blob: f1b0d4871f3a72b8cfb9e9650b6e271daf910956 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Resizable virtual memory filesystem for Linux.
3 *
4 * Copyright (C) 2000 Linus Torvalds.
5 * 2000 Transmeta Corp.
6 * 2000-2001 Christoph Rohland
7 * 2000-2001 SAP AG
8 * 2002 Red Hat Inc.
Hugh Dickins0edd73b2005-06-21 17:15:04 -07009 * Copyright (C) 2002-2005 Hugh Dickins.
10 * Copyright (C) 2002-2005 VERITAS Software Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Copyright (C) 2004 Andi Kleen, SuSE Labs
12 *
13 * Extended attribute support for tmpfs:
14 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
15 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
16 *
17 * This file is released under the GPL.
18 */
19
20/*
21 * This virtual memory filesystem is heavily based on the ramfs. It
22 * extends ramfs by the ability to use swap and honor resource limits
23 * which makes it a completely usable filesystem.
24 */
25
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/module.h>
27#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/fs.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070029#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070030#include <linux/exportfs.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070031#include <linux/generic_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/mm.h>
33#include <linux/mman.h>
34#include <linux/file.h>
35#include <linux/swap.h>
36#include <linux/pagemap.h>
37#include <linux/string.h>
38#include <linux/slab.h>
39#include <linux/backing-dev.h>
40#include <linux/shmem_fs.h>
41#include <linux/mount.h>
42#include <linux/writeback.h>
43#include <linux/vfs.h>
44#include <linux/blkdev.h>
45#include <linux/security.h>
46#include <linux/swapops.h>
47#include <linux/mempolicy.h>
48#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000049#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070050#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070051#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080052#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040053#include <linux/magic.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <asm/uaccess.h>
56#include <asm/div64.h>
57#include <asm/pgtable.h>
58
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
60#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
61#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
62
63#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
64#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
65
66#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
67
68/* info->flags needs VM_flags to handle pagein/truncate races efficiently */
69#define SHMEM_PAGEIN VM_READ
70#define SHMEM_TRUNCATE VM_WRITE
71
72/* Definition to limit shmem_truncate's steps between cond_rescheds */
73#define LATENCY_LIMIT 64
74
75/* Pretend that each entry is of this size in directory's i_size */
76#define BOGO_DIRENT_SIZE 20
77
Linus Torvalds1da177e2005-04-16 15:20:36 -070078/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
79enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 SGP_READ, /* don't exceed i_size, don't allocate page */
81 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -080082 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 SGP_WRITE, /* may exceed i_size, may allocate page */
84};
85
Andrew Mortonb76db732008-02-08 04:21:49 -080086#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080087static unsigned long shmem_default_max_blocks(void)
88{
89 return totalram_pages / 2;
90}
91
92static unsigned long shmem_default_max_inodes(void)
93{
94 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
95}
Andrew Mortonb76db732008-02-08 04:21:49 -080096#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080097
Linus Torvalds1da177e2005-04-16 15:20:36 -070098static int shmem_getpage(struct inode *inode, unsigned long idx,
99 struct page **pagep, enum sgp_type sgp, int *type);
100
Al Viro6daa0e22005-10-21 03:18:50 -0400101static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102{
103 /*
104 * The above definition of ENTRIES_PER_PAGE, and the use of
105 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
106 * might be reconsidered if it ever diverges from PAGE_SIZE.
Mel Gorman769848c2007-07-17 04:03:05 -0700107 *
Mel Gormane12ba742007-10-16 01:25:52 -0700108 * Mobility flags are masked out as swap vectors cannot move
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 */
Mel Gormane12ba742007-10-16 01:25:52 -0700110 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
Mel Gorman769848c2007-07-17 04:03:05 -0700111 PAGE_CACHE_SHIFT-PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112}
113
114static inline void shmem_dir_free(struct page *page)
115{
116 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
117}
118
119static struct page **shmem_dir_map(struct page *page)
120{
121 return (struct page **)kmap_atomic(page, KM_USER0);
122}
123
124static inline void shmem_dir_unmap(struct page **dir)
125{
126 kunmap_atomic(dir, KM_USER0);
127}
128
129static swp_entry_t *shmem_swp_map(struct page *page)
130{
131 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
132}
133
134static inline void shmem_swp_balance_unmap(void)
135{
136 /*
137 * When passing a pointer to an i_direct entry, to code which
138 * also handles indirect entries and so will shmem_swp_unmap,
139 * we must arrange for the preempt count to remain in balance.
140 * What kmap_atomic of a lowmem page does depends on config
141 * and architecture, so pretend to kmap_atomic some lowmem page.
142 */
143 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
144}
145
146static inline void shmem_swp_unmap(swp_entry_t *entry)
147{
148 kunmap_atomic(entry, KM_USER1);
149}
150
151static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
152{
153 return sb->s_fs_info;
154}
155
156/*
157 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
158 * for shared memory and for shared anonymous (/dev/zero) mappings
159 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
160 * consistent with the pre-accounting of private mappings ...
161 */
162static inline int shmem_acct_size(unsigned long flags, loff_t size)
163{
Alan Cox731572d2008-10-29 14:01:20 -0700164 return (flags & VM_ACCOUNT) ?
165 security_vm_enough_memory_kern(VM_ACCT(size)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166}
167
168static inline void shmem_unacct_size(unsigned long flags, loff_t size)
169{
170 if (flags & VM_ACCOUNT)
171 vm_unacct_memory(VM_ACCT(size));
172}
173
174/*
175 * ... whereas tmpfs objects are accounted incrementally as
176 * pages are allocated, in order to allow huge sparse files.
177 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
178 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
179 */
180static inline int shmem_acct_block(unsigned long flags)
181{
Alan Cox731572d2008-10-29 14:01:20 -0700182 return (flags & VM_ACCOUNT) ?
183 0 : security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184}
185
186static inline void shmem_unacct_blocks(unsigned long flags, long pages)
187{
188 if (!(flags & VM_ACCOUNT))
189 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
190}
191
Hugh Dickins759b9772007-03-05 00:30:28 -0800192static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700193static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800194static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800195static const struct inode_operations shmem_inode_operations;
196static const struct inode_operations shmem_dir_inode_operations;
197static const struct inode_operations shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198static struct vm_operations_struct shmem_vm_ops;
199
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700200static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 .ra_pages = 0, /* No readahead */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700202 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 .unplug_io_fn = default_unplug_io_fn,
204};
205
206static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800207static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
209static void shmem_free_blocks(struct inode *inode, long pages)
210{
211 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700212 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 spin_lock(&sbinfo->stat_lock);
214 sbinfo->free_blocks += pages;
215 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
216 spin_unlock(&sbinfo->stat_lock);
217 }
218}
219
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800220static int shmem_reserve_inode(struct super_block *sb)
221{
222 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
223 if (sbinfo->max_inodes) {
224 spin_lock(&sbinfo->stat_lock);
225 if (!sbinfo->free_inodes) {
226 spin_unlock(&sbinfo->stat_lock);
227 return -ENOSPC;
228 }
229 sbinfo->free_inodes--;
230 spin_unlock(&sbinfo->stat_lock);
231 }
232 return 0;
233}
234
235static void shmem_free_inode(struct super_block *sb)
236{
237 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
238 if (sbinfo->max_inodes) {
239 spin_lock(&sbinfo->stat_lock);
240 sbinfo->free_inodes++;
241 spin_unlock(&sbinfo->stat_lock);
242 }
243}
244
Randy Dunlap46711812008-03-19 17:00:41 -0700245/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 * shmem_recalc_inode - recalculate the size of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 * @inode: inode to recalc
248 *
249 * We have to calculate the free blocks since the mm can drop
250 * undirtied hole pages behind our back.
251 *
252 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
253 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
254 *
255 * It has to be called with the spinlock held.
256 */
257static void shmem_recalc_inode(struct inode *inode)
258{
259 struct shmem_inode_info *info = SHMEM_I(inode);
260 long freed;
261
262 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
263 if (freed > 0) {
264 info->alloced -= freed;
265 shmem_unacct_blocks(info->flags, freed);
266 shmem_free_blocks(inode, freed);
267 }
268}
269
Randy Dunlap46711812008-03-19 17:00:41 -0700270/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 * shmem_swp_entry - find the swap vector position in the info structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 * @info: info structure for the inode
273 * @index: index of the page to find
274 * @page: optional page to add to the structure. Has to be preset to
275 * all zeros
276 *
277 * If there is no space allocated yet it will return NULL when
278 * page is NULL, else it will use the page for the needed block,
279 * setting it to NULL on return to indicate that it has been used.
280 *
281 * The swap vector is organized the following way:
282 *
283 * There are SHMEM_NR_DIRECT entries directly stored in the
284 * shmem_inode_info structure. So small files do not need an addional
285 * allocation.
286 *
287 * For pages with index > SHMEM_NR_DIRECT there is the pointer
288 * i_indirect which points to a page which holds in the first half
289 * doubly indirect blocks, in the second half triple indirect blocks:
290 *
291 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
292 * following layout (for SHMEM_NR_DIRECT == 16):
293 *
294 * i_indirect -> dir --> 16-19
295 * | +-> 20-23
296 * |
297 * +-->dir2 --> 24-27
298 * | +-> 28-31
299 * | +-> 32-35
300 * | +-> 36-39
301 * |
302 * +-->dir3 --> 40-43
303 * +-> 44-47
304 * +-> 48-51
305 * +-> 52-55
306 */
307static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
308{
309 unsigned long offset;
310 struct page **dir;
311 struct page *subdir;
312
313 if (index < SHMEM_NR_DIRECT) {
314 shmem_swp_balance_unmap();
315 return info->i_direct+index;
316 }
317 if (!info->i_indirect) {
318 if (page) {
319 info->i_indirect = *page;
320 *page = NULL;
321 }
322 return NULL; /* need another page */
323 }
324
325 index -= SHMEM_NR_DIRECT;
326 offset = index % ENTRIES_PER_PAGE;
327 index /= ENTRIES_PER_PAGE;
328 dir = shmem_dir_map(info->i_indirect);
329
330 if (index >= ENTRIES_PER_PAGE/2) {
331 index -= ENTRIES_PER_PAGE/2;
332 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
333 index %= ENTRIES_PER_PAGE;
334 subdir = *dir;
335 if (!subdir) {
336 if (page) {
337 *dir = *page;
338 *page = NULL;
339 }
340 shmem_dir_unmap(dir);
341 return NULL; /* need another page */
342 }
343 shmem_dir_unmap(dir);
344 dir = shmem_dir_map(subdir);
345 }
346
347 dir += index;
348 subdir = *dir;
349 if (!subdir) {
350 if (!page || !(subdir = *page)) {
351 shmem_dir_unmap(dir);
352 return NULL; /* need a page */
353 }
354 *dir = subdir;
355 *page = NULL;
356 }
357 shmem_dir_unmap(dir);
358 return shmem_swp_map(subdir) + offset;
359}
360
361static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
362{
363 long incdec = value? 1: -1;
364
365 entry->val = value;
366 info->swapped += incdec;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700367 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
368 struct page *page = kmap_atomic_to_page(entry);
369 set_page_private(page, page_private(page) + incdec);
370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371}
372
Randy Dunlap46711812008-03-19 17:00:41 -0700373/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 * shmem_swp_alloc - get the position of the swap entry for the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * @info: info structure for the inode
376 * @index: index of the page to find
377 * @sgp: check and recheck i_size? skip allocation?
Randy Dunlap46711812008-03-19 17:00:41 -0700378 *
379 * If the entry does not exist, allocate it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 */
381static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
382{
383 struct inode *inode = &info->vfs_inode;
384 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
385 struct page *page = NULL;
386 swp_entry_t *entry;
387
388 if (sgp != SGP_WRITE &&
389 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
390 return ERR_PTR(-EINVAL);
391
392 while (!(entry = shmem_swp_entry(info, index, &page))) {
393 if (sgp == SGP_READ)
394 return shmem_swp_map(ZERO_PAGE(0));
395 /*
396 * Test free_blocks against 1 not 0, since we have 1 data
397 * page (and perhaps indirect index pages) yet to allocate:
398 * a waste to allocate index if we cannot allocate data.
399 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700400 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 spin_lock(&sbinfo->stat_lock);
402 if (sbinfo->free_blocks <= 1) {
403 spin_unlock(&sbinfo->stat_lock);
404 return ERR_PTR(-ENOSPC);
405 }
406 sbinfo->free_blocks--;
407 inode->i_blocks += BLOCKS_PER_PAGE;
408 spin_unlock(&sbinfo->stat_lock);
409 }
410
411 spin_unlock(&info->lock);
Mel Gorman769848c2007-07-17 04:03:05 -0700412 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700413 if (page)
414 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 spin_lock(&info->lock);
416
417 if (!page) {
418 shmem_free_blocks(inode, 1);
419 return ERR_PTR(-ENOMEM);
420 }
421 if (sgp != SGP_WRITE &&
422 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
423 entry = ERR_PTR(-EINVAL);
424 break;
425 }
426 if (info->next_index <= index)
427 info->next_index = index + 1;
428 }
429 if (page) {
430 /* another task gave its page, or truncated the file */
431 shmem_free_blocks(inode, 1);
432 shmem_dir_free(page);
433 }
434 if (info->next_index <= index && !IS_ERR(entry))
435 info->next_index = index + 1;
436 return entry;
437}
438
Randy Dunlap46711812008-03-19 17:00:41 -0700439/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 * shmem_free_swp - free some swap entries in a directory
Hugh Dickins1ae70002007-03-29 01:20:36 -0700441 * @dir: pointer to the directory
442 * @edir: pointer after last entry of the directory
443 * @punch_lock: pointer to spinlock when needed for the holepunch case
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 */
Hugh Dickins1ae70002007-03-29 01:20:36 -0700445static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
446 spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447{
Hugh Dickins1ae70002007-03-29 01:20:36 -0700448 spinlock_t *punch_unlock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 swp_entry_t *ptr;
450 int freed = 0;
451
452 for (ptr = dir; ptr < edir; ptr++) {
453 if (ptr->val) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700454 if (unlikely(punch_lock)) {
455 punch_unlock = punch_lock;
456 punch_lock = NULL;
457 spin_lock(punch_unlock);
458 if (!ptr->val)
459 continue;
460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 free_swap_and_cache(*ptr);
462 *ptr = (swp_entry_t){0};
463 freed++;
464 }
465 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700466 if (punch_unlock)
467 spin_unlock(punch_unlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 return freed;
469}
470
Hugh Dickins1ae70002007-03-29 01:20:36 -0700471static int shmem_map_and_free_swp(struct page *subdir, int offset,
472 int limit, struct page ***dir, spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473{
474 swp_entry_t *ptr;
475 int freed = 0;
476
477 ptr = shmem_swp_map(subdir);
478 for (; offset < limit; offset += LATENCY_LIMIT) {
479 int size = limit - offset;
480 if (size > LATENCY_LIMIT)
481 size = LATENCY_LIMIT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700482 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
483 punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 if (need_resched()) {
485 shmem_swp_unmap(ptr);
486 if (*dir) {
487 shmem_dir_unmap(*dir);
488 *dir = NULL;
489 }
490 cond_resched();
491 ptr = shmem_swp_map(subdir);
492 }
493 }
494 shmem_swp_unmap(ptr);
495 return freed;
496}
497
498static void shmem_free_pages(struct list_head *next)
499{
500 struct page *page;
501 int freed = 0;
502
503 do {
504 page = container_of(next, struct page, lru);
505 next = next->next;
506 shmem_dir_free(page);
507 freed++;
508 if (freed >= LATENCY_LIMIT) {
509 cond_resched();
510 freed = 0;
511 }
512 } while (next);
513}
514
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800515static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516{
517 struct shmem_inode_info *info = SHMEM_I(inode);
518 unsigned long idx;
519 unsigned long size;
520 unsigned long limit;
521 unsigned long stage;
522 unsigned long diroff;
523 struct page **dir;
524 struct page *topdir;
525 struct page *middir;
526 struct page *subdir;
527 swp_entry_t *ptr;
528 LIST_HEAD(pages_to_free);
529 long nr_pages_to_free = 0;
530 long nr_swaps_freed = 0;
531 int offset;
532 int freed;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700533 int punch_hole;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700534 spinlock_t *needs_lock;
535 spinlock_t *punch_lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700536 unsigned long upper_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
538 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800539 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 if (idx >= info->next_index)
541 return;
542
543 spin_lock(&info->lock);
544 info->flags |= SHMEM_TRUNCATE;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800545 if (likely(end == (loff_t) -1)) {
546 limit = info->next_index;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700547 upper_limit = SHMEM_MAX_INDEX;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800548 info->next_index = idx;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700549 needs_lock = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700550 punch_hole = 0;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800551 } else {
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700552 if (end + 1 >= inode->i_size) { /* we may free a little more */
553 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
554 PAGE_CACHE_SHIFT;
555 upper_limit = SHMEM_MAX_INDEX;
556 } else {
557 limit = (end + 1) >> PAGE_CACHE_SHIFT;
558 upper_limit = limit;
559 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700560 needs_lock = &info->lock;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800561 punch_hole = 1;
562 }
563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 topdir = info->i_indirect;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800565 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 info->i_indirect = NULL;
567 nr_pages_to_free++;
568 list_add(&topdir->lru, &pages_to_free);
569 }
570 spin_unlock(&info->lock);
571
572 if (info->swapped && idx < SHMEM_NR_DIRECT) {
573 ptr = info->i_direct;
574 size = limit;
575 if (size > SHMEM_NR_DIRECT)
576 size = SHMEM_NR_DIRECT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700577 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 }
Badari Pulavarty92a3d032006-12-22 01:06:23 -0800579
580 /*
581 * If there are no indirect blocks or we are punching a hole
582 * below indirect blocks, nothing to be done.
583 */
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700584 if (!topdir || limit <= SHMEM_NR_DIRECT)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 goto done2;
586
Hugh Dickins1ae70002007-03-29 01:20:36 -0700587 /*
588 * The truncation case has already dropped info->lock, and we're safe
589 * because i_size and next_index have already been lowered, preventing
590 * access beyond. But in the punch_hole case, we still need to take
591 * the lock when updating the swap directory, because there might be
592 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
593 * shmem_writepage. However, whenever we find we can remove a whole
594 * directory page (not at the misaligned start or end of the range),
595 * we first NULLify its pointer in the level above, and then have no
596 * need to take the lock when updating its contents: needs_lock and
597 * punch_lock (either pointing to info->lock or NULL) manage this.
598 */
599
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700600 upper_limit -= SHMEM_NR_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 limit -= SHMEM_NR_DIRECT;
602 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
603 offset = idx % ENTRIES_PER_PAGE;
604 idx -= offset;
605
606 dir = shmem_dir_map(topdir);
607 stage = ENTRIES_PER_PAGEPAGE/2;
608 if (idx < ENTRIES_PER_PAGEPAGE/2) {
609 middir = topdir;
610 diroff = idx/ENTRIES_PER_PAGE;
611 } else {
612 dir += ENTRIES_PER_PAGE/2;
613 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
614 while (stage <= idx)
615 stage += ENTRIES_PER_PAGEPAGE;
616 middir = *dir;
617 if (*dir) {
618 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
619 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700620 if (!diroff && !offset && upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700621 if (needs_lock) {
622 spin_lock(needs_lock);
623 *dir = NULL;
624 spin_unlock(needs_lock);
625 needs_lock = NULL;
626 } else
627 *dir = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 nr_pages_to_free++;
629 list_add(&middir->lru, &pages_to_free);
630 }
631 shmem_dir_unmap(dir);
632 dir = shmem_dir_map(middir);
633 } else {
634 diroff = 0;
635 offset = 0;
636 idx = stage;
637 }
638 }
639
640 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
641 if (unlikely(idx == stage)) {
642 shmem_dir_unmap(dir);
643 dir = shmem_dir_map(topdir) +
644 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
645 while (!*dir) {
646 dir++;
647 idx += ENTRIES_PER_PAGEPAGE;
648 if (idx >= limit)
649 goto done1;
650 }
651 stage = idx + ENTRIES_PER_PAGEPAGE;
652 middir = *dir;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700653 if (punch_hole)
654 needs_lock = &info->lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700655 if (upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700656 if (needs_lock) {
657 spin_lock(needs_lock);
658 *dir = NULL;
659 spin_unlock(needs_lock);
660 needs_lock = NULL;
661 } else
662 *dir = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700663 nr_pages_to_free++;
664 list_add(&middir->lru, &pages_to_free);
665 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 shmem_dir_unmap(dir);
667 cond_resched();
668 dir = shmem_dir_map(middir);
669 diroff = 0;
670 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700671 punch_lock = needs_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 subdir = dir[diroff];
Hugh Dickins1ae70002007-03-29 01:20:36 -0700673 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
674 if (needs_lock) {
675 spin_lock(needs_lock);
676 dir[diroff] = NULL;
677 spin_unlock(needs_lock);
678 punch_lock = NULL;
679 } else
680 dir[diroff] = NULL;
681 nr_pages_to_free++;
682 list_add(&subdir->lru, &pages_to_free);
683 }
684 if (subdir && page_private(subdir) /* has swap entries */) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 size = limit - idx;
686 if (size > ENTRIES_PER_PAGE)
687 size = ENTRIES_PER_PAGE;
688 freed = shmem_map_and_free_swp(subdir,
Hugh Dickins1ae70002007-03-29 01:20:36 -0700689 offset, size, &dir, punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if (!dir)
691 dir = shmem_dir_map(middir);
692 nr_swaps_freed += freed;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700693 if (offset || punch_lock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 spin_lock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700695 set_page_private(subdir,
696 page_private(subdir) - freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 spin_unlock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700698 } else
699 BUG_ON(page_private(subdir) != freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700701 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 }
703done1:
704 shmem_dir_unmap(dir);
705done2:
706 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
707 /*
708 * Call truncate_inode_pages again: racing shmem_unuse_inode
709 * may have swizzled a page in from swap since vmtruncate or
710 * generic_delete_inode did it, before we lowered next_index.
711 * Also, though shmem_getpage checks i_size before adding to
712 * cache, no recheck after: so fix the narrow window there too.
Hugh Dickins16a10012007-03-29 01:20:37 -0700713 *
714 * Recalling truncate_inode_pages_range and unmap_mapping_range
715 * every time for punch_hole (which never got a chance to clear
716 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
717 * yet hardly ever necessary: try to optimize them out later.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800719 truncate_inode_pages_range(inode->i_mapping, start, end);
Hugh Dickins16a10012007-03-29 01:20:37 -0700720 if (punch_hole)
721 unmap_mapping_range(inode->i_mapping, start,
722 end - start, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 }
724
725 spin_lock(&info->lock);
726 info->flags &= ~SHMEM_TRUNCATE;
727 info->swapped -= nr_swaps_freed;
728 if (nr_pages_to_free)
729 shmem_free_blocks(inode, nr_pages_to_free);
730 shmem_recalc_inode(inode);
731 spin_unlock(&info->lock);
732
733 /*
734 * Empty swap vector directory pages to be freed?
735 */
736 if (!list_empty(&pages_to_free)) {
737 pages_to_free.prev->next = NULL;
738 shmem_free_pages(pages_to_free.next);
739 }
740}
741
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800742static void shmem_truncate(struct inode *inode)
743{
744 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
745}
746
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
748{
749 struct inode *inode = dentry->d_inode;
750 struct page *page = NULL;
751 int error;
752
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700753 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 if (attr->ia_size < inode->i_size) {
755 /*
756 * If truncating down to a partial page, then
757 * if that page is already allocated, hold it
758 * in memory until the truncation is over, so
759 * truncate_partial_page cannnot miss it were
760 * it assigned to swap.
761 */
762 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
763 (void) shmem_getpage(inode,
764 attr->ia_size>>PAGE_CACHE_SHIFT,
765 &page, SGP_READ, NULL);
Hugh Dickinsd3602442008-02-04 22:28:44 -0800766 if (page)
767 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 }
769 /*
770 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
771 * detect if any pages might have been added to cache
772 * after truncate_inode_pages. But we needn't bother
773 * if it's being fully truncated to zero-length: the
774 * nrpages check is efficient enough in that case.
775 */
776 if (attr->ia_size) {
777 struct shmem_inode_info *info = SHMEM_I(inode);
778 spin_lock(&info->lock);
779 info->flags &= ~SHMEM_PAGEIN;
780 spin_unlock(&info->lock);
781 }
782 }
783 }
784
785 error = inode_change_ok(inode, attr);
786 if (!error)
787 error = inode_setattr(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700788#ifdef CONFIG_TMPFS_POSIX_ACL
789 if (!error && (attr->ia_valid & ATTR_MODE))
790 error = generic_acl_chmod(inode, &shmem_acl_ops);
791#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 if (page)
793 page_cache_release(page);
794 return error;
795}
796
797static void shmem_delete_inode(struct inode *inode)
798{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 struct shmem_inode_info *info = SHMEM_I(inode);
800
801 if (inode->i_op->truncate == shmem_truncate) {
Mark Fashehfef26652005-09-09 13:01:31 -0700802 truncate_inode_pages(inode->i_mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 shmem_unacct_size(info->flags, inode->i_size);
804 inode->i_size = 0;
805 shmem_truncate(inode);
806 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800807 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800809 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 }
811 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700812 BUG_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800813 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 clear_inode(inode);
815}
816
817static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
818{
819 swp_entry_t *ptr;
820
821 for (ptr = dir; ptr < edir; ptr++) {
822 if (ptr->val == entry.val)
823 return ptr - dir;
824 }
825 return -1;
826}
827
828static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
829{
830 struct inode *inode;
831 unsigned long idx;
832 unsigned long size;
833 unsigned long limit;
834 unsigned long stage;
835 struct page **dir;
836 struct page *subdir;
837 swp_entry_t *ptr;
838 int offset;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800839 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
841 idx = 0;
842 ptr = info->i_direct;
843 spin_lock(&info->lock);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800844 if (!info->swapped) {
845 list_del_init(&info->swaplist);
846 goto lost2;
847 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 limit = info->next_index;
849 size = limit;
850 if (size > SHMEM_NR_DIRECT)
851 size = SHMEM_NR_DIRECT;
852 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800853 if (offset >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 goto found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 if (!info->i_indirect)
856 goto lost2;
857
858 dir = shmem_dir_map(info->i_indirect);
859 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
860
861 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
862 if (unlikely(idx == stage)) {
863 shmem_dir_unmap(dir-1);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800864 if (cond_resched_lock(&info->lock)) {
865 /* check it has not been truncated */
866 if (limit > info->next_index) {
867 limit = info->next_index;
868 if (idx >= limit)
869 goto lost2;
870 }
871 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 dir = shmem_dir_map(info->i_indirect) +
873 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
874 while (!*dir) {
875 dir++;
876 idx += ENTRIES_PER_PAGEPAGE;
877 if (idx >= limit)
878 goto lost1;
879 }
880 stage = idx + ENTRIES_PER_PAGEPAGE;
881 subdir = *dir;
882 shmem_dir_unmap(dir);
883 dir = shmem_dir_map(subdir);
884 }
885 subdir = *dir;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700886 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 ptr = shmem_swp_map(subdir);
888 size = limit - idx;
889 if (size > ENTRIES_PER_PAGE)
890 size = ENTRIES_PER_PAGE;
891 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800892 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (offset >= 0) {
894 shmem_dir_unmap(dir);
895 goto found;
896 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 }
898 }
899lost1:
900 shmem_dir_unmap(dir-1);
901lost2:
902 spin_unlock(&info->lock);
903 return 0;
904found:
905 idx += offset;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800906 inode = igrab(&info->vfs_inode);
907 spin_unlock(&info->lock);
908
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800909 /*
910 * Move _head_ to start search for next from here.
911 * But be careful: shmem_delete_inode checks list_empty without taking
912 * mutex, and there's an instant in list_move_tail when info->swaplist
913 * would appear empty, if it were the only one on shmem_swaplist. We
914 * could avoid doing it if inode NULL; or use this minor optimization.
915 */
916 if (shmem_swaplist.next != &info->swaplist)
917 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800918 mutex_unlock(&shmem_swaplist_mutex);
919
920 error = 1;
921 if (!inode)
922 goto out;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700923 /* Precharge page using GFP_KERNEL while we can wait */
Hugh Dickins82369552008-02-07 00:14:22 -0800924 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800925 if (error)
926 goto out;
Hugh Dickins82369552008-02-07 00:14:22 -0800927 error = radix_tree_preload(GFP_KERNEL);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700928 if (error) {
929 mem_cgroup_uncharge_cache_page(page);
930 goto out;
931 }
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800932 error = 1;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800933
934 spin_lock(&info->lock);
935 ptr = shmem_swp_entry(info, idx, NULL);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700936 if (ptr && ptr->val == entry.val) {
Nick Piggine2867812008-07-25 19:45:30 -0700937 error = add_to_page_cache_locked(page, inode->i_mapping,
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800938 idx, GFP_NOWAIT);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700939 /* does mem_cgroup_uncharge_cache_page on error */
940 } else /* we must compensate for our precharge above */
941 mem_cgroup_uncharge_cache_page(page);
942
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800943 if (error == -EEXIST) {
944 struct page *filepage = find_get_page(inode->i_mapping, idx);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800945 error = 1;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800946 if (filepage) {
947 /*
948 * There might be a more uptodate page coming down
949 * from a stacked writepage: forget our swappage if so.
950 */
951 if (PageUptodate(filepage))
952 error = 0;
953 page_cache_release(filepage);
954 }
955 }
956 if (!error) {
Hugh Dickins73b12622008-02-04 22:28:50 -0800957 delete_from_swap_cache(page);
958 set_page_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 info->flags |= SHMEM_PAGEIN;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800960 shmem_swp_set(info, ptr, 0);
961 swap_free(entry);
962 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800964 if (ptr)
965 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 spin_unlock(&info->lock);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800967 radix_tree_preload_end();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800968out:
969 unlock_page(page);
970 page_cache_release(page);
971 iput(inode); /* allows for NULL */
972 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973}
974
975/*
976 * shmem_unuse() search for an eventually swapped out shmem page.
977 */
978int shmem_unuse(swp_entry_t entry, struct page *page)
979{
980 struct list_head *p, *next;
981 struct shmem_inode_info *info;
982 int found = 0;
983
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800984 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 list_for_each_safe(p, next, &shmem_swaplist) {
986 info = list_entry(p, struct shmem_inode_info, swaplist);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800987 found = shmem_unuse_inode(info, entry, page);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800988 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800989 if (found)
990 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800992 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800993out: return found; /* 0 or 1 or -ENOMEM */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994}
995
996/*
997 * Move the page from the page cache to the swap cache.
998 */
999static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1000{
1001 struct shmem_inode_info *info;
1002 swp_entry_t *entry, swap;
1003 struct address_space *mapping;
1004 unsigned long index;
1005 struct inode *inode;
1006
1007 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 mapping = page->mapping;
1009 index = page->index;
1010 inode = mapping->host;
1011 info = SHMEM_I(inode);
1012 if (info->flags & VM_LOCKED)
1013 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001014 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 goto redirty;
1016
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001017 /*
1018 * shmem_backing_dev_info's capabilities prevent regular writeback or
1019 * sync from ever calling shmem_writepage; but a stacking filesystem
1020 * may use the ->writepage of its underlying filesystem, in which case
1021 * tmpfs should write out to swap only in response to memory pressure,
1022 * and not for pdflush or sync. However, in those cases, we do still
1023 * want to check if there's a redundant swappage to be discarded.
1024 */
1025 if (wbc->for_reclaim)
1026 swap = get_swap_page();
1027 else
1028 swap.val = 0;
1029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 if (index >= info->next_index) {
1032 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1033 goto unlock;
1034 }
1035 entry = shmem_swp_entry(info, index, NULL);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001036 if (entry->val) {
1037 /*
1038 * The more uptodate page coming down from a stacked
1039 * writepage should replace our old swappage.
1040 */
1041 free_swap_and_cache(*entry);
1042 shmem_swp_set(info, entry, 0);
1043 }
1044 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001046 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickins73b12622008-02-04 22:28:50 -08001047 remove_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 shmem_swp_set(info, entry, swap.val);
1049 shmem_swp_unmap(entry);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001050 if (list_empty(&info->swaplist))
1051 inode = igrab(inode);
1052 else
1053 inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 spin_unlock(&info->lock);
Hugh Dickins73b12622008-02-04 22:28:50 -08001055 swap_duplicate(swap);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001056 BUG_ON(page_mapped(page));
Hugh Dickins73b12622008-02-04 22:28:50 -08001057 page_cache_release(page); /* pagecache ref */
1058 set_page_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 unlock_page(page);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001060 if (inode) {
1061 mutex_lock(&shmem_swaplist_mutex);
1062 /* move instead of add in case we're racing */
1063 list_move_tail(&info->swaplist, &shmem_swaplist);
1064 mutex_unlock(&shmem_swaplist_mutex);
1065 iput(inode);
1066 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 return 0;
1068 }
1069
1070 shmem_swp_unmap(entry);
1071unlock:
1072 spin_unlock(&info->lock);
1073 swap_free(swap);
1074redirty:
1075 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001076 if (wbc->for_reclaim)
1077 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1078 unlock_page(page);
1079 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080}
1081
1082#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001083#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001084static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001085{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001086 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001087
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001088 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001089 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001090
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001091 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001092
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001093 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001094}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001095
1096static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1097{
1098 struct mempolicy *mpol = NULL;
1099 if (sbinfo->mpol) {
1100 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1101 mpol = sbinfo->mpol;
1102 mpol_get(mpol);
1103 spin_unlock(&sbinfo->stat_lock);
1104 }
1105 return mpol;
1106}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001107#endif /* CONFIG_TMPFS */
1108
Hugh Dickins02098fe2008-02-04 22:28:42 -08001109static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1110 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111{
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001112 struct mempolicy mpol, *spol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 struct vm_area_struct pvma;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001114 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001116 spol = mpol_cond_copy(&mpol,
1117 mpol_shared_policy_lookup(&info->policy, idx));
1118
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 /* Create a pseudo vma that just contains the policy */
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001120 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001122 pvma.vm_ops = NULL;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001123 pvma.vm_policy = spol;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001124 page = swapin_readahead(entry, gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 return page;
1126}
1127
Hugh Dickins02098fe2008-02-04 22:28:42 -08001128static struct page *shmem_alloc_page(gfp_t gfp,
1129 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130{
1131 struct vm_area_struct pvma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001133 /* Create a pseudo vma that just contains the policy */
1134 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001136 pvma.vm_ops = NULL;
1137 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001138
1139 /*
1140 * alloc_page_vma() will drop the shared policy reference
1141 */
1142 return alloc_page_vma(gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001144#else /* !CONFIG_NUMA */
1145#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001146static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001147{
1148}
1149#endif /* CONFIG_TMPFS */
1150
Hugh Dickins02098fe2008-02-04 22:28:42 -08001151static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1152 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153{
Hugh Dickins02098fe2008-02-04 22:28:42 -08001154 return swapin_readahead(entry, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155}
1156
Hugh Dickins02098fe2008-02-04 22:28:42 -08001157static inline struct page *shmem_alloc_page(gfp_t gfp,
1158 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159{
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001160 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001162#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001164#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1165static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1166{
1167 return NULL;
1168}
1169#endif
1170
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171/*
1172 * shmem_getpage - either get the page from swap or allocate a new one
1173 *
1174 * If we allocate a new one we do not mark it dirty. That's up to the
1175 * vm. If we swap it in we mark it dirty since we also free the swap
1176 * entry since a page cannot live in both the swap and page cache
1177 */
1178static int shmem_getpage(struct inode *inode, unsigned long idx,
1179 struct page **pagep, enum sgp_type sgp, int *type)
1180{
1181 struct address_space *mapping = inode->i_mapping;
1182 struct shmem_inode_info *info = SHMEM_I(inode);
1183 struct shmem_sb_info *sbinfo;
1184 struct page *filepage = *pagep;
1185 struct page *swappage;
1186 swp_entry_t *entry;
1187 swp_entry_t swap;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001188 gfp_t gfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 int error;
1190
1191 if (idx >= SHMEM_MAX_INDEX)
1192 return -EFBIG;
Nick Piggin54cb8822007-07-19 01:46:59 -07001193
1194 if (type)
Nick Piggin83c54072007-07-19 01:47:05 -07001195 *type = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07001196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 /*
1198 * Normally, filepage is NULL on entry, and either found
1199 * uptodate immediately, or allocated and zeroed, or read
1200 * in under swappage, which is then assigned to filepage.
Hugh Dickins5402b972008-02-04 22:28:44 -08001201 * But shmem_readpage (required for splice) passes in a locked
Hugh Dickinsae976412007-06-04 10:00:39 +02001202 * filepage, which may be found not uptodate by other callers
1203 * too, and may need to be copied from the swappage read in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 */
1205repeat:
1206 if (!filepage)
1207 filepage = find_lock_page(mapping, idx);
1208 if (filepage && PageUptodate(filepage))
1209 goto done;
1210 error = 0;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001211 gfp = mapping_gfp_mask(mapping);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -08001212 if (!filepage) {
1213 /*
1214 * Try to preload while we can wait, to not make a habit of
1215 * draining atomic reserves; but don't latch on to this cpu.
1216 */
1217 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1218 if (error)
1219 goto failed;
1220 radix_tree_preload_end();
1221 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
1223 spin_lock(&info->lock);
1224 shmem_recalc_inode(inode);
1225 entry = shmem_swp_alloc(info, idx, sgp);
1226 if (IS_ERR(entry)) {
1227 spin_unlock(&info->lock);
1228 error = PTR_ERR(entry);
1229 goto failed;
1230 }
1231 swap = *entry;
1232
1233 if (swap.val) {
1234 /* Look it up and read it in.. */
1235 swappage = lookup_swap_cache(swap);
1236 if (!swappage) {
1237 shmem_swp_unmap(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 /* here we actually do the io */
Nick Piggin83c54072007-07-19 01:47:05 -07001239 if (type && !(*type & VM_FAULT_MAJOR)) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001240 __count_vm_event(PGMAJFAULT);
Nick Piggin83c54072007-07-19 01:47:05 -07001241 *type |= VM_FAULT_MAJOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 }
Christoph Lameterf8891e52006-06-30 01:55:45 -07001243 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001244 swappage = shmem_swapin(swap, gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 if (!swappage) {
1246 spin_lock(&info->lock);
1247 entry = shmem_swp_alloc(info, idx, sgp);
1248 if (IS_ERR(entry))
1249 error = PTR_ERR(entry);
1250 else {
1251 if (entry->val == swap.val)
1252 error = -ENOMEM;
1253 shmem_swp_unmap(entry);
1254 }
1255 spin_unlock(&info->lock);
1256 if (error)
1257 goto failed;
1258 goto repeat;
1259 }
1260 wait_on_page_locked(swappage);
1261 page_cache_release(swappage);
1262 goto repeat;
1263 }
1264
1265 /* We have to do this with page locked to prevent races */
Nick Piggin529ae9a2008-08-02 12:01:03 +02001266 if (!trylock_page(swappage)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 shmem_swp_unmap(entry);
1268 spin_unlock(&info->lock);
1269 wait_on_page_locked(swappage);
1270 page_cache_release(swappage);
1271 goto repeat;
1272 }
1273 if (PageWriteback(swappage)) {
1274 shmem_swp_unmap(entry);
1275 spin_unlock(&info->lock);
1276 wait_on_page_writeback(swappage);
1277 unlock_page(swappage);
1278 page_cache_release(swappage);
1279 goto repeat;
1280 }
1281 if (!PageUptodate(swappage)) {
1282 shmem_swp_unmap(entry);
1283 spin_unlock(&info->lock);
1284 unlock_page(swappage);
1285 page_cache_release(swappage);
1286 error = -EIO;
1287 goto failed;
1288 }
1289
1290 if (filepage) {
1291 shmem_swp_set(info, entry, 0);
1292 shmem_swp_unmap(entry);
1293 delete_from_swap_cache(swappage);
1294 spin_unlock(&info->lock);
1295 copy_highpage(filepage, swappage);
1296 unlock_page(swappage);
1297 page_cache_release(swappage);
1298 flush_dcache_page(filepage);
1299 SetPageUptodate(filepage);
1300 set_page_dirty(filepage);
1301 swap_free(swap);
Nick Piggine2867812008-07-25 19:45:30 -07001302 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1303 idx, GFP_NOWAIT))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 info->flags |= SHMEM_PAGEIN;
1305 shmem_swp_set(info, entry, 0);
1306 shmem_swp_unmap(entry);
Hugh Dickins73b12622008-02-04 22:28:50 -08001307 delete_from_swap_cache(swappage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 spin_unlock(&info->lock);
1309 filepage = swappage;
Hugh Dickins73b12622008-02-04 22:28:50 -08001310 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 swap_free(swap);
1312 } else {
1313 shmem_swp_unmap(entry);
1314 spin_unlock(&info->lock);
1315 unlock_page(swappage);
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001316 page_cache_release(swappage);
Hugh Dickins82369552008-02-07 00:14:22 -08001317 if (error == -ENOMEM) {
1318 /* allow reclaim from this memory cgroup */
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001319 error = mem_cgroup_shrink_usage(current->mm,
1320 gfp);
1321 if (error)
Hugh Dickins82369552008-02-07 00:14:22 -08001322 goto failed;
1323 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 goto repeat;
1325 }
1326 } else if (sgp == SGP_READ && !filepage) {
1327 shmem_swp_unmap(entry);
1328 filepage = find_get_page(mapping, idx);
1329 if (filepage &&
Nick Piggin529ae9a2008-08-02 12:01:03 +02001330 (!PageUptodate(filepage) || !trylock_page(filepage))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 spin_unlock(&info->lock);
1332 wait_on_page_locked(filepage);
1333 page_cache_release(filepage);
1334 filepage = NULL;
1335 goto repeat;
1336 }
1337 spin_unlock(&info->lock);
1338 } else {
1339 shmem_swp_unmap(entry);
1340 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001341 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 spin_lock(&sbinfo->stat_lock);
1343 if (sbinfo->free_blocks == 0 ||
1344 shmem_acct_block(info->flags)) {
1345 spin_unlock(&sbinfo->stat_lock);
1346 spin_unlock(&info->lock);
1347 error = -ENOSPC;
1348 goto failed;
1349 }
1350 sbinfo->free_blocks--;
1351 inode->i_blocks += BLOCKS_PER_PAGE;
1352 spin_unlock(&sbinfo->stat_lock);
1353 } else if (shmem_acct_block(info->flags)) {
1354 spin_unlock(&info->lock);
1355 error = -ENOSPC;
1356 goto failed;
1357 }
1358
1359 if (!filepage) {
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001360 int ret;
1361
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001363 filepage = shmem_alloc_page(gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 if (!filepage) {
1365 shmem_unacct_blocks(info->flags, 1);
1366 shmem_free_blocks(inode, 1);
1367 error = -ENOMEM;
1368 goto failed;
1369 }
Rik van Rielb2e18532008-10-18 20:26:30 -07001370 SetPageSwapBacked(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
Hugh Dickins82369552008-02-07 00:14:22 -08001372 /* Precharge page while we can wait, compensate after */
1373 error = mem_cgroup_cache_charge(filepage, current->mm,
1374 gfp & ~__GFP_HIGHMEM);
1375 if (error) {
1376 page_cache_release(filepage);
1377 shmem_unacct_blocks(info->flags, 1);
1378 shmem_free_blocks(inode, 1);
1379 filepage = NULL;
1380 goto failed;
1381 }
1382
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 spin_lock(&info->lock);
1384 entry = shmem_swp_alloc(info, idx, sgp);
1385 if (IS_ERR(entry))
1386 error = PTR_ERR(entry);
1387 else {
1388 swap = *entry;
1389 shmem_swp_unmap(entry);
1390 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001391 ret = error || swap.val;
1392 if (ret)
1393 mem_cgroup_uncharge_cache_page(filepage);
1394 else
1395 ret = add_to_page_cache_lru(filepage, mapping,
1396 idx, GFP_NOWAIT);
1397 /*
1398 * At add_to_page_cache_lru() failure, uncharge will
1399 * be done automatically.
1400 */
1401 if (ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 spin_unlock(&info->lock);
1403 page_cache_release(filepage);
1404 shmem_unacct_blocks(info->flags, 1);
1405 shmem_free_blocks(inode, 1);
1406 filepage = NULL;
1407 if (error)
1408 goto failed;
1409 goto repeat;
1410 }
1411 info->flags |= SHMEM_PAGEIN;
1412 }
1413
1414 info->alloced++;
1415 spin_unlock(&info->lock);
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001416 clear_highpage(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 flush_dcache_page(filepage);
1418 SetPageUptodate(filepage);
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001419 if (sgp == SGP_DIRTY)
1420 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 }
1422done:
Hugh Dickinsd3602442008-02-04 22:28:44 -08001423 *pagep = filepage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 return 0;
1425
1426failed:
1427 if (*pagep != filepage) {
1428 unlock_page(filepage);
1429 page_cache_release(filepage);
1430 }
1431 return error;
1432}
1433
Nick Piggind0217ac2007-07-19 01:47:03 -07001434static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001436 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 int error;
Nick Piggind0217ac2007-07-19 01:47:03 -07001438 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
Nick Piggind0217ac2007-07-19 01:47:03 -07001440 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1441 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001442
Hugh Dickins27d54b32008-02-04 22:28:43 -08001443 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
Nick Piggind0217ac2007-07-19 01:47:03 -07001444 if (error)
1445 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
Nick Piggind0217ac2007-07-19 01:47:03 -07001447 mark_page_accessed(vmf->page);
Nick Piggin83c54072007-07-19 01:47:05 -07001448 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449}
1450
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451#ifdef CONFIG_NUMA
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001452static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001454 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1456}
1457
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001458static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1459 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001461 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 unsigned long idx;
1463
1464 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1465 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1466}
1467#endif
1468
1469int shmem_lock(struct file *file, int lock, struct user_struct *user)
1470{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001471 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 struct shmem_inode_info *info = SHMEM_I(inode);
1473 int retval = -ENOMEM;
1474
1475 spin_lock(&info->lock);
1476 if (lock && !(info->flags & VM_LOCKED)) {
1477 if (!user_shm_lock(inode->i_size, user))
1478 goto out_nomem;
1479 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001480 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 }
1482 if (!lock && (info->flags & VM_LOCKED) && user) {
1483 user_shm_unlock(inode->i_size, user);
1484 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001485 mapping_clear_unevictable(file->f_mapping);
1486 scan_mapping_unevictable_pages(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 }
1488 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001489
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490out_nomem:
1491 spin_unlock(&info->lock);
1492 return retval;
1493}
1494
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001495static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496{
1497 file_accessed(file);
1498 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001499 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 return 0;
1501}
1502
1503static struct inode *
1504shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1505{
1506 struct inode *inode;
1507 struct shmem_inode_info *info;
1508 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1509
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001510 if (shmem_reserve_inode(sb))
1511 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
1513 inode = new_inode(sb);
1514 if (inode) {
1515 inode->i_mode = mode;
David Howells76aac0e2008-11-14 10:39:12 +11001516 inode->i_uid = current_fsuid();
1517 inode->i_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1520 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001521 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 info = SHMEM_I(inode);
1523 memset(info, 0, (char *)inode - (char *)info);
1524 spin_lock_init(&info->lock);
1525 INIT_LIST_HEAD(&info->swaplist);
1526
1527 switch (mode & S_IFMT) {
1528 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001529 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 init_special_inode(inode, mode, dev);
1531 break;
1532 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001533 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 inode->i_op = &shmem_inode_operations;
1535 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001536 mpol_shared_policy_init(&info->policy,
1537 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 break;
1539 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001540 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 /* Some things misbehave if size == 0 on a directory */
1542 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1543 inode->i_op = &shmem_dir_inode_operations;
1544 inode->i_fop = &simple_dir_operations;
1545 break;
1546 case S_IFLNK:
1547 /*
1548 * Must not load anything in the rbtree,
1549 * mpol_free_shared_policy will not be called.
1550 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001551 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 break;
1553 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001554 } else
1555 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 return inode;
1557}
1558
1559#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001560static const struct inode_operations shmem_symlink_inode_operations;
1561static const struct inode_operations shmem_symlink_inline_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562
1563/*
Nick Piggin800d15a2007-10-16 01:25:03 -07001564 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
Hugh Dickinsae976412007-06-04 10:00:39 +02001565 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1566 * below the loop driver, in the generic fashion that many filesystems support.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 */
Hugh Dickinsae976412007-06-04 10:00:39 +02001568static int shmem_readpage(struct file *file, struct page *page)
1569{
1570 struct inode *inode = page->mapping->host;
1571 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1572 unlock_page(page);
1573 return error;
1574}
1575
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001577shmem_write_begin(struct file *file, struct address_space *mapping,
1578 loff_t pos, unsigned len, unsigned flags,
1579 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580{
Nick Piggin800d15a2007-10-16 01:25:03 -07001581 struct inode *inode = mapping->host;
1582 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1583 *pagep = NULL;
1584 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1585}
1586
1587static int
1588shmem_write_end(struct file *file, struct address_space *mapping,
1589 loff_t pos, unsigned len, unsigned copied,
1590 struct page *page, void *fsdata)
1591{
1592 struct inode *inode = mapping->host;
1593
Hugh Dickinsd3602442008-02-04 22:28:44 -08001594 if (pos + copied > inode->i_size)
1595 i_size_write(inode, pos + copied);
1596
1597 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001598 set_page_dirty(page);
1599 page_cache_release(page);
1600
Nick Piggin800d15a2007-10-16 01:25:03 -07001601 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602}
1603
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1605{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001606 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 struct address_space *mapping = inode->i_mapping;
1608 unsigned long index, offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001609 enum sgp_type sgp = SGP_READ;
1610
1611 /*
1612 * Might this read be for a stacking filesystem? Then when reading
1613 * holes of a sparse file, we actually need to allocate those pages,
1614 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1615 */
1616 if (segment_eq(get_fs(), KERNEL_DS))
1617 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
1619 index = *ppos >> PAGE_CACHE_SHIFT;
1620 offset = *ppos & ~PAGE_CACHE_MASK;
1621
1622 for (;;) {
1623 struct page *page = NULL;
1624 unsigned long end_index, nr, ret;
1625 loff_t i_size = i_size_read(inode);
1626
1627 end_index = i_size >> PAGE_CACHE_SHIFT;
1628 if (index > end_index)
1629 break;
1630 if (index == end_index) {
1631 nr = i_size & ~PAGE_CACHE_MASK;
1632 if (nr <= offset)
1633 break;
1634 }
1635
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001636 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 if (desc->error) {
1638 if (desc->error == -EINVAL)
1639 desc->error = 0;
1640 break;
1641 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001642 if (page)
1643 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
1645 /*
1646 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001647 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 */
1649 nr = PAGE_CACHE_SIZE;
1650 i_size = i_size_read(inode);
1651 end_index = i_size >> PAGE_CACHE_SHIFT;
1652 if (index == end_index) {
1653 nr = i_size & ~PAGE_CACHE_MASK;
1654 if (nr <= offset) {
1655 if (page)
1656 page_cache_release(page);
1657 break;
1658 }
1659 }
1660 nr -= offset;
1661
1662 if (page) {
1663 /*
1664 * If users can be writing to this page using arbitrary
1665 * virtual addresses, take care about potential aliasing
1666 * before reading the page on the kernel side.
1667 */
1668 if (mapping_writably_mapped(mapping))
1669 flush_dcache_page(page);
1670 /*
1671 * Mark the page accessed if we read the beginning.
1672 */
1673 if (!offset)
1674 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001675 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001677 page_cache_get(page);
1678 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
1680 /*
1681 * Ok, we have the page, and it's up-to-date, so
1682 * now we can copy it to user space...
1683 *
1684 * The actor routine returns how many bytes were actually used..
1685 * NOTE! This may not be the same as how much of a user buffer
1686 * we filled up (we may be padding etc), so we can only update
1687 * "pos" here (the actor routine has to update the user buffer
1688 * pointers and the remaining count).
1689 */
1690 ret = actor(desc, page, offset, nr);
1691 offset += ret;
1692 index += offset >> PAGE_CACHE_SHIFT;
1693 offset &= ~PAGE_CACHE_MASK;
1694
1695 page_cache_release(page);
1696 if (ret != nr || !desc->count)
1697 break;
1698
1699 cond_resched();
1700 }
1701
1702 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1703 file_accessed(filp);
1704}
1705
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001706static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1707 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708{
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001709 struct file *filp = iocb->ki_filp;
1710 ssize_t retval;
1711 unsigned long seg;
1712 size_t count;
1713 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001715 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1716 if (retval)
1717 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001719 for (seg = 0; seg < nr_segs; seg++) {
1720 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001722 desc.written = 0;
1723 desc.arg.buf = iov[seg].iov_base;
1724 desc.count = iov[seg].iov_len;
1725 if (desc.count == 0)
1726 continue;
1727 desc.error = 0;
1728 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1729 retval += desc.written;
1730 if (desc.error) {
1731 retval = retval ?: desc.error;
1732 break;
1733 }
1734 if (desc.count > 0)
1735 break;
1736 }
1737 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738}
1739
David Howells726c3342006-06-23 02:02:58 -07001740static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741{
David Howells726c3342006-06-23 02:02:58 -07001742 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
1744 buf->f_type = TMPFS_MAGIC;
1745 buf->f_bsize = PAGE_CACHE_SIZE;
1746 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001747 spin_lock(&sbinfo->stat_lock);
1748 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 buf->f_blocks = sbinfo->max_blocks;
1750 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001751 }
1752 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 buf->f_files = sbinfo->max_inodes;
1754 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 }
1756 /* else leave those fields 0 like simple_statfs */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001757 spin_unlock(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 return 0;
1759}
1760
1761/*
1762 * File creation. Allocate an inode, and we're done..
1763 */
1764static int
1765shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1766{
1767 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1768 int error = -ENOSPC;
1769
1770 if (inode) {
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001771 error = security_inode_init_security(inode, dir, NULL, NULL,
1772 NULL);
1773 if (error) {
1774 if (error != -EOPNOTSUPP) {
1775 iput(inode);
1776 return error;
1777 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001778 }
1779 error = shmem_acl_init(inode, dir);
1780 if (error) {
1781 iput(inode);
1782 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001783 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 if (dir->i_mode & S_ISGID) {
1785 inode->i_gid = dir->i_gid;
1786 if (S_ISDIR(mode))
1787 inode->i_mode |= S_ISGID;
1788 }
1789 dir->i_size += BOGO_DIRENT_SIZE;
1790 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1791 d_instantiate(dentry, inode);
1792 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 }
1794 return error;
1795}
1796
1797static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1798{
1799 int error;
1800
1801 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1802 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001803 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 return 0;
1805}
1806
1807static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1808 struct nameidata *nd)
1809{
1810 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1811}
1812
1813/*
1814 * Link a file..
1815 */
1816static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1817{
1818 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001819 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820
1821 /*
1822 * No ordinary (disk based) filesystem counts links as inodes;
1823 * but each new link needs a new dentry, pinning lowmem, and
1824 * tmpfs dentries cannot be pruned until they are unlinked.
1825 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001826 ret = shmem_reserve_inode(inode->i_sb);
1827 if (ret)
1828 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
1830 dir->i_size += BOGO_DIRENT_SIZE;
1831 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001832 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 atomic_inc(&inode->i_count); /* New dentry reference */
1834 dget(dentry); /* Extra pinning count for the created dentry */
1835 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001836out:
1837 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838}
1839
1840static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1841{
1842 struct inode *inode = dentry->d_inode;
1843
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001844 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1845 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846
1847 dir->i_size -= BOGO_DIRENT_SIZE;
1848 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001849 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 dput(dentry); /* Undo the count from "create" - this does all the work */
1851 return 0;
1852}
1853
1854static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1855{
1856 if (!simple_empty(dentry))
1857 return -ENOTEMPTY;
1858
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001859 drop_nlink(dentry->d_inode);
1860 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 return shmem_unlink(dir, dentry);
1862}
1863
1864/*
1865 * The VFS layer already does all the dentry stuff for rename,
1866 * we just have to decrement the usage count for the target if
1867 * it exists so that the VFS layer correctly free's it when it
1868 * gets overwritten.
1869 */
1870static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1871{
1872 struct inode *inode = old_dentry->d_inode;
1873 int they_are_dirs = S_ISDIR(inode->i_mode);
1874
1875 if (!simple_empty(new_dentry))
1876 return -ENOTEMPTY;
1877
1878 if (new_dentry->d_inode) {
1879 (void) shmem_unlink(new_dir, new_dentry);
1880 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001881 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001883 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001884 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 }
1886
1887 old_dir->i_size -= BOGO_DIRENT_SIZE;
1888 new_dir->i_size += BOGO_DIRENT_SIZE;
1889 old_dir->i_ctime = old_dir->i_mtime =
1890 new_dir->i_ctime = new_dir->i_mtime =
1891 inode->i_ctime = CURRENT_TIME;
1892 return 0;
1893}
1894
1895static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1896{
1897 int error;
1898 int len;
1899 struct inode *inode;
1900 struct page *page = NULL;
1901 char *kaddr;
1902 struct shmem_inode_info *info;
1903
1904 len = strlen(symname) + 1;
1905 if (len > PAGE_CACHE_SIZE)
1906 return -ENAMETOOLONG;
1907
1908 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1909 if (!inode)
1910 return -ENOSPC;
1911
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001912 error = security_inode_init_security(inode, dir, NULL, NULL,
1913 NULL);
1914 if (error) {
1915 if (error != -EOPNOTSUPP) {
1916 iput(inode);
1917 return error;
1918 }
1919 error = 0;
1920 }
1921
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 info = SHMEM_I(inode);
1923 inode->i_size = len-1;
1924 if (len <= (char *)inode - (char *)info) {
1925 /* do it inline */
1926 memcpy(info, symname, len);
1927 inode->i_op = &shmem_symlink_inline_operations;
1928 } else {
1929 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1930 if (error) {
1931 iput(inode);
1932 return error;
1933 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001934 unlock_page(page);
Hugh Dickins14fcc232008-07-28 15:46:19 -07001935 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 inode->i_op = &shmem_symlink_inode_operations;
1937 kaddr = kmap_atomic(page, KM_USER0);
1938 memcpy(kaddr, symname, len);
1939 kunmap_atomic(kaddr, KM_USER0);
1940 set_page_dirty(page);
1941 page_cache_release(page);
1942 }
1943 if (dir->i_mode & S_ISGID)
1944 inode->i_gid = dir->i_gid;
1945 dir->i_size += BOGO_DIRENT_SIZE;
1946 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1947 d_instantiate(dentry, inode);
1948 dget(dentry);
1949 return 0;
1950}
1951
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001952static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953{
1954 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001955 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956}
1957
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001958static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959{
1960 struct page *page = NULL;
1961 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1962 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08001963 if (page)
1964 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001965 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966}
1967
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001968static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969{
1970 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001971 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 kunmap(page);
1973 mark_page_accessed(page);
1974 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 }
1976}
1977
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001978static const struct inode_operations shmem_symlink_inline_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 .readlink = generic_readlink,
1980 .follow_link = shmem_follow_link_inline,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981};
1982
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001983static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 .truncate = shmem_truncate,
1985 .readlink = generic_readlink,
1986 .follow_link = shmem_follow_link,
1987 .put_link = shmem_put_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988};
1989
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001990#ifdef CONFIG_TMPFS_POSIX_ACL
Randy Dunlap46711812008-03-19 17:00:41 -07001991/*
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001992 * Superblocks without xattr inode operations will get security.* xattr
1993 * support from the VFS "for free". As soon as we have any other xattrs
1994 * like ACLs, we also need to implement the security.* handlers at
1995 * filesystem level, though.
1996 */
1997
1998static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1999 size_t list_len, const char *name,
2000 size_t name_len)
2001{
2002 return security_inode_listsecurity(inode, list, list_len);
2003}
2004
2005static int shmem_xattr_security_get(struct inode *inode, const char *name,
2006 void *buffer, size_t size)
2007{
2008 if (strcmp(name, "") == 0)
2009 return -EINVAL;
David P. Quigley42492592008-02-04 22:29:39 -08002010 return xattr_getsecurity(inode, name, buffer, size);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002011}
2012
2013static int shmem_xattr_security_set(struct inode *inode, const char *name,
2014 const void *value, size_t size, int flags)
2015{
2016 if (strcmp(name, "") == 0)
2017 return -EINVAL;
2018 return security_inode_setsecurity(inode, name, value, size, flags);
2019}
2020
Adrian Bunk1f370a22006-12-06 20:38:21 -08002021static struct xattr_handler shmem_xattr_security_handler = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002022 .prefix = XATTR_SECURITY_PREFIX,
2023 .list = shmem_xattr_security_list,
2024 .get = shmem_xattr_security_get,
2025 .set = shmem_xattr_security_set,
2026};
2027
2028static struct xattr_handler *shmem_xattr_handlers[] = {
2029 &shmem_xattr_acl_access_handler,
2030 &shmem_xattr_acl_default_handler,
2031 &shmem_xattr_security_handler,
2032 NULL
2033};
2034#endif
2035
David M. Grimes91828a42006-10-17 00:09:45 -07002036static struct dentry *shmem_get_parent(struct dentry *child)
2037{
2038 return ERR_PTR(-ESTALE);
2039}
2040
2041static int shmem_match(struct inode *ino, void *vfh)
2042{
2043 __u32 *fh = vfh;
2044 __u64 inum = fh[2];
2045 inum = (inum << 32) | fh[1];
2046 return ino->i_ino == inum && fh[0] == ino->i_generation;
2047}
2048
Christoph Hellwig480b1162007-10-21 16:42:13 -07002049static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2050 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002051{
David M. Grimes91828a42006-10-17 00:09:45 -07002052 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002053 struct dentry *dentry = NULL;
2054 u64 inum = fid->raw[2];
2055 inum = (inum << 32) | fid->raw[1];
David M. Grimes91828a42006-10-17 00:09:45 -07002056
Christoph Hellwig480b1162007-10-21 16:42:13 -07002057 if (fh_len < 3)
2058 return NULL;
2059
2060 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2061 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002062 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002063 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002064 iput(inode);
2065 }
2066
Christoph Hellwig480b1162007-10-21 16:42:13 -07002067 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002068}
2069
2070static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2071 int connectable)
2072{
2073 struct inode *inode = dentry->d_inode;
2074
2075 if (*len < 3)
2076 return 255;
2077
2078 if (hlist_unhashed(&inode->i_hash)) {
2079 /* Unfortunately insert_inode_hash is not idempotent,
2080 * so as we hash inodes here rather than at creation
2081 * time, we need a lock to ensure we only try
2082 * to do it once
2083 */
2084 static DEFINE_SPINLOCK(lock);
2085 spin_lock(&lock);
2086 if (hlist_unhashed(&inode->i_hash))
2087 __insert_inode_hash(inode,
2088 inode->i_ino + inode->i_generation);
2089 spin_unlock(&lock);
2090 }
2091
2092 fh[0] = inode->i_generation;
2093 fh[1] = inode->i_ino;
2094 fh[2] = ((__u64)inode->i_ino) >> 32;
2095
2096 *len = 3;
2097 return 1;
2098}
2099
Christoph Hellwig39655162007-10-21 16:42:17 -07002100static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002101 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002102 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002103 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002104};
2105
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002106static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2107 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108{
2109 char *this_char, *value, *rest;
2110
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002111 while (options != NULL) {
2112 this_char = options;
2113 for (;;) {
2114 /*
2115 * NUL-terminate this option: unfortunately,
2116 * mount options form a comma-separated list,
2117 * but mpol's nodelist may also contain commas.
2118 */
2119 options = strchr(options, ',');
2120 if (options == NULL)
2121 break;
2122 options++;
2123 if (!isdigit(*options)) {
2124 options[-1] = '\0';
2125 break;
2126 }
2127 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 if (!*this_char)
2129 continue;
2130 if ((value = strchr(this_char,'=')) != NULL) {
2131 *value++ = 0;
2132 } else {
2133 printk(KERN_ERR
2134 "tmpfs: No value for mount option '%s'\n",
2135 this_char);
2136 return 1;
2137 }
2138
2139 if (!strcmp(this_char,"size")) {
2140 unsigned long long size;
2141 size = memparse(value,&rest);
2142 if (*rest == '%') {
2143 size <<= PAGE_SHIFT;
2144 size *= totalram_pages;
2145 do_div(size, 100);
2146 rest++;
2147 }
2148 if (*rest)
2149 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002150 sbinfo->max_blocks =
2151 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002153 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 if (*rest)
2155 goto bad_val;
2156 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002157 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 if (*rest)
2159 goto bad_val;
2160 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002161 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002163 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 if (*rest)
2165 goto bad_val;
2166 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002167 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002169 sbinfo->uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 if (*rest)
2171 goto bad_val;
2172 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002173 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002175 sbinfo->gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 if (*rest)
2177 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002178 } else if (!strcmp(this_char,"mpol")) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002179 if (mpol_parse_str(value, &sbinfo->mpol, 1))
Robin Holt7339ff82006-01-14 13:20:48 -08002180 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 } else {
2182 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2183 this_char);
2184 return 1;
2185 }
2186 }
2187 return 0;
2188
2189bad_val:
2190 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2191 value, this_char);
2192 return 1;
2193
2194}
2195
2196static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2197{
2198 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002199 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002200 unsigned long blocks;
2201 unsigned long inodes;
2202 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002204 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002205 return error;
2206
2207 spin_lock(&sbinfo->stat_lock);
2208 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2209 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002210 if (config.max_blocks < blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002211 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002212 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002213 goto out;
2214 /*
2215 * Those tests also disallow limited->unlimited while any are in
2216 * use, so i_blocks will always be zero when max_blocks is zero;
2217 * but we must separately disallow unlimited->limited, because
2218 * in that case we have no record of how much is already in use.
2219 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002220 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002221 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002222 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002223 goto out;
2224
2225 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002226 sbinfo->max_blocks = config.max_blocks;
2227 sbinfo->free_blocks = config.max_blocks - blocks;
2228 sbinfo->max_inodes = config.max_inodes;
2229 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002230
2231 mpol_put(sbinfo->mpol);
2232 sbinfo->mpol = config.mpol; /* transfers initial ref */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002233out:
2234 spin_unlock(&sbinfo->stat_lock);
2235 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002237
2238static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2239{
2240 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2241
2242 if (sbinfo->max_blocks != shmem_default_max_blocks())
2243 seq_printf(seq, ",size=%luk",
2244 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2245 if (sbinfo->max_inodes != shmem_default_max_inodes())
2246 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2247 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2248 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2249 if (sbinfo->uid != 0)
2250 seq_printf(seq, ",uid=%u", sbinfo->uid);
2251 if (sbinfo->gid != 0)
2252 seq_printf(seq, ",gid=%u", sbinfo->gid);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002253 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002254 return 0;
2255}
2256#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257
2258static void shmem_put_super(struct super_block *sb)
2259{
2260 kfree(sb->s_fs_info);
2261 sb->s_fs_info = NULL;
2262}
2263
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264static int shmem_fill_super(struct super_block *sb,
2265 void *data, int silent)
2266{
2267 struct inode *inode;
2268 struct dentry *root;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002269 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002270 int err = -ENOMEM;
2271
2272 /* Round up to L1_CACHE_BYTES to resist false sharing */
2273 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2274 L1_CACHE_BYTES), GFP_KERNEL);
2275 if (!sbinfo)
2276 return -ENOMEM;
2277
2278 sbinfo->max_blocks = 0;
2279 sbinfo->max_inodes = 0;
2280 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002281 sbinfo->uid = current_fsuid();
2282 sbinfo->gid = current_fsgid();
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002283 sbinfo->mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002284 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002286#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 /*
2288 * Per default we only allow half of the physical ram per
2289 * tmpfs instance, limiting inodes to one per page of lowmem;
2290 * but the internal instance is left unlimited.
2291 */
2292 if (!(sb->s_flags & MS_NOUSER)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002293 sbinfo->max_blocks = shmem_default_max_blocks();
2294 sbinfo->max_inodes = shmem_default_max_inodes();
2295 if (shmem_parse_options(data, sbinfo, false)) {
2296 err = -EINVAL;
2297 goto failed;
2298 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 }
David M. Grimes91828a42006-10-17 00:09:45 -07002300 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301#else
2302 sb->s_flags |= MS_NOUSER;
2303#endif
2304
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002305 spin_lock_init(&sbinfo->stat_lock);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002306 sbinfo->free_blocks = sbinfo->max_blocks;
2307 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002308
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 sb->s_maxbytes = SHMEM_MAX_BYTES;
2310 sb->s_blocksize = PAGE_CACHE_SIZE;
2311 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2312 sb->s_magic = TMPFS_MAGIC;
2313 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002314 sb->s_time_gran = 1;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002315#ifdef CONFIG_TMPFS_POSIX_ACL
2316 sb->s_xattr = shmem_xattr_handlers;
2317 sb->s_flags |= MS_POSIXACL;
2318#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002319
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002320 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 if (!inode)
2322 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002323 inode->i_uid = sbinfo->uid;
2324 inode->i_gid = sbinfo->gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 root = d_alloc_root(inode);
2326 if (!root)
2327 goto failed_iput;
2328 sb->s_root = root;
2329 return 0;
2330
2331failed_iput:
2332 iput(inode);
2333failed:
2334 shmem_put_super(sb);
2335 return err;
2336}
2337
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002338static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
2340static struct inode *shmem_alloc_inode(struct super_block *sb)
2341{
2342 struct shmem_inode_info *p;
Christoph Lametere94b1762006-12-06 20:33:17 -08002343 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 if (!p)
2345 return NULL;
2346 return &p->vfs_inode;
2347}
2348
2349static void shmem_destroy_inode(struct inode *inode)
2350{
2351 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2352 /* only struct inode is valid if it's an inline symlink */
2353 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2354 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002355 shmem_acl_destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2357}
2358
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002359static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360{
2361 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2362
Christoph Lametera35afb82007-05-16 22:10:57 -07002363 inode_init_once(&p->vfs_inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002364#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Lametera35afb82007-05-16 22:10:57 -07002365 p->i_acl = NULL;
2366 p->i_default_acl = NULL;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002367#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368}
2369
2370static int init_inodecache(void)
2371{
2372 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2373 sizeof(struct shmem_inode_info),
Alexey Dobriyan040b5c62007-10-16 23:26:10 -07002374 0, SLAB_PANIC, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 return 0;
2376}
2377
2378static void destroy_inodecache(void)
2379{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002380 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381}
2382
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002383static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002385 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386#ifdef CONFIG_TMPFS
Hugh Dickinsae976412007-06-04 10:00:39 +02002387 .readpage = shmem_readpage,
Nick Piggin800d15a2007-10-16 01:25:03 -07002388 .write_begin = shmem_write_begin,
2389 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002391 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392};
2393
Helge Deller15ad7cd2006-12-06 20:40:36 -08002394static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 .mmap = shmem_mmap,
2396#ifdef CONFIG_TMPFS
2397 .llseek = generic_file_llseek,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002398 .read = do_sync_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002399 .write = do_sync_write,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002400 .aio_read = shmem_file_aio_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002401 .aio_write = generic_file_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 .fsync = simple_sync_file,
Hugh Dickinsae976412007-06-04 10:00:39 +02002403 .splice_read = generic_file_splice_read,
2404 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405#endif
2406};
2407
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002408static const struct inode_operations shmem_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 .truncate = shmem_truncate,
2410 .setattr = shmem_notify_change,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002411 .truncate_range = shmem_truncate_range,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002412#ifdef CONFIG_TMPFS_POSIX_ACL
2413 .setxattr = generic_setxattr,
2414 .getxattr = generic_getxattr,
2415 .listxattr = generic_listxattr,
2416 .removexattr = generic_removexattr,
2417 .permission = shmem_permission,
2418#endif
2419
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420};
2421
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002422static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423#ifdef CONFIG_TMPFS
2424 .create = shmem_create,
2425 .lookup = simple_lookup,
2426 .link = shmem_link,
2427 .unlink = shmem_unlink,
2428 .symlink = shmem_symlink,
2429 .mkdir = shmem_mkdir,
2430 .rmdir = shmem_rmdir,
2431 .mknod = shmem_mknod,
2432 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002434#ifdef CONFIG_TMPFS_POSIX_ACL
2435 .setattr = shmem_notify_change,
2436 .setxattr = generic_setxattr,
2437 .getxattr = generic_getxattr,
2438 .listxattr = generic_listxattr,
2439 .removexattr = generic_removexattr,
2440 .permission = shmem_permission,
2441#endif
2442};
2443
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002444static const struct inode_operations shmem_special_inode_operations = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002445#ifdef CONFIG_TMPFS_POSIX_ACL
2446 .setattr = shmem_notify_change,
2447 .setxattr = generic_setxattr,
2448 .getxattr = generic_getxattr,
2449 .listxattr = generic_listxattr,
2450 .removexattr = generic_removexattr,
2451 .permission = shmem_permission,
2452#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453};
2454
Hugh Dickins759b9772007-03-05 00:30:28 -08002455static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 .alloc_inode = shmem_alloc_inode,
2457 .destroy_inode = shmem_destroy_inode,
2458#ifdef CONFIG_TMPFS
2459 .statfs = shmem_statfs,
2460 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002461 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462#endif
2463 .delete_inode = shmem_delete_inode,
2464 .drop_inode = generic_delete_inode,
2465 .put_super = shmem_put_super,
2466};
2467
2468static struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002469 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470#ifdef CONFIG_NUMA
2471 .set_policy = shmem_set_policy,
2472 .get_policy = shmem_get_policy,
2473#endif
2474};
2475
2476
David Howells454e2392006-06-23 02:02:57 -07002477static int shmem_get_sb(struct file_system_type *fs_type,
2478 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479{
David Howells454e2392006-06-23 02:02:57 -07002480 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481}
2482
2483static struct file_system_type tmpfs_fs_type = {
2484 .owner = THIS_MODULE,
2485 .name = "tmpfs",
2486 .get_sb = shmem_get_sb,
2487 .kill_sb = kill_litter_super,
2488};
2489static struct vfsmount *shm_mnt;
2490
2491static int __init init_tmpfs(void)
2492{
2493 int error;
2494
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002495 error = bdi_init(&shmem_backing_dev_info);
2496 if (error)
2497 goto out4;
2498
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 error = init_inodecache();
2500 if (error)
2501 goto out3;
2502
2503 error = register_filesystem(&tmpfs_fs_type);
2504 if (error) {
2505 printk(KERN_ERR "Could not register tmpfs\n");
2506 goto out2;
2507 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002508
Trond Myklebust1f5ce9e2006-06-09 09:34:16 -04002509 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 tmpfs_fs_type.name, NULL);
2511 if (IS_ERR(shm_mnt)) {
2512 error = PTR_ERR(shm_mnt);
2513 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2514 goto out1;
2515 }
2516 return 0;
2517
2518out1:
2519 unregister_filesystem(&tmpfs_fs_type);
2520out2:
2521 destroy_inodecache();
2522out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002523 bdi_destroy(&shmem_backing_dev_info);
2524out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 shm_mnt = ERR_PTR(error);
2526 return error;
2527}
2528module_init(init_tmpfs)
2529
Randy Dunlap46711812008-03-19 17:00:41 -07002530/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 * shmem_file_setup - get an unlinked file living in tmpfs
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 * @name: name for dentry (to be seen in /proc/<pid>/maps
2533 * @size: size to be set for the file
Randy Dunlap46711812008-03-19 17:00:41 -07002534 * @flags: vm_flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 */
2536struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2537{
2538 int error;
2539 struct file *file;
2540 struct inode *inode;
2541 struct dentry *dentry, *root;
2542 struct qstr this;
2543
2544 if (IS_ERR(shm_mnt))
2545 return (void *)shm_mnt;
2546
2547 if (size < 0 || size > SHMEM_MAX_BYTES)
2548 return ERR_PTR(-EINVAL);
2549
2550 if (shmem_acct_size(flags, size))
2551 return ERR_PTR(-ENOMEM);
2552
2553 error = -ENOMEM;
2554 this.name = name;
2555 this.len = strlen(name);
2556 this.hash = 0; /* will go */
2557 root = shm_mnt->mnt_root;
2558 dentry = d_alloc(root, &this);
2559 if (!dentry)
2560 goto put_memory;
2561
2562 error = -ENFILE;
2563 file = get_empty_filp();
2564 if (!file)
2565 goto put_dentry;
2566
2567 error = -ENOSPC;
2568 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2569 if (!inode)
2570 goto close_file;
2571
2572 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2573 d_instantiate(dentry, inode);
2574 inode->i_size = size;
2575 inode->i_nlink = 0; /* It is unlinked */
Dave Hansence8d2cd2007-10-16 23:31:13 -07002576 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
2577 &shmem_file_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 return file;
2579
2580close_file:
2581 put_filp(file);
2582put_dentry:
2583 dput(dentry);
2584put_memory:
2585 shmem_unacct_size(flags, size);
2586 return ERR_PTR(error);
2587}
Keith Packard395e0dd2008-06-20 00:08:06 -07002588EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589
Randy Dunlap46711812008-03-19 17:00:41 -07002590/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2593 */
2594int shmem_zero_setup(struct vm_area_struct *vma)
2595{
2596 struct file *file;
2597 loff_t size = vma->vm_end - vma->vm_start;
2598
2599 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2600 if (IS_ERR(file))
2601 return PTR_ERR(file);
2602
2603 if (vma->vm_file)
2604 fput(vma->vm_file);
2605 vma->vm_file = file;
2606 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 return 0;
2608}