blob: 356dd99566ecb671cd324ba3dddcee10441312f0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Resizable virtual memory filesystem for Linux.
3 *
4 * Copyright (C) 2000 Linus Torvalds.
5 * 2000 Transmeta Corp.
6 * 2000-2001 Christoph Rohland
7 * 2000-2001 SAP AG
8 * 2002 Red Hat Inc.
Hugh Dickins0edd73b2005-06-21 17:15:04 -07009 * Copyright (C) 2002-2005 Hugh Dickins.
10 * Copyright (C) 2002-2005 VERITAS Software Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Copyright (C) 2004 Andi Kleen, SuSE Labs
12 *
13 * Extended attribute support for tmpfs:
14 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
15 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
16 *
Matt Mackall853ac432009-01-06 14:40:20 -080017 * tiny-shmem:
18 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
19 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070020 * This file is released under the GPL.
21 */
22
Matt Mackall853ac432009-01-06 14:40:20 -080023#include <linux/fs.h>
24#include <linux/init.h>
25#include <linux/vfs.h>
26#include <linux/mount.h>
Hugh Dickinscaefba12009-04-13 14:40:12 -070027#include <linux/pagemap.h>
Matt Mackall853ac432009-01-06 14:40:20 -080028#include <linux/file.h>
29#include <linux/mm.h>
30#include <linux/module.h>
31#include <linux/swap.h>
Mimi Zohared850a52009-02-10 23:01:19 -050032#include <linux/ima.h>
Matt Mackall853ac432009-01-06 14:40:20 -080033
34static struct vfsmount *shm_mnt;
35
36#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/*
38 * This virtual memory filesystem is heavily based on the ramfs. It
39 * extends ramfs by the ability to use swap and honor resource limits
40 * which makes it a completely usable filesystem.
41 */
42
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070043#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070044#include <linux/exportfs.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070045#include <linux/generic_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/string.h>
48#include <linux/slab.h>
49#include <linux/backing-dev.h>
50#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/blkdev.h>
53#include <linux/security.h>
54#include <linux/swapops.h>
55#include <linux/mempolicy.h>
56#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000057#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070058#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070059#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080060#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040061#include <linux/magic.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070062
Linus Torvalds1da177e2005-04-16 15:20:36 -070063#include <asm/uaccess.h>
64#include <asm/div64.h>
65#include <asm/pgtable.h>
66
Hugh Dickinscaefba12009-04-13 14:40:12 -070067/*
68 * The maximum size of a shmem/tmpfs file is limited by the maximum size of
69 * its triple-indirect swap vector - see illustration at shmem_swp_entry().
70 *
71 * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
72 * but one eighth of that on a 64-bit kernel. With 8kB page size, maximum
73 * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
74 * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
75 *
76 * We use / and * instead of shifts in the definitions below, so that the swap
77 * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
78 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
Yuri Tikhonov61609d02009-04-13 14:40:11 -070080#define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
Hugh Dickinscaefba12009-04-13 14:40:12 -070081
82#define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
83#define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
84
85#define SHMEM_MAX_BYTES min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
86#define SHMEM_MAX_INDEX ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
90
91/* info->flags needs VM_flags to handle pagein/truncate races efficiently */
92#define SHMEM_PAGEIN VM_READ
93#define SHMEM_TRUNCATE VM_WRITE
94
95/* Definition to limit shmem_truncate's steps between cond_rescheds */
96#define LATENCY_LIMIT 64
97
98/* Pretend that each entry is of this size in directory's i_size */
99#define BOGO_DIRENT_SIZE 20
100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
102enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 SGP_READ, /* don't exceed i_size, don't allocate page */
104 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -0800105 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 SGP_WRITE, /* may exceed i_size, may allocate page */
107};
108
Andrew Mortonb76db732008-02-08 04:21:49 -0800109#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800110static unsigned long shmem_default_max_blocks(void)
111{
112 return totalram_pages / 2;
113}
114
115static unsigned long shmem_default_max_inodes(void)
116{
117 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
118}
Andrew Mortonb76db732008-02-08 04:21:49 -0800119#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121static int shmem_getpage(struct inode *inode, unsigned long idx,
122 struct page **pagep, enum sgp_type sgp, int *type);
123
Al Viro6daa0e22005-10-21 03:18:50 -0400124static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125{
126 /*
127 * The above definition of ENTRIES_PER_PAGE, and the use of
128 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
129 * might be reconsidered if it ever diverges from PAGE_SIZE.
Mel Gorman769848c2007-07-17 04:03:05 -0700130 *
Mel Gormane12ba742007-10-16 01:25:52 -0700131 * Mobility flags are masked out as swap vectors cannot move
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 */
Mel Gormane12ba742007-10-16 01:25:52 -0700133 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
Mel Gorman769848c2007-07-17 04:03:05 -0700134 PAGE_CACHE_SHIFT-PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135}
136
137static inline void shmem_dir_free(struct page *page)
138{
139 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
140}
141
142static struct page **shmem_dir_map(struct page *page)
143{
144 return (struct page **)kmap_atomic(page, KM_USER0);
145}
146
147static inline void shmem_dir_unmap(struct page **dir)
148{
149 kunmap_atomic(dir, KM_USER0);
150}
151
152static swp_entry_t *shmem_swp_map(struct page *page)
153{
154 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
155}
156
157static inline void shmem_swp_balance_unmap(void)
158{
159 /*
160 * When passing a pointer to an i_direct entry, to code which
161 * also handles indirect entries and so will shmem_swp_unmap,
162 * we must arrange for the preempt count to remain in balance.
163 * What kmap_atomic of a lowmem page does depends on config
164 * and architecture, so pretend to kmap_atomic some lowmem page.
165 */
166 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
167}
168
169static inline void shmem_swp_unmap(swp_entry_t *entry)
170{
171 kunmap_atomic(entry, KM_USER1);
172}
173
174static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
175{
176 return sb->s_fs_info;
177}
178
179/*
180 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
181 * for shared memory and for shared anonymous (/dev/zero) mappings
182 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
183 * consistent with the pre-accounting of private mappings ...
184 */
185static inline int shmem_acct_size(unsigned long flags, loff_t size)
186{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000187 return (flags & VM_NORESERVE) ?
188 0 : security_vm_enough_memory_kern(VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189}
190
191static inline void shmem_unacct_size(unsigned long flags, loff_t size)
192{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000193 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 vm_unacct_memory(VM_ACCT(size));
195}
196
197/*
198 * ... whereas tmpfs objects are accounted incrementally as
199 * pages are allocated, in order to allow huge sparse files.
200 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
201 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
202 */
203static inline int shmem_acct_block(unsigned long flags)
204{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000205 return (flags & VM_NORESERVE) ?
206 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
209static inline void shmem_unacct_blocks(unsigned long flags, long pages)
210{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000211 if (flags & VM_NORESERVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
213}
214
Hugh Dickins759b9772007-03-05 00:30:28 -0800215static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700216static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800217static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800218static const struct inode_operations shmem_inode_operations;
219static const struct inode_operations shmem_dir_inode_operations;
220static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400221static const struct vm_operations_struct shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700223static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 .ra_pages = 0, /* No readahead */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700225 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 .unplug_io_fn = default_unplug_io_fn,
227};
228
229static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800230static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
232static void shmem_free_blocks(struct inode *inode, long pages)
233{
234 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700235 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 spin_lock(&sbinfo->stat_lock);
237 sbinfo->free_blocks += pages;
238 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
239 spin_unlock(&sbinfo->stat_lock);
240 }
241}
242
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800243static int shmem_reserve_inode(struct super_block *sb)
244{
245 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
246 if (sbinfo->max_inodes) {
247 spin_lock(&sbinfo->stat_lock);
248 if (!sbinfo->free_inodes) {
249 spin_unlock(&sbinfo->stat_lock);
250 return -ENOSPC;
251 }
252 sbinfo->free_inodes--;
253 spin_unlock(&sbinfo->stat_lock);
254 }
255 return 0;
256}
257
258static void shmem_free_inode(struct super_block *sb)
259{
260 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
261 if (sbinfo->max_inodes) {
262 spin_lock(&sbinfo->stat_lock);
263 sbinfo->free_inodes++;
264 spin_unlock(&sbinfo->stat_lock);
265 }
266}
267
Randy Dunlap46711812008-03-19 17:00:41 -0700268/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 * shmem_recalc_inode - recalculate the size of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 * @inode: inode to recalc
271 *
272 * We have to calculate the free blocks since the mm can drop
273 * undirtied hole pages behind our back.
274 *
275 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
276 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
277 *
278 * It has to be called with the spinlock held.
279 */
280static void shmem_recalc_inode(struct inode *inode)
281{
282 struct shmem_inode_info *info = SHMEM_I(inode);
283 long freed;
284
285 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
286 if (freed > 0) {
287 info->alloced -= freed;
288 shmem_unacct_blocks(info->flags, freed);
289 shmem_free_blocks(inode, freed);
290 }
291}
292
Randy Dunlap46711812008-03-19 17:00:41 -0700293/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 * shmem_swp_entry - find the swap vector position in the info structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 * @info: info structure for the inode
296 * @index: index of the page to find
297 * @page: optional page to add to the structure. Has to be preset to
298 * all zeros
299 *
300 * If there is no space allocated yet it will return NULL when
301 * page is NULL, else it will use the page for the needed block,
302 * setting it to NULL on return to indicate that it has been used.
303 *
304 * The swap vector is organized the following way:
305 *
306 * There are SHMEM_NR_DIRECT entries directly stored in the
307 * shmem_inode_info structure. So small files do not need an addional
308 * allocation.
309 *
310 * For pages with index > SHMEM_NR_DIRECT there is the pointer
311 * i_indirect which points to a page which holds in the first half
312 * doubly indirect blocks, in the second half triple indirect blocks:
313 *
314 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
315 * following layout (for SHMEM_NR_DIRECT == 16):
316 *
317 * i_indirect -> dir --> 16-19
318 * | +-> 20-23
319 * |
320 * +-->dir2 --> 24-27
321 * | +-> 28-31
322 * | +-> 32-35
323 * | +-> 36-39
324 * |
325 * +-->dir3 --> 40-43
326 * +-> 44-47
327 * +-> 48-51
328 * +-> 52-55
329 */
330static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
331{
332 unsigned long offset;
333 struct page **dir;
334 struct page *subdir;
335
336 if (index < SHMEM_NR_DIRECT) {
337 shmem_swp_balance_unmap();
338 return info->i_direct+index;
339 }
340 if (!info->i_indirect) {
341 if (page) {
342 info->i_indirect = *page;
343 *page = NULL;
344 }
345 return NULL; /* need another page */
346 }
347
348 index -= SHMEM_NR_DIRECT;
349 offset = index % ENTRIES_PER_PAGE;
350 index /= ENTRIES_PER_PAGE;
351 dir = shmem_dir_map(info->i_indirect);
352
353 if (index >= ENTRIES_PER_PAGE/2) {
354 index -= ENTRIES_PER_PAGE/2;
355 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
356 index %= ENTRIES_PER_PAGE;
357 subdir = *dir;
358 if (!subdir) {
359 if (page) {
360 *dir = *page;
361 *page = NULL;
362 }
363 shmem_dir_unmap(dir);
364 return NULL; /* need another page */
365 }
366 shmem_dir_unmap(dir);
367 dir = shmem_dir_map(subdir);
368 }
369
370 dir += index;
371 subdir = *dir;
372 if (!subdir) {
373 if (!page || !(subdir = *page)) {
374 shmem_dir_unmap(dir);
375 return NULL; /* need a page */
376 }
377 *dir = subdir;
378 *page = NULL;
379 }
380 shmem_dir_unmap(dir);
381 return shmem_swp_map(subdir) + offset;
382}
383
384static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
385{
386 long incdec = value? 1: -1;
387
388 entry->val = value;
389 info->swapped += incdec;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700390 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
391 struct page *page = kmap_atomic_to_page(entry);
392 set_page_private(page, page_private(page) + incdec);
393 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
395
Randy Dunlap46711812008-03-19 17:00:41 -0700396/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 * shmem_swp_alloc - get the position of the swap entry for the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 * @info: info structure for the inode
399 * @index: index of the page to find
400 * @sgp: check and recheck i_size? skip allocation?
Randy Dunlap46711812008-03-19 17:00:41 -0700401 *
402 * If the entry does not exist, allocate it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 */
404static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
405{
406 struct inode *inode = &info->vfs_inode;
407 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
408 struct page *page = NULL;
409 swp_entry_t *entry;
410
411 if (sgp != SGP_WRITE &&
412 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
413 return ERR_PTR(-EINVAL);
414
415 while (!(entry = shmem_swp_entry(info, index, &page))) {
416 if (sgp == SGP_READ)
417 return shmem_swp_map(ZERO_PAGE(0));
418 /*
419 * Test free_blocks against 1 not 0, since we have 1 data
420 * page (and perhaps indirect index pages) yet to allocate:
421 * a waste to allocate index if we cannot allocate data.
422 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700423 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 spin_lock(&sbinfo->stat_lock);
425 if (sbinfo->free_blocks <= 1) {
426 spin_unlock(&sbinfo->stat_lock);
427 return ERR_PTR(-ENOSPC);
428 }
429 sbinfo->free_blocks--;
430 inode->i_blocks += BLOCKS_PER_PAGE;
431 spin_unlock(&sbinfo->stat_lock);
432 }
433
434 spin_unlock(&info->lock);
Mel Gorman769848c2007-07-17 04:03:05 -0700435 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700436 if (page)
437 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 spin_lock(&info->lock);
439
440 if (!page) {
441 shmem_free_blocks(inode, 1);
442 return ERR_PTR(-ENOMEM);
443 }
444 if (sgp != SGP_WRITE &&
445 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
446 entry = ERR_PTR(-EINVAL);
447 break;
448 }
449 if (info->next_index <= index)
450 info->next_index = index + 1;
451 }
452 if (page) {
453 /* another task gave its page, or truncated the file */
454 shmem_free_blocks(inode, 1);
455 shmem_dir_free(page);
456 }
457 if (info->next_index <= index && !IS_ERR(entry))
458 info->next_index = index + 1;
459 return entry;
460}
461
Randy Dunlap46711812008-03-19 17:00:41 -0700462/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 * shmem_free_swp - free some swap entries in a directory
Hugh Dickins1ae70002007-03-29 01:20:36 -0700464 * @dir: pointer to the directory
465 * @edir: pointer after last entry of the directory
466 * @punch_lock: pointer to spinlock when needed for the holepunch case
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 */
Hugh Dickins1ae70002007-03-29 01:20:36 -0700468static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
469 spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470{
Hugh Dickins1ae70002007-03-29 01:20:36 -0700471 spinlock_t *punch_unlock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 swp_entry_t *ptr;
473 int freed = 0;
474
475 for (ptr = dir; ptr < edir; ptr++) {
476 if (ptr->val) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700477 if (unlikely(punch_lock)) {
478 punch_unlock = punch_lock;
479 punch_lock = NULL;
480 spin_lock(punch_unlock);
481 if (!ptr->val)
482 continue;
483 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 free_swap_and_cache(*ptr);
485 *ptr = (swp_entry_t){0};
486 freed++;
487 }
488 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700489 if (punch_unlock)
490 spin_unlock(punch_unlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 return freed;
492}
493
Hugh Dickins1ae70002007-03-29 01:20:36 -0700494static int shmem_map_and_free_swp(struct page *subdir, int offset,
495 int limit, struct page ***dir, spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
497 swp_entry_t *ptr;
498 int freed = 0;
499
500 ptr = shmem_swp_map(subdir);
501 for (; offset < limit; offset += LATENCY_LIMIT) {
502 int size = limit - offset;
503 if (size > LATENCY_LIMIT)
504 size = LATENCY_LIMIT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700505 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
506 punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 if (need_resched()) {
508 shmem_swp_unmap(ptr);
509 if (*dir) {
510 shmem_dir_unmap(*dir);
511 *dir = NULL;
512 }
513 cond_resched();
514 ptr = shmem_swp_map(subdir);
515 }
516 }
517 shmem_swp_unmap(ptr);
518 return freed;
519}
520
521static void shmem_free_pages(struct list_head *next)
522{
523 struct page *page;
524 int freed = 0;
525
526 do {
527 page = container_of(next, struct page, lru);
528 next = next->next;
529 shmem_dir_free(page);
530 freed++;
531 if (freed >= LATENCY_LIMIT) {
532 cond_resched();
533 freed = 0;
534 }
535 } while (next);
536}
537
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800538static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539{
540 struct shmem_inode_info *info = SHMEM_I(inode);
541 unsigned long idx;
542 unsigned long size;
543 unsigned long limit;
544 unsigned long stage;
545 unsigned long diroff;
546 struct page **dir;
547 struct page *topdir;
548 struct page *middir;
549 struct page *subdir;
550 swp_entry_t *ptr;
551 LIST_HEAD(pages_to_free);
552 long nr_pages_to_free = 0;
553 long nr_swaps_freed = 0;
554 int offset;
555 int freed;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700556 int punch_hole;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700557 spinlock_t *needs_lock;
558 spinlock_t *punch_lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700559 unsigned long upper_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800562 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 if (idx >= info->next_index)
564 return;
565
566 spin_lock(&info->lock);
567 info->flags |= SHMEM_TRUNCATE;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800568 if (likely(end == (loff_t) -1)) {
569 limit = info->next_index;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700570 upper_limit = SHMEM_MAX_INDEX;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800571 info->next_index = idx;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700572 needs_lock = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700573 punch_hole = 0;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800574 } else {
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700575 if (end + 1 >= inode->i_size) { /* we may free a little more */
576 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
577 PAGE_CACHE_SHIFT;
578 upper_limit = SHMEM_MAX_INDEX;
579 } else {
580 limit = (end + 1) >> PAGE_CACHE_SHIFT;
581 upper_limit = limit;
582 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700583 needs_lock = &info->lock;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800584 punch_hole = 1;
585 }
586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 topdir = info->i_indirect;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800588 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 info->i_indirect = NULL;
590 nr_pages_to_free++;
591 list_add(&topdir->lru, &pages_to_free);
592 }
593 spin_unlock(&info->lock);
594
595 if (info->swapped && idx < SHMEM_NR_DIRECT) {
596 ptr = info->i_direct;
597 size = limit;
598 if (size > SHMEM_NR_DIRECT)
599 size = SHMEM_NR_DIRECT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700600 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 }
Badari Pulavarty92a3d032006-12-22 01:06:23 -0800602
603 /*
604 * If there are no indirect blocks or we are punching a hole
605 * below indirect blocks, nothing to be done.
606 */
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700607 if (!topdir || limit <= SHMEM_NR_DIRECT)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 goto done2;
609
Hugh Dickins1ae70002007-03-29 01:20:36 -0700610 /*
611 * The truncation case has already dropped info->lock, and we're safe
612 * because i_size and next_index have already been lowered, preventing
613 * access beyond. But in the punch_hole case, we still need to take
614 * the lock when updating the swap directory, because there might be
615 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
616 * shmem_writepage. However, whenever we find we can remove a whole
617 * directory page (not at the misaligned start or end of the range),
618 * we first NULLify its pointer in the level above, and then have no
619 * need to take the lock when updating its contents: needs_lock and
620 * punch_lock (either pointing to info->lock or NULL) manage this.
621 */
622
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700623 upper_limit -= SHMEM_NR_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 limit -= SHMEM_NR_DIRECT;
625 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
626 offset = idx % ENTRIES_PER_PAGE;
627 idx -= offset;
628
629 dir = shmem_dir_map(topdir);
630 stage = ENTRIES_PER_PAGEPAGE/2;
631 if (idx < ENTRIES_PER_PAGEPAGE/2) {
632 middir = topdir;
633 diroff = idx/ENTRIES_PER_PAGE;
634 } else {
635 dir += ENTRIES_PER_PAGE/2;
636 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
637 while (stage <= idx)
638 stage += ENTRIES_PER_PAGEPAGE;
639 middir = *dir;
640 if (*dir) {
641 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
642 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700643 if (!diroff && !offset && upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700644 if (needs_lock) {
645 spin_lock(needs_lock);
646 *dir = NULL;
647 spin_unlock(needs_lock);
648 needs_lock = NULL;
649 } else
650 *dir = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 nr_pages_to_free++;
652 list_add(&middir->lru, &pages_to_free);
653 }
654 shmem_dir_unmap(dir);
655 dir = shmem_dir_map(middir);
656 } else {
657 diroff = 0;
658 offset = 0;
659 idx = stage;
660 }
661 }
662
663 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
664 if (unlikely(idx == stage)) {
665 shmem_dir_unmap(dir);
666 dir = shmem_dir_map(topdir) +
667 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
668 while (!*dir) {
669 dir++;
670 idx += ENTRIES_PER_PAGEPAGE;
671 if (idx >= limit)
672 goto done1;
673 }
674 stage = idx + ENTRIES_PER_PAGEPAGE;
675 middir = *dir;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700676 if (punch_hole)
677 needs_lock = &info->lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700678 if (upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700679 if (needs_lock) {
680 spin_lock(needs_lock);
681 *dir = NULL;
682 spin_unlock(needs_lock);
683 needs_lock = NULL;
684 } else
685 *dir = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700686 nr_pages_to_free++;
687 list_add(&middir->lru, &pages_to_free);
688 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 shmem_dir_unmap(dir);
690 cond_resched();
691 dir = shmem_dir_map(middir);
692 diroff = 0;
693 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700694 punch_lock = needs_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 subdir = dir[diroff];
Hugh Dickins1ae70002007-03-29 01:20:36 -0700696 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
697 if (needs_lock) {
698 spin_lock(needs_lock);
699 dir[diroff] = NULL;
700 spin_unlock(needs_lock);
701 punch_lock = NULL;
702 } else
703 dir[diroff] = NULL;
704 nr_pages_to_free++;
705 list_add(&subdir->lru, &pages_to_free);
706 }
707 if (subdir && page_private(subdir) /* has swap entries */) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 size = limit - idx;
709 if (size > ENTRIES_PER_PAGE)
710 size = ENTRIES_PER_PAGE;
711 freed = shmem_map_and_free_swp(subdir,
Hugh Dickins1ae70002007-03-29 01:20:36 -0700712 offset, size, &dir, punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (!dir)
714 dir = shmem_dir_map(middir);
715 nr_swaps_freed += freed;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700716 if (offset || punch_lock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 spin_lock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700718 set_page_private(subdir,
719 page_private(subdir) - freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 spin_unlock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700721 } else
722 BUG_ON(page_private(subdir) != freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700724 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 }
726done1:
727 shmem_dir_unmap(dir);
728done2:
729 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
730 /*
731 * Call truncate_inode_pages again: racing shmem_unuse_inode
732 * may have swizzled a page in from swap since vmtruncate or
733 * generic_delete_inode did it, before we lowered next_index.
734 * Also, though shmem_getpage checks i_size before adding to
735 * cache, no recheck after: so fix the narrow window there too.
Hugh Dickins16a10012007-03-29 01:20:37 -0700736 *
737 * Recalling truncate_inode_pages_range and unmap_mapping_range
738 * every time for punch_hole (which never got a chance to clear
739 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
740 * yet hardly ever necessary: try to optimize them out later.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800742 truncate_inode_pages_range(inode->i_mapping, start, end);
Hugh Dickins16a10012007-03-29 01:20:37 -0700743 if (punch_hole)
744 unmap_mapping_range(inode->i_mapping, start,
745 end - start, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 }
747
748 spin_lock(&info->lock);
749 info->flags &= ~SHMEM_TRUNCATE;
750 info->swapped -= nr_swaps_freed;
751 if (nr_pages_to_free)
752 shmem_free_blocks(inode, nr_pages_to_free);
753 shmem_recalc_inode(inode);
754 spin_unlock(&info->lock);
755
756 /*
757 * Empty swap vector directory pages to be freed?
758 */
759 if (!list_empty(&pages_to_free)) {
760 pages_to_free.prev->next = NULL;
761 shmem_free_pages(pages_to_free.next);
762 }
763}
764
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800765static void shmem_truncate(struct inode *inode)
766{
767 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
768}
769
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
771{
772 struct inode *inode = dentry->d_inode;
773 struct page *page = NULL;
774 int error;
775
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700776 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 if (attr->ia_size < inode->i_size) {
778 /*
779 * If truncating down to a partial page, then
780 * if that page is already allocated, hold it
781 * in memory until the truncation is over, so
782 * truncate_partial_page cannnot miss it were
783 * it assigned to swap.
784 */
785 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
786 (void) shmem_getpage(inode,
787 attr->ia_size>>PAGE_CACHE_SHIFT,
788 &page, SGP_READ, NULL);
Hugh Dickinsd3602442008-02-04 22:28:44 -0800789 if (page)
790 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 }
792 /*
793 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
794 * detect if any pages might have been added to cache
795 * after truncate_inode_pages. But we needn't bother
796 * if it's being fully truncated to zero-length: the
797 * nrpages check is efficient enough in that case.
798 */
799 if (attr->ia_size) {
800 struct shmem_inode_info *info = SHMEM_I(inode);
801 spin_lock(&info->lock);
802 info->flags &= ~SHMEM_PAGEIN;
803 spin_unlock(&info->lock);
804 }
805 }
806 }
807
808 error = inode_change_ok(inode, attr);
809 if (!error)
810 error = inode_setattr(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700811#ifdef CONFIG_TMPFS_POSIX_ACL
812 if (!error && (attr->ia_valid & ATTR_MODE))
813 error = generic_acl_chmod(inode, &shmem_acl_ops);
814#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 if (page)
816 page_cache_release(page);
817 return error;
818}
819
820static void shmem_delete_inode(struct inode *inode)
821{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 struct shmem_inode_info *info = SHMEM_I(inode);
823
824 if (inode->i_op->truncate == shmem_truncate) {
Mark Fashehfef26652005-09-09 13:01:31 -0700825 truncate_inode_pages(inode->i_mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 shmem_unacct_size(info->flags, inode->i_size);
827 inode->i_size = 0;
828 shmem_truncate(inode);
829 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800830 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800832 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 }
834 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700835 BUG_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800836 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 clear_inode(inode);
838}
839
840static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
841{
842 swp_entry_t *ptr;
843
844 for (ptr = dir; ptr < edir; ptr++) {
845 if (ptr->val == entry.val)
846 return ptr - dir;
847 }
848 return -1;
849}
850
851static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
852{
853 struct inode *inode;
854 unsigned long idx;
855 unsigned long size;
856 unsigned long limit;
857 unsigned long stage;
858 struct page **dir;
859 struct page *subdir;
860 swp_entry_t *ptr;
861 int offset;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800862 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864 idx = 0;
865 ptr = info->i_direct;
866 spin_lock(&info->lock);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800867 if (!info->swapped) {
868 list_del_init(&info->swaplist);
869 goto lost2;
870 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 limit = info->next_index;
872 size = limit;
873 if (size > SHMEM_NR_DIRECT)
874 size = SHMEM_NR_DIRECT;
875 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800876 if (offset >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 goto found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 if (!info->i_indirect)
879 goto lost2;
880
881 dir = shmem_dir_map(info->i_indirect);
882 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
883
884 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
885 if (unlikely(idx == stage)) {
886 shmem_dir_unmap(dir-1);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800887 if (cond_resched_lock(&info->lock)) {
888 /* check it has not been truncated */
889 if (limit > info->next_index) {
890 limit = info->next_index;
891 if (idx >= limit)
892 goto lost2;
893 }
894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 dir = shmem_dir_map(info->i_indirect) +
896 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
897 while (!*dir) {
898 dir++;
899 idx += ENTRIES_PER_PAGEPAGE;
900 if (idx >= limit)
901 goto lost1;
902 }
903 stage = idx + ENTRIES_PER_PAGEPAGE;
904 subdir = *dir;
905 shmem_dir_unmap(dir);
906 dir = shmem_dir_map(subdir);
907 }
908 subdir = *dir;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700909 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 ptr = shmem_swp_map(subdir);
911 size = limit - idx;
912 if (size > ENTRIES_PER_PAGE)
913 size = ENTRIES_PER_PAGE;
914 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800915 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 if (offset >= 0) {
917 shmem_dir_unmap(dir);
918 goto found;
919 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 }
921 }
922lost1:
923 shmem_dir_unmap(dir-1);
924lost2:
925 spin_unlock(&info->lock);
926 return 0;
927found:
928 idx += offset;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800929 inode = igrab(&info->vfs_inode);
930 spin_unlock(&info->lock);
931
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800932 /*
933 * Move _head_ to start search for next from here.
934 * But be careful: shmem_delete_inode checks list_empty without taking
935 * mutex, and there's an instant in list_move_tail when info->swaplist
936 * would appear empty, if it were the only one on shmem_swaplist. We
937 * could avoid doing it if inode NULL; or use this minor optimization.
938 */
939 if (shmem_swaplist.next != &info->swaplist)
940 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800941 mutex_unlock(&shmem_swaplist_mutex);
942
943 error = 1;
944 if (!inode)
945 goto out;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800946 /*
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -0800947 * Charge page using GFP_KERNEL while we can wait.
948 * Charged back to the user(not to caller) when swap account is used.
949 * add_to_page_cache() will be called with GFP_NOWAIT.
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800950 */
Hugh Dickins82369552008-02-07 00:14:22 -0800951 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800952 if (error)
953 goto out;
Hugh Dickins82369552008-02-07 00:14:22 -0800954 error = radix_tree_preload(GFP_KERNEL);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700955 if (error) {
956 mem_cgroup_uncharge_cache_page(page);
957 goto out;
958 }
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800959 error = 1;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800960
961 spin_lock(&info->lock);
962 ptr = shmem_swp_entry(info, idx, NULL);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700963 if (ptr && ptr->val == entry.val) {
Nick Piggine2867812008-07-25 19:45:30 -0700964 error = add_to_page_cache_locked(page, inode->i_mapping,
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800965 idx, GFP_NOWAIT);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700966 /* does mem_cgroup_uncharge_cache_page on error */
967 } else /* we must compensate for our precharge above */
968 mem_cgroup_uncharge_cache_page(page);
969
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800970 if (error == -EEXIST) {
971 struct page *filepage = find_get_page(inode->i_mapping, idx);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800972 error = 1;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800973 if (filepage) {
974 /*
975 * There might be a more uptodate page coming down
976 * from a stacked writepage: forget our swappage if so.
977 */
978 if (PageUptodate(filepage))
979 error = 0;
980 page_cache_release(filepage);
981 }
982 }
983 if (!error) {
Hugh Dickins73b12622008-02-04 22:28:50 -0800984 delete_from_swap_cache(page);
985 set_page_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 info->flags |= SHMEM_PAGEIN;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800987 shmem_swp_set(info, ptr, 0);
988 swap_free(entry);
989 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800991 if (ptr)
992 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 spin_unlock(&info->lock);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800994 radix_tree_preload_end();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800995out:
996 unlock_page(page);
997 page_cache_release(page);
998 iput(inode); /* allows for NULL */
999 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000}
1001
1002/*
1003 * shmem_unuse() search for an eventually swapped out shmem page.
1004 */
1005int shmem_unuse(swp_entry_t entry, struct page *page)
1006{
1007 struct list_head *p, *next;
1008 struct shmem_inode_info *info;
1009 int found = 0;
1010
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001011 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 list_for_each_safe(p, next, &shmem_swaplist) {
1013 info = list_entry(p, struct shmem_inode_info, swaplist);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001014 found = shmem_unuse_inode(info, entry, page);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001015 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001016 if (found)
1017 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001019 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001020out: return found; /* 0 or 1 or -ENOMEM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021}
1022
1023/*
1024 * Move the page from the page cache to the swap cache.
1025 */
1026static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1027{
1028 struct shmem_inode_info *info;
1029 swp_entry_t *entry, swap;
1030 struct address_space *mapping;
1031 unsigned long index;
1032 struct inode *inode;
1033
1034 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 mapping = page->mapping;
1036 index = page->index;
1037 inode = mapping->host;
1038 info = SHMEM_I(inode);
1039 if (info->flags & VM_LOCKED)
1040 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001041 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 goto redirty;
1043
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001044 /*
1045 * shmem_backing_dev_info's capabilities prevent regular writeback or
1046 * sync from ever calling shmem_writepage; but a stacking filesystem
1047 * may use the ->writepage of its underlying filesystem, in which case
1048 * tmpfs should write out to swap only in response to memory pressure,
Jens Axboe5b0830c2009-09-23 19:37:09 +02001049 * and not for the writeback threads or sync. However, in those cases,
1050 * we do still want to check if there's a redundant swappage to be
1051 * discarded.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001052 */
1053 if (wbc->for_reclaim)
1054 swap = get_swap_page();
1055 else
1056 swap.val = 0;
1057
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 if (index >= info->next_index) {
1060 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1061 goto unlock;
1062 }
1063 entry = shmem_swp_entry(info, index, NULL);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001064 if (entry->val) {
1065 /*
1066 * The more uptodate page coming down from a stacked
1067 * writepage should replace our old swappage.
1068 */
1069 free_swap_and_cache(*entry);
1070 shmem_swp_set(info, entry, 0);
1071 }
1072 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001074 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickins73b12622008-02-04 22:28:50 -08001075 remove_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 shmem_swp_set(info, entry, swap.val);
1077 shmem_swp_unmap(entry);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001078 if (list_empty(&info->swaplist))
1079 inode = igrab(inode);
1080 else
1081 inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 spin_unlock(&info->lock);
Hugh Dickins73b12622008-02-04 22:28:50 -08001083 swap_duplicate(swap);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001084 BUG_ON(page_mapped(page));
Hugh Dickins73b12622008-02-04 22:28:50 -08001085 page_cache_release(page); /* pagecache ref */
Hugh Dickins9fab5612009-03-31 15:23:33 -07001086 swap_writepage(page, wbc);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001087 if (inode) {
1088 mutex_lock(&shmem_swaplist_mutex);
1089 /* move instead of add in case we're racing */
1090 list_move_tail(&info->swaplist, &shmem_swaplist);
1091 mutex_unlock(&shmem_swaplist_mutex);
1092 iput(inode);
1093 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 return 0;
1095 }
1096
1097 shmem_swp_unmap(entry);
1098unlock:
1099 spin_unlock(&info->lock);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -07001100 /*
1101 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1102 * clear SWAP_HAS_CACHE flag.
1103 */
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07001104 swapcache_free(swap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105redirty:
1106 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001107 if (wbc->for_reclaim)
1108 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1109 unlock_page(page);
1110 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111}
1112
1113#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001114#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001115static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001116{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001117 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001118
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001119 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001120 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001121
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001122 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001123
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001124 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001125}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001126
1127static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1128{
1129 struct mempolicy *mpol = NULL;
1130 if (sbinfo->mpol) {
1131 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1132 mpol = sbinfo->mpol;
1133 mpol_get(mpol);
1134 spin_unlock(&sbinfo->stat_lock);
1135 }
1136 return mpol;
1137}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001138#endif /* CONFIG_TMPFS */
1139
Hugh Dickins02098fe2008-02-04 22:28:42 -08001140static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1141 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142{
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001143 struct mempolicy mpol, *spol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 struct vm_area_struct pvma;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001145 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001147 spol = mpol_cond_copy(&mpol,
1148 mpol_shared_policy_lookup(&info->policy, idx));
1149
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 /* Create a pseudo vma that just contains the policy */
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001151 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001153 pvma.vm_ops = NULL;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001154 pvma.vm_policy = spol;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001155 page = swapin_readahead(entry, gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 return page;
1157}
1158
Hugh Dickins02098fe2008-02-04 22:28:42 -08001159static struct page *shmem_alloc_page(gfp_t gfp,
1160 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161{
1162 struct vm_area_struct pvma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001164 /* Create a pseudo vma that just contains the policy */
1165 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001167 pvma.vm_ops = NULL;
1168 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001169
1170 /*
1171 * alloc_page_vma() will drop the shared policy reference
1172 */
1173 return alloc_page_vma(gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001175#else /* !CONFIG_NUMA */
1176#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001177static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001178{
1179}
1180#endif /* CONFIG_TMPFS */
1181
Hugh Dickins02098fe2008-02-04 22:28:42 -08001182static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1183 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184{
Hugh Dickins02098fe2008-02-04 22:28:42 -08001185 return swapin_readahead(entry, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186}
1187
Hugh Dickins02098fe2008-02-04 22:28:42 -08001188static inline struct page *shmem_alloc_page(gfp_t gfp,
1189 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190{
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001191 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001193#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001195#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1196static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1197{
1198 return NULL;
1199}
1200#endif
1201
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202/*
1203 * shmem_getpage - either get the page from swap or allocate a new one
1204 *
1205 * If we allocate a new one we do not mark it dirty. That's up to the
1206 * vm. If we swap it in we mark it dirty since we also free the swap
1207 * entry since a page cannot live in both the swap and page cache
1208 */
1209static int shmem_getpage(struct inode *inode, unsigned long idx,
1210 struct page **pagep, enum sgp_type sgp, int *type)
1211{
1212 struct address_space *mapping = inode->i_mapping;
1213 struct shmem_inode_info *info = SHMEM_I(inode);
1214 struct shmem_sb_info *sbinfo;
1215 struct page *filepage = *pagep;
1216 struct page *swappage;
1217 swp_entry_t *entry;
1218 swp_entry_t swap;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001219 gfp_t gfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 int error;
1221
1222 if (idx >= SHMEM_MAX_INDEX)
1223 return -EFBIG;
Nick Piggin54cb8822007-07-19 01:46:59 -07001224
1225 if (type)
Nick Piggin83c54072007-07-19 01:47:05 -07001226 *type = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07001227
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 /*
1229 * Normally, filepage is NULL on entry, and either found
1230 * uptodate immediately, or allocated and zeroed, or read
1231 * in under swappage, which is then assigned to filepage.
Hugh Dickins5402b972008-02-04 22:28:44 -08001232 * But shmem_readpage (required for splice) passes in a locked
Hugh Dickinsae9764162007-06-04 10:00:39 +02001233 * filepage, which may be found not uptodate by other callers
1234 * too, and may need to be copied from the swappage read in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 */
1236repeat:
1237 if (!filepage)
1238 filepage = find_lock_page(mapping, idx);
1239 if (filepage && PageUptodate(filepage))
1240 goto done;
1241 error = 0;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001242 gfp = mapping_gfp_mask(mapping);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -08001243 if (!filepage) {
1244 /*
1245 * Try to preload while we can wait, to not make a habit of
1246 * draining atomic reserves; but don't latch on to this cpu.
1247 */
1248 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1249 if (error)
1250 goto failed;
1251 radix_tree_preload_end();
1252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
1254 spin_lock(&info->lock);
1255 shmem_recalc_inode(inode);
1256 entry = shmem_swp_alloc(info, idx, sgp);
1257 if (IS_ERR(entry)) {
1258 spin_unlock(&info->lock);
1259 error = PTR_ERR(entry);
1260 goto failed;
1261 }
1262 swap = *entry;
1263
1264 if (swap.val) {
1265 /* Look it up and read it in.. */
1266 swappage = lookup_swap_cache(swap);
1267 if (!swappage) {
1268 shmem_swp_unmap(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 /* here we actually do the io */
Nick Piggin83c54072007-07-19 01:47:05 -07001270 if (type && !(*type & VM_FAULT_MAJOR)) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001271 __count_vm_event(PGMAJFAULT);
Nick Piggin83c54072007-07-19 01:47:05 -07001272 *type |= VM_FAULT_MAJOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 }
Christoph Lameterf8891e52006-06-30 01:55:45 -07001274 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001275 swappage = shmem_swapin(swap, gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 if (!swappage) {
1277 spin_lock(&info->lock);
1278 entry = shmem_swp_alloc(info, idx, sgp);
1279 if (IS_ERR(entry))
1280 error = PTR_ERR(entry);
1281 else {
1282 if (entry->val == swap.val)
1283 error = -ENOMEM;
1284 shmem_swp_unmap(entry);
1285 }
1286 spin_unlock(&info->lock);
1287 if (error)
1288 goto failed;
1289 goto repeat;
1290 }
1291 wait_on_page_locked(swappage);
1292 page_cache_release(swappage);
1293 goto repeat;
1294 }
1295
1296 /* We have to do this with page locked to prevent races */
Nick Piggin529ae9a2008-08-02 12:01:03 +02001297 if (!trylock_page(swappage)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 shmem_swp_unmap(entry);
1299 spin_unlock(&info->lock);
1300 wait_on_page_locked(swappage);
1301 page_cache_release(swappage);
1302 goto repeat;
1303 }
1304 if (PageWriteback(swappage)) {
1305 shmem_swp_unmap(entry);
1306 spin_unlock(&info->lock);
1307 wait_on_page_writeback(swappage);
1308 unlock_page(swappage);
1309 page_cache_release(swappage);
1310 goto repeat;
1311 }
1312 if (!PageUptodate(swappage)) {
1313 shmem_swp_unmap(entry);
1314 spin_unlock(&info->lock);
1315 unlock_page(swappage);
1316 page_cache_release(swappage);
1317 error = -EIO;
1318 goto failed;
1319 }
1320
1321 if (filepage) {
1322 shmem_swp_set(info, entry, 0);
1323 shmem_swp_unmap(entry);
1324 delete_from_swap_cache(swappage);
1325 spin_unlock(&info->lock);
1326 copy_highpage(filepage, swappage);
1327 unlock_page(swappage);
1328 page_cache_release(swappage);
1329 flush_dcache_page(filepage);
1330 SetPageUptodate(filepage);
1331 set_page_dirty(filepage);
1332 swap_free(swap);
Nick Piggine2867812008-07-25 19:45:30 -07001333 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1334 idx, GFP_NOWAIT))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 info->flags |= SHMEM_PAGEIN;
1336 shmem_swp_set(info, entry, 0);
1337 shmem_swp_unmap(entry);
Hugh Dickins73b12622008-02-04 22:28:50 -08001338 delete_from_swap_cache(swappage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 spin_unlock(&info->lock);
1340 filepage = swappage;
Hugh Dickins73b12622008-02-04 22:28:50 -08001341 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 swap_free(swap);
1343 } else {
1344 shmem_swp_unmap(entry);
1345 spin_unlock(&info->lock);
Hugh Dickins82369552008-02-07 00:14:22 -08001346 if (error == -ENOMEM) {
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001347 /*
1348 * reclaim from proper memory cgroup and
1349 * call memcg's OOM if needed.
1350 */
1351 error = mem_cgroup_shmem_charge_fallback(
1352 swappage,
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001353 current->mm,
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001354 gfp);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001355 if (error) {
1356 unlock_page(swappage);
1357 page_cache_release(swappage);
Hugh Dickins82369552008-02-07 00:14:22 -08001358 goto failed;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001359 }
Hugh Dickins82369552008-02-07 00:14:22 -08001360 }
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001361 unlock_page(swappage);
1362 page_cache_release(swappage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 goto repeat;
1364 }
1365 } else if (sgp == SGP_READ && !filepage) {
1366 shmem_swp_unmap(entry);
1367 filepage = find_get_page(mapping, idx);
1368 if (filepage &&
Nick Piggin529ae9a2008-08-02 12:01:03 +02001369 (!PageUptodate(filepage) || !trylock_page(filepage))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 spin_unlock(&info->lock);
1371 wait_on_page_locked(filepage);
1372 page_cache_release(filepage);
1373 filepage = NULL;
1374 goto repeat;
1375 }
1376 spin_unlock(&info->lock);
1377 } else {
1378 shmem_swp_unmap(entry);
1379 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001380 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 spin_lock(&sbinfo->stat_lock);
1382 if (sbinfo->free_blocks == 0 ||
1383 shmem_acct_block(info->flags)) {
1384 spin_unlock(&sbinfo->stat_lock);
1385 spin_unlock(&info->lock);
1386 error = -ENOSPC;
1387 goto failed;
1388 }
1389 sbinfo->free_blocks--;
1390 inode->i_blocks += BLOCKS_PER_PAGE;
1391 spin_unlock(&sbinfo->stat_lock);
1392 } else if (shmem_acct_block(info->flags)) {
1393 spin_unlock(&info->lock);
1394 error = -ENOSPC;
1395 goto failed;
1396 }
1397
1398 if (!filepage) {
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001399 int ret;
1400
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001402 filepage = shmem_alloc_page(gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 if (!filepage) {
1404 shmem_unacct_blocks(info->flags, 1);
1405 shmem_free_blocks(inode, 1);
1406 error = -ENOMEM;
1407 goto failed;
1408 }
Rik van Rielb2e18532008-10-18 20:26:30 -07001409 SetPageSwapBacked(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
Hugh Dickins82369552008-02-07 00:14:22 -08001411 /* Precharge page while we can wait, compensate after */
1412 error = mem_cgroup_cache_charge(filepage, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08001413 GFP_KERNEL);
Hugh Dickins82369552008-02-07 00:14:22 -08001414 if (error) {
1415 page_cache_release(filepage);
1416 shmem_unacct_blocks(info->flags, 1);
1417 shmem_free_blocks(inode, 1);
1418 filepage = NULL;
1419 goto failed;
1420 }
1421
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 spin_lock(&info->lock);
1423 entry = shmem_swp_alloc(info, idx, sgp);
1424 if (IS_ERR(entry))
1425 error = PTR_ERR(entry);
1426 else {
1427 swap = *entry;
1428 shmem_swp_unmap(entry);
1429 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001430 ret = error || swap.val;
1431 if (ret)
1432 mem_cgroup_uncharge_cache_page(filepage);
1433 else
1434 ret = add_to_page_cache_lru(filepage, mapping,
1435 idx, GFP_NOWAIT);
1436 /*
1437 * At add_to_page_cache_lru() failure, uncharge will
1438 * be done automatically.
1439 */
1440 if (ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 spin_unlock(&info->lock);
1442 page_cache_release(filepage);
1443 shmem_unacct_blocks(info->flags, 1);
1444 shmem_free_blocks(inode, 1);
1445 filepage = NULL;
1446 if (error)
1447 goto failed;
1448 goto repeat;
1449 }
1450 info->flags |= SHMEM_PAGEIN;
1451 }
1452
1453 info->alloced++;
1454 spin_unlock(&info->lock);
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001455 clear_highpage(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 flush_dcache_page(filepage);
1457 SetPageUptodate(filepage);
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001458 if (sgp == SGP_DIRTY)
1459 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 }
1461done:
Hugh Dickinsd3602442008-02-04 22:28:44 -08001462 *pagep = filepage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 return 0;
1464
1465failed:
1466 if (*pagep != filepage) {
1467 unlock_page(filepage);
1468 page_cache_release(filepage);
1469 }
1470 return error;
1471}
1472
Nick Piggind0217ac2007-07-19 01:47:03 -07001473static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001475 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 int error;
Nick Piggind0217ac2007-07-19 01:47:03 -07001477 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
Nick Piggind0217ac2007-07-19 01:47:03 -07001479 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1480 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001481
Hugh Dickins27d54b32008-02-04 22:28:43 -08001482 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
Nick Piggind0217ac2007-07-19 01:47:03 -07001483 if (error)
1484 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
Nick Piggin83c54072007-07-19 01:47:05 -07001486 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487}
1488
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489#ifdef CONFIG_NUMA
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001490static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001492 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1494}
1495
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001496static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1497 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001499 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 unsigned long idx;
1501
1502 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1503 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1504}
1505#endif
1506
1507int shmem_lock(struct file *file, int lock, struct user_struct *user)
1508{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001509 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 struct shmem_inode_info *info = SHMEM_I(inode);
1511 int retval = -ENOMEM;
1512
1513 spin_lock(&info->lock);
1514 if (lock && !(info->flags & VM_LOCKED)) {
1515 if (!user_shm_lock(inode->i_size, user))
1516 goto out_nomem;
1517 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001518 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 }
1520 if (!lock && (info->flags & VM_LOCKED) && user) {
1521 user_shm_unlock(inode->i_size, user);
1522 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001523 mapping_clear_unevictable(file->f_mapping);
1524 scan_mapping_unevictable_pages(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 }
1526 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001527
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528out_nomem:
1529 spin_unlock(&info->lock);
1530 return retval;
1531}
1532
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001533static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534{
1535 file_accessed(file);
1536 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001537 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 return 0;
1539}
1540
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001541static struct inode *shmem_get_inode(struct super_block *sb, int mode,
1542 dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543{
1544 struct inode *inode;
1545 struct shmem_inode_info *info;
1546 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1547
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001548 if (shmem_reserve_inode(sb))
1549 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
1551 inode = new_inode(sb);
1552 if (inode) {
1553 inode->i_mode = mode;
David Howells76aac0e2008-11-14 10:39:12 +11001554 inode->i_uid = current_fsuid();
1555 inode->i_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1558 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001559 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 info = SHMEM_I(inode);
1561 memset(info, 0, (char *)inode - (char *)info);
1562 spin_lock_init(&info->lock);
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001563 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 INIT_LIST_HEAD(&info->swaplist);
Al Viro72c04902009-06-24 16:58:48 -04001565 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
1567 switch (mode & S_IFMT) {
1568 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001569 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 init_special_inode(inode, mode, dev);
1571 break;
1572 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001573 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 inode->i_op = &shmem_inode_operations;
1575 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001576 mpol_shared_policy_init(&info->policy,
1577 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 break;
1579 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001580 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 /* Some things misbehave if size == 0 on a directory */
1582 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1583 inode->i_op = &shmem_dir_inode_operations;
1584 inode->i_fop = &simple_dir_operations;
1585 break;
1586 case S_IFLNK:
1587 /*
1588 * Must not load anything in the rbtree,
1589 * mpol_free_shared_policy will not be called.
1590 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001591 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 break;
1593 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001594 } else
1595 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 return inode;
1597}
1598
1599#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001600static const struct inode_operations shmem_symlink_inode_operations;
1601static const struct inode_operations shmem_symlink_inline_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602
1603/*
Nick Piggin800d15a2007-10-16 01:25:03 -07001604 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
Hugh Dickinsae9764162007-06-04 10:00:39 +02001605 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1606 * below the loop driver, in the generic fashion that many filesystems support.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 */
Hugh Dickinsae9764162007-06-04 10:00:39 +02001608static int shmem_readpage(struct file *file, struct page *page)
1609{
1610 struct inode *inode = page->mapping->host;
1611 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1612 unlock_page(page);
1613 return error;
1614}
1615
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001617shmem_write_begin(struct file *file, struct address_space *mapping,
1618 loff_t pos, unsigned len, unsigned flags,
1619 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620{
Nick Piggin800d15a2007-10-16 01:25:03 -07001621 struct inode *inode = mapping->host;
1622 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1623 *pagep = NULL;
1624 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1625}
1626
1627static int
1628shmem_write_end(struct file *file, struct address_space *mapping,
1629 loff_t pos, unsigned len, unsigned copied,
1630 struct page *page, void *fsdata)
1631{
1632 struct inode *inode = mapping->host;
1633
Hugh Dickinsd3602442008-02-04 22:28:44 -08001634 if (pos + copied > inode->i_size)
1635 i_size_write(inode, pos + copied);
1636
Nick Piggin800d15a2007-10-16 01:25:03 -07001637 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001638 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001639 page_cache_release(page);
1640
Nick Piggin800d15a2007-10-16 01:25:03 -07001641 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642}
1643
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1645{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001646 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 struct address_space *mapping = inode->i_mapping;
1648 unsigned long index, offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001649 enum sgp_type sgp = SGP_READ;
1650
1651 /*
1652 * Might this read be for a stacking filesystem? Then when reading
1653 * holes of a sparse file, we actually need to allocate those pages,
1654 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1655 */
1656 if (segment_eq(get_fs(), KERNEL_DS))
1657 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
1659 index = *ppos >> PAGE_CACHE_SHIFT;
1660 offset = *ppos & ~PAGE_CACHE_MASK;
1661
1662 for (;;) {
1663 struct page *page = NULL;
1664 unsigned long end_index, nr, ret;
1665 loff_t i_size = i_size_read(inode);
1666
1667 end_index = i_size >> PAGE_CACHE_SHIFT;
1668 if (index > end_index)
1669 break;
1670 if (index == end_index) {
1671 nr = i_size & ~PAGE_CACHE_MASK;
1672 if (nr <= offset)
1673 break;
1674 }
1675
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001676 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 if (desc->error) {
1678 if (desc->error == -EINVAL)
1679 desc->error = 0;
1680 break;
1681 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001682 if (page)
1683 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684
1685 /*
1686 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001687 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 */
1689 nr = PAGE_CACHE_SIZE;
1690 i_size = i_size_read(inode);
1691 end_index = i_size >> PAGE_CACHE_SHIFT;
1692 if (index == end_index) {
1693 nr = i_size & ~PAGE_CACHE_MASK;
1694 if (nr <= offset) {
1695 if (page)
1696 page_cache_release(page);
1697 break;
1698 }
1699 }
1700 nr -= offset;
1701
1702 if (page) {
1703 /*
1704 * If users can be writing to this page using arbitrary
1705 * virtual addresses, take care about potential aliasing
1706 * before reading the page on the kernel side.
1707 */
1708 if (mapping_writably_mapped(mapping))
1709 flush_dcache_page(page);
1710 /*
1711 * Mark the page accessed if we read the beginning.
1712 */
1713 if (!offset)
1714 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001715 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001717 page_cache_get(page);
1718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
1720 /*
1721 * Ok, we have the page, and it's up-to-date, so
1722 * now we can copy it to user space...
1723 *
1724 * The actor routine returns how many bytes were actually used..
1725 * NOTE! This may not be the same as how much of a user buffer
1726 * we filled up (we may be padding etc), so we can only update
1727 * "pos" here (the actor routine has to update the user buffer
1728 * pointers and the remaining count).
1729 */
1730 ret = actor(desc, page, offset, nr);
1731 offset += ret;
1732 index += offset >> PAGE_CACHE_SHIFT;
1733 offset &= ~PAGE_CACHE_MASK;
1734
1735 page_cache_release(page);
1736 if (ret != nr || !desc->count)
1737 break;
1738
1739 cond_resched();
1740 }
1741
1742 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1743 file_accessed(filp);
1744}
1745
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001746static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1747 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748{
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001749 struct file *filp = iocb->ki_filp;
1750 ssize_t retval;
1751 unsigned long seg;
1752 size_t count;
1753 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001755 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1756 if (retval)
1757 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001759 for (seg = 0; seg < nr_segs; seg++) {
1760 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001762 desc.written = 0;
1763 desc.arg.buf = iov[seg].iov_base;
1764 desc.count = iov[seg].iov_len;
1765 if (desc.count == 0)
1766 continue;
1767 desc.error = 0;
1768 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1769 retval += desc.written;
1770 if (desc.error) {
1771 retval = retval ?: desc.error;
1772 break;
1773 }
1774 if (desc.count > 0)
1775 break;
1776 }
1777 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778}
1779
David Howells726c3342006-06-23 02:02:58 -07001780static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781{
David Howells726c3342006-06-23 02:02:58 -07001782 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
1784 buf->f_type = TMPFS_MAGIC;
1785 buf->f_bsize = PAGE_CACHE_SIZE;
1786 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001787 spin_lock(&sbinfo->stat_lock);
1788 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 buf->f_blocks = sbinfo->max_blocks;
1790 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001791 }
1792 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 buf->f_files = sbinfo->max_inodes;
1794 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 }
1796 /* else leave those fields 0 like simple_statfs */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001797 spin_unlock(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 return 0;
1799}
1800
1801/*
1802 * File creation. Allocate an inode, and we're done..
1803 */
1804static int
1805shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1806{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001807 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 int error = -ENOSPC;
1809
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001810 inode = shmem_get_inode(dir->i_sb, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 if (inode) {
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001812 error = security_inode_init_security(inode, dir, NULL, NULL,
1813 NULL);
1814 if (error) {
1815 if (error != -EOPNOTSUPP) {
1816 iput(inode);
1817 return error;
1818 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001819 }
1820 error = shmem_acl_init(inode, dir);
1821 if (error) {
1822 iput(inode);
1823 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001824 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 if (dir->i_mode & S_ISGID) {
1826 inode->i_gid = dir->i_gid;
1827 if (S_ISDIR(mode))
1828 inode->i_mode |= S_ISGID;
1829 }
1830 dir->i_size += BOGO_DIRENT_SIZE;
1831 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1832 d_instantiate(dentry, inode);
1833 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 }
1835 return error;
1836}
1837
1838static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1839{
1840 int error;
1841
1842 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1843 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001844 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 return 0;
1846}
1847
1848static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1849 struct nameidata *nd)
1850{
1851 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1852}
1853
1854/*
1855 * Link a file..
1856 */
1857static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1858{
1859 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001860 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
1862 /*
1863 * No ordinary (disk based) filesystem counts links as inodes;
1864 * but each new link needs a new dentry, pinning lowmem, and
1865 * tmpfs dentries cannot be pruned until they are unlinked.
1866 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001867 ret = shmem_reserve_inode(inode->i_sb);
1868 if (ret)
1869 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
1871 dir->i_size += BOGO_DIRENT_SIZE;
1872 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001873 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 atomic_inc(&inode->i_count); /* New dentry reference */
1875 dget(dentry); /* Extra pinning count for the created dentry */
1876 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001877out:
1878 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879}
1880
1881static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1882{
1883 struct inode *inode = dentry->d_inode;
1884
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001885 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1886 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
1888 dir->i_size -= BOGO_DIRENT_SIZE;
1889 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001890 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 dput(dentry); /* Undo the count from "create" - this does all the work */
1892 return 0;
1893}
1894
1895static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1896{
1897 if (!simple_empty(dentry))
1898 return -ENOTEMPTY;
1899
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001900 drop_nlink(dentry->d_inode);
1901 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 return shmem_unlink(dir, dentry);
1903}
1904
1905/*
1906 * The VFS layer already does all the dentry stuff for rename,
1907 * we just have to decrement the usage count for the target if
1908 * it exists so that the VFS layer correctly free's it when it
1909 * gets overwritten.
1910 */
1911static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1912{
1913 struct inode *inode = old_dentry->d_inode;
1914 int they_are_dirs = S_ISDIR(inode->i_mode);
1915
1916 if (!simple_empty(new_dentry))
1917 return -ENOTEMPTY;
1918
1919 if (new_dentry->d_inode) {
1920 (void) shmem_unlink(new_dir, new_dentry);
1921 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001922 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001924 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001925 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 }
1927
1928 old_dir->i_size -= BOGO_DIRENT_SIZE;
1929 new_dir->i_size += BOGO_DIRENT_SIZE;
1930 old_dir->i_ctime = old_dir->i_mtime =
1931 new_dir->i_ctime = new_dir->i_mtime =
1932 inode->i_ctime = CURRENT_TIME;
1933 return 0;
1934}
1935
1936static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1937{
1938 int error;
1939 int len;
1940 struct inode *inode;
1941 struct page *page = NULL;
1942 char *kaddr;
1943 struct shmem_inode_info *info;
1944
1945 len = strlen(symname) + 1;
1946 if (len > PAGE_CACHE_SIZE)
1947 return -ENAMETOOLONG;
1948
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001949 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 if (!inode)
1951 return -ENOSPC;
1952
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001953 error = security_inode_init_security(inode, dir, NULL, NULL,
1954 NULL);
1955 if (error) {
1956 if (error != -EOPNOTSUPP) {
1957 iput(inode);
1958 return error;
1959 }
1960 error = 0;
1961 }
1962
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 info = SHMEM_I(inode);
1964 inode->i_size = len-1;
1965 if (len <= (char *)inode - (char *)info) {
1966 /* do it inline */
1967 memcpy(info, symname, len);
1968 inode->i_op = &shmem_symlink_inline_operations;
1969 } else {
1970 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1971 if (error) {
1972 iput(inode);
1973 return error;
1974 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07001975 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 inode->i_op = &shmem_symlink_inode_operations;
1977 kaddr = kmap_atomic(page, KM_USER0);
1978 memcpy(kaddr, symname, len);
1979 kunmap_atomic(kaddr, KM_USER0);
1980 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001981 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 page_cache_release(page);
1983 }
1984 if (dir->i_mode & S_ISGID)
1985 inode->i_gid = dir->i_gid;
1986 dir->i_size += BOGO_DIRENT_SIZE;
1987 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1988 d_instantiate(dentry, inode);
1989 dget(dentry);
1990 return 0;
1991}
1992
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001993static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994{
1995 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001996 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997}
1998
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001999static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000{
2001 struct page *page = NULL;
2002 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2003 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08002004 if (page)
2005 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002006 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007}
2008
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002009static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010{
2011 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002012 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 kunmap(page);
2014 mark_page_accessed(page);
2015 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 }
2017}
2018
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002019static const struct inode_operations shmem_symlink_inline_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 .readlink = generic_readlink,
2021 .follow_link = shmem_follow_link_inline,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022};
2023
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002024static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 .truncate = shmem_truncate,
2026 .readlink = generic_readlink,
2027 .follow_link = shmem_follow_link,
2028 .put_link = shmem_put_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029};
2030
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002031#ifdef CONFIG_TMPFS_POSIX_ACL
Randy Dunlap46711812008-03-19 17:00:41 -07002032/*
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002033 * Superblocks without xattr inode operations will get security.* xattr
2034 * support from the VFS "for free". As soon as we have any other xattrs
2035 * like ACLs, we also need to implement the security.* handlers at
2036 * filesystem level, though.
2037 */
2038
2039static size_t shmem_xattr_security_list(struct inode *inode, char *list,
2040 size_t list_len, const char *name,
2041 size_t name_len)
2042{
2043 return security_inode_listsecurity(inode, list, list_len);
2044}
2045
2046static int shmem_xattr_security_get(struct inode *inode, const char *name,
2047 void *buffer, size_t size)
2048{
2049 if (strcmp(name, "") == 0)
2050 return -EINVAL;
David P. Quigley42492592008-02-04 22:29:39 -08002051 return xattr_getsecurity(inode, name, buffer, size);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002052}
2053
2054static int shmem_xattr_security_set(struct inode *inode, const char *name,
2055 const void *value, size_t size, int flags)
2056{
2057 if (strcmp(name, "") == 0)
2058 return -EINVAL;
2059 return security_inode_setsecurity(inode, name, value, size, flags);
2060}
2061
Adrian Bunk1f370a22006-12-06 20:38:21 -08002062static struct xattr_handler shmem_xattr_security_handler = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002063 .prefix = XATTR_SECURITY_PREFIX,
2064 .list = shmem_xattr_security_list,
2065 .get = shmem_xattr_security_get,
2066 .set = shmem_xattr_security_set,
2067};
2068
2069static struct xattr_handler *shmem_xattr_handlers[] = {
2070 &shmem_xattr_acl_access_handler,
2071 &shmem_xattr_acl_default_handler,
2072 &shmem_xattr_security_handler,
2073 NULL
2074};
2075#endif
2076
David M. Grimes91828a42006-10-17 00:09:45 -07002077static struct dentry *shmem_get_parent(struct dentry *child)
2078{
2079 return ERR_PTR(-ESTALE);
2080}
2081
2082static int shmem_match(struct inode *ino, void *vfh)
2083{
2084 __u32 *fh = vfh;
2085 __u64 inum = fh[2];
2086 inum = (inum << 32) | fh[1];
2087 return ino->i_ino == inum && fh[0] == ino->i_generation;
2088}
2089
Christoph Hellwig480b1162007-10-21 16:42:13 -07002090static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2091 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002092{
David M. Grimes91828a42006-10-17 00:09:45 -07002093 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002094 struct dentry *dentry = NULL;
2095 u64 inum = fid->raw[2];
2096 inum = (inum << 32) | fid->raw[1];
David M. Grimes91828a42006-10-17 00:09:45 -07002097
Christoph Hellwig480b1162007-10-21 16:42:13 -07002098 if (fh_len < 3)
2099 return NULL;
2100
2101 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2102 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002103 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002104 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002105 iput(inode);
2106 }
2107
Christoph Hellwig480b1162007-10-21 16:42:13 -07002108 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002109}
2110
2111static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2112 int connectable)
2113{
2114 struct inode *inode = dentry->d_inode;
2115
2116 if (*len < 3)
2117 return 255;
2118
2119 if (hlist_unhashed(&inode->i_hash)) {
2120 /* Unfortunately insert_inode_hash is not idempotent,
2121 * so as we hash inodes here rather than at creation
2122 * time, we need a lock to ensure we only try
2123 * to do it once
2124 */
2125 static DEFINE_SPINLOCK(lock);
2126 spin_lock(&lock);
2127 if (hlist_unhashed(&inode->i_hash))
2128 __insert_inode_hash(inode,
2129 inode->i_ino + inode->i_generation);
2130 spin_unlock(&lock);
2131 }
2132
2133 fh[0] = inode->i_generation;
2134 fh[1] = inode->i_ino;
2135 fh[2] = ((__u64)inode->i_ino) >> 32;
2136
2137 *len = 3;
2138 return 1;
2139}
2140
Christoph Hellwig39655162007-10-21 16:42:17 -07002141static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002142 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002143 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002144 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002145};
2146
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002147static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2148 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149{
2150 char *this_char, *value, *rest;
2151
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002152 while (options != NULL) {
2153 this_char = options;
2154 for (;;) {
2155 /*
2156 * NUL-terminate this option: unfortunately,
2157 * mount options form a comma-separated list,
2158 * but mpol's nodelist may also contain commas.
2159 */
2160 options = strchr(options, ',');
2161 if (options == NULL)
2162 break;
2163 options++;
2164 if (!isdigit(*options)) {
2165 options[-1] = '\0';
2166 break;
2167 }
2168 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 if (!*this_char)
2170 continue;
2171 if ((value = strchr(this_char,'=')) != NULL) {
2172 *value++ = 0;
2173 } else {
2174 printk(KERN_ERR
2175 "tmpfs: No value for mount option '%s'\n",
2176 this_char);
2177 return 1;
2178 }
2179
2180 if (!strcmp(this_char,"size")) {
2181 unsigned long long size;
2182 size = memparse(value,&rest);
2183 if (*rest == '%') {
2184 size <<= PAGE_SHIFT;
2185 size *= totalram_pages;
2186 do_div(size, 100);
2187 rest++;
2188 }
2189 if (*rest)
2190 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002191 sbinfo->max_blocks =
2192 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002194 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 if (*rest)
2196 goto bad_val;
2197 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002198 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 if (*rest)
2200 goto bad_val;
2201 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002202 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002204 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 if (*rest)
2206 goto bad_val;
2207 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002208 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002210 sbinfo->uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 if (*rest)
2212 goto bad_val;
2213 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002214 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002216 sbinfo->gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 if (*rest)
2218 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002219 } else if (!strcmp(this_char,"mpol")) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002220 if (mpol_parse_str(value, &sbinfo->mpol, 1))
Robin Holt7339ff82006-01-14 13:20:48 -08002221 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 } else {
2223 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2224 this_char);
2225 return 1;
2226 }
2227 }
2228 return 0;
2229
2230bad_val:
2231 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2232 value, this_char);
2233 return 1;
2234
2235}
2236
2237static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2238{
2239 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002240 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002241 unsigned long blocks;
2242 unsigned long inodes;
2243 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002245 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002246 return error;
2247
2248 spin_lock(&sbinfo->stat_lock);
2249 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2250 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002251 if (config.max_blocks < blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002252 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002253 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002254 goto out;
2255 /*
2256 * Those tests also disallow limited->unlimited while any are in
2257 * use, so i_blocks will always be zero when max_blocks is zero;
2258 * but we must separately disallow unlimited->limited, because
2259 * in that case we have no record of how much is already in use.
2260 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002261 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002262 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002263 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002264 goto out;
2265
2266 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002267 sbinfo->max_blocks = config.max_blocks;
2268 sbinfo->free_blocks = config.max_blocks - blocks;
2269 sbinfo->max_inodes = config.max_inodes;
2270 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002271
2272 mpol_put(sbinfo->mpol);
2273 sbinfo->mpol = config.mpol; /* transfers initial ref */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002274out:
2275 spin_unlock(&sbinfo->stat_lock);
2276 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002278
2279static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2280{
2281 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2282
2283 if (sbinfo->max_blocks != shmem_default_max_blocks())
2284 seq_printf(seq, ",size=%luk",
2285 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2286 if (sbinfo->max_inodes != shmem_default_max_inodes())
2287 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2288 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2289 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2290 if (sbinfo->uid != 0)
2291 seq_printf(seq, ",uid=%u", sbinfo->uid);
2292 if (sbinfo->gid != 0)
2293 seq_printf(seq, ",gid=%u", sbinfo->gid);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002294 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002295 return 0;
2296}
2297#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
2299static void shmem_put_super(struct super_block *sb)
2300{
2301 kfree(sb->s_fs_info);
2302 sb->s_fs_info = NULL;
2303}
2304
Kay Sievers2b2af542009-04-30 15:23:42 +02002305int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306{
2307 struct inode *inode;
2308 struct dentry *root;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002309 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002310 int err = -ENOMEM;
2311
2312 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07002313 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002314 L1_CACHE_BYTES), GFP_KERNEL);
2315 if (!sbinfo)
2316 return -ENOMEM;
2317
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002318 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002319 sbinfo->uid = current_fsuid();
2320 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002321 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002323#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 /*
2325 * Per default we only allow half of the physical ram per
2326 * tmpfs instance, limiting inodes to one per page of lowmem;
2327 * but the internal instance is left unlimited.
2328 */
2329 if (!(sb->s_flags & MS_NOUSER)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002330 sbinfo->max_blocks = shmem_default_max_blocks();
2331 sbinfo->max_inodes = shmem_default_max_inodes();
2332 if (shmem_parse_options(data, sbinfo, false)) {
2333 err = -EINVAL;
2334 goto failed;
2335 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 }
David M. Grimes91828a42006-10-17 00:09:45 -07002337 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338#else
2339 sb->s_flags |= MS_NOUSER;
2340#endif
2341
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002342 spin_lock_init(&sbinfo->stat_lock);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002343 sbinfo->free_blocks = sbinfo->max_blocks;
2344 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002345
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 sb->s_maxbytes = SHMEM_MAX_BYTES;
2347 sb->s_blocksize = PAGE_CACHE_SIZE;
2348 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2349 sb->s_magic = TMPFS_MAGIC;
2350 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002351 sb->s_time_gran = 1;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002352#ifdef CONFIG_TMPFS_POSIX_ACL
2353 sb->s_xattr = shmem_xattr_handlers;
2354 sb->s_flags |= MS_POSIXACL;
2355#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002356
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002357 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 if (!inode)
2359 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002360 inode->i_uid = sbinfo->uid;
2361 inode->i_gid = sbinfo->gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 root = d_alloc_root(inode);
2363 if (!root)
2364 goto failed_iput;
2365 sb->s_root = root;
2366 return 0;
2367
2368failed_iput:
2369 iput(inode);
2370failed:
2371 shmem_put_super(sb);
2372 return err;
2373}
2374
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002375static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376
2377static struct inode *shmem_alloc_inode(struct super_block *sb)
2378{
2379 struct shmem_inode_info *p;
Christoph Lametere94b1762006-12-06 20:33:17 -08002380 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 if (!p)
2382 return NULL;
2383 return &p->vfs_inode;
2384}
2385
2386static void shmem_destroy_inode(struct inode *inode)
2387{
2388 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2389 /* only struct inode is valid if it's an inline symlink */
2390 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2391 }
2392 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2393}
2394
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002395static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396{
2397 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2398
Christoph Lametera35afb82007-05-16 22:10:57 -07002399 inode_init_once(&p->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400}
2401
2402static int init_inodecache(void)
2403{
2404 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2405 sizeof(struct shmem_inode_info),
Alexey Dobriyan040b5c62007-10-16 23:26:10 -07002406 0, SLAB_PANIC, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 return 0;
2408}
2409
2410static void destroy_inodecache(void)
2411{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002412 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413}
2414
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002415static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002417 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418#ifdef CONFIG_TMPFS
Hugh Dickinsae9764162007-06-04 10:00:39 +02002419 .readpage = shmem_readpage,
Nick Piggin800d15a2007-10-16 01:25:03 -07002420 .write_begin = shmem_write_begin,
2421 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002423 .migratepage = migrate_page,
Andi Kleenaa261f52009-09-16 11:50:16 +02002424 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425};
2426
Helge Deller15ad7cd2006-12-06 20:40:36 -08002427static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 .mmap = shmem_mmap,
2429#ifdef CONFIG_TMPFS
2430 .llseek = generic_file_llseek,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002431 .read = do_sync_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002432 .write = do_sync_write,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002433 .aio_read = shmem_file_aio_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002434 .aio_write = generic_file_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 .fsync = simple_sync_file,
Hugh Dickinsae9764162007-06-04 10:00:39 +02002436 .splice_read = generic_file_splice_read,
2437 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438#endif
2439};
2440
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002441static const struct inode_operations shmem_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 .truncate = shmem_truncate,
2443 .setattr = shmem_notify_change,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002444 .truncate_range = shmem_truncate_range,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002445#ifdef CONFIG_TMPFS_POSIX_ACL
2446 .setxattr = generic_setxattr,
2447 .getxattr = generic_getxattr,
2448 .listxattr = generic_listxattr,
2449 .removexattr = generic_removexattr,
Linus Torvalds6d848a42009-08-28 12:04:28 -07002450 .check_acl = shmem_check_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002451#endif
2452
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453};
2454
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002455static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456#ifdef CONFIG_TMPFS
2457 .create = shmem_create,
2458 .lookup = simple_lookup,
2459 .link = shmem_link,
2460 .unlink = shmem_unlink,
2461 .symlink = shmem_symlink,
2462 .mkdir = shmem_mkdir,
2463 .rmdir = shmem_rmdir,
2464 .mknod = shmem_mknod,
2465 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002467#ifdef CONFIG_TMPFS_POSIX_ACL
2468 .setattr = shmem_notify_change,
2469 .setxattr = generic_setxattr,
2470 .getxattr = generic_getxattr,
2471 .listxattr = generic_listxattr,
2472 .removexattr = generic_removexattr,
Linus Torvalds6d848a42009-08-28 12:04:28 -07002473 .check_acl = shmem_check_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002474#endif
2475};
2476
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002477static const struct inode_operations shmem_special_inode_operations = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002478#ifdef CONFIG_TMPFS_POSIX_ACL
2479 .setattr = shmem_notify_change,
2480 .setxattr = generic_setxattr,
2481 .getxattr = generic_getxattr,
2482 .listxattr = generic_listxattr,
2483 .removexattr = generic_removexattr,
Linus Torvalds6d848a42009-08-28 12:04:28 -07002484 .check_acl = shmem_check_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002485#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486};
2487
Hugh Dickins759b9772007-03-05 00:30:28 -08002488static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 .alloc_inode = shmem_alloc_inode,
2490 .destroy_inode = shmem_destroy_inode,
2491#ifdef CONFIG_TMPFS
2492 .statfs = shmem_statfs,
2493 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002494 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495#endif
2496 .delete_inode = shmem_delete_inode,
2497 .drop_inode = generic_delete_inode,
2498 .put_super = shmem_put_super,
2499};
2500
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002501static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002502 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503#ifdef CONFIG_NUMA
2504 .set_policy = shmem_set_policy,
2505 .get_policy = shmem_get_policy,
2506#endif
2507};
2508
2509
David Howells454e2392006-06-23 02:02:57 -07002510static int shmem_get_sb(struct file_system_type *fs_type,
2511 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512{
David Howells454e2392006-06-23 02:02:57 -07002513 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514}
2515
2516static struct file_system_type tmpfs_fs_type = {
2517 .owner = THIS_MODULE,
2518 .name = "tmpfs",
2519 .get_sb = shmem_get_sb,
2520 .kill_sb = kill_litter_super,
2521};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522
Kay Sievers2b2af542009-04-30 15:23:42 +02002523int __init init_tmpfs(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524{
2525 int error;
2526
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002527 error = bdi_init(&shmem_backing_dev_info);
2528 if (error)
2529 goto out4;
2530
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 error = init_inodecache();
2532 if (error)
2533 goto out3;
2534
2535 error = register_filesystem(&tmpfs_fs_type);
2536 if (error) {
2537 printk(KERN_ERR "Could not register tmpfs\n");
2538 goto out2;
2539 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002540
Trond Myklebust1f5ce9e2006-06-09 09:34:16 -04002541 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 tmpfs_fs_type.name, NULL);
2543 if (IS_ERR(shm_mnt)) {
2544 error = PTR_ERR(shm_mnt);
2545 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2546 goto out1;
2547 }
2548 return 0;
2549
2550out1:
2551 unregister_filesystem(&tmpfs_fs_type);
2552out2:
2553 destroy_inodecache();
2554out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002555 bdi_destroy(&shmem_backing_dev_info);
2556out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 shm_mnt = ERR_PTR(error);
2558 return error;
2559}
Matt Mackall853ac432009-01-06 14:40:20 -08002560
2561#else /* !CONFIG_SHMEM */
2562
2563/*
2564 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2565 *
2566 * This is intended for small system where the benefits of the full
2567 * shmem code (swap-backed and resource-limited) are outweighed by
2568 * their complexity. On systems without swap this code should be
2569 * effectively equivalent, but much lighter weight.
2570 */
2571
2572#include <linux/ramfs.h>
2573
2574static struct file_system_type tmpfs_fs_type = {
2575 .name = "tmpfs",
2576 .get_sb = ramfs_get_sb,
2577 .kill_sb = kill_litter_super,
2578};
2579
Kay Sievers2b2af542009-04-30 15:23:42 +02002580int __init init_tmpfs(void)
Matt Mackall853ac432009-01-06 14:40:20 -08002581{
2582 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2583
2584 shm_mnt = kern_mount(&tmpfs_fs_type);
2585 BUG_ON(IS_ERR(shm_mnt));
2586
2587 return 0;
2588}
2589
2590int shmem_unuse(swp_entry_t entry, struct page *page)
2591{
2592 return 0;
2593}
2594
Hugh Dickins3f96b792009-09-21 17:03:37 -07002595int shmem_lock(struct file *file, int lock, struct user_struct *user)
2596{
2597 return 0;
2598}
2599
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002600#define shmem_vm_ops generic_file_vm_ops
2601#define shmem_file_operations ramfs_file_operations
2602#define shmem_get_inode(sb, mode, dev, flags) ramfs_get_inode(sb, mode, dev)
2603#define shmem_acct_size(flags, size) 0
2604#define shmem_unacct_size(flags, size) do {} while (0)
Hugh Dickinscaefba12009-04-13 14:40:12 -07002605#define SHMEM_MAX_BYTES MAX_LFS_FILESIZE
Matt Mackall853ac432009-01-06 14:40:20 -08002606
2607#endif /* CONFIG_SHMEM */
2608
2609/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610
Randy Dunlap46711812008-03-19 17:00:41 -07002611/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 * shmem_file_setup - get an unlinked file living in tmpfs
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 * @name: name for dentry (to be seen in /proc/<pid>/maps
2614 * @size: size to be set for the file
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002615 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 */
Sergei Trofimovich168f5ac2009-06-16 15:33:02 -07002617struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618{
2619 int error;
2620 struct file *file;
2621 struct inode *inode;
2622 struct dentry *dentry, *root;
2623 struct qstr this;
2624
2625 if (IS_ERR(shm_mnt))
2626 return (void *)shm_mnt;
2627
2628 if (size < 0 || size > SHMEM_MAX_BYTES)
2629 return ERR_PTR(-EINVAL);
2630
2631 if (shmem_acct_size(flags, size))
2632 return ERR_PTR(-ENOMEM);
2633
2634 error = -ENOMEM;
2635 this.name = name;
2636 this.len = strlen(name);
2637 this.hash = 0; /* will go */
2638 root = shm_mnt->mnt_root;
2639 dentry = d_alloc(root, &this);
2640 if (!dentry)
2641 goto put_memory;
2642
2643 error = -ENFILE;
2644 file = get_empty_filp();
2645 if (!file)
2646 goto put_dentry;
2647
2648 error = -ENOSPC;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002649 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 if (!inode)
2651 goto close_file;
2652
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 d_instantiate(dentry, inode);
2654 inode->i_size = size;
2655 inode->i_nlink = 0; /* It is unlinked */
Dave Hansence8d2cd2007-10-16 23:31:13 -07002656 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
Matt Mackall853ac432009-01-06 14:40:20 -08002657 &shmem_file_operations);
2658
2659#ifndef CONFIG_MMU
2660 error = ramfs_nommu_expand_for_mapping(inode, size);
2661 if (error)
2662 goto close_file;
2663#endif
Mimi Zoharc9d9ac52009-05-19 13:25:58 -04002664 ima_counts_get(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 return file;
2666
2667close_file:
2668 put_filp(file);
2669put_dentry:
2670 dput(dentry);
2671put_memory:
2672 shmem_unacct_size(flags, size);
2673 return ERR_PTR(error);
2674}
Keith Packard395e0dd2008-06-20 00:08:06 -07002675EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676
Randy Dunlap46711812008-03-19 17:00:41 -07002677/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2680 */
2681int shmem_zero_setup(struct vm_area_struct *vma)
2682{
2683 struct file *file;
2684 loff_t size = vma->vm_end - vma->vm_start;
2685
2686 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2687 if (IS_ERR(file))
2688 return PTR_ERR(file);
2689
2690 if (vma->vm_file)
2691 fput(vma->vm_file);
2692 vma->vm_file = file;
2693 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 return 0;
2695}