (flags & VM_NORESERVE)
157 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
160 static const struct super_operations shmem_ops;
161 static const struct address_space_operations shmem_aops;
162 static const struct file_operations shmem_file_operations;
163 static const struct inode_operations shmem_inode_operations;
164 static const struct inode_operations shmem_dir_inode_operations;
165 static const struct inode_operations shmem_special_inode_operations;
166 static const struct vm_operations_struct shmem_vm_ops;
168 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
169 .ra_pages = 0, /* No readahead */
170 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
173 static LIST_HEAD(shmem_swaplist);
174 static DEFINE_MUTEX(shmem_swaplist_mutex);
176 static int shmem_reserve_inode(struct super_block *sb)
178 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
179 if (sbinfo->max_inodes) {
180 spin_lock(&sbinfo->stat_lock);
181 if (!sbinfo->free_inodes) {
182 spin_unlock(&sbinfo->stat_lock);
185 sbinfo->free_inodes--;
186 spin_unlock(&sbinfo->stat_lock);
191 static void shmem_free_inode(struct super_block *sb)
193 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
194 if (sbinfo->max_inodes) {
195 spin_lock(&sbinfo->stat_lock);
196 sbinfo->free_inodes++;
197 spin_unlock(&sbinfo->stat_lock);
202 * shmem_recalc_inode - recalculate the block usage of an inode
203 * @inode: inode to recalc
205 * We have to calculate the free blocks since the mm can drop
206 * undirtied hole pages behind our back.
208 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
209 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
211 * It has to be called with the spinlock held.
213 static void shmem_recalc_inode(struct inode *inode)
215 struct shmem_inode_info *info = SHMEM_I(inode);
218 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
220 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
221 if (sbinfo->max_blocks)
222 percpu_counter_add(&sbinfo->used_blocks, -freed);
223 info->alloced -= freed;
224 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
225 shmem_unacct_blocks(info->flags, freed);
230 * Replace item expected in radix tree by a new item, while holding tree lock.
232 static int shmem_radix_tree_replace(struct address_space *mapping,
233 pgoff_t index, void *expected, void *replacement)
238 VM_BUG_ON(!expected);
239 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
241 item = radix_tree_deref_slot_protected(pslot,
242 &mapping->tree_lock);
243 if (item != expected)
246 radix_tree_replace_slot(pslot, replacement);
248 radix_tree_delete(&mapping->page_tree, index);
253 * Like add_to_page_cache_locked, but error if expected item has gone.
255 static int shmem_add_to_page_cache(struct page *page,
256 struct address_space *mapping,
257 pgoff_t index, gfp_t gfp, void *expected)
261 VM_BUG_ON(!PageLocked(page));
262 VM_BUG_ON(!PageSwapBacked(page));
265 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
267 page_cache_get(page);
268 page->mapping = mapping;
271 spin_lock_irq(&mapping->tree_lock);
273 error = radix_tree_insert(&mapping->page_tree,
276 error = shmem_radix_tree_replace(mapping, index,
280 __inc_zone_page_state(page, NR_FILE_PAGES);
281 __inc_zone_page_state(page, NR_SHMEM);
282 spin_unlock_irq(&mapping->tree_lock);
284 page->mapping = NULL;
285 spin_unlock_irq(&mapping->tree_lock);
286 page_cache_release(page);
289 radix_tree_preload_end();
292 mem_cgroup_uncharge_cache_page(page);
297 * Like delete_from_page_cache, but substitutes swap for page.
299 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
301 struct address_space *mapping = page->mapping;
304 spin_lock_irq(&mapping->tree_lock);
305 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
306 page->mapping = NULL;
308 __dec_zone_page_state(page, NR_FILE_PAGES);
309 __dec_zone_page_state(page, NR_SHMEM);
310 spin_unlock_irq(&mapping->tree_lock);
311 page_cache_release(page);
316 * Like find_get_pages, but collecting swap entries as well as pages.
318 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
319 pgoff_t start, unsigned int nr_pages,
320 struct page **pages, pgoff_t *indices)
324 unsigned int nr_found;
328 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
329 (void ***)pages, indices, start, nr_pages);
331 for (i = 0; i < nr_found; i++) {
334 page = radix_tree_deref_slot((void **)pages[i]);
337 if (radix_tree_exception(page)) {
338 if (radix_tree_deref_retry(page))
341 * Otherwise, we must be storing a swap entry
342 * here as an exceptional entry: so return it
343 * without attempting to raise page count.
347 if (!page_cache_get_speculative(page))
350 /* Has the page moved? */
351 if (unlikely(page != *((void **)pages[i]))) {
352 page_cache_release(page);
356 indices[ret] = indices[i];
360 if (unlikely(!ret && nr_found))
367 * Remove swap entry from radix tree, free the swap and its page cache.
369 static int shmem_free_swap(struct address_space *mapping,
370 pgoff_t index, void *radswap)
374 spin_lock_irq(&mapping->tree_lock);
375 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
376 spin_unlock_irq(&mapping->tree_lock);
378 free_swap_and_cache(radix_to_swp_entry(radswap));
383 * Pagevec may contain swap entries, so shuffle up pages before releasing.
385 static void shmem_deswap_pagevec(struct pagevec *pvec)
389 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
390 struct page *page = pvec->pages[i];
391 if (!radix_tree_exceptional_entry(page))
392 pvec->pages[j++] = page;
398 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
400 void shmem_unlock_mapping(struct address_space *mapping)
403 pgoff_t indices[PAGEVEC_SIZE];
406 pagevec_init(&pvec, 0);
408 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
410 while (!mapping_unevictable(mapping)) {
412 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
413 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
415 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
416 PAGEVEC_SIZE, pvec.pages, indices);
419 index = indices[pvec.nr - 1] + 1;
420 shmem_deswap_pagevec(&pvec);
421 check_move_unevictable_pages(pvec.pages, pvec.nr);
422 pagevec_release(&pvec);
428 * Remove range of pages and swap entries from radix tree, and free them.
430 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
432 struct address_space *mapping = inode->i_mapping;
433 struct shmem_inode_info *info = SHMEM_I(inode);
434 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
435 unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
436 pgoff_t end = (lend >> PAGE_CACHE_SHIFT);
438 pgoff_t indices[PAGEVEC_SIZE];
439 long nr_swaps_freed = 0;
443 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
445 pagevec_init(&pvec, 0);
447 while (index <= end) {
448 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
449 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
450 pvec.pages, indices);
453 mem_cgroup_uncharge_start();
454 for (i = 0; i < pagevec_count(&pvec); i++) {
455 struct page *page = pvec.pages[i];
461 if (radix_tree_exceptional_entry(page)) {
462 nr_swaps_freed += !shmem_free_swap(mapping,
467 if (!trylock_page(page))
469 if (page->mapping == mapping) {
470 VM_BUG_ON(PageWriteback(page));
471 truncate_inode_page(mapping, page);
475 shmem_deswap_pagevec(&pvec);
476 pagevec_release(&pvec);
477 mem_cgroup_uncharge_end();
483 struct page *page = NULL;
484 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
486 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
487 set_page_dirty(page);
489 page_cache_release(page);
496 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
497 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
498 pvec.pages, indices);
505 if (index == start && indices[0] > end) {
506 shmem_deswap_pagevec(&pvec);
507 pagevec_release(&pvec);
510 mem_cgroup_uncharge_start();
511 for (i = 0; i < pagevec_count(&pvec); i++) {
512 struct page *page = pvec.pages[i];
518 if (radix_tree_exceptional_entry(page)) {
519 nr_swaps_freed += !shmem_free_swap(mapping,
525 if (page->mapping == mapping) {
526 VM_BUG_ON(PageWriteback(page));
527 truncate_inode_page(mapping, page);
531 shmem_deswap_pagevec(&pvec);
532 pagevec_release(&pvec);
533 mem_cgroup_uncharge_end();
537 spin_lock(&info->lock);
538 info->swapped -= nr_swaps_freed;
539 shmem_recalc_inode(inode);
540 spin_unlock(&info->lock);
542 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
544 EXPORT_SYMBOL_GPL(shmem_truncate_range);
546 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
548 struct inode *inode = dentry->d_inode;
551 error = inode_change_ok(inode, attr);
555 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
556 loff_t oldsize = inode->i_size;
557 loff_t newsize = attr->ia_size;
559 if (newsize != oldsize) {
560 i_size_write(inode, newsize);
561 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
563 if (newsize < oldsize) {
564 loff_t holebegin = round_up(newsize, PAGE_SIZE);
565 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
566 shmem_truncate_range(inode, newsize, (loff_t)-1);
567 /* unmap again to remove racily COWed private pages */
568 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
572 setattr_copy(inode, attr);
573 #ifdef CONFIG_TMPFS_POSIX_ACL
574 if (attr->ia_valid & ATTR_MODE)
575 error = generic_acl_chmod(inode);
580 static void shmem_evict_inode(struct inode *inode)
582 struct shmem_inode_info *info = SHMEM_I(inode);
583 struct shmem_xattr *xattr, *nxattr;
585 if (inode->i_mapping->a_ops == &shmem_aops) {
586 shmem_unacct_size(info->flags, inode->i_size);
588 shmem_truncate_range(inode, 0, (loff_t)-1);
589 if (!list_empty(&info->swaplist)) {
590 mutex_lock(&shmem_swaplist_mutex);
591 list_del_init(&info->swaplist);
592 mutex_unlock(&shmem_swaplist_mutex);
595 kfree(info->symlink);
597 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
601 BUG_ON(inode->i_blocks);
602 shmem_free_inode(inode->i_sb);
607 * If swap found in inode, free it and move page from swapcache to filecache.
609 static int shmem_unuse_inode(struct shmem_inode_info *info,
610 swp_entry_t swap, struct page **pagep)
612 struct address_space *mapping = info->vfs_inode.i_mapping;
618 radswap = swp_to_radix_entry(swap);
619 index = radix_tree_locate_item(&mapping->page_tree, radswap);
624 * Move _head_ to start search for next from here.
625 * But be careful: shmem_evict_inode checks list_empty without taking
626 * mutex, and there's an instant in list_move_tail when info->swaplist
627 * would appear empty, if it were the only one on shmem_swaplist.
629 if (shmem_swaplist.next != &info->swaplist)
630 list_move_tail(&shmem_swaplist, &info->swaplist);
632 gfp = mapping_gfp_mask(mapping);
633 if (shmem_should_replace_page(*pagep, gfp)) {
634 mutex_unlock(&shmem_swaplist_mutex);
635 error = shmem_replace_page(pagep, gfp, info, index);
636 mutex_lock(&shmem_swaplist_mutex);
638 * We needed to drop mutex to make that restrictive page
639 * allocation; but the inode might already be freed by now,
640 * and we cannot refer to inode or mapping or info to check.
641 * However, we do hold page lock on the PageSwapCache page,
642 * so can check if that still has our reference remaining.
644 if (!page_swapcount(*pagep))
649 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
650 * but also to hold up shmem_evict_inode(): so inode cannot be freed
651 * beneath us (pagelock doesn't help until the page is in pagecache).
654 error = shmem_add_to_page_cache(*pagep, mapping, index,
655 GFP_NOWAIT, radswap);
656 if (error != -ENOMEM) {
658 * Truncation and eviction use free_swap_and_cache(), which
659 * only does trylock page: if we raced, best clean up here.
661 delete_from_swap_cache(*pagep);
662 set_page_dirty(*pagep);
664 spin_lock(&info->lock);
666 spin_unlock(&info->lock);
669 error = 1; /* not an error, but entry was found */
675 * Search through swapped inodes to find and replace swap by page.
677 int shmem_unuse(swp_entry_t swap, struct page *page)
679 struct list_head *this, *next;
680 struct shmem_inode_info *info;
685 * There's a faint possibility that swap page was replaced before
686 * caller locked it: it will come back later with the right page.
688 if (unlikely(!PageSwapCache(page)))
692 * Charge page using GFP_KERNEL while we can wait, before taking
693 * the shmem_swaplist_mutex which might hold up shmem_writepage().
694 * Charged back to the user (not to caller) when swap account is used.
696 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
699 /* No radix_tree_preload: swap entry keeps a place for page in tree */
701 mutex_lock(&shmem_swaplist_mutex);
702 list_for_each_safe(this, next, &shmem_swaplist) {
703 info = list_entry(this, struct shmem_inode_info, swaplist);
705 found = shmem_unuse_inode(info, swap, &page);
707 list_del_init(&info->swaplist);
712 mutex_unlock(&shmem_swaplist_mutex);
718 page_cache_release(page);
723 * Move the page from the page cache to the swap cache.
725 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
727 struct shmem_inode_info *info;
728 struct address_space *mapping;
733 BUG_ON(!PageLocked(page));
734 mapping = page->mapping;
736 inode = mapping->host;
737 info = SHMEM_I(inode);
738 if (info->flags & VM_LOCKED)
740 if (!total_swap_pages)
744 * shmem_backing_dev_info's capabilities prevent regular writeback or
745 * sync from ever calling shmem_writepage; but a stacking filesystem
746 * might use ->writepage of its underlying filesystem, in which case
747 * tmpfs should write out to swap only in response to memory pressure,
748 * and not for the writeback threads or sync.
750 if (!wbc->for_reclaim) {
751 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
754 swap = get_swap_page();
759 * Add inode to shmem_unuse()'s list of swapped-out inodes,
760 * if it's not already there. Do it now before the page is
761 * moved to swap cache, when its pagelock no longer protects
762 * the inode from eviction. But don't unlock the mutex until
763 * we've incremented swapped, because shmem_unuse_inode() will
764 * prune a !swapped inode from the swaplist under this mutex.
766 mutex_lock(&shmem_swaplist_mutex);
767 if (list_empty(&info->swaplist))
768 list_add_tail(&info->swaplist, &shmem_swaplist);
770 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
771 swap_shmem_alloc(swap);
772 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
774 spin_lock(&info->lock);
776 shmem_recalc_inode(inode);
777 spin_unlock(&info->lock);
779 mutex_unlock(&shmem_swaplist_mutex);
780 BUG_ON(page_mapped(page));
781 swap_writepage(page, wbc);
785 mutex_unlock(&shmem_swaplist_mutex);
786 swapcache_free(swap, NULL);
788 set_page_dirty(page);
789 if (wbc->for_reclaim)
790 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
797 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
801 if (!mpol || mpol->mode == MPOL_DEFAULT)
802 return; /* show nothing */
804 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
806 seq_printf(seq, ",mpol=%s", buffer);
809 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
811 struct mempolicy *mpol = NULL;
813 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
816 spin_unlock(&sbinfo->stat_lock);
820 #endif /* CONFIG_TMPFS */
822 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
823 struct shmem_inode_info *info, pgoff_t index)
825 struct mempolicy mpol, *spol;
826 struct vm_area_struct pvma;
828 spol = mpol_cond_copy(&mpol,
829 mpol_shared_policy_lookup(&info->policy, index));
831 /* Create a pseudo vma that just contains the policy */
833 pvma.vm_pgoff = index;
835 pvma.vm_policy = spol;
836 return swapin_readahead(swap, gfp, &pvma, 0);
839 static struct page *shmem_alloc_page(gfp_t gfp,
840 struct shmem_inode_info *info, pgoff_t index)
842 struct vm_area_struct pvma;
844 /* Create a pseudo vma that just contains the policy */
846 pvma.vm_pgoff = index;
848 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
851 * alloc_page_vma() will drop the shared policy reference
853 return alloc_page_vma(gfp, &pvma, 0);
855 #else /* !CONFIG_NUMA */
857 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
860 #endif /* CONFIG_TMPFS */
862 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
863 struct shmem_inode_info *info, pgoff_t index)
865 return swapin_readahead(swap, gfp, NULL, 0);
868 static inline struct page *shmem_alloc_page(gfp_t gfp,
869 struct shmem_inode_info *info, pgoff_t index)
871 return alloc_page(gfp);
873 #endif /* CONFIG_NUMA */
875 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
876 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
883 * When a page is moved from swapcache to shmem filecache (either by the
884 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
885 * shmem_unuse_inode()), it may have been read in earlier from swap, in
886 * ignorance of the mapping it belongs to. If that mapping has special
887 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
888 * we may need to copy to a suitable page before moving to filecache.
890 * In a future release, this may well be extended to respect cpuset and
891 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
892 * but for now it is a simple matter of zone.
894 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
896 return page_zonenum(page) > gfp_zone(gfp);
899 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
900 struct shmem_inode_info *info, pgoff_t index)
902 struct page *oldpage, *newpage;
903 struct address_space *swap_mapping;
908 swap_index = page_private(oldpage);
909 swap_mapping = page_mapping(oldpage);
912 * We have arrived here because our zones are constrained, so don't
913 * limit chance of success by further cpuset and node constraints.
915 gfp &= ~GFP_CONSTRAINT_MASK;
916 newpage = shmem_alloc_page(gfp, info, index);
919 VM_BUG_ON(shmem_should_replace_page(newpage, gfp));
922 page_cache_get(newpage);
923 copy_highpage(newpage, oldpage);
925 VM_BUG_ON(!PageLocked(oldpage));
926 __set_page_locked(newpage);
927 VM_BUG_ON(!PageUptodate(oldpage));
928 SetPageUptodate(newpage);
929 VM_BUG_ON(!PageSwapBacked(oldpage));
930 SetPageSwapBacked(newpage);
931 VM_BUG_ON(!swap_index);
932 set_page_private(newpage, swap_index);
933 VM_BUG_ON(!PageSwapCache(oldpage));
934 SetPageSwapCache(newpage);
937 * Our caller will very soon move newpage out of swapcache, but it's
938 * a nice clean interface for us to replace oldpage by newpage there.
940 spin_lock_irq(&swap_mapping->tree_lock);
941 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
943 __inc_zone_page_state(newpage, NR_FILE_PAGES);
944 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
945 spin_unlock_irq(&swap_mapping->tree_lock);
948 mem_cgroup_replace_page_cache(oldpage, newpage);
949 lru_cache_add_anon(newpage);
951 ClearPageSwapCache(oldpage);
952 set_page_private(oldpage, 0);
954 unlock_page(oldpage);
955 page_cache_release(oldpage);
956 page_cache_release(oldpage);
961 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
963 * If we allocate a new one we do not mark it dirty. That's up to the
964 * vm. If we swap it in we mark it dirty since we also free the swap
965 * entry since a page cannot live in both the swap and page cache
967 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
968 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
970 struct address_space *mapping = inode->i_mapping;
971 struct shmem_inode_info *info;
972 struct shmem_sb_info *sbinfo;
978 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
982 page = find_lock_page(mapping, index);
983 if (radix_tree_exceptional_entry(page)) {
984 swap = radix_to_swp_entry(page);
988 if (sgp != SGP_WRITE &&
989 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
994 if (page || (sgp == SGP_READ && !swap.val)) {
996 * Once we can get the page lock, it must be uptodate:
997 * if there were an error in reading back from swap,
998 * the page would not be inserted into the filecache.
1000 BUG_ON(page && !PageUptodate(page));
1006 * Fast cache lookup did not find it:
1007 * bring it back from swap or allocate.
1009 info = SHMEM_I(inode);
1010 sbinfo = SHMEM_SB(inode->i_sb);
1013 /* Look it up and read it in.. */
1014 page = lookup_swap_cache(swap);
1016 /* here we actually do the io */
1018 *fault_type |= VM_FAULT_MAJOR;
1019 page = shmem_swapin(swap, gfp, info, index);
1026 /* We have to do this with page locked to prevent races */
1028 if (!PageSwapCache(page) || page->mapping) {
1029 error = -EEXIST; /* try again */
1032 if (!PageUptodate(page)) {
1036 wait_on_page_writeback(page);
1038 if (shmem_should_replace_page(page, gfp)) {
1039 error = shmem_replace_page(&page, gfp, info, index);
1044 error = mem_cgroup_cache_charge(page, current->mm,
1045 gfp & GFP_RECLAIM_MASK);
1047 error = shmem_add_to_page_cache(page, mapping, index,
1048 gfp, swp_to_radix_entry(swap));
1052 spin_lock(&info->lock);
1054 shmem_recalc_inode(inode);
1055 spin_unlock(&info->lock);
1057 delete_from_swap_cache(page);
1058 set_page_dirty(page);
1062 if (shmem_acct_block(info->flags)) {
1066 if (sbinfo->max_blocks) {
1067 if (percpu_counter_compare(&sbinfo->used_blocks,
1068 sbinfo->max_blocks) >= 0) {
1072 percpu_counter_inc(&sbinfo->used_blocks);
1075 page = shmem_alloc_page(gfp, info, index);
1081 SetPageSwapBacked(page);
1082 __set_page_locked(page);
1083 error = mem_cgroup_cache_charge(page, current->mm,
1084 gfp & GFP_RECLAIM_MASK);
1086 error = shmem_add_to_page_cache(page, mapping, index,
1090 lru_cache_add_anon(page);
1092 spin_lock(&info->lock);
1094 inode->i_blocks += BLOCKS_PER_PAGE;
1095 shmem_recalc_inode(inode);
1096 spin_unlock(&info->lock);
1099 * Let SGP_WRITE caller clear ends if write does not fill page
1101 if (sgp != SGP_WRITE) {
1102 clear_highpage(page);
1103 flush_dcache_page(page);
1104 SetPageUptodate(page);
1106 if (sgp == SGP_DIRTY)
1107 set_page_dirty(page);
1110 /* Perhaps the file has been truncated since we checked */
1111 if (sgp != SGP_WRITE &&
1112 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1123 ClearPageDirty(page);
1124 delete_from_page_cache(page);
1125 spin_lock(&info->lock);
1127 inode->i_blocks -= BLOCKS_PER_PAGE;
1128 spin_unlock(&info->lock);
1130 if (sbinfo->max_blocks)
1131 percpu_counter_add(&sbinfo->used_blocks, -1);
1133 shmem_unacct_blocks(info->flags, 1);
1135 if (swap.val && error != -EINVAL) {
1136 struct page *test = find_get_page(mapping, index);
1137 if (test && !radix_tree_exceptional_entry(test))
1138 page_cache_release(test);
1139 /* Have another try if the entry has changed */
1140 if (test != swp_to_radix_entry(swap))
1145 page_cache_release(page);
1147 if (error == -ENOSPC && !once++) {
1148 info = SHMEM_I(inode);
1149 spin_lock(&info->lock);
1150 shmem_recalc_inode(inode);
1151 spin_unlock(&info->lock);
1154 if (error == -EEXIST)
1159 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1161 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1163 int ret = VM_FAULT_LOCKED;
1165 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1167 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1169 if (ret & VM_FAULT_MAJOR) {
1170 count_vm_event(PGMAJFAULT);
1171 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1177 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1179 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1180 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1183 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1186 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1189 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1190 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1194 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1196 struct inode *inode = file->f_path.dentry->d_inode;
1197 struct shmem_inode_info *info = SHMEM_I(inode);
1198 int retval = -ENOMEM;
1200 spin_lock(&info->lock);
1201 if (lock && !(info->flags & VM_LOCKED)) {
1202 if (!user_shm_lock(inode->i_size, user))
1204 info->flags |= VM_LOCKED;
1205 mapping_set_unevictable(file->f_mapping);
1207 if (!lock && (info->flags & VM_LOCKED) && user) {
1208 user_shm_unlock(inode->i_size, user);
1209 info->flags &= ~VM_LOCKED;
1210 mapping_clear_unevictable(file->f_mapping);
1215 spin_unlock(&info->lock);
1219 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1221 file_accessed(file);
1222 vma->vm_ops = &shmem_vm_ops;
1223 vma->vm_flags |= VM_CAN_NONLINEAR;
1227 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1228 umode_t mode, dev_t dev, unsigned long flags)
1230 struct inode *inode;
1231 struct shmem_inode_info *info;
1232 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1234 if (shmem_reserve_inode(sb))
1237 inode = new_inode(sb);
1239 inode->i_ino = get_next_ino();
1240 inode_init_owner(inode, dir, mode);
1241 inode->i_blocks = 0;
1242 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1243 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1244 inode->i_generation = get_seconds();
1245 info = SHMEM_I(inode);
1246 memset(info, 0, (char *)inode - (char *)info);
1247 spin_lock_init(&info->lock);
1248 info->flags = flags & VM_NORESERVE;
1249 INIT_LIST_HEAD(&info->swaplist);
1250 INIT_LIST_HEAD(&info->xattr_list);
1251 cache_no_acl(inode);
1253 switch (mode & S_IFMT) {
1255 inode->i_op = &shmem_special_inode_operations;
1256 init_special_inode(inode, mode, dev);
1259 inode->i_mapping->a_ops = &shmem_aops;
1260 inode->i_op = &shmem_inode_operations;
1261 inode->i_fop = &shmem_file_operations;
1262 mpol_shared_policy_init(&info->policy,
1263 shmem_get_sbmpol(sbinfo));
1267 /* Some things misbehave if size == 0 on a directory */
1268 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1269 inode->i_op = &shmem_dir_inode_operations;
1270 inode->i_fop = &simple_dir_operations;
1274 * Must not load anything in the rbtree,
1275 * mpol_free_shared_policy will not be called.
1277 mpol_shared_policy_init(&info->policy, NULL);
1281 shmem_free_inode(sb);
1286 static const struct inode_operations shmem_symlink_inode_operations;
1287 static const struct inode_operations shmem_short_symlink_operations;
1289 #ifdef CONFIG_TMPFS_XATTR
1290 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1292 #define shmem_initxattrs NULL
1296 shmem_write_begin(struct file *file, struct address_space *mapping,
1297 loff_t pos, unsigned len, unsigned flags,
1298 struct page **pagep, void **fsdata)
1300 struct inode *inode = mapping->host;
1301 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1302 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1306 shmem_write_end(struct file *file, struct address_space *mapping,
1307 loff_t pos, unsigned len, unsigned copied,
1308 struct page *page, void *fsdata)
1310 struct inode *inode = mapping->host;
1312 if (pos + copied > inode->i_size)
1313 i_size_write(inode, pos + copied);
1315 if (!PageUptodate(page)) {
1316 if (copied < PAGE_CACHE_SIZE) {
1317 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1318 zero_user_segments(page, 0, from,
1319 from + copied, PAGE_CACHE_SIZE);
1321 SetPageUptodate(page);
1323 set_page_dirty(page);
1325 page_cache_release(page);
1330 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1332 struct inode *inode = filp->f_path.dentry->d_inode;
1333 struct address_space *mapping = inode->i_mapping;
1335 unsigned long offset;
1336 enum sgp_type sgp = SGP_READ;
1339 * Might this read be for a stacking filesystem? Then when reading
1340 * holes of a sparse file, we actually need to allocate those pages,
1341 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1343 if (segment_eq(get_fs(), KERNEL_DS))
1346 index = *ppos >> PAGE_CACHE_SHIFT;
1347 offset = *ppos & ~PAGE_CACHE_MASK;
1350 struct page *page = NULL;
1352 unsigned long nr, ret;
1353 loff_t i_size = i_size_read(inode);
1355 end_index = i_size >> PAGE_CACHE_SHIFT;
1356 if (index > end_index)
1358 if (index == end_index) {
1359 nr = i_size & ~PAGE_CACHE_MASK;
1364 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1366 if (desc->error == -EINVAL)
1374 * We must evaluate after, since reads (unlike writes)
1375 * are called without i_mutex protection against truncate
1377 nr = PAGE_CACHE_SIZE;
1378 i_size = i_size_read(inode);
1379 end_index = i_size >> PAGE_CACHE_SHIFT;
1380 if (index == end_index) {
1381 nr = i_size & ~PAGE_CACHE_MASK;
1384 page_cache_release(page);
1392 * If users can be writing to this page using arbitrary
1393 * virtual addresses, take care about potential aliasing
1394 * before reading the page on the kernel side.
1396 if (mapping_writably_mapped(mapping))
1397 flush_dcache_page(page);
1399 * Mark the page accessed if we read the beginning.
1402 mark_page_accessed(page);
1404 page = ZERO_PAGE(0);
1405 page_cache_get(page);
1409 * Ok, we have the page, and it's up-to-date, so
1410 * now we can copy it to user space...
1412 * The actor routine returns how many bytes were actually used..
1413 * NOTE! This may not be the same as how much of a user buffer
1414 * we filled up (we may be padding etc), so we can only update
1415 * "pos" here (the actor routine has to update the user buffer
1416 * pointers and the remaining count).
1418 ret = actor(desc, page, offset, nr);
1420 index += offset >> PAGE_CACHE_SHIFT;
1421 offset &= ~PAGE_CACHE_MASK;
1423 page_cache_release(page);
1424 if (ret != nr || !desc->count)
1430 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1431 file_accessed(filp);
1434 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1435 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1437 struct file *filp = iocb->ki_filp;
1441 loff_t *ppos = &iocb->ki_pos;
1443 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1447 for (seg = 0; seg < nr_segs; seg++) {
1448 read_descriptor_t desc;
1451 desc.arg.buf = iov[seg].iov_base;
1452 desc.count = iov[seg].iov_len;
1453 if (desc.count == 0)
1456 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1457 retval += desc.written;
1459 retval = retval ?: desc.error;
1468 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1469 struct pipe_inode_info *pipe, size_t len,
1472 struct address_space *mapping = in->f_mapping;
1473 struct inode *inode = mapping->host;
1474 unsigned int loff, nr_pages, req_pages;
1475 struct page *pages[PIPE_DEF_BUFFERS];
1476 struct partial_page partial[PIPE_DEF_BUFFERS];
1478 pgoff_t index, end_index;
1481 struct splice_pipe_desc spd = {
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(pipe, &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, umode_t 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,
1615 shmem_initxattrs, NULL);
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, umode_t mode)
1643 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1649 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t 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);
1723 drop_nlink(old_dir);
1724 } else if (they_are_dirs) {
1725 drop_nlink(old_dir);
1729 old_dir->i_size -= BOGO_DIRENT_SIZE;
1730 new_dir->i_size += BOGO_DIRENT_SIZE;
1731 old_dir->i_ctime = old_dir->i_mtime =
1732 new_dir->i_ctime = new_dir->i_mtime =
1733 inode->i_ctime = CURRENT_TIME;
1737 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1741 struct inode *inode;
1744 struct shmem_inode_info *info;
1746 len = strlen(symname) + 1;
1747 if (len > PAGE_CACHE_SIZE)
1748 return -ENAMETOOLONG;
1750 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1754 error = security_inode_init_security(inode, dir, &dentry->d_name,
1755 shmem_initxattrs, NULL);
1757 if (error != -EOPNOTSUPP) {
1764 info = SHMEM_I(inode);
1765 inode->i_size = len-1;
1766 if (len <= SHORT_SYMLINK_LEN) {
1767 info->symlink = kmemdup(symname, len, GFP_KERNEL);
1768 if (!info->symlink) {
1772 inode->i_op = &shmem_short_symlink_operations;
1774 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1779 inode->i_mapping->a_ops = &shmem_aops;
1780 inode->i_op = &shmem_symlink_inode_operations;
1781 kaddr = kmap_atomic(page);
1782 memcpy(kaddr, symname, len);
1783 kunmap_atomic(kaddr);
1784 SetPageUptodate(page);
1785 set_page_dirty(page);
1787 page_cache_release(page);
1789 dir->i_size += BOGO_DIRENT_SIZE;
1790 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1791 d_instantiate(dentry, inode);
1796 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
1798 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
1802 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1804 struct page *page = NULL;
1805 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1806 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
1812 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1814 if (!IS_ERR(nd_get_link(nd))) {
1815 struct page *page = cookie;
1817 mark_page_accessed(page);
1818 page_cache_release(page);
1822 #ifdef CONFIG_TMPFS_XATTR
1824 * Superblocks without xattr inode operations may get some security.* xattr
1825 * support from the LSM "for free". As soon as we have any other xattrs
1826 * like ACLs, we also need to implement the security.* handlers at
1827 * filesystem level, though.
1831 * Allocate new xattr and copy in the value; but leave the name to callers.
1833 static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size)
1835 struct shmem_xattr *new_xattr;
1839 len = sizeof(*new_xattr) + size;
1840 if (len <= sizeof(*new_xattr))
1843 new_xattr = kmalloc(len, GFP_KERNEL);
1847 new_xattr->size = size;
1848 memcpy(new_xattr->value, value, size);
1853 * Callback for security_inode_init_security() for acquiring xattrs.
1855 static int shmem_initxattrs(struct inode *inode,
1856 const struct xattr *xattr_array,
1859 struct shmem_inode_info *info = SHMEM_I(inode);
1860 const struct xattr *xattr;
1861 struct shmem_xattr *new_xattr;
1864 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
1865 new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len);
1869 len = strlen(xattr->name) + 1;
1870 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
1872 if (!new_xattr->name) {
1877 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
1878 XATTR_SECURITY_PREFIX_LEN);
1879 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
1882 spin_lock(&info->lock);
1883 list_add(&new_xattr->list, &info->xattr_list);
1884 spin_unlock(&info->lock);
1890 static int shmem_xattr_get(struct dentry *dentry, const char *name,
1891 void *buffer, size_t size)
1893 struct shmem_inode_info *info;
1894 struct shmem_xattr *xattr;
1897 info = SHMEM_I(dentry->d_inode);
1899 spin_lock(&info->lock);
1900 list_for_each_entry(xattr, &info->xattr_list, list) {
1901 if (strcmp(name, xattr->name))
1906 if (size < xattr->size)
1909 memcpy(buffer, xattr->value, xattr->size);
1913 spin_unlock(&info->lock);
1917 static int shmem_xattr_set(struct inode *inode, const char *name,
1918 const void *value, size_t size, int flags)
1920 struct shmem_inode_info *info = SHMEM_I(inode);
1921 struct shmem_xattr *xattr;
1922 struct shmem_xattr *new_xattr = NULL;
1925 /* value == NULL means remove */
1927 new_xattr = shmem_xattr_alloc(value, size);
1931 new_xattr->name = kstrdup(name, GFP_KERNEL);
1932 if (!new_xattr->name) {
1938 spin_lock(&info->lock);
1939 list_for_each_entry(xattr, &info->xattr_list, list) {
1940 if (!strcmp(name, xattr->name)) {
1941 if (flags & XATTR_CREATE) {
1944 } else if (new_xattr) {
1945 list_replace(&xattr->list, &new_xattr->list);
1947 list_del(&xattr->list);
1952 if (flags & XATTR_REPLACE) {
1956 list_add(&new_xattr->list, &info->xattr_list);
1960 spin_unlock(&info->lock);
1967 static const struct xattr_handler *shmem_xattr_handlers[] = {
1968 #ifdef CONFIG_TMPFS_POSIX_ACL
1969 &generic_acl_access_handler,
1970 &generic_acl_default_handler,
1975 static int shmem_xattr_validate(const char *name)
1977 struct { const char *prefix; size_t len; } arr[] = {
1978 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
1979 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
1983 for (i = 0; i < ARRAY_SIZE(arr); i++) {
1984 size_t preflen = arr[i].len;
1985 if (strncmp(name, arr[i].prefix, preflen) == 0) {
1994 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
1995 void *buffer, size_t size)
2000 * If this is a request for a synthetic attribute in the system.*
2001 * namespace use the generic infrastructure to resolve a handler
2002 * for it via sb->s_xattr.
2004 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2005 return generic_getxattr(dentry, name, buffer, size);
2007 err = shmem_xattr_validate(name);
2011 return shmem_xattr_get(dentry, name, buffer, size);
2014 static int shmem_setxattr(struct dentry *dentry, const char *name,
2015 const void *value, size_t size, int flags)
2020 * If this is a request for a synthetic attribute in the system.*
2021 * namespace use the generic infrastructure to resolve a handler
2022 * for it via sb->s_xattr.
2024 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2025 return generic_setxattr(dentry, name, value, size, flags);
2027 err = shmem_xattr_validate(name);
2032 value = ""; /* empty EA, do not remove */
2034 return shmem_xattr_set(dentry->d_inode, name, value, size, flags);
2038 static int shmem_removexattr(struct dentry *dentry, const char *name)
2043 * If this is a request for a synthetic attribute in the system.*
2044 * namespace use the generic infrastructure to resolve a handler
2045 * for it via sb->s_xattr.
2047 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2048 return generic_removexattr(dentry, name);
2050 err = shmem_xattr_validate(name);
2054 return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
2057 static bool xattr_is_trusted(const char *name)
2059 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
2062 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2064 bool trusted = capable(CAP_SYS_ADMIN);
2065 struct shmem_xattr *xattr;
2066 struct shmem_inode_info *info;
2069 info = SHMEM_I(dentry->d_inode);
2071 spin_lock(&info->lock);
2072 list_for_each_entry(xattr, &info->xattr_list, list) {
2075 /* skip "trusted." attributes for unprivileged callers */
2076 if (!trusted && xattr_is_trusted(xattr->name))
2079 len = strlen(xattr->name) + 1;
2086 memcpy(buffer, xattr->name, len);
2090 spin_unlock(&info->lock);
2094 #endif /* CONFIG_TMPFS_XATTR */
2096 static const struct inode_operations shmem_short_symlink_operations = {
2097 .readlink = generic_readlink,
2098 .follow_link = shmem_follow_short_symlink,
2099 #ifdef CONFIG_TMPFS_XATTR
2100 .setxattr = shmem_setxattr,
2101 .getxattr = shmem_getxattr,
2102 .listxattr = shmem_listxattr,
2103 .removexattr = shmem_removexattr,
2107 static const struct inode_operations shmem_symlink_inode_operations = {
2108 .readlink = generic_readlink,
2109 .follow_link = shmem_follow_link,
2110 .put_link = shmem_put_link,
2111 #ifdef CONFIG_TMPFS_XATTR
2112 .setxattr = shmem_setxattr,
2113 .getxattr = shmem_getxattr,
2114 .listxattr = shmem_listxattr,
2115 .removexattr = shmem_removexattr,
2119 static struct dentry *shmem_get_parent(struct dentry *child)
2121 return ERR_PTR(-ESTALE);
2124 static int shmem_match(struct inode *ino, void *vfh)
2128 inum = (inum << 32) | fh[1];
2129 return ino->i_ino == inum && fh[0] == ino->i_generation;
2132 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2133 struct fid *fid, int fh_len, int fh_type)
2135 struct inode *inode;
2136 struct dentry *dentry = NULL;
2137 u64 inum = fid->raw[2];
2138 inum = (inum << 32) | fid->raw[1];
2143 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2144 shmem_match, fid->raw);
2146 dentry = d_find_alias(inode);
2153 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2156 struct inode *inode = dentry->d_inode;
2163 if (inode_unhashed(inode)) {
2164 /* Unfortunately insert_inode_hash is not idempotent,
2165 * so as we hash inodes here rather than at creation
2166 * time, we need a lock to ensure we only try
2169 static DEFINE_SPINLOCK(lock);
2171 if (inode_unhashed(inode))
2172 __insert_inode_hash(inode,
2173 inode->i_ino + inode->i_generation);
2177 fh[0] = inode->i_generation;
2178 fh[1] = inode->i_ino;
2179 fh[2] = ((__u64)inode->i_ino) >> 32;
2185 static const struct export_operations shmem_export_ops = {
2186 .get_parent = shmem_get_parent,
2187 .encode_fh = shmem_encode_fh,
2188 .fh_to_dentry = shmem_fh_to_dentry,
2191 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2194 char *this_char, *value, *rest;
2198 while (options != NULL) {
2199 this_char = options;
2202 * NUL-terminate this option: unfortunately,
2203 * mount options form a comma-separated list,
2204 * but mpol's nodelist may also contain commas.
2206 options = strchr(options, ',');
2207 if (options == NULL)
2210 if (!isdigit(*options)) {
2217 if ((value = strchr(this_char,'=')) != NULL) {
2221 "tmpfs: No value for mount option '%s'\n",
2226 if (!strcmp(this_char,"size")) {
2227 unsigned long long size;
2228 size = memparse(value,&rest);
2230 size <<= PAGE_SHIFT;
2231 size *= totalram_pages;
2237 sbinfo->max_blocks =
2238 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2239 } else if (!strcmp(this_char,"nr_blocks")) {
2240 sbinfo->max_blocks = memparse(value, &rest);
2243 } else if (!strcmp(this_char,"nr_inodes")) {
2244 sbinfo->max_inodes = memparse(value, &rest);
2247 } else if (!strcmp(this_char,"mode")) {
2250 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2253 } else if (!strcmp(this_char,"uid")) {
2256 uid = simple_strtoul(value, &rest, 0);
2259 sbinfo->uid = make_kuid(current_user_ns(), uid);
2260 if (!uid_valid(sbinfo->uid))
2262 } else if (!strcmp(this_char,"gid")) {
2265 gid = simple_strtoul(value, &rest, 0);
2268 sbinfo->gid = make_kgid(current_user_ns(), gid);
2269 if (!gid_valid(sbinfo->gid))
2271 } else if (!strcmp(this_char,"mpol")) {
2272 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2275 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2283 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2289 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2291 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2292 struct shmem_sb_info config = *sbinfo;
2293 unsigned long inodes;
2294 int error = -EINVAL;
2296 if (shmem_parse_options(data, &config, true))
2299 spin_lock(&sbinfo->stat_lock);
2300 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2301 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2303 if (config.max_inodes < inodes)
2306 * Those tests disallow limited->unlimited while any are in use;
2307 * but we must separately disallow unlimited->limited, because
2308 * in that case we have no record of how much is already in use.
2310 if (config.max_blocks && !sbinfo->max_blocks)
2312 if (config.max_inodes && !sbinfo->max_inodes)
2316 sbinfo->max_blocks = config.max_blocks;
2317 sbinfo->max_inodes = config.max_inodes;
2318 sbinfo->free_inodes = config.max_inodes - inodes;
2320 mpol_put(sbinfo->mpol);
2321 sbinfo->mpol = config.mpol; /* transfers initial ref */
2323 spin_unlock(&sbinfo->stat_lock);
2327 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2329 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2331 if (sbinfo->max_blocks != shmem_default_max_blocks())
2332 seq_printf(seq, ",size=%luk",
2333 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2334 if (sbinfo->max_inodes != shmem_default_max_inodes())
2335 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2336 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2337 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2338 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2339 seq_printf(seq, ",uid=%u",
2340 from_kuid_munged(&init_user_ns, sbinfo->uid));
2341 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2342 seq_printf(seq, ",gid=%u",
2343 from_kgid_munged(&init_user_ns, sbinfo->gid));
2344 shmem_show_mpol(seq, sbinfo->mpol);
2347 #endif /* CONFIG_TMPFS */
2349 static void shmem_put_super(struct super_block *sb)
2351 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2353 percpu_counter_destroy(&sbinfo->used_blocks);
2355 sb->s_fs_info = NULL;
2358 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2360 struct inode *inode;
2361 struct shmem_sb_info *sbinfo;
2364 /* Round up to L1_CACHE_BYTES to resist false sharing */
2365 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2366 L1_CACHE_BYTES), GFP_KERNEL);
2370 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2371 sbinfo->uid = current_fsuid();
2372 sbinfo->gid = current_fsgid();
2373 sb->s_fs_info = sbinfo;
2377 * Per default we only allow half of the physical ram per
2378 * tmpfs instance, limiting inodes to one per page of lowmem;
2379 * but the internal instance is left unlimited.
2381 if (!(sb->s_flags & MS_NOUSER)) {
2382 sbinfo->max_blocks = shmem_default_max_blocks();
2383 sbinfo->max_inodes = shmem_default_max_inodes();
2384 if (shmem_parse_options(data, sbinfo, false)) {
2389 sb->s_export_op = &shmem_export_ops;
2390 sb->s_flags |= MS_NOSEC;
2392 sb->s_flags |= MS_NOUSER;
2395 spin_lock_init(&sbinfo->stat_lock);
2396 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2398 sbinfo->free_inodes = sbinfo->max_inodes;
2400 sb->s_maxbytes = MAX_LFS_FILESIZE;
2401 sb->s_blocksize = PAGE_CACHE_SIZE;
2402 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2403 sb->s_magic = TMPFS_MAGIC;
2404 sb->s_op = &shmem_ops;
2405 sb->s_time_gran = 1;
2406 #ifdef CONFIG_TMPFS_XATTR
2407 sb->s_xattr = shmem_xattr_handlers;
2409 #ifdef CONFIG_TMPFS_POSIX_ACL
2410 sb->s_flags |= MS_POSIXACL;
2413 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2416 inode->i_uid = sbinfo->uid;
2417 inode->i_gid = sbinfo->gid;
2418 sb->s_root = d_make_root(inode);
2424 shmem_put_super(sb);
2428 static struct kmem_cache *shmem_inode_cachep;
2430 static struct inode *shmem_alloc_inode(struct super_block *sb)
2432 struct shmem_inode_info *info;
2433 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2436 return &info->vfs_inode;
2439 static void shmem_destroy_callback(struct rcu_head *head)
2441 struct inode *inode = container_of(head, struct inode, i_rcu);
2442 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2445 static void shmem_destroy_inode(struct inode *inode)
2447 if (S_ISREG(inode->i_mode))
2448 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2449 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2452 static void shmem_init_inode(void *foo)
2454 struct shmem_inode_info *info = foo;
2455 inode_init_once(&info->vfs_inode);
2458 static int shmem_init_inodecache(void)
2460 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2461 sizeof(struct shmem_inode_info),
2462 0, SLAB_PANIC, shmem_init_inode);
2466 static void shmem_destroy_inodecache(void)
2468 kmem_cache_destroy(shmem_inode_cachep);
2471 static const struct address_space_operations shmem_aops = {
2472 .writepage = shmem_writepage,
2473 .set_page_dirty = __set_page_dirty_no_writeback,
2475 .write_begin = shmem_write_begin,
2476 .write_end = shmem_write_end,
2478 .migratepage = migrate_page,
2479 .error_remove_page = generic_error_remove_page,
2482 static const struct file_operations shmem_file_operations = {
2485 .llseek = generic_file_llseek,
2486 .read = do_sync_read,
2487 .write = do_sync_write,
2488 .aio_read = shmem_file_aio_read,
2489 .aio_write = generic_file_aio_write,
2490 .fsync = noop_fsync,
2491 .splice_read = shmem_file_splice_read,
2492 .splice_write = generic_file_splice_write,
2496 static const struct inode_operations shmem_inode_operations = {
2497 .setattr = shmem_setattr,
2498 .truncate_range = shmem_truncate_range,
2499 #ifdef CONFIG_TMPFS_XATTR
2500 .setxattr = shmem_setxattr,
2501 .getxattr = shmem_getxattr,
2502 .listxattr = shmem_listxattr,
2503 .removexattr = shmem_removexattr,
2507 static const struct inode_operations shmem_dir_inode_operations = {
2509 .create = shmem_create,
2510 .lookup = simple_lookup,
2512 .unlink = shmem_unlink,
2513 .symlink = shmem_symlink,
2514 .mkdir = shmem_mkdir,
2515 .rmdir = shmem_rmdir,
2516 .mknod = shmem_mknod,
2517 .rename = shmem_rename,
2519 #ifdef CONFIG_TMPFS_XATTR
2520 .setxattr = shmem_setxattr,
2521 .getxattr = shmem_getxattr,
2522 .listxattr = shmem_listxattr,
2523 .removexattr = shmem_removexattr,
2525 #ifdef CONFIG_TMPFS_POSIX_ACL
2526 .setattr = shmem_setattr,
2530 static const struct inode_operations shmem_special_inode_operations = {
2531 #ifdef CONFIG_TMPFS_XATTR
2532 .setxattr = shmem_setxattr,
2533 .getxattr = shmem_getxattr,
2534 .listxattr = shmem_listxattr,
2535 .removexattr = shmem_removexattr,
2537 #ifdef CONFIG_TMPFS_POSIX_ACL
2538 .setattr = shmem_setattr,
2542 static const struct super_operations shmem_ops = {
2543 .alloc_inode = shmem_alloc_inode,
2544 .destroy_inode = shmem_destroy_inode,
2546 .statfs = shmem_statfs,
2547 .remount_fs = shmem_remount_fs,
2548 .show_options = shmem_show_options,
2550 .evict_inode = shmem_evict_inode,
2551 .drop_inode = generic_delete_inode,
2552 .put_super = shmem_put_super,
2555 static const struct vm_operations_struct shmem_vm_ops = {
2556 .fault = shmem_fault,
2558 .set_policy = shmem_set_policy,
2559 .get_policy = shmem_get_policy,
2563 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2564 int flags, const char *dev_name, void *data)
2566 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2569 static struct file_system_type shmem_fs_type = {
2570 .owner = THIS_MODULE,
2572 .mount = shmem_mount,
2573 .kill_sb = kill_litter_super,
2576 int __init shmem_init(void)
2580 error = bdi_init(&shmem_backing_dev_info);
2584 error = shmem_init_inodecache();
2588 error = register_filesystem(&shmem_fs_type);
2590 printk(KERN_ERR "Could not register tmpfs\n");
2594 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2595 shmem_fs_type.name, NULL);
2596 if (IS_ERR(shm_mnt)) {
2597 error = PTR_ERR(shm_mnt);
2598 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2604 unregister_filesystem(&shmem_fs_type);
2606 shmem_destroy_inodecache();
2608 bdi_destroy(&shmem_backing_dev_info);
2610 shm_mnt = ERR_PTR(error);
2614 #else /* !CONFIG_SHMEM */
2617 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2619 * This is intended for small system where the benefits of the full
2620 * shmem code (swap-backed and resource-limited) are outweighed by
2621 * their complexity. On systems without swap this code should be
2622 * effectively equivalent, but much lighter weight.
2625 #include <linux/ramfs.h>
2627 static struct file_system_type shmem_fs_type = {
2629 .mount = ramfs_mount,
2630 .kill_sb = kill_litter_super,
2633 int __init shmem_init(void)
2635 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2637 shm_mnt = kern_mount(&shmem_fs_type);
2638 BUG_ON(IS_ERR(shm_mnt));
2643 int shmem_unuse(swp_entry_t swap, struct page *page)
2648 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2653 void shmem_unlock_mapping(struct address_space *mapping)
2657 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2659 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2661 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2663 #define shmem_vm_ops generic_file_vm_ops
2664 #define shmem_file_operations ramfs_file_operations
2665 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2666 #define shmem_acct_size(flags, size) 0
2667 #define shmem_unacct_size(flags, size) do {} while (0)
2669 #endif /* CONFIG_SHMEM */
2674 * shmem_file_setup - get an unlinked file living in tmpfs
2675 * @name: name for dentry (to be seen in /proc/<pid>/maps
2676 * @size: size to be set for the file
2677 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2679 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2683 struct inode *inode;
2685 struct dentry *root;
2688 if (IS_ERR(shm_mnt))
2689 return (void *)shm_mnt;
2691 if (size < 0 || size > MAX_LFS_FILESIZE)
2692 return ERR_PTR(-EINVAL);
2694 if (shmem_acct_size(flags, size))
2695 return ERR_PTR(-ENOMEM);
2699 this.len = strlen(name);
2700 this.hash = 0; /* will go */
2701 root = shm_mnt->mnt_root;
2702 path.dentry = d_alloc(root, &this);
2705 path.mnt = mntget(shm_mnt);
2708 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2712 d_instantiate(path.dentry, inode);
2713 inode->i_size = size;
2714 clear_nlink(inode); /* It is unlinked */
2716 error = ramfs_nommu_expand_for_mapping(inode, size);
2722 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2723 &shmem_file_operations);
2732 shmem_unacct_size(flags, size);
2733 return ERR_PTR(error);
2735 EXPORT_SYMBOL_GPL(shmem_file_setup);
2738 * shmem_zero_setup - setup a shared anonymous mapping
2739 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2741 int shmem_zero_setup(struct vm_area_struct *vma)
2744 loff_t size = vma->vm_end - vma->vm_start;
2746 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2748 return PTR_ERR(file);
2752 vma->vm_file = file;
2753 vma->vm_ops = &shmem_vm_ops;
2754 vma->vm_flags |= VM_CAN_NONLINEAR;
2759 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2760 * @mapping: the page's address_space
2761 * @index: the page index
2762 * @gfp: the page allocator flags to use if allocating
2764 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2765 * with any new page allocations done using the specified allocation flags.
2766 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2767 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2768 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2770 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2771 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2773 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2774 pgoff_t index, gfp_t gfp)
2777 struct inode *inode = mapping->host;
2781 BUG_ON(mapping->a_ops != &shmem_aops);
2782 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2784 page = ERR_PTR(error);
2790 * The tiny !SHMEM case uses ramfs without swap
2792 return read_cache_page_gfp(mapping, index, gfp);
2795 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);