blob: 7ec78e24a30dcc549670b88643078509bbd12e4d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Resizable virtual memory filesystem for Linux.
3 *
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.
Hugh Dickins0edd73b2005-06-21 17:15:04 -07009 * Copyright (C) 2002-2005 Hugh Dickins.
10 * Copyright (C) 2002-2005 VERITAS Software Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Copyright (C) 2004 Andi Kleen, SuSE Labs
12 *
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>
16 *
Matt Mackall853ac432009-01-06 14:40:20 -080017 * tiny-shmem:
18 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
19 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070020 * This file is released under the GPL.
21 */
22
Matt Mackall853ac432009-01-06 14:40:20 -080023#include <linux/fs.h>
24#include <linux/init.h>
25#include <linux/vfs.h>
26#include <linux/mount.h>
27#include <linux/file.h>
28#include <linux/mm.h>
29#include <linux/module.h>
30#include <linux/swap.h>
Mimi Zohared850a52009-02-10 23:01:19 -050031#include <linux/ima.h>
Matt Mackall853ac432009-01-06 14:40:20 -080032
33static struct vfsmount *shm_mnt;
34
35#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070036/*
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.
40 */
41
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070042#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070043#include <linux/exportfs.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070044#include <linux/generic_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/pagemap.h>
47#include <linux/string.h>
48#include <linux/slab.h>
49#include <linux/backing-dev.h>
50#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/writeback.h>
52#include <linux/vfs.h>
53#include <linux/blkdev.h>
54#include <linux/security.h>
55#include <linux/swapops.h>
56#include <linux/mempolicy.h>
57#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000058#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070059#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070060#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080061#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040062#include <linux/magic.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070063
Linus Torvalds1da177e2005-04-16 15:20:36 -070064#include <asm/uaccess.h>
65#include <asm/div64.h>
66#include <asm/pgtable.h>
67
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
69#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
70#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
71
72#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
73#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
74
75#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
76
77/* info->flags needs VM_flags to handle pagein/truncate races efficiently */
78#define SHMEM_PAGEIN VM_READ
79#define SHMEM_TRUNCATE VM_WRITE
80
81/* Definition to limit shmem_truncate's steps between cond_rescheds */
82#define LATENCY_LIMIT 64
83
84/* Pretend that each entry is of this size in directory's i_size */
85#define BOGO_DIRENT_SIZE 20
86
Linus Torvalds1da177e2005-04-16 15:20:36 -070087/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
88enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 SGP_READ, /* don't exceed i_size, don't allocate page */
90 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -080091 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 SGP_WRITE, /* may exceed i_size, may allocate page */
93};
94
Andrew Mortonb76db732008-02-08 04:21:49 -080095#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080096static unsigned long shmem_default_max_blocks(void)
97{
98 return totalram_pages / 2;
99}
100
101static unsigned long shmem_default_max_inodes(void)
102{
103 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
104}
Andrew Mortonb76db732008-02-08 04:21:49 -0800105#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800106
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107static int shmem_getpage(struct inode *inode, unsigned long idx,
108 struct page **pagep, enum sgp_type sgp, int *type);
109
Al Viro6daa0e22005-10-21 03:18:50 -0400110static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111{
112 /*
113 * The above definition of ENTRIES_PER_PAGE, and the use of
114 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
115 * might be reconsidered if it ever diverges from PAGE_SIZE.
Mel Gorman769848c2007-07-17 04:03:05 -0700116 *
Mel Gormane12ba742007-10-16 01:25:52 -0700117 * Mobility flags are masked out as swap vectors cannot move
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 */
Mel Gormane12ba742007-10-16 01:25:52 -0700119 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
Mel Gorman769848c2007-07-17 04:03:05 -0700120 PAGE_CACHE_SHIFT-PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121}
122
123static inline void shmem_dir_free(struct page *page)
124{
125 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
126}
127
128static struct page **shmem_dir_map(struct page *page)
129{
130 return (struct page **)kmap_atomic(page, KM_USER0);
131}
132
133static inline void shmem_dir_unmap(struct page **dir)
134{
135 kunmap_atomic(dir, KM_USER0);
136}
137
138static swp_entry_t *shmem_swp_map(struct page *page)
139{
140 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
141}
142
143static inline void shmem_swp_balance_unmap(void)
144{
145 /*
146 * When passing a pointer to an i_direct entry, to code which
147 * also handles indirect entries and so will shmem_swp_unmap,
148 * we must arrange for the preempt count to remain in balance.
149 * What kmap_atomic of a lowmem page does depends on config
150 * and architecture, so pretend to kmap_atomic some lowmem page.
151 */
152 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
153}
154
155static inline void shmem_swp_unmap(swp_entry_t *entry)
156{
157 kunmap_atomic(entry, KM_USER1);
158}
159
160static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
161{
162 return sb->s_fs_info;
163}
164
165/*
166 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
167 * for shared memory and for shared anonymous (/dev/zero) mappings
168 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
169 * consistent with the pre-accounting of private mappings ...
170 */
171static inline int shmem_acct_size(unsigned long flags, loff_t size)
172{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000173 return (flags & VM_NORESERVE) ?
174 0 : security_vm_enough_memory_kern(VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175}
176
177static inline void shmem_unacct_size(unsigned long flags, loff_t size)
178{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000179 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 vm_unacct_memory(VM_ACCT(size));
181}
182
183/*
184 * ... whereas tmpfs objects are accounted incrementally as
185 * pages are allocated, in order to allow huge sparse files.
186 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
187 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
188 */
189static inline int shmem_acct_block(unsigned long flags)
190{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000191 return (flags & VM_NORESERVE) ?
192 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193}
194
195static inline void shmem_unacct_blocks(unsigned long flags, long pages)
196{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000197 if (flags & VM_NORESERVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
199}
200
Hugh Dickins759b9772007-03-05 00:30:28 -0800201static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700202static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800203static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800204static const struct inode_operations shmem_inode_operations;
205static const struct inode_operations shmem_dir_inode_operations;
206static const struct inode_operations shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207static struct vm_operations_struct shmem_vm_ops;
208
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700209static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 .ra_pages = 0, /* No readahead */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700211 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 .unplug_io_fn = default_unplug_io_fn,
213};
214
215static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800216static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
218static void shmem_free_blocks(struct inode *inode, long pages)
219{
220 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700221 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 spin_lock(&sbinfo->stat_lock);
223 sbinfo->free_blocks += pages;
224 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
225 spin_unlock(&sbinfo->stat_lock);
226 }
227}
228
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800229static int shmem_reserve_inode(struct super_block *sb)
230{
231 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
232 if (sbinfo->max_inodes) {
233 spin_lock(&sbinfo->stat_lock);
234 if (!sbinfo->free_inodes) {
235 spin_unlock(&sbinfo->stat_lock);
236 return -ENOSPC;
237 }
238 sbinfo->free_inodes--;
239 spin_unlock(&sbinfo->stat_lock);
240 }
241 return 0;
242}
243
244static void shmem_free_inode(struct super_block *sb)
245{
246 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
247 if (sbinfo->max_inodes) {
248 spin_lock(&sbinfo->stat_lock);
249 sbinfo->free_inodes++;
250 spin_unlock(&sbinfo->stat_lock);
251 }
252}
253
Randy Dunlap46711812008-03-19 17:00:41 -0700254/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 * shmem_recalc_inode - recalculate the size of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 * @inode: inode to recalc
257 *
258 * We have to calculate the free blocks since the mm can drop
259 * undirtied hole pages behind our back.
260 *
261 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
262 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
263 *
264 * It has to be called with the spinlock held.
265 */
266static void shmem_recalc_inode(struct inode *inode)
267{
268 struct shmem_inode_info *info = SHMEM_I(inode);
269 long freed;
270
271 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
272 if (freed > 0) {
273 info->alloced -= freed;
274 shmem_unacct_blocks(info->flags, freed);
275 shmem_free_blocks(inode, freed);
276 }
277}
278
Randy Dunlap46711812008-03-19 17:00:41 -0700279/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 * shmem_swp_entry - find the swap vector position in the info structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 * @info: info structure for the inode
282 * @index: index of the page to find
283 * @page: optional page to add to the structure. Has to be preset to
284 * all zeros
285 *
286 * If there is no space allocated yet it will return NULL when
287 * page is NULL, else it will use the page for the needed block,
288 * setting it to NULL on return to indicate that it has been used.
289 *
290 * The swap vector is organized the following way:
291 *
292 * There are SHMEM_NR_DIRECT entries directly stored in the
293 * shmem_inode_info structure. So small files do not need an addional
294 * allocation.
295 *
296 * For pages with index > SHMEM_NR_DIRECT there is the pointer
297 * i_indirect which points to a page which holds in the first half
298 * doubly indirect blocks, in the second half triple indirect blocks:
299 *
300 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
301 * following layout (for SHMEM_NR_DIRECT == 16):
302 *
303 * i_indirect -> dir --> 16-19
304 * | +-> 20-23
305 * |
306 * +-->dir2 --> 24-27
307 * | +-> 28-31
308 * | +-> 32-35
309 * | +-> 36-39
310 * |
311 * +-->dir3 --> 40-43
312 * +-> 44-47
313 * +-> 48-51
314 * +-> 52-55
315 */
316static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
317{
318 unsigned long offset;
319 struct page **dir;
320 struct page *subdir;
321
322 if (index < SHMEM_NR_DIRECT) {
323 shmem_swp_balance_unmap();
324 return info->i_direct+index;
325 }
326 if (!info->i_indirect) {
327 if (page) {
328 info->i_indirect = *page;
329 *page = NULL;
330 }
331 return NULL; /* need another page */
332 }
333
334 index -= SHMEM_NR_DIRECT;
335 offset = index % ENTRIES_PER_PAGE;
336 index /= ENTRIES_PER_PAGE;
337 dir = shmem_dir_map(info->i_indirect);
338
339 if (index >= ENTRIES_PER_PAGE/2) {
340 index -= ENTRIES_PER_PAGE/2;
341 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
342 index %= ENTRIES_PER_PAGE;
343 subdir = *dir;
344 if (!subdir) {
345 if (page) {
346 *dir = *page;
347 *page = NULL;
348 }
349 shmem_dir_unmap(dir);
350 return NULL; /* need another page */
351 }
352 shmem_dir_unmap(dir);
353 dir = shmem_dir_map(subdir);
354 }
355
356 dir += index;
357 subdir = *dir;
358 if (!subdir) {
359 if (!page || !(subdir = *page)) {
360 shmem_dir_unmap(dir);
361 return NULL; /* need a page */
362 }
363 *dir = subdir;
364 *page = NULL;
365 }
366 shmem_dir_unmap(dir);
367 return shmem_swp_map(subdir) + offset;
368}
369
370static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
371{
372 long incdec = value? 1: -1;
373
374 entry->val = value;
375 info->swapped += incdec;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700376 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
377 struct page *page = kmap_atomic_to_page(entry);
378 set_page_private(page, page_private(page) + incdec);
379 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380}
381
Randy Dunlap46711812008-03-19 17:00:41 -0700382/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 * shmem_swp_alloc - get the position of the swap entry for the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 * @info: info structure for the inode
385 * @index: index of the page to find
386 * @sgp: check and recheck i_size? skip allocation?
Randy Dunlap46711812008-03-19 17:00:41 -0700387 *
388 * If the entry does not exist, allocate it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 */
390static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
391{
392 struct inode *inode = &info->vfs_inode;
393 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
394 struct page *page = NULL;
395 swp_entry_t *entry;
396
397 if (sgp != SGP_WRITE &&
398 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
399 return ERR_PTR(-EINVAL);
400
401 while (!(entry = shmem_swp_entry(info, index, &page))) {
402 if (sgp == SGP_READ)
403 return shmem_swp_map(ZERO_PAGE(0));
404 /*
405 * Test free_blocks against 1 not 0, since we have 1 data
406 * page (and perhaps indirect index pages) yet to allocate:
407 * a waste to allocate index if we cannot allocate data.
408 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700409 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 spin_lock(&sbinfo->stat_lock);
411 if (sbinfo->free_blocks <= 1) {
412 spin_unlock(&sbinfo->stat_lock);
413 return ERR_PTR(-ENOSPC);
414 }
415 sbinfo->free_blocks--;
416 inode->i_blocks += BLOCKS_PER_PAGE;
417 spin_unlock(&sbinfo->stat_lock);
418 }
419
420 spin_unlock(&info->lock);
Mel Gorman769848c2007-07-17 04:03:05 -0700421 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700422 if (page)
423 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 spin_lock(&info->lock);
425
426 if (!page) {
427 shmem_free_blocks(inode, 1);
428 return ERR_PTR(-ENOMEM);
429 }
430 if (sgp != SGP_WRITE &&
431 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
432 entry = ERR_PTR(-EINVAL);
433 break;
434 }
435 if (info->next_index <= index)
436 info->next_index = index + 1;
437 }
438 if (page) {
439 /* another task gave its page, or truncated the file */
440 shmem_free_blocks(inode, 1);
441 shmem_dir_free(page);
442 }
443 if (info->next_index <= index && !IS_ERR(entry))
444 info->next_index = index + 1;
445 return entry;
446}
447
Randy Dunlap46711812008-03-19 17:00:41 -0700448/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 * shmem_free_swp - free some swap entries in a directory
Hugh Dickins1ae70002007-03-29 01:20:36 -0700450 * @dir: pointer to the directory
451 * @edir: pointer after last entry of the directory
452 * @punch_lock: pointer to spinlock when needed for the holepunch case
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 */
Hugh Dickins1ae70002007-03-29 01:20:36 -0700454static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
455 spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456{
Hugh Dickins1ae70002007-03-29 01:20:36 -0700457 spinlock_t *punch_unlock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 swp_entry_t *ptr;
459 int freed = 0;
460
461 for (ptr = dir; ptr < edir; ptr++) {
462 if (ptr->val) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700463 if (unlikely(punch_lock)) {
464 punch_unlock = punch_lock;
465 punch_lock = NULL;
466 spin_lock(punch_unlock);
467 if (!ptr->val)
468 continue;
469 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 free_swap_and_cache(*ptr);
471 *ptr = (swp_entry_t){0};
472 freed++;
473 }
474 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700475 if (punch_unlock)
476 spin_unlock(punch_unlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 return freed;
478}
479
Hugh Dickins1ae70002007-03-29 01:20:36 -0700480static int shmem_map_and_free_swp(struct page *subdir, int offset,
481 int limit, struct page ***dir, spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482{
483 swp_entry_t *ptr;
484 int freed = 0;
485
486 ptr = shmem_swp_map(subdir);
487 for (; offset < limit; offset += LATENCY_LIMIT) {
488 int size = limit - offset;
489 if (size > LATENCY_LIMIT)
490 size = LATENCY_LIMIT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700491 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
492 punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 if (need_resched()) {
494 shmem_swp_unmap(ptr);
495 if (*dir) {
496 shmem_dir_unmap(*dir);
497 *dir = NULL;
498 }
499 cond_resched();
500 ptr = shmem_swp_map(subdir);
501 }
502 }
503 shmem_swp_unmap(ptr);
504 return freed;
505}
506
507static void shmem_free_pages(struct list_head *next)
508{
509 struct page *page;
510 int freed = 0;
511
512 do {
513 page = container_of(next, struct page, lru);
514 next = next->next;
515 shmem_dir_free(page);
516 freed++;
517 if (freed >= LATENCY_LIMIT) {
518 cond_resched();
519 freed = 0;
520 }
521 } while (next);
522}
523
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800524static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
526 struct shmem_inode_info *info = SHMEM_I(inode);
527 unsigned long idx;
528 unsigned long size;
529 unsigned long limit;
530 unsigned long stage;
531 unsigned long diroff;
532 struct page **dir;
533 struct page *topdir;
534 struct page *middir;
535 struct page *subdir;
536 swp_entry_t *ptr;
537 LIST_HEAD(pages_to_free);
538 long nr_pages_to_free = 0;
539 long nr_swaps_freed = 0;
540 int offset;
541 int freed;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700542 int punch_hole;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700543 spinlock_t *needs_lock;
544 spinlock_t *punch_lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700545 unsigned long upper_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
547 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800548 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 if (idx >= info->next_index)
550 return;
551
552 spin_lock(&info->lock);
553 info->flags |= SHMEM_TRUNCATE;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800554 if (likely(end == (loff_t) -1)) {
555 limit = info->next_index;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700556 upper_limit = SHMEM_MAX_INDEX;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800557 info->next_index = idx;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700558 needs_lock = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700559 punch_hole = 0;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800560 } else {
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700561 if (end + 1 >= inode->i_size) { /* we may free a little more */
562 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
563 PAGE_CACHE_SHIFT;
564 upper_limit = SHMEM_MAX_INDEX;
565 } else {
566 limit = (end + 1) >> PAGE_CACHE_SHIFT;
567 upper_limit = limit;
568 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700569 needs_lock = &info->lock;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800570 punch_hole = 1;
571 }
572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 topdir = info->i_indirect;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800574 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 info->i_indirect = NULL;
576 nr_pages_to_free++;
577 list_add(&topdir->lru, &pages_to_free);
578 }
579 spin_unlock(&info->lock);
580
581 if (info->swapped && idx < SHMEM_NR_DIRECT) {
582 ptr = info->i_direct;
583 size = limit;
584 if (size > SHMEM_NR_DIRECT)
585 size = SHMEM_NR_DIRECT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700586 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 }
Badari Pulavarty92a3d032006-12-22 01:06:23 -0800588
589 /*
590 * If there are no indirect blocks or we are punching a hole
591 * below indirect blocks, nothing to be done.
592 */
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700593 if (!topdir || limit <= SHMEM_NR_DIRECT)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 goto done2;
595
Hugh Dickins1ae70002007-03-29 01:20:36 -0700596 /*
597 * The truncation case has already dropped info->lock, and we're safe
598 * because i_size and next_index have already been lowered, preventing
599 * access beyond. But in the punch_hole case, we still need to take
600 * the lock when updating the swap directory, because there might be
601 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
602 * shmem_writepage. However, whenever we find we can remove a whole
603 * directory page (not at the misaligned start or end of the range),
604 * we first NULLify its pointer in the level above, and then have no
605 * need to take the lock when updating its contents: needs_lock and
606 * punch_lock (either pointing to info->lock or NULL) manage this.
607 */
608
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700609 upper_limit -= SHMEM_NR_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 limit -= SHMEM_NR_DIRECT;
611 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
612 offset = idx % ENTRIES_PER_PAGE;
613 idx -= offset;
614
615 dir = shmem_dir_map(topdir);
616 stage = ENTRIES_PER_PAGEPAGE/2;
617 if (idx < ENTRIES_PER_PAGEPAGE/2) {
618 middir = topdir;
619 diroff = idx/ENTRIES_PER_PAGE;
620 } else {
621 dir += ENTRIES_PER_PAGE/2;
622 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
623 while (stage <= idx)
624 stage += ENTRIES_PER_PAGEPAGE;
625 middir = *dir;
626 if (*dir) {
627 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
628 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700629 if (!diroff && !offset && upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700630 if (needs_lock) {
631 spin_lock(needs_lock);
632 *dir = NULL;
633 spin_unlock(needs_lock);
634 needs_lock = NULL;
635 } else
636 *dir = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 nr_pages_to_free++;
638 list_add(&middir->lru, &pages_to_free);
639 }
640 shmem_dir_unmap(dir);
641 dir = shmem_dir_map(middir);
642 } else {
643 diroff = 0;
644 offset = 0;
645 idx = stage;
646 }
647 }
648
649 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
650 if (unlikely(idx == stage)) {
651 shmem_dir_unmap(dir);
652 dir = shmem_dir_map(topdir) +
653 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
654 while (!*dir) {
655 dir++;
656 idx += ENTRIES_PER_PAGEPAGE;
657 if (idx >= limit)
658 goto done1;
659 }
660 stage = idx + ENTRIES_PER_PAGEPAGE;
661 middir = *dir;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700662 if (punch_hole)
663 needs_lock = &info->lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700664 if (upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700665 if (needs_lock) {
666 spin_lock(needs_lock);
667 *dir = NULL;
668 spin_unlock(needs_lock);
669 needs_lock = NULL;
670 } else
671 *dir = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700672 nr_pages_to_free++;
673 list_add(&middir->lru, &pages_to_free);
674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 shmem_dir_unmap(dir);
676 cond_resched();
677 dir = shmem_dir_map(middir);
678 diroff = 0;
679 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700680 punch_lock = needs_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 subdir = dir[diroff];
Hugh Dickins1ae70002007-03-29 01:20:36 -0700682 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
683 if (needs_lock) {
684 spin_lock(needs_lock);
685 dir[diroff] = NULL;
686 spin_unlock(needs_lock);
687 punch_lock = NULL;
688 } else
689 dir[diroff] = NULL;
690 nr_pages_to_free++;
691 list_add(&subdir->lru, &pages_to_free);
692 }
693 if (subdir && page_private(subdir) /* has swap entries */) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 size = limit - idx;
695 if (size > ENTRIES_PER_PAGE)
696 size = ENTRIES_PER_PAGE;
697 freed = shmem_map_and_free_swp(subdir,
Hugh Dickins1ae70002007-03-29 01:20:36 -0700698 offset, size, &dir, punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 if (!dir)
700 dir = shmem_dir_map(middir);
701 nr_swaps_freed += freed;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700702 if (offset || punch_lock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 spin_lock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700704 set_page_private(subdir,
705 page_private(subdir) - freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 spin_unlock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700707 } else
708 BUG_ON(page_private(subdir) != freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700710 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 }
712done1:
713 shmem_dir_unmap(dir);
714done2:
715 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
716 /*
717 * Call truncate_inode_pages again: racing shmem_unuse_inode
718 * may have swizzled a page in from swap since vmtruncate or
719 * generic_delete_inode did it, before we lowered next_index.
720 * Also, though shmem_getpage checks i_size before adding to
721 * cache, no recheck after: so fix the narrow window there too.
Hugh Dickins16a10012007-03-29 01:20:37 -0700722 *
723 * Recalling truncate_inode_pages_range and unmap_mapping_range
724 * every time for punch_hole (which never got a chance to clear
725 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
726 * yet hardly ever necessary: try to optimize them out later.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800728 truncate_inode_pages_range(inode->i_mapping, start, end);
Hugh Dickins16a10012007-03-29 01:20:37 -0700729 if (punch_hole)
730 unmap_mapping_range(inode->i_mapping, start,
731 end - start, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 }
733
734 spin_lock(&info->lock);
735 info->flags &= ~SHMEM_TRUNCATE;
736 info->swapped -= nr_swaps_freed;
737 if (nr_pages_to_free)
738 shmem_free_blocks(inode, nr_pages_to_free);
739 shmem_recalc_inode(inode);
740 spin_unlock(&info->lock);
741
742 /*
743 * Empty swap vector directory pages to be freed?
744 */
745 if (!list_empty(&pages_to_free)) {
746 pages_to_free.prev->next = NULL;
747 shmem_free_pages(pages_to_free.next);
748 }
749}
750
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800751static void shmem_truncate(struct inode *inode)
752{
753 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
754}
755
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
757{
758 struct inode *inode = dentry->d_inode;
759 struct page *page = NULL;
760 int error;
761
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700762 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 if (attr->ia_size < inode->i_size) {
764 /*
765 * If truncating down to a partial page, then
766 * if that page is already allocated, hold it
767 * in memory until the truncation is over, so
768 * truncate_partial_page cannnot miss it were
769 * it assigned to swap.
770 */
771 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
772 (void) shmem_getpage(inode,
773 attr->ia_size>>PAGE_CACHE_SHIFT,
774 &page, SGP_READ, NULL);
Hugh Dickinsd3602442008-02-04 22:28:44 -0800775 if (page)
776 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 }
778 /*
779 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
780 * detect if any pages might have been added to cache
781 * after truncate_inode_pages. But we needn't bother
782 * if it's being fully truncated to zero-length: the
783 * nrpages check is efficient enough in that case.
784 */
785 if (attr->ia_size) {
786 struct shmem_inode_info *info = SHMEM_I(inode);
787 spin_lock(&info->lock);
788 info->flags &= ~SHMEM_PAGEIN;
789 spin_unlock(&info->lock);
790 }
791 }
792 }
793
794 error = inode_change_ok(inode, attr);
795 if (!error)
796 error = inode_setattr(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700797#ifdef CONFIG_TMPFS_POSIX_ACL
798 if (!error && (attr->ia_valid & ATTR_MODE))
799 error = generic_acl_chmod(inode, &shmem_acl_ops);
800#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 if (page)
802 page_cache_release(page);
803 return error;
804}
805
806static void shmem_delete_inode(struct inode *inode)
807{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 struct shmem_inode_info *info = SHMEM_I(inode);
809
810 if (inode->i_op->truncate == shmem_truncate) {
Mark Fashehfef26652005-09-09 13:01:31 -0700811 truncate_inode_pages(inode->i_mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 shmem_unacct_size(info->flags, inode->i_size);
813 inode->i_size = 0;
814 shmem_truncate(inode);
815 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800816 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800818 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 }
820 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700821 BUG_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800822 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 clear_inode(inode);
824}
825
826static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
827{
828 swp_entry_t *ptr;
829
830 for (ptr = dir; ptr < edir; ptr++) {
831 if (ptr->val == entry.val)
832 return ptr - dir;
833 }
834 return -1;
835}
836
837static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
838{
839 struct inode *inode;
840 unsigned long idx;
841 unsigned long size;
842 unsigned long limit;
843 unsigned long stage;
844 struct page **dir;
845 struct page *subdir;
846 swp_entry_t *ptr;
847 int offset;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800848 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
850 idx = 0;
851 ptr = info->i_direct;
852 spin_lock(&info->lock);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800853 if (!info->swapped) {
854 list_del_init(&info->swaplist);
855 goto lost2;
856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 limit = info->next_index;
858 size = limit;
859 if (size > SHMEM_NR_DIRECT)
860 size = SHMEM_NR_DIRECT;
861 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800862 if (offset >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 goto found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 if (!info->i_indirect)
865 goto lost2;
866
867 dir = shmem_dir_map(info->i_indirect);
868 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
869
870 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
871 if (unlikely(idx == stage)) {
872 shmem_dir_unmap(dir-1);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800873 if (cond_resched_lock(&info->lock)) {
874 /* check it has not been truncated */
875 if (limit > info->next_index) {
876 limit = info->next_index;
877 if (idx >= limit)
878 goto lost2;
879 }
880 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 dir = shmem_dir_map(info->i_indirect) +
882 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
883 while (!*dir) {
884 dir++;
885 idx += ENTRIES_PER_PAGEPAGE;
886 if (idx >= limit)
887 goto lost1;
888 }
889 stage = idx + ENTRIES_PER_PAGEPAGE;
890 subdir = *dir;
891 shmem_dir_unmap(dir);
892 dir = shmem_dir_map(subdir);
893 }
894 subdir = *dir;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700895 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 ptr = shmem_swp_map(subdir);
897 size = limit - idx;
898 if (size > ENTRIES_PER_PAGE)
899 size = ENTRIES_PER_PAGE;
900 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800901 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 if (offset >= 0) {
903 shmem_dir_unmap(dir);
904 goto found;
905 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 }
907 }
908lost1:
909 shmem_dir_unmap(dir-1);
910lost2:
911 spin_unlock(&info->lock);
912 return 0;
913found:
914 idx += offset;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800915 inode = igrab(&info->vfs_inode);
916 spin_unlock(&info->lock);
917
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800918 /*
919 * Move _head_ to start search for next from here.
920 * But be careful: shmem_delete_inode checks list_empty without taking
921 * mutex, and there's an instant in list_move_tail when info->swaplist
922 * would appear empty, if it were the only one on shmem_swaplist. We
923 * could avoid doing it if inode NULL; or use this minor optimization.
924 */
925 if (shmem_swaplist.next != &info->swaplist)
926 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800927 mutex_unlock(&shmem_swaplist_mutex);
928
929 error = 1;
930 if (!inode)
931 goto out;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800932 /*
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -0800933 * Charge page using GFP_KERNEL while we can wait.
934 * Charged back to the user(not to caller) when swap account is used.
935 * add_to_page_cache() will be called with GFP_NOWAIT.
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800936 */
Hugh Dickins82369552008-02-07 00:14:22 -0800937 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800938 if (error)
939 goto out;
Hugh Dickins82369552008-02-07 00:14:22 -0800940 error = radix_tree_preload(GFP_KERNEL);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700941 if (error) {
942 mem_cgroup_uncharge_cache_page(page);
943 goto out;
944 }
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800945 error = 1;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800946
947 spin_lock(&info->lock);
948 ptr = shmem_swp_entry(info, idx, NULL);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700949 if (ptr && ptr->val == entry.val) {
Nick Piggine2867812008-07-25 19:45:30 -0700950 error = add_to_page_cache_locked(page, inode->i_mapping,
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800951 idx, GFP_NOWAIT);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700952 /* does mem_cgroup_uncharge_cache_page on error */
953 } else /* we must compensate for our precharge above */
954 mem_cgroup_uncharge_cache_page(page);
955
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800956 if (error == -EEXIST) {
957 struct page *filepage = find_get_page(inode->i_mapping, idx);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800958 error = 1;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800959 if (filepage) {
960 /*
961 * There might be a more uptodate page coming down
962 * from a stacked writepage: forget our swappage if so.
963 */
964 if (PageUptodate(filepage))
965 error = 0;
966 page_cache_release(filepage);
967 }
968 }
969 if (!error) {
Hugh Dickins73b12622008-02-04 22:28:50 -0800970 delete_from_swap_cache(page);
971 set_page_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 info->flags |= SHMEM_PAGEIN;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800973 shmem_swp_set(info, ptr, 0);
974 swap_free(entry);
975 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800977 if (ptr)
978 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 spin_unlock(&info->lock);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800980 radix_tree_preload_end();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800981out:
982 unlock_page(page);
983 page_cache_release(page);
984 iput(inode); /* allows for NULL */
985 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986}
987
988/*
989 * shmem_unuse() search for an eventually swapped out shmem page.
990 */
991int shmem_unuse(swp_entry_t entry, struct page *page)
992{
993 struct list_head *p, *next;
994 struct shmem_inode_info *info;
995 int found = 0;
996
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800997 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 list_for_each_safe(p, next, &shmem_swaplist) {
999 info = list_entry(p, struct shmem_inode_info, swaplist);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001000 found = shmem_unuse_inode(info, entry, page);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001001 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001002 if (found)
1003 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001005 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001006out: return found; /* 0 or 1 or -ENOMEM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007}
1008
1009/*
1010 * Move the page from the page cache to the swap cache.
1011 */
1012static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1013{
1014 struct shmem_inode_info *info;
1015 swp_entry_t *entry, swap;
1016 struct address_space *mapping;
1017 unsigned long index;
1018 struct inode *inode;
1019
1020 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 mapping = page->mapping;
1022 index = page->index;
1023 inode = mapping->host;
1024 info = SHMEM_I(inode);
1025 if (info->flags & VM_LOCKED)
1026 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001027 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 goto redirty;
1029
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001030 /*
1031 * shmem_backing_dev_info's capabilities prevent regular writeback or
1032 * sync from ever calling shmem_writepage; but a stacking filesystem
1033 * may use the ->writepage of its underlying filesystem, in which case
1034 * tmpfs should write out to swap only in response to memory pressure,
1035 * and not for pdflush or sync. However, in those cases, we do still
1036 * want to check if there's a redundant swappage to be discarded.
1037 */
1038 if (wbc->for_reclaim)
1039 swap = get_swap_page();
1040 else
1041 swap.val = 0;
1042
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 if (index >= info->next_index) {
1045 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1046 goto unlock;
1047 }
1048 entry = shmem_swp_entry(info, index, NULL);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001049 if (entry->val) {
1050 /*
1051 * The more uptodate page coming down from a stacked
1052 * writepage should replace our old swappage.
1053 */
1054 free_swap_and_cache(*entry);
1055 shmem_swp_set(info, entry, 0);
1056 }
1057 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001059 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickins73b12622008-02-04 22:28:50 -08001060 remove_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 shmem_swp_set(info, entry, swap.val);
1062 shmem_swp_unmap(entry);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001063 if (list_empty(&info->swaplist))
1064 inode = igrab(inode);
1065 else
1066 inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 spin_unlock(&info->lock);
Hugh Dickins73b12622008-02-04 22:28:50 -08001068 swap_duplicate(swap);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001069 BUG_ON(page_mapped(page));
Hugh Dickins73b12622008-02-04 22:28:50 -08001070 page_cache_release(page); /* pagecache ref */
1071 set_page_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 unlock_page(page);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001073 if (inode) {
1074 mutex_lock(&shmem_swaplist_mutex);
1075 /* move instead of add in case we're racing */
1076 list_move_tail(&info->swaplist, &shmem_swaplist);
1077 mutex_unlock(&shmem_swaplist_mutex);
1078 iput(inode);
1079 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 return 0;
1081 }
1082
1083 shmem_swp_unmap(entry);
1084unlock:
1085 spin_unlock(&info->lock);
1086 swap_free(swap);
1087redirty:
1088 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001089 if (wbc->for_reclaim)
1090 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1091 unlock_page(page);
1092 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093}
1094
1095#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001096#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001097static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001098{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001099 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001100
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001101 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001102 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001103
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001104 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001105
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001106 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001107}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001108
1109static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1110{
1111 struct mempolicy *mpol = NULL;
1112 if (sbinfo->mpol) {
1113 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1114 mpol = sbinfo->mpol;
1115 mpol_get(mpol);
1116 spin_unlock(&sbinfo->stat_lock);
1117 }
1118 return mpol;
1119}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001120#endif /* CONFIG_TMPFS */
1121
Hugh Dickins02098fe2008-02-04 22:28:42 -08001122static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1123 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124{
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001125 struct mempolicy mpol, *spol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 struct vm_area_struct pvma;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001127 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001129 spol = mpol_cond_copy(&mpol,
1130 mpol_shared_policy_lookup(&info->policy, idx));
1131
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 /* Create a pseudo vma that just contains the policy */
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001133 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001135 pvma.vm_ops = NULL;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001136 pvma.vm_policy = spol;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001137 page = swapin_readahead(entry, gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 return page;
1139}
1140
Hugh Dickins02098fe2008-02-04 22:28:42 -08001141static struct page *shmem_alloc_page(gfp_t gfp,
1142 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143{
1144 struct vm_area_struct pvma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001146 /* Create a pseudo vma that just contains the policy */
1147 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001149 pvma.vm_ops = NULL;
1150 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001151
1152 /*
1153 * alloc_page_vma() will drop the shared policy reference
1154 */
1155 return alloc_page_vma(gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001157#else /* !CONFIG_NUMA */
1158#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001159static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001160{
1161}
1162#endif /* CONFIG_TMPFS */
1163
Hugh Dickins02098fe2008-02-04 22:28:42 -08001164static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1165 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166{
Hugh Dickins02098fe2008-02-04 22:28:42 -08001167 return swapin_readahead(entry, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168}
1169
Hugh Dickins02098fe2008-02-04 22:28:42 -08001170static inline struct page *shmem_alloc_page(gfp_t gfp,
1171 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172{
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001173 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001175#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001177#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1178static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1179{
1180 return NULL;
1181}
1182#endif
1183
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184/*
1185 * shmem_getpage - either get the page from swap or allocate a new one
1186 *
1187 * If we allocate a new one we do not mark it dirty. That's up to the
1188 * vm. If we swap it in we mark it dirty since we also free the swap
1189 * entry since a page cannot live in both the swap and page cache
1190 */
1191static int shmem_getpage(struct inode *inode, unsigned long idx,
1192 struct page **pagep, enum sgp_type sgp, int *type)
1193{
1194 struct address_space *mapping = inode->i_mapping;
1195 struct shmem_inode_info *info = SHMEM_I(inode);
1196 struct shmem_sb_info *sbinfo;
1197 struct page *filepage = *pagep;
1198 struct page *swappage;
1199 swp_entry_t *entry;
1200 swp_entry_t swap;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001201 gfp_t gfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 int error;
1203
1204 if (idx >= SHMEM_MAX_INDEX)
1205 return -EFBIG;
Nick Piggin54cb8822007-07-19 01:46:59 -07001206
1207 if (type)
Nick Piggin83c54072007-07-19 01:47:05 -07001208 *type = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07001209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 /*
1211 * Normally, filepage is NULL on entry, and either found
1212 * uptodate immediately, or allocated and zeroed, or read
1213 * in under swappage, which is then assigned to filepage.
Hugh Dickins5402b972008-02-04 22:28:44 -08001214 * But shmem_readpage (required for splice) passes in a locked
Hugh Dickinsae9764162007-06-04 10:00:39 +02001215 * filepage, which may be found not uptodate by other callers
1216 * too, and may need to be copied from the swappage read in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 */
1218repeat:
1219 if (!filepage)
1220 filepage = find_lock_page(mapping, idx);
1221 if (filepage && PageUptodate(filepage))
1222 goto done;
1223 error = 0;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001224 gfp = mapping_gfp_mask(mapping);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -08001225 if (!filepage) {
1226 /*
1227 * Try to preload while we can wait, to not make a habit of
1228 * draining atomic reserves; but don't latch on to this cpu.
1229 */
1230 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1231 if (error)
1232 goto failed;
1233 radix_tree_preload_end();
1234 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235
1236 spin_lock(&info->lock);
1237 shmem_recalc_inode(inode);
1238 entry = shmem_swp_alloc(info, idx, sgp);
1239 if (IS_ERR(entry)) {
1240 spin_unlock(&info->lock);
1241 error = PTR_ERR(entry);
1242 goto failed;
1243 }
1244 swap = *entry;
1245
1246 if (swap.val) {
1247 /* Look it up and read it in.. */
1248 swappage = lookup_swap_cache(swap);
1249 if (!swappage) {
1250 shmem_swp_unmap(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 /* here we actually do the io */
Nick Piggin83c54072007-07-19 01:47:05 -07001252 if (type && !(*type & VM_FAULT_MAJOR)) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001253 __count_vm_event(PGMAJFAULT);
Nick Piggin83c54072007-07-19 01:47:05 -07001254 *type |= VM_FAULT_MAJOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
Christoph Lameterf8891e52006-06-30 01:55:45 -07001256 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001257 swappage = shmem_swapin(swap, gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 if (!swappage) {
1259 spin_lock(&info->lock);
1260 entry = shmem_swp_alloc(info, idx, sgp);
1261 if (IS_ERR(entry))
1262 error = PTR_ERR(entry);
1263 else {
1264 if (entry->val == swap.val)
1265 error = -ENOMEM;
1266 shmem_swp_unmap(entry);
1267 }
1268 spin_unlock(&info->lock);
1269 if (error)
1270 goto failed;
1271 goto repeat;
1272 }
1273 wait_on_page_locked(swappage);
1274 page_cache_release(swappage);
1275 goto repeat;
1276 }
1277
1278 /* We have to do this with page locked to prevent races */
Nick Piggin529ae9a2008-08-02 12:01:03 +02001279 if (!trylock_page(swappage)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 shmem_swp_unmap(entry);
1281 spin_unlock(&info->lock);
1282 wait_on_page_locked(swappage);
1283 page_cache_release(swappage);
1284 goto repeat;
1285 }
1286 if (PageWriteback(swappage)) {
1287 shmem_swp_unmap(entry);
1288 spin_unlock(&info->lock);
1289 wait_on_page_writeback(swappage);
1290 unlock_page(swappage);
1291 page_cache_release(swappage);
1292 goto repeat;
1293 }
1294 if (!PageUptodate(swappage)) {
1295 shmem_swp_unmap(entry);
1296 spin_unlock(&info->lock);
1297 unlock_page(swappage);
1298 page_cache_release(swappage);
1299 error = -EIO;
1300 goto failed;
1301 }
1302
1303 if (filepage) {
1304 shmem_swp_set(info, entry, 0);
1305 shmem_swp_unmap(entry);
1306 delete_from_swap_cache(swappage);
1307 spin_unlock(&info->lock);
1308 copy_highpage(filepage, swappage);
1309 unlock_page(swappage);
1310 page_cache_release(swappage);
1311 flush_dcache_page(filepage);
1312 SetPageUptodate(filepage);
1313 set_page_dirty(filepage);
1314 swap_free(swap);
Nick Piggine2867812008-07-25 19:45:30 -07001315 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1316 idx, GFP_NOWAIT))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 info->flags |= SHMEM_PAGEIN;
1318 shmem_swp_set(info, entry, 0);
1319 shmem_swp_unmap(entry);
Hugh Dickins73b12622008-02-04 22:28:50 -08001320 delete_from_swap_cache(swappage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 spin_unlock(&info->lock);
1322 filepage = swappage;
Hugh Dickins73b12622008-02-04 22:28:50 -08001323 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 swap_free(swap);
1325 } else {
1326 shmem_swp_unmap(entry);
1327 spin_unlock(&info->lock);
Hugh Dickins82369552008-02-07 00:14:22 -08001328 if (error == -ENOMEM) {
1329 /* allow reclaim from this memory cgroup */
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001330 error = mem_cgroup_shrink_usage(swappage,
1331 current->mm,
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001332 gfp);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001333 if (error) {
1334 unlock_page(swappage);
1335 page_cache_release(swappage);
Hugh Dickins82369552008-02-07 00:14:22 -08001336 goto failed;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001337 }
Hugh Dickins82369552008-02-07 00:14:22 -08001338 }
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001339 unlock_page(swappage);
1340 page_cache_release(swappage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 goto repeat;
1342 }
1343 } else if (sgp == SGP_READ && !filepage) {
1344 shmem_swp_unmap(entry);
1345 filepage = find_get_page(mapping, idx);
1346 if (filepage &&
Nick Piggin529ae9a2008-08-02 12:01:03 +02001347 (!PageUptodate(filepage) || !trylock_page(filepage))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 spin_unlock(&info->lock);
1349 wait_on_page_locked(filepage);
1350 page_cache_release(filepage);
1351 filepage = NULL;
1352 goto repeat;
1353 }
1354 spin_unlock(&info->lock);
1355 } else {
1356 shmem_swp_unmap(entry);
1357 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001358 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 spin_lock(&sbinfo->stat_lock);
1360 if (sbinfo->free_blocks == 0 ||
1361 shmem_acct_block(info->flags)) {
1362 spin_unlock(&sbinfo->stat_lock);
1363 spin_unlock(&info->lock);
1364 error = -ENOSPC;
1365 goto failed;
1366 }
1367 sbinfo->free_blocks--;
1368 inode->i_blocks += BLOCKS_PER_PAGE;
1369 spin_unlock(&sbinfo->stat_lock);
1370 } else if (shmem_acct_block(info->flags)) {
1371 spin_unlock(&info->lock);
1372 error = -ENOSPC;
1373 goto failed;
1374 }
1375
1376 if (!filepage) {
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001377 int ret;
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001380 filepage = shmem_alloc_page(gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 if (!filepage) {
1382 shmem_unacct_blocks(info->flags, 1);
1383 shmem_free_blocks(inode, 1);
1384 error = -ENOMEM;
1385 goto failed;
1386 }
Rik van Rielb2e18532008-10-18 20:26:30 -07001387 SetPageSwapBacked(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
Hugh Dickins82369552008-02-07 00:14:22 -08001389 /* Precharge page while we can wait, compensate after */
1390 error = mem_cgroup_cache_charge(filepage, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08001391 GFP_KERNEL);
Hugh Dickins82369552008-02-07 00:14:22 -08001392 if (error) {
1393 page_cache_release(filepage);
1394 shmem_unacct_blocks(info->flags, 1);
1395 shmem_free_blocks(inode, 1);
1396 filepage = NULL;
1397 goto failed;
1398 }
1399
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 spin_lock(&info->lock);
1401 entry = shmem_swp_alloc(info, idx, sgp);
1402 if (IS_ERR(entry))
1403 error = PTR_ERR(entry);
1404 else {
1405 swap = *entry;
1406 shmem_swp_unmap(entry);
1407 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001408 ret = error || swap.val;
1409 if (ret)
1410 mem_cgroup_uncharge_cache_page(filepage);
1411 else
1412 ret = add_to_page_cache_lru(filepage, mapping,
1413 idx, GFP_NOWAIT);
1414 /*
1415 * At add_to_page_cache_lru() failure, uncharge will
1416 * be done automatically.
1417 */
1418 if (ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 spin_unlock(&info->lock);
1420 page_cache_release(filepage);
1421 shmem_unacct_blocks(info->flags, 1);
1422 shmem_free_blocks(inode, 1);
1423 filepage = NULL;
1424 if (error)
1425 goto failed;
1426 goto repeat;
1427 }
1428 info->flags |= SHMEM_PAGEIN;
1429 }
1430
1431 info->alloced++;
1432 spin_unlock(&info->lock);
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001433 clear_highpage(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 flush_dcache_page(filepage);
1435 SetPageUptodate(filepage);
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001436 if (sgp == SGP_DIRTY)
1437 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 }
1439done:
Hugh Dickinsd3602442008-02-04 22:28:44 -08001440 *pagep = filepage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 return 0;
1442
1443failed:
1444 if (*pagep != filepage) {
1445 unlock_page(filepage);
1446 page_cache_release(filepage);
1447 }
1448 return error;
1449}
1450
Nick Piggind0217ac2007-07-19 01:47:03 -07001451static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001453 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 int error;
Nick Piggind0217ac2007-07-19 01:47:03 -07001455 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
Nick Piggind0217ac2007-07-19 01:47:03 -07001457 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1458 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001459
Hugh Dickins27d54b32008-02-04 22:28:43 -08001460 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
Nick Piggind0217ac2007-07-19 01:47:03 -07001461 if (error)
1462 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463
Nick Piggin83c54072007-07-19 01:47:05 -07001464 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465}
1466
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467#ifdef CONFIG_NUMA
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001468static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001470 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1472}
1473
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001474static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1475 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001477 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 unsigned long idx;
1479
1480 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1481 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1482}
1483#endif
1484
1485int shmem_lock(struct file *file, int lock, struct user_struct *user)
1486{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001487 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 struct shmem_inode_info *info = SHMEM_I(inode);
1489 int retval = -ENOMEM;
1490
1491 spin_lock(&info->lock);
1492 if (lock && !(info->flags & VM_LOCKED)) {
1493 if (!user_shm_lock(inode->i_size, user))
1494 goto out_nomem;
1495 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001496 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 }
1498 if (!lock && (info->flags & VM_LOCKED) && user) {
1499 user_shm_unlock(inode->i_size, user);
1500 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001501 mapping_clear_unevictable(file->f_mapping);
1502 scan_mapping_unevictable_pages(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 }
1504 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001505
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506out_nomem:
1507 spin_unlock(&info->lock);
1508 return retval;
1509}
1510
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001511static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512{
1513 file_accessed(file);
1514 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001515 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 return 0;
1517}
1518
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001519static struct inode *shmem_get_inode(struct super_block *sb, int mode,
1520 dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521{
1522 struct inode *inode;
1523 struct shmem_inode_info *info;
1524 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1525
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001526 if (shmem_reserve_inode(sb))
1527 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
1529 inode = new_inode(sb);
1530 if (inode) {
1531 inode->i_mode = mode;
David Howells76aac0e2008-11-14 10:39:12 +11001532 inode->i_uid = current_fsuid();
1533 inode->i_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1536 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001537 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 info = SHMEM_I(inode);
1539 memset(info, 0, (char *)inode - (char *)info);
1540 spin_lock_init(&info->lock);
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001541 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 INIT_LIST_HEAD(&info->swaplist);
1543
1544 switch (mode & S_IFMT) {
1545 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001546 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 init_special_inode(inode, mode, dev);
1548 break;
1549 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001550 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 inode->i_op = &shmem_inode_operations;
1552 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001553 mpol_shared_policy_init(&info->policy,
1554 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 break;
1556 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001557 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 /* Some things misbehave if size == 0 on a directory */
1559 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1560 inode->i_op = &shmem_dir_inode_operations;
1561 inode->i_fop = &simple_dir_operations;
1562 break;
1563 case S_IFLNK:
1564 /*
1565 * Must not load anything in the rbtree,
1566 * mpol_free_shared_policy will not be called.
1567 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001568 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 break;
1570 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001571 } else
1572 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 return inode;
1574}
1575
1576#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001577static const struct inode_operations shmem_symlink_inode_operations;
1578static const struct inode_operations shmem_symlink_inline_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
1580/*
Nick Piggin800d15a2007-10-16 01:25:03 -07001581 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
Hugh Dickinsae9764162007-06-04 10:00:39 +02001582 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1583 * below the loop driver, in the generic fashion that many filesystems support.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 */
Hugh Dickinsae9764162007-06-04 10:00:39 +02001585static int shmem_readpage(struct file *file, struct page *page)
1586{
1587 struct inode *inode = page->mapping->host;
1588 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1589 unlock_page(page);
1590 return error;
1591}
1592
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001594shmem_write_begin(struct file *file, struct address_space *mapping,
1595 loff_t pos, unsigned len, unsigned flags,
1596 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597{
Nick Piggin800d15a2007-10-16 01:25:03 -07001598 struct inode *inode = mapping->host;
1599 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1600 *pagep = NULL;
1601 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1602}
1603
1604static int
1605shmem_write_end(struct file *file, struct address_space *mapping,
1606 loff_t pos, unsigned len, unsigned copied,
1607 struct page *page, void *fsdata)
1608{
1609 struct inode *inode = mapping->host;
1610
Hugh Dickinsd3602442008-02-04 22:28:44 -08001611 if (pos + copied > inode->i_size)
1612 i_size_write(inode, pos + copied);
1613
1614 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001615 set_page_dirty(page);
1616 page_cache_release(page);
1617
Nick Piggin800d15a2007-10-16 01:25:03 -07001618 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619}
1620
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1622{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001623 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 struct address_space *mapping = inode->i_mapping;
1625 unsigned long index, offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001626 enum sgp_type sgp = SGP_READ;
1627
1628 /*
1629 * Might this read be for a stacking filesystem? Then when reading
1630 * holes of a sparse file, we actually need to allocate those pages,
1631 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1632 */
1633 if (segment_eq(get_fs(), KERNEL_DS))
1634 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
1636 index = *ppos >> PAGE_CACHE_SHIFT;
1637 offset = *ppos & ~PAGE_CACHE_MASK;
1638
1639 for (;;) {
1640 struct page *page = NULL;
1641 unsigned long end_index, nr, ret;
1642 loff_t i_size = i_size_read(inode);
1643
1644 end_index = i_size >> PAGE_CACHE_SHIFT;
1645 if (index > end_index)
1646 break;
1647 if (index == end_index) {
1648 nr = i_size & ~PAGE_CACHE_MASK;
1649 if (nr <= offset)
1650 break;
1651 }
1652
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001653 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 if (desc->error) {
1655 if (desc->error == -EINVAL)
1656 desc->error = 0;
1657 break;
1658 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001659 if (page)
1660 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
1662 /*
1663 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001664 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 */
1666 nr = PAGE_CACHE_SIZE;
1667 i_size = i_size_read(inode);
1668 end_index = i_size >> PAGE_CACHE_SHIFT;
1669 if (index == end_index) {
1670 nr = i_size & ~PAGE_CACHE_MASK;
1671 if (nr <= offset) {
1672 if (page)
1673 page_cache_release(page);
1674 break;
1675 }
1676 }
1677 nr -= offset;
1678
1679 if (page) {
1680 /*
1681 * If users can be writing to this page using arbitrary
1682 * virtual addresses, take care about potential aliasing
1683 * before reading the page on the kernel side.
1684 */
1685 if (mapping_writably_mapped(mapping))
1686 flush_dcache_page(page);
1687 /*
1688 * Mark the page accessed if we read the beginning.
1689 */
1690 if (!offset)
1691 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001692 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001694 page_cache_get(page);
1695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696
1697 /*
1698 * Ok, we have the page, and it's up-to-date, so
1699 * now we can copy it to user space...
1700 *
1701 * The actor routine returns how many bytes were actually used..
1702 * NOTE! This may not be the same as how much of a user buffer
1703 * we filled up (we may be padding etc), so we can only update
1704 * "pos" here (the actor routine has to update the user buffer
1705 * pointers and the remaining count).
1706 */
1707 ret = actor(desc, page, offset, nr);
1708 offset += ret;
1709 index += offset >> PAGE_CACHE_SHIFT;
1710 offset &= ~PAGE_CACHE_MASK;
1711
1712 page_cache_release(page);
1713 if (ret != nr || !desc->count)
1714 break;
1715
1716 cond_resched();
1717 }
1718
1719 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1720 file_accessed(filp);
1721}
1722
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001723static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1724 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725{
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001726 struct file *filp = iocb->ki_filp;
1727 ssize_t retval;
1728 unsigned long seg;
1729 size_t count;
1730 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001732 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1733 if (retval)
1734 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001736 for (seg = 0; seg < nr_segs; seg++) {
1737 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001739 desc.written = 0;
1740 desc.arg.buf = iov[seg].iov_base;
1741 desc.count = iov[seg].iov_len;
1742 if (desc.count == 0)
1743 continue;
1744 desc.error = 0;
1745 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1746 retval += desc.written;
1747 if (desc.error) {
1748 retval = retval ?: desc.error;
1749 break;
1750 }
1751 if (desc.count > 0)
1752 break;
1753 }
1754 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755}
1756
David Howells726c3342006-06-23 02:02:58 -07001757static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758{
David Howells726c3342006-06-23 02:02:58 -07001759 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
1761 buf->f_type = TMPFS_MAGIC;
1762 buf->f_bsize = PAGE_CACHE_SIZE;
1763 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001764 spin_lock(&sbinfo->stat_lock);
1765 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 buf->f_blocks = sbinfo->max_blocks;
1767 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001768 }
1769 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 buf->f_files = sbinfo->max_inodes;
1771 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 }
1773 /* else leave those fields 0 like simple_statfs */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001774 spin_unlock(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 return 0;
1776}
1777
1778/*
1779 * File creation. Allocate an inode, and we're done..
1780 */
1781static int
1782shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1783{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001784 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 int error = -ENOSPC;
1786
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001787 inode = shmem_get_inode(dir->i_sb, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 if (inode) {
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001789 error = security_inode_init_security(inode, dir, NULL, NULL,
1790 NULL);
1791 if (error) {
1792 if (error != -EOPNOTSUPP) {
1793 iput(inode);
1794 return error;
1795 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001796 }
1797 error = shmem_acl_init(inode, dir);
1798 if (error) {
1799 iput(inode);
1800 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001801 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 if (dir->i_mode & S_ISGID) {
1803 inode->i_gid = dir->i_gid;
1804 if (S_ISDIR(mode))
1805 inode->i_mode |= S_ISGID;
1806 }
1807 dir->i_size += BOGO_DIRENT_SIZE;
1808 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1809 d_instantiate(dentry, inode);
1810 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 }
1812 return error;
1813}
1814
1815static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1816{
1817 int error;
1818
1819 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1820 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001821 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 return 0;
1823}
1824
1825static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1826 struct nameidata *nd)
1827{
1828 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1829}
1830
1831/*
1832 * Link a file..
1833 */
1834static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1835{
1836 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001837 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838
1839 /*
1840 * No ordinary (disk based) filesystem counts links as inodes;
1841 * but each new link needs a new dentry, pinning lowmem, and
1842 * tmpfs dentries cannot be pruned until they are unlinked.
1843 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001844 ret = shmem_reserve_inode(inode->i_sb);
1845 if (ret)
1846 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
1848 dir->i_size += BOGO_DIRENT_SIZE;
1849 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001850 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 atomic_inc(&inode->i_count); /* New dentry reference */
1852 dget(dentry); /* Extra pinning count for the created dentry */
1853 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001854out:
1855 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856}
1857
1858static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1859{
1860 struct inode *inode = dentry->d_inode;
1861
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001862 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1863 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
1865 dir->i_size -= BOGO_DIRENT_SIZE;
1866 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001867 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 dput(dentry); /* Undo the count from "create" - this does all the work */
1869 return 0;
1870}
1871
1872static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1873{
1874 if (!simple_empty(dentry))
1875 return -ENOTEMPTY;
1876
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001877 drop_nlink(dentry->d_inode);
1878 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 return shmem_unlink(dir, dentry);
1880}
1881
1882/*
1883 * The VFS layer already does all the dentry stuff for rename,
1884 * we just have to decrement the usage count for the target if
1885 * it exists so that the VFS layer correctly free's it when it
1886 * gets overwritten.
1887 */
1888static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1889{
1890 struct inode *inode = old_dentry->d_inode;
1891 int they_are_dirs = S_ISDIR(inode->i_mode);
1892
1893 if (!simple_empty(new_dentry))
1894 return -ENOTEMPTY;
1895
1896 if (new_dentry->d_inode) {
1897 (void) shmem_unlink(new_dir, new_dentry);
1898 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001899 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001901 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001902 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 }
1904
1905 old_dir->i_size -= BOGO_DIRENT_SIZE;
1906 new_dir->i_size += BOGO_DIRENT_SIZE;
1907 old_dir->i_ctime = old_dir->i_mtime =
1908 new_dir->i_ctime = new_dir->i_mtime =
1909 inode->i_ctime = CURRENT_TIME;
1910 return 0;
1911}
1912
1913static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1914{
1915 int error;
1916 int len;
1917 struct inode *inode;
1918 struct page *page = NULL;
1919 char *kaddr;
1920 struct shmem_inode_info *info;
1921
1922 len = strlen(symname) + 1;
1923 if (len > PAGE_CACHE_SIZE)
1924 return -ENAMETOOLONG;
1925
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001926 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 if (!inode)
1928 return -ENOSPC;
1929
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001930 error = security_inode_init_security(inode, dir, NULL, NULL,
1931 NULL);
1932 if (error) {
1933 if (error != -EOPNOTSUPP) {
1934 iput(inode);
1935 return error;
1936 }
1937 error = 0;
1938 }
1939
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 info = SHMEM_I(inode);
1941 inode->i_size = len-1;
1942 if (len <= (char *)inode - (char *)info) {
1943 /* do it inline */
1944 memcpy(info, symname, len);
1945 inode->i_op = &shmem_symlink_inline_operations;
1946 } else {
1947 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1948 if (error) {
1949 iput(inode);
1950 return error;
1951 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001952 unlock_page(page);
Hugh Dickins14fcc232008-07-28 15:46:19 -07001953 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 inode->i_op = &shmem_symlink_inode_operations;
1955 kaddr = kmap_atomic(page, KM_USER0);
1956 memcpy(kaddr, symname, len);
1957 kunmap_atomic(kaddr, KM_USER0);
1958 set_page_dirty(page);
1959 page_cache_release(page);
1960 }
1961 if (dir->i_mode & S_ISGID)
1962 inode->i_gid = dir->i_gid;
1963 dir->i_size += BOGO_DIRENT_SIZE;
1964 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1965 d_instantiate(dentry, inode);
1966 dget(dentry);
1967 return 0;
1968}
1969
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001970static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971{
1972 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001973 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974}
1975
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001976static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977{
1978 struct page *page = NULL;
1979 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1980 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08001981 if (page)
1982 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001983 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984}
1985
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001986static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987{
1988 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001989 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 kunmap(page);
1991 mark_page_accessed(page);
1992 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 }
1994}
1995
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001996static const struct inode_operations shmem_symlink_inline_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 .readlink = generic_readlink,
1998 .follow_link = shmem_follow_link_inline,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999};
2000
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002001static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 .truncate = shmem_truncate,
2003 .readlink = generic_readlink,
2004 .follow_link = shmem_follow_link,
2005 .put_link = shmem_put_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006};
2007
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002008#ifdef CONFIG_TMPFS_POSIX_ACL
Randy Dunlap46711812008-03-19 17:00:41 -07002009/*
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002010 * Superblocks without xattr inode operations will get security.* xattr
2011 * support from the VFS "for free". As soon as we have any other xattrs
2012 * like ACLs, we also need to implement the security.* handlers at
2013 * filesystem level, though.
2014 */
2015
2016static size_t shmem_xattr_security_list(struct inode *inode, char *list,
2017 size_t list_len, const char *name,
2018 size_t name_len)
2019{
2020 return security_inode_listsecurity(inode, list, list_len);
2021}
2022
2023static int shmem_xattr_security_get(struct inode *inode, const char *name,
2024 void *buffer, size_t size)
2025{
2026 if (strcmp(name, "") == 0)
2027 return -EINVAL;
David P. Quigley42492592008-02-04 22:29:39 -08002028 return xattr_getsecurity(inode, name, buffer, size);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002029}
2030
2031static int shmem_xattr_security_set(struct inode *inode, const char *name,
2032 const void *value, size_t size, int flags)
2033{
2034 if (strcmp(name, "") == 0)
2035 return -EINVAL;
2036 return security_inode_setsecurity(inode, name, value, size, flags);
2037}
2038
Adrian Bunk1f370a22006-12-06 20:38:21 -08002039static struct xattr_handler shmem_xattr_security_handler = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002040 .prefix = XATTR_SECURITY_PREFIX,
2041 .list = shmem_xattr_security_list,
2042 .get = shmem_xattr_security_get,
2043 .set = shmem_xattr_security_set,
2044};
2045
2046static struct xattr_handler *shmem_xattr_handlers[] = {
2047 &shmem_xattr_acl_access_handler,
2048 &shmem_xattr_acl_default_handler,
2049 &shmem_xattr_security_handler,
2050 NULL
2051};
2052#endif
2053
David M. Grimes91828a42006-10-17 00:09:45 -07002054static struct dentry *shmem_get_parent(struct dentry *child)
2055{
2056 return ERR_PTR(-ESTALE);
2057}
2058
2059static int shmem_match(struct inode *ino, void *vfh)
2060{
2061 __u32 *fh = vfh;
2062 __u64 inum = fh[2];
2063 inum = (inum << 32) | fh[1];
2064 return ino->i_ino == inum && fh[0] == ino->i_generation;
2065}
2066
Christoph Hellwig480b1162007-10-21 16:42:13 -07002067static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2068 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002069{
David M. Grimes91828a42006-10-17 00:09:45 -07002070 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002071 struct dentry *dentry = NULL;
2072 u64 inum = fid->raw[2];
2073 inum = (inum << 32) | fid->raw[1];
David M. Grimes91828a42006-10-17 00:09:45 -07002074
Christoph Hellwig480b1162007-10-21 16:42:13 -07002075 if (fh_len < 3)
2076 return NULL;
2077
2078 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2079 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002080 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002081 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002082 iput(inode);
2083 }
2084
Christoph Hellwig480b1162007-10-21 16:42:13 -07002085 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002086}
2087
2088static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2089 int connectable)
2090{
2091 struct inode *inode = dentry->d_inode;
2092
2093 if (*len < 3)
2094 return 255;
2095
2096 if (hlist_unhashed(&inode->i_hash)) {
2097 /* Unfortunately insert_inode_hash is not idempotent,
2098 * so as we hash inodes here rather than at creation
2099 * time, we need a lock to ensure we only try
2100 * to do it once
2101 */
2102 static DEFINE_SPINLOCK(lock);
2103 spin_lock(&lock);
2104 if (hlist_unhashed(&inode->i_hash))
2105 __insert_inode_hash(inode,
2106 inode->i_ino + inode->i_generation);
2107 spin_unlock(&lock);
2108 }
2109
2110 fh[0] = inode->i_generation;
2111 fh[1] = inode->i_ino;
2112 fh[2] = ((__u64)inode->i_ino) >> 32;
2113
2114 *len = 3;
2115 return 1;
2116}
2117
Christoph Hellwig39655162007-10-21 16:42:17 -07002118static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002119 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002120 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002121 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002122};
2123
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002124static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2125 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126{
2127 char *this_char, *value, *rest;
2128
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002129 while (options != NULL) {
2130 this_char = options;
2131 for (;;) {
2132 /*
2133 * NUL-terminate this option: unfortunately,
2134 * mount options form a comma-separated list,
2135 * but mpol's nodelist may also contain commas.
2136 */
2137 options = strchr(options, ',');
2138 if (options == NULL)
2139 break;
2140 options++;
2141 if (!isdigit(*options)) {
2142 options[-1] = '\0';
2143 break;
2144 }
2145 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 if (!*this_char)
2147 continue;
2148 if ((value = strchr(this_char,'=')) != NULL) {
2149 *value++ = 0;
2150 } else {
2151 printk(KERN_ERR
2152 "tmpfs: No value for mount option '%s'\n",
2153 this_char);
2154 return 1;
2155 }
2156
2157 if (!strcmp(this_char,"size")) {
2158 unsigned long long size;
2159 size = memparse(value,&rest);
2160 if (*rest == '%') {
2161 size <<= PAGE_SHIFT;
2162 size *= totalram_pages;
2163 do_div(size, 100);
2164 rest++;
2165 }
2166 if (*rest)
2167 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002168 sbinfo->max_blocks =
2169 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002171 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 if (*rest)
2173 goto bad_val;
2174 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002175 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 if (*rest)
2177 goto bad_val;
2178 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002179 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002181 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 if (*rest)
2183 goto bad_val;
2184 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002185 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002187 sbinfo->uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 if (*rest)
2189 goto bad_val;
2190 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002191 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002193 sbinfo->gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 if (*rest)
2195 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002196 } else if (!strcmp(this_char,"mpol")) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002197 if (mpol_parse_str(value, &sbinfo->mpol, 1))
Robin Holt7339ff82006-01-14 13:20:48 -08002198 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 } else {
2200 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2201 this_char);
2202 return 1;
2203 }
2204 }
2205 return 0;
2206
2207bad_val:
2208 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2209 value, this_char);
2210 return 1;
2211
2212}
2213
2214static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2215{
2216 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002217 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002218 unsigned long blocks;
2219 unsigned long inodes;
2220 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002222 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002223 return error;
2224
2225 spin_lock(&sbinfo->stat_lock);
2226 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2227 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002228 if (config.max_blocks < blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002229 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002230 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002231 goto out;
2232 /*
2233 * Those tests also disallow limited->unlimited while any are in
2234 * use, so i_blocks will always be zero when max_blocks is zero;
2235 * but we must separately disallow unlimited->limited, because
2236 * in that case we have no record of how much is already in use.
2237 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002238 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002239 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002240 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002241 goto out;
2242
2243 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002244 sbinfo->max_blocks = config.max_blocks;
2245 sbinfo->free_blocks = config.max_blocks - blocks;
2246 sbinfo->max_inodes = config.max_inodes;
2247 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002248
2249 mpol_put(sbinfo->mpol);
2250 sbinfo->mpol = config.mpol; /* transfers initial ref */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002251out:
2252 spin_unlock(&sbinfo->stat_lock);
2253 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002255
2256static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2257{
2258 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2259
2260 if (sbinfo->max_blocks != shmem_default_max_blocks())
2261 seq_printf(seq, ",size=%luk",
2262 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2263 if (sbinfo->max_inodes != shmem_default_max_inodes())
2264 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2265 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2266 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2267 if (sbinfo->uid != 0)
2268 seq_printf(seq, ",uid=%u", sbinfo->uid);
2269 if (sbinfo->gid != 0)
2270 seq_printf(seq, ",gid=%u", sbinfo->gid);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002271 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002272 return 0;
2273}
2274#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275
2276static void shmem_put_super(struct super_block *sb)
2277{
2278 kfree(sb->s_fs_info);
2279 sb->s_fs_info = NULL;
2280}
2281
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282static int shmem_fill_super(struct super_block *sb,
2283 void *data, int silent)
2284{
2285 struct inode *inode;
2286 struct dentry *root;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002287 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002288 int err = -ENOMEM;
2289
2290 /* Round up to L1_CACHE_BYTES to resist false sharing */
2291 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2292 L1_CACHE_BYTES), GFP_KERNEL);
2293 if (!sbinfo)
2294 return -ENOMEM;
2295
2296 sbinfo->max_blocks = 0;
2297 sbinfo->max_inodes = 0;
2298 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002299 sbinfo->uid = current_fsuid();
2300 sbinfo->gid = current_fsgid();
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002301 sbinfo->mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002302 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002304#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 /*
2306 * Per default we only allow half of the physical ram per
2307 * tmpfs instance, limiting inodes to one per page of lowmem;
2308 * but the internal instance is left unlimited.
2309 */
2310 if (!(sb->s_flags & MS_NOUSER)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002311 sbinfo->max_blocks = shmem_default_max_blocks();
2312 sbinfo->max_inodes = shmem_default_max_inodes();
2313 if (shmem_parse_options(data, sbinfo, false)) {
2314 err = -EINVAL;
2315 goto failed;
2316 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 }
David M. Grimes91828a42006-10-17 00:09:45 -07002318 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319#else
2320 sb->s_flags |= MS_NOUSER;
2321#endif
2322
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002323 spin_lock_init(&sbinfo->stat_lock);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002324 sbinfo->free_blocks = sbinfo->max_blocks;
2325 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002326
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 sb->s_maxbytes = SHMEM_MAX_BYTES;
2328 sb->s_blocksize = PAGE_CACHE_SIZE;
2329 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2330 sb->s_magic = TMPFS_MAGIC;
2331 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002332 sb->s_time_gran = 1;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002333#ifdef CONFIG_TMPFS_POSIX_ACL
2334 sb->s_xattr = shmem_xattr_handlers;
2335 sb->s_flags |= MS_POSIXACL;
2336#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002337
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002338 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 if (!inode)
2340 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002341 inode->i_uid = sbinfo->uid;
2342 inode->i_gid = sbinfo->gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 root = d_alloc_root(inode);
2344 if (!root)
2345 goto failed_iput;
2346 sb->s_root = root;
2347 return 0;
2348
2349failed_iput:
2350 iput(inode);
2351failed:
2352 shmem_put_super(sb);
2353 return err;
2354}
2355
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002356static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357
2358static struct inode *shmem_alloc_inode(struct super_block *sb)
2359{
2360 struct shmem_inode_info *p;
Christoph Lametere94b1762006-12-06 20:33:17 -08002361 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 if (!p)
2363 return NULL;
2364 return &p->vfs_inode;
2365}
2366
2367static void shmem_destroy_inode(struct inode *inode)
2368{
2369 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2370 /* only struct inode is valid if it's an inline symlink */
2371 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2372 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002373 shmem_acl_destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2375}
2376
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002377static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378{
2379 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2380
Christoph Lametera35afb82007-05-16 22:10:57 -07002381 inode_init_once(&p->vfs_inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002382#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Lametera35afb82007-05-16 22:10:57 -07002383 p->i_acl = NULL;
2384 p->i_default_acl = NULL;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002385#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386}
2387
2388static int init_inodecache(void)
2389{
2390 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2391 sizeof(struct shmem_inode_info),
Alexey Dobriyan040b5c62007-10-16 23:26:10 -07002392 0, SLAB_PANIC, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 return 0;
2394}
2395
2396static void destroy_inodecache(void)
2397{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002398 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399}
2400
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002401static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002403 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404#ifdef CONFIG_TMPFS
Hugh Dickinsae9764162007-06-04 10:00:39 +02002405 .readpage = shmem_readpage,
Nick Piggin800d15a2007-10-16 01:25:03 -07002406 .write_begin = shmem_write_begin,
2407 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002409 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410};
2411
Helge Deller15ad7cd2006-12-06 20:40:36 -08002412static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 .mmap = shmem_mmap,
2414#ifdef CONFIG_TMPFS
2415 .llseek = generic_file_llseek,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002416 .read = do_sync_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002417 .write = do_sync_write,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002418 .aio_read = shmem_file_aio_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002419 .aio_write = generic_file_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 .fsync = simple_sync_file,
Hugh Dickinsae9764162007-06-04 10:00:39 +02002421 .splice_read = generic_file_splice_read,
2422 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423#endif
2424};
2425
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002426static const struct inode_operations shmem_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 .truncate = shmem_truncate,
2428 .setattr = shmem_notify_change,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002429 .truncate_range = shmem_truncate_range,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002430#ifdef CONFIG_TMPFS_POSIX_ACL
2431 .setxattr = generic_setxattr,
2432 .getxattr = generic_getxattr,
2433 .listxattr = generic_listxattr,
2434 .removexattr = generic_removexattr,
2435 .permission = shmem_permission,
2436#endif
2437
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438};
2439
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002440static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441#ifdef CONFIG_TMPFS
2442 .create = shmem_create,
2443 .lookup = simple_lookup,
2444 .link = shmem_link,
2445 .unlink = shmem_unlink,
2446 .symlink = shmem_symlink,
2447 .mkdir = shmem_mkdir,
2448 .rmdir = shmem_rmdir,
2449 .mknod = shmem_mknod,
2450 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002452#ifdef CONFIG_TMPFS_POSIX_ACL
2453 .setattr = shmem_notify_change,
2454 .setxattr = generic_setxattr,
2455 .getxattr = generic_getxattr,
2456 .listxattr = generic_listxattr,
2457 .removexattr = generic_removexattr,
2458 .permission = shmem_permission,
2459#endif
2460};
2461
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002462static const struct inode_operations shmem_special_inode_operations = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002463#ifdef CONFIG_TMPFS_POSIX_ACL
2464 .setattr = shmem_notify_change,
2465 .setxattr = generic_setxattr,
2466 .getxattr = generic_getxattr,
2467 .listxattr = generic_listxattr,
2468 .removexattr = generic_removexattr,
2469 .permission = shmem_permission,
2470#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471};
2472
Hugh Dickins759b9772007-03-05 00:30:28 -08002473static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 .alloc_inode = shmem_alloc_inode,
2475 .destroy_inode = shmem_destroy_inode,
2476#ifdef CONFIG_TMPFS
2477 .statfs = shmem_statfs,
2478 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002479 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480#endif
2481 .delete_inode = shmem_delete_inode,
2482 .drop_inode = generic_delete_inode,
2483 .put_super = shmem_put_super,
2484};
2485
2486static struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002487 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488#ifdef CONFIG_NUMA
2489 .set_policy = shmem_set_policy,
2490 .get_policy = shmem_get_policy,
2491#endif
2492};
2493
2494
David Howells454e2392006-06-23 02:02:57 -07002495static int shmem_get_sb(struct file_system_type *fs_type,
2496 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497{
David Howells454e2392006-06-23 02:02:57 -07002498 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499}
2500
2501static struct file_system_type tmpfs_fs_type = {
2502 .owner = THIS_MODULE,
2503 .name = "tmpfs",
2504 .get_sb = shmem_get_sb,
2505 .kill_sb = kill_litter_super,
2506};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507
2508static int __init init_tmpfs(void)
2509{
2510 int error;
2511
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002512 error = bdi_init(&shmem_backing_dev_info);
2513 if (error)
2514 goto out4;
2515
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 error = init_inodecache();
2517 if (error)
2518 goto out3;
2519
2520 error = register_filesystem(&tmpfs_fs_type);
2521 if (error) {
2522 printk(KERN_ERR "Could not register tmpfs\n");
2523 goto out2;
2524 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002525
Trond Myklebust1f5ce9e2006-06-09 09:34:16 -04002526 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 tmpfs_fs_type.name, NULL);
2528 if (IS_ERR(shm_mnt)) {
2529 error = PTR_ERR(shm_mnt);
2530 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2531 goto out1;
2532 }
2533 return 0;
2534
2535out1:
2536 unregister_filesystem(&tmpfs_fs_type);
2537out2:
2538 destroy_inodecache();
2539out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002540 bdi_destroy(&shmem_backing_dev_info);
2541out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 shm_mnt = ERR_PTR(error);
2543 return error;
2544}
Matt Mackall853ac432009-01-06 14:40:20 -08002545
2546#else /* !CONFIG_SHMEM */
2547
2548/*
2549 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2550 *
2551 * This is intended for small system where the benefits of the full
2552 * shmem code (swap-backed and resource-limited) are outweighed by
2553 * their complexity. On systems without swap this code should be
2554 * effectively equivalent, but much lighter weight.
2555 */
2556
2557#include <linux/ramfs.h>
2558
2559static struct file_system_type tmpfs_fs_type = {
2560 .name = "tmpfs",
2561 .get_sb = ramfs_get_sb,
2562 .kill_sb = kill_litter_super,
2563};
2564
2565static int __init init_tmpfs(void)
2566{
2567 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2568
2569 shm_mnt = kern_mount(&tmpfs_fs_type);
2570 BUG_ON(IS_ERR(shm_mnt));
2571
2572 return 0;
2573}
2574
2575int shmem_unuse(swp_entry_t entry, struct page *page)
2576{
2577 return 0;
2578}
2579
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002580#define shmem_vm_ops generic_file_vm_ops
2581#define shmem_file_operations ramfs_file_operations
2582#define shmem_get_inode(sb, mode, dev, flags) ramfs_get_inode(sb, mode, dev)
2583#define shmem_acct_size(flags, size) 0
2584#define shmem_unacct_size(flags, size) do {} while (0)
2585#define SHMEM_MAX_BYTES LLONG_MAX
Matt Mackall853ac432009-01-06 14:40:20 -08002586
2587#endif /* CONFIG_SHMEM */
2588
2589/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590
Randy Dunlap46711812008-03-19 17:00:41 -07002591/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 * shmem_file_setup - get an unlinked file living in tmpfs
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 * @name: name for dentry (to be seen in /proc/<pid>/maps
2594 * @size: size to be set for the file
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002595 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 */
2597struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2598{
2599 int error;
2600 struct file *file;
2601 struct inode *inode;
2602 struct dentry *dentry, *root;
2603 struct qstr this;
2604
2605 if (IS_ERR(shm_mnt))
2606 return (void *)shm_mnt;
2607
2608 if (size < 0 || size > SHMEM_MAX_BYTES)
2609 return ERR_PTR(-EINVAL);
2610
2611 if (shmem_acct_size(flags, size))
2612 return ERR_PTR(-ENOMEM);
2613
2614 error = -ENOMEM;
2615 this.name = name;
2616 this.len = strlen(name);
2617 this.hash = 0; /* will go */
2618 root = shm_mnt->mnt_root;
2619 dentry = d_alloc(root, &this);
2620 if (!dentry)
2621 goto put_memory;
2622
2623 error = -ENFILE;
2624 file = get_empty_filp();
2625 if (!file)
2626 goto put_dentry;
2627
2628 error = -ENOSPC;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002629 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 if (!inode)
2631 goto close_file;
2632
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 d_instantiate(dentry, inode);
2634 inode->i_size = size;
2635 inode->i_nlink = 0; /* It is unlinked */
Dave Hansence8d2cd2007-10-16 23:31:13 -07002636 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
Matt Mackall853ac432009-01-06 14:40:20 -08002637 &shmem_file_operations);
2638
2639#ifndef CONFIG_MMU
2640 error = ramfs_nommu_expand_for_mapping(inode, size);
2641 if (error)
2642 goto close_file;
2643#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 return file;
2645
2646close_file:
2647 put_filp(file);
2648put_dentry:
2649 dput(dentry);
2650put_memory:
2651 shmem_unacct_size(flags, size);
2652 return ERR_PTR(error);
2653}
Keith Packard395e0dd2008-06-20 00:08:06 -07002654EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655
Randy Dunlap46711812008-03-19 17:00:41 -07002656/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2659 */
2660int shmem_zero_setup(struct vm_area_struct *vma)
2661{
2662 struct file *file;
2663 loff_t size = vma->vm_end - vma->vm_start;
2664
2665 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2666 if (IS_ERR(file))
2667 return PTR_ERR(file);
2668
Mimi Zohar1df9f0a2009-02-04 09:07:02 -05002669 ima_shm_check(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 if (vma->vm_file)
2671 fput(vma->vm_file);
2672 vma->vm_file = file;
2673 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 return 0;
2675}
Matt Mackall853ac432009-01-06 14:40:20 -08002676
2677module_init(init_tmpfs)