leds: add LED driver for Network Space v2 LEDs
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / mm / shmem.c
blobf65f84062db554ed2b3e9452cc787f8ab4f1b0d3
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>
33 static struct vfsmount *shm_mnt;
35 #ifdef CONFIG_SHMEM
37 * This virtual memory filesystem is heavily based on the ramfs. It
38 * extends ramfs by the ability to use swap and honor resource limits
39 * which makes it a completely usable filesystem.
42 #include <linux/xattr.h>
43 #include <linux/exportfs.h>
44 #include <linux/posix_acl.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 spin_lock(&info->lock);
438 if (!page) {
439 shmem_free_blocks(inode, 1);
440 return ERR_PTR(-ENOMEM);
442 if (sgp != SGP_WRITE &&
443 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
444 entry = ERR_PTR(-EINVAL);
445 break;
447 if (info->next_index <= index)
448 info->next_index = index + 1;
450 if (page) {
451 /* another task gave its page, or truncated the file */
452 shmem_free_blocks(inode, 1);
453 shmem_dir_free(page);
455 if (info->next_index <= index && !IS_ERR(entry))
456 info->next_index = index + 1;
457 return entry;
461 * shmem_free_swp - free some swap entries in a directory
462 * @dir: pointer to the directory
463 * @edir: pointer after last entry of the directory
464 * @punch_lock: pointer to spinlock when needed for the holepunch case
466 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
467 spinlock_t *punch_lock)
469 spinlock_t *punch_unlock = NULL;
470 swp_entry_t *ptr;
471 int freed = 0;
473 for (ptr = dir; ptr < edir; ptr++) {
474 if (ptr->val) {
475 if (unlikely(punch_lock)) {
476 punch_unlock = punch_lock;
477 punch_lock = NULL;
478 spin_lock(punch_unlock);
479 if (!ptr->val)
480 continue;
482 free_swap_and_cache(*ptr);
483 *ptr = (swp_entry_t){0};
484 freed++;
487 if (punch_unlock)
488 spin_unlock(punch_unlock);
489 return freed;
492 static int shmem_map_and_free_swp(struct page *subdir, int offset,
493 int limit, struct page ***dir, spinlock_t *punch_lock)
495 swp_entry_t *ptr;
496 int freed = 0;
498 ptr = shmem_swp_map(subdir);
499 for (; offset < limit; offset += LATENCY_LIMIT) {
500 int size = limit - offset;
501 if (size > LATENCY_LIMIT)
502 size = LATENCY_LIMIT;
503 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
504 punch_lock);
505 if (need_resched()) {
506 shmem_swp_unmap(ptr);
507 if (*dir) {
508 shmem_dir_unmap(*dir);
509 *dir = NULL;
511 cond_resched();
512 ptr = shmem_swp_map(subdir);
515 shmem_swp_unmap(ptr);
516 return freed;
519 static void shmem_free_pages(struct list_head *next)
521 struct page *page;
522 int freed = 0;
524 do {
525 page = container_of(next, struct page, lru);
526 next = next->next;
527 shmem_dir_free(page);
528 freed++;
529 if (freed >= LATENCY_LIMIT) {
530 cond_resched();
531 freed = 0;
533 } while (next);
536 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
538 struct shmem_inode_info *info = SHMEM_I(inode);
539 unsigned long idx;
540 unsigned long size;
541 unsigned long limit;
542 unsigned long stage;
543 unsigned long diroff;
544 struct page **dir;
545 struct page *topdir;
546 struct page *middir;
547 struct page *subdir;
548 swp_entry_t *ptr;
549 LIST_HEAD(pages_to_free);
550 long nr_pages_to_free = 0;
551 long nr_swaps_freed = 0;
552 int offset;
553 int freed;
554 int punch_hole;
555 spinlock_t *needs_lock;
556 spinlock_t *punch_lock;
557 unsigned long upper_limit;
559 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
560 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
561 if (idx >= info->next_index)
562 return;
564 spin_lock(&info->lock);
565 info->flags |= SHMEM_TRUNCATE;
566 if (likely(end == (loff_t) -1)) {
567 limit = info->next_index;
568 upper_limit = SHMEM_MAX_INDEX;
569 info->next_index = idx;
570 needs_lock = NULL;
571 punch_hole = 0;
572 } else {
573 if (end + 1 >= inode->i_size) { /* we may free a little more */
574 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
575 PAGE_CACHE_SHIFT;
576 upper_limit = SHMEM_MAX_INDEX;
577 } else {
578 limit = (end + 1) >> PAGE_CACHE_SHIFT;
579 upper_limit = limit;
581 needs_lock = &info->lock;
582 punch_hole = 1;
585 topdir = info->i_indirect;
586 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
587 info->i_indirect = NULL;
588 nr_pages_to_free++;
589 list_add(&topdir->lru, &pages_to_free);
591 spin_unlock(&info->lock);
593 if (info->swapped && idx < SHMEM_NR_DIRECT) {
594 ptr = info->i_direct;
595 size = limit;
596 if (size > SHMEM_NR_DIRECT)
597 size = SHMEM_NR_DIRECT;
598 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
602 * If there are no indirect blocks or we are punching a hole
603 * below indirect blocks, nothing to be done.
605 if (!topdir || limit <= SHMEM_NR_DIRECT)
606 goto done2;
609 * The truncation case has already dropped info->lock, and we're safe
610 * because i_size and next_index have already been lowered, preventing
611 * access beyond. But in the punch_hole case, we still need to take
612 * the lock when updating the swap directory, because there might be
613 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
614 * shmem_writepage. However, whenever we find we can remove a whole
615 * directory page (not at the misaligned start or end of the range),
616 * we first NULLify its pointer in the level above, and then have no
617 * need to take the lock when updating its contents: needs_lock and
618 * punch_lock (either pointing to info->lock or NULL) manage this.
621 upper_limit -= SHMEM_NR_DIRECT;
622 limit -= SHMEM_NR_DIRECT;
623 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
624 offset = idx % ENTRIES_PER_PAGE;
625 idx -= offset;
627 dir = shmem_dir_map(topdir);
628 stage = ENTRIES_PER_PAGEPAGE/2;
629 if (idx < ENTRIES_PER_PAGEPAGE/2) {
630 middir = topdir;
631 diroff = idx/ENTRIES_PER_PAGE;
632 } else {
633 dir += ENTRIES_PER_PAGE/2;
634 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
635 while (stage <= idx)
636 stage += ENTRIES_PER_PAGEPAGE;
637 middir = *dir;
638 if (*dir) {
639 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
640 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
641 if (!diroff && !offset && upper_limit >= stage) {
642 if (needs_lock) {
643 spin_lock(needs_lock);
644 *dir = NULL;
645 spin_unlock(needs_lock);
646 needs_lock = NULL;
647 } else
648 *dir = NULL;
649 nr_pages_to_free++;
650 list_add(&middir->lru, &pages_to_free);
652 shmem_dir_unmap(dir);
653 dir = shmem_dir_map(middir);
654 } else {
655 diroff = 0;
656 offset = 0;
657 idx = stage;
661 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
662 if (unlikely(idx == stage)) {
663 shmem_dir_unmap(dir);
664 dir = shmem_dir_map(topdir) +
665 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
666 while (!*dir) {
667 dir++;
668 idx += ENTRIES_PER_PAGEPAGE;
669 if (idx >= limit)
670 goto done1;
672 stage = idx + ENTRIES_PER_PAGEPAGE;
673 middir = *dir;
674 if (punch_hole)
675 needs_lock = &info->lock;
676 if (upper_limit >= stage) {
677 if (needs_lock) {
678 spin_lock(needs_lock);
679 *dir = NULL;
680 spin_unlock(needs_lock);
681 needs_lock = NULL;
682 } else
683 *dir = NULL;
684 nr_pages_to_free++;
685 list_add(&middir->lru, &pages_to_free);
687 shmem_dir_unmap(dir);
688 cond_resched();
689 dir = shmem_dir_map(middir);
690 diroff = 0;
692 punch_lock = needs_lock;
693 subdir = dir[diroff];
694 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
695 if (needs_lock) {
696 spin_lock(needs_lock);
697 dir[diroff] = NULL;
698 spin_unlock(needs_lock);
699 punch_lock = NULL;
700 } else
701 dir[diroff] = NULL;
702 nr_pages_to_free++;
703 list_add(&subdir->lru, &pages_to_free);
705 if (subdir && page_private(subdir) /* has swap entries */) {
706 size = limit - idx;
707 if (size > ENTRIES_PER_PAGE)
708 size = ENTRIES_PER_PAGE;
709 freed = shmem_map_and_free_swp(subdir,
710 offset, size, &dir, punch_lock);
711 if (!dir)
712 dir = shmem_dir_map(middir);
713 nr_swaps_freed += freed;
714 if (offset || punch_lock) {
715 spin_lock(&info->lock);
716 set_page_private(subdir,
717 page_private(subdir) - freed);
718 spin_unlock(&info->lock);
719 } else
720 BUG_ON(page_private(subdir) != freed);
722 offset = 0;
724 done1:
725 shmem_dir_unmap(dir);
726 done2:
727 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
729 * Call truncate_inode_pages again: racing shmem_unuse_inode
730 * may have swizzled a page in from swap since
731 * truncate_pagecache or generic_delete_inode did it, before we
732 * lowered next_index. Also, though shmem_getpage checks
733 * i_size before adding to cache, no recheck after: so fix the
734 * narrow window there too.
736 * Recalling truncate_inode_pages_range and unmap_mapping_range
737 * every time for punch_hole (which never got a chance to clear
738 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
739 * yet hardly ever necessary: try to optimize them out later.
741 truncate_inode_pages_range(inode->i_mapping, start, end);
742 if (punch_hole)
743 unmap_mapping_range(inode->i_mapping, start,
744 end - start, 1);
747 spin_lock(&info->lock);
748 info->flags &= ~SHMEM_TRUNCATE;
749 info->swapped -= nr_swaps_freed;
750 if (nr_pages_to_free)
751 shmem_free_blocks(inode, nr_pages_to_free);
752 shmem_recalc_inode(inode);
753 spin_unlock(&info->lock);
756 * Empty swap vector directory pages to be freed?
758 if (!list_empty(&pages_to_free)) {
759 pages_to_free.prev->next = NULL;
760 shmem_free_pages(pages_to_free.next);
764 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
766 struct inode *inode = dentry->d_inode;
767 loff_t newsize = attr->ia_size;
768 int error;
770 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)
771 && newsize != inode->i_size) {
772 struct page *page = NULL;
774 if (newsize < inode->i_size) {
776 * If truncating down to a partial page, then
777 * if that page is already allocated, hold it
778 * in memory until the truncation is over, so
779 * truncate_partial_page cannnot miss it were
780 * it assigned to swap.
782 if (newsize & (PAGE_CACHE_SIZE-1)) {
783 (void) shmem_getpage(inode,
784 newsize >> PAGE_CACHE_SHIFT,
785 &page, SGP_READ, NULL);
786 if (page)
787 unlock_page(page);
790 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
791 * detect if any pages might have been added to cache
792 * after truncate_inode_pages. But we needn't bother
793 * if it's being fully truncated to zero-length: the
794 * nrpages check is efficient enough in that case.
796 if (newsize) {
797 struct shmem_inode_info *info = SHMEM_I(inode);
798 spin_lock(&info->lock);
799 info->flags &= ~SHMEM_PAGEIN;
800 spin_unlock(&info->lock);
804 error = simple_setsize(inode, newsize);
805 if (page)
806 page_cache_release(page);
807 if (error)
808 return error;
809 shmem_truncate_range(inode, newsize, (loff_t)-1);
812 error = inode_change_ok(inode, attr);
813 if (!error)
814 generic_setattr(inode, attr);
815 #ifdef CONFIG_TMPFS_POSIX_ACL
816 if (!error && (attr->ia_valid & ATTR_MODE))
817 error = generic_acl_chmod(inode);
818 #endif
819 return error;
822 static void shmem_delete_inode(struct inode *inode)
824 struct shmem_inode_info *info = SHMEM_I(inode);
826 if (inode->i_mapping->a_ops == &shmem_aops) {
827 truncate_inode_pages(inode->i_mapping, 0);
828 shmem_unacct_size(info->flags, inode->i_size);
829 inode->i_size = 0;
830 shmem_truncate_range(inode, 0, (loff_t)-1);
831 if (!list_empty(&info->swaplist)) {
832 mutex_lock(&shmem_swaplist_mutex);
833 list_del_init(&info->swaplist);
834 mutex_unlock(&shmem_swaplist_mutex);
837 BUG_ON(inode->i_blocks);
838 shmem_free_inode(inode->i_sb);
839 clear_inode(inode);
842 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
844 swp_entry_t *ptr;
846 for (ptr = dir; ptr < edir; ptr++) {
847 if (ptr->val == entry.val)
848 return ptr - dir;
850 return -1;
853 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
855 struct inode *inode;
856 unsigned long idx;
857 unsigned long size;
858 unsigned long limit;
859 unsigned long stage;
860 struct page **dir;
861 struct page *subdir;
862 swp_entry_t *ptr;
863 int offset;
864 int error;
866 idx = 0;
867 ptr = info->i_direct;
868 spin_lock(&info->lock);
869 if (!info->swapped) {
870 list_del_init(&info->swaplist);
871 goto lost2;
873 limit = info->next_index;
874 size = limit;
875 if (size > SHMEM_NR_DIRECT)
876 size = SHMEM_NR_DIRECT;
877 offset = shmem_find_swp(entry, ptr, ptr+size);
878 if (offset >= 0)
879 goto found;
880 if (!info->i_indirect)
881 goto lost2;
883 dir = shmem_dir_map(info->i_indirect);
884 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
886 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
887 if (unlikely(idx == stage)) {
888 shmem_dir_unmap(dir-1);
889 if (cond_resched_lock(&info->lock)) {
890 /* check it has not been truncated */
891 if (limit > info->next_index) {
892 limit = info->next_index;
893 if (idx >= limit)
894 goto lost2;
897 dir = shmem_dir_map(info->i_indirect) +
898 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
899 while (!*dir) {
900 dir++;
901 idx += ENTRIES_PER_PAGEPAGE;
902 if (idx >= limit)
903 goto lost1;
905 stage = idx + ENTRIES_PER_PAGEPAGE;
906 subdir = *dir;
907 shmem_dir_unmap(dir);
908 dir = shmem_dir_map(subdir);
910 subdir = *dir;
911 if (subdir && page_private(subdir)) {
912 ptr = shmem_swp_map(subdir);
913 size = limit - idx;
914 if (size > ENTRIES_PER_PAGE)
915 size = ENTRIES_PER_PAGE;
916 offset = shmem_find_swp(entry, ptr, ptr+size);
917 shmem_swp_unmap(ptr);
918 if (offset >= 0) {
919 shmem_dir_unmap(dir);
920 goto found;
924 lost1:
925 shmem_dir_unmap(dir-1);
926 lost2:
927 spin_unlock(&info->lock);
928 return 0;
929 found:
930 idx += offset;
931 inode = igrab(&info->vfs_inode);
932 spin_unlock(&info->lock);
935 * Move _head_ to start search for next from here.
936 * But be careful: shmem_delete_inode checks list_empty without taking
937 * mutex, and there's an instant in list_move_tail when info->swaplist
938 * would appear empty, if it were the only one on shmem_swaplist. We
939 * could avoid doing it if inode NULL; or use this minor optimization.
941 if (shmem_swaplist.next != &info->swaplist)
942 list_move_tail(&shmem_swaplist, &info->swaplist);
943 mutex_unlock(&shmem_swaplist_mutex);
945 error = 1;
946 if (!inode)
947 goto out;
949 * Charge page using GFP_KERNEL while we can wait.
950 * Charged back to the user(not to caller) when swap account is used.
951 * add_to_page_cache() will be called with GFP_NOWAIT.
953 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
954 if (error)
955 goto out;
956 error = radix_tree_preload(GFP_KERNEL);
957 if (error) {
958 mem_cgroup_uncharge_cache_page(page);
959 goto out;
961 error = 1;
963 spin_lock(&info->lock);
964 ptr = shmem_swp_entry(info, idx, NULL);
965 if (ptr && ptr->val == entry.val) {
966 error = add_to_page_cache_locked(page, inode->i_mapping,
967 idx, GFP_NOWAIT);
968 /* does mem_cgroup_uncharge_cache_page on error */
969 } else /* we must compensate for our precharge above */
970 mem_cgroup_uncharge_cache_page(page);
972 if (error == -EEXIST) {
973 struct page *filepage = find_get_page(inode->i_mapping, idx);
974 error = 1;
975 if (filepage) {
977 * There might be a more uptodate page coming down
978 * from a stacked writepage: forget our swappage if so.
980 if (PageUptodate(filepage))
981 error = 0;
982 page_cache_release(filepage);
985 if (!error) {
986 delete_from_swap_cache(page);
987 set_page_dirty(page);
988 info->flags |= SHMEM_PAGEIN;
989 shmem_swp_set(info, ptr, 0);
990 swap_free(entry);
991 error = 1; /* not an error, but entry was found */
993 if (ptr)
994 shmem_swp_unmap(ptr);
995 spin_unlock(&info->lock);
996 radix_tree_preload_end();
997 out:
998 unlock_page(page);
999 page_cache_release(page);
1000 iput(inode); /* allows for NULL */
1001 return error;
1005 * shmem_unuse() search for an eventually swapped out shmem page.
1007 int shmem_unuse(swp_entry_t entry, struct page *page)
1009 struct list_head *p, *next;
1010 struct shmem_inode_info *info;
1011 int found = 0;
1013 mutex_lock(&shmem_swaplist_mutex);
1014 list_for_each_safe(p, next, &shmem_swaplist) {
1015 info = list_entry(p, struct shmem_inode_info, swaplist);
1016 found = shmem_unuse_inode(info, entry, page);
1017 cond_resched();
1018 if (found)
1019 goto out;
1021 mutex_unlock(&shmem_swaplist_mutex);
1023 * Can some race bring us here? We've been holding page lock,
1024 * so I think not; but would rather try again later than BUG()
1026 unlock_page(page);
1027 page_cache_release(page);
1028 out:
1029 return (found < 0) ? found : 0;
1033 * Move the page from the page cache to the swap cache.
1035 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1037 struct shmem_inode_info *info;
1038 swp_entry_t *entry, swap;
1039 struct address_space *mapping;
1040 unsigned long index;
1041 struct inode *inode;
1043 BUG_ON(!PageLocked(page));
1044 mapping = page->mapping;
1045 index = page->index;
1046 inode = mapping->host;
1047 info = SHMEM_I(inode);
1048 if (info->flags & VM_LOCKED)
1049 goto redirty;
1050 if (!total_swap_pages)
1051 goto redirty;
1054 * shmem_backing_dev_info's capabilities prevent regular writeback or
1055 * sync from ever calling shmem_writepage; but a stacking filesystem
1056 * may use the ->writepage of its underlying filesystem, in which case
1057 * tmpfs should write out to swap only in response to memory pressure,
1058 * and not for the writeback threads or sync. However, in those cases,
1059 * we do still want to check if there's a redundant swappage to be
1060 * discarded.
1062 if (wbc->for_reclaim)
1063 swap = get_swap_page();
1064 else
1065 swap.val = 0;
1067 spin_lock(&info->lock);
1068 if (index >= info->next_index) {
1069 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1070 goto unlock;
1072 entry = shmem_swp_entry(info, index, NULL);
1073 if (entry->val) {
1075 * The more uptodate page coming down from a stacked
1076 * writepage should replace our old swappage.
1078 free_swap_and_cache(*entry);
1079 shmem_swp_set(info, entry, 0);
1081 shmem_recalc_inode(inode);
1083 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1084 remove_from_page_cache(page);
1085 shmem_swp_set(info, entry, swap.val);
1086 shmem_swp_unmap(entry);
1087 if (list_empty(&info->swaplist))
1088 inode = igrab(inode);
1089 else
1090 inode = NULL;
1091 spin_unlock(&info->lock);
1092 swap_shmem_alloc(swap);
1093 BUG_ON(page_mapped(page));
1094 page_cache_release(page); /* pagecache ref */
1095 swap_writepage(page, wbc);
1096 if (inode) {
1097 mutex_lock(&shmem_swaplist_mutex);
1098 /* move instead of add in case we're racing */
1099 list_move_tail(&info->swaplist, &shmem_swaplist);
1100 mutex_unlock(&shmem_swaplist_mutex);
1101 iput(inode);
1103 return 0;
1106 shmem_swp_unmap(entry);
1107 unlock:
1108 spin_unlock(&info->lock);
1110 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1111 * clear SWAP_HAS_CACHE flag.
1113 swapcache_free(swap, NULL);
1114 redirty:
1115 set_page_dirty(page);
1116 if (wbc->for_reclaim)
1117 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1118 unlock_page(page);
1119 return 0;
1122 #ifdef CONFIG_NUMA
1123 #ifdef CONFIG_TMPFS
1124 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1126 char buffer[64];
1128 if (!mpol || mpol->mode == MPOL_DEFAULT)
1129 return; /* show nothing */
1131 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1133 seq_printf(seq, ",mpol=%s", buffer);
1136 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1138 struct mempolicy *mpol = NULL;
1139 if (sbinfo->mpol) {
1140 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1141 mpol = sbinfo->mpol;
1142 mpol_get(mpol);
1143 spin_unlock(&sbinfo->stat_lock);
1145 return mpol;
1147 #endif /* CONFIG_TMPFS */
1149 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1150 struct shmem_inode_info *info, unsigned long idx)
1152 struct mempolicy mpol, *spol;
1153 struct vm_area_struct pvma;
1154 struct page *page;
1156 spol = mpol_cond_copy(&mpol,
1157 mpol_shared_policy_lookup(&info->policy, idx));
1159 /* Create a pseudo vma that just contains the policy */
1160 pvma.vm_start = 0;
1161 pvma.vm_pgoff = idx;
1162 pvma.vm_ops = NULL;
1163 pvma.vm_policy = spol;
1164 page = swapin_readahead(entry, gfp, &pvma, 0);
1165 return page;
1168 static struct page *shmem_alloc_page(gfp_t gfp,
1169 struct shmem_inode_info *info, unsigned long idx)
1171 struct vm_area_struct pvma;
1173 /* Create a pseudo vma that just contains the policy */
1174 pvma.vm_start = 0;
1175 pvma.vm_pgoff = idx;
1176 pvma.vm_ops = NULL;
1177 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1180 * alloc_page_vma() will drop the shared policy reference
1182 return alloc_page_vma(gfp, &pvma, 0);
1184 #else /* !CONFIG_NUMA */
1185 #ifdef CONFIG_TMPFS
1186 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1189 #endif /* CONFIG_TMPFS */
1191 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1192 struct shmem_inode_info *info, unsigned long idx)
1194 return swapin_readahead(entry, gfp, NULL, 0);
1197 static inline struct page *shmem_alloc_page(gfp_t gfp,
1198 struct shmem_inode_info *info, unsigned long idx)
1200 return alloc_page(gfp);
1202 #endif /* CONFIG_NUMA */
1204 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1205 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1207 return NULL;
1209 #endif
1212 * shmem_getpage - either get the page from swap or allocate a new one
1214 * If we allocate a new one we do not mark it dirty. That's up to the
1215 * vm. If we swap it in we mark it dirty since we also free the swap
1216 * entry since a page cannot live in both the swap and page cache
1218 static int shmem_getpage(struct inode *inode, unsigned long idx,
1219 struct page **pagep, enum sgp_type sgp, int *type)
1221 struct address_space *mapping = inode->i_mapping;
1222 struct shmem_inode_info *info = SHMEM_I(inode);
1223 struct shmem_sb_info *sbinfo;
1224 struct page *filepage = *pagep;
1225 struct page *swappage;
1226 swp_entry_t *entry;
1227 swp_entry_t swap;
1228 gfp_t gfp;
1229 int error;
1231 if (idx >= SHMEM_MAX_INDEX)
1232 return -EFBIG;
1234 if (type)
1235 *type = 0;
1238 * Normally, filepage is NULL on entry, and either found
1239 * uptodate immediately, or allocated and zeroed, or read
1240 * in under swappage, which is then assigned to filepage.
1241 * But shmem_readpage (required for splice) passes in a locked
1242 * filepage, which may be found not uptodate by other callers
1243 * too, and may need to be copied from the swappage read in.
1245 repeat:
1246 if (!filepage)
1247 filepage = find_lock_page(mapping, idx);
1248 if (filepage && PageUptodate(filepage))
1249 goto done;
1250 error = 0;
1251 gfp = mapping_gfp_mask(mapping);
1252 if (!filepage) {
1254 * Try to preload while we can wait, to not make a habit of
1255 * draining atomic reserves; but don't latch on to this cpu.
1257 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1258 if (error)
1259 goto failed;
1260 radix_tree_preload_end();
1263 spin_lock(&info->lock);
1264 shmem_recalc_inode(inode);
1265 entry = shmem_swp_alloc(info, idx, sgp);
1266 if (IS_ERR(entry)) {
1267 spin_unlock(&info->lock);
1268 error = PTR_ERR(entry);
1269 goto failed;
1271 swap = *entry;
1273 if (swap.val) {
1274 /* Look it up and read it in.. */
1275 swappage = lookup_swap_cache(swap);
1276 if (!swappage) {
1277 shmem_swp_unmap(entry);
1278 /* here we actually do the io */
1279 if (type && !(*type & VM_FAULT_MAJOR)) {
1280 __count_vm_event(PGMAJFAULT);
1281 *type |= VM_FAULT_MAJOR;
1283 spin_unlock(&info->lock);
1284 swappage = shmem_swapin(swap, gfp, info, idx);
1285 if (!swappage) {
1286 spin_lock(&info->lock);
1287 entry = shmem_swp_alloc(info, idx, sgp);
1288 if (IS_ERR(entry))
1289 error = PTR_ERR(entry);
1290 else {
1291 if (entry->val == swap.val)
1292 error = -ENOMEM;
1293 shmem_swp_unmap(entry);
1295 spin_unlock(&info->lock);
1296 if (error)
1297 goto failed;
1298 goto repeat;
1300 wait_on_page_locked(swappage);
1301 page_cache_release(swappage);
1302 goto repeat;
1305 /* We have to do this with page locked to prevent races */
1306 if (!trylock_page(swappage)) {
1307 shmem_swp_unmap(entry);
1308 spin_unlock(&info->lock);
1309 wait_on_page_locked(swappage);
1310 page_cache_release(swappage);
1311 goto repeat;
1313 if (PageWriteback(swappage)) {
1314 shmem_swp_unmap(entry);
1315 spin_unlock(&info->lock);
1316 wait_on_page_writeback(swappage);
1317 unlock_page(swappage);
1318 page_cache_release(swappage);
1319 goto repeat;
1321 if (!PageUptodate(swappage)) {
1322 shmem_swp_unmap(entry);
1323 spin_unlock(&info->lock);
1324 unlock_page(swappage);
1325 page_cache_release(swappage);
1326 error = -EIO;
1327 goto failed;
1330 if (filepage) {
1331 shmem_swp_set(info, entry, 0);
1332 shmem_swp_unmap(entry);
1333 delete_from_swap_cache(swappage);
1334 spin_unlock(&info->lock);
1335 copy_highpage(filepage, swappage);
1336 unlock_page(swappage);
1337 page_cache_release(swappage);
1338 flush_dcache_page(filepage);
1339 SetPageUptodate(filepage);
1340 set_page_dirty(filepage);
1341 swap_free(swap);
1342 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1343 idx, GFP_NOWAIT))) {
1344 info->flags |= SHMEM_PAGEIN;
1345 shmem_swp_set(info, entry, 0);
1346 shmem_swp_unmap(entry);
1347 delete_from_swap_cache(swappage);
1348 spin_unlock(&info->lock);
1349 filepage = swappage;
1350 set_page_dirty(filepage);
1351 swap_free(swap);
1352 } else {
1353 shmem_swp_unmap(entry);
1354 spin_unlock(&info->lock);
1355 if (error == -ENOMEM) {
1357 * reclaim from proper memory cgroup and
1358 * call memcg's OOM if needed.
1360 error = mem_cgroup_shmem_charge_fallback(
1361 swappage,
1362 current->mm,
1363 gfp);
1364 if (error) {
1365 unlock_page(swappage);
1366 page_cache_release(swappage);
1367 goto failed;
1370 unlock_page(swappage);
1371 page_cache_release(swappage);
1372 goto repeat;
1374 } else if (sgp == SGP_READ && !filepage) {
1375 shmem_swp_unmap(entry);
1376 filepage = find_get_page(mapping, idx);
1377 if (filepage &&
1378 (!PageUptodate(filepage) || !trylock_page(filepage))) {
1379 spin_unlock(&info->lock);
1380 wait_on_page_locked(filepage);
1381 page_cache_release(filepage);
1382 filepage = NULL;
1383 goto repeat;
1385 spin_unlock(&info->lock);
1386 } else {
1387 shmem_swp_unmap(entry);
1388 sbinfo = SHMEM_SB(inode->i_sb);
1389 if (sbinfo->max_blocks) {
1390 spin_lock(&sbinfo->stat_lock);
1391 if (sbinfo->free_blocks == 0 ||
1392 shmem_acct_block(info->flags)) {
1393 spin_unlock(&sbinfo->stat_lock);
1394 spin_unlock(&info->lock);
1395 error = -ENOSPC;
1396 goto failed;
1398 sbinfo->free_blocks--;
1399 inode->i_blocks += BLOCKS_PER_PAGE;
1400 spin_unlock(&sbinfo->stat_lock);
1401 } else if (shmem_acct_block(info->flags)) {
1402 spin_unlock(&info->lock);
1403 error = -ENOSPC;
1404 goto failed;
1407 if (!filepage) {
1408 int ret;
1410 spin_unlock(&info->lock);
1411 filepage = shmem_alloc_page(gfp, info, idx);
1412 if (!filepage) {
1413 shmem_unacct_blocks(info->flags, 1);
1414 shmem_free_blocks(inode, 1);
1415 error = -ENOMEM;
1416 goto failed;
1418 SetPageSwapBacked(filepage);
1420 /* Precharge page while we can wait, compensate after */
1421 error = mem_cgroup_cache_charge(filepage, current->mm,
1422 GFP_KERNEL);
1423 if (error) {
1424 page_cache_release(filepage);
1425 shmem_unacct_blocks(info->flags, 1);
1426 shmem_free_blocks(inode, 1);
1427 filepage = NULL;
1428 goto failed;
1431 spin_lock(&info->lock);
1432 entry = shmem_swp_alloc(info, idx, sgp);
1433 if (IS_ERR(entry))
1434 error = PTR_ERR(entry);
1435 else {
1436 swap = *entry;
1437 shmem_swp_unmap(entry);
1439 ret = error || swap.val;
1440 if (ret)
1441 mem_cgroup_uncharge_cache_page(filepage);
1442 else
1443 ret = add_to_page_cache_lru(filepage, mapping,
1444 idx, GFP_NOWAIT);
1446 * At add_to_page_cache_lru() failure, uncharge will
1447 * be done automatically.
1449 if (ret) {
1450 spin_unlock(&info->lock);
1451 page_cache_release(filepage);
1452 shmem_unacct_blocks(info->flags, 1);
1453 shmem_free_blocks(inode, 1);
1454 filepage = NULL;
1455 if (error)
1456 goto failed;
1457 goto repeat;
1459 info->flags |= SHMEM_PAGEIN;
1462 info->alloced++;
1463 spin_unlock(&info->lock);
1464 clear_highpage(filepage);
1465 flush_dcache_page(filepage);
1466 SetPageUptodate(filepage);
1467 if (sgp == SGP_DIRTY)
1468 set_page_dirty(filepage);
1470 done:
1471 *pagep = filepage;
1472 return 0;
1474 failed:
1475 if (*pagep != filepage) {
1476 unlock_page(filepage);
1477 page_cache_release(filepage);
1479 return error;
1482 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1484 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1485 int error;
1486 int ret;
1488 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1489 return VM_FAULT_SIGBUS;
1491 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1492 if (error)
1493 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1495 return ret | VM_FAULT_LOCKED;
1498 #ifdef CONFIG_NUMA
1499 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1501 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1502 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1505 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1506 unsigned long addr)
1508 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1509 unsigned long idx;
1511 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1512 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1514 #endif
1516 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1518 struct inode *inode = file->f_path.dentry->d_inode;
1519 struct shmem_inode_info *info = SHMEM_I(inode);
1520 int retval = -ENOMEM;
1522 spin_lock(&info->lock);
1523 if (lock && !(info->flags & VM_LOCKED)) {
1524 if (!user_shm_lock(inode->i_size, user))
1525 goto out_nomem;
1526 info->flags |= VM_LOCKED;
1527 mapping_set_unevictable(file->f_mapping);
1529 if (!lock && (info->flags & VM_LOCKED) && user) {
1530 user_shm_unlock(inode->i_size, user);
1531 info->flags &= ~VM_LOCKED;
1532 mapping_clear_unevictable(file->f_mapping);
1533 scan_mapping_unevictable_pages(file->f_mapping);
1535 retval = 0;
1537 out_nomem:
1538 spin_unlock(&info->lock);
1539 return retval;
1542 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1544 file_accessed(file);
1545 vma->vm_ops = &shmem_vm_ops;
1546 vma->vm_flags |= VM_CAN_NONLINEAR;
1547 return 0;
1550 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1551 int mode, dev_t dev, unsigned long flags)
1553 struct inode *inode;
1554 struct shmem_inode_info *info;
1555 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1557 if (shmem_reserve_inode(sb))
1558 return NULL;
1560 inode = new_inode(sb);
1561 if (inode) {
1562 inode_init_owner(inode, dir, mode);
1563 inode->i_blocks = 0;
1564 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1565 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1566 inode->i_generation = get_seconds();
1567 info = SHMEM_I(inode);
1568 memset(info, 0, (char *)inode - (char *)info);
1569 spin_lock_init(&info->lock);
1570 info->flags = flags & VM_NORESERVE;
1571 INIT_LIST_HEAD(&info->swaplist);
1572 cache_no_acl(inode);
1574 switch (mode & S_IFMT) {
1575 default:
1576 inode->i_op = &shmem_special_inode_operations;
1577 init_special_inode(inode, mode, dev);
1578 break;
1579 case S_IFREG:
1580 inode->i_mapping->a_ops = &shmem_aops;
1581 inode->i_op = &shmem_inode_operations;
1582 inode->i_fop = &shmem_file_operations;
1583 mpol_shared_policy_init(&info->policy,
1584 shmem_get_sbmpol(sbinfo));
1585 break;
1586 case S_IFDIR:
1587 inc_nlink(inode);
1588 /* Some things misbehave if size == 0 on a directory */
1589 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1590 inode->i_op = &shmem_dir_inode_operations;
1591 inode->i_fop = &simple_dir_operations;
1592 break;
1593 case S_IFLNK:
1595 * Must not load anything in the rbtree,
1596 * mpol_free_shared_policy will not be called.
1598 mpol_shared_policy_init(&info->policy, NULL);
1599 break;
1601 } else
1602 shmem_free_inode(sb);
1603 return inode;
1606 #ifdef CONFIG_TMPFS
1607 static const struct inode_operations shmem_symlink_inode_operations;
1608 static const struct inode_operations shmem_symlink_inline_operations;
1611 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1612 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1613 * below the loop driver, in the generic fashion that many filesystems support.
1615 static int shmem_readpage(struct file *file, struct page *page)
1617 struct inode *inode = page->mapping->host;
1618 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1619 unlock_page(page);
1620 return error;
1623 static int
1624 shmem_write_begin(struct file *file, struct address_space *mapping,
1625 loff_t pos, unsigned len, unsigned flags,
1626 struct page **pagep, void **fsdata)
1628 struct inode *inode = mapping->host;
1629 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1630 *pagep = NULL;
1631 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1634 static int
1635 shmem_write_end(struct file *file, struct address_space *mapping,
1636 loff_t pos, unsigned len, unsigned copied,
1637 struct page *page, void *fsdata)
1639 struct inode *inode = mapping->host;
1641 if (pos + copied > inode->i_size)
1642 i_size_write(inode, pos + copied);
1644 set_page_dirty(page);
1645 unlock_page(page);
1646 page_cache_release(page);
1648 return copied;
1651 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1653 struct inode *inode = filp->f_path.dentry->d_inode;
1654 struct address_space *mapping = inode->i_mapping;
1655 unsigned long index, offset;
1656 enum sgp_type sgp = SGP_READ;
1659 * Might this read be for a stacking filesystem? Then when reading
1660 * holes of a sparse file, we actually need to allocate those pages,
1661 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1663 if (segment_eq(get_fs(), KERNEL_DS))
1664 sgp = SGP_DIRTY;
1666 index = *ppos >> PAGE_CACHE_SHIFT;
1667 offset = *ppos & ~PAGE_CACHE_MASK;
1669 for (;;) {
1670 struct page *page = NULL;
1671 unsigned long end_index, nr, ret;
1672 loff_t i_size = i_size_read(inode);
1674 end_index = i_size >> PAGE_CACHE_SHIFT;
1675 if (index > end_index)
1676 break;
1677 if (index == end_index) {
1678 nr = i_size & ~PAGE_CACHE_MASK;
1679 if (nr <= offset)
1680 break;
1683 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1684 if (desc->error) {
1685 if (desc->error == -EINVAL)
1686 desc->error = 0;
1687 break;
1689 if (page)
1690 unlock_page(page);
1693 * We must evaluate after, since reads (unlike writes)
1694 * are called without i_mutex protection against truncate
1696 nr = PAGE_CACHE_SIZE;
1697 i_size = i_size_read(inode);
1698 end_index = i_size >> PAGE_CACHE_SHIFT;
1699 if (index == end_index) {
1700 nr = i_size & ~PAGE_CACHE_MASK;
1701 if (nr <= offset) {
1702 if (page)
1703 page_cache_release(page);
1704 break;
1707 nr -= offset;
1709 if (page) {
1711 * If users can be writing to this page using arbitrary
1712 * virtual addresses, take care about potential aliasing
1713 * before reading the page on the kernel side.
1715 if (mapping_writably_mapped(mapping))
1716 flush_dcache_page(page);
1718 * Mark the page accessed if we read the beginning.
1720 if (!offset)
1721 mark_page_accessed(page);
1722 } else {
1723 page = ZERO_PAGE(0);
1724 page_cache_get(page);
1728 * Ok, we have the page, and it's up-to-date, so
1729 * now we can copy it to user space...
1731 * The actor routine returns how many bytes were actually used..
1732 * NOTE! This may not be the same as how much of a user buffer
1733 * we filled up (we may be padding etc), so we can only update
1734 * "pos" here (the actor routine has to update the user buffer
1735 * pointers and the remaining count).
1737 ret = actor(desc, page, offset, nr);
1738 offset += ret;
1739 index += offset >> PAGE_CACHE_SHIFT;
1740 offset &= ~PAGE_CACHE_MASK;
1742 page_cache_release(page);
1743 if (ret != nr || !desc->count)
1744 break;
1746 cond_resched();
1749 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1750 file_accessed(filp);
1753 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1754 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1756 struct file *filp = iocb->ki_filp;
1757 ssize_t retval;
1758 unsigned long seg;
1759 size_t count;
1760 loff_t *ppos = &iocb->ki_pos;
1762 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1763 if (retval)
1764 return retval;
1766 for (seg = 0; seg < nr_segs; seg++) {
1767 read_descriptor_t desc;
1769 desc.written = 0;
1770 desc.arg.buf = iov[seg].iov_base;
1771 desc.count = iov[seg].iov_len;
1772 if (desc.count == 0)
1773 continue;
1774 desc.error = 0;
1775 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1776 retval += desc.written;
1777 if (desc.error) {
1778 retval = retval ?: desc.error;
1779 break;
1781 if (desc.count > 0)
1782 break;
1784 return retval;
1787 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1789 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1791 buf->f_type = TMPFS_MAGIC;
1792 buf->f_bsize = PAGE_CACHE_SIZE;
1793 buf->f_namelen = NAME_MAX;
1794 spin_lock(&sbinfo->stat_lock);
1795 if (sbinfo->max_blocks) {
1796 buf->f_blocks = sbinfo->max_blocks;
1797 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1799 if (sbinfo->max_inodes) {
1800 buf->f_files = sbinfo->max_inodes;
1801 buf->f_ffree = sbinfo->free_inodes;
1803 /* else leave those fields 0 like simple_statfs */
1804 spin_unlock(&sbinfo->stat_lock);
1805 return 0;
1809 * File creation. Allocate an inode, and we're done..
1811 static int
1812 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1814 struct inode *inode;
1815 int error = -ENOSPC;
1817 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1818 if (inode) {
1819 error = security_inode_init_security(inode, dir, NULL, NULL,
1820 NULL);
1821 if (error) {
1822 if (error != -EOPNOTSUPP) {
1823 iput(inode);
1824 return error;
1827 #ifdef CONFIG_TMPFS_POSIX_ACL
1828 error = generic_acl_init(inode, dir);
1829 if (error) {
1830 iput(inode);
1831 return error;
1833 #else
1834 error = 0;
1835 #endif
1836 dir->i_size += BOGO_DIRENT_SIZE;
1837 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1838 d_instantiate(dentry, inode);
1839 dget(dentry); /* Extra count - pin the dentry in core */
1841 return error;
1844 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1846 int error;
1848 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1849 return error;
1850 inc_nlink(dir);
1851 return 0;
1854 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1855 struct nameidata *nd)
1857 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1861 * Link a file..
1863 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1865 struct inode *inode = old_dentry->d_inode;
1866 int ret;
1869 * No ordinary (disk based) filesystem counts links as inodes;
1870 * but each new link needs a new dentry, pinning lowmem, and
1871 * tmpfs dentries cannot be pruned until they are unlinked.
1873 ret = shmem_reserve_inode(inode->i_sb);
1874 if (ret)
1875 goto out;
1877 dir->i_size += BOGO_DIRENT_SIZE;
1878 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1879 inc_nlink(inode);
1880 atomic_inc(&inode->i_count); /* New dentry reference */
1881 dget(dentry); /* Extra pinning count for the created dentry */
1882 d_instantiate(dentry, inode);
1883 out:
1884 return ret;
1887 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1889 struct inode *inode = dentry->d_inode;
1891 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1892 shmem_free_inode(inode->i_sb);
1894 dir->i_size -= BOGO_DIRENT_SIZE;
1895 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1896 drop_nlink(inode);
1897 dput(dentry); /* Undo the count from "create" - this does all the work */
1898 return 0;
1901 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1903 if (!simple_empty(dentry))
1904 return -ENOTEMPTY;
1906 drop_nlink(dentry->d_inode);
1907 drop_nlink(dir);
1908 return shmem_unlink(dir, dentry);
1912 * The VFS layer already does all the dentry stuff for rename,
1913 * we just have to decrement the usage count for the target if
1914 * it exists so that the VFS layer correctly free's it when it
1915 * gets overwritten.
1917 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1919 struct inode *inode = old_dentry->d_inode;
1920 int they_are_dirs = S_ISDIR(inode->i_mode);
1922 if (!simple_empty(new_dentry))
1923 return -ENOTEMPTY;
1925 if (new_dentry->d_inode) {
1926 (void) shmem_unlink(new_dir, new_dentry);
1927 if (they_are_dirs)
1928 drop_nlink(old_dir);
1929 } else if (they_are_dirs) {
1930 drop_nlink(old_dir);
1931 inc_nlink(new_dir);
1934 old_dir->i_size -= BOGO_DIRENT_SIZE;
1935 new_dir->i_size += BOGO_DIRENT_SIZE;
1936 old_dir->i_ctime = old_dir->i_mtime =
1937 new_dir->i_ctime = new_dir->i_mtime =
1938 inode->i_ctime = CURRENT_TIME;
1939 return 0;
1942 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1944 int error;
1945 int len;
1946 struct inode *inode;
1947 struct page *page = NULL;
1948 char *kaddr;
1949 struct shmem_inode_info *info;
1951 len = strlen(symname) + 1;
1952 if (len > PAGE_CACHE_SIZE)
1953 return -ENAMETOOLONG;
1955 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1956 if (!inode)
1957 return -ENOSPC;
1959 error = security_inode_init_security(inode, dir, NULL, NULL,
1960 NULL);
1961 if (error) {
1962 if (error != -EOPNOTSUPP) {
1963 iput(inode);
1964 return error;
1966 error = 0;
1969 info = SHMEM_I(inode);
1970 inode->i_size = len-1;
1971 if (len <= (char *)inode - (char *)info) {
1972 /* do it inline */
1973 memcpy(info, symname, len);
1974 inode->i_op = &shmem_symlink_inline_operations;
1975 } else {
1976 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1977 if (error) {
1978 iput(inode);
1979 return error;
1981 inode->i_mapping->a_ops = &shmem_aops;
1982 inode->i_op = &shmem_symlink_inode_operations;
1983 kaddr = kmap_atomic(page, KM_USER0);
1984 memcpy(kaddr, symname, len);
1985 kunmap_atomic(kaddr, KM_USER0);
1986 set_page_dirty(page);
1987 unlock_page(page);
1988 page_cache_release(page);
1990 dir->i_size += BOGO_DIRENT_SIZE;
1991 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1992 d_instantiate(dentry, inode);
1993 dget(dentry);
1994 return 0;
1997 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1999 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
2000 return NULL;
2003 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2005 struct page *page = NULL;
2006 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2007 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
2008 if (page)
2009 unlock_page(page);
2010 return page;
2013 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2015 if (!IS_ERR(nd_get_link(nd))) {
2016 struct page *page = cookie;
2017 kunmap(page);
2018 mark_page_accessed(page);
2019 page_cache_release(page);
2023 static const struct inode_operations shmem_symlink_inline_operations = {
2024 .readlink = generic_readlink,
2025 .follow_link = shmem_follow_link_inline,
2028 static const struct inode_operations shmem_symlink_inode_operations = {
2029 .readlink = generic_readlink,
2030 .follow_link = shmem_follow_link,
2031 .put_link = shmem_put_link,
2034 #ifdef CONFIG_TMPFS_POSIX_ACL
2036 * Superblocks without xattr inode operations will get security.* xattr
2037 * support from the VFS "for free". As soon as we have any other xattrs
2038 * like ACLs, we also need to implement the security.* handlers at
2039 * filesystem level, though.
2042 static size_t shmem_xattr_security_list(struct dentry *dentry, char *list,
2043 size_t list_len, const char *name,
2044 size_t name_len, int handler_flags)
2046 return security_inode_listsecurity(dentry->d_inode, list, list_len);
2049 static int shmem_xattr_security_get(struct dentry *dentry, const char *name,
2050 void *buffer, size_t size, int handler_flags)
2052 if (strcmp(name, "") == 0)
2053 return -EINVAL;
2054 return xattr_getsecurity(dentry->d_inode, name, buffer, size);
2057 static int shmem_xattr_security_set(struct dentry *dentry, const char *name,
2058 const void *value, size_t size, int flags, int handler_flags)
2060 if (strcmp(name, "") == 0)
2061 return -EINVAL;
2062 return security_inode_setsecurity(dentry->d_inode, name, value,
2063 size, flags);
2066 static const struct xattr_handler shmem_xattr_security_handler = {
2067 .prefix = XATTR_SECURITY_PREFIX,
2068 .list = shmem_xattr_security_list,
2069 .get = shmem_xattr_security_get,
2070 .set = shmem_xattr_security_set,
2073 static const struct xattr_handler *shmem_xattr_handlers[] = {
2074 &generic_acl_access_handler,
2075 &generic_acl_default_handler,
2076 &shmem_xattr_security_handler,
2077 NULL
2079 #endif
2081 static struct dentry *shmem_get_parent(struct dentry *child)
2083 return ERR_PTR(-ESTALE);
2086 static int shmem_match(struct inode *ino, void *vfh)
2088 __u32 *fh = vfh;
2089 __u64 inum = fh[2];
2090 inum = (inum << 32) | fh[1];
2091 return ino->i_ino == inum && fh[0] == ino->i_generation;
2094 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2095 struct fid *fid, int fh_len, int fh_type)
2097 struct inode *inode;
2098 struct dentry *dentry = NULL;
2099 u64 inum = fid->raw[2];
2100 inum = (inum << 32) | fid->raw[1];
2102 if (fh_len < 3)
2103 return NULL;
2105 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2106 shmem_match, fid->raw);
2107 if (inode) {
2108 dentry = d_find_alias(inode);
2109 iput(inode);
2112 return dentry;
2115 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2116 int connectable)
2118 struct inode *inode = dentry->d_inode;
2120 if (*len < 3)
2121 return 255;
2123 if (hlist_unhashed(&inode->i_hash)) {
2124 /* Unfortunately insert_inode_hash is not idempotent,
2125 * so as we hash inodes here rather than at creation
2126 * time, we need a lock to ensure we only try
2127 * to do it once
2129 static DEFINE_SPINLOCK(lock);
2130 spin_lock(&lock);
2131 if (hlist_unhashed(&inode->i_hash))
2132 __insert_inode_hash(inode,
2133 inode->i_ino + inode->i_generation);
2134 spin_unlock(&lock);
2137 fh[0] = inode->i_generation;
2138 fh[1] = inode->i_ino;
2139 fh[2] = ((__u64)inode->i_ino) >> 32;
2141 *len = 3;
2142 return 1;
2145 static const struct export_operations shmem_export_ops = {
2146 .get_parent = shmem_get_parent,
2147 .encode_fh = shmem_encode_fh,
2148 .fh_to_dentry = shmem_fh_to_dentry,
2151 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2152 bool remount)
2154 char *this_char, *value, *rest;
2156 while (options != NULL) {
2157 this_char = options;
2158 for (;;) {
2160 * NUL-terminate this option: unfortunately,
2161 * mount options form a comma-separated list,
2162 * but mpol's nodelist may also contain commas.
2164 options = strchr(options, ',');
2165 if (options == NULL)
2166 break;
2167 options++;
2168 if (!isdigit(*options)) {
2169 options[-1] = '\0';
2170 break;
2173 if (!*this_char)
2174 continue;
2175 if ((value = strchr(this_char,'=')) != NULL) {
2176 *value++ = 0;
2177 } else {
2178 printk(KERN_ERR
2179 "tmpfs: No value for mount option '%s'\n",
2180 this_char);
2181 return 1;
2184 if (!strcmp(this_char,"size")) {
2185 unsigned long long size;
2186 size = memparse(value,&rest);
2187 if (*rest == '%') {
2188 size <<= PAGE_SHIFT;
2189 size *= totalram_pages;
2190 do_div(size, 100);
2191 rest++;
2193 if (*rest)
2194 goto bad_val;
2195 sbinfo->max_blocks =
2196 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2197 } else if (!strcmp(this_char,"nr_blocks")) {
2198 sbinfo->max_blocks = memparse(value, &rest);
2199 if (*rest)
2200 goto bad_val;
2201 } else if (!strcmp(this_char,"nr_inodes")) {
2202 sbinfo->max_inodes = memparse(value, &rest);
2203 if (*rest)
2204 goto bad_val;
2205 } else if (!strcmp(this_char,"mode")) {
2206 if (remount)
2207 continue;
2208 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2209 if (*rest)
2210 goto bad_val;
2211 } else if (!strcmp(this_char,"uid")) {
2212 if (remount)
2213 continue;
2214 sbinfo->uid = simple_strtoul(value, &rest, 0);
2215 if (*rest)
2216 goto bad_val;
2217 } else if (!strcmp(this_char,"gid")) {
2218 if (remount)
2219 continue;
2220 sbinfo->gid = simple_strtoul(value, &rest, 0);
2221 if (*rest)
2222 goto bad_val;
2223 } else if (!strcmp(this_char,"mpol")) {
2224 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2225 goto bad_val;
2226 } else {
2227 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2228 this_char);
2229 return 1;
2232 return 0;
2234 bad_val:
2235 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2236 value, this_char);
2237 return 1;
2241 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2243 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2244 struct shmem_sb_info config = *sbinfo;
2245 unsigned long blocks;
2246 unsigned long inodes;
2247 int error = -EINVAL;
2249 if (shmem_parse_options(data, &config, true))
2250 return error;
2252 spin_lock(&sbinfo->stat_lock);
2253 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2254 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2255 if (config.max_blocks < blocks)
2256 goto out;
2257 if (config.max_inodes < inodes)
2258 goto out;
2260 * Those tests also disallow limited->unlimited while any are in
2261 * use, so i_blocks will always be zero when max_blocks is zero;
2262 * but we must separately disallow unlimited->limited, because
2263 * in that case we have no record of how much is already in use.
2265 if (config.max_blocks && !sbinfo->max_blocks)
2266 goto out;
2267 if (config.max_inodes && !sbinfo->max_inodes)
2268 goto out;
2270 error = 0;
2271 sbinfo->max_blocks = config.max_blocks;
2272 sbinfo->free_blocks = config.max_blocks - blocks;
2273 sbinfo->max_inodes = config.max_inodes;
2274 sbinfo->free_inodes = config.max_inodes - inodes;
2276 mpol_put(sbinfo->mpol);
2277 sbinfo->mpol = config.mpol; /* transfers initial ref */
2278 out:
2279 spin_unlock(&sbinfo->stat_lock);
2280 return error;
2283 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2285 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2287 if (sbinfo->max_blocks != shmem_default_max_blocks())
2288 seq_printf(seq, ",size=%luk",
2289 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2290 if (sbinfo->max_inodes != shmem_default_max_inodes())
2291 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2292 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2293 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2294 if (sbinfo->uid != 0)
2295 seq_printf(seq, ",uid=%u", sbinfo->uid);
2296 if (sbinfo->gid != 0)
2297 seq_printf(seq, ",gid=%u", sbinfo->gid);
2298 shmem_show_mpol(seq, sbinfo->mpol);
2299 return 0;
2301 #endif /* CONFIG_TMPFS */
2303 static void shmem_put_super(struct super_block *sb)
2305 kfree(sb->s_fs_info);
2306 sb->s_fs_info = NULL;
2309 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2311 struct inode *inode;
2312 struct dentry *root;
2313 struct shmem_sb_info *sbinfo;
2314 int err = -ENOMEM;
2316 /* Round up to L1_CACHE_BYTES to resist false sharing */
2317 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2318 L1_CACHE_BYTES), GFP_KERNEL);
2319 if (!sbinfo)
2320 return -ENOMEM;
2322 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2323 sbinfo->uid = current_fsuid();
2324 sbinfo->gid = current_fsgid();
2325 sb->s_fs_info = sbinfo;
2327 #ifdef CONFIG_TMPFS
2329 * Per default we only allow half of the physical ram per
2330 * tmpfs instance, limiting inodes to one per page of lowmem;
2331 * but the internal instance is left unlimited.
2333 if (!(sb->s_flags & MS_NOUSER)) {
2334 sbinfo->max_blocks = shmem_default_max_blocks();
2335 sbinfo->max_inodes = shmem_default_max_inodes();
2336 if (shmem_parse_options(data, sbinfo, false)) {
2337 err = -EINVAL;
2338 goto failed;
2341 sb->s_export_op = &shmem_export_ops;
2342 #else
2343 sb->s_flags |= MS_NOUSER;
2344 #endif
2346 spin_lock_init(&sbinfo->stat_lock);
2347 sbinfo->free_blocks = sbinfo->max_blocks;
2348 sbinfo->free_inodes = sbinfo->max_inodes;
2350 sb->s_maxbytes = SHMEM_MAX_BYTES;
2351 sb->s_blocksize = PAGE_CACHE_SIZE;
2352 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2353 sb->s_magic = TMPFS_MAGIC;
2354 sb->s_op = &shmem_ops;
2355 sb->s_time_gran = 1;
2356 #ifdef CONFIG_TMPFS_POSIX_ACL
2357 sb->s_xattr = shmem_xattr_handlers;
2358 sb->s_flags |= MS_POSIXACL;
2359 #endif
2361 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2362 if (!inode)
2363 goto failed;
2364 inode->i_uid = sbinfo->uid;
2365 inode->i_gid = sbinfo->gid;
2366 root = d_alloc_root(inode);
2367 if (!root)
2368 goto failed_iput;
2369 sb->s_root = root;
2370 return 0;
2372 failed_iput:
2373 iput(inode);
2374 failed:
2375 shmem_put_super(sb);
2376 return err;
2379 static struct kmem_cache *shmem_inode_cachep;
2381 static struct inode *shmem_alloc_inode(struct super_block *sb)
2383 struct shmem_inode_info *p;
2384 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2385 if (!p)
2386 return NULL;
2387 return &p->vfs_inode;
2390 static void shmem_destroy_inode(struct inode *inode)
2392 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2393 /* only struct inode is valid if it's an inline symlink */
2394 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2396 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2399 static void init_once(void *foo)
2401 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2403 inode_init_once(&p->vfs_inode);
2406 static int init_inodecache(void)
2408 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2409 sizeof(struct shmem_inode_info),
2410 0, SLAB_PANIC, init_once);
2411 return 0;
2414 static void destroy_inodecache(void)
2416 kmem_cache_destroy(shmem_inode_cachep);
2419 static const struct address_space_operations shmem_aops = {
2420 .writepage = shmem_writepage,
2421 .set_page_dirty = __set_page_dirty_no_writeback,
2422 #ifdef CONFIG_TMPFS
2423 .readpage = shmem_readpage,
2424 .write_begin = shmem_write_begin,
2425 .write_end = shmem_write_end,
2426 #endif
2427 .migratepage = migrate_page,
2428 .error_remove_page = generic_error_remove_page,
2431 static const struct file_operations shmem_file_operations = {
2432 .mmap = shmem_mmap,
2433 #ifdef CONFIG_TMPFS
2434 .llseek = generic_file_llseek,
2435 .read = do_sync_read,
2436 .write = do_sync_write,
2437 .aio_read = shmem_file_aio_read,
2438 .aio_write = generic_file_aio_write,
2439 .fsync = noop_fsync,
2440 .splice_read = generic_file_splice_read,
2441 .splice_write = generic_file_splice_write,
2442 #endif
2445 static const struct inode_operations shmem_inode_operations = {
2446 .setattr = shmem_notify_change,
2447 .truncate_range = shmem_truncate_range,
2448 #ifdef CONFIG_TMPFS_POSIX_ACL
2449 .setxattr = generic_setxattr,
2450 .getxattr = generic_getxattr,
2451 .listxattr = generic_listxattr,
2452 .removexattr = generic_removexattr,
2453 .check_acl = generic_check_acl,
2454 #endif
2458 static const struct inode_operations shmem_dir_inode_operations = {
2459 #ifdef CONFIG_TMPFS
2460 .create = shmem_create,
2461 .lookup = simple_lookup,
2462 .link = shmem_link,
2463 .unlink = shmem_unlink,
2464 .symlink = shmem_symlink,
2465 .mkdir = shmem_mkdir,
2466 .rmdir = shmem_rmdir,
2467 .mknod = shmem_mknod,
2468 .rename = shmem_rename,
2469 #endif
2470 #ifdef CONFIG_TMPFS_POSIX_ACL
2471 .setattr = shmem_notify_change,
2472 .setxattr = generic_setxattr,
2473 .getxattr = generic_getxattr,
2474 .listxattr = generic_listxattr,
2475 .removexattr = generic_removexattr,
2476 .check_acl = generic_check_acl,
2477 #endif
2480 static const struct inode_operations shmem_special_inode_operations = {
2481 #ifdef CONFIG_TMPFS_POSIX_ACL
2482 .setattr = shmem_notify_change,
2483 .setxattr = generic_setxattr,
2484 .getxattr = generic_getxattr,
2485 .listxattr = generic_listxattr,
2486 .removexattr = generic_removexattr,
2487 .check_acl = generic_check_acl,
2488 #endif
2491 static const struct super_operations shmem_ops = {
2492 .alloc_inode = shmem_alloc_inode,
2493 .destroy_inode = shmem_destroy_inode,
2494 #ifdef CONFIG_TMPFS
2495 .statfs = shmem_statfs,
2496 .remount_fs = shmem_remount_fs,
2497 .show_options = shmem_show_options,
2498 #endif
2499 .delete_inode = shmem_delete_inode,
2500 .drop_inode = generic_delete_inode,
2501 .put_super = shmem_put_super,
2504 static const struct vm_operations_struct shmem_vm_ops = {
2505 .fault = shmem_fault,
2506 #ifdef CONFIG_NUMA
2507 .set_policy = shmem_set_policy,
2508 .get_policy = shmem_get_policy,
2509 #endif
2513 static int shmem_get_sb(struct file_system_type *fs_type,
2514 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2516 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2519 static struct file_system_type tmpfs_fs_type = {
2520 .owner = THIS_MODULE,
2521 .name = "tmpfs",
2522 .get_sb = shmem_get_sb,
2523 .kill_sb = kill_litter_super,
2526 int __init init_tmpfs(void)
2528 int error;
2530 error = bdi_init(&shmem_backing_dev_info);
2531 if (error)
2532 goto out4;
2534 error = init_inodecache();
2535 if (error)
2536 goto out3;
2538 error = register_filesystem(&tmpfs_fs_type);
2539 if (error) {
2540 printk(KERN_ERR "Could not register tmpfs\n");
2541 goto out2;
2544 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2545 tmpfs_fs_type.name, NULL);
2546 if (IS_ERR(shm_mnt)) {
2547 error = PTR_ERR(shm_mnt);
2548 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2549 goto out1;
2551 return 0;
2553 out1:
2554 unregister_filesystem(&tmpfs_fs_type);
2555 out2:
2556 destroy_inodecache();
2557 out3:
2558 bdi_destroy(&shmem_backing_dev_info);
2559 out4:
2560 shm_mnt = ERR_PTR(error);
2561 return error;
2564 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2566 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2567 * @inode: the inode to be searched
2568 * @pgoff: the offset to be searched
2569 * @pagep: the pointer for the found page to be stored
2570 * @ent: the pointer for the found swap entry to be stored
2572 * If a page is found, refcount of it is incremented. Callers should handle
2573 * these refcount.
2575 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2576 struct page **pagep, swp_entry_t *ent)
2578 swp_entry_t entry = { .val = 0 }, *ptr;
2579 struct page *page = NULL;
2580 struct shmem_inode_info *info = SHMEM_I(inode);
2582 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2583 goto out;
2585 spin_lock(&info->lock);
2586 ptr = shmem_swp_entry(info, pgoff, NULL);
2587 #ifdef CONFIG_SWAP
2588 if (ptr && ptr->val) {
2589 entry.val = ptr->val;
2590 page = find_get_page(&swapper_space, entry.val);
2591 } else
2592 #endif
2593 page = find_get_page(inode->i_mapping, pgoff);
2594 if (ptr)
2595 shmem_swp_unmap(ptr);
2596 spin_unlock(&info->lock);
2597 out:
2598 *pagep = page;
2599 *ent = entry;
2601 #endif
2603 #else /* !CONFIG_SHMEM */
2606 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2608 * This is intended for small system where the benefits of the full
2609 * shmem code (swap-backed and resource-limited) are outweighed by
2610 * their complexity. On systems without swap this code should be
2611 * effectively equivalent, but much lighter weight.
2614 #include <linux/ramfs.h>
2616 static struct file_system_type tmpfs_fs_type = {
2617 .name = "tmpfs",
2618 .get_sb = ramfs_get_sb,
2619 .kill_sb = kill_litter_super,
2622 int __init init_tmpfs(void)
2624 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2626 shm_mnt = kern_mount(&tmpfs_fs_type);
2627 BUG_ON(IS_ERR(shm_mnt));
2629 return 0;
2632 int shmem_unuse(swp_entry_t entry, struct page *page)
2634 return 0;
2637 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2639 return 0;
2642 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2644 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2645 * @inode: the inode to be searched
2646 * @pgoff: the offset to be searched
2647 * @pagep: the pointer for the found page to be stored
2648 * @ent: the pointer for the found swap entry to be stored
2650 * If a page is found, refcount of it is incremented. Callers should handle
2651 * these refcount.
2653 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2654 struct page **pagep, swp_entry_t *ent)
2656 struct page *page = NULL;
2658 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2659 goto out;
2660 page = find_get_page(inode->i_mapping, pgoff);
2661 out:
2662 *pagep = page;
2663 *ent = (swp_entry_t){ .val = 0 };
2665 #endif
2667 #define shmem_vm_ops generic_file_vm_ops
2668 #define shmem_file_operations ramfs_file_operations
2669 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2670 #define shmem_acct_size(flags, size) 0
2671 #define shmem_unacct_size(flags, size) do {} while (0)
2672 #define SHMEM_MAX_BYTES MAX_LFS_FILESIZE
2674 #endif /* CONFIG_SHMEM */
2676 /* common code */
2679 * shmem_file_setup - get an unlinked file living in tmpfs
2680 * @name: name for dentry (to be seen in /proc/<pid>/maps
2681 * @size: size to be set for the file
2682 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2684 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2686 int error;
2687 struct file *file;
2688 struct inode *inode;
2689 struct path path;
2690 struct dentry *root;
2691 struct qstr this;
2693 if (IS_ERR(shm_mnt))
2694 return (void *)shm_mnt;
2696 if (size < 0 || size > SHMEM_MAX_BYTES)
2697 return ERR_PTR(-EINVAL);
2699 if (shmem_acct_size(flags, size))
2700 return ERR_PTR(-ENOMEM);
2702 error = -ENOMEM;
2703 this.name = name;
2704 this.len = strlen(name);
2705 this.hash = 0; /* will go */
2706 root = shm_mnt->mnt_root;
2707 path.dentry = d_alloc(root, &this);
2708 if (!path.dentry)
2709 goto put_memory;
2710 path.mnt = mntget(shm_mnt);
2712 error = -ENOSPC;
2713 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2714 if (!inode)
2715 goto put_dentry;
2717 d_instantiate(path.dentry, inode);
2718 inode->i_size = size;
2719 inode->i_nlink = 0; /* It is unlinked */
2720 #ifndef CONFIG_MMU
2721 error = ramfs_nommu_expand_for_mapping(inode, size);
2722 if (error)
2723 goto put_dentry;
2724 #endif
2726 error = -ENFILE;
2727 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2728 &shmem_file_operations);
2729 if (!file)
2730 goto put_dentry;
2732 return file;
2734 put_dentry:
2735 path_put(&path);
2736 put_memory:
2737 shmem_unacct_size(flags, size);
2738 return ERR_PTR(error);
2740 EXPORT_SYMBOL_GPL(shmem_file_setup);
2743 * shmem_zero_setup - setup a shared anonymous mapping
2744 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2746 int shmem_zero_setup(struct vm_area_struct *vma)
2748 struct file *file;
2749 loff_t size = vma->vm_end - vma->vm_start;
2751 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2752 if (IS_ERR(file))
2753 return PTR_ERR(file);
2755 if (vma->vm_file)
2756 fput(vma->vm_file);
2757 vma->vm_file = file;
2758 vma->vm_ops = &shmem_vm_ops;
2759 return 0;