2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/aio.h>
36 static struct vfsmount *shm_mnt;
40 * This virtual memory filesystem is heavily based on the ramfs. It
41 * extends ramfs by the ability to use swap and honor resource limits
42 * which makes it a completely usable filesystem.
45 #include <linux/xattr.h>
46 #include <linux/exportfs.h>
47 #include <linux/posix_acl.h>
48 #include <linux/posix_acl_xattr.h>
49 #include <linux/mman.h>
50 #include <linux/string.h>
51 #include <linux/slab.h>
52 #include <linux/backing-dev.h>
53 #include <linux/shmem_fs.h>
54 #include <linux/writeback.h>
55 #include <linux/blkdev.h>
56 #include <linux/pagevec.h>
57 #include <linux/percpu_counter.h>
58 #include <linux/falloc.h>
59 #include <linux/splice.h>
60 #include <linux/security.h>
61 #include <linux/swapops.h>
62 #include <linux/mempolicy.h>
63 #include <linux/namei.h>
64 #include <linux/ctype.h>
65 #include <linux/migrate.h>
66 #include <linux/highmem.h>
67 #include <linux/seq_file.h>
68 #include <linux/magic.h>
70 #include <asm/uaccess.h>
71 #include <asm/pgtable.h>
73 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
74 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
76 /* Pretend that each entry is of this size in directory's i_size */
77 #define BOGO_DIRENT_SIZE 20
79 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
80 #define SHORT_SYMLINK_LEN 128
83 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
84 * inode->i_private (with i_mutex making sure that it has only one user at
85 * a time): we would prefer not to enlarge the shmem inode just for that.
88 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
89 pgoff_t start; /* start of range currently being fallocated */
90 pgoff_t next; /* the next page offset to be fallocated */
91 pgoff_t nr_falloced; /* how many new pages have been fallocated */
92 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
95 /* Flag allocation requirements to shmem_getpage */
97 SGP_READ, /* don't exceed i_size, don't allocate page */
98 SGP_CACHE, /* don't exceed i_size, may allocate page */
99 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
100 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
101 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
105 static unsigned long shmem_default_max_blocks(void)
107 return totalram_pages / 2;
110 static unsigned long shmem_default_max_inodes(void)
112 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
116 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
117 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
118 struct shmem_inode_info *info, pgoff_t index);
119 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
120 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
122 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
123 struct page **pagep, enum sgp_type sgp, int *fault_type)
125 return shmem_getpage_gfp(inode, index, pagep, sgp,
126 mapping_gfp_mask(inode->i_mapping), fault_type);
129 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
131 return sb->s_fs_info;
135 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
136 * for shared memory and for shared anonymous (/dev/zero) mappings
137 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
138 * consistent with the pre-accounting of private mappings ...
140 static inline int shmem_acct_size(unsigned long flags, loff_t size)
142 return (flags & VM_NORESERVE) ?
143 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
146 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
148 if (!(flags & VM_NORESERVE))
149 vm_unacct_memory(VM_ACCT(size));
153 * ... whereas tmpfs objects are accounted incrementally as
154 * pages are allocated, in order to allow huge sparse files.
155 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
156 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
158 static inline int shmem_acct_block(unsigned long flags)
160 return (flags & VM_NORESERVE) ?
161 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
164 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
166 if (flags & VM_NORESERVE)
167 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
170 static const struct super_operations shmem_ops;
171 static const struct address_space_operations shmem_aops;
172 static const struct file_operations shmem_file_operations;
173 static const struct inode_operations shmem_inode_operations;
174 static const struct inode_operations shmem_dir_inode_operations;
175 static const struct inode_operations shmem_special_inode_operations;
176 static const struct vm_operations_struct shmem_vm_ops;
178 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
179 .ra_pages = 0, /* No readahead */
180 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
183 static LIST_HEAD(shmem_swaplist);
184 static DEFINE_MUTEX(shmem_swaplist_mutex);
186 static int shmem_reserve_inode(struct super_block *sb)
188 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
189 if (sbinfo->max_inodes) {
190 spin_lock(&sbinfo->stat_lock);
191 if (!sbinfo->free_inodes) {
192 spin_unlock(&sbinfo->stat_lock);
195 sbinfo->free_inodes--;
196 spin_unlock(&sbinfo->stat_lock);
201 static void shmem_free_inode(struct super_block *sb)
203 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
204 if (sbinfo->max_inodes) {
205 spin_lock(&sbinfo->stat_lock);
206 sbinfo->free_inodes++;
207 spin_unlock(&sbinfo->stat_lock);
212 * shmem_recalc_inode - recalculate the block usage of an inode
213 * @inode: inode to recalc
215 * We have to calculate the free blocks since the mm can drop
216 * undirtied hole pages behind our back.
218 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
219 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
221 * It has to be called with the spinlock held.
223 static void shmem_recalc_inode(struct inode *inode)
225 struct shmem_inode_info *info = SHMEM_I(inode);
228 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
230 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
231 if (sbinfo->max_blocks)
232 percpu_counter_add(&sbinfo->used_blocks, -freed);
233 info->alloced -= freed;
234 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
235 shmem_unacct_blocks(info->flags, freed);
240 * Replace item expected in radix tree by a new item, while holding tree lock.
242 static int shmem_radix_tree_replace(struct address_space *mapping,
243 pgoff_t index, void *expected, void *replacement)
248 VM_BUG_ON(!expected);
249 VM_BUG_ON(!replacement);
250 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
253 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
254 if (item != expected)
256 radix_tree_replace_slot(pslot, replacement);
261 * Sometimes, before we decide whether to proceed or to fail, we must check
262 * that an entry was not already brought back from swap by a racing thread.
264 * Checking page is not enough: by the time a SwapCache page is locked, it
265 * might be reused, and again be SwapCache, using the same swap as before.
267 static bool shmem_confirm_swap(struct address_space *mapping,
268 pgoff_t index, swp_entry_t swap)
273 item = radix_tree_lookup(&mapping->page_tree, index);
275 return item == swp_to_radix_entry(swap);
279 * Like add_to_page_cache_locked, but error if expected item has gone.
281 static int shmem_add_to_page_cache(struct page *page,
282 struct address_space *mapping,
283 pgoff_t index, gfp_t gfp, void *expected)
287 VM_BUG_ON_PAGE(!PageLocked(page), page);
288 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
290 page_cache_get(page);
291 page->mapping = mapping;
294 spin_lock_irq(&mapping->tree_lock);
296 error = radix_tree_insert(&mapping->page_tree, index, page);
298 error = shmem_radix_tree_replace(mapping, index, expected,
302 __inc_zone_page_state(page, NR_FILE_PAGES);
303 __inc_zone_page_state(page, NR_SHMEM);
304 spin_unlock_irq(&mapping->tree_lock);
306 page->mapping = NULL;
307 spin_unlock_irq(&mapping->tree_lock);
308 page_cache_release(page);
314 * Like delete_from_page_cache, but substitutes swap for page.
316 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
318 struct address_space *mapping = page->mapping;
321 spin_lock_irq(&mapping->tree_lock);
322 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
323 page->mapping = NULL;
325 __dec_zone_page_state(page, NR_FILE_PAGES);
326 __dec_zone_page_state(page, NR_SHMEM);
327 spin_unlock_irq(&mapping->tree_lock);
328 page_cache_release(page);
333 * Remove swap entry from radix tree, free the swap and its page cache.
335 static int shmem_free_swap(struct address_space *mapping,
336 pgoff_t index, void *radswap)
340 spin_lock_irq(&mapping->tree_lock);
341 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
342 spin_unlock_irq(&mapping->tree_lock);
345 free_swap_and_cache(radix_to_swp_entry(radswap));
350 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
352 void shmem_unlock_mapping(struct address_space *mapping)
355 pgoff_t indices[PAGEVEC_SIZE];
358 pagevec_init(&pvec, 0);
360 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
362 while (!mapping_unevictable(mapping)) {
364 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
365 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
367 pvec.nr = find_get_entries(mapping, index,
368 PAGEVEC_SIZE, pvec.pages, indices);
371 index = indices[pvec.nr - 1] + 1;
372 pagevec_remove_exceptionals(&pvec);
373 check_move_unevictable_pages(pvec.pages, pvec.nr);
374 pagevec_release(&pvec);
380 * Remove range of pages and swap entries from radix tree, and free them.
381 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
383 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
386 struct address_space *mapping = inode->i_mapping;
387 struct shmem_inode_info *info = SHMEM_I(inode);
388 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
389 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
390 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
391 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
393 pgoff_t indices[PAGEVEC_SIZE];
394 long nr_swaps_freed = 0;
399 end = -1; /* unsigned, so actually very big */
401 pagevec_init(&pvec, 0);
403 while (index < end) {
404 pvec.nr = find_get_entries(mapping, index,
405 min(end - index, (pgoff_t)PAGEVEC_SIZE),
406 pvec.pages, indices);
409 mem_cgroup_uncharge_start();
410 for (i = 0; i < pagevec_count(&pvec); i++) {
411 struct page *page = pvec.pages[i];
417 if (radix_tree_exceptional_entry(page)) {
420 nr_swaps_freed += !shmem_free_swap(mapping,
425 if (!trylock_page(page))
427 if (!unfalloc || !PageUptodate(page)) {
428 if (page->mapping == mapping) {
429 VM_BUG_ON_PAGE(PageWriteback(page), page);
430 truncate_inode_page(mapping, page);
435 pagevec_remove_exceptionals(&pvec);
436 pagevec_release(&pvec);
437 mem_cgroup_uncharge_end();
443 struct page *page = NULL;
444 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
446 unsigned int top = PAGE_CACHE_SIZE;
451 zero_user_segment(page, partial_start, top);
452 set_page_dirty(page);
454 page_cache_release(page);
458 struct page *page = NULL;
459 shmem_getpage(inode, end, &page, SGP_READ, NULL);
461 zero_user_segment(page, 0, partial_end);
462 set_page_dirty(page);
464 page_cache_release(page);
474 pvec.nr = find_get_entries(mapping, index,
475 min(end - index, (pgoff_t)PAGEVEC_SIZE),
476 pvec.pages, indices);
478 if (index == start || unfalloc)
483 if ((index == start || unfalloc) && indices[0] >= end) {
484 pagevec_remove_exceptionals(&pvec);
485 pagevec_release(&pvec);
488 mem_cgroup_uncharge_start();
489 for (i = 0; i < pagevec_count(&pvec); i++) {
490 struct page *page = pvec.pages[i];
496 if (radix_tree_exceptional_entry(page)) {
499 nr_swaps_freed += !shmem_free_swap(mapping,
505 if (!unfalloc || !PageUptodate(page)) {
506 if (page->mapping == mapping) {
507 VM_BUG_ON_PAGE(PageWriteback(page), page);
508 truncate_inode_page(mapping, page);
513 pagevec_remove_exceptionals(&pvec);
514 pagevec_release(&pvec);
515 mem_cgroup_uncharge_end();
519 spin_lock(&info->lock);
520 info->swapped -= nr_swaps_freed;
521 shmem_recalc_inode(inode);
522 spin_unlock(&info->lock);
525 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
527 shmem_undo_range(inode, lstart, lend, false);
528 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
530 EXPORT_SYMBOL_GPL(shmem_truncate_range);
532 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
534 struct inode *inode = dentry->d_inode;
537 error = inode_change_ok(inode, attr);
541 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
542 loff_t oldsize = inode->i_size;
543 loff_t newsize = attr->ia_size;
545 if (newsize != oldsize) {
546 i_size_write(inode, newsize);
547 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
549 if (newsize < oldsize) {
550 loff_t holebegin = round_up(newsize, PAGE_SIZE);
551 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
552 shmem_truncate_range(inode, newsize, (loff_t)-1);
553 /* unmap again to remove racily COWed private pages */
554 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
558 setattr_copy(inode, attr);
559 if (attr->ia_valid & ATTR_MODE)
560 error = posix_acl_chmod(inode, inode->i_mode);
564 static void shmem_evict_inode(struct inode *inode)
566 struct shmem_inode_info *info = SHMEM_I(inode);
568 if (inode->i_mapping->a_ops == &shmem_aops) {
569 shmem_unacct_size(info->flags, inode->i_size);
571 shmem_truncate_range(inode, 0, (loff_t)-1);
572 if (!list_empty(&info->swaplist)) {
573 mutex_lock(&shmem_swaplist_mutex);
574 list_del_init(&info->swaplist);
575 mutex_unlock(&shmem_swaplist_mutex);
578 kfree(info->symlink);
580 simple_xattrs_free(&info->xattrs);
581 WARN_ON(inode->i_blocks);
582 shmem_free_inode(inode->i_sb);
587 * If swap found in inode, free it and move page from swapcache to filecache.
589 static int shmem_unuse_inode(struct shmem_inode_info *info,
590 swp_entry_t swap, struct page **pagep)
592 struct address_space *mapping = info->vfs_inode.i_mapping;
598 radswap = swp_to_radix_entry(swap);
599 index = radix_tree_locate_item(&mapping->page_tree, radswap);
604 * Move _head_ to start search for next from here.
605 * But be careful: shmem_evict_inode checks list_empty without taking
606 * mutex, and there's an instant in list_move_tail when info->swaplist
607 * would appear empty, if it were the only one on shmem_swaplist.
609 if (shmem_swaplist.next != &info->swaplist)
610 list_move_tail(&shmem_swaplist, &info->swaplist);
612 gfp = mapping_gfp_mask(mapping);
613 if (shmem_should_replace_page(*pagep, gfp)) {
614 mutex_unlock(&shmem_swaplist_mutex);
615 error = shmem_replace_page(pagep, gfp, info, index);
616 mutex_lock(&shmem_swaplist_mutex);
618 * We needed to drop mutex to make that restrictive page
619 * allocation, but the inode might have been freed while we
620 * dropped it: although a racing shmem_evict_inode() cannot
621 * complete without emptying the radix_tree, our page lock
622 * on this swapcache page is not enough to prevent that -
623 * free_swap_and_cache() of our swap entry will only
624 * trylock_page(), removing swap from radix_tree whatever.
626 * We must not proceed to shmem_add_to_page_cache() if the
627 * inode has been freed, but of course we cannot rely on
628 * inode or mapping or info to check that. However, we can
629 * safely check if our swap entry is still in use (and here
630 * it can't have got reused for another page): if it's still
631 * in use, then the inode cannot have been freed yet, and we
632 * can safely proceed (if it's no longer in use, that tells
633 * nothing about the inode, but we don't need to unuse swap).
635 if (!page_swapcount(*pagep))
640 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
641 * but also to hold up shmem_evict_inode(): so inode cannot be freed
642 * beneath us (pagelock doesn't help until the page is in pagecache).
645 error = shmem_add_to_page_cache(*pagep, mapping, index,
646 GFP_NOWAIT, radswap);
647 if (error != -ENOMEM) {
649 * Truncation and eviction use free_swap_and_cache(), which
650 * only does trylock page: if we raced, best clean up here.
652 delete_from_swap_cache(*pagep);
653 set_page_dirty(*pagep);
655 spin_lock(&info->lock);
657 spin_unlock(&info->lock);
660 error = 1; /* not an error, but entry was found */
666 * Search through swapped inodes to find and replace swap by page.
668 int shmem_unuse(swp_entry_t swap, struct page *page)
670 struct list_head *this, *next;
671 struct shmem_inode_info *info;
676 * There's a faint possibility that swap page was replaced before
677 * caller locked it: caller will come back later with the right page.
679 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
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_charge_file(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);
709 page_cache_release(page);
714 * Move the page from the page cache to the swap cache.
716 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
718 struct shmem_inode_info *info;
719 struct address_space *mapping;
724 BUG_ON(!PageLocked(page));
725 mapping = page->mapping;
727 inode = mapping->host;
728 info = SHMEM_I(inode);
729 if (info->flags & VM_LOCKED)
731 if (!total_swap_pages)
735 * shmem_backing_dev_info's capabilities prevent regular writeback or
736 * sync from ever calling shmem_writepage; but a stacking filesystem
737 * might use ->writepage of its underlying filesystem, in which case
738 * tmpfs should write out to swap only in response to memory pressure,
739 * and not for the writeback threads or sync.
741 if (!wbc->for_reclaim) {
742 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
747 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
748 * value into swapfile.c, the only way we can correctly account for a
749 * fallocated page arriving here is now to initialize it and write it.
751 * That's okay for a page already fallocated earlier, but if we have
752 * not yet completed the fallocation, then (a) we want to keep track
753 * of this page in case we have to undo it, and (b) it may not be a
754 * good idea to continue anyway, once we're pushing into swap. So
755 * reactivate the page, and let shmem_fallocate() quit when too many.
757 if (!PageUptodate(page)) {
758 if (inode->i_private) {
759 struct shmem_falloc *shmem_falloc;
760 spin_lock(&inode->i_lock);
761 shmem_falloc = inode->i_private;
763 !shmem_falloc->waitq &&
764 index >= shmem_falloc->start &&
765 index < shmem_falloc->next)
766 shmem_falloc->nr_unswapped++;
769 spin_unlock(&inode->i_lock);
773 clear_highpage(page);
774 flush_dcache_page(page);
775 SetPageUptodate(page);
778 swap = get_swap_page();
783 * Add inode to shmem_unuse()'s list of swapped-out inodes,
784 * if it's not already there. Do it now before the page is
785 * moved to swap cache, when its pagelock no longer protects
786 * the inode from eviction. But don't unlock the mutex until
787 * we've incremented swapped, because shmem_unuse_inode() will
788 * prune a !swapped inode from the swaplist under this mutex.
790 mutex_lock(&shmem_swaplist_mutex);
791 if (list_empty(&info->swaplist))
792 list_add_tail(&info->swaplist, &shmem_swaplist);
794 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
795 swap_shmem_alloc(swap);
796 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
798 spin_lock(&info->lock);
800 shmem_recalc_inode(inode);
801 spin_unlock(&info->lock);
803 mutex_unlock(&shmem_swaplist_mutex);
804 BUG_ON(page_mapped(page));
805 swap_writepage(page, wbc);
809 mutex_unlock(&shmem_swaplist_mutex);
810 swapcache_free(swap, NULL);
812 set_page_dirty(page);
813 if (wbc->for_reclaim)
814 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
821 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
825 if (!mpol || mpol->mode == MPOL_DEFAULT)
826 return; /* show nothing */
828 mpol_to_str(buffer, sizeof(buffer), mpol);
830 seq_printf(seq, ",mpol=%s", buffer);
833 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
835 struct mempolicy *mpol = NULL;
837 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
840 spin_unlock(&sbinfo->stat_lock);
844 #endif /* CONFIG_TMPFS */
846 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
847 struct shmem_inode_info *info, pgoff_t index)
849 struct vm_area_struct pvma;
852 /* Create a pseudo vma that just contains the policy */
854 /* Bias interleave by inode number to distribute better across nodes */
855 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
857 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
859 page = swapin_readahead(swap, gfp, &pvma, 0);
861 /* Drop reference taken by mpol_shared_policy_lookup() */
862 mpol_cond_put(pvma.vm_policy);
867 static struct page *shmem_alloc_page(gfp_t gfp,
868 struct shmem_inode_info *info, pgoff_t index)
870 struct vm_area_struct pvma;
873 /* Create a pseudo vma that just contains the policy */
875 /* Bias interleave by inode number to distribute better across nodes */
876 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
878 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
880 page = alloc_page_vma(gfp, &pvma, 0);
882 /* Drop reference taken by mpol_shared_policy_lookup() */
883 mpol_cond_put(pvma.vm_policy);
887 #else /* !CONFIG_NUMA */
889 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
892 #endif /* CONFIG_TMPFS */
894 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
895 struct shmem_inode_info *info, pgoff_t index)
897 return swapin_readahead(swap, gfp, NULL, 0);
900 static inline struct page *shmem_alloc_page(gfp_t gfp,
901 struct shmem_inode_info *info, pgoff_t index)
903 return alloc_page(gfp);
905 #endif /* CONFIG_NUMA */
907 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
908 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
915 * When a page is moved from swapcache to shmem filecache (either by the
916 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
917 * shmem_unuse_inode()), it may have been read in earlier from swap, in
918 * ignorance of the mapping it belongs to. If that mapping has special
919 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
920 * we may need to copy to a suitable page before moving to filecache.
922 * In a future release, this may well be extended to respect cpuset and
923 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
924 * but for now it is a simple matter of zone.
926 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
928 return page_zonenum(page) > gfp_zone(gfp);
931 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
932 struct shmem_inode_info *info, pgoff_t index)
934 struct page *oldpage, *newpage;
935 struct address_space *swap_mapping;
940 swap_index = page_private(oldpage);
941 swap_mapping = page_mapping(oldpage);
944 * We have arrived here because our zones are constrained, so don't
945 * limit chance of success by further cpuset and node constraints.
947 gfp &= ~GFP_CONSTRAINT_MASK;
948 newpage = shmem_alloc_page(gfp, info, index);
952 page_cache_get(newpage);
953 copy_highpage(newpage, oldpage);
954 flush_dcache_page(newpage);
956 __set_page_locked(newpage);
957 SetPageUptodate(newpage);
958 SetPageSwapBacked(newpage);
959 set_page_private(newpage, swap_index);
960 SetPageSwapCache(newpage);
963 * Our caller will very soon move newpage out of swapcache, but it's
964 * a nice clean interface for us to replace oldpage by newpage there.
966 spin_lock_irq(&swap_mapping->tree_lock);
967 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
970 __inc_zone_page_state(newpage, NR_FILE_PAGES);
971 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
973 spin_unlock_irq(&swap_mapping->tree_lock);
975 if (unlikely(error)) {
977 * Is this possible? I think not, now that our callers check
978 * both PageSwapCache and page_private after getting page lock;
979 * but be defensive. Reverse old to newpage for clear and free.
983 mem_cgroup_replace_page_cache(oldpage, newpage);
984 lru_cache_add_anon(newpage);
988 ClearPageSwapCache(oldpage);
989 set_page_private(oldpage, 0);
991 unlock_page(oldpage);
992 page_cache_release(oldpage);
993 page_cache_release(oldpage);
998 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1000 * If we allocate a new one we do not mark it dirty. That's up to the
1001 * vm. If we swap it in we mark it dirty since we also free the swap
1002 * entry since a page cannot live in both the swap and page cache
1004 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1005 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1007 struct address_space *mapping = inode->i_mapping;
1008 struct shmem_inode_info *info;
1009 struct shmem_sb_info *sbinfo;
1016 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1020 page = find_lock_entry(mapping, index);
1021 if (radix_tree_exceptional_entry(page)) {
1022 swap = radix_to_swp_entry(page);
1026 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1027 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1032 if (page && sgp == SGP_WRITE)
1033 mark_page_accessed(page);
1035 /* fallocated page? */
1036 if (page && !PageUptodate(page)) {
1037 if (sgp != SGP_READ)
1040 page_cache_release(page);
1043 if (page || (sgp == SGP_READ && !swap.val)) {
1049 * Fast cache lookup did not find it:
1050 * bring it back from swap or allocate.
1052 info = SHMEM_I(inode);
1053 sbinfo = SHMEM_SB(inode->i_sb);
1056 /* Look it up and read it in.. */
1057 page = lookup_swap_cache(swap);
1059 /* here we actually do the io */
1061 *fault_type |= VM_FAULT_MAJOR;
1062 page = shmem_swapin(swap, gfp, info, index);
1069 /* We have to do this with page locked to prevent races */
1071 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1072 !shmem_confirm_swap(mapping, index, swap)) {
1073 error = -EEXIST; /* try again */
1076 if (!PageUptodate(page)) {
1080 wait_on_page_writeback(page);
1082 if (shmem_should_replace_page(page, gfp)) {
1083 error = shmem_replace_page(&page, gfp, info, index);
1088 error = mem_cgroup_charge_file(page, current->mm,
1089 gfp & GFP_RECLAIM_MASK);
1091 error = shmem_add_to_page_cache(page, mapping, index,
1092 gfp, swp_to_radix_entry(swap));
1094 * We already confirmed swap under page lock, and make
1095 * no memory allocation here, so usually no possibility
1096 * of error; but free_swap_and_cache() only trylocks a
1097 * page, so it is just possible that the entry has been
1098 * truncated or holepunched since swap was confirmed.
1099 * shmem_undo_range() will have done some of the
1100 * unaccounting, now delete_from_swap_cache() will do
1101 * the rest (including mem_cgroup_uncharge_swapcache).
1102 * Reset swap.val? No, leave it so "failed" goes back to
1103 * "repeat": reading a hole and writing should succeed.
1106 delete_from_swap_cache(page);
1111 spin_lock(&info->lock);
1113 shmem_recalc_inode(inode);
1114 spin_unlock(&info->lock);
1116 if (sgp == SGP_WRITE)
1117 mark_page_accessed(page);
1119 delete_from_swap_cache(page);
1120 set_page_dirty(page);
1124 if (shmem_acct_block(info->flags)) {
1128 if (sbinfo->max_blocks) {
1129 if (percpu_counter_compare(&sbinfo->used_blocks,
1130 sbinfo->max_blocks) >= 0) {
1134 percpu_counter_inc(&sbinfo->used_blocks);
1137 page = shmem_alloc_page(gfp, info, index);
1143 __SetPageSwapBacked(page);
1144 __set_page_locked(page);
1145 if (sgp == SGP_WRITE)
1146 init_page_accessed(page);
1148 error = mem_cgroup_charge_file(page, current->mm,
1149 gfp & GFP_RECLAIM_MASK);
1152 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
1154 error = shmem_add_to_page_cache(page, mapping, index,
1156 radix_tree_preload_end();
1159 mem_cgroup_uncharge_cache_page(page);
1162 lru_cache_add_anon(page);
1164 spin_lock(&info->lock);
1166 inode->i_blocks += BLOCKS_PER_PAGE;
1167 shmem_recalc_inode(inode);
1168 spin_unlock(&info->lock);
1172 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1174 if (sgp == SGP_FALLOC)
1178 * Let SGP_WRITE caller clear ends if write does not fill page;
1179 * but SGP_FALLOC on a page fallocated earlier must initialize
1180 * it now, lest undo on failure cancel our earlier guarantee.
1182 if (sgp != SGP_WRITE) {
1183 clear_highpage(page);
1184 flush_dcache_page(page);
1185 SetPageUptodate(page);
1187 if (sgp == SGP_DIRTY)
1188 set_page_dirty(page);
1191 /* Perhaps the file has been truncated since we checked */
1192 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1193 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1207 info = SHMEM_I(inode);
1208 ClearPageDirty(page);
1209 delete_from_page_cache(page);
1210 spin_lock(&info->lock);
1212 inode->i_blocks -= BLOCKS_PER_PAGE;
1213 spin_unlock(&info->lock);
1215 sbinfo = SHMEM_SB(inode->i_sb);
1216 if (sbinfo->max_blocks)
1217 percpu_counter_add(&sbinfo->used_blocks, -1);
1219 shmem_unacct_blocks(info->flags, 1);
1221 if (swap.val && error != -EINVAL &&
1222 !shmem_confirm_swap(mapping, index, swap))
1227 page_cache_release(page);
1229 if (error == -ENOSPC && !once++) {
1230 info = SHMEM_I(inode);
1231 spin_lock(&info->lock);
1232 shmem_recalc_inode(inode);
1233 spin_unlock(&info->lock);
1236 if (error == -EEXIST) /* from above or from radix_tree_insert */
1241 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1243 struct inode *inode = file_inode(vma->vm_file);
1245 int ret = VM_FAULT_LOCKED;
1248 * Trinity finds that probing a hole which tmpfs is punching can
1249 * prevent the hole-punch from ever completing: which in turn
1250 * locks writers out with its hold on i_mutex. So refrain from
1251 * faulting pages into the hole while it's being punched. Although
1252 * shmem_undo_range() does remove the additions, it may be unable to
1253 * keep up, as each new page needs its own unmap_mapping_range() call,
1254 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1256 * It does not matter if we sometimes reach this check just before the
1257 * hole-punch begins, so that one fault then races with the punch:
1258 * we just need to make racing faults a rare case.
1260 * The implementation below would be much simpler if we just used a
1261 * standard mutex or completion: but we cannot take i_mutex in fault,
1262 * and bloating every shmem inode for this unlikely case would be sad.
1264 if (unlikely(inode->i_private)) {
1265 struct shmem_falloc *shmem_falloc;
1267 spin_lock(&inode->i_lock);
1268 shmem_falloc = inode->i_private;
1270 shmem_falloc->waitq &&
1271 vmf->pgoff >= shmem_falloc->start &&
1272 vmf->pgoff < shmem_falloc->next) {
1273 wait_queue_head_t *shmem_falloc_waitq;
1274 DEFINE_WAIT(shmem_fault_wait);
1276 ret = VM_FAULT_NOPAGE;
1277 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1278 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1279 /* It's polite to up mmap_sem if we can */
1280 up_read(&vma->vm_mm->mmap_sem);
1281 ret = VM_FAULT_RETRY;
1284 shmem_falloc_waitq = shmem_falloc->waitq;
1285 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1286 TASK_UNINTERRUPTIBLE);
1287 spin_unlock(&inode->i_lock);
1291 * shmem_falloc_waitq points into the shmem_fallocate()
1292 * stack of the hole-punching task: shmem_falloc_waitq
1293 * is usually invalid by the time we reach here, but
1294 * finish_wait() does not dereference it in that case;
1295 * though i_lock needed lest racing with wake_up_all().
1297 spin_lock(&inode->i_lock);
1298 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1299 spin_unlock(&inode->i_lock);
1302 spin_unlock(&inode->i_lock);
1305 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1307 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1309 if (ret & VM_FAULT_MAJOR) {
1310 count_vm_event(PGMAJFAULT);
1311 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1317 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1319 struct inode *inode = file_inode(vma->vm_file);
1320 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1323 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1326 struct inode *inode = file_inode(vma->vm_file);
1329 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1330 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1334 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1336 struct inode *inode = file_inode(file);
1337 struct shmem_inode_info *info = SHMEM_I(inode);
1338 int retval = -ENOMEM;
1340 spin_lock(&info->lock);
1341 if (lock && !(info->flags & VM_LOCKED)) {
1342 if (!user_shm_lock(inode->i_size, user))
1344 info->flags |= VM_LOCKED;
1345 mapping_set_unevictable(file->f_mapping);
1347 if (!lock && (info->flags & VM_LOCKED) && user) {
1348 user_shm_unlock(inode->i_size, user);
1349 info->flags &= ~VM_LOCKED;
1350 mapping_clear_unevictable(file->f_mapping);
1355 spin_unlock(&info->lock);
1359 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1361 file_accessed(file);
1362 vma->vm_ops = &shmem_vm_ops;
1366 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1367 umode_t mode, dev_t dev, unsigned long flags)
1369 struct inode *inode;
1370 struct shmem_inode_info *info;
1371 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1373 if (shmem_reserve_inode(sb))
1376 inode = new_inode(sb);
1378 inode->i_ino = get_next_ino();
1379 inode_init_owner(inode, dir, mode);
1380 inode->i_blocks = 0;
1381 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1382 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1383 inode->i_generation = get_seconds();
1384 info = SHMEM_I(inode);
1385 memset(info, 0, (char *)inode - (char *)info);
1386 spin_lock_init(&info->lock);
1387 info->flags = flags & VM_NORESERVE;
1388 INIT_LIST_HEAD(&info->swaplist);
1389 simple_xattrs_init(&info->xattrs);
1390 cache_no_acl(inode);
1392 switch (mode & S_IFMT) {
1394 inode->i_op = &shmem_special_inode_operations;
1395 init_special_inode(inode, mode, dev);
1398 inode->i_mapping->a_ops = &shmem_aops;
1399 inode->i_op = &shmem_inode_operations;
1400 inode->i_fop = &shmem_file_operations;
1401 mpol_shared_policy_init(&info->policy,
1402 shmem_get_sbmpol(sbinfo));
1406 /* Some things misbehave if size == 0 on a directory */
1407 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1408 inode->i_op = &shmem_dir_inode_operations;
1409 inode->i_fop = &simple_dir_operations;
1413 * Must not load anything in the rbtree,
1414 * mpol_free_shared_policy will not be called.
1416 mpol_shared_policy_init(&info->policy, NULL);
1420 shmem_free_inode(sb);
1424 bool shmem_mapping(struct address_space *mapping)
1426 return mapping->backing_dev_info == &shmem_backing_dev_info;
1430 static const struct inode_operations shmem_symlink_inode_operations;
1431 static const struct inode_operations shmem_short_symlink_operations;
1433 #ifdef CONFIG_TMPFS_XATTR
1434 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1436 #define shmem_initxattrs NULL
1440 shmem_write_begin(struct file *file, struct address_space *mapping,
1441 loff_t pos, unsigned len, unsigned flags,
1442 struct page **pagep, void **fsdata)
1444 struct inode *inode = mapping->host;
1445 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1446 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1450 shmem_write_end(struct file *file, struct address_space *mapping,
1451 loff_t pos, unsigned len, unsigned copied,
1452 struct page *page, void *fsdata)
1454 struct inode *inode = mapping->host;
1456 if (pos + copied > inode->i_size)
1457 i_size_write(inode, pos + copied);
1459 if (!PageUptodate(page)) {
1460 if (copied < PAGE_CACHE_SIZE) {
1461 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1462 zero_user_segments(page, 0, from,
1463 from + copied, PAGE_CACHE_SIZE);
1465 SetPageUptodate(page);
1467 set_page_dirty(page);
1469 page_cache_release(page);
1474 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1476 struct file *file = iocb->ki_filp;
1477 struct inode *inode = file_inode(file);
1478 struct address_space *mapping = inode->i_mapping;
1480 unsigned long offset;
1481 enum sgp_type sgp = SGP_READ;
1484 loff_t *ppos = &iocb->ki_pos;
1487 * Might this read be for a stacking filesystem? Then when reading
1488 * holes of a sparse file, we actually need to allocate those pages,
1489 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1491 if (segment_eq(get_fs(), KERNEL_DS))
1494 index = *ppos >> PAGE_CACHE_SHIFT;
1495 offset = *ppos & ~PAGE_CACHE_MASK;
1498 struct page *page = NULL;
1500 unsigned long nr, ret;
1501 loff_t i_size = i_size_read(inode);
1503 end_index = i_size >> PAGE_CACHE_SHIFT;
1504 if (index > end_index)
1506 if (index == end_index) {
1507 nr = i_size & ~PAGE_CACHE_MASK;
1512 error = shmem_getpage(inode, index, &page, sgp, NULL);
1514 if (error == -EINVAL)
1522 * We must evaluate after, since reads (unlike writes)
1523 * are called without i_mutex protection against truncate
1525 nr = PAGE_CACHE_SIZE;
1526 i_size = i_size_read(inode);
1527 end_index = i_size >> PAGE_CACHE_SHIFT;
1528 if (index == end_index) {
1529 nr = i_size & ~PAGE_CACHE_MASK;
1532 page_cache_release(page);
1540 * If users can be writing to this page using arbitrary
1541 * virtual addresses, take care about potential aliasing
1542 * before reading the page on the kernel side.
1544 if (mapping_writably_mapped(mapping))
1545 flush_dcache_page(page);
1547 * Mark the page accessed if we read the beginning.
1550 mark_page_accessed(page);
1552 page = ZERO_PAGE(0);
1553 page_cache_get(page);
1557 * Ok, we have the page, and it's up-to-date, so
1558 * now we can copy it to user space...
1560 ret = copy_page_to_iter(page, offset, nr, to);
1563 index += offset >> PAGE_CACHE_SHIFT;
1564 offset &= ~PAGE_CACHE_MASK;
1566 page_cache_release(page);
1567 if (!iov_iter_count(to))
1576 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1577 file_accessed(file);
1578 return retval ? retval : error;
1581 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1582 struct pipe_inode_info *pipe, size_t len,
1585 struct address_space *mapping = in->f_mapping;
1586 struct inode *inode = mapping->host;
1587 unsigned int loff, nr_pages, req_pages;
1588 struct page *pages[PIPE_DEF_BUFFERS];
1589 struct partial_page partial[PIPE_DEF_BUFFERS];
1591 pgoff_t index, end_index;
1594 struct splice_pipe_desc spd = {
1597 .nr_pages_max = PIPE_DEF_BUFFERS,
1599 .ops = &page_cache_pipe_buf_ops,
1600 .spd_release = spd_release_page,
1603 isize = i_size_read(inode);
1604 if (unlikely(*ppos >= isize))
1607 left = isize - *ppos;
1608 if (unlikely(left < len))
1611 if (splice_grow_spd(pipe, &spd))
1614 index = *ppos >> PAGE_CACHE_SHIFT;
1615 loff = *ppos & ~PAGE_CACHE_MASK;
1616 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1617 nr_pages = min(req_pages, spd.nr_pages_max);
1619 spd.nr_pages = find_get_pages_contig(mapping, index,
1620 nr_pages, spd.pages);
1621 index += spd.nr_pages;
1624 while (spd.nr_pages < nr_pages) {
1625 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1629 spd.pages[spd.nr_pages++] = page;
1633 index = *ppos >> PAGE_CACHE_SHIFT;
1634 nr_pages = spd.nr_pages;
1637 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1638 unsigned int this_len;
1643 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1644 page = spd.pages[page_nr];
1646 if (!PageUptodate(page) || page->mapping != mapping) {
1647 error = shmem_getpage(inode, index, &page,
1652 page_cache_release(spd.pages[page_nr]);
1653 spd.pages[page_nr] = page;
1656 isize = i_size_read(inode);
1657 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1658 if (unlikely(!isize || index > end_index))
1661 if (end_index == index) {
1664 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1668 this_len = min(this_len, plen - loff);
1672 spd.partial[page_nr].offset = loff;
1673 spd.partial[page_nr].len = this_len;
1680 while (page_nr < nr_pages)
1681 page_cache_release(spd.pages[page_nr++]);
1684 error = splice_to_pipe(pipe, &spd);
1686 splice_shrink_spd(&spd);
1696 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1698 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1699 pgoff_t index, pgoff_t end, int whence)
1702 struct pagevec pvec;
1703 pgoff_t indices[PAGEVEC_SIZE];
1707 pagevec_init(&pvec, 0);
1708 pvec.nr = 1; /* start small: we may be there already */
1710 pvec.nr = find_get_entries(mapping, index,
1711 pvec.nr, pvec.pages, indices);
1713 if (whence == SEEK_DATA)
1717 for (i = 0; i < pvec.nr; i++, index++) {
1718 if (index < indices[i]) {
1719 if (whence == SEEK_HOLE) {
1725 page = pvec.pages[i];
1726 if (page && !radix_tree_exceptional_entry(page)) {
1727 if (!PageUptodate(page))
1731 (page && whence == SEEK_DATA) ||
1732 (!page && whence == SEEK_HOLE)) {
1737 pagevec_remove_exceptionals(&pvec);
1738 pagevec_release(&pvec);
1739 pvec.nr = PAGEVEC_SIZE;
1745 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
1747 struct address_space *mapping = file->f_mapping;
1748 struct inode *inode = mapping->host;
1752 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1753 return generic_file_llseek_size(file, offset, whence,
1754 MAX_LFS_FILESIZE, i_size_read(inode));
1755 mutex_lock(&inode->i_mutex);
1756 /* We're holding i_mutex so we can access i_size directly */
1760 else if (offset >= inode->i_size)
1763 start = offset >> PAGE_CACHE_SHIFT;
1764 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1765 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
1766 new_offset <<= PAGE_CACHE_SHIFT;
1767 if (new_offset > offset) {
1768 if (new_offset < inode->i_size)
1769 offset = new_offset;
1770 else if (whence == SEEK_DATA)
1773 offset = inode->i_size;
1778 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
1779 mutex_unlock(&inode->i_mutex);
1783 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1786 struct inode *inode = file_inode(file);
1787 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1788 struct shmem_falloc shmem_falloc;
1789 pgoff_t start, index, end;
1792 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1795 mutex_lock(&inode->i_mutex);
1797 if (mode & FALLOC_FL_PUNCH_HOLE) {
1798 struct address_space *mapping = file->f_mapping;
1799 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1800 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1801 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
1803 shmem_falloc.waitq = &shmem_falloc_waitq;
1804 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
1805 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
1806 spin_lock(&inode->i_lock);
1807 inode->i_private = &shmem_falloc;
1808 spin_unlock(&inode->i_lock);
1810 if ((u64)unmap_end > (u64)unmap_start)
1811 unmap_mapping_range(mapping, unmap_start,
1812 1 + unmap_end - unmap_start, 0);
1813 shmem_truncate_range(inode, offset, offset + len - 1);
1814 /* No need to unmap again: hole-punching leaves COWed pages */
1816 spin_lock(&inode->i_lock);
1817 inode->i_private = NULL;
1818 wake_up_all(&shmem_falloc_waitq);
1819 spin_unlock(&inode->i_lock);
1824 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1825 error = inode_newsize_ok(inode, offset + len);
1829 start = offset >> PAGE_CACHE_SHIFT;
1830 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1831 /* Try to avoid a swapstorm if len is impossible to satisfy */
1832 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1837 shmem_falloc.waitq = NULL;
1838 shmem_falloc.start = start;
1839 shmem_falloc.next = start;
1840 shmem_falloc.nr_falloced = 0;
1841 shmem_falloc.nr_unswapped = 0;
1842 spin_lock(&inode->i_lock);
1843 inode->i_private = &shmem_falloc;
1844 spin_unlock(&inode->i_lock);
1846 for (index = start; index < end; index++) {
1850 * Good, the fallocate(2) manpage permits EINTR: we may have
1851 * been interrupted because we are using up too much memory.
1853 if (signal_pending(current))
1855 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1858 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1861 /* Remove the !PageUptodate pages we added */
1862 shmem_undo_range(inode,
1863 (loff_t)start << PAGE_CACHE_SHIFT,
1864 (loff_t)index << PAGE_CACHE_SHIFT, true);
1869 * Inform shmem_writepage() how far we have reached.
1870 * No need for lock or barrier: we have the page lock.
1872 shmem_falloc.next++;
1873 if (!PageUptodate(page))
1874 shmem_falloc.nr_falloced++;
1877 * If !PageUptodate, leave it that way so that freeable pages
1878 * can be recognized if we need to rollback on error later.
1879 * But set_page_dirty so that memory pressure will swap rather
1880 * than free the pages we are allocating (and SGP_CACHE pages
1881 * might still be clean: we now need to mark those dirty too).
1883 set_page_dirty(page);
1885 page_cache_release(page);
1889 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1890 i_size_write(inode, offset + len);
1891 inode->i_ctime = CURRENT_TIME;
1893 spin_lock(&inode->i_lock);
1894 inode->i_private = NULL;
1895 spin_unlock(&inode->i_lock);
1897 mutex_unlock(&inode->i_mutex);
1901 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1903 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1905 buf->f_type = TMPFS_MAGIC;
1906 buf->f_bsize = PAGE_CACHE_SIZE;
1907 buf->f_namelen = NAME_MAX;
1908 if (sbinfo->max_blocks) {
1909 buf->f_blocks = sbinfo->max_blocks;
1911 buf->f_bfree = sbinfo->max_blocks -
1912 percpu_counter_sum(&sbinfo->used_blocks);
1914 if (sbinfo->max_inodes) {
1915 buf->f_files = sbinfo->max_inodes;
1916 buf->f_ffree = sbinfo->free_inodes;
1918 /* else leave those fields 0 like simple_statfs */
1923 * File creation. Allocate an inode, and we're done..
1926 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1928 struct inode *inode;
1929 int error = -ENOSPC;
1931 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1933 error = simple_acl_create(dir, inode);
1936 error = security_inode_init_security(inode, dir,
1938 shmem_initxattrs, NULL);
1939 if (error && error != -EOPNOTSUPP)
1943 dir->i_size += BOGO_DIRENT_SIZE;
1944 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1945 d_instantiate(dentry, inode);
1946 dget(dentry); /* Extra count - pin the dentry in core */
1955 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
1957 struct inode *inode;
1958 int error = -ENOSPC;
1960 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
1962 error = security_inode_init_security(inode, dir,
1964 shmem_initxattrs, NULL);
1965 if (error && error != -EOPNOTSUPP)
1967 error = simple_acl_create(dir, inode);
1970 d_tmpfile(dentry, inode);
1978 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1982 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1988 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1991 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1997 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1999 struct inode *inode = old_dentry->d_inode;
2003 * No ordinary (disk based) filesystem counts links as inodes;
2004 * but each new link needs a new dentry, pinning lowmem, and
2005 * tmpfs dentries cannot be pruned until they are unlinked.
2007 ret = shmem_reserve_inode(inode->i_sb);
2011 dir->i_size += BOGO_DIRENT_SIZE;
2012 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2014 ihold(inode); /* New dentry reference */
2015 dget(dentry); /* Extra pinning count for the created dentry */
2016 d_instantiate(dentry, inode);
2021 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2023 struct inode *inode = dentry->d_inode;
2025 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2026 shmem_free_inode(inode->i_sb);
2028 dir->i_size -= BOGO_DIRENT_SIZE;
2029 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2031 dput(dentry); /* Undo the count from "create" - this does all the work */
2035 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2037 if (!simple_empty(dentry))
2040 drop_nlink(dentry->d_inode);
2042 return shmem_unlink(dir, dentry);
2046 * The VFS layer already does all the dentry stuff for rename,
2047 * we just have to decrement the usage count for the target if
2048 * it exists so that the VFS layer correctly free's it when it
2051 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2053 struct inode *inode = old_dentry->d_inode;
2054 int they_are_dirs = S_ISDIR(inode->i_mode);
2056 if (!simple_empty(new_dentry))
2059 if (new_dentry->d_inode) {
2060 (void) shmem_unlink(new_dir, new_dentry);
2062 drop_nlink(old_dir);
2063 } else if (they_are_dirs) {
2064 drop_nlink(old_dir);
2068 old_dir->i_size -= BOGO_DIRENT_SIZE;
2069 new_dir->i_size += BOGO_DIRENT_SIZE;
2070 old_dir->i_ctime = old_dir->i_mtime =
2071 new_dir->i_ctime = new_dir->i_mtime =
2072 inode->i_ctime = CURRENT_TIME;
2076 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2080 struct inode *inode;
2083 struct shmem_inode_info *info;
2085 len = strlen(symname) + 1;
2086 if (len > PAGE_CACHE_SIZE)
2087 return -ENAMETOOLONG;
2089 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2093 error = security_inode_init_security(inode, dir, &dentry->d_name,
2094 shmem_initxattrs, NULL);
2096 if (error != -EOPNOTSUPP) {
2103 info = SHMEM_I(inode);
2104 inode->i_size = len-1;
2105 if (len <= SHORT_SYMLINK_LEN) {
2106 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2107 if (!info->symlink) {
2111 inode->i_op = &shmem_short_symlink_operations;
2113 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2118 inode->i_mapping->a_ops = &shmem_aops;
2119 inode->i_op = &shmem_symlink_inode_operations;
2120 kaddr = kmap_atomic(page);
2121 memcpy(kaddr, symname, len);
2122 kunmap_atomic(kaddr);
2123 SetPageUptodate(page);
2124 set_page_dirty(page);
2126 page_cache_release(page);
2128 dir->i_size += BOGO_DIRENT_SIZE;
2129 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2130 d_instantiate(dentry, inode);
2135 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2137 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2141 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2143 struct page *page = NULL;
2144 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2145 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2151 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2153 if (!IS_ERR(nd_get_link(nd))) {
2154 struct page *page = cookie;
2156 mark_page_accessed(page);
2157 page_cache_release(page);
2161 #ifdef CONFIG_TMPFS_XATTR
2163 * Superblocks without xattr inode operations may get some security.* xattr
2164 * support from the LSM "for free". As soon as we have any other xattrs
2165 * like ACLs, we also need to implement the security.* handlers at
2166 * filesystem level, though.
2170 * Callback for security_inode_init_security() for acquiring xattrs.
2172 static int shmem_initxattrs(struct inode *inode,
2173 const struct xattr *xattr_array,
2176 struct shmem_inode_info *info = SHMEM_I(inode);
2177 const struct xattr *xattr;
2178 struct simple_xattr *new_xattr;
2181 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2182 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
2186 len = strlen(xattr->name) + 1;
2187 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2189 if (!new_xattr->name) {
2194 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2195 XATTR_SECURITY_PREFIX_LEN);
2196 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2199 simple_xattr_list_add(&info->xattrs, new_xattr);
2205 static const struct xattr_handler *shmem_xattr_handlers[] = {
2206 #ifdef CONFIG_TMPFS_POSIX_ACL
2207 &posix_acl_access_xattr_handler,
2208 &posix_acl_default_xattr_handler,
2213 static int shmem_xattr_validate(const char *name)
2215 struct { const char *prefix; size_t len; } arr[] = {
2216 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2217 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2221 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2222 size_t preflen = arr[i].len;
2223 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2232 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2233 void *buffer, size_t size)
2235 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2239 * If this is a request for a synthetic attribute in the system.*
2240 * namespace use the generic infrastructure to resolve a handler
2241 * for it via sb->s_xattr.
2243 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2244 return generic_getxattr(dentry, name, buffer, size);
2246 err = shmem_xattr_validate(name);
2250 return simple_xattr_get(&info->xattrs, name, buffer, size);
2253 static int shmem_setxattr(struct dentry *dentry, const char *name,
2254 const void *value, size_t size, int flags)
2256 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2260 * If this is a request for a synthetic attribute in the system.*
2261 * namespace use the generic infrastructure to resolve a handler
2262 * for it via sb->s_xattr.
2264 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2265 return generic_setxattr(dentry, name, value, size, flags);
2267 err = shmem_xattr_validate(name);
2271 return simple_xattr_set(&info->xattrs, name, value, size, flags);
2274 static int shmem_removexattr(struct dentry *dentry, const char *name)
2276 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2280 * If this is a request for a synthetic attribute in the system.*
2281 * namespace use the generic infrastructure to resolve a handler
2282 * for it via sb->s_xattr.
2284 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2285 return generic_removexattr(dentry, name);
2287 err = shmem_xattr_validate(name);
2291 return simple_xattr_remove(&info->xattrs, name);
2294 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2296 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2297 return simple_xattr_list(&info->xattrs, buffer, size);
2299 #endif /* CONFIG_TMPFS_XATTR */
2301 static const struct inode_operations shmem_short_symlink_operations = {
2302 .readlink = generic_readlink,
2303 .follow_link = shmem_follow_short_symlink,
2304 #ifdef CONFIG_TMPFS_XATTR
2305 .setxattr = shmem_setxattr,
2306 .getxattr = shmem_getxattr,
2307 .listxattr = shmem_listxattr,
2308 .removexattr = shmem_removexattr,
2312 static const struct inode_operations shmem_symlink_inode_operations = {
2313 .readlink = generic_readlink,
2314 .follow_link = shmem_follow_link,
2315 .put_link = shmem_put_link,
2316 #ifdef CONFIG_TMPFS_XATTR
2317 .setxattr = shmem_setxattr,
2318 .getxattr = shmem_getxattr,
2319 .listxattr = shmem_listxattr,
2320 .removexattr = shmem_removexattr,
2324 static struct dentry *shmem_get_parent(struct dentry *child)
2326 return ERR_PTR(-ESTALE);
2329 static int shmem_match(struct inode *ino, void *vfh)
2333 inum = (inum << 32) | fh[1];
2334 return ino->i_ino == inum && fh[0] == ino->i_generation;
2337 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2338 struct fid *fid, int fh_len, int fh_type)
2340 struct inode *inode;
2341 struct dentry *dentry = NULL;
2348 inum = (inum << 32) | fid->raw[1];
2350 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2351 shmem_match, fid->raw);
2353 dentry = d_find_alias(inode);
2360 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2361 struct inode *parent)
2365 return FILEID_INVALID;
2368 if (inode_unhashed(inode)) {
2369 /* Unfortunately insert_inode_hash is not idempotent,
2370 * so as we hash inodes here rather than at creation
2371 * time, we need a lock to ensure we only try
2374 static DEFINE_SPINLOCK(lock);
2376 if (inode_unhashed(inode))
2377 __insert_inode_hash(inode,
2378 inode->i_ino + inode->i_generation);
2382 fh[0] = inode->i_generation;
2383 fh[1] = inode->i_ino;
2384 fh[2] = ((__u64)inode->i_ino) >> 32;
2390 static const struct export_operations shmem_export_ops = {
2391 .get_parent = shmem_get_parent,
2392 .encode_fh = shmem_encode_fh,
2393 .fh_to_dentry = shmem_fh_to_dentry,
2396 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2399 char *this_char, *value, *rest;
2400 struct mempolicy *mpol = NULL;
2404 while (options != NULL) {
2405 this_char = options;
2408 * NUL-terminate this option: unfortunately,
2409 * mount options form a comma-separated list,
2410 * but mpol's nodelist may also contain commas.
2412 options = strchr(options, ',');
2413 if (options == NULL)
2416 if (!isdigit(*options)) {
2423 if ((value = strchr(this_char,'=')) != NULL) {
2427 "tmpfs: No value for mount option '%s'\n",
2432 if (!strcmp(this_char,"size")) {
2433 unsigned long long size;
2434 size = memparse(value,&rest);
2436 size <<= PAGE_SHIFT;
2437 size *= totalram_pages;
2443 sbinfo->max_blocks =
2444 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2445 } else if (!strcmp(this_char,"nr_blocks")) {
2446 sbinfo->max_blocks = memparse(value, &rest);
2449 } else if (!strcmp(this_char,"nr_inodes")) {
2450 sbinfo->max_inodes = memparse(value, &rest);
2453 } else if (!strcmp(this_char,"mode")) {
2456 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2459 } else if (!strcmp(this_char,"uid")) {
2462 uid = simple_strtoul(value, &rest, 0);
2465 sbinfo->uid = make_kuid(current_user_ns(), uid);
2466 if (!uid_valid(sbinfo->uid))
2468 } else if (!strcmp(this_char,"gid")) {
2471 gid = simple_strtoul(value, &rest, 0);
2474 sbinfo->gid = make_kgid(current_user_ns(), gid);
2475 if (!gid_valid(sbinfo->gid))
2477 } else if (!strcmp(this_char,"mpol")) {
2480 if (mpol_parse_str(value, &mpol))
2483 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2488 sbinfo->mpol = mpol;
2492 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2500 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2502 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2503 struct shmem_sb_info config = *sbinfo;
2504 unsigned long inodes;
2505 int error = -EINVAL;
2508 if (shmem_parse_options(data, &config, true))
2511 spin_lock(&sbinfo->stat_lock);
2512 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2513 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2515 if (config.max_inodes < inodes)
2518 * Those tests disallow limited->unlimited while any are in use;
2519 * but we must separately disallow unlimited->limited, because
2520 * in that case we have no record of how much is already in use.
2522 if (config.max_blocks && !sbinfo->max_blocks)
2524 if (config.max_inodes && !sbinfo->max_inodes)
2528 sbinfo->max_blocks = config.max_blocks;
2529 sbinfo->max_inodes = config.max_inodes;
2530 sbinfo->free_inodes = config.max_inodes - inodes;
2533 * Preserve previous mempolicy unless mpol remount option was specified.
2536 mpol_put(sbinfo->mpol);
2537 sbinfo->mpol = config.mpol; /* transfers initial ref */
2540 spin_unlock(&sbinfo->stat_lock);
2544 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2546 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2548 if (sbinfo->max_blocks != shmem_default_max_blocks())
2549 seq_printf(seq, ",size=%luk",
2550 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2551 if (sbinfo->max_inodes != shmem_default_max_inodes())
2552 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2553 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2554 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2555 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2556 seq_printf(seq, ",uid=%u",
2557 from_kuid_munged(&init_user_ns, sbinfo->uid));
2558 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2559 seq_printf(seq, ",gid=%u",
2560 from_kgid_munged(&init_user_ns, sbinfo->gid));
2561 shmem_show_mpol(seq, sbinfo->mpol);
2564 #endif /* CONFIG_TMPFS */
2566 static void shmem_put_super(struct super_block *sb)
2568 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2570 percpu_counter_destroy(&sbinfo->used_blocks);
2571 mpol_put(sbinfo->mpol);
2573 sb->s_fs_info = NULL;
2576 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2578 struct inode *inode;
2579 struct shmem_sb_info *sbinfo;
2582 /* Round up to L1_CACHE_BYTES to resist false sharing */
2583 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2584 L1_CACHE_BYTES), GFP_KERNEL);
2588 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2589 sbinfo->uid = current_fsuid();
2590 sbinfo->gid = current_fsgid();
2591 sb->s_fs_info = sbinfo;
2595 * Per default we only allow half of the physical ram per
2596 * tmpfs instance, limiting inodes to one per page of lowmem;
2597 * but the internal instance is left unlimited.
2599 if (!(sb->s_flags & MS_KERNMOUNT)) {
2600 sbinfo->max_blocks = shmem_default_max_blocks();
2601 sbinfo->max_inodes = shmem_default_max_inodes();
2602 if (shmem_parse_options(data, sbinfo, false)) {
2607 sb->s_flags |= MS_NOUSER;
2609 sb->s_export_op = &shmem_export_ops;
2610 sb->s_flags |= MS_NOSEC;
2612 sb->s_flags |= MS_NOUSER;
2615 spin_lock_init(&sbinfo->stat_lock);
2616 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2618 sbinfo->free_inodes = sbinfo->max_inodes;
2620 sb->s_maxbytes = MAX_LFS_FILESIZE;
2621 sb->s_blocksize = PAGE_CACHE_SIZE;
2622 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2623 sb->s_magic = TMPFS_MAGIC;
2624 sb->s_op = &shmem_ops;
2625 sb->s_time_gran = 1;
2626 #ifdef CONFIG_TMPFS_XATTR
2627 sb->s_xattr = shmem_xattr_handlers;
2629 #ifdef CONFIG_TMPFS_POSIX_ACL
2630 sb->s_flags |= MS_POSIXACL;
2633 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2636 inode->i_uid = sbinfo->uid;
2637 inode->i_gid = sbinfo->gid;
2638 sb->s_root = d_make_root(inode);
2644 shmem_put_super(sb);
2648 static struct kmem_cache *shmem_inode_cachep;
2650 static struct inode *shmem_alloc_inode(struct super_block *sb)
2652 struct shmem_inode_info *info;
2653 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2656 return &info->vfs_inode;
2659 static void shmem_destroy_callback(struct rcu_head *head)
2661 struct inode *inode = container_of(head, struct inode, i_rcu);
2662 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2665 static void shmem_destroy_inode(struct inode *inode)
2667 if (S_ISREG(inode->i_mode))
2668 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2669 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2672 static void shmem_init_inode(void *foo)
2674 struct shmem_inode_info *info = foo;
2675 inode_init_once(&info->vfs_inode);
2678 static int shmem_init_inodecache(void)
2680 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2681 sizeof(struct shmem_inode_info),
2682 0, SLAB_PANIC, shmem_init_inode);
2686 static void shmem_destroy_inodecache(void)
2688 kmem_cache_destroy(shmem_inode_cachep);
2691 static const struct address_space_operations shmem_aops = {
2692 .writepage = shmem_writepage,
2693 .set_page_dirty = __set_page_dirty_no_writeback,
2695 .write_begin = shmem_write_begin,
2696 .write_end = shmem_write_end,
2698 .migratepage = migrate_page,
2699 .error_remove_page = generic_error_remove_page,
2702 static const struct file_operations shmem_file_operations = {
2705 .llseek = shmem_file_llseek,
2706 .read = new_sync_read,
2707 .write = new_sync_write,
2708 .read_iter = shmem_file_read_iter,
2709 .write_iter = generic_file_write_iter,
2710 .fsync = noop_fsync,
2711 .splice_read = shmem_file_splice_read,
2712 .splice_write = iter_file_splice_write,
2713 .fallocate = shmem_fallocate,
2717 static const struct inode_operations shmem_inode_operations = {
2718 .setattr = shmem_setattr,
2719 #ifdef CONFIG_TMPFS_XATTR
2720 .setxattr = shmem_setxattr,
2721 .getxattr = shmem_getxattr,
2722 .listxattr = shmem_listxattr,
2723 .removexattr = shmem_removexattr,
2724 .set_acl = simple_set_acl,
2728 static const struct inode_operations shmem_dir_inode_operations = {
2730 .create = shmem_create,
2731 .lookup = simple_lookup,
2733 .unlink = shmem_unlink,
2734 .symlink = shmem_symlink,
2735 .mkdir = shmem_mkdir,
2736 .rmdir = shmem_rmdir,
2737 .mknod = shmem_mknod,
2738 .rename = shmem_rename,
2739 .tmpfile = shmem_tmpfile,
2741 #ifdef CONFIG_TMPFS_XATTR
2742 .setxattr = shmem_setxattr,
2743 .getxattr = shmem_getxattr,
2744 .listxattr = shmem_listxattr,
2745 .removexattr = shmem_removexattr,
2747 #ifdef CONFIG_TMPFS_POSIX_ACL
2748 .setattr = shmem_setattr,
2749 .set_acl = simple_set_acl,
2753 static const struct inode_operations shmem_special_inode_operations = {
2754 #ifdef CONFIG_TMPFS_XATTR
2755 .setxattr = shmem_setxattr,
2756 .getxattr = shmem_getxattr,
2757 .listxattr = shmem_listxattr,
2758 .removexattr = shmem_removexattr,
2760 #ifdef CONFIG_TMPFS_POSIX_ACL
2761 .setattr = shmem_setattr,
2762 .set_acl = simple_set_acl,
2766 static const struct super_operations shmem_ops = {
2767 .alloc_inode = shmem_alloc_inode,
2768 .destroy_inode = shmem_destroy_inode,
2770 .statfs = shmem_statfs,
2771 .remount_fs = shmem_remount_fs,
2772 .show_options = shmem_show_options,
2774 .evict_inode = shmem_evict_inode,
2775 .drop_inode = generic_delete_inode,
2776 .put_super = shmem_put_super,
2779 static const struct vm_operations_struct shmem_vm_ops = {
2780 .fault = shmem_fault,
2781 .map_pages = filemap_map_pages,
2783 .set_policy = shmem_set_policy,
2784 .get_policy = shmem_get_policy,
2786 .remap_pages = generic_file_remap_pages,
2789 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2790 int flags, const char *dev_name, void *data)
2792 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2795 static struct file_system_type shmem_fs_type = {
2796 .owner = THIS_MODULE,
2798 .mount = shmem_mount,
2799 .kill_sb = kill_litter_super,
2800 .fs_flags = FS_USERNS_MOUNT,
2803 int __init shmem_init(void)
2807 /* If rootfs called this, don't re-init */
2808 if (shmem_inode_cachep)
2811 error = bdi_init(&shmem_backing_dev_info);
2815 error = shmem_init_inodecache();
2819 error = register_filesystem(&shmem_fs_type);
2821 printk(KERN_ERR "Could not register tmpfs\n");
2825 shm_mnt = kern_mount(&shmem_fs_type);
2826 if (IS_ERR(shm_mnt)) {
2827 error = PTR_ERR(shm_mnt);
2828 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2834 unregister_filesystem(&shmem_fs_type);
2836 shmem_destroy_inodecache();
2838 bdi_destroy(&shmem_backing_dev_info);
2840 shm_mnt = ERR_PTR(error);
2844 #else /* !CONFIG_SHMEM */
2847 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2849 * This is intended for small system where the benefits of the full
2850 * shmem code (swap-backed and resource-limited) are outweighed by
2851 * their complexity. On systems without swap this code should be
2852 * effectively equivalent, but much lighter weight.
2855 static struct file_system_type shmem_fs_type = {
2857 .mount = ramfs_mount,
2858 .kill_sb = kill_litter_super,
2859 .fs_flags = FS_USERNS_MOUNT,
2862 int __init shmem_init(void)
2864 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2866 shm_mnt = kern_mount(&shmem_fs_type);
2867 BUG_ON(IS_ERR(shm_mnt));
2872 int shmem_unuse(swp_entry_t swap, struct page *page)
2877 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2882 void shmem_unlock_mapping(struct address_space *mapping)
2886 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2888 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2890 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2892 #define shmem_vm_ops generic_file_vm_ops
2893 #define shmem_file_operations ramfs_file_operations
2894 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2895 #define shmem_acct_size(flags, size) 0
2896 #define shmem_unacct_size(flags, size) do {} while (0)
2898 #endif /* CONFIG_SHMEM */
2902 static struct dentry_operations anon_ops = {
2903 .d_dname = simple_dname
2906 static struct file *__shmem_file_setup(const char *name, loff_t size,
2907 unsigned long flags, unsigned int i_flags)
2910 struct inode *inode;
2912 struct super_block *sb;
2915 if (IS_ERR(shm_mnt))
2916 return ERR_CAST(shm_mnt);
2918 if (size < 0 || size > MAX_LFS_FILESIZE)
2919 return ERR_PTR(-EINVAL);
2921 if (shmem_acct_size(flags, size))
2922 return ERR_PTR(-ENOMEM);
2924 res = ERR_PTR(-ENOMEM);
2926 this.len = strlen(name);
2927 this.hash = 0; /* will go */
2928 sb = shm_mnt->mnt_sb;
2929 path.dentry = d_alloc_pseudo(sb, &this);
2932 d_set_d_op(path.dentry, &anon_ops);
2933 path.mnt = mntget(shm_mnt);
2935 res = ERR_PTR(-ENOSPC);
2936 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2940 inode->i_flags |= i_flags;
2941 d_instantiate(path.dentry, inode);
2942 inode->i_size = size;
2943 clear_nlink(inode); /* It is unlinked */
2944 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
2948 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2949 &shmem_file_operations);
2958 shmem_unacct_size(flags, size);
2963 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
2964 * kernel internal. There will be NO LSM permission checks against the
2965 * underlying inode. So users of this interface must do LSM checks at a
2966 * higher layer. The one user is the big_key implementation. LSM checks
2967 * are provided at the key level rather than the inode level.
2968 * @name: name for dentry (to be seen in /proc/<pid>/maps
2969 * @size: size to be set for the file
2970 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2972 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
2974 return __shmem_file_setup(name, size, flags, S_PRIVATE);
2978 * shmem_file_setup - get an unlinked file living in tmpfs
2979 * @name: name for dentry (to be seen in /proc/<pid>/maps
2980 * @size: size to be set for the file
2981 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2983 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2985 return __shmem_file_setup(name, size, flags, 0);
2987 EXPORT_SYMBOL_GPL(shmem_file_setup);
2990 * shmem_zero_setup - setup a shared anonymous mapping
2991 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2993 int shmem_zero_setup(struct vm_area_struct *vma)
2996 loff_t size = vma->vm_end - vma->vm_start;
2998 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3000 return PTR_ERR(file);
3004 vma->vm_file = file;
3005 vma->vm_ops = &shmem_vm_ops;
3010 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3011 * @mapping: the page's address_space
3012 * @index: the page index
3013 * @gfp: the page allocator flags to use if allocating
3015 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3016 * with any new page allocations done using the specified allocation flags.
3017 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3018 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3019 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3021 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3022 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3024 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3025 pgoff_t index, gfp_t gfp)
3028 struct inode *inode = mapping->host;
3032 BUG_ON(mapping->a_ops != &shmem_aops);
3033 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3035 page = ERR_PTR(error);
3041 * The tiny !SHMEM case uses ramfs without swap
3043 return read_cache_page_gfp(mapping, index, gfp);
3046 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);