fs: check for statfs overflow
[linux-2.6/mini2440.git] / mm / shmem.c
blob9ffbea9b79e15370b540ffcce2207e5714091e0e
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 * This file is released under the GPL.
21 * This virtual memory filesystem is heavily based on the ramfs. It
22 * extends ramfs by the ability to use swap and honor resource limits
23 * which makes it a completely usable filesystem.
26 #include <linux/module.h>
27 #include <linux/init.h>
28 #include <linux/fs.h>
29 #include <linux/xattr.h>
30 #include <linux/exportfs.h>
31 #include <linux/generic_acl.h>
32 #include <linux/mm.h>
33 #include <linux/mman.h>
34 #include <linux/file.h>
35 #include <linux/swap.h>
36 #include <linux/pagemap.h>
37 #include <linux/string.h>
38 #include <linux/slab.h>
39 #include <linux/backing-dev.h>
40 #include <linux/shmem_fs.h>
41 #include <linux/mount.h>
42 #include <linux/writeback.h>
43 #include <linux/vfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/security.h>
46 #include <linux/swapops.h>
47 #include <linux/mempolicy.h>
48 #include <linux/namei.h>
49 #include <linux/ctype.h>
50 #include <linux/migrate.h>
51 #include <linux/highmem.h>
52 #include <linux/seq_file.h>
54 #include <asm/uaccess.h>
55 #include <asm/div64.h>
56 #include <asm/pgtable.h>
58 /* This magic number is used in glibc for posix shared memory */
59 #define TMPFS_MAGIC 0x01021994
61 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
62 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
63 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
65 #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
66 #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
68 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
70 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
71 #define SHMEM_PAGEIN VM_READ
72 #define SHMEM_TRUNCATE VM_WRITE
74 /* Definition to limit shmem_truncate's steps between cond_rescheds */
75 #define LATENCY_LIMIT 64
77 /* Pretend that each entry is of this size in directory's i_size */
78 #define BOGO_DIRENT_SIZE 20
80 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
81 enum sgp_type {
82 SGP_READ, /* don't exceed i_size, don't allocate page */
83 SGP_CACHE, /* don't exceed i_size, may allocate page */
84 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
85 SGP_WRITE, /* may exceed i_size, may allocate page */
88 #ifdef CONFIG_TMPFS
89 static unsigned long shmem_default_max_blocks(void)
91 return totalram_pages / 2;
94 static unsigned long shmem_default_max_inodes(void)
96 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
98 #endif
100 static int shmem_getpage(struct inode *inode, unsigned long idx,
101 struct page **pagep, enum sgp_type sgp, int *type);
103 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
106 * The above definition of ENTRIES_PER_PAGE, and the use of
107 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
108 * might be reconsidered if it ever diverges from PAGE_SIZE.
110 * Mobility flags are masked out as swap vectors cannot move
112 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
113 PAGE_CACHE_SHIFT-PAGE_SHIFT);
116 static inline void shmem_dir_free(struct page *page)
118 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
121 static struct page **shmem_dir_map(struct page *page)
123 return (struct page **)kmap_atomic(page, KM_USER0);
126 static inline void shmem_dir_unmap(struct page **dir)
128 kunmap_atomic(dir, KM_USER0);
131 static swp_entry_t *shmem_swp_map(struct page *page)
133 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
136 static inline void shmem_swp_balance_unmap(void)
139 * When passing a pointer to an i_direct entry, to code which
140 * also handles indirect entries and so will shmem_swp_unmap,
141 * we must arrange for the preempt count to remain in balance.
142 * What kmap_atomic of a lowmem page does depends on config
143 * and architecture, so pretend to kmap_atomic some lowmem page.
145 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
148 static inline void shmem_swp_unmap(swp_entry_t *entry)
150 kunmap_atomic(entry, KM_USER1);
153 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
155 return sb->s_fs_info;
159 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
160 * for shared memory and for shared anonymous (/dev/zero) mappings
161 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
162 * consistent with the pre-accounting of private mappings ...
164 static inline int shmem_acct_size(unsigned long flags, loff_t size)
166 return (flags & VM_ACCOUNT)?
167 security_vm_enough_memory(VM_ACCT(size)): 0;
170 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
172 if (flags & VM_ACCOUNT)
173 vm_unacct_memory(VM_ACCT(size));
177 * ... whereas tmpfs objects are accounted incrementally as
178 * pages are allocated, in order to allow huge sparse files.
179 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
180 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
182 static inline int shmem_acct_block(unsigned long flags)
184 return (flags & VM_ACCOUNT)?
185 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
188 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
190 if (!(flags & VM_ACCOUNT))
191 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
194 static const struct super_operations shmem_ops;
195 static const struct address_space_operations shmem_aops;
196 static const struct file_operations shmem_file_operations;
197 static const struct inode_operations shmem_inode_operations;
198 static const struct inode_operations shmem_dir_inode_operations;
199 static const struct inode_operations shmem_special_inode_operations;
200 static struct vm_operations_struct shmem_vm_ops;
202 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
203 .ra_pages = 0, /* No readahead */
204 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
205 .unplug_io_fn = default_unplug_io_fn,
208 static LIST_HEAD(shmem_swaplist);
209 static DEFINE_MUTEX(shmem_swaplist_mutex);
211 static void shmem_free_blocks(struct inode *inode, long pages)
213 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
214 if (sbinfo->max_blocks) {
215 spin_lock(&sbinfo->stat_lock);
216 sbinfo->free_blocks += pages;
217 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
218 spin_unlock(&sbinfo->stat_lock);
222 static int shmem_reserve_inode(struct super_block *sb)
224 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
225 if (sbinfo->max_inodes) {
226 spin_lock(&sbinfo->stat_lock);
227 if (!sbinfo->free_inodes) {
228 spin_unlock(&sbinfo->stat_lock);
229 return -ENOSPC;
231 sbinfo->free_inodes--;
232 spin_unlock(&sbinfo->stat_lock);
234 return 0;
237 static void shmem_free_inode(struct super_block *sb)
239 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
240 if (sbinfo->max_inodes) {
241 spin_lock(&sbinfo->stat_lock);
242 sbinfo->free_inodes++;
243 spin_unlock(&sbinfo->stat_lock);
248 * shmem_recalc_inode - recalculate the size of an inode
249 * @inode: inode to recalc
251 * We have to calculate the free blocks since the mm can drop
252 * undirtied hole pages behind our back.
254 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
255 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
257 * It has to be called with the spinlock held.
259 static void shmem_recalc_inode(struct inode *inode)
261 struct shmem_inode_info *info = SHMEM_I(inode);
262 long freed;
264 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
265 if (freed > 0) {
266 info->alloced -= freed;
267 shmem_unacct_blocks(info->flags, freed);
268 shmem_free_blocks(inode, freed);
273 * shmem_swp_entry - find the swap vector position in the info structure
274 * @info: info structure for the inode
275 * @index: index of the page to find
276 * @page: optional page to add to the structure. Has to be preset to
277 * all zeros
279 * If there is no space allocated yet it will return NULL when
280 * page is NULL, else it will use the page for the needed block,
281 * setting it to NULL on return to indicate that it has been used.
283 * The swap vector is organized the following way:
285 * There are SHMEM_NR_DIRECT entries directly stored in the
286 * shmem_inode_info structure. So small files do not need an addional
287 * allocation.
289 * For pages with index > SHMEM_NR_DIRECT there is the pointer
290 * i_indirect which points to a page which holds in the first half
291 * doubly indirect blocks, in the second half triple indirect blocks:
293 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
294 * following layout (for SHMEM_NR_DIRECT == 16):
296 * i_indirect -> dir --> 16-19
297 * | +-> 20-23
299 * +-->dir2 --> 24-27
300 * | +-> 28-31
301 * | +-> 32-35
302 * | +-> 36-39
304 * +-->dir3 --> 40-43
305 * +-> 44-47
306 * +-> 48-51
307 * +-> 52-55
309 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
311 unsigned long offset;
312 struct page **dir;
313 struct page *subdir;
315 if (index < SHMEM_NR_DIRECT) {
316 shmem_swp_balance_unmap();
317 return info->i_direct+index;
319 if (!info->i_indirect) {
320 if (page) {
321 info->i_indirect = *page;
322 *page = NULL;
324 return NULL; /* need another page */
327 index -= SHMEM_NR_DIRECT;
328 offset = index % ENTRIES_PER_PAGE;
329 index /= ENTRIES_PER_PAGE;
330 dir = shmem_dir_map(info->i_indirect);
332 if (index >= ENTRIES_PER_PAGE/2) {
333 index -= ENTRIES_PER_PAGE/2;
334 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
335 index %= ENTRIES_PER_PAGE;
336 subdir = *dir;
337 if (!subdir) {
338 if (page) {
339 *dir = *page;
340 *page = NULL;
342 shmem_dir_unmap(dir);
343 return NULL; /* need another page */
345 shmem_dir_unmap(dir);
346 dir = shmem_dir_map(subdir);
349 dir += index;
350 subdir = *dir;
351 if (!subdir) {
352 if (!page || !(subdir = *page)) {
353 shmem_dir_unmap(dir);
354 return NULL; /* need a page */
356 *dir = subdir;
357 *page = NULL;
359 shmem_dir_unmap(dir);
360 return shmem_swp_map(subdir) + offset;
363 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
365 long incdec = value? 1: -1;
367 entry->val = value;
368 info->swapped += incdec;
369 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
370 struct page *page = kmap_atomic_to_page(entry);
371 set_page_private(page, page_private(page) + incdec);
376 * shmem_swp_alloc - get the position of the swap entry for the page.
377 * @info: info structure for the inode
378 * @index: index of the page to find
379 * @sgp: check and recheck i_size? skip allocation?
381 * If the entry does not exist, allocate it.
383 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
385 struct inode *inode = &info->vfs_inode;
386 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
387 struct page *page = NULL;
388 swp_entry_t *entry;
390 if (sgp != SGP_WRITE &&
391 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
392 return ERR_PTR(-EINVAL);
394 while (!(entry = shmem_swp_entry(info, index, &page))) {
395 if (sgp == SGP_READ)
396 return shmem_swp_map(ZERO_PAGE(0));
398 * Test free_blocks against 1 not 0, since we have 1 data
399 * page (and perhaps indirect index pages) yet to allocate:
400 * a waste to allocate index if we cannot allocate data.
402 if (sbinfo->max_blocks) {
403 spin_lock(&sbinfo->stat_lock);
404 if (sbinfo->free_blocks <= 1) {
405 spin_unlock(&sbinfo->stat_lock);
406 return ERR_PTR(-ENOSPC);
408 sbinfo->free_blocks--;
409 inode->i_blocks += BLOCKS_PER_PAGE;
410 spin_unlock(&sbinfo->stat_lock);
413 spin_unlock(&info->lock);
414 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
415 if (page)
416 set_page_private(page, 0);
417 spin_lock(&info->lock);
419 if (!page) {
420 shmem_free_blocks(inode, 1);
421 return ERR_PTR(-ENOMEM);
423 if (sgp != SGP_WRITE &&
424 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
425 entry = ERR_PTR(-EINVAL);
426 break;
428 if (info->next_index <= index)
429 info->next_index = index + 1;
431 if (page) {
432 /* another task gave its page, or truncated the file */
433 shmem_free_blocks(inode, 1);
434 shmem_dir_free(page);
436 if (info->next_index <= index && !IS_ERR(entry))
437 info->next_index = index + 1;
438 return entry;
442 * shmem_free_swp - free some swap entries in a directory
443 * @dir: pointer to the directory
444 * @edir: pointer after last entry of the directory
445 * @punch_lock: pointer to spinlock when needed for the holepunch case
447 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
448 spinlock_t *punch_lock)
450 spinlock_t *punch_unlock = NULL;
451 swp_entry_t *ptr;
452 int freed = 0;
454 for (ptr = dir; ptr < edir; ptr++) {
455 if (ptr->val) {
456 if (unlikely(punch_lock)) {
457 punch_unlock = punch_lock;
458 punch_lock = NULL;
459 spin_lock(punch_unlock);
460 if (!ptr->val)
461 continue;
463 free_swap_and_cache(*ptr);
464 *ptr = (swp_entry_t){0};
465 freed++;
468 if (punch_unlock)
469 spin_unlock(punch_unlock);
470 return freed;
473 static int shmem_map_and_free_swp(struct page *subdir, int offset,
474 int limit, struct page ***dir, spinlock_t *punch_lock)
476 swp_entry_t *ptr;
477 int freed = 0;
479 ptr = shmem_swp_map(subdir);
480 for (; offset < limit; offset += LATENCY_LIMIT) {
481 int size = limit - offset;
482 if (size > LATENCY_LIMIT)
483 size = LATENCY_LIMIT;
484 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
485 punch_lock);
486 if (need_resched()) {
487 shmem_swp_unmap(ptr);
488 if (*dir) {
489 shmem_dir_unmap(*dir);
490 *dir = NULL;
492 cond_resched();
493 ptr = shmem_swp_map(subdir);
496 shmem_swp_unmap(ptr);
497 return freed;
500 static void shmem_free_pages(struct list_head *next)
502 struct page *page;
503 int freed = 0;
505 do {
506 page = container_of(next, struct page, lru);
507 next = next->next;
508 shmem_dir_free(page);
509 freed++;
510 if (freed >= LATENCY_LIMIT) {
511 cond_resched();
512 freed = 0;
514 } while (next);
517 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
519 struct shmem_inode_info *info = SHMEM_I(inode);
520 unsigned long idx;
521 unsigned long size;
522 unsigned long limit;
523 unsigned long stage;
524 unsigned long diroff;
525 struct page **dir;
526 struct page *topdir;
527 struct page *middir;
528 struct page *subdir;
529 swp_entry_t *ptr;
530 LIST_HEAD(pages_to_free);
531 long nr_pages_to_free = 0;
532 long nr_swaps_freed = 0;
533 int offset;
534 int freed;
535 int punch_hole;
536 spinlock_t *needs_lock;
537 spinlock_t *punch_lock;
538 unsigned long upper_limit;
540 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
541 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
542 if (idx >= info->next_index)
543 return;
545 spin_lock(&info->lock);
546 info->flags |= SHMEM_TRUNCATE;
547 if (likely(end == (loff_t) -1)) {
548 limit = info->next_index;
549 upper_limit = SHMEM_MAX_INDEX;
550 info->next_index = idx;
551 needs_lock = NULL;
552 punch_hole = 0;
553 } else {
554 if (end + 1 >= inode->i_size) { /* we may free a little more */
555 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
556 PAGE_CACHE_SHIFT;
557 upper_limit = SHMEM_MAX_INDEX;
558 } else {
559 limit = (end + 1) >> PAGE_CACHE_SHIFT;
560 upper_limit = limit;
562 needs_lock = &info->lock;
563 punch_hole = 1;
566 topdir = info->i_indirect;
567 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
568 info->i_indirect = NULL;
569 nr_pages_to_free++;
570 list_add(&topdir->lru, &pages_to_free);
572 spin_unlock(&info->lock);
574 if (info->swapped && idx < SHMEM_NR_DIRECT) {
575 ptr = info->i_direct;
576 size = limit;
577 if (size > SHMEM_NR_DIRECT)
578 size = SHMEM_NR_DIRECT;
579 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
583 * If there are no indirect blocks or we are punching a hole
584 * below indirect blocks, nothing to be done.
586 if (!topdir || limit <= SHMEM_NR_DIRECT)
587 goto done2;
590 * The truncation case has already dropped info->lock, and we're safe
591 * because i_size and next_index have already been lowered, preventing
592 * access beyond. But in the punch_hole case, we still need to take
593 * the lock when updating the swap directory, because there might be
594 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
595 * shmem_writepage. However, whenever we find we can remove a whole
596 * directory page (not at the misaligned start or end of the range),
597 * we first NULLify its pointer in the level above, and then have no
598 * need to take the lock when updating its contents: needs_lock and
599 * punch_lock (either pointing to info->lock or NULL) manage this.
602 upper_limit -= SHMEM_NR_DIRECT;
603 limit -= SHMEM_NR_DIRECT;
604 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
605 offset = idx % ENTRIES_PER_PAGE;
606 idx -= offset;
608 dir = shmem_dir_map(topdir);
609 stage = ENTRIES_PER_PAGEPAGE/2;
610 if (idx < ENTRIES_PER_PAGEPAGE/2) {
611 middir = topdir;
612 diroff = idx/ENTRIES_PER_PAGE;
613 } else {
614 dir += ENTRIES_PER_PAGE/2;
615 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
616 while (stage <= idx)
617 stage += ENTRIES_PER_PAGEPAGE;
618 middir = *dir;
619 if (*dir) {
620 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
621 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
622 if (!diroff && !offset && upper_limit >= stage) {
623 if (needs_lock) {
624 spin_lock(needs_lock);
625 *dir = NULL;
626 spin_unlock(needs_lock);
627 needs_lock = NULL;
628 } else
629 *dir = NULL;
630 nr_pages_to_free++;
631 list_add(&middir->lru, &pages_to_free);
633 shmem_dir_unmap(dir);
634 dir = shmem_dir_map(middir);
635 } else {
636 diroff = 0;
637 offset = 0;
638 idx = stage;
642 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
643 if (unlikely(idx == stage)) {
644 shmem_dir_unmap(dir);
645 dir = shmem_dir_map(topdir) +
646 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
647 while (!*dir) {
648 dir++;
649 idx += ENTRIES_PER_PAGEPAGE;
650 if (idx >= limit)
651 goto done1;
653 stage = idx + ENTRIES_PER_PAGEPAGE;
654 middir = *dir;
655 if (punch_hole)
656 needs_lock = &info->lock;
657 if (upper_limit >= stage) {
658 if (needs_lock) {
659 spin_lock(needs_lock);
660 *dir = NULL;
661 spin_unlock(needs_lock);
662 needs_lock = NULL;
663 } else
664 *dir = NULL;
665 nr_pages_to_free++;
666 list_add(&middir->lru, &pages_to_free);
668 shmem_dir_unmap(dir);
669 cond_resched();
670 dir = shmem_dir_map(middir);
671 diroff = 0;
673 punch_lock = needs_lock;
674 subdir = dir[diroff];
675 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
676 if (needs_lock) {
677 spin_lock(needs_lock);
678 dir[diroff] = NULL;
679 spin_unlock(needs_lock);
680 punch_lock = NULL;
681 } else
682 dir[diroff] = NULL;
683 nr_pages_to_free++;
684 list_add(&subdir->lru, &pages_to_free);
686 if (subdir && page_private(subdir) /* has swap entries */) {
687 size = limit - idx;
688 if (size > ENTRIES_PER_PAGE)
689 size = ENTRIES_PER_PAGE;
690 freed = shmem_map_and_free_swp(subdir,
691 offset, size, &dir, punch_lock);
692 if (!dir)
693 dir = shmem_dir_map(middir);
694 nr_swaps_freed += freed;
695 if (offset || punch_lock) {
696 spin_lock(&info->lock);
697 set_page_private(subdir,
698 page_private(subdir) - freed);
699 spin_unlock(&info->lock);
700 } else
701 BUG_ON(page_private(subdir) != freed);
703 offset = 0;
705 done1:
706 shmem_dir_unmap(dir);
707 done2:
708 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
710 * Call truncate_inode_pages again: racing shmem_unuse_inode
711 * may have swizzled a page in from swap since vmtruncate or
712 * generic_delete_inode did it, before we lowered next_index.
713 * Also, though shmem_getpage checks i_size before adding to
714 * cache, no recheck after: so fix the narrow window there too.
716 * Recalling truncate_inode_pages_range and unmap_mapping_range
717 * every time for punch_hole (which never got a chance to clear
718 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
719 * yet hardly ever necessary: try to optimize them out later.
721 truncate_inode_pages_range(inode->i_mapping, start, end);
722 if (punch_hole)
723 unmap_mapping_range(inode->i_mapping, start,
724 end - start, 1);
727 spin_lock(&info->lock);
728 info->flags &= ~SHMEM_TRUNCATE;
729 info->swapped -= nr_swaps_freed;
730 if (nr_pages_to_free)
731 shmem_free_blocks(inode, nr_pages_to_free);
732 shmem_recalc_inode(inode);
733 spin_unlock(&info->lock);
736 * Empty swap vector directory pages to be freed?
738 if (!list_empty(&pages_to_free)) {
739 pages_to_free.prev->next = NULL;
740 shmem_free_pages(pages_to_free.next);
744 static void shmem_truncate(struct inode *inode)
746 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
749 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
751 struct inode *inode = dentry->d_inode;
752 struct page *page = NULL;
753 int error;
755 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
756 if (attr->ia_size < inode->i_size) {
758 * If truncating down to a partial page, then
759 * if that page is already allocated, hold it
760 * in memory until the truncation is over, so
761 * truncate_partial_page cannnot miss it were
762 * it assigned to swap.
764 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
765 (void) shmem_getpage(inode,
766 attr->ia_size>>PAGE_CACHE_SHIFT,
767 &page, SGP_READ, NULL);
768 if (page)
769 unlock_page(page);
772 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
773 * detect if any pages might have been added to cache
774 * after truncate_inode_pages. But we needn't bother
775 * if it's being fully truncated to zero-length: the
776 * nrpages check is efficient enough in that case.
778 if (attr->ia_size) {
779 struct shmem_inode_info *info = SHMEM_I(inode);
780 spin_lock(&info->lock);
781 info->flags &= ~SHMEM_PAGEIN;
782 spin_unlock(&info->lock);
787 error = inode_change_ok(inode, attr);
788 if (!error)
789 error = inode_setattr(inode, attr);
790 #ifdef CONFIG_TMPFS_POSIX_ACL
791 if (!error && (attr->ia_valid & ATTR_MODE))
792 error = generic_acl_chmod(inode, &shmem_acl_ops);
793 #endif
794 if (page)
795 page_cache_release(page);
796 return error;
799 static void shmem_delete_inode(struct inode *inode)
801 struct shmem_inode_info *info = SHMEM_I(inode);
803 if (inode->i_op->truncate == shmem_truncate) {
804 truncate_inode_pages(inode->i_mapping, 0);
805 shmem_unacct_size(info->flags, inode->i_size);
806 inode->i_size = 0;
807 shmem_truncate(inode);
808 if (!list_empty(&info->swaplist)) {
809 mutex_lock(&shmem_swaplist_mutex);
810 list_del_init(&info->swaplist);
811 mutex_unlock(&shmem_swaplist_mutex);
814 BUG_ON(inode->i_blocks);
815 shmem_free_inode(inode->i_sb);
816 clear_inode(inode);
819 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
821 swp_entry_t *ptr;
823 for (ptr = dir; ptr < edir; ptr++) {
824 if (ptr->val == entry.val)
825 return ptr - dir;
827 return -1;
830 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
832 struct inode *inode;
833 unsigned long idx;
834 unsigned long size;
835 unsigned long limit;
836 unsigned long stage;
837 struct page **dir;
838 struct page *subdir;
839 swp_entry_t *ptr;
840 int offset;
841 int error;
843 idx = 0;
844 ptr = info->i_direct;
845 spin_lock(&info->lock);
846 if (!info->swapped) {
847 list_del_init(&info->swaplist);
848 goto lost2;
850 limit = info->next_index;
851 size = limit;
852 if (size > SHMEM_NR_DIRECT)
853 size = SHMEM_NR_DIRECT;
854 offset = shmem_find_swp(entry, ptr, ptr+size);
855 if (offset >= 0)
856 goto found;
857 if (!info->i_indirect)
858 goto lost2;
860 dir = shmem_dir_map(info->i_indirect);
861 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
863 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
864 if (unlikely(idx == stage)) {
865 shmem_dir_unmap(dir-1);
866 if (cond_resched_lock(&info->lock)) {
867 /* check it has not been truncated */
868 if (limit > info->next_index) {
869 limit = info->next_index;
870 if (idx >= limit)
871 goto lost2;
874 dir = shmem_dir_map(info->i_indirect) +
875 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
876 while (!*dir) {
877 dir++;
878 idx += ENTRIES_PER_PAGEPAGE;
879 if (idx >= limit)
880 goto lost1;
882 stage = idx + ENTRIES_PER_PAGEPAGE;
883 subdir = *dir;
884 shmem_dir_unmap(dir);
885 dir = shmem_dir_map(subdir);
887 subdir = *dir;
888 if (subdir && page_private(subdir)) {
889 ptr = shmem_swp_map(subdir);
890 size = limit - idx;
891 if (size > ENTRIES_PER_PAGE)
892 size = ENTRIES_PER_PAGE;
893 offset = shmem_find_swp(entry, ptr, ptr+size);
894 shmem_swp_unmap(ptr);
895 if (offset >= 0) {
896 shmem_dir_unmap(dir);
897 goto found;
901 lost1:
902 shmem_dir_unmap(dir-1);
903 lost2:
904 spin_unlock(&info->lock);
905 return 0;
906 found:
907 idx += offset;
908 inode = igrab(&info->vfs_inode);
909 spin_unlock(&info->lock);
912 * Move _head_ to start search for next from here.
913 * But be careful: shmem_delete_inode checks list_empty without taking
914 * mutex, and there's an instant in list_move_tail when info->swaplist
915 * would appear empty, if it were the only one on shmem_swaplist. We
916 * could avoid doing it if inode NULL; or use this minor optimization.
918 if (shmem_swaplist.next != &info->swaplist)
919 list_move_tail(&shmem_swaplist, &info->swaplist);
920 mutex_unlock(&shmem_swaplist_mutex);
922 error = 1;
923 if (!inode)
924 goto out;
925 /* Precharge page while we can wait, compensate afterwards */
926 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
927 if (error)
928 goto out;
929 error = radix_tree_preload(GFP_KERNEL);
930 if (error)
931 goto uncharge;
932 error = 1;
934 spin_lock(&info->lock);
935 ptr = shmem_swp_entry(info, idx, NULL);
936 if (ptr && ptr->val == entry.val)
937 error = add_to_page_cache(page, inode->i_mapping,
938 idx, GFP_NOWAIT);
939 if (error == -EEXIST) {
940 struct page *filepage = find_get_page(inode->i_mapping, idx);
941 error = 1;
942 if (filepage) {
944 * There might be a more uptodate page coming down
945 * from a stacked writepage: forget our swappage if so.
947 if (PageUptodate(filepage))
948 error = 0;
949 page_cache_release(filepage);
952 if (!error) {
953 delete_from_swap_cache(page);
954 set_page_dirty(page);
955 info->flags |= SHMEM_PAGEIN;
956 shmem_swp_set(info, ptr, 0);
957 swap_free(entry);
958 error = 1; /* not an error, but entry was found */
960 if (ptr)
961 shmem_swp_unmap(ptr);
962 spin_unlock(&info->lock);
963 radix_tree_preload_end();
964 uncharge:
965 mem_cgroup_uncharge_page(page);
966 out:
967 unlock_page(page);
968 page_cache_release(page);
969 iput(inode); /* allows for NULL */
970 return error;
974 * shmem_unuse() search for an eventually swapped out shmem page.
976 int shmem_unuse(swp_entry_t entry, struct page *page)
978 struct list_head *p, *next;
979 struct shmem_inode_info *info;
980 int found = 0;
982 mutex_lock(&shmem_swaplist_mutex);
983 list_for_each_safe(p, next, &shmem_swaplist) {
984 info = list_entry(p, struct shmem_inode_info, swaplist);
985 found = shmem_unuse_inode(info, entry, page);
986 cond_resched();
987 if (found)
988 goto out;
990 mutex_unlock(&shmem_swaplist_mutex);
991 out: return found; /* 0 or 1 or -ENOMEM */
995 * Move the page from the page cache to the swap cache.
997 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
999 struct shmem_inode_info *info;
1000 swp_entry_t *entry, swap;
1001 struct address_space *mapping;
1002 unsigned long index;
1003 struct inode *inode;
1005 BUG_ON(!PageLocked(page));
1006 mapping = page->mapping;
1007 index = page->index;
1008 inode = mapping->host;
1009 info = SHMEM_I(inode);
1010 if (info->flags & VM_LOCKED)
1011 goto redirty;
1012 if (!total_swap_pages)
1013 goto redirty;
1016 * shmem_backing_dev_info's capabilities prevent regular writeback or
1017 * sync from ever calling shmem_writepage; but a stacking filesystem
1018 * may use the ->writepage of its underlying filesystem, in which case
1019 * tmpfs should write out to swap only in response to memory pressure,
1020 * and not for pdflush or sync. However, in those cases, we do still
1021 * want to check if there's a redundant swappage to be discarded.
1023 if (wbc->for_reclaim)
1024 swap = get_swap_page();
1025 else
1026 swap.val = 0;
1028 spin_lock(&info->lock);
1029 if (index >= info->next_index) {
1030 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1031 goto unlock;
1033 entry = shmem_swp_entry(info, index, NULL);
1034 if (entry->val) {
1036 * The more uptodate page coming down from a stacked
1037 * writepage should replace our old swappage.
1039 free_swap_and_cache(*entry);
1040 shmem_swp_set(info, entry, 0);
1042 shmem_recalc_inode(inode);
1044 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1045 remove_from_page_cache(page);
1046 shmem_swp_set(info, entry, swap.val);
1047 shmem_swp_unmap(entry);
1048 if (list_empty(&info->swaplist))
1049 inode = igrab(inode);
1050 else
1051 inode = NULL;
1052 spin_unlock(&info->lock);
1053 swap_duplicate(swap);
1054 BUG_ON(page_mapped(page));
1055 page_cache_release(page); /* pagecache ref */
1056 set_page_dirty(page);
1057 unlock_page(page);
1058 if (inode) {
1059 mutex_lock(&shmem_swaplist_mutex);
1060 /* move instead of add in case we're racing */
1061 list_move_tail(&info->swaplist, &shmem_swaplist);
1062 mutex_unlock(&shmem_swaplist_mutex);
1063 iput(inode);
1065 return 0;
1068 shmem_swp_unmap(entry);
1069 unlock:
1070 spin_unlock(&info->lock);
1071 swap_free(swap);
1072 redirty:
1073 set_page_dirty(page);
1074 if (wbc->for_reclaim)
1075 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1076 unlock_page(page);
1077 return 0;
1080 #ifdef CONFIG_NUMA
1081 #ifdef CONFIG_TMPFS
1082 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1084 char buffer[64];
1086 if (!mpol || mpol->mode == MPOL_DEFAULT)
1087 return; /* show nothing */
1089 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1091 seq_printf(seq, ",mpol=%s", buffer);
1094 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1096 struct mempolicy *mpol = NULL;
1097 if (sbinfo->mpol) {
1098 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1099 mpol = sbinfo->mpol;
1100 mpol_get(mpol);
1101 spin_unlock(&sbinfo->stat_lock);
1103 return mpol;
1105 #endif /* CONFIG_TMPFS */
1107 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1108 struct shmem_inode_info *info, unsigned long idx)
1110 struct mempolicy mpol, *spol;
1111 struct vm_area_struct pvma;
1112 struct page *page;
1114 spol = mpol_cond_copy(&mpol,
1115 mpol_shared_policy_lookup(&info->policy, idx));
1117 /* Create a pseudo vma that just contains the policy */
1118 pvma.vm_start = 0;
1119 pvma.vm_pgoff = idx;
1120 pvma.vm_ops = NULL;
1121 pvma.vm_policy = spol;
1122 page = swapin_readahead(entry, gfp, &pvma, 0);
1123 return page;
1126 static struct page *shmem_alloc_page(gfp_t gfp,
1127 struct shmem_inode_info *info, unsigned long idx)
1129 struct vm_area_struct pvma;
1131 /* Create a pseudo vma that just contains the policy */
1132 pvma.vm_start = 0;
1133 pvma.vm_pgoff = idx;
1134 pvma.vm_ops = NULL;
1135 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1138 * alloc_page_vma() will drop the shared policy reference
1140 return alloc_page_vma(gfp, &pvma, 0);
1142 #else /* !CONFIG_NUMA */
1143 #ifdef CONFIG_TMPFS
1144 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1147 #endif /* CONFIG_TMPFS */
1149 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1150 struct shmem_inode_info *info, unsigned long idx)
1152 return swapin_readahead(entry, gfp, NULL, 0);
1155 static inline struct page *shmem_alloc_page(gfp_t gfp,
1156 struct shmem_inode_info *info, unsigned long idx)
1158 return alloc_page(gfp);
1160 #endif /* CONFIG_NUMA */
1162 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1163 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1165 return NULL;
1167 #endif
1170 * shmem_getpage - either get the page from swap or allocate a new one
1172 * If we allocate a new one we do not mark it dirty. That's up to the
1173 * vm. If we swap it in we mark it dirty since we also free the swap
1174 * entry since a page cannot live in both the swap and page cache
1176 static int shmem_getpage(struct inode *inode, unsigned long idx,
1177 struct page **pagep, enum sgp_type sgp, int *type)
1179 struct address_space *mapping = inode->i_mapping;
1180 struct shmem_inode_info *info = SHMEM_I(inode);
1181 struct shmem_sb_info *sbinfo;
1182 struct page *filepage = *pagep;
1183 struct page *swappage;
1184 swp_entry_t *entry;
1185 swp_entry_t swap;
1186 gfp_t gfp;
1187 int error;
1189 if (idx >= SHMEM_MAX_INDEX)
1190 return -EFBIG;
1192 if (type)
1193 *type = 0;
1196 * Normally, filepage is NULL on entry, and either found
1197 * uptodate immediately, or allocated and zeroed, or read
1198 * in under swappage, which is then assigned to filepage.
1199 * But shmem_readpage (required for splice) passes in a locked
1200 * filepage, which may be found not uptodate by other callers
1201 * too, and may need to be copied from the swappage read in.
1203 repeat:
1204 if (!filepage)
1205 filepage = find_lock_page(mapping, idx);
1206 if (filepage && PageUptodate(filepage))
1207 goto done;
1208 error = 0;
1209 gfp = mapping_gfp_mask(mapping);
1210 if (!filepage) {
1212 * Try to preload while we can wait, to not make a habit of
1213 * draining atomic reserves; but don't latch on to this cpu.
1215 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1216 if (error)
1217 goto failed;
1218 radix_tree_preload_end();
1221 spin_lock(&info->lock);
1222 shmem_recalc_inode(inode);
1223 entry = shmem_swp_alloc(info, idx, sgp);
1224 if (IS_ERR(entry)) {
1225 spin_unlock(&info->lock);
1226 error = PTR_ERR(entry);
1227 goto failed;
1229 swap = *entry;
1231 if (swap.val) {
1232 /* Look it up and read it in.. */
1233 swappage = lookup_swap_cache(swap);
1234 if (!swappage) {
1235 shmem_swp_unmap(entry);
1236 /* here we actually do the io */
1237 if (type && !(*type & VM_FAULT_MAJOR)) {
1238 __count_vm_event(PGMAJFAULT);
1239 *type |= VM_FAULT_MAJOR;
1241 spin_unlock(&info->lock);
1242 swappage = shmem_swapin(swap, gfp, info, idx);
1243 if (!swappage) {
1244 spin_lock(&info->lock);
1245 entry = shmem_swp_alloc(info, idx, sgp);
1246 if (IS_ERR(entry))
1247 error = PTR_ERR(entry);
1248 else {
1249 if (entry->val == swap.val)
1250 error = -ENOMEM;
1251 shmem_swp_unmap(entry);
1253 spin_unlock(&info->lock);
1254 if (error)
1255 goto failed;
1256 goto repeat;
1258 wait_on_page_locked(swappage);
1259 page_cache_release(swappage);
1260 goto repeat;
1263 /* We have to do this with page locked to prevent races */
1264 if (TestSetPageLocked(swappage)) {
1265 shmem_swp_unmap(entry);
1266 spin_unlock(&info->lock);
1267 wait_on_page_locked(swappage);
1268 page_cache_release(swappage);
1269 goto repeat;
1271 if (PageWriteback(swappage)) {
1272 shmem_swp_unmap(entry);
1273 spin_unlock(&info->lock);
1274 wait_on_page_writeback(swappage);
1275 unlock_page(swappage);
1276 page_cache_release(swappage);
1277 goto repeat;
1279 if (!PageUptodate(swappage)) {
1280 shmem_swp_unmap(entry);
1281 spin_unlock(&info->lock);
1282 unlock_page(swappage);
1283 page_cache_release(swappage);
1284 error = -EIO;
1285 goto failed;
1288 if (filepage) {
1289 shmem_swp_set(info, entry, 0);
1290 shmem_swp_unmap(entry);
1291 delete_from_swap_cache(swappage);
1292 spin_unlock(&info->lock);
1293 copy_highpage(filepage, swappage);
1294 unlock_page(swappage);
1295 page_cache_release(swappage);
1296 flush_dcache_page(filepage);
1297 SetPageUptodate(filepage);
1298 set_page_dirty(filepage);
1299 swap_free(swap);
1300 } else if (!(error = add_to_page_cache(
1301 swappage, mapping, idx, GFP_NOWAIT))) {
1302 info->flags |= SHMEM_PAGEIN;
1303 shmem_swp_set(info, entry, 0);
1304 shmem_swp_unmap(entry);
1305 delete_from_swap_cache(swappage);
1306 spin_unlock(&info->lock);
1307 filepage = swappage;
1308 set_page_dirty(filepage);
1309 swap_free(swap);
1310 } else {
1311 shmem_swp_unmap(entry);
1312 spin_unlock(&info->lock);
1313 unlock_page(swappage);
1314 if (error == -ENOMEM) {
1315 /* allow reclaim from this memory cgroup */
1316 error = mem_cgroup_cache_charge(swappage,
1317 current->mm, gfp & ~__GFP_HIGHMEM);
1318 if (error) {
1319 page_cache_release(swappage);
1320 goto failed;
1322 mem_cgroup_uncharge_page(swappage);
1324 page_cache_release(swappage);
1325 goto repeat;
1327 } else if (sgp == SGP_READ && !filepage) {
1328 shmem_swp_unmap(entry);
1329 filepage = find_get_page(mapping, idx);
1330 if (filepage &&
1331 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1332 spin_unlock(&info->lock);
1333 wait_on_page_locked(filepage);
1334 page_cache_release(filepage);
1335 filepage = NULL;
1336 goto repeat;
1338 spin_unlock(&info->lock);
1339 } else {
1340 shmem_swp_unmap(entry);
1341 sbinfo = SHMEM_SB(inode->i_sb);
1342 if (sbinfo->max_blocks) {
1343 spin_lock(&sbinfo->stat_lock);
1344 if (sbinfo->free_blocks == 0 ||
1345 shmem_acct_block(info->flags)) {
1346 spin_unlock(&sbinfo->stat_lock);
1347 spin_unlock(&info->lock);
1348 error = -ENOSPC;
1349 goto failed;
1351 sbinfo->free_blocks--;
1352 inode->i_blocks += BLOCKS_PER_PAGE;
1353 spin_unlock(&sbinfo->stat_lock);
1354 } else if (shmem_acct_block(info->flags)) {
1355 spin_unlock(&info->lock);
1356 error = -ENOSPC;
1357 goto failed;
1360 if (!filepage) {
1361 spin_unlock(&info->lock);
1362 filepage = shmem_alloc_page(gfp, info, idx);
1363 if (!filepage) {
1364 shmem_unacct_blocks(info->flags, 1);
1365 shmem_free_blocks(inode, 1);
1366 error = -ENOMEM;
1367 goto failed;
1370 /* Precharge page while we can wait, compensate after */
1371 error = mem_cgroup_cache_charge(filepage, current->mm,
1372 gfp & ~__GFP_HIGHMEM);
1373 if (error) {
1374 page_cache_release(filepage);
1375 shmem_unacct_blocks(info->flags, 1);
1376 shmem_free_blocks(inode, 1);
1377 filepage = NULL;
1378 goto failed;
1381 spin_lock(&info->lock);
1382 entry = shmem_swp_alloc(info, idx, sgp);
1383 if (IS_ERR(entry))
1384 error = PTR_ERR(entry);
1385 else {
1386 swap = *entry;
1387 shmem_swp_unmap(entry);
1389 if (error || swap.val || 0 != add_to_page_cache_lru(
1390 filepage, mapping, idx, GFP_NOWAIT)) {
1391 spin_unlock(&info->lock);
1392 mem_cgroup_uncharge_page(filepage);
1393 page_cache_release(filepage);
1394 shmem_unacct_blocks(info->flags, 1);
1395 shmem_free_blocks(inode, 1);
1396 filepage = NULL;
1397 if (error)
1398 goto failed;
1399 goto repeat;
1401 mem_cgroup_uncharge_page(filepage);
1402 info->flags |= SHMEM_PAGEIN;
1405 info->alloced++;
1406 spin_unlock(&info->lock);
1407 clear_highpage(filepage);
1408 flush_dcache_page(filepage);
1409 SetPageUptodate(filepage);
1410 if (sgp == SGP_DIRTY)
1411 set_page_dirty(filepage);
1413 done:
1414 *pagep = filepage;
1415 return 0;
1417 failed:
1418 if (*pagep != filepage) {
1419 unlock_page(filepage);
1420 page_cache_release(filepage);
1422 return error;
1425 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1427 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1428 int error;
1429 int ret;
1431 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1432 return VM_FAULT_SIGBUS;
1434 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1435 if (error)
1436 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1438 mark_page_accessed(vmf->page);
1439 return ret | VM_FAULT_LOCKED;
1442 #ifdef CONFIG_NUMA
1443 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1445 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1446 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1449 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1450 unsigned long addr)
1452 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1453 unsigned long idx;
1455 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1456 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1458 #endif
1460 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1462 struct inode *inode = file->f_path.dentry->d_inode;
1463 struct shmem_inode_info *info = SHMEM_I(inode);
1464 int retval = -ENOMEM;
1466 spin_lock(&info->lock);
1467 if (lock && !(info->flags & VM_LOCKED)) {
1468 if (!user_shm_lock(inode->i_size, user))
1469 goto out_nomem;
1470 info->flags |= VM_LOCKED;
1472 if (!lock && (info->flags & VM_LOCKED) && user) {
1473 user_shm_unlock(inode->i_size, user);
1474 info->flags &= ~VM_LOCKED;
1476 retval = 0;
1477 out_nomem:
1478 spin_unlock(&info->lock);
1479 return retval;
1482 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1484 file_accessed(file);
1485 vma->vm_ops = &shmem_vm_ops;
1486 vma->vm_flags |= VM_CAN_NONLINEAR;
1487 return 0;
1490 static struct inode *
1491 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1493 struct inode *inode;
1494 struct shmem_inode_info *info;
1495 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1497 if (shmem_reserve_inode(sb))
1498 return NULL;
1500 inode = new_inode(sb);
1501 if (inode) {
1502 inode->i_mode = mode;
1503 inode->i_uid = current->fsuid;
1504 inode->i_gid = current->fsgid;
1505 inode->i_blocks = 0;
1506 inode->i_mapping->a_ops = &shmem_aops;
1507 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1508 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1509 inode->i_generation = get_seconds();
1510 info = SHMEM_I(inode);
1511 memset(info, 0, (char *)inode - (char *)info);
1512 spin_lock_init(&info->lock);
1513 INIT_LIST_HEAD(&info->swaplist);
1515 switch (mode & S_IFMT) {
1516 default:
1517 inode->i_op = &shmem_special_inode_operations;
1518 init_special_inode(inode, mode, dev);
1519 break;
1520 case S_IFREG:
1521 inode->i_op = &shmem_inode_operations;
1522 inode->i_fop = &shmem_file_operations;
1523 mpol_shared_policy_init(&info->policy,
1524 shmem_get_sbmpol(sbinfo));
1525 break;
1526 case S_IFDIR:
1527 inc_nlink(inode);
1528 /* Some things misbehave if size == 0 on a directory */
1529 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1530 inode->i_op = &shmem_dir_inode_operations;
1531 inode->i_fop = &simple_dir_operations;
1532 break;
1533 case S_IFLNK:
1535 * Must not load anything in the rbtree,
1536 * mpol_free_shared_policy will not be called.
1538 mpol_shared_policy_init(&info->policy, NULL);
1539 break;
1541 } else
1542 shmem_free_inode(sb);
1543 return inode;
1546 #ifdef CONFIG_TMPFS
1547 static const struct inode_operations shmem_symlink_inode_operations;
1548 static const struct inode_operations shmem_symlink_inline_operations;
1551 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1552 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1553 * below the loop driver, in the generic fashion that many filesystems support.
1555 static int shmem_readpage(struct file *file, struct page *page)
1557 struct inode *inode = page->mapping->host;
1558 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1559 unlock_page(page);
1560 return error;
1563 static int
1564 shmem_write_begin(struct file *file, struct address_space *mapping,
1565 loff_t pos, unsigned len, unsigned flags,
1566 struct page **pagep, void **fsdata)
1568 struct inode *inode = mapping->host;
1569 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1570 *pagep = NULL;
1571 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1574 static int
1575 shmem_write_end(struct file *file, struct address_space *mapping,
1576 loff_t pos, unsigned len, unsigned copied,
1577 struct page *page, void *fsdata)
1579 struct inode *inode = mapping->host;
1581 if (pos + copied > inode->i_size)
1582 i_size_write(inode, pos + copied);
1584 unlock_page(page);
1585 set_page_dirty(page);
1586 page_cache_release(page);
1588 return copied;
1591 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1593 struct inode *inode = filp->f_path.dentry->d_inode;
1594 struct address_space *mapping = inode->i_mapping;
1595 unsigned long index, offset;
1596 enum sgp_type sgp = SGP_READ;
1599 * Might this read be for a stacking filesystem? Then when reading
1600 * holes of a sparse file, we actually need to allocate those pages,
1601 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1603 if (segment_eq(get_fs(), KERNEL_DS))
1604 sgp = SGP_DIRTY;
1606 index = *ppos >> PAGE_CACHE_SHIFT;
1607 offset = *ppos & ~PAGE_CACHE_MASK;
1609 for (;;) {
1610 struct page *page = NULL;
1611 unsigned long end_index, nr, ret;
1612 loff_t i_size = i_size_read(inode);
1614 end_index = i_size >> PAGE_CACHE_SHIFT;
1615 if (index > end_index)
1616 break;
1617 if (index == end_index) {
1618 nr = i_size & ~PAGE_CACHE_MASK;
1619 if (nr <= offset)
1620 break;
1623 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1624 if (desc->error) {
1625 if (desc->error == -EINVAL)
1626 desc->error = 0;
1627 break;
1629 if (page)
1630 unlock_page(page);
1633 * We must evaluate after, since reads (unlike writes)
1634 * are called without i_mutex protection against truncate
1636 nr = PAGE_CACHE_SIZE;
1637 i_size = i_size_read(inode);
1638 end_index = i_size >> PAGE_CACHE_SHIFT;
1639 if (index == end_index) {
1640 nr = i_size & ~PAGE_CACHE_MASK;
1641 if (nr <= offset) {
1642 if (page)
1643 page_cache_release(page);
1644 break;
1647 nr -= offset;
1649 if (page) {
1651 * If users can be writing to this page using arbitrary
1652 * virtual addresses, take care about potential aliasing
1653 * before reading the page on the kernel side.
1655 if (mapping_writably_mapped(mapping))
1656 flush_dcache_page(page);
1658 * Mark the page accessed if we read the beginning.
1660 if (!offset)
1661 mark_page_accessed(page);
1662 } else {
1663 page = ZERO_PAGE(0);
1664 page_cache_get(page);
1668 * Ok, we have the page, and it's up-to-date, so
1669 * now we can copy it to user space...
1671 * The actor routine returns how many bytes were actually used..
1672 * NOTE! This may not be the same as how much of a user buffer
1673 * we filled up (we may be padding etc), so we can only update
1674 * "pos" here (the actor routine has to update the user buffer
1675 * pointers and the remaining count).
1677 ret = actor(desc, page, offset, nr);
1678 offset += ret;
1679 index += offset >> PAGE_CACHE_SHIFT;
1680 offset &= ~PAGE_CACHE_MASK;
1682 page_cache_release(page);
1683 if (ret != nr || !desc->count)
1684 break;
1686 cond_resched();
1689 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1690 file_accessed(filp);
1693 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1694 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1696 struct file *filp = iocb->ki_filp;
1697 ssize_t retval;
1698 unsigned long seg;
1699 size_t count;
1700 loff_t *ppos = &iocb->ki_pos;
1702 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1703 if (retval)
1704 return retval;
1706 for (seg = 0; seg < nr_segs; seg++) {
1707 read_descriptor_t desc;
1709 desc.written = 0;
1710 desc.arg.buf = iov[seg].iov_base;
1711 desc.count = iov[seg].iov_len;
1712 if (desc.count == 0)
1713 continue;
1714 desc.error = 0;
1715 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1716 retval += desc.written;
1717 if (desc.error) {
1718 retval = retval ?: desc.error;
1719 break;
1721 if (desc.count > 0)
1722 break;
1724 return retval;
1727 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1729 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1731 buf->f_type = TMPFS_MAGIC;
1732 buf->f_bsize = PAGE_CACHE_SIZE;
1733 buf->f_namelen = NAME_MAX;
1734 spin_lock(&sbinfo->stat_lock);
1735 if (sbinfo->max_blocks) {
1736 buf->f_blocks = sbinfo->max_blocks;
1737 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1739 if (sbinfo->max_inodes) {
1740 buf->f_files = sbinfo->max_inodes;
1741 buf->f_ffree = sbinfo->free_inodes;
1743 /* else leave those fields 0 like simple_statfs */
1744 spin_unlock(&sbinfo->stat_lock);
1745 return 0;
1749 * File creation. Allocate an inode, and we're done..
1751 static int
1752 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1754 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1755 int error = -ENOSPC;
1757 if (inode) {
1758 error = security_inode_init_security(inode, dir, NULL, NULL,
1759 NULL);
1760 if (error) {
1761 if (error != -EOPNOTSUPP) {
1762 iput(inode);
1763 return error;
1766 error = shmem_acl_init(inode, dir);
1767 if (error) {
1768 iput(inode);
1769 return error;
1771 if (dir->i_mode & S_ISGID) {
1772 inode->i_gid = dir->i_gid;
1773 if (S_ISDIR(mode))
1774 inode->i_mode |= S_ISGID;
1776 dir->i_size += BOGO_DIRENT_SIZE;
1777 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1778 d_instantiate(dentry, inode);
1779 dget(dentry); /* Extra count - pin the dentry in core */
1781 return error;
1784 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1786 int error;
1788 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1789 return error;
1790 inc_nlink(dir);
1791 return 0;
1794 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1795 struct nameidata *nd)
1797 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1801 * Link a file..
1803 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1805 struct inode *inode = old_dentry->d_inode;
1806 int ret;
1809 * No ordinary (disk based) filesystem counts links as inodes;
1810 * but each new link needs a new dentry, pinning lowmem, and
1811 * tmpfs dentries cannot be pruned until they are unlinked.
1813 ret = shmem_reserve_inode(inode->i_sb);
1814 if (ret)
1815 goto out;
1817 dir->i_size += BOGO_DIRENT_SIZE;
1818 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1819 inc_nlink(inode);
1820 atomic_inc(&inode->i_count); /* New dentry reference */
1821 dget(dentry); /* Extra pinning count for the created dentry */
1822 d_instantiate(dentry, inode);
1823 out:
1824 return ret;
1827 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1829 struct inode *inode = dentry->d_inode;
1831 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1832 shmem_free_inode(inode->i_sb);
1834 dir->i_size -= BOGO_DIRENT_SIZE;
1835 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1836 drop_nlink(inode);
1837 dput(dentry); /* Undo the count from "create" - this does all the work */
1838 return 0;
1841 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1843 if (!simple_empty(dentry))
1844 return -ENOTEMPTY;
1846 drop_nlink(dentry->d_inode);
1847 drop_nlink(dir);
1848 return shmem_unlink(dir, dentry);
1852 * The VFS layer already does all the dentry stuff for rename,
1853 * we just have to decrement the usage count for the target if
1854 * it exists so that the VFS layer correctly free's it when it
1855 * gets overwritten.
1857 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1859 struct inode *inode = old_dentry->d_inode;
1860 int they_are_dirs = S_ISDIR(inode->i_mode);
1862 if (!simple_empty(new_dentry))
1863 return -ENOTEMPTY;
1865 if (new_dentry->d_inode) {
1866 (void) shmem_unlink(new_dir, new_dentry);
1867 if (they_are_dirs)
1868 drop_nlink(old_dir);
1869 } else if (they_are_dirs) {
1870 drop_nlink(old_dir);
1871 inc_nlink(new_dir);
1874 old_dir->i_size -= BOGO_DIRENT_SIZE;
1875 new_dir->i_size += BOGO_DIRENT_SIZE;
1876 old_dir->i_ctime = old_dir->i_mtime =
1877 new_dir->i_ctime = new_dir->i_mtime =
1878 inode->i_ctime = CURRENT_TIME;
1879 return 0;
1882 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1884 int error;
1885 int len;
1886 struct inode *inode;
1887 struct page *page = NULL;
1888 char *kaddr;
1889 struct shmem_inode_info *info;
1891 len = strlen(symname) + 1;
1892 if (len > PAGE_CACHE_SIZE)
1893 return -ENAMETOOLONG;
1895 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1896 if (!inode)
1897 return -ENOSPC;
1899 error = security_inode_init_security(inode, dir, NULL, NULL,
1900 NULL);
1901 if (error) {
1902 if (error != -EOPNOTSUPP) {
1903 iput(inode);
1904 return error;
1906 error = 0;
1909 info = SHMEM_I(inode);
1910 inode->i_size = len-1;
1911 if (len <= (char *)inode - (char *)info) {
1912 /* do it inline */
1913 memcpy(info, symname, len);
1914 inode->i_op = &shmem_symlink_inline_operations;
1915 } else {
1916 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1917 if (error) {
1918 iput(inode);
1919 return error;
1921 unlock_page(page);
1922 inode->i_op = &shmem_symlink_inode_operations;
1923 kaddr = kmap_atomic(page, KM_USER0);
1924 memcpy(kaddr, symname, len);
1925 kunmap_atomic(kaddr, KM_USER0);
1926 set_page_dirty(page);
1927 page_cache_release(page);
1929 if (dir->i_mode & S_ISGID)
1930 inode->i_gid = dir->i_gid;
1931 dir->i_size += BOGO_DIRENT_SIZE;
1932 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1933 d_instantiate(dentry, inode);
1934 dget(dentry);
1935 return 0;
1938 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1940 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1941 return NULL;
1944 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1946 struct page *page = NULL;
1947 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1948 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1949 if (page)
1950 unlock_page(page);
1951 return page;
1954 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1956 if (!IS_ERR(nd_get_link(nd))) {
1957 struct page *page = cookie;
1958 kunmap(page);
1959 mark_page_accessed(page);
1960 page_cache_release(page);
1964 static const struct inode_operations shmem_symlink_inline_operations = {
1965 .readlink = generic_readlink,
1966 .follow_link = shmem_follow_link_inline,
1969 static const struct inode_operations shmem_symlink_inode_operations = {
1970 .truncate = shmem_truncate,
1971 .readlink = generic_readlink,
1972 .follow_link = shmem_follow_link,
1973 .put_link = shmem_put_link,
1976 #ifdef CONFIG_TMPFS_POSIX_ACL
1978 * Superblocks without xattr inode operations will get security.* xattr
1979 * support from the VFS "for free". As soon as we have any other xattrs
1980 * like ACLs, we also need to implement the security.* handlers at
1981 * filesystem level, though.
1984 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1985 size_t list_len, const char *name,
1986 size_t name_len)
1988 return security_inode_listsecurity(inode, list, list_len);
1991 static int shmem_xattr_security_get(struct inode *inode, const char *name,
1992 void *buffer, size_t size)
1994 if (strcmp(name, "") == 0)
1995 return -EINVAL;
1996 return xattr_getsecurity(inode, name, buffer, size);
1999 static int shmem_xattr_security_set(struct inode *inode, const char *name,
2000 const void *value, size_t size, int flags)
2002 if (strcmp(name, "") == 0)
2003 return -EINVAL;
2004 return security_inode_setsecurity(inode, name, value, size, flags);
2007 static struct xattr_handler shmem_xattr_security_handler = {
2008 .prefix = XATTR_SECURITY_PREFIX,
2009 .list = shmem_xattr_security_list,
2010 .get = shmem_xattr_security_get,
2011 .set = shmem_xattr_security_set,
2014 static struct xattr_handler *shmem_xattr_handlers[] = {
2015 &shmem_xattr_acl_access_handler,
2016 &shmem_xattr_acl_default_handler,
2017 &shmem_xattr_security_handler,
2018 NULL
2020 #endif
2022 static struct dentry *shmem_get_parent(struct dentry *child)
2024 return ERR_PTR(-ESTALE);
2027 static int shmem_match(struct inode *ino, void *vfh)
2029 __u32 *fh = vfh;
2030 __u64 inum = fh[2];
2031 inum = (inum << 32) | fh[1];
2032 return ino->i_ino == inum && fh[0] == ino->i_generation;
2035 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2036 struct fid *fid, int fh_len, int fh_type)
2038 struct inode *inode;
2039 struct dentry *dentry = NULL;
2040 u64 inum = fid->raw[2];
2041 inum = (inum << 32) | fid->raw[1];
2043 if (fh_len < 3)
2044 return NULL;
2046 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2047 shmem_match, fid->raw);
2048 if (inode) {
2049 dentry = d_find_alias(inode);
2050 iput(inode);
2053 return dentry;
2056 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2057 int connectable)
2059 struct inode *inode = dentry->d_inode;
2061 if (*len < 3)
2062 return 255;
2064 if (hlist_unhashed(&inode->i_hash)) {
2065 /* Unfortunately insert_inode_hash is not idempotent,
2066 * so as we hash inodes here rather than at creation
2067 * time, we need a lock to ensure we only try
2068 * to do it once
2070 static DEFINE_SPINLOCK(lock);
2071 spin_lock(&lock);
2072 if (hlist_unhashed(&inode->i_hash))
2073 __insert_inode_hash(inode,
2074 inode->i_ino + inode->i_generation);
2075 spin_unlock(&lock);
2078 fh[0] = inode->i_generation;
2079 fh[1] = inode->i_ino;
2080 fh[2] = ((__u64)inode->i_ino) >> 32;
2082 *len = 3;
2083 return 1;
2086 static const struct export_operations shmem_export_ops = {
2087 .get_parent = shmem_get_parent,
2088 .encode_fh = shmem_encode_fh,
2089 .fh_to_dentry = shmem_fh_to_dentry,
2092 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2093 bool remount)
2095 char *this_char, *value, *rest;
2097 while (options != NULL) {
2098 this_char = options;
2099 for (;;) {
2101 * NUL-terminate this option: unfortunately,
2102 * mount options form a comma-separated list,
2103 * but mpol's nodelist may also contain commas.
2105 options = strchr(options, ',');
2106 if (options == NULL)
2107 break;
2108 options++;
2109 if (!isdigit(*options)) {
2110 options[-1] = '\0';
2111 break;
2114 if (!*this_char)
2115 continue;
2116 if ((value = strchr(this_char,'=')) != NULL) {
2117 *value++ = 0;
2118 } else {
2119 printk(KERN_ERR
2120 "tmpfs: No value for mount option '%s'\n",
2121 this_char);
2122 return 1;
2125 if (!strcmp(this_char,"size")) {
2126 unsigned long long size;
2127 size = memparse(value,&rest);
2128 if (*rest == '%') {
2129 size <<= PAGE_SHIFT;
2130 size *= totalram_pages;
2131 do_div(size, 100);
2132 rest++;
2134 if (*rest)
2135 goto bad_val;
2136 sbinfo->max_blocks =
2137 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2138 } else if (!strcmp(this_char,"nr_blocks")) {
2139 sbinfo->max_blocks = memparse(value, &rest);
2140 if (*rest)
2141 goto bad_val;
2142 } else if (!strcmp(this_char,"nr_inodes")) {
2143 sbinfo->max_inodes = memparse(value, &rest);
2144 if (*rest)
2145 goto bad_val;
2146 } else if (!strcmp(this_char,"mode")) {
2147 if (remount)
2148 continue;
2149 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2150 if (*rest)
2151 goto bad_val;
2152 } else if (!strcmp(this_char,"uid")) {
2153 if (remount)
2154 continue;
2155 sbinfo->uid = simple_strtoul(value, &rest, 0);
2156 if (*rest)
2157 goto bad_val;
2158 } else if (!strcmp(this_char,"gid")) {
2159 if (remount)
2160 continue;
2161 sbinfo->gid = simple_strtoul(value, &rest, 0);
2162 if (*rest)
2163 goto bad_val;
2164 } else if (!strcmp(this_char,"mpol")) {
2165 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2166 goto bad_val;
2167 } else {
2168 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2169 this_char);
2170 return 1;
2173 return 0;
2175 bad_val:
2176 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2177 value, this_char);
2178 return 1;
2182 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2184 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2185 struct shmem_sb_info config = *sbinfo;
2186 unsigned long blocks;
2187 unsigned long inodes;
2188 int error = -EINVAL;
2190 if (shmem_parse_options(data, &config, true))
2191 return error;
2193 spin_lock(&sbinfo->stat_lock);
2194 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2195 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2196 if (config.max_blocks < blocks)
2197 goto out;
2198 if (config.max_inodes < inodes)
2199 goto out;
2201 * Those tests also disallow limited->unlimited while any are in
2202 * use, so i_blocks will always be zero when max_blocks is zero;
2203 * but we must separately disallow unlimited->limited, because
2204 * in that case we have no record of how much is already in use.
2206 if (config.max_blocks && !sbinfo->max_blocks)
2207 goto out;
2208 if (config.max_inodes && !sbinfo->max_inodes)
2209 goto out;
2211 error = 0;
2212 sbinfo->max_blocks = config.max_blocks;
2213 sbinfo->free_blocks = config.max_blocks - blocks;
2214 sbinfo->max_inodes = config.max_inodes;
2215 sbinfo->free_inodes = config.max_inodes - inodes;
2217 mpol_put(sbinfo->mpol);
2218 sbinfo->mpol = config.mpol; /* transfers initial ref */
2219 out:
2220 spin_unlock(&sbinfo->stat_lock);
2221 return error;
2224 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2226 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2228 if (sbinfo->max_blocks != shmem_default_max_blocks())
2229 seq_printf(seq, ",size=%luk",
2230 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2231 if (sbinfo->max_inodes != shmem_default_max_inodes())
2232 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2233 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2234 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2235 if (sbinfo->uid != 0)
2236 seq_printf(seq, ",uid=%u", sbinfo->uid);
2237 if (sbinfo->gid != 0)
2238 seq_printf(seq, ",gid=%u", sbinfo->gid);
2239 shmem_show_mpol(seq, sbinfo->mpol);
2240 return 0;
2242 #endif /* CONFIG_TMPFS */
2244 static void shmem_put_super(struct super_block *sb)
2246 kfree(sb->s_fs_info);
2247 sb->s_fs_info = NULL;
2250 static int shmem_fill_super(struct super_block *sb,
2251 void *data, int silent)
2253 struct inode *inode;
2254 struct dentry *root;
2255 struct shmem_sb_info *sbinfo;
2256 int err = -ENOMEM;
2258 /* Round up to L1_CACHE_BYTES to resist false sharing */
2259 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2260 L1_CACHE_BYTES), GFP_KERNEL);
2261 if (!sbinfo)
2262 return -ENOMEM;
2264 sbinfo->max_blocks = 0;
2265 sbinfo->max_inodes = 0;
2266 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2267 sbinfo->uid = current->fsuid;
2268 sbinfo->gid = current->fsgid;
2269 sbinfo->mpol = NULL;
2270 sb->s_fs_info = sbinfo;
2272 #ifdef CONFIG_TMPFS
2274 * Per default we only allow half of the physical ram per
2275 * tmpfs instance, limiting inodes to one per page of lowmem;
2276 * but the internal instance is left unlimited.
2278 if (!(sb->s_flags & MS_NOUSER)) {
2279 sbinfo->max_blocks = shmem_default_max_blocks();
2280 sbinfo->max_inodes = shmem_default_max_inodes();
2281 if (shmem_parse_options(data, sbinfo, false)) {
2282 err = -EINVAL;
2283 goto failed;
2286 sb->s_export_op = &shmem_export_ops;
2287 #else
2288 sb->s_flags |= MS_NOUSER;
2289 #endif
2291 spin_lock_init(&sbinfo->stat_lock);
2292 sbinfo->free_blocks = sbinfo->max_blocks;
2293 sbinfo->free_inodes = sbinfo->max_inodes;
2295 sb->s_maxbytes = SHMEM_MAX_BYTES;
2296 sb->s_blocksize = PAGE_CACHE_SIZE;
2297 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2298 sb->s_magic = TMPFS_MAGIC;
2299 sb->s_op = &shmem_ops;
2300 sb->s_time_gran = 1;
2301 #ifdef CONFIG_TMPFS_POSIX_ACL
2302 sb->s_xattr = shmem_xattr_handlers;
2303 sb->s_flags |= MS_POSIXACL;
2304 #endif
2306 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0);
2307 if (!inode)
2308 goto failed;
2309 inode->i_uid = sbinfo->uid;
2310 inode->i_gid = sbinfo->gid;
2311 root = d_alloc_root(inode);
2312 if (!root)
2313 goto failed_iput;
2314 sb->s_root = root;
2315 return 0;
2317 failed_iput:
2318 iput(inode);
2319 failed:
2320 shmem_put_super(sb);
2321 return err;
2324 static struct kmem_cache *shmem_inode_cachep;
2326 static struct inode *shmem_alloc_inode(struct super_block *sb)
2328 struct shmem_inode_info *p;
2329 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2330 if (!p)
2331 return NULL;
2332 return &p->vfs_inode;
2335 static void shmem_destroy_inode(struct inode *inode)
2337 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2338 /* only struct inode is valid if it's an inline symlink */
2339 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2341 shmem_acl_destroy_inode(inode);
2342 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2345 static void init_once(struct kmem_cache *cachep, void *foo)
2347 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2349 inode_init_once(&p->vfs_inode);
2350 #ifdef CONFIG_TMPFS_POSIX_ACL
2351 p->i_acl = NULL;
2352 p->i_default_acl = NULL;
2353 #endif
2356 static int init_inodecache(void)
2358 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2359 sizeof(struct shmem_inode_info),
2360 0, SLAB_PANIC, init_once);
2361 return 0;
2364 static void destroy_inodecache(void)
2366 kmem_cache_destroy(shmem_inode_cachep);
2369 static const struct address_space_operations shmem_aops = {
2370 .writepage = shmem_writepage,
2371 .set_page_dirty = __set_page_dirty_no_writeback,
2372 #ifdef CONFIG_TMPFS
2373 .readpage = shmem_readpage,
2374 .write_begin = shmem_write_begin,
2375 .write_end = shmem_write_end,
2376 #endif
2377 .migratepage = migrate_page,
2380 static const struct file_operations shmem_file_operations = {
2381 .mmap = shmem_mmap,
2382 #ifdef CONFIG_TMPFS
2383 .llseek = generic_file_llseek,
2384 .read = do_sync_read,
2385 .write = do_sync_write,
2386 .aio_read = shmem_file_aio_read,
2387 .aio_write = generic_file_aio_write,
2388 .fsync = simple_sync_file,
2389 .splice_read = generic_file_splice_read,
2390 .splice_write = generic_file_splice_write,
2391 #endif
2394 static const struct inode_operations shmem_inode_operations = {
2395 .truncate = shmem_truncate,
2396 .setattr = shmem_notify_change,
2397 .truncate_range = shmem_truncate_range,
2398 #ifdef CONFIG_TMPFS_POSIX_ACL
2399 .setxattr = generic_setxattr,
2400 .getxattr = generic_getxattr,
2401 .listxattr = generic_listxattr,
2402 .removexattr = generic_removexattr,
2403 .permission = shmem_permission,
2404 #endif
2408 static const struct inode_operations shmem_dir_inode_operations = {
2409 #ifdef CONFIG_TMPFS
2410 .create = shmem_create,
2411 .lookup = simple_lookup,
2412 .link = shmem_link,
2413 .unlink = shmem_unlink,
2414 .symlink = shmem_symlink,
2415 .mkdir = shmem_mkdir,
2416 .rmdir = shmem_rmdir,
2417 .mknod = shmem_mknod,
2418 .rename = shmem_rename,
2419 #endif
2420 #ifdef CONFIG_TMPFS_POSIX_ACL
2421 .setattr = shmem_notify_change,
2422 .setxattr = generic_setxattr,
2423 .getxattr = generic_getxattr,
2424 .listxattr = generic_listxattr,
2425 .removexattr = generic_removexattr,
2426 .permission = shmem_permission,
2427 #endif
2430 static const struct inode_operations shmem_special_inode_operations = {
2431 #ifdef CONFIG_TMPFS_POSIX_ACL
2432 .setattr = shmem_notify_change,
2433 .setxattr = generic_setxattr,
2434 .getxattr = generic_getxattr,
2435 .listxattr = generic_listxattr,
2436 .removexattr = generic_removexattr,
2437 .permission = shmem_permission,
2438 #endif
2441 static const struct super_operations shmem_ops = {
2442 .alloc_inode = shmem_alloc_inode,
2443 .destroy_inode = shmem_destroy_inode,
2444 #ifdef CONFIG_TMPFS
2445 .statfs = shmem_statfs,
2446 .remount_fs = shmem_remount_fs,
2447 .show_options = shmem_show_options,
2448 #endif
2449 .delete_inode = shmem_delete_inode,
2450 .drop_inode = generic_delete_inode,
2451 .put_super = shmem_put_super,
2454 static struct vm_operations_struct shmem_vm_ops = {
2455 .fault = shmem_fault,
2456 #ifdef CONFIG_NUMA
2457 .set_policy = shmem_set_policy,
2458 .get_policy = shmem_get_policy,
2459 #endif
2463 static int shmem_get_sb(struct file_system_type *fs_type,
2464 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2466 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2469 static struct file_system_type tmpfs_fs_type = {
2470 .owner = THIS_MODULE,
2471 .name = "tmpfs",
2472 .get_sb = shmem_get_sb,
2473 .kill_sb = kill_litter_super,
2475 static struct vfsmount *shm_mnt;
2477 static int __init init_tmpfs(void)
2479 int error;
2481 error = bdi_init(&shmem_backing_dev_info);
2482 if (error)
2483 goto out4;
2485 error = init_inodecache();
2486 if (error)
2487 goto out3;
2489 error = register_filesystem(&tmpfs_fs_type);
2490 if (error) {
2491 printk(KERN_ERR "Could not register tmpfs\n");
2492 goto out2;
2495 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2496 tmpfs_fs_type.name, NULL);
2497 if (IS_ERR(shm_mnt)) {
2498 error = PTR_ERR(shm_mnt);
2499 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2500 goto out1;
2502 return 0;
2504 out1:
2505 unregister_filesystem(&tmpfs_fs_type);
2506 out2:
2507 destroy_inodecache();
2508 out3:
2509 bdi_destroy(&shmem_backing_dev_info);
2510 out4:
2511 shm_mnt = ERR_PTR(error);
2512 return error;
2514 module_init(init_tmpfs)
2517 * shmem_file_setup - get an unlinked file living in tmpfs
2518 * @name: name for dentry (to be seen in /proc/<pid>/maps
2519 * @size: size to be set for the file
2520 * @flags: vm_flags
2522 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2524 int error;
2525 struct file *file;
2526 struct inode *inode;
2527 struct dentry *dentry, *root;
2528 struct qstr this;
2530 if (IS_ERR(shm_mnt))
2531 return (void *)shm_mnt;
2533 if (size < 0 || size > SHMEM_MAX_BYTES)
2534 return ERR_PTR(-EINVAL);
2536 if (shmem_acct_size(flags, size))
2537 return ERR_PTR(-ENOMEM);
2539 error = -ENOMEM;
2540 this.name = name;
2541 this.len = strlen(name);
2542 this.hash = 0; /* will go */
2543 root = shm_mnt->mnt_root;
2544 dentry = d_alloc(root, &this);
2545 if (!dentry)
2546 goto put_memory;
2548 error = -ENFILE;
2549 file = get_empty_filp();
2550 if (!file)
2551 goto put_dentry;
2553 error = -ENOSPC;
2554 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2555 if (!inode)
2556 goto close_file;
2558 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2559 d_instantiate(dentry, inode);
2560 inode->i_size = size;
2561 inode->i_nlink = 0; /* It is unlinked */
2562 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
2563 &shmem_file_operations);
2564 return file;
2566 close_file:
2567 put_filp(file);
2568 put_dentry:
2569 dput(dentry);
2570 put_memory:
2571 shmem_unacct_size(flags, size);
2572 return ERR_PTR(error);
2576 * shmem_zero_setup - setup a shared anonymous mapping
2577 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2579 int shmem_zero_setup(struct vm_area_struct *vma)
2581 struct file *file;
2582 loff_t size = vma->vm_end - vma->vm_start;
2584 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2585 if (IS_ERR(file))
2586 return PTR_ERR(file);
2588 if (vma->vm_file)
2589 fput(vma->vm_file);
2590 vma->vm_file = file;
2591 vma->vm_ops = &shmem_vm_ops;
2592 return 0;