Merge branch 'docs-next' of git://git.lwn.net/linux-2.6
[linux-2.6/next.git] / mm / shmem.c
blob356dd99566ecb671cd324ba3dddcee10441312f0
1 /*
2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
5 * 2000 Transmeta Corp.
6 * 2000-2001 Christoph Rohland
7 * 2000-2001 SAP AG
8 * 2002 Red Hat Inc.
9 * Copyright (C) 2002-2005 Hugh Dickins.
10 * Copyright (C) 2002-2005 VERITAS Software Corporation.
11 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 * Extended attribute support for tmpfs:
14 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
15 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17 * tiny-shmem:
18 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 * This file is released under the GPL.
23 #include <linux/fs.h>
24 #include <linux/init.h>
25 #include <linux/vfs.h>
26 #include <linux/mount.h>
27 #include <linux/pagemap.h>
28 #include <linux/file.h>
29 #include <linux/mm.h>
30 #include <linux/module.h>
31 #include <linux/swap.h>
32 #include <linux/ima.h>
34 static struct vfsmount *shm_mnt;
36 #ifdef CONFIG_SHMEM
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/generic_acl.h>
46 #include <linux/mman.h>
47 #include <linux/string.h>
48 #include <linux/slab.h>
49 #include <linux/backing-dev.h>
50 #include <linux/shmem_fs.h>
51 #include <linux/writeback.h>
52 #include <linux/blkdev.h>
53 #include <linux/security.h>
54 #include <linux/swapops.h>
55 #include <linux/mempolicy.h>
56 #include <linux/namei.h>
57 #include <linux/ctype.h>
58 #include <linux/migrate.h>
59 #include <linux/highmem.h>
60 #include <linux/seq_file.h>
61 #include <linux/magic.h>
63 #include <asm/uaccess.h>
64 #include <asm/div64.h>
65 #include <asm/pgtable.h>
68 * The maximum size of a shmem/tmpfs file is limited by the maximum size of
69 * its triple-indirect swap vector - see illustration at shmem_swp_entry().
71 * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
72 * but one eighth of that on a 64-bit kernel. With 8kB page size, maximum
73 * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
74 * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
76 * We use / and * instead of shifts in the definitions below, so that the swap
77 * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
79 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
80 #define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
82 #define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
83 #define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
85 #define SHMEM_MAX_BYTES min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
86 #define SHMEM_MAX_INDEX ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
88 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
89 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
91 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
92 #define SHMEM_PAGEIN VM_READ
93 #define SHMEM_TRUNCATE VM_WRITE
95 /* Definition to limit shmem_truncate's steps between cond_rescheds */
96 #define LATENCY_LIMIT 64
98 /* Pretend that each entry is of this size in directory's i_size */
99 #define BOGO_DIRENT_SIZE 20
101 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
102 enum sgp_type {
103 SGP_READ, /* don't exceed i_size, don't allocate page */
104 SGP_CACHE, /* don't exceed i_size, may allocate page */
105 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
106 SGP_WRITE, /* may exceed i_size, may allocate page */
109 #ifdef CONFIG_TMPFS
110 static unsigned long shmem_default_max_blocks(void)
112 return totalram_pages / 2;
115 static unsigned long shmem_default_max_inodes(void)
117 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
119 #endif
121 static int shmem_getpage(struct inode *inode, unsigned long idx,
122 struct page **pagep, enum sgp_type sgp, int *type);
124 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
127 * The above definition of ENTRIES_PER_PAGE, and the use of
128 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
129 * might be reconsidered if it ever diverges from PAGE_SIZE.
131 * Mobility flags are masked out as swap vectors cannot move
133 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
134 PAGE_CACHE_SHIFT-PAGE_SHIFT);
137 static inline void shmem_dir_free(struct page *page)
139 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
142 static struct page **shmem_dir_map(struct page *page)
144 return (struct page **)kmap_atomic(page, KM_USER0);
147 static inline void shmem_dir_unmap(struct page **dir)
149 kunmap_atomic(dir, KM_USER0);
152 static swp_entry_t *shmem_swp_map(struct page *page)
154 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
157 static inline void shmem_swp_balance_unmap(void)
160 * When passing a pointer to an i_direct entry, to code which
161 * also handles indirect entries and so will shmem_swp_unmap,
162 * we must arrange for the preempt count to remain in balance.
163 * What kmap_atomic of a lowmem page does depends on config
164 * and architecture, so pretend to kmap_atomic some lowmem page.
166 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
169 static inline void shmem_swp_unmap(swp_entry_t *entry)
171 kunmap_atomic(entry, KM_USER1);
174 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
176 return sb->s_fs_info;
180 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
181 * for shared memory and for shared anonymous (/dev/zero) mappings
182 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
183 * consistent with the pre-accounting of private mappings ...
185 static inline int shmem_acct_size(unsigned long flags, loff_t size)
187 return (flags & VM_NORESERVE) ?
188 0 : security_vm_enough_memory_kern(VM_ACCT(size));
191 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
193 if (!(flags & VM_NORESERVE))
194 vm_unacct_memory(VM_ACCT(size));
198 * ... whereas tmpfs objects are accounted incrementally as
199 * pages are allocated, in order to allow huge sparse files.
200 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
201 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
203 static inline int shmem_acct_block(unsigned long flags)
205 return (flags & VM_NORESERVE) ?
206 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
209 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
211 if (flags & VM_NORESERVE)
212 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
215 static const struct super_operations shmem_ops;
216 static const struct address_space_operations shmem_aops;
217 static const struct file_operations shmem_file_operations;
218 static const struct inode_operations shmem_inode_operations;
219 static const struct inode_operations shmem_dir_inode_operations;
220 static const struct inode_operations shmem_special_inode_operations;
221 static const struct vm_operations_struct shmem_vm_ops;
223 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
224 .ra_pages = 0, /* No readahead */
225 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
226 .unplug_io_fn = default_unplug_io_fn,
229 static LIST_HEAD(shmem_swaplist);
230 static DEFINE_MUTEX(shmem_swaplist_mutex);
232 static void shmem_free_blocks(struct inode *inode, long pages)
234 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
235 if (sbinfo->max_blocks) {
236 spin_lock(&sbinfo->stat_lock);
237 sbinfo->free_blocks += pages;
238 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
239 spin_unlock(&sbinfo->stat_lock);
243 static int shmem_reserve_inode(struct super_block *sb)
245 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
246 if (sbinfo->max_inodes) {
247 spin_lock(&sbinfo->stat_lock);
248 if (!sbinfo->free_inodes) {
249 spin_unlock(&sbinfo->stat_lock);
250 return -ENOSPC;
252 sbinfo->free_inodes--;
253 spin_unlock(&sbinfo->stat_lock);
255 return 0;
258 static void shmem_free_inode(struct super_block *sb)
260 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
261 if (sbinfo->max_inodes) {
262 spin_lock(&sbinfo->stat_lock);
263 sbinfo->free_inodes++;
264 spin_unlock(&sbinfo->stat_lock);
269 * shmem_recalc_inode - recalculate the size of an inode
270 * @inode: inode to recalc
272 * We have to calculate the free blocks since the mm can drop
273 * undirtied hole pages behind our back.
275 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
276 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
278 * It has to be called with the spinlock held.
280 static void shmem_recalc_inode(struct inode *inode)
282 struct shmem_inode_info *info = SHMEM_I(inode);
283 long freed;
285 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
286 if (freed > 0) {
287 info->alloced -= freed;
288 shmem_unacct_blocks(info->flags, freed);
289 shmem_free_blocks(inode, freed);
294 * shmem_swp_entry - find the swap vector position in the info structure
295 * @info: info structure for the inode
296 * @index: index of the page to find
297 * @page: optional page to add to the structure. Has to be preset to
298 * all zeros
300 * If there is no space allocated yet it will return NULL when
301 * page is NULL, else it will use the page for the needed block,
302 * setting it to NULL on return to indicate that it has been used.
304 * The swap vector is organized the following way:
306 * There are SHMEM_NR_DIRECT entries directly stored in the
307 * shmem_inode_info structure. So small files do not need an addional
308 * allocation.
310 * For pages with index > SHMEM_NR_DIRECT there is the pointer
311 * i_indirect which points to a page which holds in the first half
312 * doubly indirect blocks, in the second half triple indirect blocks:
314 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
315 * following layout (for SHMEM_NR_DIRECT == 16):
317 * i_indirect -> dir --> 16-19
318 * | +-> 20-23
320 * +-->dir2 --> 24-27
321 * | +-> 28-31
322 * | +-> 32-35
323 * | +-> 36-39
325 * +-->dir3 --> 40-43
326 * +-> 44-47
327 * +-> 48-51
328 * +-> 52-55
330 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
332 unsigned long offset;
333 struct page **dir;
334 struct page *subdir;
336 if (index < SHMEM_NR_DIRECT) {
337 shmem_swp_balance_unmap();
338 return info->i_direct+index;
340 if (!info->i_indirect) {
341 if (page) {
342 info->i_indirect = *page;
343 *page = NULL;
345 return NULL; /* need another page */
348 index -= SHMEM_NR_DIRECT;
349 offset = index % ENTRIES_PER_PAGE;
350 index /= ENTRIES_PER_PAGE;
351 dir = shmem_dir_map(info->i_indirect);
353 if (index >= ENTRIES_PER_PAGE/2) {
354 index -= ENTRIES_PER_PAGE/2;
355 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
356 index %= ENTRIES_PER_PAGE;
357 subdir = *dir;
358 if (!subdir) {
359 if (page) {
360 *dir = *page;
361 *page = NULL;
363 shmem_dir_unmap(dir);
364 return NULL; /* need another page */
366 shmem_dir_unmap(dir);
367 dir = shmem_dir_map(subdir);
370 dir += index;
371 subdir = *dir;
372 if (!subdir) {
373 if (!page || !(subdir = *page)) {
374 shmem_dir_unmap(dir);
375 return NULL; /* need a page */
377 *dir = subdir;
378 *page = NULL;
380 shmem_dir_unmap(dir);
381 return shmem_swp_map(subdir) + offset;
384 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
386 long incdec = value? 1: -1;
388 entry->val = value;
389 info->swapped += incdec;
390 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
391 struct page *page = kmap_atomic_to_page(entry);
392 set_page_private(page, page_private(page) + incdec);
397 * shmem_swp_alloc - get the position of the swap entry for the page.
398 * @info: info structure for the inode
399 * @index: index of the page to find
400 * @sgp: check and recheck i_size? skip allocation?
402 * If the entry does not exist, allocate it.
404 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
406 struct inode *inode = &info->vfs_inode;
407 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
408 struct page *page = NULL;
409 swp_entry_t *entry;
411 if (sgp != SGP_WRITE &&
412 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
413 return ERR_PTR(-EINVAL);
415 while (!(entry = shmem_swp_entry(info, index, &page))) {
416 if (sgp == SGP_READ)
417 return shmem_swp_map(ZERO_PAGE(0));
419 * Test free_blocks against 1 not 0, since we have 1 data
420 * page (and perhaps indirect index pages) yet to allocate:
421 * a waste to allocate index if we cannot allocate data.
423 if (sbinfo->max_blocks) {
424 spin_lock(&sbinfo->stat_lock);
425 if (sbinfo->free_blocks <= 1) {
426 spin_unlock(&sbinfo->stat_lock);
427 return ERR_PTR(-ENOSPC);
429 sbinfo->free_blocks--;
430 inode->i_blocks += BLOCKS_PER_PAGE;
431 spin_unlock(&sbinfo->stat_lock);
434 spin_unlock(&info->lock);
435 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
436 if (page)
437 set_page_private(page, 0);
438 spin_lock(&info->lock);
440 if (!page) {
441 shmem_free_blocks(inode, 1);
442 return ERR_PTR(-ENOMEM);
444 if (sgp != SGP_WRITE &&
445 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
446 entry = ERR_PTR(-EINVAL);
447 break;
449 if (info->next_index <= index)
450 info->next_index = index + 1;
452 if (page) {
453 /* another task gave its page, or truncated the file */
454 shmem_free_blocks(inode, 1);
455 shmem_dir_free(page);
457 if (info->next_index <= index && !IS_ERR(entry))
458 info->next_index = index + 1;
459 return entry;
463 * shmem_free_swp - free some swap entries in a directory
464 * @dir: pointer to the directory
465 * @edir: pointer after last entry of the directory
466 * @punch_lock: pointer to spinlock when needed for the holepunch case
468 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
469 spinlock_t *punch_lock)
471 spinlock_t *punch_unlock = NULL;
472 swp_entry_t *ptr;
473 int freed = 0;
475 for (ptr = dir; ptr < edir; ptr++) {
476 if (ptr->val) {
477 if (unlikely(punch_lock)) {
478 punch_unlock = punch_lock;
479 punch_lock = NULL;
480 spin_lock(punch_unlock);
481 if (!ptr->val)
482 continue;
484 free_swap_and_cache(*ptr);
485 *ptr = (swp_entry_t){0};
486 freed++;
489 if (punch_unlock)
490 spin_unlock(punch_unlock);
491 return freed;
494 static int shmem_map_and_free_swp(struct page *subdir, int offset,
495 int limit, struct page ***dir, spinlock_t *punch_lock)
497 swp_entry_t *ptr;
498 int freed = 0;
500 ptr = shmem_swp_map(subdir);
501 for (; offset < limit; offset += LATENCY_LIMIT) {
502 int size = limit - offset;
503 if (size > LATENCY_LIMIT)
504 size = LATENCY_LIMIT;
505 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
506 punch_lock);
507 if (need_resched()) {
508 shmem_swp_unmap(ptr);
509 if (*dir) {
510 shmem_dir_unmap(*dir);
511 *dir = NULL;
513 cond_resched();
514 ptr = shmem_swp_map(subdir);
517 shmem_swp_unmap(ptr);
518 return freed;
521 static void shmem_free_pages(struct list_head *next)
523 struct page *page;
524 int freed = 0;
526 do {
527 page = container_of(next, struct page, lru);
528 next = next->next;
529 shmem_dir_free(page);
530 freed++;
531 if (freed >= LATENCY_LIMIT) {
532 cond_resched();
533 freed = 0;
535 } while (next);
538 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
540 struct shmem_inode_info *info = SHMEM_I(inode);
541 unsigned long idx;
542 unsigned long size;
543 unsigned long limit;
544 unsigned long stage;
545 unsigned long diroff;
546 struct page **dir;
547 struct page *topdir;
548 struct page *middir;
549 struct page *subdir;
550 swp_entry_t *ptr;
551 LIST_HEAD(pages_to_free);
552 long nr_pages_to_free = 0;
553 long nr_swaps_freed = 0;
554 int offset;
555 int freed;
556 int punch_hole;
557 spinlock_t *needs_lock;
558 spinlock_t *punch_lock;
559 unsigned long upper_limit;
561 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
562 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
563 if (idx >= info->next_index)
564 return;
566 spin_lock(&info->lock);
567 info->flags |= SHMEM_TRUNCATE;
568 if (likely(end == (loff_t) -1)) {
569 limit = info->next_index;
570 upper_limit = SHMEM_MAX_INDEX;
571 info->next_index = idx;
572 needs_lock = NULL;
573 punch_hole = 0;
574 } else {
575 if (end + 1 >= inode->i_size) { /* we may free a little more */
576 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
577 PAGE_CACHE_SHIFT;
578 upper_limit = SHMEM_MAX_INDEX;
579 } else {
580 limit = (end + 1) >> PAGE_CACHE_SHIFT;
581 upper_limit = limit;
583 needs_lock = &info->lock;
584 punch_hole = 1;
587 topdir = info->i_indirect;
588 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
589 info->i_indirect = NULL;
590 nr_pages_to_free++;
591 list_add(&topdir->lru, &pages_to_free);
593 spin_unlock(&info->lock);
595 if (info->swapped && idx < SHMEM_NR_DIRECT) {
596 ptr = info->i_direct;
597 size = limit;
598 if (size > SHMEM_NR_DIRECT)
599 size = SHMEM_NR_DIRECT;
600 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
604 * If there are no indirect blocks or we are punching a hole
605 * below indirect blocks, nothing to be done.
607 if (!topdir || limit <= SHMEM_NR_DIRECT)
608 goto done2;
611 * The truncation case has already dropped info->lock, and we're safe
612 * because i_size and next_index have already been lowered, preventing
613 * access beyond. But in the punch_hole case, we still need to take
614 * the lock when updating the swap directory, because there might be
615 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
616 * shmem_writepage. However, whenever we find we can remove a whole
617 * directory page (not at the misaligned start or end of the range),
618 * we first NULLify its pointer in the level above, and then have no
619 * need to take the lock when updating its contents: needs_lock and
620 * punch_lock (either pointing to info->lock or NULL) manage this.
623 upper_limit -= SHMEM_NR_DIRECT;
624 limit -= SHMEM_NR_DIRECT;
625 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
626 offset = idx % ENTRIES_PER_PAGE;
627 idx -= offset;
629 dir = shmem_dir_map(topdir);
630 stage = ENTRIES_PER_PAGEPAGE/2;
631 if (idx < ENTRIES_PER_PAGEPAGE/2) {
632 middir = topdir;
633 diroff = idx/ENTRIES_PER_PAGE;
634 } else {
635 dir += ENTRIES_PER_PAGE/2;
636 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
637 while (stage <= idx)
638 stage += ENTRIES_PER_PAGEPAGE;
639 middir = *dir;
640 if (*dir) {
641 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
642 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
643 if (!diroff && !offset && upper_limit >= stage) {
644 if (needs_lock) {
645 spin_lock(needs_lock);
646 *dir = NULL;
647 spin_unlock(needs_lock);
648 needs_lock = NULL;
649 } else
650 *dir = NULL;
651 nr_pages_to_free++;
652 list_add(&middir->lru, &pages_to_free);
654 shmem_dir_unmap(dir);
655 dir = shmem_dir_map(middir);
656 } else {
657 diroff = 0;
658 offset = 0;
659 idx = stage;
663 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
664 if (unlikely(idx == stage)) {
665 shmem_dir_unmap(dir);
666 dir = shmem_dir_map(topdir) +
667 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
668 while (!*dir) {
669 dir++;
670 idx += ENTRIES_PER_PAGEPAGE;
671 if (idx >= limit)
672 goto done1;
674 stage = idx + ENTRIES_PER_PAGEPAGE;
675 middir = *dir;
676 if (punch_hole)
677 needs_lock = &info->lock;
678 if (upper_limit >= stage) {
679 if (needs_lock) {
680 spin_lock(needs_lock);
681 *dir = NULL;
682 spin_unlock(needs_lock);
683 needs_lock = NULL;
684 } else
685 *dir = NULL;
686 nr_pages_to_free++;
687 list_add(&middir->lru, &pages_to_free);
689 shmem_dir_unmap(dir);
690 cond_resched();
691 dir = shmem_dir_map(middir);
692 diroff = 0;
694 punch_lock = needs_lock;
695 subdir = dir[diroff];
696 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
697 if (needs_lock) {
698 spin_lock(needs_lock);
699 dir[diroff] = NULL;
700 spin_unlock(needs_lock);
701 punch_lock = NULL;
702 } else
703 dir[diroff] = NULL;
704 nr_pages_to_free++;
705 list_add(&subdir->lru, &pages_to_free);
707 if (subdir && page_private(subdir) /* has swap entries */) {
708 size = limit - idx;
709 if (size > ENTRIES_PER_PAGE)
710 size = ENTRIES_PER_PAGE;
711 freed = shmem_map_and_free_swp(subdir,
712 offset, size, &dir, punch_lock);
713 if (!dir)
714 dir = shmem_dir_map(middir);
715 nr_swaps_freed += freed;
716 if (offset || punch_lock) {
717 spin_lock(&info->lock);
718 set_page_private(subdir,
719 page_private(subdir) - freed);
720 spin_unlock(&info->lock);
721 } else
722 BUG_ON(page_private(subdir) != freed);
724 offset = 0;
726 done1:
727 shmem_dir_unmap(dir);
728 done2:
729 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
731 * Call truncate_inode_pages again: racing shmem_unuse_inode
732 * may have swizzled a page in from swap since vmtruncate or
733 * generic_delete_inode did it, before we lowered next_index.
734 * Also, though shmem_getpage checks i_size before adding to
735 * cache, no recheck after: so fix the narrow window there too.
737 * Recalling truncate_inode_pages_range and unmap_mapping_range
738 * every time for punch_hole (which never got a chance to clear
739 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
740 * yet hardly ever necessary: try to optimize them out later.
742 truncate_inode_pages_range(inode->i_mapping, start, end);
743 if (punch_hole)
744 unmap_mapping_range(inode->i_mapping, start,
745 end - start, 1);
748 spin_lock(&info->lock);
749 info->flags &= ~SHMEM_TRUNCATE;
750 info->swapped -= nr_swaps_freed;
751 if (nr_pages_to_free)
752 shmem_free_blocks(inode, nr_pages_to_free);
753 shmem_recalc_inode(inode);
754 spin_unlock(&info->lock);
757 * Empty swap vector directory pages to be freed?
759 if (!list_empty(&pages_to_free)) {
760 pages_to_free.prev->next = NULL;
761 shmem_free_pages(pages_to_free.next);
765 static void shmem_truncate(struct inode *inode)
767 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
770 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
772 struct inode *inode = dentry->d_inode;
773 struct page *page = NULL;
774 int error;
776 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
777 if (attr->ia_size < inode->i_size) {
779 * If truncating down to a partial page, then
780 * if that page is already allocated, hold it
781 * in memory until the truncation is over, so
782 * truncate_partial_page cannnot miss it were
783 * it assigned to swap.
785 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
786 (void) shmem_getpage(inode,
787 attr->ia_size>>PAGE_CACHE_SHIFT,
788 &page, SGP_READ, NULL);
789 if (page)
790 unlock_page(page);
793 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
794 * detect if any pages might have been added to cache
795 * after truncate_inode_pages. But we needn't bother
796 * if it's being fully truncated to zero-length: the
797 * nrpages check is efficient enough in that case.
799 if (attr->ia_size) {
800 struct shmem_inode_info *info = SHMEM_I(inode);
801 spin_lock(&info->lock);
802 info->flags &= ~SHMEM_PAGEIN;
803 spin_unlock(&info->lock);
808 error = inode_change_ok(inode, attr);
809 if (!error)
810 error = inode_setattr(inode, attr);
811 #ifdef CONFIG_TMPFS_POSIX_ACL
812 if (!error && (attr->ia_valid & ATTR_MODE))
813 error = generic_acl_chmod(inode, &shmem_acl_ops);
814 #endif
815 if (page)
816 page_cache_release(page);
817 return error;
820 static void shmem_delete_inode(struct inode *inode)
822 struct shmem_inode_info *info = SHMEM_I(inode);
824 if (inode->i_op->truncate == shmem_truncate) {
825 truncate_inode_pages(inode->i_mapping, 0);
826 shmem_unacct_size(info->flags, inode->i_size);
827 inode->i_size = 0;
828 shmem_truncate(inode);
829 if (!list_empty(&info->swaplist)) {
830 mutex_lock(&shmem_swaplist_mutex);
831 list_del_init(&info->swaplist);
832 mutex_unlock(&shmem_swaplist_mutex);
835 BUG_ON(inode->i_blocks);
836 shmem_free_inode(inode->i_sb);
837 clear_inode(inode);
840 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
842 swp_entry_t *ptr;
844 for (ptr = dir; ptr < edir; ptr++) {
845 if (ptr->val == entry.val)
846 return ptr - dir;
848 return -1;
851 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
853 struct inode *inode;
854 unsigned long idx;
855 unsigned long size;
856 unsigned long limit;
857 unsigned long stage;
858 struct page **dir;
859 struct page *subdir;
860 swp_entry_t *ptr;
861 int offset;
862 int error;
864 idx = 0;
865 ptr = info->i_direct;
866 spin_lock(&info->lock);
867 if (!info->swapped) {
868 list_del_init(&info->swaplist);
869 goto lost2;
871 limit = info->next_index;
872 size = limit;
873 if (size > SHMEM_NR_DIRECT)
874 size = SHMEM_NR_DIRECT;
875 offset = shmem_find_swp(entry, ptr, ptr+size);
876 if (offset >= 0)
877 goto found;
878 if (!info->i_indirect)
879 goto lost2;
881 dir = shmem_dir_map(info->i_indirect);
882 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
884 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
885 if (unlikely(idx == stage)) {
886 shmem_dir_unmap(dir-1);
887 if (cond_resched_lock(&info->lock)) {
888 /* check it has not been truncated */
889 if (limit > info->next_index) {
890 limit = info->next_index;
891 if (idx >= limit)
892 goto lost2;
895 dir = shmem_dir_map(info->i_indirect) +
896 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
897 while (!*dir) {
898 dir++;
899 idx += ENTRIES_PER_PAGEPAGE;
900 if (idx >= limit)
901 goto lost1;
903 stage = idx + ENTRIES_PER_PAGEPAGE;
904 subdir = *dir;
905 shmem_dir_unmap(dir);
906 dir = shmem_dir_map(subdir);
908 subdir = *dir;
909 if (subdir && page_private(subdir)) {
910 ptr = shmem_swp_map(subdir);
911 size = limit - idx;
912 if (size > ENTRIES_PER_PAGE)
913 size = ENTRIES_PER_PAGE;
914 offset = shmem_find_swp(entry, ptr, ptr+size);
915 shmem_swp_unmap(ptr);
916 if (offset >= 0) {
917 shmem_dir_unmap(dir);
918 goto found;
922 lost1:
923 shmem_dir_unmap(dir-1);
924 lost2:
925 spin_unlock(&info->lock);
926 return 0;
927 found:
928 idx += offset;
929 inode = igrab(&info->vfs_inode);
930 spin_unlock(&info->lock);
933 * Move _head_ to start search for next from here.
934 * But be careful: shmem_delete_inode checks list_empty without taking
935 * mutex, and there's an instant in list_move_tail when info->swaplist
936 * would appear empty, if it were the only one on shmem_swaplist. We
937 * could avoid doing it if inode NULL; or use this minor optimization.
939 if (shmem_swaplist.next != &info->swaplist)
940 list_move_tail(&shmem_swaplist, &info->swaplist);
941 mutex_unlock(&shmem_swaplist_mutex);
943 error = 1;
944 if (!inode)
945 goto out;
947 * Charge page using GFP_KERNEL while we can wait.
948 * Charged back to the user(not to caller) when swap account is used.
949 * add_to_page_cache() will be called with GFP_NOWAIT.
951 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
952 if (error)
953 goto out;
954 error = radix_tree_preload(GFP_KERNEL);
955 if (error) {
956 mem_cgroup_uncharge_cache_page(page);
957 goto out;
959 error = 1;
961 spin_lock(&info->lock);
962 ptr = shmem_swp_entry(info, idx, NULL);
963 if (ptr && ptr->val == entry.val) {
964 error = add_to_page_cache_locked(page, inode->i_mapping,
965 idx, GFP_NOWAIT);
966 /* does mem_cgroup_uncharge_cache_page on error */
967 } else /* we must compensate for our precharge above */
968 mem_cgroup_uncharge_cache_page(page);
970 if (error == -EEXIST) {
971 struct page *filepage = find_get_page(inode->i_mapping, idx);
972 error = 1;
973 if (filepage) {
975 * There might be a more uptodate page coming down
976 * from a stacked writepage: forget our swappage if so.
978 if (PageUptodate(filepage))
979 error = 0;
980 page_cache_release(filepage);
983 if (!error) {
984 delete_from_swap_cache(page);
985 set_page_dirty(page);
986 info->flags |= SHMEM_PAGEIN;
987 shmem_swp_set(info, ptr, 0);
988 swap_free(entry);
989 error = 1; /* not an error, but entry was found */
991 if (ptr)
992 shmem_swp_unmap(ptr);
993 spin_unlock(&info->lock);
994 radix_tree_preload_end();
995 out:
996 unlock_page(page);
997 page_cache_release(page);
998 iput(inode); /* allows for NULL */
999 return error;
1003 * shmem_unuse() search for an eventually swapped out shmem page.
1005 int shmem_unuse(swp_entry_t entry, struct page *page)
1007 struct list_head *p, *next;
1008 struct shmem_inode_info *info;
1009 int found = 0;
1011 mutex_lock(&shmem_swaplist_mutex);
1012 list_for_each_safe(p, next, &shmem_swaplist) {
1013 info = list_entry(p, struct shmem_inode_info, swaplist);
1014 found = shmem_unuse_inode(info, entry, page);
1015 cond_resched();
1016 if (found)
1017 goto out;
1019 mutex_unlock(&shmem_swaplist_mutex);
1020 out: return found; /* 0 or 1 or -ENOMEM */
1024 * Move the page from the page cache to the swap cache.
1026 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1028 struct shmem_inode_info *info;
1029 swp_entry_t *entry, swap;
1030 struct address_space *mapping;
1031 unsigned long index;
1032 struct inode *inode;
1034 BUG_ON(!PageLocked(page));
1035 mapping = page->mapping;
1036 index = page->index;
1037 inode = mapping->host;
1038 info = SHMEM_I(inode);
1039 if (info->flags & VM_LOCKED)
1040 goto redirty;
1041 if (!total_swap_pages)
1042 goto redirty;
1045 * shmem_backing_dev_info's capabilities prevent regular writeback or
1046 * sync from ever calling shmem_writepage; but a stacking filesystem
1047 * may use the ->writepage of its underlying filesystem, in which case
1048 * tmpfs should write out to swap only in response to memory pressure,
1049 * and not for the writeback threads or sync. However, in those cases,
1050 * we do still want to check if there's a redundant swappage to be
1051 * discarded.
1053 if (wbc->for_reclaim)
1054 swap = get_swap_page();
1055 else
1056 swap.val = 0;
1058 spin_lock(&info->lock);
1059 if (index >= info->next_index) {
1060 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1061 goto unlock;
1063 entry = shmem_swp_entry(info, index, NULL);
1064 if (entry->val) {
1066 * The more uptodate page coming down from a stacked
1067 * writepage should replace our old swappage.
1069 free_swap_and_cache(*entry);
1070 shmem_swp_set(info, entry, 0);
1072 shmem_recalc_inode(inode);
1074 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1075 remove_from_page_cache(page);
1076 shmem_swp_set(info, entry, swap.val);
1077 shmem_swp_unmap(entry);
1078 if (list_empty(&info->swaplist))
1079 inode = igrab(inode);
1080 else
1081 inode = NULL;
1082 spin_unlock(&info->lock);
1083 swap_duplicate(swap);
1084 BUG_ON(page_mapped(page));
1085 page_cache_release(page); /* pagecache ref */
1086 swap_writepage(page, wbc);
1087 if (inode) {
1088 mutex_lock(&shmem_swaplist_mutex);
1089 /* move instead of add in case we're racing */
1090 list_move_tail(&info->swaplist, &shmem_swaplist);
1091 mutex_unlock(&shmem_swaplist_mutex);
1092 iput(inode);
1094 return 0;
1097 shmem_swp_unmap(entry);
1098 unlock:
1099 spin_unlock(&info->lock);
1101 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1102 * clear SWAP_HAS_CACHE flag.
1104 swapcache_free(swap, NULL);
1105 redirty:
1106 set_page_dirty(page);
1107 if (wbc->for_reclaim)
1108 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1109 unlock_page(page);
1110 return 0;
1113 #ifdef CONFIG_NUMA
1114 #ifdef CONFIG_TMPFS
1115 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1117 char buffer[64];
1119 if (!mpol || mpol->mode == MPOL_DEFAULT)
1120 return; /* show nothing */
1122 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1124 seq_printf(seq, ",mpol=%s", buffer);
1127 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1129 struct mempolicy *mpol = NULL;
1130 if (sbinfo->mpol) {
1131 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1132 mpol = sbinfo->mpol;
1133 mpol_get(mpol);
1134 spin_unlock(&sbinfo->stat_lock);
1136 return mpol;
1138 #endif /* CONFIG_TMPFS */
1140 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1141 struct shmem_inode_info *info, unsigned long idx)
1143 struct mempolicy mpol, *spol;
1144 struct vm_area_struct pvma;
1145 struct page *page;
1147 spol = mpol_cond_copy(&mpol,
1148 mpol_shared_policy_lookup(&info->policy, idx));
1150 /* Create a pseudo vma that just contains the policy */
1151 pvma.vm_start = 0;
1152 pvma.vm_pgoff = idx;
1153 pvma.vm_ops = NULL;
1154 pvma.vm_policy = spol;
1155 page = swapin_readahead(entry, gfp, &pvma, 0);
1156 return page;
1159 static struct page *shmem_alloc_page(gfp_t gfp,
1160 struct shmem_inode_info *info, unsigned long idx)
1162 struct vm_area_struct pvma;
1164 /* Create a pseudo vma that just contains the policy */
1165 pvma.vm_start = 0;
1166 pvma.vm_pgoff = idx;
1167 pvma.vm_ops = NULL;
1168 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1171 * alloc_page_vma() will drop the shared policy reference
1173 return alloc_page_vma(gfp, &pvma, 0);
1175 #else /* !CONFIG_NUMA */
1176 #ifdef CONFIG_TMPFS
1177 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1180 #endif /* CONFIG_TMPFS */
1182 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1183 struct shmem_inode_info *info, unsigned long idx)
1185 return swapin_readahead(entry, gfp, NULL, 0);
1188 static inline struct page *shmem_alloc_page(gfp_t gfp,
1189 struct shmem_inode_info *info, unsigned long idx)
1191 return alloc_page(gfp);
1193 #endif /* CONFIG_NUMA */
1195 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1196 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1198 return NULL;
1200 #endif
1203 * shmem_getpage - either get the page from swap or allocate a new one
1205 * If we allocate a new one we do not mark it dirty. That's up to the
1206 * vm. If we swap it in we mark it dirty since we also free the swap
1207 * entry since a page cannot live in both the swap and page cache
1209 static int shmem_getpage(struct inode *inode, unsigned long idx,
1210 struct page **pagep, enum sgp_type sgp, int *type)
1212 struct address_space *mapping = inode->i_mapping;
1213 struct shmem_inode_info *info = SHMEM_I(inode);
1214 struct shmem_sb_info *sbinfo;
1215 struct page *filepage = *pagep;
1216 struct page *swappage;
1217 swp_entry_t *entry;
1218 swp_entry_t swap;
1219 gfp_t gfp;
1220 int error;
1222 if (idx >= SHMEM_MAX_INDEX)
1223 return -EFBIG;
1225 if (type)
1226 *type = 0;
1229 * Normally, filepage is NULL on entry, and either found
1230 * uptodate immediately, or allocated and zeroed, or read
1231 * in under swappage, which is then assigned to filepage.
1232 * But shmem_readpage (required for splice) passes in a locked
1233 * filepage, which may be found not uptodate by other callers
1234 * too, and may need to be copied from the swappage read in.
1236 repeat:
1237 if (!filepage)
1238 filepage = find_lock_page(mapping, idx);
1239 if (filepage && PageUptodate(filepage))
1240 goto done;
1241 error = 0;
1242 gfp = mapping_gfp_mask(mapping);
1243 if (!filepage) {
1245 * Try to preload while we can wait, to not make a habit of
1246 * draining atomic reserves; but don't latch on to this cpu.
1248 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1249 if (error)
1250 goto failed;
1251 radix_tree_preload_end();
1254 spin_lock(&info->lock);
1255 shmem_recalc_inode(inode);
1256 entry = shmem_swp_alloc(info, idx, sgp);
1257 if (IS_ERR(entry)) {
1258 spin_unlock(&info->lock);
1259 error = PTR_ERR(entry);
1260 goto failed;
1262 swap = *entry;
1264 if (swap.val) {
1265 /* Look it up and read it in.. */
1266 swappage = lookup_swap_cache(swap);
1267 if (!swappage) {
1268 shmem_swp_unmap(entry);
1269 /* here we actually do the io */
1270 if (type && !(*type & VM_FAULT_MAJOR)) {
1271 __count_vm_event(PGMAJFAULT);
1272 *type |= VM_FAULT_MAJOR;
1274 spin_unlock(&info->lock);
1275 swappage = shmem_swapin(swap, gfp, info, idx);
1276 if (!swappage) {
1277 spin_lock(&info->lock);
1278 entry = shmem_swp_alloc(info, idx, sgp);
1279 if (IS_ERR(entry))
1280 error = PTR_ERR(entry);
1281 else {
1282 if (entry->val == swap.val)
1283 error = -ENOMEM;
1284 shmem_swp_unmap(entry);
1286 spin_unlock(&info->lock);
1287 if (error)
1288 goto failed;
1289 goto repeat;
1291 wait_on_page_locked(swappage);
1292 page_cache_release(swappage);
1293 goto repeat;
1296 /* We have to do this with page locked to prevent races */
1297 if (!trylock_page(swappage)) {
1298 shmem_swp_unmap(entry);
1299 spin_unlock(&info->lock);
1300 wait_on_page_locked(swappage);
1301 page_cache_release(swappage);
1302 goto repeat;
1304 if (PageWriteback(swappage)) {
1305 shmem_swp_unmap(entry);
1306 spin_unlock(&info->lock);
1307 wait_on_page_writeback(swappage);
1308 unlock_page(swappage);
1309 page_cache_release(swappage);
1310 goto repeat;
1312 if (!PageUptodate(swappage)) {
1313 shmem_swp_unmap(entry);
1314 spin_unlock(&info->lock);
1315 unlock_page(swappage);
1316 page_cache_release(swappage);
1317 error = -EIO;
1318 goto failed;
1321 if (filepage) {
1322 shmem_swp_set(info, entry, 0);
1323 shmem_swp_unmap(entry);
1324 delete_from_swap_cache(swappage);
1325 spin_unlock(&info->lock);
1326 copy_highpage(filepage, swappage);
1327 unlock_page(swappage);
1328 page_cache_release(swappage);
1329 flush_dcache_page(filepage);
1330 SetPageUptodate(filepage);
1331 set_page_dirty(filepage);
1332 swap_free(swap);
1333 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1334 idx, GFP_NOWAIT))) {
1335 info->flags |= SHMEM_PAGEIN;
1336 shmem_swp_set(info, entry, 0);
1337 shmem_swp_unmap(entry);
1338 delete_from_swap_cache(swappage);
1339 spin_unlock(&info->lock);
1340 filepage = swappage;
1341 set_page_dirty(filepage);
1342 swap_free(swap);
1343 } else {
1344 shmem_swp_unmap(entry);
1345 spin_unlock(&info->lock);
1346 if (error == -ENOMEM) {
1348 * reclaim from proper memory cgroup and
1349 * call memcg's OOM if needed.
1351 error = mem_cgroup_shmem_charge_fallback(
1352 swappage,
1353 current->mm,
1354 gfp);
1355 if (error) {
1356 unlock_page(swappage);
1357 page_cache_release(swappage);
1358 goto failed;
1361 unlock_page(swappage);
1362 page_cache_release(swappage);
1363 goto repeat;
1365 } else if (sgp == SGP_READ && !filepage) {
1366 shmem_swp_unmap(entry);
1367 filepage = find_get_page(mapping, idx);
1368 if (filepage &&
1369 (!PageUptodate(filepage) || !trylock_page(filepage))) {
1370 spin_unlock(&info->lock);
1371 wait_on_page_locked(filepage);
1372 page_cache_release(filepage);
1373 filepage = NULL;
1374 goto repeat;
1376 spin_unlock(&info->lock);
1377 } else {
1378 shmem_swp_unmap(entry);
1379 sbinfo = SHMEM_SB(inode->i_sb);
1380 if (sbinfo->max_blocks) {
1381 spin_lock(&sbinfo->stat_lock);
1382 if (sbinfo->free_blocks == 0 ||
1383 shmem_acct_block(info->flags)) {
1384 spin_unlock(&sbinfo->stat_lock);
1385 spin_unlock(&info->lock);
1386 error = -ENOSPC;
1387 goto failed;
1389 sbinfo->free_blocks--;
1390 inode->i_blocks += BLOCKS_PER_PAGE;
1391 spin_unlock(&sbinfo->stat_lock);
1392 } else if (shmem_acct_block(info->flags)) {
1393 spin_unlock(&info->lock);
1394 error = -ENOSPC;
1395 goto failed;
1398 if (!filepage) {
1399 int ret;
1401 spin_unlock(&info->lock);
1402 filepage = shmem_alloc_page(gfp, info, idx);
1403 if (!filepage) {
1404 shmem_unacct_blocks(info->flags, 1);
1405 shmem_free_blocks(inode, 1);
1406 error = -ENOMEM;
1407 goto failed;
1409 SetPageSwapBacked(filepage);
1411 /* Precharge page while we can wait, compensate after */
1412 error = mem_cgroup_cache_charge(filepage, current->mm,
1413 GFP_KERNEL);
1414 if (error) {
1415 page_cache_release(filepage);
1416 shmem_unacct_blocks(info->flags, 1);
1417 shmem_free_blocks(inode, 1);
1418 filepage = NULL;
1419 goto failed;
1422 spin_lock(&info->lock);
1423 entry = shmem_swp_alloc(info, idx, sgp);
1424 if (IS_ERR(entry))
1425 error = PTR_ERR(entry);
1426 else {
1427 swap = *entry;
1428 shmem_swp_unmap(entry);
1430 ret = error || swap.val;
1431 if (ret)
1432 mem_cgroup_uncharge_cache_page(filepage);
1433 else
1434 ret = add_to_page_cache_lru(filepage, mapping,
1435 idx, GFP_NOWAIT);
1437 * At add_to_page_cache_lru() failure, uncharge will
1438 * be done automatically.
1440 if (ret) {
1441 spin_unlock(&info->lock);
1442 page_cache_release(filepage);
1443 shmem_unacct_blocks(info->flags, 1);
1444 shmem_free_blocks(inode, 1);
1445 filepage = NULL;
1446 if (error)
1447 goto failed;
1448 goto repeat;
1450 info->flags |= SHMEM_PAGEIN;
1453 info->alloced++;
1454 spin_unlock(&info->lock);
1455 clear_highpage(filepage);
1456 flush_dcache_page(filepage);
1457 SetPageUptodate(filepage);
1458 if (sgp == SGP_DIRTY)
1459 set_page_dirty(filepage);
1461 done:
1462 *pagep = filepage;
1463 return 0;
1465 failed:
1466 if (*pagep != filepage) {
1467 unlock_page(filepage);
1468 page_cache_release(filepage);
1470 return error;
1473 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1475 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1476 int error;
1477 int ret;
1479 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1480 return VM_FAULT_SIGBUS;
1482 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1483 if (error)
1484 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1486 return ret | VM_FAULT_LOCKED;
1489 #ifdef CONFIG_NUMA
1490 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1492 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1493 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1496 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1497 unsigned long addr)
1499 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1500 unsigned long idx;
1502 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1503 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1505 #endif
1507 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1509 struct inode *inode = file->f_path.dentry->d_inode;
1510 struct shmem_inode_info *info = SHMEM_I(inode);
1511 int retval = -ENOMEM;
1513 spin_lock(&info->lock);
1514 if (lock && !(info->flags & VM_LOCKED)) {
1515 if (!user_shm_lock(inode->i_size, user))
1516 goto out_nomem;
1517 info->flags |= VM_LOCKED;
1518 mapping_set_unevictable(file->f_mapping);
1520 if (!lock && (info->flags & VM_LOCKED) && user) {
1521 user_shm_unlock(inode->i_size, user);
1522 info->flags &= ~VM_LOCKED;
1523 mapping_clear_unevictable(file->f_mapping);
1524 scan_mapping_unevictable_pages(file->f_mapping);
1526 retval = 0;
1528 out_nomem:
1529 spin_unlock(&info->lock);
1530 return retval;
1533 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1535 file_accessed(file);
1536 vma->vm_ops = &shmem_vm_ops;
1537 vma->vm_flags |= VM_CAN_NONLINEAR;
1538 return 0;
1541 static struct inode *shmem_get_inode(struct super_block *sb, int mode,
1542 dev_t dev, unsigned long flags)
1544 struct inode *inode;
1545 struct shmem_inode_info *info;
1546 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1548 if (shmem_reserve_inode(sb))
1549 return NULL;
1551 inode = new_inode(sb);
1552 if (inode) {
1553 inode->i_mode = mode;
1554 inode->i_uid = current_fsuid();
1555 inode->i_gid = current_fsgid();
1556 inode->i_blocks = 0;
1557 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1558 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1559 inode->i_generation = get_seconds();
1560 info = SHMEM_I(inode);
1561 memset(info, 0, (char *)inode - (char *)info);
1562 spin_lock_init(&info->lock);
1563 info->flags = flags & VM_NORESERVE;
1564 INIT_LIST_HEAD(&info->swaplist);
1565 cache_no_acl(inode);
1567 switch (mode & S_IFMT) {
1568 default:
1569 inode->i_op = &shmem_special_inode_operations;
1570 init_special_inode(inode, mode, dev);
1571 break;
1572 case S_IFREG:
1573 inode->i_mapping->a_ops = &shmem_aops;
1574 inode->i_op = &shmem_inode_operations;
1575 inode->i_fop = &shmem_file_operations;
1576 mpol_shared_policy_init(&info->policy,
1577 shmem_get_sbmpol(sbinfo));
1578 break;
1579 case S_IFDIR:
1580 inc_nlink(inode);
1581 /* Some things misbehave if size == 0 on a directory */
1582 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1583 inode->i_op = &shmem_dir_inode_operations;
1584 inode->i_fop = &simple_dir_operations;
1585 break;
1586 case S_IFLNK:
1588 * Must not load anything in the rbtree,
1589 * mpol_free_shared_policy will not be called.
1591 mpol_shared_policy_init(&info->policy, NULL);
1592 break;
1594 } else
1595 shmem_free_inode(sb);
1596 return inode;
1599 #ifdef CONFIG_TMPFS
1600 static const struct inode_operations shmem_symlink_inode_operations;
1601 static const struct inode_operations shmem_symlink_inline_operations;
1604 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1605 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1606 * below the loop driver, in the generic fashion that many filesystems support.
1608 static int shmem_readpage(struct file *file, struct page *page)
1610 struct inode *inode = page->mapping->host;
1611 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1612 unlock_page(page);
1613 return error;
1616 static int
1617 shmem_write_begin(struct file *file, struct address_space *mapping,
1618 loff_t pos, unsigned len, unsigned flags,
1619 struct page **pagep, void **fsdata)
1621 struct inode *inode = mapping->host;
1622 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1623 *pagep = NULL;
1624 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1627 static int
1628 shmem_write_end(struct file *file, struct address_space *mapping,
1629 loff_t pos, unsigned len, unsigned copied,
1630 struct page *page, void *fsdata)
1632 struct inode *inode = mapping->host;
1634 if (pos + copied > inode->i_size)
1635 i_size_write(inode, pos + copied);
1637 set_page_dirty(page);
1638 unlock_page(page);
1639 page_cache_release(page);
1641 return copied;
1644 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1646 struct inode *inode = filp->f_path.dentry->d_inode;
1647 struct address_space *mapping = inode->i_mapping;
1648 unsigned long index, offset;
1649 enum sgp_type sgp = SGP_READ;
1652 * Might this read be for a stacking filesystem? Then when reading
1653 * holes of a sparse file, we actually need to allocate those pages,
1654 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1656 if (segment_eq(get_fs(), KERNEL_DS))
1657 sgp = SGP_DIRTY;
1659 index = *ppos >> PAGE_CACHE_SHIFT;
1660 offset = *ppos & ~PAGE_CACHE_MASK;
1662 for (;;) {
1663 struct page *page = NULL;
1664 unsigned long end_index, nr, ret;
1665 loff_t i_size = i_size_read(inode);
1667 end_index = i_size >> PAGE_CACHE_SHIFT;
1668 if (index > end_index)
1669 break;
1670 if (index == end_index) {
1671 nr = i_size & ~PAGE_CACHE_MASK;
1672 if (nr <= offset)
1673 break;
1676 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1677 if (desc->error) {
1678 if (desc->error == -EINVAL)
1679 desc->error = 0;
1680 break;
1682 if (page)
1683 unlock_page(page);
1686 * We must evaluate after, since reads (unlike writes)
1687 * are called without i_mutex protection against truncate
1689 nr = PAGE_CACHE_SIZE;
1690 i_size = i_size_read(inode);
1691 end_index = i_size >> PAGE_CACHE_SHIFT;
1692 if (index == end_index) {
1693 nr = i_size & ~PAGE_CACHE_MASK;
1694 if (nr <= offset) {
1695 if (page)
1696 page_cache_release(page);
1697 break;
1700 nr -= offset;
1702 if (page) {
1704 * If users can be writing to this page using arbitrary
1705 * virtual addresses, take care about potential aliasing
1706 * before reading the page on the kernel side.
1708 if (mapping_writably_mapped(mapping))
1709 flush_dcache_page(page);
1711 * Mark the page accessed if we read the beginning.
1713 if (!offset)
1714 mark_page_accessed(page);
1715 } else {
1716 page = ZERO_PAGE(0);
1717 page_cache_get(page);
1721 * Ok, we have the page, and it's up-to-date, so
1722 * now we can copy it to user space...
1724 * The actor routine returns how many bytes were actually used..
1725 * NOTE! This may not be the same as how much of a user buffer
1726 * we filled up (we may be padding etc), so we can only update
1727 * "pos" here (the actor routine has to update the user buffer
1728 * pointers and the remaining count).
1730 ret = actor(desc, page, offset, nr);
1731 offset += ret;
1732 index += offset >> PAGE_CACHE_SHIFT;
1733 offset &= ~PAGE_CACHE_MASK;
1735 page_cache_release(page);
1736 if (ret != nr || !desc->count)
1737 break;
1739 cond_resched();
1742 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1743 file_accessed(filp);
1746 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1747 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1749 struct file *filp = iocb->ki_filp;
1750 ssize_t retval;
1751 unsigned long seg;
1752 size_t count;
1753 loff_t *ppos = &iocb->ki_pos;
1755 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1756 if (retval)
1757 return retval;
1759 for (seg = 0; seg < nr_segs; seg++) {
1760 read_descriptor_t desc;
1762 desc.written = 0;
1763 desc.arg.buf = iov[seg].iov_base;
1764 desc.count = iov[seg].iov_len;
1765 if (desc.count == 0)
1766 continue;
1767 desc.error = 0;
1768 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1769 retval += desc.written;
1770 if (desc.error) {
1771 retval = retval ?: desc.error;
1772 break;
1774 if (desc.count > 0)
1775 break;
1777 return retval;
1780 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1782 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1784 buf->f_type = TMPFS_MAGIC;
1785 buf->f_bsize = PAGE_CACHE_SIZE;
1786 buf->f_namelen = NAME_MAX;
1787 spin_lock(&sbinfo->stat_lock);
1788 if (sbinfo->max_blocks) {
1789 buf->f_blocks = sbinfo->max_blocks;
1790 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1792 if (sbinfo->max_inodes) {
1793 buf->f_files = sbinfo->max_inodes;
1794 buf->f_ffree = sbinfo->free_inodes;
1796 /* else leave those fields 0 like simple_statfs */
1797 spin_unlock(&sbinfo->stat_lock);
1798 return 0;
1802 * File creation. Allocate an inode, and we're done..
1804 static int
1805 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1807 struct inode *inode;
1808 int error = -ENOSPC;
1810 inode = shmem_get_inode(dir->i_sb, mode, dev, VM_NORESERVE);
1811 if (inode) {
1812 error = security_inode_init_security(inode, dir, NULL, NULL,
1813 NULL);
1814 if (error) {
1815 if (error != -EOPNOTSUPP) {
1816 iput(inode);
1817 return error;
1820 error = shmem_acl_init(inode, dir);
1821 if (error) {
1822 iput(inode);
1823 return error;
1825 if (dir->i_mode & S_ISGID) {
1826 inode->i_gid = dir->i_gid;
1827 if (S_ISDIR(mode))
1828 inode->i_mode |= S_ISGID;
1830 dir->i_size += BOGO_DIRENT_SIZE;
1831 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1832 d_instantiate(dentry, inode);
1833 dget(dentry); /* Extra count - pin the dentry in core */
1835 return error;
1838 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1840 int error;
1842 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1843 return error;
1844 inc_nlink(dir);
1845 return 0;
1848 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1849 struct nameidata *nd)
1851 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1855 * Link a file..
1857 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1859 struct inode *inode = old_dentry->d_inode;
1860 int ret;
1863 * No ordinary (disk based) filesystem counts links as inodes;
1864 * but each new link needs a new dentry, pinning lowmem, and
1865 * tmpfs dentries cannot be pruned until they are unlinked.
1867 ret = shmem_reserve_inode(inode->i_sb);
1868 if (ret)
1869 goto out;
1871 dir->i_size += BOGO_DIRENT_SIZE;
1872 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1873 inc_nlink(inode);
1874 atomic_inc(&inode->i_count); /* New dentry reference */
1875 dget(dentry); /* Extra pinning count for the created dentry */
1876 d_instantiate(dentry, inode);
1877 out:
1878 return ret;
1881 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1883 struct inode *inode = dentry->d_inode;
1885 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1886 shmem_free_inode(inode->i_sb);
1888 dir->i_size -= BOGO_DIRENT_SIZE;
1889 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1890 drop_nlink(inode);
1891 dput(dentry); /* Undo the count from "create" - this does all the work */
1892 return 0;
1895 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1897 if (!simple_empty(dentry))
1898 return -ENOTEMPTY;
1900 drop_nlink(dentry->d_inode);
1901 drop_nlink(dir);
1902 return shmem_unlink(dir, dentry);
1906 * The VFS layer already does all the dentry stuff for rename,
1907 * we just have to decrement the usage count for the target if
1908 * it exists so that the VFS layer correctly free's it when it
1909 * gets overwritten.
1911 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1913 struct inode *inode = old_dentry->d_inode;
1914 int they_are_dirs = S_ISDIR(inode->i_mode);
1916 if (!simple_empty(new_dentry))
1917 return -ENOTEMPTY;
1919 if (new_dentry->d_inode) {
1920 (void) shmem_unlink(new_dir, new_dentry);
1921 if (they_are_dirs)
1922 drop_nlink(old_dir);
1923 } else if (they_are_dirs) {
1924 drop_nlink(old_dir);
1925 inc_nlink(new_dir);
1928 old_dir->i_size -= BOGO_DIRENT_SIZE;
1929 new_dir->i_size += BOGO_DIRENT_SIZE;
1930 old_dir->i_ctime = old_dir->i_mtime =
1931 new_dir->i_ctime = new_dir->i_mtime =
1932 inode->i_ctime = CURRENT_TIME;
1933 return 0;
1936 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1938 int error;
1939 int len;
1940 struct inode *inode;
1941 struct page *page = NULL;
1942 char *kaddr;
1943 struct shmem_inode_info *info;
1945 len = strlen(symname) + 1;
1946 if (len > PAGE_CACHE_SIZE)
1947 return -ENAMETOOLONG;
1949 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1950 if (!inode)
1951 return -ENOSPC;
1953 error = security_inode_init_security(inode, dir, NULL, NULL,
1954 NULL);
1955 if (error) {
1956 if (error != -EOPNOTSUPP) {
1957 iput(inode);
1958 return error;
1960 error = 0;
1963 info = SHMEM_I(inode);
1964 inode->i_size = len-1;
1965 if (len <= (char *)inode - (char *)info) {
1966 /* do it inline */
1967 memcpy(info, symname, len);
1968 inode->i_op = &shmem_symlink_inline_operations;
1969 } else {
1970 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1971 if (error) {
1972 iput(inode);
1973 return error;
1975 inode->i_mapping->a_ops = &shmem_aops;
1976 inode->i_op = &shmem_symlink_inode_operations;
1977 kaddr = kmap_atomic(page, KM_USER0);
1978 memcpy(kaddr, symname, len);
1979 kunmap_atomic(kaddr, KM_USER0);
1980 set_page_dirty(page);
1981 unlock_page(page);
1982 page_cache_release(page);
1984 if (dir->i_mode & S_ISGID)
1985 inode->i_gid = dir->i_gid;
1986 dir->i_size += BOGO_DIRENT_SIZE;
1987 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1988 d_instantiate(dentry, inode);
1989 dget(dentry);
1990 return 0;
1993 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1995 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1996 return NULL;
1999 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2001 struct page *page = NULL;
2002 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2003 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
2004 if (page)
2005 unlock_page(page);
2006 return page;
2009 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2011 if (!IS_ERR(nd_get_link(nd))) {
2012 struct page *page = cookie;
2013 kunmap(page);
2014 mark_page_accessed(page);
2015 page_cache_release(page);
2019 static const struct inode_operations shmem_symlink_inline_operations = {
2020 .readlink = generic_readlink,
2021 .follow_link = shmem_follow_link_inline,
2024 static const struct inode_operations shmem_symlink_inode_operations = {
2025 .truncate = shmem_truncate,
2026 .readlink = generic_readlink,
2027 .follow_link = shmem_follow_link,
2028 .put_link = shmem_put_link,
2031 #ifdef CONFIG_TMPFS_POSIX_ACL
2033 * Superblocks without xattr inode operations will get security.* xattr
2034 * support from the VFS "for free". As soon as we have any other xattrs
2035 * like ACLs, we also need to implement the security.* handlers at
2036 * filesystem level, though.
2039 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
2040 size_t list_len, const char *name,
2041 size_t name_len)
2043 return security_inode_listsecurity(inode, list, list_len);
2046 static int shmem_xattr_security_get(struct inode *inode, const char *name,
2047 void *buffer, size_t size)
2049 if (strcmp(name, "") == 0)
2050 return -EINVAL;
2051 return xattr_getsecurity(inode, name, buffer, size);
2054 static int shmem_xattr_security_set(struct inode *inode, const char *name,
2055 const void *value, size_t size, int flags)
2057 if (strcmp(name, "") == 0)
2058 return -EINVAL;
2059 return security_inode_setsecurity(inode, name, value, size, flags);
2062 static struct xattr_handler shmem_xattr_security_handler = {
2063 .prefix = XATTR_SECURITY_PREFIX,
2064 .list = shmem_xattr_security_list,
2065 .get = shmem_xattr_security_get,
2066 .set = shmem_xattr_security_set,
2069 static struct xattr_handler *shmem_xattr_handlers[] = {
2070 &shmem_xattr_acl_access_handler,
2071 &shmem_xattr_acl_default_handler,
2072 &shmem_xattr_security_handler,
2073 NULL
2075 #endif
2077 static struct dentry *shmem_get_parent(struct dentry *child)
2079 return ERR_PTR(-ESTALE);
2082 static int shmem_match(struct inode *ino, void *vfh)
2084 __u32 *fh = vfh;
2085 __u64 inum = fh[2];
2086 inum = (inum << 32) | fh[1];
2087 return ino->i_ino == inum && fh[0] == ino->i_generation;
2090 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2091 struct fid *fid, int fh_len, int fh_type)
2093 struct inode *inode;
2094 struct dentry *dentry = NULL;
2095 u64 inum = fid->raw[2];
2096 inum = (inum << 32) | fid->raw[1];
2098 if (fh_len < 3)
2099 return NULL;
2101 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2102 shmem_match, fid->raw);
2103 if (inode) {
2104 dentry = d_find_alias(inode);
2105 iput(inode);
2108 return dentry;
2111 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2112 int connectable)
2114 struct inode *inode = dentry->d_inode;
2116 if (*len < 3)
2117 return 255;
2119 if (hlist_unhashed(&inode->i_hash)) {
2120 /* Unfortunately insert_inode_hash is not idempotent,
2121 * so as we hash inodes here rather than at creation
2122 * time, we need a lock to ensure we only try
2123 * to do it once
2125 static DEFINE_SPINLOCK(lock);
2126 spin_lock(&lock);
2127 if (hlist_unhashed(&inode->i_hash))
2128 __insert_inode_hash(inode,
2129 inode->i_ino + inode->i_generation);
2130 spin_unlock(&lock);
2133 fh[0] = inode->i_generation;
2134 fh[1] = inode->i_ino;
2135 fh[2] = ((__u64)inode->i_ino) >> 32;
2137 *len = 3;
2138 return 1;
2141 static const struct export_operations shmem_export_ops = {
2142 .get_parent = shmem_get_parent,
2143 .encode_fh = shmem_encode_fh,
2144 .fh_to_dentry = shmem_fh_to_dentry,
2147 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2148 bool remount)
2150 char *this_char, *value, *rest;
2152 while (options != NULL) {
2153 this_char = options;
2154 for (;;) {
2156 * NUL-terminate this option: unfortunately,
2157 * mount options form a comma-separated list,
2158 * but mpol's nodelist may also contain commas.
2160 options = strchr(options, ',');
2161 if (options == NULL)
2162 break;
2163 options++;
2164 if (!isdigit(*options)) {
2165 options[-1] = '\0';
2166 break;
2169 if (!*this_char)
2170 continue;
2171 if ((value = strchr(this_char,'=')) != NULL) {
2172 *value++ = 0;
2173 } else {
2174 printk(KERN_ERR
2175 "tmpfs: No value for mount option '%s'\n",
2176 this_char);
2177 return 1;
2180 if (!strcmp(this_char,"size")) {
2181 unsigned long long size;
2182 size = memparse(value,&rest);
2183 if (*rest == '%') {
2184 size <<= PAGE_SHIFT;
2185 size *= totalram_pages;
2186 do_div(size, 100);
2187 rest++;
2189 if (*rest)
2190 goto bad_val;
2191 sbinfo->max_blocks =
2192 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2193 } else if (!strcmp(this_char,"nr_blocks")) {
2194 sbinfo->max_blocks = memparse(value, &rest);
2195 if (*rest)
2196 goto bad_val;
2197 } else if (!strcmp(this_char,"nr_inodes")) {
2198 sbinfo->max_inodes = memparse(value, &rest);
2199 if (*rest)
2200 goto bad_val;
2201 } else if (!strcmp(this_char,"mode")) {
2202 if (remount)
2203 continue;
2204 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2205 if (*rest)
2206 goto bad_val;
2207 } else if (!strcmp(this_char,"uid")) {
2208 if (remount)
2209 continue;
2210 sbinfo->uid = simple_strtoul(value, &rest, 0);
2211 if (*rest)
2212 goto bad_val;
2213 } else if (!strcmp(this_char,"gid")) {
2214 if (remount)
2215 continue;
2216 sbinfo->gid = simple_strtoul(value, &rest, 0);
2217 if (*rest)
2218 goto bad_val;
2219 } else if (!strcmp(this_char,"mpol")) {
2220 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2221 goto bad_val;
2222 } else {
2223 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2224 this_char);
2225 return 1;
2228 return 0;
2230 bad_val:
2231 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2232 value, this_char);
2233 return 1;
2237 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2239 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2240 struct shmem_sb_info config = *sbinfo;
2241 unsigned long blocks;
2242 unsigned long inodes;
2243 int error = -EINVAL;
2245 if (shmem_parse_options(data, &config, true))
2246 return error;
2248 spin_lock(&sbinfo->stat_lock);
2249 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2250 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2251 if (config.max_blocks < blocks)
2252 goto out;
2253 if (config.max_inodes < inodes)
2254 goto out;
2256 * Those tests also disallow limited->unlimited while any are in
2257 * use, so i_blocks will always be zero when max_blocks is zero;
2258 * but we must separately disallow unlimited->limited, because
2259 * in that case we have no record of how much is already in use.
2261 if (config.max_blocks && !sbinfo->max_blocks)
2262 goto out;
2263 if (config.max_inodes && !sbinfo->max_inodes)
2264 goto out;
2266 error = 0;
2267 sbinfo->max_blocks = config.max_blocks;
2268 sbinfo->free_blocks = config.max_blocks - blocks;
2269 sbinfo->max_inodes = config.max_inodes;
2270 sbinfo->free_inodes = config.max_inodes - inodes;
2272 mpol_put(sbinfo->mpol);
2273 sbinfo->mpol = config.mpol; /* transfers initial ref */
2274 out:
2275 spin_unlock(&sbinfo->stat_lock);
2276 return error;
2279 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2281 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2283 if (sbinfo->max_blocks != shmem_default_max_blocks())
2284 seq_printf(seq, ",size=%luk",
2285 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2286 if (sbinfo->max_inodes != shmem_default_max_inodes())
2287 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2288 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2289 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2290 if (sbinfo->uid != 0)
2291 seq_printf(seq, ",uid=%u", sbinfo->uid);
2292 if (sbinfo->gid != 0)
2293 seq_printf(seq, ",gid=%u", sbinfo->gid);
2294 shmem_show_mpol(seq, sbinfo->mpol);
2295 return 0;
2297 #endif /* CONFIG_TMPFS */
2299 static void shmem_put_super(struct super_block *sb)
2301 kfree(sb->s_fs_info);
2302 sb->s_fs_info = NULL;
2305 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2307 struct inode *inode;
2308 struct dentry *root;
2309 struct shmem_sb_info *sbinfo;
2310 int err = -ENOMEM;
2312 /* Round up to L1_CACHE_BYTES to resist false sharing */
2313 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2314 L1_CACHE_BYTES), GFP_KERNEL);
2315 if (!sbinfo)
2316 return -ENOMEM;
2318 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2319 sbinfo->uid = current_fsuid();
2320 sbinfo->gid = current_fsgid();
2321 sb->s_fs_info = sbinfo;
2323 #ifdef CONFIG_TMPFS
2325 * Per default we only allow half of the physical ram per
2326 * tmpfs instance, limiting inodes to one per page of lowmem;
2327 * but the internal instance is left unlimited.
2329 if (!(sb->s_flags & MS_NOUSER)) {
2330 sbinfo->max_blocks = shmem_default_max_blocks();
2331 sbinfo->max_inodes = shmem_default_max_inodes();
2332 if (shmem_parse_options(data, sbinfo, false)) {
2333 err = -EINVAL;
2334 goto failed;
2337 sb->s_export_op = &shmem_export_ops;
2338 #else
2339 sb->s_flags |= MS_NOUSER;
2340 #endif
2342 spin_lock_init(&sbinfo->stat_lock);
2343 sbinfo->free_blocks = sbinfo->max_blocks;
2344 sbinfo->free_inodes = sbinfo->max_inodes;
2346 sb->s_maxbytes = SHMEM_MAX_BYTES;
2347 sb->s_blocksize = PAGE_CACHE_SIZE;
2348 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2349 sb->s_magic = TMPFS_MAGIC;
2350 sb->s_op = &shmem_ops;
2351 sb->s_time_gran = 1;
2352 #ifdef CONFIG_TMPFS_POSIX_ACL
2353 sb->s_xattr = shmem_xattr_handlers;
2354 sb->s_flags |= MS_POSIXACL;
2355 #endif
2357 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2358 if (!inode)
2359 goto failed;
2360 inode->i_uid = sbinfo->uid;
2361 inode->i_gid = sbinfo->gid;
2362 root = d_alloc_root(inode);
2363 if (!root)
2364 goto failed_iput;
2365 sb->s_root = root;
2366 return 0;
2368 failed_iput:
2369 iput(inode);
2370 failed:
2371 shmem_put_super(sb);
2372 return err;
2375 static struct kmem_cache *shmem_inode_cachep;
2377 static struct inode *shmem_alloc_inode(struct super_block *sb)
2379 struct shmem_inode_info *p;
2380 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2381 if (!p)
2382 return NULL;
2383 return &p->vfs_inode;
2386 static void shmem_destroy_inode(struct inode *inode)
2388 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2389 /* only struct inode is valid if it's an inline symlink */
2390 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2392 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2395 static void init_once(void *foo)
2397 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2399 inode_init_once(&p->vfs_inode);
2402 static int init_inodecache(void)
2404 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2405 sizeof(struct shmem_inode_info),
2406 0, SLAB_PANIC, init_once);
2407 return 0;
2410 static void destroy_inodecache(void)
2412 kmem_cache_destroy(shmem_inode_cachep);
2415 static const struct address_space_operations shmem_aops = {
2416 .writepage = shmem_writepage,
2417 .set_page_dirty = __set_page_dirty_no_writeback,
2418 #ifdef CONFIG_TMPFS
2419 .readpage = shmem_readpage,
2420 .write_begin = shmem_write_begin,
2421 .write_end = shmem_write_end,
2422 #endif
2423 .migratepage = migrate_page,
2424 .error_remove_page = generic_error_remove_page,
2427 static const struct file_operations shmem_file_operations = {
2428 .mmap = shmem_mmap,
2429 #ifdef CONFIG_TMPFS
2430 .llseek = generic_file_llseek,
2431 .read = do_sync_read,
2432 .write = do_sync_write,
2433 .aio_read = shmem_file_aio_read,
2434 .aio_write = generic_file_aio_write,
2435 .fsync = simple_sync_file,
2436 .splice_read = generic_file_splice_read,
2437 .splice_write = generic_file_splice_write,
2438 #endif
2441 static const struct inode_operations shmem_inode_operations = {
2442 .truncate = shmem_truncate,
2443 .setattr = shmem_notify_change,
2444 .truncate_range = shmem_truncate_range,
2445 #ifdef CONFIG_TMPFS_POSIX_ACL
2446 .setxattr = generic_setxattr,
2447 .getxattr = generic_getxattr,
2448 .listxattr = generic_listxattr,
2449 .removexattr = generic_removexattr,
2450 .check_acl = shmem_check_acl,
2451 #endif
2455 static const struct inode_operations shmem_dir_inode_operations = {
2456 #ifdef CONFIG_TMPFS
2457 .create = shmem_create,
2458 .lookup = simple_lookup,
2459 .link = shmem_link,
2460 .unlink = shmem_unlink,
2461 .symlink = shmem_symlink,
2462 .mkdir = shmem_mkdir,
2463 .rmdir = shmem_rmdir,
2464 .mknod = shmem_mknod,
2465 .rename = shmem_rename,
2466 #endif
2467 #ifdef CONFIG_TMPFS_POSIX_ACL
2468 .setattr = shmem_notify_change,
2469 .setxattr = generic_setxattr,
2470 .getxattr = generic_getxattr,
2471 .listxattr = generic_listxattr,
2472 .removexattr = generic_removexattr,
2473 .check_acl = shmem_check_acl,
2474 #endif
2477 static const struct inode_operations shmem_special_inode_operations = {
2478 #ifdef CONFIG_TMPFS_POSIX_ACL
2479 .setattr = shmem_notify_change,
2480 .setxattr = generic_setxattr,
2481 .getxattr = generic_getxattr,
2482 .listxattr = generic_listxattr,
2483 .removexattr = generic_removexattr,
2484 .check_acl = shmem_check_acl,
2485 #endif
2488 static const struct super_operations shmem_ops = {
2489 .alloc_inode = shmem_alloc_inode,
2490 .destroy_inode = shmem_destroy_inode,
2491 #ifdef CONFIG_TMPFS
2492 .statfs = shmem_statfs,
2493 .remount_fs = shmem_remount_fs,
2494 .show_options = shmem_show_options,
2495 #endif
2496 .delete_inode = shmem_delete_inode,
2497 .drop_inode = generic_delete_inode,
2498 .put_super = shmem_put_super,
2501 static const struct vm_operations_struct shmem_vm_ops = {
2502 .fault = shmem_fault,
2503 #ifdef CONFIG_NUMA
2504 .set_policy = shmem_set_policy,
2505 .get_policy = shmem_get_policy,
2506 #endif
2510 static int shmem_get_sb(struct file_system_type *fs_type,
2511 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2513 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2516 static struct file_system_type tmpfs_fs_type = {
2517 .owner = THIS_MODULE,
2518 .name = "tmpfs",
2519 .get_sb = shmem_get_sb,
2520 .kill_sb = kill_litter_super,
2523 int __init init_tmpfs(void)
2525 int error;
2527 error = bdi_init(&shmem_backing_dev_info);
2528 if (error)
2529 goto out4;
2531 error = init_inodecache();
2532 if (error)
2533 goto out3;
2535 error = register_filesystem(&tmpfs_fs_type);
2536 if (error) {
2537 printk(KERN_ERR "Could not register tmpfs\n");
2538 goto out2;
2541 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2542 tmpfs_fs_type.name, NULL);
2543 if (IS_ERR(shm_mnt)) {
2544 error = PTR_ERR(shm_mnt);
2545 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2546 goto out1;
2548 return 0;
2550 out1:
2551 unregister_filesystem(&tmpfs_fs_type);
2552 out2:
2553 destroy_inodecache();
2554 out3:
2555 bdi_destroy(&shmem_backing_dev_info);
2556 out4:
2557 shm_mnt = ERR_PTR(error);
2558 return error;
2561 #else /* !CONFIG_SHMEM */
2564 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2566 * This is intended for small system where the benefits of the full
2567 * shmem code (swap-backed and resource-limited) are outweighed by
2568 * their complexity. On systems without swap this code should be
2569 * effectively equivalent, but much lighter weight.
2572 #include <linux/ramfs.h>
2574 static struct file_system_type tmpfs_fs_type = {
2575 .name = "tmpfs",
2576 .get_sb = ramfs_get_sb,
2577 .kill_sb = kill_litter_super,
2580 int __init init_tmpfs(void)
2582 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2584 shm_mnt = kern_mount(&tmpfs_fs_type);
2585 BUG_ON(IS_ERR(shm_mnt));
2587 return 0;
2590 int shmem_unuse(swp_entry_t entry, struct page *page)
2592 return 0;
2595 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2597 return 0;
2600 #define shmem_vm_ops generic_file_vm_ops
2601 #define shmem_file_operations ramfs_file_operations
2602 #define shmem_get_inode(sb, mode, dev, flags) ramfs_get_inode(sb, mode, dev)
2603 #define shmem_acct_size(flags, size) 0
2604 #define shmem_unacct_size(flags, size) do {} while (0)
2605 #define SHMEM_MAX_BYTES MAX_LFS_FILESIZE
2607 #endif /* CONFIG_SHMEM */
2609 /* common code */
2612 * shmem_file_setup - get an unlinked file living in tmpfs
2613 * @name: name for dentry (to be seen in /proc/<pid>/maps
2614 * @size: size to be set for the file
2615 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2617 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2619 int error;
2620 struct file *file;
2621 struct inode *inode;
2622 struct dentry *dentry, *root;
2623 struct qstr this;
2625 if (IS_ERR(shm_mnt))
2626 return (void *)shm_mnt;
2628 if (size < 0 || size > SHMEM_MAX_BYTES)
2629 return ERR_PTR(-EINVAL);
2631 if (shmem_acct_size(flags, size))
2632 return ERR_PTR(-ENOMEM);
2634 error = -ENOMEM;
2635 this.name = name;
2636 this.len = strlen(name);
2637 this.hash = 0; /* will go */
2638 root = shm_mnt->mnt_root;
2639 dentry = d_alloc(root, &this);
2640 if (!dentry)
2641 goto put_memory;
2643 error = -ENFILE;
2644 file = get_empty_filp();
2645 if (!file)
2646 goto put_dentry;
2648 error = -ENOSPC;
2649 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0, flags);
2650 if (!inode)
2651 goto close_file;
2653 d_instantiate(dentry, inode);
2654 inode->i_size = size;
2655 inode->i_nlink = 0; /* It is unlinked */
2656 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
2657 &shmem_file_operations);
2659 #ifndef CONFIG_MMU
2660 error = ramfs_nommu_expand_for_mapping(inode, size);
2661 if (error)
2662 goto close_file;
2663 #endif
2664 ima_counts_get(file);
2665 return file;
2667 close_file:
2668 put_filp(file);
2669 put_dentry:
2670 dput(dentry);
2671 put_memory:
2672 shmem_unacct_size(flags, size);
2673 return ERR_PTR(error);
2675 EXPORT_SYMBOL_GPL(shmem_file_setup);
2678 * shmem_zero_setup - setup a shared anonymous mapping
2679 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2681 int shmem_zero_setup(struct vm_area_struct *vma)
2683 struct file *file;
2684 loff_t size = vma->vm_end - vma->vm_start;
2686 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2687 if (IS_ERR(file))
2688 return PTR_ERR(file);
2690 if (vma->vm_file)
2691 fput(vma->vm_file);
2692 vma->vm_file = file;
2693 vma->vm_ops = &shmem_vm_ops;
2694 return 0;