blob: e2a6ae1a44e9a5e8b5e6594d11f76b11a02ea49f [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 *
17 * This file is released under the GPL.
18 */
19
20/*
21 * This virtual memory filesystem is heavily based on the ramfs. It
22 * extends ramfs by the ability to use swap and honor resource limits
23 * which makes it a completely usable filesystem.
24 */
25
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/module.h>
27#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/fs.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070029#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070030#include <linux/exportfs.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070031#include <linux/generic_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/mm.h>
33#include <linux/mman.h>
34#include <linux/file.h>
35#include <linux/swap.h>
36#include <linux/pagemap.h>
37#include <linux/string.h>
38#include <linux/slab.h>
39#include <linux/backing-dev.h>
40#include <linux/shmem_fs.h>
41#include <linux/mount.h>
42#include <linux/writeback.h>
43#include <linux/vfs.h>
44#include <linux/blkdev.h>
45#include <linux/security.h>
46#include <linux/swapops.h>
47#include <linux/mempolicy.h>
48#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000049#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070050#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070051#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080052#include <linux/seq_file.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070053
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include <asm/uaccess.h>
55#include <asm/div64.h>
56#include <asm/pgtable.h>
57
58/* This magic number is used in glibc for posix shared memory */
59#define TMPFS_MAGIC 0x01021994
60
61#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
62#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
63#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
64
65#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
66#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
67
68#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
69
70/* info->flags needs VM_flags to handle pagein/truncate races efficiently */
71#define SHMEM_PAGEIN VM_READ
72#define SHMEM_TRUNCATE VM_WRITE
73
74/* Definition to limit shmem_truncate's steps between cond_rescheds */
75#define LATENCY_LIMIT 64
76
77/* Pretend that each entry is of this size in directory's i_size */
78#define BOGO_DIRENT_SIZE 20
79
Linus Torvalds1da177e2005-04-16 15:20:36 -070080/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
81enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 SGP_READ, /* don't exceed i_size, don't allocate page */
83 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -080084 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 SGP_WRITE, /* may exceed i_size, may allocate page */
86};
87
Andrew Mortonb76db732008-02-08 04:21:49 -080088#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080089static unsigned long shmem_default_max_blocks(void)
90{
91 return totalram_pages / 2;
92}
93
94static unsigned long shmem_default_max_inodes(void)
95{
96 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
97}
Andrew Mortonb76db732008-02-08 04:21:49 -080098#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080099
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100static int shmem_getpage(struct inode *inode, unsigned long idx,
101 struct page **pagep, enum sgp_type sgp, int *type);
102
Al Viro6daa0e22005-10-21 03:18:50 -0400103static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
105 /*
106 * The above definition of ENTRIES_PER_PAGE, and the use of
107 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
108 * might be reconsidered if it ever diverges from PAGE_SIZE.
Mel Gorman769848c2007-07-17 04:03:05 -0700109 *
Mel Gormane12ba742007-10-16 01:25:52 -0700110 * Mobility flags are masked out as swap vectors cannot move
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 */
Mel Gormane12ba742007-10-16 01:25:52 -0700112 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
Mel Gorman769848c2007-07-17 04:03:05 -0700113 PAGE_CACHE_SHIFT-PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114}
115
116static inline void shmem_dir_free(struct page *page)
117{
118 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
119}
120
121static struct page **shmem_dir_map(struct page *page)
122{
123 return (struct page **)kmap_atomic(page, KM_USER0);
124}
125
126static inline void shmem_dir_unmap(struct page **dir)
127{
128 kunmap_atomic(dir, KM_USER0);
129}
130
131static swp_entry_t *shmem_swp_map(struct page *page)
132{
133 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
134}
135
136static inline void shmem_swp_balance_unmap(void)
137{
138 /*
139 * When passing a pointer to an i_direct entry, to code which
140 * also handles indirect entries and so will shmem_swp_unmap,
141 * we must arrange for the preempt count to remain in balance.
142 * What kmap_atomic of a lowmem page does depends on config
143 * and architecture, so pretend to kmap_atomic some lowmem page.
144 */
145 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
146}
147
148static inline void shmem_swp_unmap(swp_entry_t *entry)
149{
150 kunmap_atomic(entry, KM_USER1);
151}
152
153static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
154{
155 return sb->s_fs_info;
156}
157
158/*
159 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
160 * for shared memory and for shared anonymous (/dev/zero) mappings
161 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
162 * consistent with the pre-accounting of private mappings ...
163 */
164static inline int shmem_acct_size(unsigned long flags, loff_t size)
165{
166 return (flags & VM_ACCOUNT)?
167 security_vm_enough_memory(VM_ACCT(size)): 0;
168}
169
170static inline void shmem_unacct_size(unsigned long flags, loff_t size)
171{
172 if (flags & VM_ACCOUNT)
173 vm_unacct_memory(VM_ACCT(size));
174}
175
176/*
177 * ... whereas tmpfs objects are accounted incrementally as
178 * pages are allocated, in order to allow huge sparse files.
179 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
180 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
181 */
182static inline int shmem_acct_block(unsigned long flags)
183{
184 return (flags & VM_ACCOUNT)?
185 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
186}
187
188static inline void shmem_unacct_blocks(unsigned long flags, long pages)
189{
190 if (!(flags & VM_ACCOUNT))
191 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
192}
193
Hugh Dickins759b9772007-03-05 00:30:28 -0800194static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700195static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800196static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800197static const struct inode_operations shmem_inode_operations;
198static const struct inode_operations shmem_dir_inode_operations;
199static const struct inode_operations shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200static struct vm_operations_struct shmem_vm_ops;
201
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700202static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 .ra_pages = 0, /* No readahead */
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700204 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 .unplug_io_fn = default_unplug_io_fn,
206};
207
208static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800209static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
211static void shmem_free_blocks(struct inode *inode, long pages)
212{
213 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700214 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 spin_lock(&sbinfo->stat_lock);
216 sbinfo->free_blocks += pages;
217 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
218 spin_unlock(&sbinfo->stat_lock);
219 }
220}
221
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800222static int shmem_reserve_inode(struct super_block *sb)
223{
224 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
225 if (sbinfo->max_inodes) {
226 spin_lock(&sbinfo->stat_lock);
227 if (!sbinfo->free_inodes) {
228 spin_unlock(&sbinfo->stat_lock);
229 return -ENOSPC;
230 }
231 sbinfo->free_inodes--;
232 spin_unlock(&sbinfo->stat_lock);
233 }
234 return 0;
235}
236
237static void shmem_free_inode(struct super_block *sb)
238{
239 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
240 if (sbinfo->max_inodes) {
241 spin_lock(&sbinfo->stat_lock);
242 sbinfo->free_inodes++;
243 spin_unlock(&sbinfo->stat_lock);
244 }
245}
246
Randy Dunlap46711812008-03-19 17:00:41 -0700247/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 * shmem_recalc_inode - recalculate the size of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 * @inode: inode to recalc
250 *
251 * We have to calculate the free blocks since the mm can drop
252 * undirtied hole pages behind our back.
253 *
254 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
255 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
256 *
257 * It has to be called with the spinlock held.
258 */
259static void shmem_recalc_inode(struct inode *inode)
260{
261 struct shmem_inode_info *info = SHMEM_I(inode);
262 long freed;
263
264 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
265 if (freed > 0) {
266 info->alloced -= freed;
267 shmem_unacct_blocks(info->flags, freed);
268 shmem_free_blocks(inode, freed);
269 }
270}
271
Randy Dunlap46711812008-03-19 17:00:41 -0700272/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 * shmem_swp_entry - find the swap vector position in the info structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 * @info: info structure for the inode
275 * @index: index of the page to find
276 * @page: optional page to add to the structure. Has to be preset to
277 * all zeros
278 *
279 * If there is no space allocated yet it will return NULL when
280 * page is NULL, else it will use the page for the needed block,
281 * setting it to NULL on return to indicate that it has been used.
282 *
283 * The swap vector is organized the following way:
284 *
285 * There are SHMEM_NR_DIRECT entries directly stored in the
286 * shmem_inode_info structure. So small files do not need an addional
287 * allocation.
288 *
289 * For pages with index > SHMEM_NR_DIRECT there is the pointer
290 * i_indirect which points to a page which holds in the first half
291 * doubly indirect blocks, in the second half triple indirect blocks:
292 *
293 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
294 * following layout (for SHMEM_NR_DIRECT == 16):
295 *
296 * i_indirect -> dir --> 16-19
297 * | +-> 20-23
298 * |
299 * +-->dir2 --> 24-27
300 * | +-> 28-31
301 * | +-> 32-35
302 * | +-> 36-39
303 * |
304 * +-->dir3 --> 40-43
305 * +-> 44-47
306 * +-> 48-51
307 * +-> 52-55
308 */
309static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
310{
311 unsigned long offset;
312 struct page **dir;
313 struct page *subdir;
314
315 if (index < SHMEM_NR_DIRECT) {
316 shmem_swp_balance_unmap();
317 return info->i_direct+index;
318 }
319 if (!info->i_indirect) {
320 if (page) {
321 info->i_indirect = *page;
322 *page = NULL;
323 }
324 return NULL; /* need another page */
325 }
326
327 index -= SHMEM_NR_DIRECT;
328 offset = index % ENTRIES_PER_PAGE;
329 index /= ENTRIES_PER_PAGE;
330 dir = shmem_dir_map(info->i_indirect);
331
332 if (index >= ENTRIES_PER_PAGE/2) {
333 index -= ENTRIES_PER_PAGE/2;
334 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
335 index %= ENTRIES_PER_PAGE;
336 subdir = *dir;
337 if (!subdir) {
338 if (page) {
339 *dir = *page;
340 *page = NULL;
341 }
342 shmem_dir_unmap(dir);
343 return NULL; /* need another page */
344 }
345 shmem_dir_unmap(dir);
346 dir = shmem_dir_map(subdir);
347 }
348
349 dir += index;
350 subdir = *dir;
351 if (!subdir) {
352 if (!page || !(subdir = *page)) {
353 shmem_dir_unmap(dir);
354 return NULL; /* need a page */
355 }
356 *dir = subdir;
357 *page = NULL;
358 }
359 shmem_dir_unmap(dir);
360 return shmem_swp_map(subdir) + offset;
361}
362
363static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
364{
365 long incdec = value? 1: -1;
366
367 entry->val = value;
368 info->swapped += incdec;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700369 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
370 struct page *page = kmap_atomic_to_page(entry);
371 set_page_private(page, page_private(page) + incdec);
372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373}
374
Randy Dunlap46711812008-03-19 17:00:41 -0700375/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 * shmem_swp_alloc - get the position of the swap entry for the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 * @info: info structure for the inode
378 * @index: index of the page to find
379 * @sgp: check and recheck i_size? skip allocation?
Randy Dunlap46711812008-03-19 17:00:41 -0700380 *
381 * If the entry does not exist, allocate it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 */
383static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
384{
385 struct inode *inode = &info->vfs_inode;
386 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
387 struct page *page = NULL;
388 swp_entry_t *entry;
389
390 if (sgp != SGP_WRITE &&
391 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
392 return ERR_PTR(-EINVAL);
393
394 while (!(entry = shmem_swp_entry(info, index, &page))) {
395 if (sgp == SGP_READ)
396 return shmem_swp_map(ZERO_PAGE(0));
397 /*
398 * Test free_blocks against 1 not 0, since we have 1 data
399 * page (and perhaps indirect index pages) yet to allocate:
400 * a waste to allocate index if we cannot allocate data.
401 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700402 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 spin_lock(&sbinfo->stat_lock);
404 if (sbinfo->free_blocks <= 1) {
405 spin_unlock(&sbinfo->stat_lock);
406 return ERR_PTR(-ENOSPC);
407 }
408 sbinfo->free_blocks--;
409 inode->i_blocks += BLOCKS_PER_PAGE;
410 spin_unlock(&sbinfo->stat_lock);
411 }
412
413 spin_unlock(&info->lock);
Mel Gorman769848c2007-07-17 04:03:05 -0700414 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700415 if (page)
416 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 spin_lock(&info->lock);
418
419 if (!page) {
420 shmem_free_blocks(inode, 1);
421 return ERR_PTR(-ENOMEM);
422 }
423 if (sgp != SGP_WRITE &&
424 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
425 entry = ERR_PTR(-EINVAL);
426 break;
427 }
428 if (info->next_index <= index)
429 info->next_index = index + 1;
430 }
431 if (page) {
432 /* another task gave its page, or truncated the file */
433 shmem_free_blocks(inode, 1);
434 shmem_dir_free(page);
435 }
436 if (info->next_index <= index && !IS_ERR(entry))
437 info->next_index = index + 1;
438 return entry;
439}
440
Randy Dunlap46711812008-03-19 17:00:41 -0700441/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 * shmem_free_swp - free some swap entries in a directory
Hugh Dickins1ae70002007-03-29 01:20:36 -0700443 * @dir: pointer to the directory
444 * @edir: pointer after last entry of the directory
445 * @punch_lock: pointer to spinlock when needed for the holepunch case
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 */
Hugh Dickins1ae70002007-03-29 01:20:36 -0700447static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
448 spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449{
Hugh Dickins1ae70002007-03-29 01:20:36 -0700450 spinlock_t *punch_unlock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 swp_entry_t *ptr;
452 int freed = 0;
453
454 for (ptr = dir; ptr < edir; ptr++) {
455 if (ptr->val) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700456 if (unlikely(punch_lock)) {
457 punch_unlock = punch_lock;
458 punch_lock = NULL;
459 spin_lock(punch_unlock);
460 if (!ptr->val)
461 continue;
462 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 free_swap_and_cache(*ptr);
464 *ptr = (swp_entry_t){0};
465 freed++;
466 }
467 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700468 if (punch_unlock)
469 spin_unlock(punch_unlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 return freed;
471}
472
Hugh Dickins1ae70002007-03-29 01:20:36 -0700473static int shmem_map_and_free_swp(struct page *subdir, int offset,
474 int limit, struct page ***dir, spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475{
476 swp_entry_t *ptr;
477 int freed = 0;
478
479 ptr = shmem_swp_map(subdir);
480 for (; offset < limit; offset += LATENCY_LIMIT) {
481 int size = limit - offset;
482 if (size > LATENCY_LIMIT)
483 size = LATENCY_LIMIT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700484 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
485 punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 if (need_resched()) {
487 shmem_swp_unmap(ptr);
488 if (*dir) {
489 shmem_dir_unmap(*dir);
490 *dir = NULL;
491 }
492 cond_resched();
493 ptr = shmem_swp_map(subdir);
494 }
495 }
496 shmem_swp_unmap(ptr);
497 return freed;
498}
499
500static void shmem_free_pages(struct list_head *next)
501{
502 struct page *page;
503 int freed = 0;
504
505 do {
506 page = container_of(next, struct page, lru);
507 next = next->next;
508 shmem_dir_free(page);
509 freed++;
510 if (freed >= LATENCY_LIMIT) {
511 cond_resched();
512 freed = 0;
513 }
514 } while (next);
515}
516
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800517static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518{
519 struct shmem_inode_info *info = SHMEM_I(inode);
520 unsigned long idx;
521 unsigned long size;
522 unsigned long limit;
523 unsigned long stage;
524 unsigned long diroff;
525 struct page **dir;
526 struct page *topdir;
527 struct page *middir;
528 struct page *subdir;
529 swp_entry_t *ptr;
530 LIST_HEAD(pages_to_free);
531 long nr_pages_to_free = 0;
532 long nr_swaps_freed = 0;
533 int offset;
534 int freed;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700535 int punch_hole;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700536 spinlock_t *needs_lock;
537 spinlock_t *punch_lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700538 unsigned long upper_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
540 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800541 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 if (idx >= info->next_index)
543 return;
544
545 spin_lock(&info->lock);
546 info->flags |= SHMEM_TRUNCATE;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800547 if (likely(end == (loff_t) -1)) {
548 limit = info->next_index;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700549 upper_limit = SHMEM_MAX_INDEX;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800550 info->next_index = idx;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700551 needs_lock = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700552 punch_hole = 0;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800553 } else {
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700554 if (end + 1 >= inode->i_size) { /* we may free a little more */
555 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
556 PAGE_CACHE_SHIFT;
557 upper_limit = SHMEM_MAX_INDEX;
558 } else {
559 limit = (end + 1) >> PAGE_CACHE_SHIFT;
560 upper_limit = limit;
561 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700562 needs_lock = &info->lock;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800563 punch_hole = 1;
564 }
565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 topdir = info->i_indirect;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800567 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 info->i_indirect = NULL;
569 nr_pages_to_free++;
570 list_add(&topdir->lru, &pages_to_free);
571 }
572 spin_unlock(&info->lock);
573
574 if (info->swapped && idx < SHMEM_NR_DIRECT) {
575 ptr = info->i_direct;
576 size = limit;
577 if (size > SHMEM_NR_DIRECT)
578 size = SHMEM_NR_DIRECT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700579 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 }
Badari Pulavarty92a3d032006-12-22 01:06:23 -0800581
582 /*
583 * If there are no indirect blocks or we are punching a hole
584 * below indirect blocks, nothing to be done.
585 */
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700586 if (!topdir || limit <= SHMEM_NR_DIRECT)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 goto done2;
588
Hugh Dickins1ae70002007-03-29 01:20:36 -0700589 /*
590 * The truncation case has already dropped info->lock, and we're safe
591 * because i_size and next_index have already been lowered, preventing
592 * access beyond. But in the punch_hole case, we still need to take
593 * the lock when updating the swap directory, because there might be
594 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
595 * shmem_writepage. However, whenever we find we can remove a whole
596 * directory page (not at the misaligned start or end of the range),
597 * we first NULLify its pointer in the level above, and then have no
598 * need to take the lock when updating its contents: needs_lock and
599 * punch_lock (either pointing to info->lock or NULL) manage this.
600 */
601
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700602 upper_limit -= SHMEM_NR_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 limit -= SHMEM_NR_DIRECT;
604 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
605 offset = idx % ENTRIES_PER_PAGE;
606 idx -= offset;
607
608 dir = shmem_dir_map(topdir);
609 stage = ENTRIES_PER_PAGEPAGE/2;
610 if (idx < ENTRIES_PER_PAGEPAGE/2) {
611 middir = topdir;
612 diroff = idx/ENTRIES_PER_PAGE;
613 } else {
614 dir += ENTRIES_PER_PAGE/2;
615 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
616 while (stage <= idx)
617 stage += ENTRIES_PER_PAGEPAGE;
618 middir = *dir;
619 if (*dir) {
620 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
621 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700622 if (!diroff && !offset && upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700623 if (needs_lock) {
624 spin_lock(needs_lock);
625 *dir = NULL;
626 spin_unlock(needs_lock);
627 needs_lock = NULL;
628 } else
629 *dir = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 nr_pages_to_free++;
631 list_add(&middir->lru, &pages_to_free);
632 }
633 shmem_dir_unmap(dir);
634 dir = shmem_dir_map(middir);
635 } else {
636 diroff = 0;
637 offset = 0;
638 idx = stage;
639 }
640 }
641
642 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
643 if (unlikely(idx == stage)) {
644 shmem_dir_unmap(dir);
645 dir = shmem_dir_map(topdir) +
646 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
647 while (!*dir) {
648 dir++;
649 idx += ENTRIES_PER_PAGEPAGE;
650 if (idx >= limit)
651 goto done1;
652 }
653 stage = idx + ENTRIES_PER_PAGEPAGE;
654 middir = *dir;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700655 if (punch_hole)
656 needs_lock = &info->lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700657 if (upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700658 if (needs_lock) {
659 spin_lock(needs_lock);
660 *dir = NULL;
661 spin_unlock(needs_lock);
662 needs_lock = NULL;
663 } else
664 *dir = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700665 nr_pages_to_free++;
666 list_add(&middir->lru, &pages_to_free);
667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 shmem_dir_unmap(dir);
669 cond_resched();
670 dir = shmem_dir_map(middir);
671 diroff = 0;
672 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700673 punch_lock = needs_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 subdir = dir[diroff];
Hugh Dickins1ae70002007-03-29 01:20:36 -0700675 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
676 if (needs_lock) {
677 spin_lock(needs_lock);
678 dir[diroff] = NULL;
679 spin_unlock(needs_lock);
680 punch_lock = NULL;
681 } else
682 dir[diroff] = NULL;
683 nr_pages_to_free++;
684 list_add(&subdir->lru, &pages_to_free);
685 }
686 if (subdir && page_private(subdir) /* has swap entries */) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 size = limit - idx;
688 if (size > ENTRIES_PER_PAGE)
689 size = ENTRIES_PER_PAGE;
690 freed = shmem_map_and_free_swp(subdir,
Hugh Dickins1ae70002007-03-29 01:20:36 -0700691 offset, size, &dir, punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 if (!dir)
693 dir = shmem_dir_map(middir);
694 nr_swaps_freed += freed;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700695 if (offset || punch_lock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 spin_lock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700697 set_page_private(subdir,
698 page_private(subdir) - freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 spin_unlock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700700 } else
701 BUG_ON(page_private(subdir) != freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700703 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 }
705done1:
706 shmem_dir_unmap(dir);
707done2:
708 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
709 /*
710 * Call truncate_inode_pages again: racing shmem_unuse_inode
711 * may have swizzled a page in from swap since vmtruncate or
712 * generic_delete_inode did it, before we lowered next_index.
713 * Also, though shmem_getpage checks i_size before adding to
714 * cache, no recheck after: so fix the narrow window there too.
Hugh Dickins16a10012007-03-29 01:20:37 -0700715 *
716 * Recalling truncate_inode_pages_range and unmap_mapping_range
717 * every time for punch_hole (which never got a chance to clear
718 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
719 * yet hardly ever necessary: try to optimize them out later.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800721 truncate_inode_pages_range(inode->i_mapping, start, end);
Hugh Dickins16a10012007-03-29 01:20:37 -0700722 if (punch_hole)
723 unmap_mapping_range(inode->i_mapping, start,
724 end - start, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 }
726
727 spin_lock(&info->lock);
728 info->flags &= ~SHMEM_TRUNCATE;
729 info->swapped -= nr_swaps_freed;
730 if (nr_pages_to_free)
731 shmem_free_blocks(inode, nr_pages_to_free);
732 shmem_recalc_inode(inode);
733 spin_unlock(&info->lock);
734
735 /*
736 * Empty swap vector directory pages to be freed?
737 */
738 if (!list_empty(&pages_to_free)) {
739 pages_to_free.prev->next = NULL;
740 shmem_free_pages(pages_to_free.next);
741 }
742}
743
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800744static void shmem_truncate(struct inode *inode)
745{
746 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
747}
748
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
750{
751 struct inode *inode = dentry->d_inode;
752 struct page *page = NULL;
753 int error;
754
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700755 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 if (attr->ia_size < inode->i_size) {
757 /*
758 * If truncating down to a partial page, then
759 * if that page is already allocated, hold it
760 * in memory until the truncation is over, so
761 * truncate_partial_page cannnot miss it were
762 * it assigned to swap.
763 */
764 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
765 (void) shmem_getpage(inode,
766 attr->ia_size>>PAGE_CACHE_SHIFT,
767 &page, SGP_READ, NULL);
Hugh Dickinsd3602442008-02-04 22:28:44 -0800768 if (page)
769 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 }
771 /*
772 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
773 * detect if any pages might have been added to cache
774 * after truncate_inode_pages. But we needn't bother
775 * if it's being fully truncated to zero-length: the
776 * nrpages check is efficient enough in that case.
777 */
778 if (attr->ia_size) {
779 struct shmem_inode_info *info = SHMEM_I(inode);
780 spin_lock(&info->lock);
781 info->flags &= ~SHMEM_PAGEIN;
782 spin_unlock(&info->lock);
783 }
784 }
785 }
786
787 error = inode_change_ok(inode, attr);
788 if (!error)
789 error = inode_setattr(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700790#ifdef CONFIG_TMPFS_POSIX_ACL
791 if (!error && (attr->ia_valid & ATTR_MODE))
792 error = generic_acl_chmod(inode, &shmem_acl_ops);
793#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 if (page)
795 page_cache_release(page);
796 return error;
797}
798
799static void shmem_delete_inode(struct inode *inode)
800{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 struct shmem_inode_info *info = SHMEM_I(inode);
802
803 if (inode->i_op->truncate == shmem_truncate) {
Mark Fashehfef26652005-09-09 13:01:31 -0700804 truncate_inode_pages(inode->i_mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 shmem_unacct_size(info->flags, inode->i_size);
806 inode->i_size = 0;
807 shmem_truncate(inode);
808 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800809 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800811 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 }
813 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700814 BUG_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800815 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 clear_inode(inode);
817}
818
819static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
820{
821 swp_entry_t *ptr;
822
823 for (ptr = dir; ptr < edir; ptr++) {
824 if (ptr->val == entry.val)
825 return ptr - dir;
826 }
827 return -1;
828}
829
830static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
831{
832 struct inode *inode;
833 unsigned long idx;
834 unsigned long size;
835 unsigned long limit;
836 unsigned long stage;
837 struct page **dir;
838 struct page *subdir;
839 swp_entry_t *ptr;
840 int offset;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800841 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
843 idx = 0;
844 ptr = info->i_direct;
845 spin_lock(&info->lock);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800846 if (!info->swapped) {
847 list_del_init(&info->swaplist);
848 goto lost2;
849 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 limit = info->next_index;
851 size = limit;
852 if (size > SHMEM_NR_DIRECT)
853 size = SHMEM_NR_DIRECT;
854 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800855 if (offset >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 goto found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 if (!info->i_indirect)
858 goto lost2;
859
860 dir = shmem_dir_map(info->i_indirect);
861 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
862
863 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
864 if (unlikely(idx == stage)) {
865 shmem_dir_unmap(dir-1);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800866 if (cond_resched_lock(&info->lock)) {
867 /* check it has not been truncated */
868 if (limit > info->next_index) {
869 limit = info->next_index;
870 if (idx >= limit)
871 goto lost2;
872 }
873 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 dir = shmem_dir_map(info->i_indirect) +
875 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
876 while (!*dir) {
877 dir++;
878 idx += ENTRIES_PER_PAGEPAGE;
879 if (idx >= limit)
880 goto lost1;
881 }
882 stage = idx + ENTRIES_PER_PAGEPAGE;
883 subdir = *dir;
884 shmem_dir_unmap(dir);
885 dir = shmem_dir_map(subdir);
886 }
887 subdir = *dir;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700888 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 ptr = shmem_swp_map(subdir);
890 size = limit - idx;
891 if (size > ENTRIES_PER_PAGE)
892 size = ENTRIES_PER_PAGE;
893 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800894 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 if (offset >= 0) {
896 shmem_dir_unmap(dir);
897 goto found;
898 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 }
900 }
901lost1:
902 shmem_dir_unmap(dir-1);
903lost2:
904 spin_unlock(&info->lock);
905 return 0;
906found:
907 idx += offset;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800908 inode = igrab(&info->vfs_inode);
909 spin_unlock(&info->lock);
910
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800911 /*
912 * Move _head_ to start search for next from here.
913 * But be careful: shmem_delete_inode checks list_empty without taking
914 * mutex, and there's an instant in list_move_tail when info->swaplist
915 * would appear empty, if it were the only one on shmem_swaplist. We
916 * could avoid doing it if inode NULL; or use this minor optimization.
917 */
918 if (shmem_swaplist.next != &info->swaplist)
919 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800920 mutex_unlock(&shmem_swaplist_mutex);
921
922 error = 1;
923 if (!inode)
924 goto out;
Hugh Dickins82369552008-02-07 00:14:22 -0800925 /* Precharge page while we can wait, compensate afterwards */
926 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800927 if (error)
928 goto out;
Hugh Dickins82369552008-02-07 00:14:22 -0800929 error = radix_tree_preload(GFP_KERNEL);
930 if (error)
931 goto uncharge;
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800932 error = 1;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800933
934 spin_lock(&info->lock);
935 ptr = shmem_swp_entry(info, idx, NULL);
936 if (ptr && ptr->val == entry.val)
937 error = add_to_page_cache(page, inode->i_mapping,
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800938 idx, GFP_NOWAIT);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800939 if (error == -EEXIST) {
940 struct page *filepage = find_get_page(inode->i_mapping, idx);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800941 error = 1;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800942 if (filepage) {
943 /*
944 * There might be a more uptodate page coming down
945 * from a stacked writepage: forget our swappage if so.
946 */
947 if (PageUptodate(filepage))
948 error = 0;
949 page_cache_release(filepage);
950 }
951 }
952 if (!error) {
Hugh Dickins73b12622008-02-04 22:28:50 -0800953 delete_from_swap_cache(page);
954 set_page_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 info->flags |= SHMEM_PAGEIN;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800956 shmem_swp_set(info, ptr, 0);
957 swap_free(entry);
958 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800960 if (ptr)
961 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 spin_unlock(&info->lock);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800963 radix_tree_preload_end();
Hugh Dickins82369552008-02-07 00:14:22 -0800964uncharge:
965 mem_cgroup_uncharge_page(page);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800966out:
967 unlock_page(page);
968 page_cache_release(page);
969 iput(inode); /* allows for NULL */
970 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971}
972
973/*
974 * shmem_unuse() search for an eventually swapped out shmem page.
975 */
976int shmem_unuse(swp_entry_t entry, struct page *page)
977{
978 struct list_head *p, *next;
979 struct shmem_inode_info *info;
980 int found = 0;
981
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800982 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 list_for_each_safe(p, next, &shmem_swaplist) {
984 info = list_entry(p, struct shmem_inode_info, swaplist);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800985 found = shmem_unuse_inode(info, entry, page);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800986 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800987 if (found)
988 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800990 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800991out: return found; /* 0 or 1 or -ENOMEM */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992}
993
994/*
995 * Move the page from the page cache to the swap cache.
996 */
997static int shmem_writepage(struct page *page, struct writeback_control *wbc)
998{
999 struct shmem_inode_info *info;
1000 swp_entry_t *entry, swap;
1001 struct address_space *mapping;
1002 unsigned long index;
1003 struct inode *inode;
1004
1005 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 mapping = page->mapping;
1007 index = page->index;
1008 inode = mapping->host;
1009 info = SHMEM_I(inode);
1010 if (info->flags & VM_LOCKED)
1011 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001012 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 goto redirty;
1014
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001015 /*
1016 * shmem_backing_dev_info's capabilities prevent regular writeback or
1017 * sync from ever calling shmem_writepage; but a stacking filesystem
1018 * may use the ->writepage of its underlying filesystem, in which case
1019 * tmpfs should write out to swap only in response to memory pressure,
1020 * and not for pdflush or sync. However, in those cases, we do still
1021 * want to check if there's a redundant swappage to be discarded.
1022 */
1023 if (wbc->for_reclaim)
1024 swap = get_swap_page();
1025 else
1026 swap.val = 0;
1027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 if (index >= info->next_index) {
1030 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1031 goto unlock;
1032 }
1033 entry = shmem_swp_entry(info, index, NULL);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001034 if (entry->val) {
1035 /*
1036 * The more uptodate page coming down from a stacked
1037 * writepage should replace our old swappage.
1038 */
1039 free_swap_and_cache(*entry);
1040 shmem_swp_set(info, entry, 0);
1041 }
1042 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001044 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickins73b12622008-02-04 22:28:50 -08001045 remove_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 shmem_swp_set(info, entry, swap.val);
1047 shmem_swp_unmap(entry);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001048 if (list_empty(&info->swaplist))
1049 inode = igrab(inode);
1050 else
1051 inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 spin_unlock(&info->lock);
Hugh Dickins73b12622008-02-04 22:28:50 -08001053 swap_duplicate(swap);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001054 BUG_ON(page_mapped(page));
Hugh Dickins73b12622008-02-04 22:28:50 -08001055 page_cache_release(page); /* pagecache ref */
1056 set_page_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 unlock_page(page);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001058 if (inode) {
1059 mutex_lock(&shmem_swaplist_mutex);
1060 /* move instead of add in case we're racing */
1061 list_move_tail(&info->swaplist, &shmem_swaplist);
1062 mutex_unlock(&shmem_swaplist_mutex);
1063 iput(inode);
1064 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 return 0;
1066 }
1067
1068 shmem_swp_unmap(entry);
1069unlock:
1070 spin_unlock(&info->lock);
1071 swap_free(swap);
1072redirty:
1073 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001074 if (wbc->for_reclaim)
1075 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1076 unlock_page(page);
1077 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078}
1079
1080#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001081#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001082static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001083{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001084 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001085
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001086 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001087 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001088
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001089 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001090
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001091 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001092}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001093
1094static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1095{
1096 struct mempolicy *mpol = NULL;
1097 if (sbinfo->mpol) {
1098 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1099 mpol = sbinfo->mpol;
1100 mpol_get(mpol);
1101 spin_unlock(&sbinfo->stat_lock);
1102 }
1103 return mpol;
1104}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001105#endif /* CONFIG_TMPFS */
1106
Hugh Dickins02098fe2008-02-04 22:28:42 -08001107static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1108 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109{
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001110 struct mempolicy mpol, *spol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 struct vm_area_struct pvma;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001112 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001114 spol = mpol_cond_copy(&mpol,
1115 mpol_shared_policy_lookup(&info->policy, idx));
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 /* Create a pseudo vma that just contains the policy */
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001118 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001120 pvma.vm_ops = NULL;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001121 pvma.vm_policy = spol;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001122 page = swapin_readahead(entry, gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 return page;
1124}
1125
Hugh Dickins02098fe2008-02-04 22:28:42 -08001126static struct page *shmem_alloc_page(gfp_t gfp,
1127 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128{
1129 struct vm_area_struct pvma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001131 /* Create a pseudo vma that just contains the policy */
1132 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001134 pvma.vm_ops = NULL;
1135 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001136
1137 /*
1138 * alloc_page_vma() will drop the shared policy reference
1139 */
1140 return alloc_page_vma(gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001142#else /* !CONFIG_NUMA */
1143#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001144static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001145{
1146}
1147#endif /* CONFIG_TMPFS */
1148
Hugh Dickins02098fe2008-02-04 22:28:42 -08001149static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1150 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151{
Hugh Dickins02098fe2008-02-04 22:28:42 -08001152 return swapin_readahead(entry, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153}
1154
Hugh Dickins02098fe2008-02-04 22:28:42 -08001155static inline struct page *shmem_alloc_page(gfp_t gfp,
1156 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157{
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001158 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001160#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001162#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1163static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1164{
1165 return NULL;
1166}
1167#endif
1168
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169/*
1170 * shmem_getpage - either get the page from swap or allocate a new one
1171 *
1172 * If we allocate a new one we do not mark it dirty. That's up to the
1173 * vm. If we swap it in we mark it dirty since we also free the swap
1174 * entry since a page cannot live in both the swap and page cache
1175 */
1176static int shmem_getpage(struct inode *inode, unsigned long idx,
1177 struct page **pagep, enum sgp_type sgp, int *type)
1178{
1179 struct address_space *mapping = inode->i_mapping;
1180 struct shmem_inode_info *info = SHMEM_I(inode);
1181 struct shmem_sb_info *sbinfo;
1182 struct page *filepage = *pagep;
1183 struct page *swappage;
1184 swp_entry_t *entry;
1185 swp_entry_t swap;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001186 gfp_t gfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 int error;
1188
1189 if (idx >= SHMEM_MAX_INDEX)
1190 return -EFBIG;
Nick Piggin54cb8822007-07-19 01:46:59 -07001191
1192 if (type)
Nick Piggin83c54072007-07-19 01:47:05 -07001193 *type = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07001194
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 /*
1196 * Normally, filepage is NULL on entry, and either found
1197 * uptodate immediately, or allocated and zeroed, or read
1198 * in under swappage, which is then assigned to filepage.
Hugh Dickins5402b972008-02-04 22:28:44 -08001199 * But shmem_readpage (required for splice) passes in a locked
Hugh Dickinsae976412007-06-04 10:00:39 +02001200 * filepage, which may be found not uptodate by other callers
1201 * too, and may need to be copied from the swappage read in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 */
1203repeat:
1204 if (!filepage)
1205 filepage = find_lock_page(mapping, idx);
1206 if (filepage && PageUptodate(filepage))
1207 goto done;
1208 error = 0;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001209 gfp = mapping_gfp_mask(mapping);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -08001210 if (!filepage) {
1211 /*
1212 * Try to preload while we can wait, to not make a habit of
1213 * draining atomic reserves; but don't latch on to this cpu.
1214 */
1215 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1216 if (error)
1217 goto failed;
1218 radix_tree_preload_end();
1219 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220
1221 spin_lock(&info->lock);
1222 shmem_recalc_inode(inode);
1223 entry = shmem_swp_alloc(info, idx, sgp);
1224 if (IS_ERR(entry)) {
1225 spin_unlock(&info->lock);
1226 error = PTR_ERR(entry);
1227 goto failed;
1228 }
1229 swap = *entry;
1230
1231 if (swap.val) {
1232 /* Look it up and read it in.. */
1233 swappage = lookup_swap_cache(swap);
1234 if (!swappage) {
1235 shmem_swp_unmap(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 /* here we actually do the io */
Nick Piggin83c54072007-07-19 01:47:05 -07001237 if (type && !(*type & VM_FAULT_MAJOR)) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001238 __count_vm_event(PGMAJFAULT);
Nick Piggin83c54072007-07-19 01:47:05 -07001239 *type |= VM_FAULT_MAJOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 }
Christoph Lameterf8891e52006-06-30 01:55:45 -07001241 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001242 swappage = shmem_swapin(swap, gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 if (!swappage) {
1244 spin_lock(&info->lock);
1245 entry = shmem_swp_alloc(info, idx, sgp);
1246 if (IS_ERR(entry))
1247 error = PTR_ERR(entry);
1248 else {
1249 if (entry->val == swap.val)
1250 error = -ENOMEM;
1251 shmem_swp_unmap(entry);
1252 }
1253 spin_unlock(&info->lock);
1254 if (error)
1255 goto failed;
1256 goto repeat;
1257 }
1258 wait_on_page_locked(swappage);
1259 page_cache_release(swappage);
1260 goto repeat;
1261 }
1262
1263 /* We have to do this with page locked to prevent races */
1264 if (TestSetPageLocked(swappage)) {
1265 shmem_swp_unmap(entry);
1266 spin_unlock(&info->lock);
1267 wait_on_page_locked(swappage);
1268 page_cache_release(swappage);
1269 goto repeat;
1270 }
1271 if (PageWriteback(swappage)) {
1272 shmem_swp_unmap(entry);
1273 spin_unlock(&info->lock);
1274 wait_on_page_writeback(swappage);
1275 unlock_page(swappage);
1276 page_cache_release(swappage);
1277 goto repeat;
1278 }
1279 if (!PageUptodate(swappage)) {
1280 shmem_swp_unmap(entry);
1281 spin_unlock(&info->lock);
1282 unlock_page(swappage);
1283 page_cache_release(swappage);
1284 error = -EIO;
1285 goto failed;
1286 }
1287
1288 if (filepage) {
1289 shmem_swp_set(info, entry, 0);
1290 shmem_swp_unmap(entry);
1291 delete_from_swap_cache(swappage);
1292 spin_unlock(&info->lock);
1293 copy_highpage(filepage, swappage);
1294 unlock_page(swappage);
1295 page_cache_release(swappage);
1296 flush_dcache_page(filepage);
1297 SetPageUptodate(filepage);
1298 set_page_dirty(filepage);
1299 swap_free(swap);
Hugh Dickins73b12622008-02-04 22:28:50 -08001300 } else if (!(error = add_to_page_cache(
Hugh Dickinsb409f9f2008-02-04 22:28:54 -08001301 swappage, mapping, idx, GFP_NOWAIT))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 info->flags |= SHMEM_PAGEIN;
1303 shmem_swp_set(info, entry, 0);
1304 shmem_swp_unmap(entry);
Hugh Dickins73b12622008-02-04 22:28:50 -08001305 delete_from_swap_cache(swappage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 spin_unlock(&info->lock);
1307 filepage = swappage;
Hugh Dickins73b12622008-02-04 22:28:50 -08001308 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 swap_free(swap);
1310 } else {
1311 shmem_swp_unmap(entry);
1312 spin_unlock(&info->lock);
1313 unlock_page(swappage);
Hugh Dickins82369552008-02-07 00:14:22 -08001314 if (error == -ENOMEM) {
1315 /* allow reclaim from this memory cgroup */
Hugh Dickins7e924aa2008-03-04 14:29:08 -08001316 error = mem_cgroup_cache_charge(swappage,
Hugh Dickins82369552008-02-07 00:14:22 -08001317 current->mm, gfp & ~__GFP_HIGHMEM);
Hugh Dickins7e924aa2008-03-04 14:29:08 -08001318 if (error) {
1319 page_cache_release(swappage);
Hugh Dickins82369552008-02-07 00:14:22 -08001320 goto failed;
Hugh Dickins7e924aa2008-03-04 14:29:08 -08001321 }
1322 mem_cgroup_uncharge_page(swappage);
Hugh Dickins82369552008-02-07 00:14:22 -08001323 }
Hugh Dickins7e924aa2008-03-04 14:29:08 -08001324 page_cache_release(swappage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 goto repeat;
1326 }
1327 } else if (sgp == SGP_READ && !filepage) {
1328 shmem_swp_unmap(entry);
1329 filepage = find_get_page(mapping, idx);
1330 if (filepage &&
1331 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1332 spin_unlock(&info->lock);
1333 wait_on_page_locked(filepage);
1334 page_cache_release(filepage);
1335 filepage = NULL;
1336 goto repeat;
1337 }
1338 spin_unlock(&info->lock);
1339 } else {
1340 shmem_swp_unmap(entry);
1341 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001342 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 spin_lock(&sbinfo->stat_lock);
1344 if (sbinfo->free_blocks == 0 ||
1345 shmem_acct_block(info->flags)) {
1346 spin_unlock(&sbinfo->stat_lock);
1347 spin_unlock(&info->lock);
1348 error = -ENOSPC;
1349 goto failed;
1350 }
1351 sbinfo->free_blocks--;
1352 inode->i_blocks += BLOCKS_PER_PAGE;
1353 spin_unlock(&sbinfo->stat_lock);
1354 } else if (shmem_acct_block(info->flags)) {
1355 spin_unlock(&info->lock);
1356 error = -ENOSPC;
1357 goto failed;
1358 }
1359
1360 if (!filepage) {
1361 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001362 filepage = shmem_alloc_page(gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 if (!filepage) {
1364 shmem_unacct_blocks(info->flags, 1);
1365 shmem_free_blocks(inode, 1);
1366 error = -ENOMEM;
1367 goto failed;
1368 }
1369
Hugh Dickins82369552008-02-07 00:14:22 -08001370 /* Precharge page while we can wait, compensate after */
1371 error = mem_cgroup_cache_charge(filepage, current->mm,
1372 gfp & ~__GFP_HIGHMEM);
1373 if (error) {
1374 page_cache_release(filepage);
1375 shmem_unacct_blocks(info->flags, 1);
1376 shmem_free_blocks(inode, 1);
1377 filepage = NULL;
1378 goto failed;
1379 }
1380
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 spin_lock(&info->lock);
1382 entry = shmem_swp_alloc(info, idx, sgp);
1383 if (IS_ERR(entry))
1384 error = PTR_ERR(entry);
1385 else {
1386 swap = *entry;
1387 shmem_swp_unmap(entry);
1388 }
1389 if (error || swap.val || 0 != add_to_page_cache_lru(
Hugh Dickinsb409f9f2008-02-04 22:28:54 -08001390 filepage, mapping, idx, GFP_NOWAIT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 spin_unlock(&info->lock);
Hugh Dickins82369552008-02-07 00:14:22 -08001392 mem_cgroup_uncharge_page(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 page_cache_release(filepage);
1394 shmem_unacct_blocks(info->flags, 1);
1395 shmem_free_blocks(inode, 1);
1396 filepage = NULL;
1397 if (error)
1398 goto failed;
1399 goto repeat;
1400 }
Hugh Dickins82369552008-02-07 00:14:22 -08001401 mem_cgroup_uncharge_page(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 info->flags |= SHMEM_PAGEIN;
1403 }
1404
1405 info->alloced++;
1406 spin_unlock(&info->lock);
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001407 clear_highpage(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 flush_dcache_page(filepage);
1409 SetPageUptodate(filepage);
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001410 if (sgp == SGP_DIRTY)
1411 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 }
1413done:
Hugh Dickinsd3602442008-02-04 22:28:44 -08001414 *pagep = filepage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 return 0;
1416
1417failed:
1418 if (*pagep != filepage) {
1419 unlock_page(filepage);
1420 page_cache_release(filepage);
1421 }
1422 return error;
1423}
1424
Nick Piggind0217ac2007-07-19 01:47:03 -07001425static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001427 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 int error;
Nick Piggind0217ac2007-07-19 01:47:03 -07001429 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
Nick Piggind0217ac2007-07-19 01:47:03 -07001431 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1432 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001433
Hugh Dickins27d54b32008-02-04 22:28:43 -08001434 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
Nick Piggind0217ac2007-07-19 01:47:03 -07001435 if (error)
1436 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
Nick Piggind0217ac2007-07-19 01:47:03 -07001438 mark_page_accessed(vmf->page);
Nick Piggin83c54072007-07-19 01:47:05 -07001439 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440}
1441
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442#ifdef CONFIG_NUMA
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001443static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001445 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1447}
1448
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001449static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1450 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001452 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 unsigned long idx;
1454
1455 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1456 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1457}
1458#endif
1459
1460int shmem_lock(struct file *file, int lock, struct user_struct *user)
1461{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001462 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 struct shmem_inode_info *info = SHMEM_I(inode);
1464 int retval = -ENOMEM;
1465
1466 spin_lock(&info->lock);
1467 if (lock && !(info->flags & VM_LOCKED)) {
1468 if (!user_shm_lock(inode->i_size, user))
1469 goto out_nomem;
1470 info->flags |= VM_LOCKED;
1471 }
1472 if (!lock && (info->flags & VM_LOCKED) && user) {
1473 user_shm_unlock(inode->i_size, user);
1474 info->flags &= ~VM_LOCKED;
1475 }
1476 retval = 0;
1477out_nomem:
1478 spin_unlock(&info->lock);
1479 return retval;
1480}
1481
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001482static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483{
1484 file_accessed(file);
1485 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001486 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 return 0;
1488}
1489
1490static struct inode *
1491shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1492{
1493 struct inode *inode;
1494 struct shmem_inode_info *info;
1495 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1496
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001497 if (shmem_reserve_inode(sb))
1498 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
1500 inode = new_inode(sb);
1501 if (inode) {
1502 inode->i_mode = mode;
1503 inode->i_uid = current->fsuid;
1504 inode->i_gid = current->fsgid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 inode->i_blocks = 0;
1506 inode->i_mapping->a_ops = &shmem_aops;
1507 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1508 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001509 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 info = SHMEM_I(inode);
1511 memset(info, 0, (char *)inode - (char *)info);
1512 spin_lock_init(&info->lock);
1513 INIT_LIST_HEAD(&info->swaplist);
1514
1515 switch (mode & S_IFMT) {
1516 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001517 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 init_special_inode(inode, mode, dev);
1519 break;
1520 case S_IFREG:
1521 inode->i_op = &shmem_inode_operations;
1522 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001523 mpol_shared_policy_init(&info->policy,
1524 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 break;
1526 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001527 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 /* Some things misbehave if size == 0 on a directory */
1529 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1530 inode->i_op = &shmem_dir_inode_operations;
1531 inode->i_fop = &simple_dir_operations;
1532 break;
1533 case S_IFLNK:
1534 /*
1535 * Must not load anything in the rbtree,
1536 * mpol_free_shared_policy will not be called.
1537 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001538 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 break;
1540 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001541 } else
1542 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 return inode;
1544}
1545
1546#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001547static const struct inode_operations shmem_symlink_inode_operations;
1548static const struct inode_operations shmem_symlink_inline_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
1550/*
Nick Piggin800d15a2007-10-16 01:25:03 -07001551 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
Hugh Dickinsae976412007-06-04 10:00:39 +02001552 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1553 * below the loop driver, in the generic fashion that many filesystems support.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 */
Hugh Dickinsae976412007-06-04 10:00:39 +02001555static int shmem_readpage(struct file *file, struct page *page)
1556{
1557 struct inode *inode = page->mapping->host;
1558 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1559 unlock_page(page);
1560 return error;
1561}
1562
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001564shmem_write_begin(struct file *file, struct address_space *mapping,
1565 loff_t pos, unsigned len, unsigned flags,
1566 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567{
Nick Piggin800d15a2007-10-16 01:25:03 -07001568 struct inode *inode = mapping->host;
1569 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1570 *pagep = NULL;
1571 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1572}
1573
1574static int
1575shmem_write_end(struct file *file, struct address_space *mapping,
1576 loff_t pos, unsigned len, unsigned copied,
1577 struct page *page, void *fsdata)
1578{
1579 struct inode *inode = mapping->host;
1580
Hugh Dickinsd3602442008-02-04 22:28:44 -08001581 if (pos + copied > inode->i_size)
1582 i_size_write(inode, pos + copied);
1583
1584 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001585 set_page_dirty(page);
1586 page_cache_release(page);
1587
Nick Piggin800d15a2007-10-16 01:25:03 -07001588 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589}
1590
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1592{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001593 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 struct address_space *mapping = inode->i_mapping;
1595 unsigned long index, offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001596 enum sgp_type sgp = SGP_READ;
1597
1598 /*
1599 * Might this read be for a stacking filesystem? Then when reading
1600 * holes of a sparse file, we actually need to allocate those pages,
1601 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1602 */
1603 if (segment_eq(get_fs(), KERNEL_DS))
1604 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
1606 index = *ppos >> PAGE_CACHE_SHIFT;
1607 offset = *ppos & ~PAGE_CACHE_MASK;
1608
1609 for (;;) {
1610 struct page *page = NULL;
1611 unsigned long end_index, nr, ret;
1612 loff_t i_size = i_size_read(inode);
1613
1614 end_index = i_size >> PAGE_CACHE_SHIFT;
1615 if (index > end_index)
1616 break;
1617 if (index == end_index) {
1618 nr = i_size & ~PAGE_CACHE_MASK;
1619 if (nr <= offset)
1620 break;
1621 }
1622
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001623 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 if (desc->error) {
1625 if (desc->error == -EINVAL)
1626 desc->error = 0;
1627 break;
1628 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001629 if (page)
1630 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
1632 /*
1633 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001634 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 */
1636 nr = PAGE_CACHE_SIZE;
1637 i_size = i_size_read(inode);
1638 end_index = i_size >> PAGE_CACHE_SHIFT;
1639 if (index == end_index) {
1640 nr = i_size & ~PAGE_CACHE_MASK;
1641 if (nr <= offset) {
1642 if (page)
1643 page_cache_release(page);
1644 break;
1645 }
1646 }
1647 nr -= offset;
1648
1649 if (page) {
1650 /*
1651 * If users can be writing to this page using arbitrary
1652 * virtual addresses, take care about potential aliasing
1653 * before reading the page on the kernel side.
1654 */
1655 if (mapping_writably_mapped(mapping))
1656 flush_dcache_page(page);
1657 /*
1658 * Mark the page accessed if we read the beginning.
1659 */
1660 if (!offset)
1661 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001662 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001664 page_cache_get(page);
1665 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
1667 /*
1668 * Ok, we have the page, and it's up-to-date, so
1669 * now we can copy it to user space...
1670 *
1671 * The actor routine returns how many bytes were actually used..
1672 * NOTE! This may not be the same as how much of a user buffer
1673 * we filled up (we may be padding etc), so we can only update
1674 * "pos" here (the actor routine has to update the user buffer
1675 * pointers and the remaining count).
1676 */
1677 ret = actor(desc, page, offset, nr);
1678 offset += ret;
1679 index += offset >> PAGE_CACHE_SHIFT;
1680 offset &= ~PAGE_CACHE_MASK;
1681
1682 page_cache_release(page);
1683 if (ret != nr || !desc->count)
1684 break;
1685
1686 cond_resched();
1687 }
1688
1689 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1690 file_accessed(filp);
1691}
1692
1693static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1694{
1695 read_descriptor_t desc;
1696
1697 if ((ssize_t) count < 0)
1698 return -EINVAL;
1699 if (!access_ok(VERIFY_WRITE, buf, count))
1700 return -EFAULT;
1701 if (!count)
1702 return 0;
1703
1704 desc.written = 0;
1705 desc.count = count;
1706 desc.arg.buf = buf;
1707 desc.error = 0;
1708
1709 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1710 if (desc.written)
1711 return desc.written;
1712 return desc.error;
1713}
1714
David Howells726c3342006-06-23 02:02:58 -07001715static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716{
David Howells726c3342006-06-23 02:02:58 -07001717 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
1719 buf->f_type = TMPFS_MAGIC;
1720 buf->f_bsize = PAGE_CACHE_SIZE;
1721 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001722 spin_lock(&sbinfo->stat_lock);
1723 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 buf->f_blocks = sbinfo->max_blocks;
1725 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001726 }
1727 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 buf->f_files = sbinfo->max_inodes;
1729 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 }
1731 /* else leave those fields 0 like simple_statfs */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001732 spin_unlock(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 return 0;
1734}
1735
1736/*
1737 * File creation. Allocate an inode, and we're done..
1738 */
1739static int
1740shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1741{
1742 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1743 int error = -ENOSPC;
1744
1745 if (inode) {
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001746 error = security_inode_init_security(inode, dir, NULL, NULL,
1747 NULL);
1748 if (error) {
1749 if (error != -EOPNOTSUPP) {
1750 iput(inode);
1751 return error;
1752 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001753 }
1754 error = shmem_acl_init(inode, dir);
1755 if (error) {
1756 iput(inode);
1757 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001758 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 if (dir->i_mode & S_ISGID) {
1760 inode->i_gid = dir->i_gid;
1761 if (S_ISDIR(mode))
1762 inode->i_mode |= S_ISGID;
1763 }
1764 dir->i_size += BOGO_DIRENT_SIZE;
1765 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1766 d_instantiate(dentry, inode);
1767 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 }
1769 return error;
1770}
1771
1772static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1773{
1774 int error;
1775
1776 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1777 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001778 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 return 0;
1780}
1781
1782static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1783 struct nameidata *nd)
1784{
1785 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1786}
1787
1788/*
1789 * Link a file..
1790 */
1791static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1792{
1793 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001794 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795
1796 /*
1797 * No ordinary (disk based) filesystem counts links as inodes;
1798 * but each new link needs a new dentry, pinning lowmem, and
1799 * tmpfs dentries cannot be pruned until they are unlinked.
1800 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001801 ret = shmem_reserve_inode(inode->i_sb);
1802 if (ret)
1803 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
1805 dir->i_size += BOGO_DIRENT_SIZE;
1806 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001807 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 atomic_inc(&inode->i_count); /* New dentry reference */
1809 dget(dentry); /* Extra pinning count for the created dentry */
1810 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001811out:
1812 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813}
1814
1815static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1816{
1817 struct inode *inode = dentry->d_inode;
1818
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001819 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1820 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821
1822 dir->i_size -= BOGO_DIRENT_SIZE;
1823 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001824 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 dput(dentry); /* Undo the count from "create" - this does all the work */
1826 return 0;
1827}
1828
1829static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1830{
1831 if (!simple_empty(dentry))
1832 return -ENOTEMPTY;
1833
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001834 drop_nlink(dentry->d_inode);
1835 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 return shmem_unlink(dir, dentry);
1837}
1838
1839/*
1840 * The VFS layer already does all the dentry stuff for rename,
1841 * we just have to decrement the usage count for the target if
1842 * it exists so that the VFS layer correctly free's it when it
1843 * gets overwritten.
1844 */
1845static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1846{
1847 struct inode *inode = old_dentry->d_inode;
1848 int they_are_dirs = S_ISDIR(inode->i_mode);
1849
1850 if (!simple_empty(new_dentry))
1851 return -ENOTEMPTY;
1852
1853 if (new_dentry->d_inode) {
1854 (void) shmem_unlink(new_dir, new_dentry);
1855 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001856 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001858 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001859 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 }
1861
1862 old_dir->i_size -= BOGO_DIRENT_SIZE;
1863 new_dir->i_size += BOGO_DIRENT_SIZE;
1864 old_dir->i_ctime = old_dir->i_mtime =
1865 new_dir->i_ctime = new_dir->i_mtime =
1866 inode->i_ctime = CURRENT_TIME;
1867 return 0;
1868}
1869
1870static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1871{
1872 int error;
1873 int len;
1874 struct inode *inode;
1875 struct page *page = NULL;
1876 char *kaddr;
1877 struct shmem_inode_info *info;
1878
1879 len = strlen(symname) + 1;
1880 if (len > PAGE_CACHE_SIZE)
1881 return -ENAMETOOLONG;
1882
1883 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1884 if (!inode)
1885 return -ENOSPC;
1886
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001887 error = security_inode_init_security(inode, dir, NULL, NULL,
1888 NULL);
1889 if (error) {
1890 if (error != -EOPNOTSUPP) {
1891 iput(inode);
1892 return error;
1893 }
1894 error = 0;
1895 }
1896
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 info = SHMEM_I(inode);
1898 inode->i_size = len-1;
1899 if (len <= (char *)inode - (char *)info) {
1900 /* do it inline */
1901 memcpy(info, symname, len);
1902 inode->i_op = &shmem_symlink_inline_operations;
1903 } else {
1904 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1905 if (error) {
1906 iput(inode);
1907 return error;
1908 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001909 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 inode->i_op = &shmem_symlink_inode_operations;
1911 kaddr = kmap_atomic(page, KM_USER0);
1912 memcpy(kaddr, symname, len);
1913 kunmap_atomic(kaddr, KM_USER0);
1914 set_page_dirty(page);
1915 page_cache_release(page);
1916 }
1917 if (dir->i_mode & S_ISGID)
1918 inode->i_gid = dir->i_gid;
1919 dir->i_size += BOGO_DIRENT_SIZE;
1920 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1921 d_instantiate(dentry, inode);
1922 dget(dentry);
1923 return 0;
1924}
1925
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001926static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927{
1928 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001929 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930}
1931
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001932static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933{
1934 struct page *page = NULL;
1935 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1936 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08001937 if (page)
1938 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001939 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940}
1941
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001942static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943{
1944 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001945 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 kunmap(page);
1947 mark_page_accessed(page);
1948 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 }
1950}
1951
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001952static const struct inode_operations shmem_symlink_inline_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 .readlink = generic_readlink,
1954 .follow_link = shmem_follow_link_inline,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955};
1956
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001957static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 .truncate = shmem_truncate,
1959 .readlink = generic_readlink,
1960 .follow_link = shmem_follow_link,
1961 .put_link = shmem_put_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962};
1963
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001964#ifdef CONFIG_TMPFS_POSIX_ACL
Randy Dunlap46711812008-03-19 17:00:41 -07001965/*
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001966 * Superblocks without xattr inode operations will get security.* xattr
1967 * support from the VFS "for free". As soon as we have any other xattrs
1968 * like ACLs, we also need to implement the security.* handlers at
1969 * filesystem level, though.
1970 */
1971
1972static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1973 size_t list_len, const char *name,
1974 size_t name_len)
1975{
1976 return security_inode_listsecurity(inode, list, list_len);
1977}
1978
1979static int shmem_xattr_security_get(struct inode *inode, const char *name,
1980 void *buffer, size_t size)
1981{
1982 if (strcmp(name, "") == 0)
1983 return -EINVAL;
David P. Quigley42492592008-02-04 22:29:39 -08001984 return xattr_getsecurity(inode, name, buffer, size);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001985}
1986
1987static int shmem_xattr_security_set(struct inode *inode, const char *name,
1988 const void *value, size_t size, int flags)
1989{
1990 if (strcmp(name, "") == 0)
1991 return -EINVAL;
1992 return security_inode_setsecurity(inode, name, value, size, flags);
1993}
1994
Adrian Bunk1f370a22006-12-06 20:38:21 -08001995static struct xattr_handler shmem_xattr_security_handler = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001996 .prefix = XATTR_SECURITY_PREFIX,
1997 .list = shmem_xattr_security_list,
1998 .get = shmem_xattr_security_get,
1999 .set = shmem_xattr_security_set,
2000};
2001
2002static struct xattr_handler *shmem_xattr_handlers[] = {
2003 &shmem_xattr_acl_access_handler,
2004 &shmem_xattr_acl_default_handler,
2005 &shmem_xattr_security_handler,
2006 NULL
2007};
2008#endif
2009
David M. Grimes91828a42006-10-17 00:09:45 -07002010static struct dentry *shmem_get_parent(struct dentry *child)
2011{
2012 return ERR_PTR(-ESTALE);
2013}
2014
2015static int shmem_match(struct inode *ino, void *vfh)
2016{
2017 __u32 *fh = vfh;
2018 __u64 inum = fh[2];
2019 inum = (inum << 32) | fh[1];
2020 return ino->i_ino == inum && fh[0] == ino->i_generation;
2021}
2022
Christoph Hellwig480b1162007-10-21 16:42:13 -07002023static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2024 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002025{
David M. Grimes91828a42006-10-17 00:09:45 -07002026 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002027 struct dentry *dentry = NULL;
2028 u64 inum = fid->raw[2];
2029 inum = (inum << 32) | fid->raw[1];
David M. Grimes91828a42006-10-17 00:09:45 -07002030
Christoph Hellwig480b1162007-10-21 16:42:13 -07002031 if (fh_len < 3)
2032 return NULL;
2033
2034 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2035 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002036 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002037 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002038 iput(inode);
2039 }
2040
Christoph Hellwig480b1162007-10-21 16:42:13 -07002041 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002042}
2043
2044static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2045 int connectable)
2046{
2047 struct inode *inode = dentry->d_inode;
2048
2049 if (*len < 3)
2050 return 255;
2051
2052 if (hlist_unhashed(&inode->i_hash)) {
2053 /* Unfortunately insert_inode_hash is not idempotent,
2054 * so as we hash inodes here rather than at creation
2055 * time, we need a lock to ensure we only try
2056 * to do it once
2057 */
2058 static DEFINE_SPINLOCK(lock);
2059 spin_lock(&lock);
2060 if (hlist_unhashed(&inode->i_hash))
2061 __insert_inode_hash(inode,
2062 inode->i_ino + inode->i_generation);
2063 spin_unlock(&lock);
2064 }
2065
2066 fh[0] = inode->i_generation;
2067 fh[1] = inode->i_ino;
2068 fh[2] = ((__u64)inode->i_ino) >> 32;
2069
2070 *len = 3;
2071 return 1;
2072}
2073
Christoph Hellwig39655162007-10-21 16:42:17 -07002074static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002075 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002076 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002077 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002078};
2079
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002080static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2081 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082{
2083 char *this_char, *value, *rest;
2084
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002085 while (options != NULL) {
2086 this_char = options;
2087 for (;;) {
2088 /*
2089 * NUL-terminate this option: unfortunately,
2090 * mount options form a comma-separated list,
2091 * but mpol's nodelist may also contain commas.
2092 */
2093 options = strchr(options, ',');
2094 if (options == NULL)
2095 break;
2096 options++;
2097 if (!isdigit(*options)) {
2098 options[-1] = '\0';
2099 break;
2100 }
2101 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 if (!*this_char)
2103 continue;
2104 if ((value = strchr(this_char,'=')) != NULL) {
2105 *value++ = 0;
2106 } else {
2107 printk(KERN_ERR
2108 "tmpfs: No value for mount option '%s'\n",
2109 this_char);
2110 return 1;
2111 }
2112
2113 if (!strcmp(this_char,"size")) {
2114 unsigned long long size;
2115 size = memparse(value,&rest);
2116 if (*rest == '%') {
2117 size <<= PAGE_SHIFT;
2118 size *= totalram_pages;
2119 do_div(size, 100);
2120 rest++;
2121 }
2122 if (*rest)
2123 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002124 sbinfo->max_blocks =
2125 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002127 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 if (*rest)
2129 goto bad_val;
2130 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002131 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 if (*rest)
2133 goto bad_val;
2134 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002135 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002137 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 if (*rest)
2139 goto bad_val;
2140 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002141 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002143 sbinfo->uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 if (*rest)
2145 goto bad_val;
2146 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002147 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002149 sbinfo->gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 if (*rest)
2151 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002152 } else if (!strcmp(this_char,"mpol")) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002153 if (mpol_parse_str(value, &sbinfo->mpol, 1))
Robin Holt7339ff82006-01-14 13:20:48 -08002154 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 } else {
2156 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2157 this_char);
2158 return 1;
2159 }
2160 }
2161 return 0;
2162
2163bad_val:
2164 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2165 value, this_char);
2166 return 1;
2167
2168}
2169
2170static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2171{
2172 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002173 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002174 unsigned long blocks;
2175 unsigned long inodes;
2176 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002178 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002179 return error;
2180
2181 spin_lock(&sbinfo->stat_lock);
2182 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2183 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002184 if (config.max_blocks < blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002185 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002186 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002187 goto out;
2188 /*
2189 * Those tests also disallow limited->unlimited while any are in
2190 * use, so i_blocks will always be zero when max_blocks is zero;
2191 * but we must separately disallow unlimited->limited, because
2192 * in that case we have no record of how much is already in use.
2193 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002194 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002195 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002196 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002197 goto out;
2198
2199 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002200 sbinfo->max_blocks = config.max_blocks;
2201 sbinfo->free_blocks = config.max_blocks - blocks;
2202 sbinfo->max_inodes = config.max_inodes;
2203 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002204
2205 mpol_put(sbinfo->mpol);
2206 sbinfo->mpol = config.mpol; /* transfers initial ref */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002207out:
2208 spin_unlock(&sbinfo->stat_lock);
2209 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002211
2212static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2213{
2214 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2215
2216 if (sbinfo->max_blocks != shmem_default_max_blocks())
2217 seq_printf(seq, ",size=%luk",
2218 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2219 if (sbinfo->max_inodes != shmem_default_max_inodes())
2220 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2221 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2222 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2223 if (sbinfo->uid != 0)
2224 seq_printf(seq, ",uid=%u", sbinfo->uid);
2225 if (sbinfo->gid != 0)
2226 seq_printf(seq, ",gid=%u", sbinfo->gid);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002227 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002228 return 0;
2229}
2230#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231
2232static void shmem_put_super(struct super_block *sb)
2233{
2234 kfree(sb->s_fs_info);
2235 sb->s_fs_info = NULL;
2236}
2237
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238static int shmem_fill_super(struct super_block *sb,
2239 void *data, int silent)
2240{
2241 struct inode *inode;
2242 struct dentry *root;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002243 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002244 int err = -ENOMEM;
2245
2246 /* Round up to L1_CACHE_BYTES to resist false sharing */
2247 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2248 L1_CACHE_BYTES), GFP_KERNEL);
2249 if (!sbinfo)
2250 return -ENOMEM;
2251
2252 sbinfo->max_blocks = 0;
2253 sbinfo->max_inodes = 0;
2254 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2255 sbinfo->uid = current->fsuid;
2256 sbinfo->gid = current->fsgid;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002257 sbinfo->mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002258 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002260#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 /*
2262 * Per default we only allow half of the physical ram per
2263 * tmpfs instance, limiting inodes to one per page of lowmem;
2264 * but the internal instance is left unlimited.
2265 */
2266 if (!(sb->s_flags & MS_NOUSER)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002267 sbinfo->max_blocks = shmem_default_max_blocks();
2268 sbinfo->max_inodes = shmem_default_max_inodes();
2269 if (shmem_parse_options(data, sbinfo, false)) {
2270 err = -EINVAL;
2271 goto failed;
2272 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 }
David M. Grimes91828a42006-10-17 00:09:45 -07002274 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275#else
2276 sb->s_flags |= MS_NOUSER;
2277#endif
2278
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002279 spin_lock_init(&sbinfo->stat_lock);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002280 sbinfo->free_blocks = sbinfo->max_blocks;
2281 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002282
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 sb->s_maxbytes = SHMEM_MAX_BYTES;
2284 sb->s_blocksize = PAGE_CACHE_SIZE;
2285 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2286 sb->s_magic = TMPFS_MAGIC;
2287 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002288 sb->s_time_gran = 1;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002289#ifdef CONFIG_TMPFS_POSIX_ACL
2290 sb->s_xattr = shmem_xattr_handlers;
2291 sb->s_flags |= MS_POSIXACL;
2292#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002293
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002294 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 if (!inode)
2296 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002297 inode->i_uid = sbinfo->uid;
2298 inode->i_gid = sbinfo->gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 root = d_alloc_root(inode);
2300 if (!root)
2301 goto failed_iput;
2302 sb->s_root = root;
2303 return 0;
2304
2305failed_iput:
2306 iput(inode);
2307failed:
2308 shmem_put_super(sb);
2309 return err;
2310}
2311
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002312static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313
2314static struct inode *shmem_alloc_inode(struct super_block *sb)
2315{
2316 struct shmem_inode_info *p;
Christoph Lametere94b1762006-12-06 20:33:17 -08002317 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 if (!p)
2319 return NULL;
2320 return &p->vfs_inode;
2321}
2322
2323static void shmem_destroy_inode(struct inode *inode)
2324{
2325 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2326 /* only struct inode is valid if it's an inline symlink */
2327 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2328 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002329 shmem_acl_destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2331}
2332
Christoph Lameter4ba9b9d2007-10-16 23:25:51 -07002333static void init_once(struct kmem_cache *cachep, void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334{
2335 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2336
Christoph Lametera35afb82007-05-16 22:10:57 -07002337 inode_init_once(&p->vfs_inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002338#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Lametera35afb82007-05-16 22:10:57 -07002339 p->i_acl = NULL;
2340 p->i_default_acl = NULL;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002341#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342}
2343
2344static int init_inodecache(void)
2345{
2346 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2347 sizeof(struct shmem_inode_info),
Alexey Dobriyan040b5c62007-10-16 23:26:10 -07002348 0, SLAB_PANIC, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 return 0;
2350}
2351
2352static void destroy_inodecache(void)
2353{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002354 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355}
2356
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002357static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002359 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360#ifdef CONFIG_TMPFS
Hugh Dickinsae976412007-06-04 10:00:39 +02002361 .readpage = shmem_readpage,
Nick Piggin800d15a2007-10-16 01:25:03 -07002362 .write_begin = shmem_write_begin,
2363 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002365 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366};
2367
Helge Deller15ad7cd2006-12-06 20:40:36 -08002368static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 .mmap = shmem_mmap,
2370#ifdef CONFIG_TMPFS
2371 .llseek = generic_file_llseek,
2372 .read = shmem_file_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002373 .write = do_sync_write,
2374 .aio_write = generic_file_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 .fsync = simple_sync_file,
Hugh Dickinsae976412007-06-04 10:00:39 +02002376 .splice_read = generic_file_splice_read,
2377 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378#endif
2379};
2380
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002381static const struct inode_operations shmem_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 .truncate = shmem_truncate,
2383 .setattr = shmem_notify_change,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002384 .truncate_range = shmem_truncate_range,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002385#ifdef CONFIG_TMPFS_POSIX_ACL
2386 .setxattr = generic_setxattr,
2387 .getxattr = generic_getxattr,
2388 .listxattr = generic_listxattr,
2389 .removexattr = generic_removexattr,
2390 .permission = shmem_permission,
2391#endif
2392
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393};
2394
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002395static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396#ifdef CONFIG_TMPFS
2397 .create = shmem_create,
2398 .lookup = simple_lookup,
2399 .link = shmem_link,
2400 .unlink = shmem_unlink,
2401 .symlink = shmem_symlink,
2402 .mkdir = shmem_mkdir,
2403 .rmdir = shmem_rmdir,
2404 .mknod = shmem_mknod,
2405 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002407#ifdef CONFIG_TMPFS_POSIX_ACL
2408 .setattr = shmem_notify_change,
2409 .setxattr = generic_setxattr,
2410 .getxattr = generic_getxattr,
2411 .listxattr = generic_listxattr,
2412 .removexattr = generic_removexattr,
2413 .permission = shmem_permission,
2414#endif
2415};
2416
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002417static const struct inode_operations shmem_special_inode_operations = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002418#ifdef CONFIG_TMPFS_POSIX_ACL
2419 .setattr = shmem_notify_change,
2420 .setxattr = generic_setxattr,
2421 .getxattr = generic_getxattr,
2422 .listxattr = generic_listxattr,
2423 .removexattr = generic_removexattr,
2424 .permission = shmem_permission,
2425#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426};
2427
Hugh Dickins759b9772007-03-05 00:30:28 -08002428static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 .alloc_inode = shmem_alloc_inode,
2430 .destroy_inode = shmem_destroy_inode,
2431#ifdef CONFIG_TMPFS
2432 .statfs = shmem_statfs,
2433 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002434 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435#endif
2436 .delete_inode = shmem_delete_inode,
2437 .drop_inode = generic_delete_inode,
2438 .put_super = shmem_put_super,
2439};
2440
2441static struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002442 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443#ifdef CONFIG_NUMA
2444 .set_policy = shmem_set_policy,
2445 .get_policy = shmem_get_policy,
2446#endif
2447};
2448
2449
David Howells454e2392006-06-23 02:02:57 -07002450static int shmem_get_sb(struct file_system_type *fs_type,
2451 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452{
David Howells454e2392006-06-23 02:02:57 -07002453 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454}
2455
2456static struct file_system_type tmpfs_fs_type = {
2457 .owner = THIS_MODULE,
2458 .name = "tmpfs",
2459 .get_sb = shmem_get_sb,
2460 .kill_sb = kill_litter_super,
2461};
2462static struct vfsmount *shm_mnt;
2463
2464static int __init init_tmpfs(void)
2465{
2466 int error;
2467
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002468 error = bdi_init(&shmem_backing_dev_info);
2469 if (error)
2470 goto out4;
2471
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 error = init_inodecache();
2473 if (error)
2474 goto out3;
2475
2476 error = register_filesystem(&tmpfs_fs_type);
2477 if (error) {
2478 printk(KERN_ERR "Could not register tmpfs\n");
2479 goto out2;
2480 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002481
Trond Myklebust1f5ce9e2006-06-09 09:34:16 -04002482 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 tmpfs_fs_type.name, NULL);
2484 if (IS_ERR(shm_mnt)) {
2485 error = PTR_ERR(shm_mnt);
2486 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2487 goto out1;
2488 }
2489 return 0;
2490
2491out1:
2492 unregister_filesystem(&tmpfs_fs_type);
2493out2:
2494 destroy_inodecache();
2495out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002496 bdi_destroy(&shmem_backing_dev_info);
2497out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 shm_mnt = ERR_PTR(error);
2499 return error;
2500}
2501module_init(init_tmpfs)
2502
Randy Dunlap46711812008-03-19 17:00:41 -07002503/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 * shmem_file_setup - get an unlinked file living in tmpfs
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 * @name: name for dentry (to be seen in /proc/<pid>/maps
2506 * @size: size to be set for the file
Randy Dunlap46711812008-03-19 17:00:41 -07002507 * @flags: vm_flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 */
2509struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2510{
2511 int error;
2512 struct file *file;
2513 struct inode *inode;
2514 struct dentry *dentry, *root;
2515 struct qstr this;
2516
2517 if (IS_ERR(shm_mnt))
2518 return (void *)shm_mnt;
2519
2520 if (size < 0 || size > SHMEM_MAX_BYTES)
2521 return ERR_PTR(-EINVAL);
2522
2523 if (shmem_acct_size(flags, size))
2524 return ERR_PTR(-ENOMEM);
2525
2526 error = -ENOMEM;
2527 this.name = name;
2528 this.len = strlen(name);
2529 this.hash = 0; /* will go */
2530 root = shm_mnt->mnt_root;
2531 dentry = d_alloc(root, &this);
2532 if (!dentry)
2533 goto put_memory;
2534
2535 error = -ENFILE;
2536 file = get_empty_filp();
2537 if (!file)
2538 goto put_dentry;
2539
2540 error = -ENOSPC;
2541 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2542 if (!inode)
2543 goto close_file;
2544
2545 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2546 d_instantiate(dentry, inode);
2547 inode->i_size = size;
2548 inode->i_nlink = 0; /* It is unlinked */
Dave Hansence8d2cd2007-10-16 23:31:13 -07002549 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
2550 &shmem_file_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 return file;
2552
2553close_file:
2554 put_filp(file);
2555put_dentry:
2556 dput(dentry);
2557put_memory:
2558 shmem_unacct_size(flags, size);
2559 return ERR_PTR(error);
2560}
2561
Randy Dunlap46711812008-03-19 17:00:41 -07002562/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2565 */
2566int shmem_zero_setup(struct vm_area_struct *vma)
2567{
2568 struct file *file;
2569 loff_t size = vma->vm_end - vma->vm_start;
2570
2571 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2572 if (IS_ERR(file))
2573 return PTR_ERR(file);
2574
2575 if (vma->vm_file)
2576 fput(vma->vm_file);
2577 vma->vm_file = file;
2578 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 return 0;
2580}