Merge branch 'fix/hda' into for-linus
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / mm / shmem.c
blobeef4ebea51581f9aa8a6d2244f8ce5fa57f99c22
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 if (page)
437 set_page_private(page, 0);
438 spin_lock(&info->lock);
440 if (!page) {
441 shmem_free_blocks(inode, 1);
442 return ERR_PTR(-ENOMEM);
444 if (sgp != SGP_WRITE &&
445 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
446 entry = ERR_PTR(-EINVAL);
447 break;
449 if (info->next_index <= index)
450 info->next_index = index + 1;
452 if (page) {
453 /* another task gave its page, or truncated the file */
454 shmem_free_blocks(inode, 1);
455 shmem_dir_free(page);
457 if (info->next_index <= index && !IS_ERR(entry))
458 info->next_index = index + 1;
459 return entry;
463 * shmem_free_swp - free some swap entries in a directory
464 * @dir: pointer to the directory
465 * @edir: pointer after last entry of the directory
466 * @punch_lock: pointer to spinlock when needed for the holepunch case
468 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
469 spinlock_t *punch_lock)
471 spinlock_t *punch_unlock = NULL;
472 swp_entry_t *ptr;
473 int freed = 0;
475 for (ptr = dir; ptr < edir; ptr++) {
476 if (ptr->val) {
477 if (unlikely(punch_lock)) {
478 punch_unlock = punch_lock;
479 punch_lock = NULL;
480 spin_lock(punch_unlock);
481 if (!ptr->val)
482 continue;
484 free_swap_and_cache(*ptr);
485 *ptr = (swp_entry_t){0};
486 freed++;
489 if (punch_unlock)
490 spin_unlock(punch_unlock);
491 return freed;
494 static int shmem_map_and_free_swp(struct page *subdir, int offset,
495 int limit, struct page ***dir, spinlock_t *punch_lock)
497 swp_entry_t *ptr;
498 int freed = 0;
500 ptr = shmem_swp_map(subdir);
501 for (; offset < limit; offset += LATENCY_LIMIT) {
502 int size = limit - offset;
503 if (size > LATENCY_LIMIT)
504 size = LATENCY_LIMIT;
505 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
506 punch_lock);
507 if (need_resched()) {
508 shmem_swp_unmap(ptr);
509 if (*dir) {
510 shmem_dir_unmap(*dir);
511 *dir = NULL;
513 cond_resched();
514 ptr = shmem_swp_map(subdir);
517 shmem_swp_unmap(ptr);
518 return freed;
521 static void shmem_free_pages(struct list_head *next)
523 struct page *page;
524 int freed = 0;
526 do {
527 page = container_of(next, struct page, lru);
528 next = next->next;
529 shmem_dir_free(page);
530 freed++;
531 if (freed >= LATENCY_LIMIT) {
532 cond_resched();
533 freed = 0;
535 } while (next);
538 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
540 struct shmem_inode_info *info = SHMEM_I(inode);
541 unsigned long idx;
542 unsigned long size;
543 unsigned long limit;
544 unsigned long stage;
545 unsigned long diroff;
546 struct page **dir;
547 struct page *topdir;
548 struct page *middir;
549 struct page *subdir;
550 swp_entry_t *ptr;
551 LIST_HEAD(pages_to_free);
552 long nr_pages_to_free = 0;
553 long nr_swaps_freed = 0;
554 int offset;
555 int freed;
556 int punch_hole;
557 spinlock_t *needs_lock;
558 spinlock_t *punch_lock;
559 unsigned long upper_limit;
561 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
562 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
563 if (idx >= info->next_index)
564 return;
566 spin_lock(&info->lock);
567 info->flags |= SHMEM_TRUNCATE;
568 if (likely(end == (loff_t) -1)) {
569 limit = info->next_index;
570 upper_limit = SHMEM_MAX_INDEX;
571 info->next_index = idx;
572 needs_lock = NULL;
573 punch_hole = 0;
574 } else {
575 if (end + 1 >= inode->i_size) { /* we may free a little more */
576 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
577 PAGE_CACHE_SHIFT;
578 upper_limit = SHMEM_MAX_INDEX;
579 } else {
580 limit = (end + 1) >> PAGE_CACHE_SHIFT;
581 upper_limit = limit;
583 needs_lock = &info->lock;
584 punch_hole = 1;
587 topdir = info->i_indirect;
588 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
589 info->i_indirect = NULL;
590 nr_pages_to_free++;
591 list_add(&topdir->lru, &pages_to_free);
593 spin_unlock(&info->lock);
595 if (info->swapped && idx < SHMEM_NR_DIRECT) {
596 ptr = info->i_direct;
597 size = limit;
598 if (size > SHMEM_NR_DIRECT)
599 size = SHMEM_NR_DIRECT;
600 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
604 * If there are no indirect blocks or we are punching a hole
605 * below indirect blocks, nothing to be done.
607 if (!topdir || limit <= SHMEM_NR_DIRECT)
608 goto done2;
611 * The truncation case has already dropped info->lock, and we're safe
612 * because i_size and next_index have already been lowered, preventing
613 * access beyond. But in the punch_hole case, we still need to take
614 * the lock when updating the swap directory, because there might be
615 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
616 * shmem_writepage. However, whenever we find we can remove a whole
617 * directory page (not at the misaligned start or end of the range),
618 * we first NULLify its pointer in the level above, and then have no
619 * need to take the lock when updating its contents: needs_lock and
620 * punch_lock (either pointing to info->lock or NULL) manage this.
623 upper_limit -= SHMEM_NR_DIRECT;
624 limit -= SHMEM_NR_DIRECT;
625 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
626 offset = idx % ENTRIES_PER_PAGE;
627 idx -= offset;
629 dir = shmem_dir_map(topdir);
630 stage = ENTRIES_PER_PAGEPAGE/2;
631 if (idx < ENTRIES_PER_PAGEPAGE/2) {
632 middir = topdir;
633 diroff = idx/ENTRIES_PER_PAGE;
634 } else {
635 dir += ENTRIES_PER_PAGE/2;
636 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
637 while (stage <= idx)
638 stage += ENTRIES_PER_PAGEPAGE;
639 middir = *dir;
640 if (*dir) {
641 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
642 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
643 if (!diroff && !offset && upper_limit >= stage) {
644 if (needs_lock) {
645 spin_lock(needs_lock);
646 *dir = NULL;
647 spin_unlock(needs_lock);
648 needs_lock = NULL;
649 } else
650 *dir = NULL;
651 nr_pages_to_free++;
652 list_add(&middir->lru, &pages_to_free);
654 shmem_dir_unmap(dir);
655 dir = shmem_dir_map(middir);
656 } else {
657 diroff = 0;
658 offset = 0;
659 idx = stage;
663 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
664 if (unlikely(idx == stage)) {
665 shmem_dir_unmap(dir);
666 dir = shmem_dir_map(topdir) +
667 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
668 while (!*dir) {
669 dir++;
670 idx += ENTRIES_PER_PAGEPAGE;
671 if (idx >= limit)
672 goto done1;
674 stage = idx + ENTRIES_PER_PAGEPAGE;
675 middir = *dir;
676 if (punch_hole)
677 needs_lock = &info->lock;
678 if (upper_limit >= stage) {
679 if (needs_lock) {
680 spin_lock(needs_lock);
681 *dir = NULL;
682 spin_unlock(needs_lock);
683 needs_lock = NULL;
684 } else
685 *dir = NULL;
686 nr_pages_to_free++;
687 list_add(&middir->lru, &pages_to_free);
689 shmem_dir_unmap(dir);
690 cond_resched();
691 dir = shmem_dir_map(middir);
692 diroff = 0;
694 punch_lock = needs_lock;
695 subdir = dir[diroff];
696 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
697 if (needs_lock) {
698 spin_lock(needs_lock);
699 dir[diroff] = NULL;
700 spin_unlock(needs_lock);
701 punch_lock = NULL;
702 } else
703 dir[diroff] = NULL;
704 nr_pages_to_free++;
705 list_add(&subdir->lru, &pages_to_free);
707 if (subdir && page_private(subdir) /* has swap entries */) {
708 size = limit - idx;
709 if (size > ENTRIES_PER_PAGE)
710 size = ENTRIES_PER_PAGE;
711 freed = shmem_map_and_free_swp(subdir,
712 offset, size, &dir, punch_lock);
713 if (!dir)
714 dir = shmem_dir_map(middir);
715 nr_swaps_freed += freed;
716 if (offset || punch_lock) {
717 spin_lock(&info->lock);
718 set_page_private(subdir,
719 page_private(subdir) - freed);
720 spin_unlock(&info->lock);
721 } else
722 BUG_ON(page_private(subdir) != freed);
724 offset = 0;
726 done1:
727 shmem_dir_unmap(dir);
728 done2:
729 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
731 * Call truncate_inode_pages again: racing shmem_unuse_inode
732 * may have swizzled a page in from swap since vmtruncate or
733 * generic_delete_inode did it, before we lowered next_index.
734 * Also, though shmem_getpage checks i_size before adding to
735 * cache, no recheck after: so fix the narrow window there too.
737 * Recalling truncate_inode_pages_range and unmap_mapping_range
738 * every time for punch_hole (which never got a chance to clear
739 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
740 * yet hardly ever necessary: try to optimize them out later.
742 truncate_inode_pages_range(inode->i_mapping, start, end);
743 if (punch_hole)
744 unmap_mapping_range(inode->i_mapping, start,
745 end - start, 1);
748 spin_lock(&info->lock);
749 info->flags &= ~SHMEM_TRUNCATE;
750 info->swapped -= nr_swaps_freed;
751 if (nr_pages_to_free)
752 shmem_free_blocks(inode, nr_pages_to_free);
753 shmem_recalc_inode(inode);
754 spin_unlock(&info->lock);
757 * Empty swap vector directory pages to be freed?
759 if (!list_empty(&pages_to_free)) {
760 pages_to_free.prev->next = NULL;
761 shmem_free_pages(pages_to_free.next);
765 static void shmem_truncate(struct inode *inode)
767 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
770 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
772 struct inode *inode = dentry->d_inode;
773 struct page *page = NULL;
774 int error;
776 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
777 if (attr->ia_size < inode->i_size) {
779 * If truncating down to a partial page, then
780 * if that page is already allocated, hold it
781 * in memory until the truncation is over, so
782 * truncate_partial_page cannnot miss it were
783 * it assigned to swap.
785 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
786 (void) shmem_getpage(inode,
787 attr->ia_size>>PAGE_CACHE_SHIFT,
788 &page, SGP_READ, NULL);
789 if (page)
790 unlock_page(page);
793 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
794 * detect if any pages might have been added to cache
795 * after truncate_inode_pages. But we needn't bother
796 * if it's being fully truncated to zero-length: the
797 * nrpages check is efficient enough in that case.
799 if (attr->ia_size) {
800 struct shmem_inode_info *info = SHMEM_I(inode);
801 spin_lock(&info->lock);
802 info->flags &= ~SHMEM_PAGEIN;
803 spin_unlock(&info->lock);
808 error = inode_change_ok(inode, attr);
809 if (!error)
810 error = inode_setattr(inode, attr);
811 #ifdef CONFIG_TMPFS_POSIX_ACL
812 if (!error && (attr->ia_valid & ATTR_MODE))
813 error = generic_acl_chmod(inode);
814 #endif
815 if (page)
816 page_cache_release(page);
817 return error;
820 static void shmem_delete_inode(struct inode *inode)
822 struct shmem_inode_info *info = SHMEM_I(inode);
824 if (inode->i_op->truncate == shmem_truncate) {
825 truncate_inode_pages(inode->i_mapping, 0);
826 shmem_unacct_size(info->flags, inode->i_size);
827 inode->i_size = 0;
828 shmem_truncate(inode);
829 if (!list_empty(&info->swaplist)) {
830 mutex_lock(&shmem_swaplist_mutex);
831 list_del_init(&info->swaplist);
832 mutex_unlock(&shmem_swaplist_mutex);
835 BUG_ON(inode->i_blocks);
836 shmem_free_inode(inode->i_sb);
837 clear_inode(inode);
840 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
842 swp_entry_t *ptr;
844 for (ptr = dir; ptr < edir; ptr++) {
845 if (ptr->val == entry.val)
846 return ptr - dir;
848 return -1;
851 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
853 struct inode *inode;
854 unsigned long idx;
855 unsigned long size;
856 unsigned long limit;
857 unsigned long stage;
858 struct page **dir;
859 struct page *subdir;
860 swp_entry_t *ptr;
861 int offset;
862 int error;
864 idx = 0;
865 ptr = info->i_direct;
866 spin_lock(&info->lock);
867 if (!info->swapped) {
868 list_del_init(&info->swaplist);
869 goto lost2;
871 limit = info->next_index;
872 size = limit;
873 if (size > SHMEM_NR_DIRECT)
874 size = SHMEM_NR_DIRECT;
875 offset = shmem_find_swp(entry, ptr, ptr+size);
876 if (offset >= 0)
877 goto found;
878 if (!info->i_indirect)
879 goto lost2;
881 dir = shmem_dir_map(info->i_indirect);
882 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
884 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
885 if (unlikely(idx == stage)) {
886 shmem_dir_unmap(dir-1);
887 if (cond_resched_lock(&info->lock)) {
888 /* check it has not been truncated */
889 if (limit > info->next_index) {
890 limit = info->next_index;
891 if (idx >= limit)
892 goto lost2;
895 dir = shmem_dir_map(info->i_indirect) +
896 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
897 while (!*dir) {
898 dir++;
899 idx += ENTRIES_PER_PAGEPAGE;
900 if (idx >= limit)
901 goto lost1;
903 stage = idx + ENTRIES_PER_PAGEPAGE;
904 subdir = *dir;
905 shmem_dir_unmap(dir);
906 dir = shmem_dir_map(subdir);
908 subdir = *dir;
909 if (subdir && page_private(subdir)) {
910 ptr = shmem_swp_map(subdir);
911 size = limit - idx;
912 if (size > ENTRIES_PER_PAGE)
913 size = ENTRIES_PER_PAGE;
914 offset = shmem_find_swp(entry, ptr, ptr+size);
915 shmem_swp_unmap(ptr);
916 if (offset >= 0) {
917 shmem_dir_unmap(dir);
918 goto found;
922 lost1:
923 shmem_dir_unmap(dir-1);
924 lost2:
925 spin_unlock(&info->lock);
926 return 0;
927 found:
928 idx += offset;
929 inode = igrab(&info->vfs_inode);
930 spin_unlock(&info->lock);
933 * Move _head_ to start search for next from here.
934 * But be careful: shmem_delete_inode checks list_empty without taking
935 * mutex, and there's an instant in list_move_tail when info->swaplist
936 * would appear empty, if it were the only one on shmem_swaplist. We
937 * could avoid doing it if inode NULL; or use this minor optimization.
939 if (shmem_swaplist.next != &info->swaplist)
940 list_move_tail(&shmem_swaplist, &info->swaplist);
941 mutex_unlock(&shmem_swaplist_mutex);
943 error = 1;
944 if (!inode)
945 goto out;
947 * Charge page using GFP_KERNEL while we can wait.
948 * Charged back to the user(not to caller) when swap account is used.
949 * add_to_page_cache() will be called with GFP_NOWAIT.
951 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
952 if (error)
953 goto out;
954 error = radix_tree_preload(GFP_KERNEL);
955 if (error) {
956 mem_cgroup_uncharge_cache_page(page);
957 goto out;
959 error = 1;
961 spin_lock(&info->lock);
962 ptr = shmem_swp_entry(info, idx, NULL);
963 if (ptr && ptr->val == entry.val) {
964 error = add_to_page_cache_locked(page, inode->i_mapping,
965 idx, GFP_NOWAIT);
966 /* does mem_cgroup_uncharge_cache_page on error */
967 } else /* we must compensate for our precharge above */
968 mem_cgroup_uncharge_cache_page(page);
970 if (error == -EEXIST) {
971 struct page *filepage = find_get_page(inode->i_mapping, idx);
972 error = 1;
973 if (filepage) {
975 * There might be a more uptodate page coming down
976 * from a stacked writepage: forget our swappage if so.
978 if (PageUptodate(filepage))
979 error = 0;
980 page_cache_release(filepage);
983 if (!error) {
984 delete_from_swap_cache(page);
985 set_page_dirty(page);
986 info->flags |= SHMEM_PAGEIN;
987 shmem_swp_set(info, ptr, 0);
988 swap_free(entry);
989 error = 1; /* not an error, but entry was found */
991 if (ptr)
992 shmem_swp_unmap(ptr);
993 spin_unlock(&info->lock);
994 radix_tree_preload_end();
995 out:
996 unlock_page(page);
997 page_cache_release(page);
998 iput(inode); /* allows for NULL */
999 return error;
1003 * shmem_unuse() search for an eventually swapped out shmem page.
1005 int shmem_unuse(swp_entry_t entry, struct page *page)
1007 struct list_head *p, *next;
1008 struct shmem_inode_info *info;
1009 int found = 0;
1011 mutex_lock(&shmem_swaplist_mutex);
1012 list_for_each_safe(p, next, &shmem_swaplist) {
1013 info = list_entry(p, struct shmem_inode_info, swaplist);
1014 found = shmem_unuse_inode(info, entry, page);
1015 cond_resched();
1016 if (found)
1017 goto out;
1019 mutex_unlock(&shmem_swaplist_mutex);
1021 * Can some race bring us here? We've been holding page lock,
1022 * so I think not; but would rather try again later than BUG()
1024 unlock_page(page);
1025 page_cache_release(page);
1026 out:
1027 return (found < 0) ? found : 0;
1031 * Move the page from the page cache to the swap cache.
1033 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1035 struct shmem_inode_info *info;
1036 swp_entry_t *entry, swap;
1037 struct address_space *mapping;
1038 unsigned long index;
1039 struct inode *inode;
1041 BUG_ON(!PageLocked(page));
1042 mapping = page->mapping;
1043 index = page->index;
1044 inode = mapping->host;
1045 info = SHMEM_I(inode);
1046 if (info->flags & VM_LOCKED)
1047 goto redirty;
1048 if (!total_swap_pages)
1049 goto redirty;
1052 * shmem_backing_dev_info's capabilities prevent regular writeback or
1053 * sync from ever calling shmem_writepage; but a stacking filesystem
1054 * may use the ->writepage of its underlying filesystem, in which case
1055 * tmpfs should write out to swap only in response to memory pressure,
1056 * and not for the writeback threads or sync. However, in those cases,
1057 * we do still want to check if there's a redundant swappage to be
1058 * discarded.
1060 if (wbc->for_reclaim)
1061 swap = get_swap_page();
1062 else
1063 swap.val = 0;
1065 spin_lock(&info->lock);
1066 if (index >= info->next_index) {
1067 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1068 goto unlock;
1070 entry = shmem_swp_entry(info, index, NULL);
1071 if (entry->val) {
1073 * The more uptodate page coming down from a stacked
1074 * writepage should replace our old swappage.
1076 free_swap_and_cache(*entry);
1077 shmem_swp_set(info, entry, 0);
1079 shmem_recalc_inode(inode);
1081 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1082 remove_from_page_cache(page);
1083 shmem_swp_set(info, entry, swap.val);
1084 shmem_swp_unmap(entry);
1085 if (list_empty(&info->swaplist))
1086 inode = igrab(inode);
1087 else
1088 inode = NULL;
1089 spin_unlock(&info->lock);
1090 swap_shmem_alloc(swap);
1091 BUG_ON(page_mapped(page));
1092 page_cache_release(page); /* pagecache ref */
1093 swap_writepage(page, wbc);
1094 if (inode) {
1095 mutex_lock(&shmem_swaplist_mutex);
1096 /* move instead of add in case we're racing */
1097 list_move_tail(&info->swaplist, &shmem_swaplist);
1098 mutex_unlock(&shmem_swaplist_mutex);
1099 iput(inode);
1101 return 0;
1104 shmem_swp_unmap(entry);
1105 unlock:
1106 spin_unlock(&info->lock);
1108 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1109 * clear SWAP_HAS_CACHE flag.
1111 swapcache_free(swap, NULL);
1112 redirty:
1113 set_page_dirty(page);
1114 if (wbc->for_reclaim)
1115 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1116 unlock_page(page);
1117 return 0;
1120 #ifdef CONFIG_NUMA
1121 #ifdef CONFIG_TMPFS
1122 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1124 char buffer[64];
1126 if (!mpol || mpol->mode == MPOL_DEFAULT)
1127 return; /* show nothing */
1129 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1131 seq_printf(seq, ",mpol=%s", buffer);
1134 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1136 struct mempolicy *mpol = NULL;
1137 if (sbinfo->mpol) {
1138 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1139 mpol = sbinfo->mpol;
1140 mpol_get(mpol);
1141 spin_unlock(&sbinfo->stat_lock);
1143 return mpol;
1145 #endif /* CONFIG_TMPFS */
1147 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1148 struct shmem_inode_info *info, unsigned long idx)
1150 struct mempolicy mpol, *spol;
1151 struct vm_area_struct pvma;
1152 struct page *page;
1154 spol = mpol_cond_copy(&mpol,
1155 mpol_shared_policy_lookup(&info->policy, idx));
1157 /* Create a pseudo vma that just contains the policy */
1158 pvma.vm_start = 0;
1159 pvma.vm_pgoff = idx;
1160 pvma.vm_ops = NULL;
1161 pvma.vm_policy = spol;
1162 page = swapin_readahead(entry, gfp, &pvma, 0);
1163 return page;
1166 static struct page *shmem_alloc_page(gfp_t gfp,
1167 struct shmem_inode_info *info, unsigned long idx)
1169 struct vm_area_struct pvma;
1171 /* Create a pseudo vma that just contains the policy */
1172 pvma.vm_start = 0;
1173 pvma.vm_pgoff = idx;
1174 pvma.vm_ops = NULL;
1175 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1178 * alloc_page_vma() will drop the shared policy reference
1180 return alloc_page_vma(gfp, &pvma, 0);
1182 #else /* !CONFIG_NUMA */
1183 #ifdef CONFIG_TMPFS
1184 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1187 #endif /* CONFIG_TMPFS */
1189 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1190 struct shmem_inode_info *info, unsigned long idx)
1192 return swapin_readahead(entry, gfp, NULL, 0);
1195 static inline struct page *shmem_alloc_page(gfp_t gfp,
1196 struct shmem_inode_info *info, unsigned long idx)
1198 return alloc_page(gfp);
1200 #endif /* CONFIG_NUMA */
1202 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1203 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1205 return NULL;
1207 #endif
1210 * shmem_getpage - either get the page from swap or allocate a new one
1212 * If we allocate a new one we do not mark it dirty. That's up to the
1213 * vm. If we swap it in we mark it dirty since we also free the swap
1214 * entry since a page cannot live in both the swap and page cache
1216 static int shmem_getpage(struct inode *inode, unsigned long idx,
1217 struct page **pagep, enum sgp_type sgp, int *type)
1219 struct address_space *mapping = inode->i_mapping;
1220 struct shmem_inode_info *info = SHMEM_I(inode);
1221 struct shmem_sb_info *sbinfo;
1222 struct page *filepage = *pagep;
1223 struct page *swappage;
1224 swp_entry_t *entry;
1225 swp_entry_t swap;
1226 gfp_t gfp;
1227 int error;
1229 if (idx >= SHMEM_MAX_INDEX)
1230 return -EFBIG;
1232 if (type)
1233 *type = 0;
1236 * Normally, filepage is NULL on entry, and either found
1237 * uptodate immediately, or allocated and zeroed, or read
1238 * in under swappage, which is then assigned to filepage.
1239 * But shmem_readpage (required for splice) passes in a locked
1240 * filepage, which may be found not uptodate by other callers
1241 * too, and may need to be copied from the swappage read in.
1243 repeat:
1244 if (!filepage)
1245 filepage = find_lock_page(mapping, idx);
1246 if (filepage && PageUptodate(filepage))
1247 goto done;
1248 error = 0;
1249 gfp = mapping_gfp_mask(mapping);
1250 if (!filepage) {
1252 * Try to preload while we can wait, to not make a habit of
1253 * draining atomic reserves; but don't latch on to this cpu.
1255 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1256 if (error)
1257 goto failed;
1258 radix_tree_preload_end();
1261 spin_lock(&info->lock);
1262 shmem_recalc_inode(inode);
1263 entry = shmem_swp_alloc(info, idx, sgp);
1264 if (IS_ERR(entry)) {
1265 spin_unlock(&info->lock);
1266 error = PTR_ERR(entry);
1267 goto failed;
1269 swap = *entry;
1271 if (swap.val) {
1272 /* Look it up and read it in.. */
1273 swappage = lookup_swap_cache(swap);
1274 if (!swappage) {
1275 shmem_swp_unmap(entry);
1276 /* here we actually do the io */
1277 if (type && !(*type & VM_FAULT_MAJOR)) {
1278 __count_vm_event(PGMAJFAULT);
1279 *type |= VM_FAULT_MAJOR;
1281 spin_unlock(&info->lock);
1282 swappage = shmem_swapin(swap, gfp, info, idx);
1283 if (!swappage) {
1284 spin_lock(&info->lock);
1285 entry = shmem_swp_alloc(info, idx, sgp);
1286 if (IS_ERR(entry))
1287 error = PTR_ERR(entry);
1288 else {
1289 if (entry->val == swap.val)
1290 error = -ENOMEM;
1291 shmem_swp_unmap(entry);
1293 spin_unlock(&info->lock);
1294 if (error)
1295 goto failed;
1296 goto repeat;
1298 wait_on_page_locked(swappage);
1299 page_cache_release(swappage);
1300 goto repeat;
1303 /* We have to do this with page locked to prevent races */
1304 if (!trylock_page(swappage)) {
1305 shmem_swp_unmap(entry);
1306 spin_unlock(&info->lock);
1307 wait_on_page_locked(swappage);
1308 page_cache_release(swappage);
1309 goto repeat;
1311 if (PageWriteback(swappage)) {
1312 shmem_swp_unmap(entry);
1313 spin_unlock(&info->lock);
1314 wait_on_page_writeback(swappage);
1315 unlock_page(swappage);
1316 page_cache_release(swappage);
1317 goto repeat;
1319 if (!PageUptodate(swappage)) {
1320 shmem_swp_unmap(entry);
1321 spin_unlock(&info->lock);
1322 unlock_page(swappage);
1323 page_cache_release(swappage);
1324 error = -EIO;
1325 goto failed;
1328 if (filepage) {
1329 shmem_swp_set(info, entry, 0);
1330 shmem_swp_unmap(entry);
1331 delete_from_swap_cache(swappage);
1332 spin_unlock(&info->lock);
1333 copy_highpage(filepage, swappage);
1334 unlock_page(swappage);
1335 page_cache_release(swappage);
1336 flush_dcache_page(filepage);
1337 SetPageUptodate(filepage);
1338 set_page_dirty(filepage);
1339 swap_free(swap);
1340 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1341 idx, GFP_NOWAIT))) {
1342 info->flags |= SHMEM_PAGEIN;
1343 shmem_swp_set(info, entry, 0);
1344 shmem_swp_unmap(entry);
1345 delete_from_swap_cache(swappage);
1346 spin_unlock(&info->lock);
1347 filepage = swappage;
1348 set_page_dirty(filepage);
1349 swap_free(swap);
1350 } else {
1351 shmem_swp_unmap(entry);
1352 spin_unlock(&info->lock);
1353 if (error == -ENOMEM) {
1355 * reclaim from proper memory cgroup and
1356 * call memcg's OOM if needed.
1358 error = mem_cgroup_shmem_charge_fallback(
1359 swappage,
1360 current->mm,
1361 gfp);
1362 if (error) {
1363 unlock_page(swappage);
1364 page_cache_release(swappage);
1365 goto failed;
1368 unlock_page(swappage);
1369 page_cache_release(swappage);
1370 goto repeat;
1372 } else if (sgp == SGP_READ && !filepage) {
1373 shmem_swp_unmap(entry);
1374 filepage = find_get_page(mapping, idx);
1375 if (filepage &&
1376 (!PageUptodate(filepage) || !trylock_page(filepage))) {
1377 spin_unlock(&info->lock);
1378 wait_on_page_locked(filepage);
1379 page_cache_release(filepage);
1380 filepage = NULL;
1381 goto repeat;
1383 spin_unlock(&info->lock);
1384 } else {
1385 shmem_swp_unmap(entry);
1386 sbinfo = SHMEM_SB(inode->i_sb);
1387 if (sbinfo->max_blocks) {
1388 spin_lock(&sbinfo->stat_lock);
1389 if (sbinfo->free_blocks == 0 ||
1390 shmem_acct_block(info->flags)) {
1391 spin_unlock(&sbinfo->stat_lock);
1392 spin_unlock(&info->lock);
1393 error = -ENOSPC;
1394 goto failed;
1396 sbinfo->free_blocks--;
1397 inode->i_blocks += BLOCKS_PER_PAGE;
1398 spin_unlock(&sbinfo->stat_lock);
1399 } else if (shmem_acct_block(info->flags)) {
1400 spin_unlock(&info->lock);
1401 error = -ENOSPC;
1402 goto failed;
1405 if (!filepage) {
1406 int ret;
1408 spin_unlock(&info->lock);
1409 filepage = shmem_alloc_page(gfp, info, idx);
1410 if (!filepage) {
1411 shmem_unacct_blocks(info->flags, 1);
1412 shmem_free_blocks(inode, 1);
1413 error = -ENOMEM;
1414 goto failed;
1416 SetPageSwapBacked(filepage);
1418 /* Precharge page while we can wait, compensate after */
1419 error = mem_cgroup_cache_charge(filepage, current->mm,
1420 GFP_KERNEL);
1421 if (error) {
1422 page_cache_release(filepage);
1423 shmem_unacct_blocks(info->flags, 1);
1424 shmem_free_blocks(inode, 1);
1425 filepage = NULL;
1426 goto failed;
1429 spin_lock(&info->lock);
1430 entry = shmem_swp_alloc(info, idx, sgp);
1431 if (IS_ERR(entry))
1432 error = PTR_ERR(entry);
1433 else {
1434 swap = *entry;
1435 shmem_swp_unmap(entry);
1437 ret = error || swap.val;
1438 if (ret)
1439 mem_cgroup_uncharge_cache_page(filepage);
1440 else
1441 ret = add_to_page_cache_lru(filepage, mapping,
1442 idx, GFP_NOWAIT);
1444 * At add_to_page_cache_lru() failure, uncharge will
1445 * be done automatically.
1447 if (ret) {
1448 spin_unlock(&info->lock);
1449 page_cache_release(filepage);
1450 shmem_unacct_blocks(info->flags, 1);
1451 shmem_free_blocks(inode, 1);
1452 filepage = NULL;
1453 if (error)
1454 goto failed;
1455 goto repeat;
1457 info->flags |= SHMEM_PAGEIN;
1460 info->alloced++;
1461 spin_unlock(&info->lock);
1462 clear_highpage(filepage);
1463 flush_dcache_page(filepage);
1464 SetPageUptodate(filepage);
1465 if (sgp == SGP_DIRTY)
1466 set_page_dirty(filepage);
1468 done:
1469 *pagep = filepage;
1470 return 0;
1472 failed:
1473 if (*pagep != filepage) {
1474 unlock_page(filepage);
1475 page_cache_release(filepage);
1477 return error;
1480 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1482 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1483 int error;
1484 int ret;
1486 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1487 return VM_FAULT_SIGBUS;
1489 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1490 if (error)
1491 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1493 return ret | VM_FAULT_LOCKED;
1496 #ifdef CONFIG_NUMA
1497 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1499 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1500 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1503 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1504 unsigned long addr)
1506 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1507 unsigned long idx;
1509 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1510 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1512 #endif
1514 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1516 struct inode *inode = file->f_path.dentry->d_inode;
1517 struct shmem_inode_info *info = SHMEM_I(inode);
1518 int retval = -ENOMEM;
1520 spin_lock(&info->lock);
1521 if (lock && !(info->flags & VM_LOCKED)) {
1522 if (!user_shm_lock(inode->i_size, user))
1523 goto out_nomem;
1524 info->flags |= VM_LOCKED;
1525 mapping_set_unevictable(file->f_mapping);
1527 if (!lock && (info->flags & VM_LOCKED) && user) {
1528 user_shm_unlock(inode->i_size, user);
1529 info->flags &= ~VM_LOCKED;
1530 mapping_clear_unevictable(file->f_mapping);
1531 scan_mapping_unevictable_pages(file->f_mapping);
1533 retval = 0;
1535 out_nomem:
1536 spin_unlock(&info->lock);
1537 return retval;
1540 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1542 file_accessed(file);
1543 vma->vm_ops = &shmem_vm_ops;
1544 vma->vm_flags |= VM_CAN_NONLINEAR;
1545 return 0;
1548 static struct inode *shmem_get_inode(struct super_block *sb, int mode,
1549 dev_t dev, unsigned long flags)
1551 struct inode *inode;
1552 struct shmem_inode_info *info;
1553 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1555 if (shmem_reserve_inode(sb))
1556 return NULL;
1558 inode = new_inode(sb);
1559 if (inode) {
1560 inode->i_mode = mode;
1561 inode->i_uid = current_fsuid();
1562 inode->i_gid = current_fsgid();
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, 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 if (dir->i_mode & S_ISGID) {
1837 inode->i_gid = dir->i_gid;
1838 if (S_ISDIR(mode))
1839 inode->i_mode |= S_ISGID;
1841 dir->i_size += BOGO_DIRENT_SIZE;
1842 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1843 d_instantiate(dentry, inode);
1844 dget(dentry); /* Extra count - pin the dentry in core */
1846 return error;
1849 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1851 int error;
1853 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1854 return error;
1855 inc_nlink(dir);
1856 return 0;
1859 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1860 struct nameidata *nd)
1862 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1866 * Link a file..
1868 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1870 struct inode *inode = old_dentry->d_inode;
1871 int ret;
1874 * No ordinary (disk based) filesystem counts links as inodes;
1875 * but each new link needs a new dentry, pinning lowmem, and
1876 * tmpfs dentries cannot be pruned until they are unlinked.
1878 ret = shmem_reserve_inode(inode->i_sb);
1879 if (ret)
1880 goto out;
1882 dir->i_size += BOGO_DIRENT_SIZE;
1883 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1884 inc_nlink(inode);
1885 atomic_inc(&inode->i_count); /* New dentry reference */
1886 dget(dentry); /* Extra pinning count for the created dentry */
1887 d_instantiate(dentry, inode);
1888 out:
1889 return ret;
1892 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1894 struct inode *inode = dentry->d_inode;
1896 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1897 shmem_free_inode(inode->i_sb);
1899 dir->i_size -= BOGO_DIRENT_SIZE;
1900 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1901 drop_nlink(inode);
1902 dput(dentry); /* Undo the count from "create" - this does all the work */
1903 return 0;
1906 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1908 if (!simple_empty(dentry))
1909 return -ENOTEMPTY;
1911 drop_nlink(dentry->d_inode);
1912 drop_nlink(dir);
1913 return shmem_unlink(dir, dentry);
1917 * The VFS layer already does all the dentry stuff for rename,
1918 * we just have to decrement the usage count for the target if
1919 * it exists so that the VFS layer correctly free's it when it
1920 * gets overwritten.
1922 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1924 struct inode *inode = old_dentry->d_inode;
1925 int they_are_dirs = S_ISDIR(inode->i_mode);
1927 if (!simple_empty(new_dentry))
1928 return -ENOTEMPTY;
1930 if (new_dentry->d_inode) {
1931 (void) shmem_unlink(new_dir, new_dentry);
1932 if (they_are_dirs)
1933 drop_nlink(old_dir);
1934 } else if (they_are_dirs) {
1935 drop_nlink(old_dir);
1936 inc_nlink(new_dir);
1939 old_dir->i_size -= BOGO_DIRENT_SIZE;
1940 new_dir->i_size += BOGO_DIRENT_SIZE;
1941 old_dir->i_ctime = old_dir->i_mtime =
1942 new_dir->i_ctime = new_dir->i_mtime =
1943 inode->i_ctime = CURRENT_TIME;
1944 return 0;
1947 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1949 int error;
1950 int len;
1951 struct inode *inode;
1952 struct page *page = NULL;
1953 char *kaddr;
1954 struct shmem_inode_info *info;
1956 len = strlen(symname) + 1;
1957 if (len > PAGE_CACHE_SIZE)
1958 return -ENAMETOOLONG;
1960 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1961 if (!inode)
1962 return -ENOSPC;
1964 error = security_inode_init_security(inode, dir, NULL, NULL,
1965 NULL);
1966 if (error) {
1967 if (error != -EOPNOTSUPP) {
1968 iput(inode);
1969 return error;
1971 error = 0;
1974 info = SHMEM_I(inode);
1975 inode->i_size = len-1;
1976 if (len <= (char *)inode - (char *)info) {
1977 /* do it inline */
1978 memcpy(info, symname, len);
1979 inode->i_op = &shmem_symlink_inline_operations;
1980 } else {
1981 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1982 if (error) {
1983 iput(inode);
1984 return error;
1986 inode->i_mapping->a_ops = &shmem_aops;
1987 inode->i_op = &shmem_symlink_inode_operations;
1988 kaddr = kmap_atomic(page, KM_USER0);
1989 memcpy(kaddr, symname, len);
1990 kunmap_atomic(kaddr, KM_USER0);
1991 set_page_dirty(page);
1992 unlock_page(page);
1993 page_cache_release(page);
1995 if (dir->i_mode & S_ISGID)
1996 inode->i_gid = dir->i_gid;
1997 dir->i_size += BOGO_DIRENT_SIZE;
1998 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1999 d_instantiate(dentry, inode);
2000 dget(dentry);
2001 return 0;
2004 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
2006 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
2007 return NULL;
2010 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2012 struct page *page = NULL;
2013 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2014 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
2015 if (page)
2016 unlock_page(page);
2017 return page;
2020 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2022 if (!IS_ERR(nd_get_link(nd))) {
2023 struct page *page = cookie;
2024 kunmap(page);
2025 mark_page_accessed(page);
2026 page_cache_release(page);
2030 static const struct inode_operations shmem_symlink_inline_operations = {
2031 .readlink = generic_readlink,
2032 .follow_link = shmem_follow_link_inline,
2035 static const struct inode_operations shmem_symlink_inode_operations = {
2036 .truncate = shmem_truncate,
2037 .readlink = generic_readlink,
2038 .follow_link = shmem_follow_link,
2039 .put_link = shmem_put_link,
2042 #ifdef CONFIG_TMPFS_POSIX_ACL
2044 * Superblocks without xattr inode operations will get security.* xattr
2045 * support from the VFS "for free". As soon as we have any other xattrs
2046 * like ACLs, we also need to implement the security.* handlers at
2047 * filesystem level, though.
2050 static size_t shmem_xattr_security_list(struct dentry *dentry, char *list,
2051 size_t list_len, const char *name,
2052 size_t name_len, int handler_flags)
2054 return security_inode_listsecurity(dentry->d_inode, list, list_len);
2057 static int shmem_xattr_security_get(struct dentry *dentry, const char *name,
2058 void *buffer, size_t size, int handler_flags)
2060 if (strcmp(name, "") == 0)
2061 return -EINVAL;
2062 return xattr_getsecurity(dentry->d_inode, name, buffer, size);
2065 static int shmem_xattr_security_set(struct dentry *dentry, const char *name,
2066 const void *value, size_t size, int flags, int handler_flags)
2068 if (strcmp(name, "") == 0)
2069 return -EINVAL;
2070 return security_inode_setsecurity(dentry->d_inode, name, value,
2071 size, flags);
2074 static struct xattr_handler shmem_xattr_security_handler = {
2075 .prefix = XATTR_SECURITY_PREFIX,
2076 .list = shmem_xattr_security_list,
2077 .get = shmem_xattr_security_get,
2078 .set = shmem_xattr_security_set,
2081 static struct xattr_handler *shmem_xattr_handlers[] = {
2082 &generic_acl_access_handler,
2083 &generic_acl_default_handler,
2084 &shmem_xattr_security_handler,
2085 NULL
2087 #endif
2089 static struct dentry *shmem_get_parent(struct dentry *child)
2091 return ERR_PTR(-ESTALE);
2094 static int shmem_match(struct inode *ino, void *vfh)
2096 __u32 *fh = vfh;
2097 __u64 inum = fh[2];
2098 inum = (inum << 32) | fh[1];
2099 return ino->i_ino == inum && fh[0] == ino->i_generation;
2102 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2103 struct fid *fid, int fh_len, int fh_type)
2105 struct inode *inode;
2106 struct dentry *dentry = NULL;
2107 u64 inum = fid->raw[2];
2108 inum = (inum << 32) | fid->raw[1];
2110 if (fh_len < 3)
2111 return NULL;
2113 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2114 shmem_match, fid->raw);
2115 if (inode) {
2116 dentry = d_find_alias(inode);
2117 iput(inode);
2120 return dentry;
2123 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2124 int connectable)
2126 struct inode *inode = dentry->d_inode;
2128 if (*len < 3)
2129 return 255;
2131 if (hlist_unhashed(&inode->i_hash)) {
2132 /* Unfortunately insert_inode_hash is not idempotent,
2133 * so as we hash inodes here rather than at creation
2134 * time, we need a lock to ensure we only try
2135 * to do it once
2137 static DEFINE_SPINLOCK(lock);
2138 spin_lock(&lock);
2139 if (hlist_unhashed(&inode->i_hash))
2140 __insert_inode_hash(inode,
2141 inode->i_ino + inode->i_generation);
2142 spin_unlock(&lock);
2145 fh[0] = inode->i_generation;
2146 fh[1] = inode->i_ino;
2147 fh[2] = ((__u64)inode->i_ino) >> 32;
2149 *len = 3;
2150 return 1;
2153 static const struct export_operations shmem_export_ops = {
2154 .get_parent = shmem_get_parent,
2155 .encode_fh = shmem_encode_fh,
2156 .fh_to_dentry = shmem_fh_to_dentry,
2159 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2160 bool remount)
2162 char *this_char, *value, *rest;
2164 while (options != NULL) {
2165 this_char = options;
2166 for (;;) {
2168 * NUL-terminate this option: unfortunately,
2169 * mount options form a comma-separated list,
2170 * but mpol's nodelist may also contain commas.
2172 options = strchr(options, ',');
2173 if (options == NULL)
2174 break;
2175 options++;
2176 if (!isdigit(*options)) {
2177 options[-1] = '\0';
2178 break;
2181 if (!*this_char)
2182 continue;
2183 if ((value = strchr(this_char,'=')) != NULL) {
2184 *value++ = 0;
2185 } else {
2186 printk(KERN_ERR
2187 "tmpfs: No value for mount option '%s'\n",
2188 this_char);
2189 return 1;
2192 if (!strcmp(this_char,"size")) {
2193 unsigned long long size;
2194 size = memparse(value,&rest);
2195 if (*rest == '%') {
2196 size <<= PAGE_SHIFT;
2197 size *= totalram_pages;
2198 do_div(size, 100);
2199 rest++;
2201 if (*rest)
2202 goto bad_val;
2203 sbinfo->max_blocks =
2204 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2205 } else if (!strcmp(this_char,"nr_blocks")) {
2206 sbinfo->max_blocks = memparse(value, &rest);
2207 if (*rest)
2208 goto bad_val;
2209 } else if (!strcmp(this_char,"nr_inodes")) {
2210 sbinfo->max_inodes = memparse(value, &rest);
2211 if (*rest)
2212 goto bad_val;
2213 } else if (!strcmp(this_char,"mode")) {
2214 if (remount)
2215 continue;
2216 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2217 if (*rest)
2218 goto bad_val;
2219 } else if (!strcmp(this_char,"uid")) {
2220 if (remount)
2221 continue;
2222 sbinfo->uid = simple_strtoul(value, &rest, 0);
2223 if (*rest)
2224 goto bad_val;
2225 } else if (!strcmp(this_char,"gid")) {
2226 if (remount)
2227 continue;
2228 sbinfo->gid = simple_strtoul(value, &rest, 0);
2229 if (*rest)
2230 goto bad_val;
2231 } else if (!strcmp(this_char,"mpol")) {
2232 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2233 goto bad_val;
2234 } else {
2235 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2236 this_char);
2237 return 1;
2240 return 0;
2242 bad_val:
2243 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2244 value, this_char);
2245 return 1;
2249 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2251 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2252 struct shmem_sb_info config = *sbinfo;
2253 unsigned long blocks;
2254 unsigned long inodes;
2255 int error = -EINVAL;
2257 if (shmem_parse_options(data, &config, true))
2258 return error;
2260 spin_lock(&sbinfo->stat_lock);
2261 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2262 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2263 if (config.max_blocks < blocks)
2264 goto out;
2265 if (config.max_inodes < inodes)
2266 goto out;
2268 * Those tests also disallow limited->unlimited while any are in
2269 * use, so i_blocks will always be zero when max_blocks is zero;
2270 * but we must separately disallow unlimited->limited, because
2271 * in that case we have no record of how much is already in use.
2273 if (config.max_blocks && !sbinfo->max_blocks)
2274 goto out;
2275 if (config.max_inodes && !sbinfo->max_inodes)
2276 goto out;
2278 error = 0;
2279 sbinfo->max_blocks = config.max_blocks;
2280 sbinfo->free_blocks = config.max_blocks - blocks;
2281 sbinfo->max_inodes = config.max_inodes;
2282 sbinfo->free_inodes = config.max_inodes - inodes;
2284 mpol_put(sbinfo->mpol);
2285 sbinfo->mpol = config.mpol; /* transfers initial ref */
2286 out:
2287 spin_unlock(&sbinfo->stat_lock);
2288 return error;
2291 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2293 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2295 if (sbinfo->max_blocks != shmem_default_max_blocks())
2296 seq_printf(seq, ",size=%luk",
2297 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2298 if (sbinfo->max_inodes != shmem_default_max_inodes())
2299 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2300 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2301 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2302 if (sbinfo->uid != 0)
2303 seq_printf(seq, ",uid=%u", sbinfo->uid);
2304 if (sbinfo->gid != 0)
2305 seq_printf(seq, ",gid=%u", sbinfo->gid);
2306 shmem_show_mpol(seq, sbinfo->mpol);
2307 return 0;
2309 #endif /* CONFIG_TMPFS */
2311 static void shmem_put_super(struct super_block *sb)
2313 kfree(sb->s_fs_info);
2314 sb->s_fs_info = NULL;
2317 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2319 struct inode *inode;
2320 struct dentry *root;
2321 struct shmem_sb_info *sbinfo;
2322 int err = -ENOMEM;
2324 /* Round up to L1_CACHE_BYTES to resist false sharing */
2325 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2326 L1_CACHE_BYTES), GFP_KERNEL);
2327 if (!sbinfo)
2328 return -ENOMEM;
2330 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2331 sbinfo->uid = current_fsuid();
2332 sbinfo->gid = current_fsgid();
2333 sb->s_fs_info = sbinfo;
2335 #ifdef CONFIG_TMPFS
2337 * Per default we only allow half of the physical ram per
2338 * tmpfs instance, limiting inodes to one per page of lowmem;
2339 * but the internal instance is left unlimited.
2341 if (!(sb->s_flags & MS_NOUSER)) {
2342 sbinfo->max_blocks = shmem_default_max_blocks();
2343 sbinfo->max_inodes = shmem_default_max_inodes();
2344 if (shmem_parse_options(data, sbinfo, false)) {
2345 err = -EINVAL;
2346 goto failed;
2349 sb->s_export_op = &shmem_export_ops;
2350 #else
2351 sb->s_flags |= MS_NOUSER;
2352 #endif
2354 spin_lock_init(&sbinfo->stat_lock);
2355 sbinfo->free_blocks = sbinfo->max_blocks;
2356 sbinfo->free_inodes = sbinfo->max_inodes;
2358 sb->s_maxbytes = SHMEM_MAX_BYTES;
2359 sb->s_blocksize = PAGE_CACHE_SIZE;
2360 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2361 sb->s_magic = TMPFS_MAGIC;
2362 sb->s_op = &shmem_ops;
2363 sb->s_time_gran = 1;
2364 #ifdef CONFIG_TMPFS_POSIX_ACL
2365 sb->s_xattr = shmem_xattr_handlers;
2366 sb->s_flags |= MS_POSIXACL;
2367 #endif
2369 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2370 if (!inode)
2371 goto failed;
2372 inode->i_uid = sbinfo->uid;
2373 inode->i_gid = sbinfo->gid;
2374 root = d_alloc_root(inode);
2375 if (!root)
2376 goto failed_iput;
2377 sb->s_root = root;
2378 return 0;
2380 failed_iput:
2381 iput(inode);
2382 failed:
2383 shmem_put_super(sb);
2384 return err;
2387 static struct kmem_cache *shmem_inode_cachep;
2389 static struct inode *shmem_alloc_inode(struct super_block *sb)
2391 struct shmem_inode_info *p;
2392 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2393 if (!p)
2394 return NULL;
2395 return &p->vfs_inode;
2398 static void shmem_destroy_inode(struct inode *inode)
2400 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2401 /* only struct inode is valid if it's an inline symlink */
2402 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2404 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2407 static void init_once(void *foo)
2409 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2411 inode_init_once(&p->vfs_inode);
2414 static int init_inodecache(void)
2416 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2417 sizeof(struct shmem_inode_info),
2418 0, SLAB_PANIC, init_once);
2419 return 0;
2422 static void destroy_inodecache(void)
2424 kmem_cache_destroy(shmem_inode_cachep);
2427 static const struct address_space_operations shmem_aops = {
2428 .writepage = shmem_writepage,
2429 .set_page_dirty = __set_page_dirty_no_writeback,
2430 #ifdef CONFIG_TMPFS
2431 .readpage = shmem_readpage,
2432 .write_begin = shmem_write_begin,
2433 .write_end = shmem_write_end,
2434 #endif
2435 .migratepage = migrate_page,
2436 .error_remove_page = generic_error_remove_page,
2439 static const struct file_operations shmem_file_operations = {
2440 .mmap = shmem_mmap,
2441 #ifdef CONFIG_TMPFS
2442 .llseek = generic_file_llseek,
2443 .read = do_sync_read,
2444 .write = do_sync_write,
2445 .aio_read = shmem_file_aio_read,
2446 .aio_write = generic_file_aio_write,
2447 .fsync = simple_sync_file,
2448 .splice_read = generic_file_splice_read,
2449 .splice_write = generic_file_splice_write,
2450 #endif
2453 static const struct inode_operations shmem_inode_operations = {
2454 .truncate = shmem_truncate,
2455 .setattr = shmem_notify_change,
2456 .truncate_range = shmem_truncate_range,
2457 #ifdef CONFIG_TMPFS_POSIX_ACL
2458 .setxattr = generic_setxattr,
2459 .getxattr = generic_getxattr,
2460 .listxattr = generic_listxattr,
2461 .removexattr = generic_removexattr,
2462 .check_acl = generic_check_acl,
2463 #endif
2467 static const struct inode_operations shmem_dir_inode_operations = {
2468 #ifdef CONFIG_TMPFS
2469 .create = shmem_create,
2470 .lookup = simple_lookup,
2471 .link = shmem_link,
2472 .unlink = shmem_unlink,
2473 .symlink = shmem_symlink,
2474 .mkdir = shmem_mkdir,
2475 .rmdir = shmem_rmdir,
2476 .mknod = shmem_mknod,
2477 .rename = shmem_rename,
2478 #endif
2479 #ifdef CONFIG_TMPFS_POSIX_ACL
2480 .setattr = shmem_notify_change,
2481 .setxattr = generic_setxattr,
2482 .getxattr = generic_getxattr,
2483 .listxattr = generic_listxattr,
2484 .removexattr = generic_removexattr,
2485 .check_acl = generic_check_acl,
2486 #endif
2489 static const struct inode_operations shmem_special_inode_operations = {
2490 #ifdef CONFIG_TMPFS_POSIX_ACL
2491 .setattr = shmem_notify_change,
2492 .setxattr = generic_setxattr,
2493 .getxattr = generic_getxattr,
2494 .listxattr = generic_listxattr,
2495 .removexattr = generic_removexattr,
2496 .check_acl = generic_check_acl,
2497 #endif
2500 static const struct super_operations shmem_ops = {
2501 .alloc_inode = shmem_alloc_inode,
2502 .destroy_inode = shmem_destroy_inode,
2503 #ifdef CONFIG_TMPFS
2504 .statfs = shmem_statfs,
2505 .remount_fs = shmem_remount_fs,
2506 .show_options = shmem_show_options,
2507 #endif
2508 .delete_inode = shmem_delete_inode,
2509 .drop_inode = generic_delete_inode,
2510 .put_super = shmem_put_super,
2513 static const struct vm_operations_struct shmem_vm_ops = {
2514 .fault = shmem_fault,
2515 #ifdef CONFIG_NUMA
2516 .set_policy = shmem_set_policy,
2517 .get_policy = shmem_get_policy,
2518 #endif
2522 static int shmem_get_sb(struct file_system_type *fs_type,
2523 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2525 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2528 static struct file_system_type tmpfs_fs_type = {
2529 .owner = THIS_MODULE,
2530 .name = "tmpfs",
2531 .get_sb = shmem_get_sb,
2532 .kill_sb = kill_litter_super,
2535 int __init init_tmpfs(void)
2537 int error;
2539 error = bdi_init(&shmem_backing_dev_info);
2540 if (error)
2541 goto out4;
2543 error = init_inodecache();
2544 if (error)
2545 goto out3;
2547 error = register_filesystem(&tmpfs_fs_type);
2548 if (error) {
2549 printk(KERN_ERR "Could not register tmpfs\n");
2550 goto out2;
2553 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2554 tmpfs_fs_type.name, NULL);
2555 if (IS_ERR(shm_mnt)) {
2556 error = PTR_ERR(shm_mnt);
2557 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2558 goto out1;
2560 return 0;
2562 out1:
2563 unregister_filesystem(&tmpfs_fs_type);
2564 out2:
2565 destroy_inodecache();
2566 out3:
2567 bdi_destroy(&shmem_backing_dev_info);
2568 out4:
2569 shm_mnt = ERR_PTR(error);
2570 return error;
2573 #else /* !CONFIG_SHMEM */
2576 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2578 * This is intended for small system where the benefits of the full
2579 * shmem code (swap-backed and resource-limited) are outweighed by
2580 * their complexity. On systems without swap this code should be
2581 * effectively equivalent, but much lighter weight.
2584 #include <linux/ramfs.h>
2586 static struct file_system_type tmpfs_fs_type = {
2587 .name = "tmpfs",
2588 .get_sb = ramfs_get_sb,
2589 .kill_sb = kill_litter_super,
2592 int __init init_tmpfs(void)
2594 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2596 shm_mnt = kern_mount(&tmpfs_fs_type);
2597 BUG_ON(IS_ERR(shm_mnt));
2599 return 0;
2602 int shmem_unuse(swp_entry_t entry, struct page *page)
2604 return 0;
2607 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2609 return 0;
2612 #define shmem_vm_ops generic_file_vm_ops
2613 #define shmem_file_operations ramfs_file_operations
2614 #define shmem_get_inode(sb, mode, dev, flags) ramfs_get_inode(sb, mode, dev)
2615 #define shmem_acct_size(flags, size) 0
2616 #define shmem_unacct_size(flags, size) do {} while (0)
2617 #define SHMEM_MAX_BYTES MAX_LFS_FILESIZE
2619 #endif /* CONFIG_SHMEM */
2621 /* common code */
2624 * shmem_file_setup - get an unlinked file living in tmpfs
2625 * @name: name for dentry (to be seen in /proc/<pid>/maps
2626 * @size: size to be set for the file
2627 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2629 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2631 int error;
2632 struct file *file;
2633 struct inode *inode;
2634 struct path path;
2635 struct dentry *root;
2636 struct qstr this;
2638 if (IS_ERR(shm_mnt))
2639 return (void *)shm_mnt;
2641 if (size < 0 || size > SHMEM_MAX_BYTES)
2642 return ERR_PTR(-EINVAL);
2644 if (shmem_acct_size(flags, size))
2645 return ERR_PTR(-ENOMEM);
2647 error = -ENOMEM;
2648 this.name = name;
2649 this.len = strlen(name);
2650 this.hash = 0; /* will go */
2651 root = shm_mnt->mnt_root;
2652 path.dentry = d_alloc(root, &this);
2653 if (!path.dentry)
2654 goto put_memory;
2655 path.mnt = mntget(shm_mnt);
2657 error = -ENOSPC;
2658 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0, flags);
2659 if (!inode)
2660 goto put_dentry;
2662 d_instantiate(path.dentry, inode);
2663 inode->i_size = size;
2664 inode->i_nlink = 0; /* It is unlinked */
2665 #ifndef CONFIG_MMU
2666 error = ramfs_nommu_expand_for_mapping(inode, size);
2667 if (error)
2668 goto put_dentry;
2669 #endif
2671 error = -ENFILE;
2672 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2673 &shmem_file_operations);
2674 if (!file)
2675 goto put_dentry;
2677 return file;
2679 put_dentry:
2680 path_put(&path);
2681 put_memory:
2682 shmem_unacct_size(flags, size);
2683 return ERR_PTR(error);
2685 EXPORT_SYMBOL_GPL(shmem_file_setup);
2688 * shmem_zero_setup - setup a shared anonymous mapping
2689 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2691 int shmem_zero_setup(struct vm_area_struct *vma)
2693 struct file *file;
2694 loff_t size = vma->vm_end - vma->vm_start;
2696 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2697 if (IS_ERR(file))
2698 return PTR_ERR(file);
2700 if (vma->vm_file)
2701 fput(vma->vm_file);
2702 vma->vm_file = file;
2703 vma->vm_ops = &shmem_vm_ops;
2704 return 0;