]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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. | |
0edd73b3 HD |
9 | * Copyright (C) 2002-2005 Hugh Dickins. |
10 | * Copyright (C) 2002-2005 VERITAS Software Corporation. | |
1da177e4 LT |
11 | * 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 | ||
1da177e4 LT |
26 | #include <linux/module.h> |
27 | #include <linux/init.h> | |
1da177e4 | 28 | #include <linux/fs.h> |
39f0247d | 29 | #include <linux/xattr.h> |
a5694255 | 30 | #include <linux/exportfs.h> |
39f0247d | 31 | #include <linux/generic_acl.h> |
1da177e4 LT |
32 | #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> | |
b00dc3ad | 49 | #include <linux/ctype.h> |
304dbdb7 | 50 | #include <linux/migrate.h> |
c1f60a5a | 51 | #include <linux/highmem.h> |
680d794b | 52 | #include <linux/seq_file.h> |
304dbdb7 | 53 | |
1da177e4 LT |
54 | #include <asm/uaccess.h> |
55 | #include <asm/div64.h> | |
56 | #include <asm/pgtable.h> | |
57 | ||
58 | /* This magic number is used in glibc for posix shared memory */ | |
59 | #define TMPFS_MAGIC 0x01021994 | |
60 | ||
61 | #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long)) | |
62 | #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE) | |
63 | #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) | |
64 | ||
65 | #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1)) | |
66 | #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT) | |
67 | ||
68 | #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) | |
69 | ||
70 | /* info->flags needs VM_flags to handle pagein/truncate races efficiently */ | |
71 | #define SHMEM_PAGEIN VM_READ | |
72 | #define SHMEM_TRUNCATE VM_WRITE | |
73 | ||
74 | /* Definition to limit shmem_truncate's steps between cond_rescheds */ | |
75 | #define LATENCY_LIMIT 64 | |
76 | ||
77 | /* Pretend that each entry is of this size in directory's i_size */ | |
78 | #define BOGO_DIRENT_SIZE 20 | |
79 | ||
1da177e4 LT |
80 | /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */ |
81 | enum sgp_type { | |
1da177e4 LT |
82 | SGP_READ, /* don't exceed i_size, don't allocate page */ |
83 | SGP_CACHE, /* don't exceed i_size, may allocate page */ | |
a0ee5ec5 | 84 | SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ |
1da177e4 LT |
85 | SGP_WRITE, /* may exceed i_size, may allocate page */ |
86 | }; | |
87 | ||
b76db735 | 88 | #ifdef CONFIG_TMPFS |
680d794b | 89 | static unsigned long shmem_default_max_blocks(void) |
90 | { | |
91 | return totalram_pages / 2; | |
92 | } | |
93 | ||
94 | static unsigned long shmem_default_max_inodes(void) | |
95 | { | |
96 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
97 | } | |
b76db735 | 98 | #endif |
680d794b | 99 | |
1da177e4 LT |
100 | static int shmem_getpage(struct inode *inode, unsigned long idx, |
101 | struct page **pagep, enum sgp_type sgp, int *type); | |
102 | ||
6daa0e28 | 103 | static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) |
1da177e4 LT |
104 | { |
105 | /* | |
106 | * The above definition of ENTRIES_PER_PAGE, and the use of | |
107 | * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: | |
108 | * might be reconsidered if it ever diverges from PAGE_SIZE. | |
769848c0 | 109 | * |
e12ba74d | 110 | * Mobility flags are masked out as swap vectors cannot move |
1da177e4 | 111 | */ |
e12ba74d | 112 | return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO, |
769848c0 | 113 | PAGE_CACHE_SHIFT-PAGE_SHIFT); |
1da177e4 LT |
114 | } |
115 | ||
116 | static inline void shmem_dir_free(struct page *page) | |
117 | { | |
118 | __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); | |
119 | } | |
120 | ||
121 | static struct page **shmem_dir_map(struct page *page) | |
122 | { | |
123 | return (struct page **)kmap_atomic(page, KM_USER0); | |
124 | } | |
125 | ||
126 | static inline void shmem_dir_unmap(struct page **dir) | |
127 | { | |
128 | kunmap_atomic(dir, KM_USER0); | |
129 | } | |
130 | ||
131 | static swp_entry_t *shmem_swp_map(struct page *page) | |
132 | { | |
133 | return (swp_entry_t *)kmap_atomic(page, KM_USER1); | |
134 | } | |
135 | ||
136 | static inline void shmem_swp_balance_unmap(void) | |
137 | { | |
138 | /* | |
139 | * When passing a pointer to an i_direct entry, to code which | |
140 | * also handles indirect entries and so will shmem_swp_unmap, | |
141 | * we must arrange for the preempt count to remain in balance. | |
142 | * What kmap_atomic of a lowmem page does depends on config | |
143 | * and architecture, so pretend to kmap_atomic some lowmem page. | |
144 | */ | |
145 | (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); | |
146 | } | |
147 | ||
148 | static inline void shmem_swp_unmap(swp_entry_t *entry) | |
149 | { | |
150 | kunmap_atomic(entry, KM_USER1); | |
151 | } | |
152 | ||
153 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) | |
154 | { | |
155 | return sb->s_fs_info; | |
156 | } | |
157 | ||
158 | /* | |
159 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
160 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
161 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
162 | * consistent with the pre-accounting of private mappings ... | |
163 | */ | |
164 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
165 | { | |
166 | return (flags & VM_ACCOUNT)? | |
167 | security_vm_enough_memory(VM_ACCT(size)): 0; | |
168 | } | |
169 | ||
170 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
171 | { | |
172 | if (flags & VM_ACCOUNT) | |
173 | vm_unacct_memory(VM_ACCT(size)); | |
174 | } | |
175 | ||
176 | /* | |
177 | * ... whereas tmpfs objects are accounted incrementally as | |
178 | * pages are allocated, in order to allow huge sparse files. | |
179 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, | |
180 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
181 | */ | |
182 | static inline int shmem_acct_block(unsigned long flags) | |
183 | { | |
184 | return (flags & VM_ACCOUNT)? | |
185 | 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE)); | |
186 | } | |
187 | ||
188 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
189 | { | |
190 | if (!(flags & VM_ACCOUNT)) | |
191 | vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); | |
192 | } | |
193 | ||
759b9775 | 194 | static const struct super_operations shmem_ops; |
f5e54d6e | 195 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 196 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
197 | static const struct inode_operations shmem_inode_operations; |
198 | static const struct inode_operations shmem_dir_inode_operations; | |
199 | static const struct inode_operations shmem_special_inode_operations; | |
1da177e4 LT |
200 | static struct vm_operations_struct shmem_vm_ops; |
201 | ||
6c231b7b | 202 | static struct backing_dev_info shmem_backing_dev_info __read_mostly = { |
1da177e4 LT |
203 | .ra_pages = 0, /* No readahead */ |
204 | .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, | |
205 | .unplug_io_fn = default_unplug_io_fn, | |
206 | }; | |
207 | ||
208 | static LIST_HEAD(shmem_swaplist); | |
cb5f7b9a | 209 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 LT |
210 | |
211 | static void shmem_free_blocks(struct inode *inode, long pages) | |
212 | { | |
213 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
0edd73b3 | 214 | if (sbinfo->max_blocks) { |
1da177e4 LT |
215 | spin_lock(&sbinfo->stat_lock); |
216 | sbinfo->free_blocks += pages; | |
217 | inode->i_blocks -= pages*BLOCKS_PER_PAGE; | |
218 | spin_unlock(&sbinfo->stat_lock); | |
219 | } | |
220 | } | |
221 | ||
5b04c689 PE |
222 | static int shmem_reserve_inode(struct super_block *sb) |
223 | { | |
224 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
225 | if (sbinfo->max_inodes) { | |
226 | spin_lock(&sbinfo->stat_lock); | |
227 | if (!sbinfo->free_inodes) { | |
228 | spin_unlock(&sbinfo->stat_lock); | |
229 | return -ENOSPC; | |
230 | } | |
231 | sbinfo->free_inodes--; | |
232 | spin_unlock(&sbinfo->stat_lock); | |
233 | } | |
234 | return 0; | |
235 | } | |
236 | ||
237 | static void shmem_free_inode(struct super_block *sb) | |
238 | { | |
239 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
240 | if (sbinfo->max_inodes) { | |
241 | spin_lock(&sbinfo->stat_lock); | |
242 | sbinfo->free_inodes++; | |
243 | spin_unlock(&sbinfo->stat_lock); | |
244 | } | |
245 | } | |
246 | ||
46711810 | 247 | /** |
1da177e4 | 248 | * shmem_recalc_inode - recalculate the size of an inode |
1da177e4 LT |
249 | * @inode: inode to recalc |
250 | * | |
251 | * We have to calculate the free blocks since the mm can drop | |
252 | * undirtied hole pages behind our back. | |
253 | * | |
254 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
255 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
256 | * | |
257 | * It has to be called with the spinlock held. | |
258 | */ | |
259 | static void shmem_recalc_inode(struct inode *inode) | |
260 | { | |
261 | struct shmem_inode_info *info = SHMEM_I(inode); | |
262 | long freed; | |
263 | ||
264 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
265 | if (freed > 0) { | |
266 | info->alloced -= freed; | |
267 | shmem_unacct_blocks(info->flags, freed); | |
268 | shmem_free_blocks(inode, freed); | |
269 | } | |
270 | } | |
271 | ||
46711810 | 272 | /** |
1da177e4 | 273 | * shmem_swp_entry - find the swap vector position in the info structure |
1da177e4 LT |
274 | * @info: info structure for the inode |
275 | * @index: index of the page to find | |
276 | * @page: optional page to add to the structure. Has to be preset to | |
277 | * all zeros | |
278 | * | |
279 | * If there is no space allocated yet it will return NULL when | |
280 | * page is NULL, else it will use the page for the needed block, | |
281 | * setting it to NULL on return to indicate that it has been used. | |
282 | * | |
283 | * The swap vector is organized the following way: | |
284 | * | |
285 | * There are SHMEM_NR_DIRECT entries directly stored in the | |
286 | * shmem_inode_info structure. So small files do not need an addional | |
287 | * allocation. | |
288 | * | |
289 | * For pages with index > SHMEM_NR_DIRECT there is the pointer | |
290 | * i_indirect which points to a page which holds in the first half | |
291 | * doubly indirect blocks, in the second half triple indirect blocks: | |
292 | * | |
293 | * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the | |
294 | * following layout (for SHMEM_NR_DIRECT == 16): | |
295 | * | |
296 | * i_indirect -> dir --> 16-19 | |
297 | * | +-> 20-23 | |
298 | * | | |
299 | * +-->dir2 --> 24-27 | |
300 | * | +-> 28-31 | |
301 | * | +-> 32-35 | |
302 | * | +-> 36-39 | |
303 | * | | |
304 | * +-->dir3 --> 40-43 | |
305 | * +-> 44-47 | |
306 | * +-> 48-51 | |
307 | * +-> 52-55 | |
308 | */ | |
309 | static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) | |
310 | { | |
311 | unsigned long offset; | |
312 | struct page **dir; | |
313 | struct page *subdir; | |
314 | ||
315 | if (index < SHMEM_NR_DIRECT) { | |
316 | shmem_swp_balance_unmap(); | |
317 | return info->i_direct+index; | |
318 | } | |
319 | if (!info->i_indirect) { | |
320 | if (page) { | |
321 | info->i_indirect = *page; | |
322 | *page = NULL; | |
323 | } | |
324 | return NULL; /* need another page */ | |
325 | } | |
326 | ||
327 | index -= SHMEM_NR_DIRECT; | |
328 | offset = index % ENTRIES_PER_PAGE; | |
329 | index /= ENTRIES_PER_PAGE; | |
330 | dir = shmem_dir_map(info->i_indirect); | |
331 | ||
332 | if (index >= ENTRIES_PER_PAGE/2) { | |
333 | index -= ENTRIES_PER_PAGE/2; | |
334 | dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; | |
335 | index %= ENTRIES_PER_PAGE; | |
336 | subdir = *dir; | |
337 | if (!subdir) { | |
338 | if (page) { | |
339 | *dir = *page; | |
340 | *page = NULL; | |
341 | } | |
342 | shmem_dir_unmap(dir); | |
343 | return NULL; /* need another page */ | |
344 | } | |
345 | shmem_dir_unmap(dir); | |
346 | dir = shmem_dir_map(subdir); | |
347 | } | |
348 | ||
349 | dir += index; | |
350 | subdir = *dir; | |
351 | if (!subdir) { | |
352 | if (!page || !(subdir = *page)) { | |
353 | shmem_dir_unmap(dir); | |
354 | return NULL; /* need a page */ | |
355 | } | |
356 | *dir = subdir; | |
357 | *page = NULL; | |
358 | } | |
359 | shmem_dir_unmap(dir); | |
360 | return shmem_swp_map(subdir) + offset; | |
361 | } | |
362 | ||
363 | static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) | |
364 | { | |
365 | long incdec = value? 1: -1; | |
366 | ||
367 | entry->val = value; | |
368 | info->swapped += incdec; | |
4c21e2f2 HD |
369 | if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { |
370 | struct page *page = kmap_atomic_to_page(entry); | |
371 | set_page_private(page, page_private(page) + incdec); | |
372 | } | |
1da177e4 LT |
373 | } |
374 | ||
46711810 | 375 | /** |
1da177e4 | 376 | * shmem_swp_alloc - get the position of the swap entry for the page. |
1da177e4 LT |
377 | * @info: info structure for the inode |
378 | * @index: index of the page to find | |
379 | * @sgp: check and recheck i_size? skip allocation? | |
46711810 RD |
380 | * |
381 | * If the entry does not exist, allocate it. | |
1da177e4 LT |
382 | */ |
383 | static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) | |
384 | { | |
385 | struct inode *inode = &info->vfs_inode; | |
386 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
387 | struct page *page = NULL; | |
388 | swp_entry_t *entry; | |
389 | ||
390 | if (sgp != SGP_WRITE && | |
391 | ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) | |
392 | return ERR_PTR(-EINVAL); | |
393 | ||
394 | while (!(entry = shmem_swp_entry(info, index, &page))) { | |
395 | if (sgp == SGP_READ) | |
396 | return shmem_swp_map(ZERO_PAGE(0)); | |
397 | /* | |
398 | * Test free_blocks against 1 not 0, since we have 1 data | |
399 | * page (and perhaps indirect index pages) yet to allocate: | |
400 | * a waste to allocate index if we cannot allocate data. | |
401 | */ | |
0edd73b3 | 402 | if (sbinfo->max_blocks) { |
1da177e4 LT |
403 | spin_lock(&sbinfo->stat_lock); |
404 | if (sbinfo->free_blocks <= 1) { | |
405 | spin_unlock(&sbinfo->stat_lock); | |
406 | return ERR_PTR(-ENOSPC); | |
407 | } | |
408 | sbinfo->free_blocks--; | |
409 | inode->i_blocks += BLOCKS_PER_PAGE; | |
410 | spin_unlock(&sbinfo->stat_lock); | |
411 | } | |
412 | ||
413 | spin_unlock(&info->lock); | |
769848c0 | 414 | page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); |
4c21e2f2 HD |
415 | if (page) |
416 | set_page_private(page, 0); | |
1da177e4 LT |
417 | spin_lock(&info->lock); |
418 | ||
419 | if (!page) { | |
420 | shmem_free_blocks(inode, 1); | |
421 | return ERR_PTR(-ENOMEM); | |
422 | } | |
423 | if (sgp != SGP_WRITE && | |
424 | ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { | |
425 | entry = ERR_PTR(-EINVAL); | |
426 | break; | |
427 | } | |
428 | if (info->next_index <= index) | |
429 | info->next_index = index + 1; | |
430 | } | |
431 | if (page) { | |
432 | /* another task gave its page, or truncated the file */ | |
433 | shmem_free_blocks(inode, 1); | |
434 | shmem_dir_free(page); | |
435 | } | |
436 | if (info->next_index <= index && !IS_ERR(entry)) | |
437 | info->next_index = index + 1; | |
438 | return entry; | |
439 | } | |
440 | ||
46711810 | 441 | /** |
1da177e4 | 442 | * shmem_free_swp - free some swap entries in a directory |
1ae70006 HD |
443 | * @dir: pointer to the directory |
444 | * @edir: pointer after last entry of the directory | |
445 | * @punch_lock: pointer to spinlock when needed for the holepunch case | |
1da177e4 | 446 | */ |
1ae70006 HD |
447 | static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, |
448 | spinlock_t *punch_lock) | |
1da177e4 | 449 | { |
1ae70006 | 450 | spinlock_t *punch_unlock = NULL; |
1da177e4 LT |
451 | swp_entry_t *ptr; |
452 | int freed = 0; | |
453 | ||
454 | for (ptr = dir; ptr < edir; ptr++) { | |
455 | if (ptr->val) { | |
1ae70006 HD |
456 | if (unlikely(punch_lock)) { |
457 | punch_unlock = punch_lock; | |
458 | punch_lock = NULL; | |
459 | spin_lock(punch_unlock); | |
460 | if (!ptr->val) | |
461 | continue; | |
462 | } | |
1da177e4 LT |
463 | free_swap_and_cache(*ptr); |
464 | *ptr = (swp_entry_t){0}; | |
465 | freed++; | |
466 | } | |
467 | } | |
1ae70006 HD |
468 | if (punch_unlock) |
469 | spin_unlock(punch_unlock); | |
1da177e4 LT |
470 | return freed; |
471 | } | |
472 | ||
1ae70006 HD |
473 | static int shmem_map_and_free_swp(struct page *subdir, int offset, |
474 | int limit, struct page ***dir, spinlock_t *punch_lock) | |
1da177e4 LT |
475 | { |
476 | swp_entry_t *ptr; | |
477 | int freed = 0; | |
478 | ||
479 | ptr = shmem_swp_map(subdir); | |
480 | for (; offset < limit; offset += LATENCY_LIMIT) { | |
481 | int size = limit - offset; | |
482 | if (size > LATENCY_LIMIT) | |
483 | size = LATENCY_LIMIT; | |
1ae70006 HD |
484 | freed += shmem_free_swp(ptr+offset, ptr+offset+size, |
485 | punch_lock); | |
1da177e4 LT |
486 | if (need_resched()) { |
487 | shmem_swp_unmap(ptr); | |
488 | if (*dir) { | |
489 | shmem_dir_unmap(*dir); | |
490 | *dir = NULL; | |
491 | } | |
492 | cond_resched(); | |
493 | ptr = shmem_swp_map(subdir); | |
494 | } | |
495 | } | |
496 | shmem_swp_unmap(ptr); | |
497 | return freed; | |
498 | } | |
499 | ||
500 | static void shmem_free_pages(struct list_head *next) | |
501 | { | |
502 | struct page *page; | |
503 | int freed = 0; | |
504 | ||
505 | do { | |
506 | page = container_of(next, struct page, lru); | |
507 | next = next->next; | |
508 | shmem_dir_free(page); | |
509 | freed++; | |
510 | if (freed >= LATENCY_LIMIT) { | |
511 | cond_resched(); | |
512 | freed = 0; | |
513 | } | |
514 | } while (next); | |
515 | } | |
516 | ||
f6b3ec23 | 517 | static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) |
1da177e4 LT |
518 | { |
519 | struct shmem_inode_info *info = SHMEM_I(inode); | |
520 | unsigned long idx; | |
521 | unsigned long size; | |
522 | unsigned long limit; | |
523 | unsigned long stage; | |
524 | unsigned long diroff; | |
525 | struct page **dir; | |
526 | struct page *topdir; | |
527 | struct page *middir; | |
528 | struct page *subdir; | |
529 | swp_entry_t *ptr; | |
530 | LIST_HEAD(pages_to_free); | |
531 | long nr_pages_to_free = 0; | |
532 | long nr_swaps_freed = 0; | |
533 | int offset; | |
534 | int freed; | |
a2646d1e | 535 | int punch_hole; |
1ae70006 HD |
536 | spinlock_t *needs_lock; |
537 | spinlock_t *punch_lock; | |
a2646d1e | 538 | unsigned long upper_limit; |
1da177e4 LT |
539 | |
540 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
f6b3ec23 | 541 | idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
1da177e4 LT |
542 | if (idx >= info->next_index) |
543 | return; | |
544 | ||
545 | spin_lock(&info->lock); | |
546 | info->flags |= SHMEM_TRUNCATE; | |
f6b3ec23 BP |
547 | if (likely(end == (loff_t) -1)) { |
548 | limit = info->next_index; | |
a2646d1e | 549 | upper_limit = SHMEM_MAX_INDEX; |
f6b3ec23 | 550 | info->next_index = idx; |
1ae70006 | 551 | needs_lock = NULL; |
a2646d1e | 552 | punch_hole = 0; |
f6b3ec23 | 553 | } else { |
a2646d1e HD |
554 | if (end + 1 >= inode->i_size) { /* we may free a little more */ |
555 | limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> | |
556 | PAGE_CACHE_SHIFT; | |
557 | upper_limit = SHMEM_MAX_INDEX; | |
558 | } else { | |
559 | limit = (end + 1) >> PAGE_CACHE_SHIFT; | |
560 | upper_limit = limit; | |
561 | } | |
1ae70006 | 562 | needs_lock = &info->lock; |
f6b3ec23 BP |
563 | punch_hole = 1; |
564 | } | |
565 | ||
1da177e4 | 566 | topdir = info->i_indirect; |
f6b3ec23 | 567 | if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { |
1da177e4 LT |
568 | info->i_indirect = NULL; |
569 | nr_pages_to_free++; | |
570 | list_add(&topdir->lru, &pages_to_free); | |
571 | } | |
572 | spin_unlock(&info->lock); | |
573 | ||
574 | if (info->swapped && idx < SHMEM_NR_DIRECT) { | |
575 | ptr = info->i_direct; | |
576 | size = limit; | |
577 | if (size > SHMEM_NR_DIRECT) | |
578 | size = SHMEM_NR_DIRECT; | |
1ae70006 | 579 | nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); |
1da177e4 | 580 | } |
92a3d03a BP |
581 | |
582 | /* | |
583 | * If there are no indirect blocks or we are punching a hole | |
584 | * below indirect blocks, nothing to be done. | |
585 | */ | |
a2646d1e | 586 | if (!topdir || limit <= SHMEM_NR_DIRECT) |
1da177e4 LT |
587 | goto done2; |
588 | ||
1ae70006 HD |
589 | /* |
590 | * The truncation case has already dropped info->lock, and we're safe | |
591 | * because i_size and next_index have already been lowered, preventing | |
592 | * access beyond. But in the punch_hole case, we still need to take | |
593 | * the lock when updating the swap directory, because there might be | |
594 | * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or | |
595 | * shmem_writepage. However, whenever we find we can remove a whole | |
596 | * directory page (not at the misaligned start or end of the range), | |
597 | * we first NULLify its pointer in the level above, and then have no | |
598 | * need to take the lock when updating its contents: needs_lock and | |
599 | * punch_lock (either pointing to info->lock or NULL) manage this. | |
600 | */ | |
601 | ||
a2646d1e | 602 | upper_limit -= SHMEM_NR_DIRECT; |
1da177e4 LT |
603 | limit -= SHMEM_NR_DIRECT; |
604 | idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; | |
605 | offset = idx % ENTRIES_PER_PAGE; | |
606 | idx -= offset; | |
607 | ||
608 | dir = shmem_dir_map(topdir); | |
609 | stage = ENTRIES_PER_PAGEPAGE/2; | |
610 | if (idx < ENTRIES_PER_PAGEPAGE/2) { | |
611 | middir = topdir; | |
612 | diroff = idx/ENTRIES_PER_PAGE; | |
613 | } else { | |
614 | dir += ENTRIES_PER_PAGE/2; | |
615 | dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; | |
616 | while (stage <= idx) | |
617 | stage += ENTRIES_PER_PAGEPAGE; | |
618 | middir = *dir; | |
619 | if (*dir) { | |
620 | diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % | |
621 | ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; | |
a2646d1e | 622 | if (!diroff && !offset && upper_limit >= stage) { |
1ae70006 HD |
623 | if (needs_lock) { |
624 | spin_lock(needs_lock); | |
625 | *dir = NULL; | |
626 | spin_unlock(needs_lock); | |
627 | needs_lock = NULL; | |
628 | } else | |
629 | *dir = NULL; | |
1da177e4 LT |
630 | nr_pages_to_free++; |
631 | list_add(&middir->lru, &pages_to_free); | |
632 | } | |
633 | shmem_dir_unmap(dir); | |
634 | dir = shmem_dir_map(middir); | |
635 | } else { | |
636 | diroff = 0; | |
637 | offset = 0; | |
638 | idx = stage; | |
639 | } | |
640 | } | |
641 | ||
642 | for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { | |
643 | if (unlikely(idx == stage)) { | |
644 | shmem_dir_unmap(dir); | |
645 | dir = shmem_dir_map(topdir) + | |
646 | ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; | |
647 | while (!*dir) { | |
648 | dir++; | |
649 | idx += ENTRIES_PER_PAGEPAGE; | |
650 | if (idx >= limit) | |
651 | goto done1; | |
652 | } | |
653 | stage = idx + ENTRIES_PER_PAGEPAGE; | |
654 | middir = *dir; | |
1ae70006 HD |
655 | if (punch_hole) |
656 | needs_lock = &info->lock; | |
a2646d1e | 657 | if (upper_limit >= stage) { |
1ae70006 HD |
658 | if (needs_lock) { |
659 | spin_lock(needs_lock); | |
660 | *dir = NULL; | |
661 | spin_unlock(needs_lock); | |
662 | needs_lock = NULL; | |
663 | } else | |
664 | *dir = NULL; | |
a2646d1e HD |
665 | nr_pages_to_free++; |
666 | list_add(&middir->lru, &pages_to_free); | |
667 | } | |
1da177e4 LT |
668 | shmem_dir_unmap(dir); |
669 | cond_resched(); | |
670 | dir = shmem_dir_map(middir); | |
671 | diroff = 0; | |
672 | } | |
1ae70006 | 673 | punch_lock = needs_lock; |
1da177e4 | 674 | subdir = dir[diroff]; |
1ae70006 HD |
675 | if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { |
676 | if (needs_lock) { | |
677 | spin_lock(needs_lock); | |
678 | dir[diroff] = NULL; | |
679 | spin_unlock(needs_lock); | |
680 | punch_lock = NULL; | |
681 | } else | |
682 | dir[diroff] = NULL; | |
683 | nr_pages_to_free++; | |
684 | list_add(&subdir->lru, &pages_to_free); | |
685 | } | |
686 | if (subdir && page_private(subdir) /* has swap entries */) { | |
1da177e4 LT |
687 | size = limit - idx; |
688 | if (size > ENTRIES_PER_PAGE) | |
689 | size = ENTRIES_PER_PAGE; | |
690 | freed = shmem_map_and_free_swp(subdir, | |
1ae70006 | 691 | offset, size, &dir, punch_lock); |
1da177e4 LT |
692 | if (!dir) |
693 | dir = shmem_dir_map(middir); | |
694 | nr_swaps_freed += freed; | |
1ae70006 | 695 | if (offset || punch_lock) { |
1da177e4 | 696 | spin_lock(&info->lock); |
1ae70006 HD |
697 | set_page_private(subdir, |
698 | page_private(subdir) - freed); | |
1da177e4 | 699 | spin_unlock(&info->lock); |
1ae70006 HD |
700 | } else |
701 | BUG_ON(page_private(subdir) != freed); | |
1da177e4 | 702 | } |
1ae70006 | 703 | offset = 0; |
1da177e4 LT |
704 | } |
705 | done1: | |
706 | shmem_dir_unmap(dir); | |
707 | done2: | |
708 | if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { | |
709 | /* | |
710 | * Call truncate_inode_pages again: racing shmem_unuse_inode | |
711 | * may have swizzled a page in from swap since vmtruncate or | |
712 | * generic_delete_inode did it, before we lowered next_index. | |
713 | * Also, though shmem_getpage checks i_size before adding to | |
714 | * cache, no recheck after: so fix the narrow window there too. | |
16a10019 HD |
715 | * |
716 | * Recalling truncate_inode_pages_range and unmap_mapping_range | |
717 | * every time for punch_hole (which never got a chance to clear | |
718 | * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, | |
719 | * yet hardly ever necessary: try to optimize them out later. | |
1da177e4 | 720 | */ |
f6b3ec23 | 721 | truncate_inode_pages_range(inode->i_mapping, start, end); |
16a10019 HD |
722 | if (punch_hole) |
723 | unmap_mapping_range(inode->i_mapping, start, | |
724 | end - start, 1); | |
1da177e4 LT |
725 | } |
726 | ||
727 | spin_lock(&info->lock); | |
728 | info->flags &= ~SHMEM_TRUNCATE; | |
729 | info->swapped -= nr_swaps_freed; | |
730 | if (nr_pages_to_free) | |
731 | shmem_free_blocks(inode, nr_pages_to_free); | |
732 | shmem_recalc_inode(inode); | |
733 | spin_unlock(&info->lock); | |
734 | ||
735 | /* | |
736 | * Empty swap vector directory pages to be freed? | |
737 | */ | |
738 | if (!list_empty(&pages_to_free)) { | |
739 | pages_to_free.prev->next = NULL; | |
740 | shmem_free_pages(pages_to_free.next); | |
741 | } | |
742 | } | |
743 | ||
f6b3ec23 BP |
744 | static void shmem_truncate(struct inode *inode) |
745 | { | |
746 | shmem_truncate_range(inode, inode->i_size, (loff_t)-1); | |
747 | } | |
748 | ||
1da177e4 LT |
749 | static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) |
750 | { | |
751 | struct inode *inode = dentry->d_inode; | |
752 | struct page *page = NULL; | |
753 | int error; | |
754 | ||
39f0247d | 755 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
1da177e4 LT |
756 | if (attr->ia_size < inode->i_size) { |
757 | /* | |
758 | * If truncating down to a partial page, then | |
759 | * if that page is already allocated, hold it | |
760 | * in memory until the truncation is over, so | |
761 | * truncate_partial_page cannnot miss it were | |
762 | * it assigned to swap. | |
763 | */ | |
764 | if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { | |
765 | (void) shmem_getpage(inode, | |
766 | attr->ia_size>>PAGE_CACHE_SHIFT, | |
767 | &page, SGP_READ, NULL); | |
d3602444 HD |
768 | if (page) |
769 | unlock_page(page); | |
1da177e4 LT |
770 | } |
771 | /* | |
772 | * Reset SHMEM_PAGEIN flag so that shmem_truncate can | |
773 | * detect if any pages might have been added to cache | |
774 | * after truncate_inode_pages. But we needn't bother | |
775 | * if it's being fully truncated to zero-length: the | |
776 | * nrpages check is efficient enough in that case. | |
777 | */ | |
778 | if (attr->ia_size) { | |
779 | struct shmem_inode_info *info = SHMEM_I(inode); | |
780 | spin_lock(&info->lock); | |
781 | info->flags &= ~SHMEM_PAGEIN; | |
782 | spin_unlock(&info->lock); | |
783 | } | |
784 | } | |
785 | } | |
786 | ||
787 | error = inode_change_ok(inode, attr); | |
788 | if (!error) | |
789 | error = inode_setattr(inode, attr); | |
39f0247d AG |
790 | #ifdef CONFIG_TMPFS_POSIX_ACL |
791 | if (!error && (attr->ia_valid & ATTR_MODE)) | |
792 | error = generic_acl_chmod(inode, &shmem_acl_ops); | |
793 | #endif | |
1da177e4 LT |
794 | if (page) |
795 | page_cache_release(page); | |
796 | return error; | |
797 | } | |
798 | ||
799 | static void shmem_delete_inode(struct inode *inode) | |
800 | { | |
1da177e4 LT |
801 | struct shmem_inode_info *info = SHMEM_I(inode); |
802 | ||
803 | if (inode->i_op->truncate == shmem_truncate) { | |
fef26658 | 804 | truncate_inode_pages(inode->i_mapping, 0); |
1da177e4 LT |
805 | shmem_unacct_size(info->flags, inode->i_size); |
806 | inode->i_size = 0; | |
807 | shmem_truncate(inode); | |
808 | if (!list_empty(&info->swaplist)) { | |
cb5f7b9a | 809 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 810 | list_del_init(&info->swaplist); |
cb5f7b9a | 811 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 LT |
812 | } |
813 | } | |
0edd73b3 | 814 | BUG_ON(inode->i_blocks); |
5b04c689 | 815 | shmem_free_inode(inode->i_sb); |
1da177e4 LT |
816 | clear_inode(inode); |
817 | } | |
818 | ||
819 | static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) | |
820 | { | |
821 | swp_entry_t *ptr; | |
822 | ||
823 | for (ptr = dir; ptr < edir; ptr++) { | |
824 | if (ptr->val == entry.val) | |
825 | return ptr - dir; | |
826 | } | |
827 | return -1; | |
828 | } | |
829 | ||
830 | static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) | |
831 | { | |
832 | struct inode *inode; | |
833 | unsigned long idx; | |
834 | unsigned long size; | |
835 | unsigned long limit; | |
836 | unsigned long stage; | |
837 | struct page **dir; | |
838 | struct page *subdir; | |
839 | swp_entry_t *ptr; | |
840 | int offset; | |
d9fe526a | 841 | int error; |
1da177e4 LT |
842 | |
843 | idx = 0; | |
844 | ptr = info->i_direct; | |
845 | spin_lock(&info->lock); | |
1b1b32f2 HD |
846 | if (!info->swapped) { |
847 | list_del_init(&info->swaplist); | |
848 | goto lost2; | |
849 | } | |
1da177e4 LT |
850 | limit = info->next_index; |
851 | size = limit; | |
852 | if (size > SHMEM_NR_DIRECT) | |
853 | size = SHMEM_NR_DIRECT; | |
854 | offset = shmem_find_swp(entry, ptr, ptr+size); | |
2e0e26c7 | 855 | if (offset >= 0) |
1da177e4 | 856 | goto found; |
1da177e4 LT |
857 | if (!info->i_indirect) |
858 | goto lost2; | |
859 | ||
860 | dir = shmem_dir_map(info->i_indirect); | |
861 | stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; | |
862 | ||
863 | for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { | |
864 | if (unlikely(idx == stage)) { | |
865 | shmem_dir_unmap(dir-1); | |
cb5f7b9a HD |
866 | if (cond_resched_lock(&info->lock)) { |
867 | /* check it has not been truncated */ | |
868 | if (limit > info->next_index) { | |
869 | limit = info->next_index; | |
870 | if (idx >= limit) | |
871 | goto lost2; | |
872 | } | |
873 | } | |
1da177e4 LT |
874 | dir = shmem_dir_map(info->i_indirect) + |
875 | ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; | |
876 | while (!*dir) { | |
877 | dir++; | |
878 | idx += ENTRIES_PER_PAGEPAGE; | |
879 | if (idx >= limit) | |
880 | goto lost1; | |
881 | } | |
882 | stage = idx + ENTRIES_PER_PAGEPAGE; | |
883 | subdir = *dir; | |
884 | shmem_dir_unmap(dir); | |
885 | dir = shmem_dir_map(subdir); | |
886 | } | |
887 | subdir = *dir; | |
4c21e2f2 | 888 | if (subdir && page_private(subdir)) { |
1da177e4 LT |
889 | ptr = shmem_swp_map(subdir); |
890 | size = limit - idx; | |
891 | if (size > ENTRIES_PER_PAGE) | |
892 | size = ENTRIES_PER_PAGE; | |
893 | offset = shmem_find_swp(entry, ptr, ptr+size); | |
2e0e26c7 | 894 | shmem_swp_unmap(ptr); |
1da177e4 LT |
895 | if (offset >= 0) { |
896 | shmem_dir_unmap(dir); | |
897 | goto found; | |
898 | } | |
1da177e4 LT |
899 | } |
900 | } | |
901 | lost1: | |
902 | shmem_dir_unmap(dir-1); | |
903 | lost2: | |
904 | spin_unlock(&info->lock); | |
905 | return 0; | |
906 | found: | |
907 | idx += offset; | |
2e0e26c7 HD |
908 | inode = igrab(&info->vfs_inode); |
909 | spin_unlock(&info->lock); | |
910 | ||
1b1b32f2 HD |
911 | /* |
912 | * Move _head_ to start search for next from here. | |
913 | * But be careful: shmem_delete_inode checks list_empty without taking | |
914 | * mutex, and there's an instant in list_move_tail when info->swaplist | |
915 | * would appear empty, if it were the only one on shmem_swaplist. We | |
916 | * could avoid doing it if inode NULL; or use this minor optimization. | |
917 | */ | |
918 | if (shmem_swaplist.next != &info->swaplist) | |
919 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 HD |
920 | mutex_unlock(&shmem_swaplist_mutex); |
921 | ||
922 | error = 1; | |
923 | if (!inode) | |
924 | goto out; | |
82369553 HD |
925 | /* Precharge page while we can wait, compensate afterwards */ |
926 | error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); | |
b409f9fc HD |
927 | if (error) |
928 | goto out; | |
82369553 HD |
929 | error = radix_tree_preload(GFP_KERNEL); |
930 | if (error) | |
931 | goto uncharge; | |
b409f9fc | 932 | error = 1; |
2e0e26c7 HD |
933 | |
934 | spin_lock(&info->lock); | |
935 | ptr = shmem_swp_entry(info, idx, NULL); | |
936 | if (ptr && ptr->val == entry.val) | |
937 | error = add_to_page_cache(page, inode->i_mapping, | |
b409f9fc | 938 | idx, GFP_NOWAIT); |
d9fe526a HD |
939 | if (error == -EEXIST) { |
940 | struct page *filepage = find_get_page(inode->i_mapping, idx); | |
2e0e26c7 | 941 | error = 1; |
d9fe526a HD |
942 | if (filepage) { |
943 | /* | |
944 | * There might be a more uptodate page coming down | |
945 | * from a stacked writepage: forget our swappage if so. | |
946 | */ | |
947 | if (PageUptodate(filepage)) | |
948 | error = 0; | |
949 | page_cache_release(filepage); | |
950 | } | |
951 | } | |
952 | if (!error) { | |
73b1262f HD |
953 | delete_from_swap_cache(page); |
954 | set_page_dirty(page); | |
1da177e4 | 955 | info->flags |= SHMEM_PAGEIN; |
2e0e26c7 HD |
956 | shmem_swp_set(info, ptr, 0); |
957 | swap_free(entry); | |
958 | error = 1; /* not an error, but entry was found */ | |
1da177e4 | 959 | } |
2e0e26c7 HD |
960 | if (ptr) |
961 | shmem_swp_unmap(ptr); | |
1da177e4 | 962 | spin_unlock(&info->lock); |
b409f9fc | 963 | radix_tree_preload_end(); |
82369553 HD |
964 | uncharge: |
965 | mem_cgroup_uncharge_page(page); | |
2e0e26c7 HD |
966 | out: |
967 | unlock_page(page); | |
968 | page_cache_release(page); | |
969 | iput(inode); /* allows for NULL */ | |
970 | return error; | |
1da177e4 LT |
971 | } |
972 | ||
973 | /* | |
974 | * shmem_unuse() search for an eventually swapped out shmem page. | |
975 | */ | |
976 | int shmem_unuse(swp_entry_t entry, struct page *page) | |
977 | { | |
978 | struct list_head *p, *next; | |
979 | struct shmem_inode_info *info; | |
980 | int found = 0; | |
981 | ||
cb5f7b9a | 982 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 LT |
983 | list_for_each_safe(p, next, &shmem_swaplist) { |
984 | info = list_entry(p, struct shmem_inode_info, swaplist); | |
1b1b32f2 | 985 | found = shmem_unuse_inode(info, entry, page); |
cb5f7b9a | 986 | cond_resched(); |
2e0e26c7 HD |
987 | if (found) |
988 | goto out; | |
1da177e4 | 989 | } |
cb5f7b9a | 990 | mutex_unlock(&shmem_swaplist_mutex); |
2e0e26c7 | 991 | out: return found; /* 0 or 1 or -ENOMEM */ |
1da177e4 LT |
992 | } |
993 | ||
994 | /* | |
995 | * Move the page from the page cache to the swap cache. | |
996 | */ | |
997 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
998 | { | |
999 | struct shmem_inode_info *info; | |
1000 | swp_entry_t *entry, swap; | |
1001 | struct address_space *mapping; | |
1002 | unsigned long index; | |
1003 | struct inode *inode; | |
1004 | ||
1005 | BUG_ON(!PageLocked(page)); | |
1da177e4 LT |
1006 | mapping = page->mapping; |
1007 | index = page->index; | |
1008 | inode = mapping->host; | |
1009 | info = SHMEM_I(inode); | |
1010 | if (info->flags & VM_LOCKED) | |
1011 | goto redirty; | |
d9fe526a | 1012 | if (!total_swap_pages) |
1da177e4 LT |
1013 | goto redirty; |
1014 | ||
d9fe526a HD |
1015 | /* |
1016 | * shmem_backing_dev_info's capabilities prevent regular writeback or | |
1017 | * sync from ever calling shmem_writepage; but a stacking filesystem | |
1018 | * may use the ->writepage of its underlying filesystem, in which case | |
1019 | * tmpfs should write out to swap only in response to memory pressure, | |
1020 | * and not for pdflush or sync. However, in those cases, we do still | |
1021 | * want to check if there's a redundant swappage to be discarded. | |
1022 | */ | |
1023 | if (wbc->for_reclaim) | |
1024 | swap = get_swap_page(); | |
1025 | else | |
1026 | swap.val = 0; | |
1027 | ||
1da177e4 | 1028 | spin_lock(&info->lock); |
1da177e4 LT |
1029 | if (index >= info->next_index) { |
1030 | BUG_ON(!(info->flags & SHMEM_TRUNCATE)); | |
1031 | goto unlock; | |
1032 | } | |
1033 | entry = shmem_swp_entry(info, index, NULL); | |
d9fe526a HD |
1034 | if (entry->val) { |
1035 | /* | |
1036 | * The more uptodate page coming down from a stacked | |
1037 | * writepage should replace our old swappage. | |
1038 | */ | |
1039 | free_swap_and_cache(*entry); | |
1040 | shmem_swp_set(info, entry, 0); | |
1041 | } | |
1042 | shmem_recalc_inode(inode); | |
1da177e4 | 1043 | |
d9fe526a | 1044 | if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
73b1262f | 1045 | remove_from_page_cache(page); |
1da177e4 LT |
1046 | shmem_swp_set(info, entry, swap.val); |
1047 | shmem_swp_unmap(entry); | |
1b1b32f2 HD |
1048 | if (list_empty(&info->swaplist)) |
1049 | inode = igrab(inode); | |
1050 | else | |
1051 | inode = NULL; | |
1da177e4 | 1052 | spin_unlock(&info->lock); |
73b1262f | 1053 | swap_duplicate(swap); |
d9fe526a | 1054 | BUG_ON(page_mapped(page)); |
73b1262f HD |
1055 | page_cache_release(page); /* pagecache ref */ |
1056 | set_page_dirty(page); | |
1da177e4 | 1057 | unlock_page(page); |
1b1b32f2 HD |
1058 | if (inode) { |
1059 | mutex_lock(&shmem_swaplist_mutex); | |
1060 | /* move instead of add in case we're racing */ | |
1061 | list_move_tail(&info->swaplist, &shmem_swaplist); | |
1062 | mutex_unlock(&shmem_swaplist_mutex); | |
1063 | iput(inode); | |
1064 | } | |
1da177e4 LT |
1065 | return 0; |
1066 | } | |
1067 | ||
1068 | shmem_swp_unmap(entry); | |
1069 | unlock: | |
1070 | spin_unlock(&info->lock); | |
1071 | swap_free(swap); | |
1072 | redirty: | |
1073 | set_page_dirty(page); | |
d9fe526a HD |
1074 | if (wbc->for_reclaim) |
1075 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1076 | unlock_page(page); | |
1077 | return 0; | |
1da177e4 LT |
1078 | } |
1079 | ||
1080 | #ifdef CONFIG_NUMA | |
680d794b | 1081 | #ifdef CONFIG_TMPFS |
a3b51e01 | 1082 | static int shmem_parse_mpol(char *value, unsigned short *policy, |
028fec41 | 1083 | unsigned short *mode_flags, nodemask_t *policy_nodes) |
b00dc3ad HD |
1084 | { |
1085 | char *nodelist = strchr(value, ':'); | |
028fec41 | 1086 | char *flags = strchr(value, '='); |
b00dc3ad HD |
1087 | int err = 1; |
1088 | ||
1089 | if (nodelist) { | |
1090 | /* NUL-terminate policy string */ | |
1091 | *nodelist++ = '\0'; | |
1092 | if (nodelist_parse(nodelist, *policy_nodes)) | |
1093 | goto out; | |
37b07e41 | 1094 | if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY])) |
a210906c | 1095 | goto out; |
b00dc3ad | 1096 | } |
028fec41 DR |
1097 | if (flags) |
1098 | *flags++ = '\0'; | |
b00dc3ad HD |
1099 | if (!strcmp(value, "default")) { |
1100 | *policy = MPOL_DEFAULT; | |
1101 | /* Don't allow a nodelist */ | |
1102 | if (!nodelist) | |
1103 | err = 0; | |
1104 | } else if (!strcmp(value, "prefer")) { | |
1105 | *policy = MPOL_PREFERRED; | |
1106 | /* Insist on a nodelist of one node only */ | |
1107 | if (nodelist) { | |
1108 | char *rest = nodelist; | |
1109 | while (isdigit(*rest)) | |
1110 | rest++; | |
1111 | if (!*rest) | |
1112 | err = 0; | |
1113 | } | |
1114 | } else if (!strcmp(value, "bind")) { | |
1115 | *policy = MPOL_BIND; | |
1116 | /* Insist on a nodelist */ | |
1117 | if (nodelist) | |
1118 | err = 0; | |
1119 | } else if (!strcmp(value, "interleave")) { | |
1120 | *policy = MPOL_INTERLEAVE; | |
37b07e41 LS |
1121 | /* |
1122 | * Default to online nodes with memory if no nodelist | |
1123 | */ | |
b00dc3ad | 1124 | if (!nodelist) |
37b07e41 | 1125 | *policy_nodes = node_states[N_HIGH_MEMORY]; |
b00dc3ad HD |
1126 | err = 0; |
1127 | } | |
a43361cf LS |
1128 | |
1129 | *mode_flags = 0; | |
028fec41 | 1130 | if (flags) { |
a43361cf LS |
1131 | /* |
1132 | * Currently, we only support two mutually exclusive | |
1133 | * mode flags. | |
1134 | */ | |
f5b087b5 DR |
1135 | if (!strcmp(flags, "static")) |
1136 | *mode_flags |= MPOL_F_STATIC_NODES; | |
a43361cf | 1137 | else if (!strcmp(flags, "relative")) |
4c50bc01 | 1138 | *mode_flags |= MPOL_F_RELATIVE_NODES; |
a43361cf LS |
1139 | else |
1140 | err = 1; /* unrecognized flag */ | |
028fec41 | 1141 | } |
b00dc3ad HD |
1142 | out: |
1143 | /* Restore string for error message */ | |
1144 | if (nodelist) | |
1145 | *--nodelist = ':'; | |
a43361cf LS |
1146 | if (flags) |
1147 | *--flags = '='; | |
b00dc3ad HD |
1148 | return err; |
1149 | } | |
1150 | ||
a3b51e01 | 1151 | static void shmem_show_mpol(struct seq_file *seq, unsigned short policy, |
028fec41 | 1152 | unsigned short flags, const nodemask_t policy_nodes) |
680d794b | 1153 | { |
1154 | char *policy_string; | |
1155 | ||
1156 | switch (policy) { | |
1157 | case MPOL_PREFERRED: | |
1158 | policy_string = "prefer"; | |
1159 | break; | |
1160 | case MPOL_BIND: | |
1161 | policy_string = "bind"; | |
1162 | break; | |
1163 | case MPOL_INTERLEAVE: | |
1164 | policy_string = "interleave"; | |
1165 | break; | |
1166 | default: | |
1167 | /* MPOL_DEFAULT */ | |
1168 | return; | |
1169 | } | |
1170 | ||
1171 | seq_printf(seq, ",mpol=%s", policy_string); | |
1172 | ||
1173 | if (policy != MPOL_INTERLEAVE || | |
1174 | !nodes_equal(policy_nodes, node_states[N_HIGH_MEMORY])) { | |
1175 | char buffer[64]; | |
1176 | int len; | |
1177 | ||
1178 | len = nodelist_scnprintf(buffer, sizeof(buffer), policy_nodes); | |
1179 | if (len < sizeof(buffer)) | |
1180 | seq_printf(seq, ":%s", buffer); | |
1181 | else | |
1182 | seq_printf(seq, ":?"); | |
1183 | } | |
1184 | } | |
1185 | #endif /* CONFIG_TMPFS */ | |
1186 | ||
02098fea HD |
1187 | static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, |
1188 | struct shmem_inode_info *info, unsigned long idx) | |
1da177e4 | 1189 | { |
52cd3b07 | 1190 | struct mempolicy mpol, *spol; |
1da177e4 | 1191 | struct vm_area_struct pvma; |
c4cc6d07 | 1192 | struct page *page; |
1da177e4 | 1193 | |
52cd3b07 LS |
1194 | spol = mpol_cond_copy(&mpol, |
1195 | mpol_shared_policy_lookup(&info->policy, idx)); | |
1196 | ||
1da177e4 | 1197 | /* Create a pseudo vma that just contains the policy */ |
c4cc6d07 | 1198 | pvma.vm_start = 0; |
1da177e4 | 1199 | pvma.vm_pgoff = idx; |
c4cc6d07 | 1200 | pvma.vm_ops = NULL; |
52cd3b07 | 1201 | pvma.vm_policy = spol; |
02098fea | 1202 | page = swapin_readahead(entry, gfp, &pvma, 0); |
1da177e4 LT |
1203 | return page; |
1204 | } | |
1205 | ||
02098fea HD |
1206 | static struct page *shmem_alloc_page(gfp_t gfp, |
1207 | struct shmem_inode_info *info, unsigned long idx) | |
1da177e4 LT |
1208 | { |
1209 | struct vm_area_struct pvma; | |
1da177e4 | 1210 | |
c4cc6d07 HD |
1211 | /* Create a pseudo vma that just contains the policy */ |
1212 | pvma.vm_start = 0; | |
1da177e4 | 1213 | pvma.vm_pgoff = idx; |
c4cc6d07 HD |
1214 | pvma.vm_ops = NULL; |
1215 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); | |
52cd3b07 LS |
1216 | |
1217 | /* | |
1218 | * alloc_page_vma() will drop the shared policy reference | |
1219 | */ | |
1220 | return alloc_page_vma(gfp, &pvma, 0); | |
1da177e4 | 1221 | } |
680d794b | 1222 | #else /* !CONFIG_NUMA */ |
1223 | #ifdef CONFIG_TMPFS | |
a3b51e01 | 1224 | static inline int shmem_parse_mpol(char *value, unsigned short *policy, |
028fec41 | 1225 | unsigned short *mode_flags, nodemask_t *policy_nodes) |
b00dc3ad HD |
1226 | { |
1227 | return 1; | |
1228 | } | |
1229 | ||
a3b51e01 | 1230 | static inline void shmem_show_mpol(struct seq_file *seq, unsigned short policy, |
028fec41 | 1231 | unsigned short flags, const nodemask_t policy_nodes) |
680d794b | 1232 | { |
1233 | } | |
1234 | #endif /* CONFIG_TMPFS */ | |
1235 | ||
02098fea HD |
1236 | static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, |
1237 | struct shmem_inode_info *info, unsigned long idx) | |
1da177e4 | 1238 | { |
02098fea | 1239 | return swapin_readahead(entry, gfp, NULL, 0); |
1da177e4 LT |
1240 | } |
1241 | ||
02098fea HD |
1242 | static inline struct page *shmem_alloc_page(gfp_t gfp, |
1243 | struct shmem_inode_info *info, unsigned long idx) | |
1da177e4 | 1244 | { |
e84e2e13 | 1245 | return alloc_page(gfp); |
1da177e4 | 1246 | } |
680d794b | 1247 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
1248 | |
1249 | /* | |
1250 | * shmem_getpage - either get the page from swap or allocate a new one | |
1251 | * | |
1252 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1253 | * vm. If we swap it in we mark it dirty since we also free the swap | |
1254 | * entry since a page cannot live in both the swap and page cache | |
1255 | */ | |
1256 | static int shmem_getpage(struct inode *inode, unsigned long idx, | |
1257 | struct page **pagep, enum sgp_type sgp, int *type) | |
1258 | { | |
1259 | struct address_space *mapping = inode->i_mapping; | |
1260 | struct shmem_inode_info *info = SHMEM_I(inode); | |
1261 | struct shmem_sb_info *sbinfo; | |
1262 | struct page *filepage = *pagep; | |
1263 | struct page *swappage; | |
1264 | swp_entry_t *entry; | |
1265 | swp_entry_t swap; | |
02098fea | 1266 | gfp_t gfp; |
1da177e4 LT |
1267 | int error; |
1268 | ||
1269 | if (idx >= SHMEM_MAX_INDEX) | |
1270 | return -EFBIG; | |
54cb8821 NP |
1271 | |
1272 | if (type) | |
83c54070 | 1273 | *type = 0; |
54cb8821 | 1274 | |
1da177e4 LT |
1275 | /* |
1276 | * Normally, filepage is NULL on entry, and either found | |
1277 | * uptodate immediately, or allocated and zeroed, or read | |
1278 | * in under swappage, which is then assigned to filepage. | |
5402b976 | 1279 | * But shmem_readpage (required for splice) passes in a locked |
ae976416 HD |
1280 | * filepage, which may be found not uptodate by other callers |
1281 | * too, and may need to be copied from the swappage read in. | |
1da177e4 LT |
1282 | */ |
1283 | repeat: | |
1284 | if (!filepage) | |
1285 | filepage = find_lock_page(mapping, idx); | |
1286 | if (filepage && PageUptodate(filepage)) | |
1287 | goto done; | |
1288 | error = 0; | |
02098fea | 1289 | gfp = mapping_gfp_mask(mapping); |
b409f9fc HD |
1290 | if (!filepage) { |
1291 | /* | |
1292 | * Try to preload while we can wait, to not make a habit of | |
1293 | * draining atomic reserves; but don't latch on to this cpu. | |
1294 | */ | |
1295 | error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); | |
1296 | if (error) | |
1297 | goto failed; | |
1298 | radix_tree_preload_end(); | |
1299 | } | |
1da177e4 LT |
1300 | |
1301 | spin_lock(&info->lock); | |
1302 | shmem_recalc_inode(inode); | |
1303 | entry = shmem_swp_alloc(info, idx, sgp); | |
1304 | if (IS_ERR(entry)) { | |
1305 | spin_unlock(&info->lock); | |
1306 | error = PTR_ERR(entry); | |
1307 | goto failed; | |
1308 | } | |
1309 | swap = *entry; | |
1310 | ||
1311 | if (swap.val) { | |
1312 | /* Look it up and read it in.. */ | |
1313 | swappage = lookup_swap_cache(swap); | |
1314 | if (!swappage) { | |
1315 | shmem_swp_unmap(entry); | |
1da177e4 | 1316 | /* here we actually do the io */ |
83c54070 | 1317 | if (type && !(*type & VM_FAULT_MAJOR)) { |
f8891e5e | 1318 | __count_vm_event(PGMAJFAULT); |
83c54070 | 1319 | *type |= VM_FAULT_MAJOR; |
1da177e4 | 1320 | } |
f8891e5e | 1321 | spin_unlock(&info->lock); |
02098fea | 1322 | swappage = shmem_swapin(swap, gfp, info, idx); |
1da177e4 LT |
1323 | if (!swappage) { |
1324 | spin_lock(&info->lock); | |
1325 | entry = shmem_swp_alloc(info, idx, sgp); | |
1326 | if (IS_ERR(entry)) | |
1327 | error = PTR_ERR(entry); | |
1328 | else { | |
1329 | if (entry->val == swap.val) | |
1330 | error = -ENOMEM; | |
1331 | shmem_swp_unmap(entry); | |
1332 | } | |
1333 | spin_unlock(&info->lock); | |
1334 | if (error) | |
1335 | goto failed; | |
1336 | goto repeat; | |
1337 | } | |
1338 | wait_on_page_locked(swappage); | |
1339 | page_cache_release(swappage); | |
1340 | goto repeat; | |
1341 | } | |
1342 | ||
1343 | /* We have to do this with page locked to prevent races */ | |
1344 | if (TestSetPageLocked(swappage)) { | |
1345 | shmem_swp_unmap(entry); | |
1346 | spin_unlock(&info->lock); | |
1347 | wait_on_page_locked(swappage); | |
1348 | page_cache_release(swappage); | |
1349 | goto repeat; | |
1350 | } | |
1351 | if (PageWriteback(swappage)) { | |
1352 | shmem_swp_unmap(entry); | |
1353 | spin_unlock(&info->lock); | |
1354 | wait_on_page_writeback(swappage); | |
1355 | unlock_page(swappage); | |
1356 | page_cache_release(swappage); | |
1357 | goto repeat; | |
1358 | } | |
1359 | if (!PageUptodate(swappage)) { | |
1360 | shmem_swp_unmap(entry); | |
1361 | spin_unlock(&info->lock); | |
1362 | unlock_page(swappage); | |
1363 | page_cache_release(swappage); | |
1364 | error = -EIO; | |
1365 | goto failed; | |
1366 | } | |
1367 | ||
1368 | if (filepage) { | |
1369 | shmem_swp_set(info, entry, 0); | |
1370 | shmem_swp_unmap(entry); | |
1371 | delete_from_swap_cache(swappage); | |
1372 | spin_unlock(&info->lock); | |
1373 | copy_highpage(filepage, swappage); | |
1374 | unlock_page(swappage); | |
1375 | page_cache_release(swappage); | |
1376 | flush_dcache_page(filepage); | |
1377 | SetPageUptodate(filepage); | |
1378 | set_page_dirty(filepage); | |
1379 | swap_free(swap); | |
73b1262f | 1380 | } else if (!(error = add_to_page_cache( |
b409f9fc | 1381 | swappage, mapping, idx, GFP_NOWAIT))) { |
1da177e4 LT |
1382 | info->flags |= SHMEM_PAGEIN; |
1383 | shmem_swp_set(info, entry, 0); | |
1384 | shmem_swp_unmap(entry); | |
73b1262f | 1385 | delete_from_swap_cache(swappage); |
1da177e4 LT |
1386 | spin_unlock(&info->lock); |
1387 | filepage = swappage; | |
73b1262f | 1388 | set_page_dirty(filepage); |
1da177e4 LT |
1389 | swap_free(swap); |
1390 | } else { | |
1391 | shmem_swp_unmap(entry); | |
1392 | spin_unlock(&info->lock); | |
1393 | unlock_page(swappage); | |
82369553 HD |
1394 | if (error == -ENOMEM) { |
1395 | /* allow reclaim from this memory cgroup */ | |
7e924aaf | 1396 | error = mem_cgroup_cache_charge(swappage, |
82369553 | 1397 | current->mm, gfp & ~__GFP_HIGHMEM); |
7e924aaf HD |
1398 | if (error) { |
1399 | page_cache_release(swappage); | |
82369553 | 1400 | goto failed; |
7e924aaf HD |
1401 | } |
1402 | mem_cgroup_uncharge_page(swappage); | |
82369553 | 1403 | } |
7e924aaf | 1404 | page_cache_release(swappage); |
1da177e4 LT |
1405 | goto repeat; |
1406 | } | |
1407 | } else if (sgp == SGP_READ && !filepage) { | |
1408 | shmem_swp_unmap(entry); | |
1409 | filepage = find_get_page(mapping, idx); | |
1410 | if (filepage && | |
1411 | (!PageUptodate(filepage) || TestSetPageLocked(filepage))) { | |
1412 | spin_unlock(&info->lock); | |
1413 | wait_on_page_locked(filepage); | |
1414 | page_cache_release(filepage); | |
1415 | filepage = NULL; | |
1416 | goto repeat; | |
1417 | } | |
1418 | spin_unlock(&info->lock); | |
1419 | } else { | |
1420 | shmem_swp_unmap(entry); | |
1421 | sbinfo = SHMEM_SB(inode->i_sb); | |
0edd73b3 | 1422 | if (sbinfo->max_blocks) { |
1da177e4 LT |
1423 | spin_lock(&sbinfo->stat_lock); |
1424 | if (sbinfo->free_blocks == 0 || | |
1425 | shmem_acct_block(info->flags)) { | |
1426 | spin_unlock(&sbinfo->stat_lock); | |
1427 | spin_unlock(&info->lock); | |
1428 | error = -ENOSPC; | |
1429 | goto failed; | |
1430 | } | |
1431 | sbinfo->free_blocks--; | |
1432 | inode->i_blocks += BLOCKS_PER_PAGE; | |
1433 | spin_unlock(&sbinfo->stat_lock); | |
1434 | } else if (shmem_acct_block(info->flags)) { | |
1435 | spin_unlock(&info->lock); | |
1436 | error = -ENOSPC; | |
1437 | goto failed; | |
1438 | } | |
1439 | ||
1440 | if (!filepage) { | |
1441 | spin_unlock(&info->lock); | |
02098fea | 1442 | filepage = shmem_alloc_page(gfp, info, idx); |
1da177e4 LT |
1443 | if (!filepage) { |
1444 | shmem_unacct_blocks(info->flags, 1); | |
1445 | shmem_free_blocks(inode, 1); | |
1446 | error = -ENOMEM; | |
1447 | goto failed; | |
1448 | } | |
1449 | ||
82369553 HD |
1450 | /* Precharge page while we can wait, compensate after */ |
1451 | error = mem_cgroup_cache_charge(filepage, current->mm, | |
1452 | gfp & ~__GFP_HIGHMEM); | |
1453 | if (error) { | |
1454 | page_cache_release(filepage); | |
1455 | shmem_unacct_blocks(info->flags, 1); | |
1456 | shmem_free_blocks(inode, 1); | |
1457 | filepage = NULL; | |
1458 | goto failed; | |
1459 | } | |
1460 | ||
1da177e4 LT |
1461 | spin_lock(&info->lock); |
1462 | entry = shmem_swp_alloc(info, idx, sgp); | |
1463 | if (IS_ERR(entry)) | |
1464 | error = PTR_ERR(entry); | |
1465 | else { | |
1466 | swap = *entry; | |
1467 | shmem_swp_unmap(entry); | |
1468 | } | |
1469 | if (error || swap.val || 0 != add_to_page_cache_lru( | |
b409f9fc | 1470 | filepage, mapping, idx, GFP_NOWAIT)) { |
1da177e4 | 1471 | spin_unlock(&info->lock); |
82369553 | 1472 | mem_cgroup_uncharge_page(filepage); |
1da177e4 LT |
1473 | page_cache_release(filepage); |
1474 | shmem_unacct_blocks(info->flags, 1); | |
1475 | shmem_free_blocks(inode, 1); | |
1476 | filepage = NULL; | |
1477 | if (error) | |
1478 | goto failed; | |
1479 | goto repeat; | |
1480 | } | |
82369553 | 1481 | mem_cgroup_uncharge_page(filepage); |
1da177e4 LT |
1482 | info->flags |= SHMEM_PAGEIN; |
1483 | } | |
1484 | ||
1485 | info->alloced++; | |
1486 | spin_unlock(&info->lock); | |
e84e2e13 | 1487 | clear_highpage(filepage); |
1da177e4 LT |
1488 | flush_dcache_page(filepage); |
1489 | SetPageUptodate(filepage); | |
a0ee5ec5 HD |
1490 | if (sgp == SGP_DIRTY) |
1491 | set_page_dirty(filepage); | |
1da177e4 LT |
1492 | } |
1493 | done: | |
d3602444 | 1494 | *pagep = filepage; |
1da177e4 LT |
1495 | return 0; |
1496 | ||
1497 | failed: | |
1498 | if (*pagep != filepage) { | |
1499 | unlock_page(filepage); | |
1500 | page_cache_release(filepage); | |
1501 | } | |
1502 | return error; | |
1503 | } | |
1504 | ||
d0217ac0 | 1505 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1506 | { |
d3ac7f89 | 1507 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 | 1508 | int error; |
d0217ac0 | 1509 | int ret; |
1da177e4 | 1510 | |
d0217ac0 NP |
1511 | if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) |
1512 | return VM_FAULT_SIGBUS; | |
d00806b1 | 1513 | |
27d54b39 | 1514 | error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); |
d0217ac0 NP |
1515 | if (error) |
1516 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
1da177e4 | 1517 | |
d0217ac0 | 1518 | mark_page_accessed(vmf->page); |
83c54070 | 1519 | return ret | VM_FAULT_LOCKED; |
1da177e4 LT |
1520 | } |
1521 | ||
1da177e4 | 1522 | #ifdef CONFIG_NUMA |
d8dc74f2 | 1523 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) |
1da177e4 | 1524 | { |
d3ac7f89 | 1525 | struct inode *i = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 LT |
1526 | return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); |
1527 | } | |
1528 | ||
d8dc74f2 AB |
1529 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
1530 | unsigned long addr) | |
1da177e4 | 1531 | { |
d3ac7f89 | 1532 | struct inode *i = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 LT |
1533 | unsigned long idx; |
1534 | ||
1535 | idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
1536 | return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); | |
1537 | } | |
1538 | #endif | |
1539 | ||
1540 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
1541 | { | |
d3ac7f89 | 1542 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
1543 | struct shmem_inode_info *info = SHMEM_I(inode); |
1544 | int retval = -ENOMEM; | |
1545 | ||
1546 | spin_lock(&info->lock); | |
1547 | if (lock && !(info->flags & VM_LOCKED)) { | |
1548 | if (!user_shm_lock(inode->i_size, user)) | |
1549 | goto out_nomem; | |
1550 | info->flags |= VM_LOCKED; | |
1551 | } | |
1552 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
1553 | user_shm_unlock(inode->i_size, user); | |
1554 | info->flags &= ~VM_LOCKED; | |
1555 | } | |
1556 | retval = 0; | |
1557 | out_nomem: | |
1558 | spin_unlock(&info->lock); | |
1559 | return retval; | |
1560 | } | |
1561 | ||
9b83a6a8 | 1562 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
1563 | { |
1564 | file_accessed(file); | |
1565 | vma->vm_ops = &shmem_vm_ops; | |
d0217ac0 | 1566 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
1567 | return 0; |
1568 | } | |
1569 | ||
1570 | static struct inode * | |
1571 | shmem_get_inode(struct super_block *sb, int mode, dev_t dev) | |
1572 | { | |
1573 | struct inode *inode; | |
1574 | struct shmem_inode_info *info; | |
1575 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
1576 | ||
5b04c689 PE |
1577 | if (shmem_reserve_inode(sb)) |
1578 | return NULL; | |
1da177e4 LT |
1579 | |
1580 | inode = new_inode(sb); | |
1581 | if (inode) { | |
1582 | inode->i_mode = mode; | |
1583 | inode->i_uid = current->fsuid; | |
1584 | inode->i_gid = current->fsgid; | |
1da177e4 LT |
1585 | inode->i_blocks = 0; |
1586 | inode->i_mapping->a_ops = &shmem_aops; | |
1587 | inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; | |
1588 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
91828a40 | 1589 | inode->i_generation = get_seconds(); |
1da177e4 LT |
1590 | info = SHMEM_I(inode); |
1591 | memset(info, 0, (char *)inode - (char *)info); | |
1592 | spin_lock_init(&info->lock); | |
1593 | INIT_LIST_HEAD(&info->swaplist); | |
1594 | ||
1595 | switch (mode & S_IFMT) { | |
1596 | default: | |
39f0247d | 1597 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
1598 | init_special_inode(inode, mode, dev); |
1599 | break; | |
1600 | case S_IFREG: | |
1601 | inode->i_op = &shmem_inode_operations; | |
1602 | inode->i_fop = &shmem_file_operations; | |
7339ff83 | 1603 | mpol_shared_policy_init(&info->policy, sbinfo->policy, |
028fec41 | 1604 | sbinfo->flags, &sbinfo->policy_nodes); |
1da177e4 LT |
1605 | break; |
1606 | case S_IFDIR: | |
d8c76e6f | 1607 | inc_nlink(inode); |
1da177e4 LT |
1608 | /* Some things misbehave if size == 0 on a directory */ |
1609 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
1610 | inode->i_op = &shmem_dir_inode_operations; | |
1611 | inode->i_fop = &simple_dir_operations; | |
1612 | break; | |
1613 | case S_IFLNK: | |
1614 | /* | |
1615 | * Must not load anything in the rbtree, | |
1616 | * mpol_free_shared_policy will not be called. | |
1617 | */ | |
028fec41 | 1618 | mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 0, |
7339ff83 | 1619 | NULL); |
1da177e4 LT |
1620 | break; |
1621 | } | |
5b04c689 PE |
1622 | } else |
1623 | shmem_free_inode(sb); | |
1da177e4 LT |
1624 | return inode; |
1625 | } | |
1626 | ||
1627 | #ifdef CONFIG_TMPFS | |
92e1d5be AV |
1628 | static const struct inode_operations shmem_symlink_inode_operations; |
1629 | static const struct inode_operations shmem_symlink_inline_operations; | |
1da177e4 LT |
1630 | |
1631 | /* | |
800d15a5 | 1632 | * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; |
ae976416 HD |
1633 | * but providing them allows a tmpfs file to be used for splice, sendfile, and |
1634 | * below the loop driver, in the generic fashion that many filesystems support. | |
1da177e4 | 1635 | */ |
ae976416 HD |
1636 | static int shmem_readpage(struct file *file, struct page *page) |
1637 | { | |
1638 | struct inode *inode = page->mapping->host; | |
1639 | int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); | |
1640 | unlock_page(page); | |
1641 | return error; | |
1642 | } | |
1643 | ||
1da177e4 | 1644 | static int |
800d15a5 NP |
1645 | shmem_write_begin(struct file *file, struct address_space *mapping, |
1646 | loff_t pos, unsigned len, unsigned flags, | |
1647 | struct page **pagep, void **fsdata) | |
1da177e4 | 1648 | { |
800d15a5 NP |
1649 | struct inode *inode = mapping->host; |
1650 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
1651 | *pagep = NULL; | |
1652 | return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); | |
1653 | } | |
1654 | ||
1655 | static int | |
1656 | shmem_write_end(struct file *file, struct address_space *mapping, | |
1657 | loff_t pos, unsigned len, unsigned copied, | |
1658 | struct page *page, void *fsdata) | |
1659 | { | |
1660 | struct inode *inode = mapping->host; | |
1661 | ||
d3602444 HD |
1662 | if (pos + copied > inode->i_size) |
1663 | i_size_write(inode, pos + copied); | |
1664 | ||
1665 | unlock_page(page); | |
800d15a5 NP |
1666 | set_page_dirty(page); |
1667 | page_cache_release(page); | |
1668 | ||
800d15a5 | 1669 | return copied; |
1da177e4 LT |
1670 | } |
1671 | ||
1da177e4 LT |
1672 | static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) |
1673 | { | |
d3ac7f89 | 1674 | struct inode *inode = filp->f_path.dentry->d_inode; |
1da177e4 LT |
1675 | struct address_space *mapping = inode->i_mapping; |
1676 | unsigned long index, offset; | |
a0ee5ec5 HD |
1677 | enum sgp_type sgp = SGP_READ; |
1678 | ||
1679 | /* | |
1680 | * Might this read be for a stacking filesystem? Then when reading | |
1681 | * holes of a sparse file, we actually need to allocate those pages, | |
1682 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
1683 | */ | |
1684 | if (segment_eq(get_fs(), KERNEL_DS)) | |
1685 | sgp = SGP_DIRTY; | |
1da177e4 LT |
1686 | |
1687 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1688 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1689 | ||
1690 | for (;;) { | |
1691 | struct page *page = NULL; | |
1692 | unsigned long end_index, nr, ret; | |
1693 | loff_t i_size = i_size_read(inode); | |
1694 | ||
1695 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1696 | if (index > end_index) | |
1697 | break; | |
1698 | if (index == end_index) { | |
1699 | nr = i_size & ~PAGE_CACHE_MASK; | |
1700 | if (nr <= offset) | |
1701 | break; | |
1702 | } | |
1703 | ||
a0ee5ec5 | 1704 | desc->error = shmem_getpage(inode, index, &page, sgp, NULL); |
1da177e4 LT |
1705 | if (desc->error) { |
1706 | if (desc->error == -EINVAL) | |
1707 | desc->error = 0; | |
1708 | break; | |
1709 | } | |
d3602444 HD |
1710 | if (page) |
1711 | unlock_page(page); | |
1da177e4 LT |
1712 | |
1713 | /* | |
1714 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 1715 | * are called without i_mutex protection against truncate |
1da177e4 LT |
1716 | */ |
1717 | nr = PAGE_CACHE_SIZE; | |
1718 | i_size = i_size_read(inode); | |
1719 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1720 | if (index == end_index) { | |
1721 | nr = i_size & ~PAGE_CACHE_MASK; | |
1722 | if (nr <= offset) { | |
1723 | if (page) | |
1724 | page_cache_release(page); | |
1725 | break; | |
1726 | } | |
1727 | } | |
1728 | nr -= offset; | |
1729 | ||
1730 | if (page) { | |
1731 | /* | |
1732 | * If users can be writing to this page using arbitrary | |
1733 | * virtual addresses, take care about potential aliasing | |
1734 | * before reading the page on the kernel side. | |
1735 | */ | |
1736 | if (mapping_writably_mapped(mapping)) | |
1737 | flush_dcache_page(page); | |
1738 | /* | |
1739 | * Mark the page accessed if we read the beginning. | |
1740 | */ | |
1741 | if (!offset) | |
1742 | mark_page_accessed(page); | |
b5810039 | 1743 | } else { |
1da177e4 | 1744 | page = ZERO_PAGE(0); |
b5810039 NP |
1745 | page_cache_get(page); |
1746 | } | |
1da177e4 LT |
1747 | |
1748 | /* | |
1749 | * Ok, we have the page, and it's up-to-date, so | |
1750 | * now we can copy it to user space... | |
1751 | * | |
1752 | * The actor routine returns how many bytes were actually used.. | |
1753 | * NOTE! This may not be the same as how much of a user buffer | |
1754 | * we filled up (we may be padding etc), so we can only update | |
1755 | * "pos" here (the actor routine has to update the user buffer | |
1756 | * pointers and the remaining count). | |
1757 | */ | |
1758 | ret = actor(desc, page, offset, nr); | |
1759 | offset += ret; | |
1760 | index += offset >> PAGE_CACHE_SHIFT; | |
1761 | offset &= ~PAGE_CACHE_MASK; | |
1762 | ||
1763 | page_cache_release(page); | |
1764 | if (ret != nr || !desc->count) | |
1765 | break; | |
1766 | ||
1767 | cond_resched(); | |
1768 | } | |
1769 | ||
1770 | *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; | |
1771 | file_accessed(filp); | |
1772 | } | |
1773 | ||
1774 | static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) | |
1775 | { | |
1776 | read_descriptor_t desc; | |
1777 | ||
1778 | if ((ssize_t) count < 0) | |
1779 | return -EINVAL; | |
1780 | if (!access_ok(VERIFY_WRITE, buf, count)) | |
1781 | return -EFAULT; | |
1782 | if (!count) | |
1783 | return 0; | |
1784 | ||
1785 | desc.written = 0; | |
1786 | desc.count = count; | |
1787 | desc.arg.buf = buf; | |
1788 | desc.error = 0; | |
1789 | ||
1790 | do_shmem_file_read(filp, ppos, &desc, file_read_actor); | |
1791 | if (desc.written) | |
1792 | return desc.written; | |
1793 | return desc.error; | |
1794 | } | |
1795 | ||
726c3342 | 1796 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 1797 | { |
726c3342 | 1798 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
1799 | |
1800 | buf->f_type = TMPFS_MAGIC; | |
1801 | buf->f_bsize = PAGE_CACHE_SIZE; | |
1802 | buf->f_namelen = NAME_MAX; | |
0edd73b3 HD |
1803 | spin_lock(&sbinfo->stat_lock); |
1804 | if (sbinfo->max_blocks) { | |
1da177e4 LT |
1805 | buf->f_blocks = sbinfo->max_blocks; |
1806 | buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; | |
0edd73b3 HD |
1807 | } |
1808 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
1809 | buf->f_files = sbinfo->max_inodes; |
1810 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
1811 | } |
1812 | /* else leave those fields 0 like simple_statfs */ | |
0edd73b3 | 1813 | spin_unlock(&sbinfo->stat_lock); |
1da177e4 LT |
1814 | return 0; |
1815 | } | |
1816 | ||
1817 | /* | |
1818 | * File creation. Allocate an inode, and we're done.. | |
1819 | */ | |
1820 | static int | |
1821 | shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) | |
1822 | { | |
1823 | struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); | |
1824 | int error = -ENOSPC; | |
1825 | ||
1826 | if (inode) { | |
570bc1c2 SS |
1827 | error = security_inode_init_security(inode, dir, NULL, NULL, |
1828 | NULL); | |
1829 | if (error) { | |
1830 | if (error != -EOPNOTSUPP) { | |
1831 | iput(inode); | |
1832 | return error; | |
1833 | } | |
39f0247d AG |
1834 | } |
1835 | error = shmem_acl_init(inode, dir); | |
1836 | if (error) { | |
1837 | iput(inode); | |
1838 | return error; | |
570bc1c2 | 1839 | } |
1da177e4 LT |
1840 | if (dir->i_mode & S_ISGID) { |
1841 | inode->i_gid = dir->i_gid; | |
1842 | if (S_ISDIR(mode)) | |
1843 | inode->i_mode |= S_ISGID; | |
1844 | } | |
1845 | dir->i_size += BOGO_DIRENT_SIZE; | |
1846 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
1847 | d_instantiate(dentry, inode); | |
1848 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
1849 | } |
1850 | return error; | |
1851 | } | |
1852 | ||
1853 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
1854 | { | |
1855 | int error; | |
1856 | ||
1857 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
1858 | return error; | |
d8c76e6f | 1859 | inc_nlink(dir); |
1da177e4 LT |
1860 | return 0; |
1861 | } | |
1862 | ||
1863 | static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, | |
1864 | struct nameidata *nd) | |
1865 | { | |
1866 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
1867 | } | |
1868 | ||
1869 | /* | |
1870 | * Link a file.. | |
1871 | */ | |
1872 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1873 | { | |
1874 | struct inode *inode = old_dentry->d_inode; | |
5b04c689 | 1875 | int ret; |
1da177e4 LT |
1876 | |
1877 | /* | |
1878 | * No ordinary (disk based) filesystem counts links as inodes; | |
1879 | * but each new link needs a new dentry, pinning lowmem, and | |
1880 | * tmpfs dentries cannot be pruned until they are unlinked. | |
1881 | */ | |
5b04c689 PE |
1882 | ret = shmem_reserve_inode(inode->i_sb); |
1883 | if (ret) | |
1884 | goto out; | |
1da177e4 LT |
1885 | |
1886 | dir->i_size += BOGO_DIRENT_SIZE; | |
1887 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 1888 | inc_nlink(inode); |
1da177e4 LT |
1889 | atomic_inc(&inode->i_count); /* New dentry reference */ |
1890 | dget(dentry); /* Extra pinning count for the created dentry */ | |
1891 | d_instantiate(dentry, inode); | |
5b04c689 PE |
1892 | out: |
1893 | return ret; | |
1da177e4 LT |
1894 | } |
1895 | ||
1896 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
1897 | { | |
1898 | struct inode *inode = dentry->d_inode; | |
1899 | ||
5b04c689 PE |
1900 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
1901 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
1902 | |
1903 | dir->i_size -= BOGO_DIRENT_SIZE; | |
1904 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 1905 | drop_nlink(inode); |
1da177e4 LT |
1906 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
1907 | return 0; | |
1908 | } | |
1909 | ||
1910 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
1911 | { | |
1912 | if (!simple_empty(dentry)) | |
1913 | return -ENOTEMPTY; | |
1914 | ||
9a53c3a7 DH |
1915 | drop_nlink(dentry->d_inode); |
1916 | drop_nlink(dir); | |
1da177e4 LT |
1917 | return shmem_unlink(dir, dentry); |
1918 | } | |
1919 | ||
1920 | /* | |
1921 | * The VFS layer already does all the dentry stuff for rename, | |
1922 | * we just have to decrement the usage count for the target if | |
1923 | * it exists so that the VFS layer correctly free's it when it | |
1924 | * gets overwritten. | |
1925 | */ | |
1926 | static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) | |
1927 | { | |
1928 | struct inode *inode = old_dentry->d_inode; | |
1929 | int they_are_dirs = S_ISDIR(inode->i_mode); | |
1930 | ||
1931 | if (!simple_empty(new_dentry)) | |
1932 | return -ENOTEMPTY; | |
1933 | ||
1934 | if (new_dentry->d_inode) { | |
1935 | (void) shmem_unlink(new_dir, new_dentry); | |
1936 | if (they_are_dirs) | |
9a53c3a7 | 1937 | drop_nlink(old_dir); |
1da177e4 | 1938 | } else if (they_are_dirs) { |
9a53c3a7 | 1939 | drop_nlink(old_dir); |
d8c76e6f | 1940 | inc_nlink(new_dir); |
1da177e4 LT |
1941 | } |
1942 | ||
1943 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
1944 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
1945 | old_dir->i_ctime = old_dir->i_mtime = | |
1946 | new_dir->i_ctime = new_dir->i_mtime = | |
1947 | inode->i_ctime = CURRENT_TIME; | |
1948 | return 0; | |
1949 | } | |
1950 | ||
1951 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1952 | { | |
1953 | int error; | |
1954 | int len; | |
1955 | struct inode *inode; | |
1956 | struct page *page = NULL; | |
1957 | char *kaddr; | |
1958 | struct shmem_inode_info *info; | |
1959 | ||
1960 | len = strlen(symname) + 1; | |
1961 | if (len > PAGE_CACHE_SIZE) | |
1962 | return -ENAMETOOLONG; | |
1963 | ||
1964 | inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); | |
1965 | if (!inode) | |
1966 | return -ENOSPC; | |
1967 | ||
570bc1c2 SS |
1968 | error = security_inode_init_security(inode, dir, NULL, NULL, |
1969 | NULL); | |
1970 | if (error) { | |
1971 | if (error != -EOPNOTSUPP) { | |
1972 | iput(inode); | |
1973 | return error; | |
1974 | } | |
1975 | error = 0; | |
1976 | } | |
1977 | ||
1da177e4 LT |
1978 | info = SHMEM_I(inode); |
1979 | inode->i_size = len-1; | |
1980 | if (len <= (char *)inode - (char *)info) { | |
1981 | /* do it inline */ | |
1982 | memcpy(info, symname, len); | |
1983 | inode->i_op = &shmem_symlink_inline_operations; | |
1984 | } else { | |
1985 | error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); | |
1986 | if (error) { | |
1987 | iput(inode); | |
1988 | return error; | |
1989 | } | |
d3602444 | 1990 | unlock_page(page); |
1da177e4 LT |
1991 | inode->i_op = &shmem_symlink_inode_operations; |
1992 | kaddr = kmap_atomic(page, KM_USER0); | |
1993 | memcpy(kaddr, symname, len); | |
1994 | kunmap_atomic(kaddr, KM_USER0); | |
1995 | set_page_dirty(page); | |
1996 | page_cache_release(page); | |
1997 | } | |
1998 | if (dir->i_mode & S_ISGID) | |
1999 | inode->i_gid = dir->i_gid; | |
2000 | dir->i_size += BOGO_DIRENT_SIZE; | |
2001 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2002 | d_instantiate(dentry, inode); | |
2003 | dget(dentry); | |
2004 | return 0; | |
2005 | } | |
2006 | ||
cc314eef | 2007 | static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) |
1da177e4 LT |
2008 | { |
2009 | nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); | |
cc314eef | 2010 | return NULL; |
1da177e4 LT |
2011 | } |
2012 | ||
cc314eef | 2013 | static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) |
1da177e4 LT |
2014 | { |
2015 | struct page *page = NULL; | |
2016 | int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); | |
2017 | nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); | |
d3602444 HD |
2018 | if (page) |
2019 | unlock_page(page); | |
cc314eef | 2020 | return page; |
1da177e4 LT |
2021 | } |
2022 | ||
cc314eef | 2023 | static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 LT |
2024 | { |
2025 | if (!IS_ERR(nd_get_link(nd))) { | |
cc314eef | 2026 | struct page *page = cookie; |
1da177e4 LT |
2027 | kunmap(page); |
2028 | mark_page_accessed(page); | |
2029 | page_cache_release(page); | |
1da177e4 LT |
2030 | } |
2031 | } | |
2032 | ||
92e1d5be | 2033 | static const struct inode_operations shmem_symlink_inline_operations = { |
1da177e4 LT |
2034 | .readlink = generic_readlink, |
2035 | .follow_link = shmem_follow_link_inline, | |
1da177e4 LT |
2036 | }; |
2037 | ||
92e1d5be | 2038 | static const struct inode_operations shmem_symlink_inode_operations = { |
1da177e4 LT |
2039 | .truncate = shmem_truncate, |
2040 | .readlink = generic_readlink, | |
2041 | .follow_link = shmem_follow_link, | |
2042 | .put_link = shmem_put_link, | |
1da177e4 LT |
2043 | }; |
2044 | ||
39f0247d | 2045 | #ifdef CONFIG_TMPFS_POSIX_ACL |
46711810 | 2046 | /* |
39f0247d AG |
2047 | * Superblocks without xattr inode operations will get security.* xattr |
2048 | * support from the VFS "for free". As soon as we have any other xattrs | |
2049 | * like ACLs, we also need to implement the security.* handlers at | |
2050 | * filesystem level, though. | |
2051 | */ | |
2052 | ||
2053 | static size_t shmem_xattr_security_list(struct inode *inode, char *list, | |
2054 | size_t list_len, const char *name, | |
2055 | size_t name_len) | |
2056 | { | |
2057 | return security_inode_listsecurity(inode, list, list_len); | |
2058 | } | |
2059 | ||
2060 | static int shmem_xattr_security_get(struct inode *inode, const char *name, | |
2061 | void *buffer, size_t size) | |
2062 | { | |
2063 | if (strcmp(name, "") == 0) | |
2064 | return -EINVAL; | |
42492594 | 2065 | return xattr_getsecurity(inode, name, buffer, size); |
39f0247d AG |
2066 | } |
2067 | ||
2068 | static int shmem_xattr_security_set(struct inode *inode, const char *name, | |
2069 | const void *value, size_t size, int flags) | |
2070 | { | |
2071 | if (strcmp(name, "") == 0) | |
2072 | return -EINVAL; | |
2073 | return security_inode_setsecurity(inode, name, value, size, flags); | |
2074 | } | |
2075 | ||
1f370a23 | 2076 | static struct xattr_handler shmem_xattr_security_handler = { |
39f0247d AG |
2077 | .prefix = XATTR_SECURITY_PREFIX, |
2078 | .list = shmem_xattr_security_list, | |
2079 | .get = shmem_xattr_security_get, | |
2080 | .set = shmem_xattr_security_set, | |
2081 | }; | |
2082 | ||
2083 | static struct xattr_handler *shmem_xattr_handlers[] = { | |
2084 | &shmem_xattr_acl_access_handler, | |
2085 | &shmem_xattr_acl_default_handler, | |
2086 | &shmem_xattr_security_handler, | |
2087 | NULL | |
2088 | }; | |
2089 | #endif | |
2090 | ||
91828a40 DG |
2091 | static struct dentry *shmem_get_parent(struct dentry *child) |
2092 | { | |
2093 | return ERR_PTR(-ESTALE); | |
2094 | } | |
2095 | ||
2096 | static int shmem_match(struct inode *ino, void *vfh) | |
2097 | { | |
2098 | __u32 *fh = vfh; | |
2099 | __u64 inum = fh[2]; | |
2100 | inum = (inum << 32) | fh[1]; | |
2101 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
2102 | } | |
2103 | ||
480b116c CH |
2104 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
2105 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 2106 | { |
91828a40 | 2107 | struct inode *inode; |
480b116c CH |
2108 | struct dentry *dentry = NULL; |
2109 | u64 inum = fid->raw[2]; | |
2110 | inum = (inum << 32) | fid->raw[1]; | |
2111 | ||
2112 | if (fh_len < 3) | |
2113 | return NULL; | |
91828a40 | 2114 | |
480b116c CH |
2115 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
2116 | shmem_match, fid->raw); | |
91828a40 | 2117 | if (inode) { |
480b116c | 2118 | dentry = d_find_alias(inode); |
91828a40 DG |
2119 | iput(inode); |
2120 | } | |
2121 | ||
480b116c | 2122 | return dentry; |
91828a40 DG |
2123 | } |
2124 | ||
2125 | static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, | |
2126 | int connectable) | |
2127 | { | |
2128 | struct inode *inode = dentry->d_inode; | |
2129 | ||
2130 | if (*len < 3) | |
2131 | return 255; | |
2132 | ||
2133 | if (hlist_unhashed(&inode->i_hash)) { | |
2134 | /* Unfortunately insert_inode_hash is not idempotent, | |
2135 | * so as we hash inodes here rather than at creation | |
2136 | * time, we need a lock to ensure we only try | |
2137 | * to do it once | |
2138 | */ | |
2139 | static DEFINE_SPINLOCK(lock); | |
2140 | spin_lock(&lock); | |
2141 | if (hlist_unhashed(&inode->i_hash)) | |
2142 | __insert_inode_hash(inode, | |
2143 | inode->i_ino + inode->i_generation); | |
2144 | spin_unlock(&lock); | |
2145 | } | |
2146 | ||
2147 | fh[0] = inode->i_generation; | |
2148 | fh[1] = inode->i_ino; | |
2149 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
2150 | ||
2151 | *len = 3; | |
2152 | return 1; | |
2153 | } | |
2154 | ||
39655164 | 2155 | static const struct export_operations shmem_export_ops = { |
91828a40 | 2156 | .get_parent = shmem_get_parent, |
91828a40 | 2157 | .encode_fh = shmem_encode_fh, |
480b116c | 2158 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
2159 | }; |
2160 | ||
680d794b | 2161 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
2162 | bool remount) | |
1da177e4 LT |
2163 | { |
2164 | char *this_char, *value, *rest; | |
2165 | ||
b00dc3ad HD |
2166 | while (options != NULL) { |
2167 | this_char = options; | |
2168 | for (;;) { | |
2169 | /* | |
2170 | * NUL-terminate this option: unfortunately, | |
2171 | * mount options form a comma-separated list, | |
2172 | * but mpol's nodelist may also contain commas. | |
2173 | */ | |
2174 | options = strchr(options, ','); | |
2175 | if (options == NULL) | |
2176 | break; | |
2177 | options++; | |
2178 | if (!isdigit(*options)) { | |
2179 | options[-1] = '\0'; | |
2180 | break; | |
2181 | } | |
2182 | } | |
1da177e4 LT |
2183 | if (!*this_char) |
2184 | continue; | |
2185 | if ((value = strchr(this_char,'=')) != NULL) { | |
2186 | *value++ = 0; | |
2187 | } else { | |
2188 | printk(KERN_ERR | |
2189 | "tmpfs: No value for mount option '%s'\n", | |
2190 | this_char); | |
2191 | return 1; | |
2192 | } | |
2193 | ||
2194 | if (!strcmp(this_char,"size")) { | |
2195 | unsigned long long size; | |
2196 | size = memparse(value,&rest); | |
2197 | if (*rest == '%') { | |
2198 | size <<= PAGE_SHIFT; | |
2199 | size *= totalram_pages; | |
2200 | do_div(size, 100); | |
2201 | rest++; | |
2202 | } | |
2203 | if (*rest) | |
2204 | goto bad_val; | |
680d794b | 2205 | sbinfo->max_blocks = |
2206 | DIV_ROUND_UP(size, PAGE_CACHE_SIZE); | |
1da177e4 | 2207 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 2208 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
2209 | if (*rest) |
2210 | goto bad_val; | |
2211 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 2212 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
2213 | if (*rest) |
2214 | goto bad_val; | |
2215 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 2216 | if (remount) |
1da177e4 | 2217 | continue; |
680d794b | 2218 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
2219 | if (*rest) |
2220 | goto bad_val; | |
2221 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 2222 | if (remount) |
1da177e4 | 2223 | continue; |
680d794b | 2224 | sbinfo->uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2225 | if (*rest) |
2226 | goto bad_val; | |
2227 | } else if (!strcmp(this_char,"gid")) { | |
680d794b | 2228 | if (remount) |
1da177e4 | 2229 | continue; |
680d794b | 2230 | sbinfo->gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2231 | if (*rest) |
2232 | goto bad_val; | |
7339ff83 | 2233 | } else if (!strcmp(this_char,"mpol")) { |
680d794b | 2234 | if (shmem_parse_mpol(value, &sbinfo->policy, |
028fec41 | 2235 | &sbinfo->flags, &sbinfo->policy_nodes)) |
7339ff83 | 2236 | goto bad_val; |
1da177e4 LT |
2237 | } else { |
2238 | printk(KERN_ERR "tmpfs: Bad mount option %s\n", | |
2239 | this_char); | |
2240 | return 1; | |
2241 | } | |
2242 | } | |
2243 | return 0; | |
2244 | ||
2245 | bad_val: | |
2246 | printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", | |
2247 | value, this_char); | |
2248 | return 1; | |
2249 | ||
2250 | } | |
2251 | ||
2252 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
2253 | { | |
2254 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 2255 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
2256 | unsigned long blocks; |
2257 | unsigned long inodes; | |
2258 | int error = -EINVAL; | |
2259 | ||
680d794b | 2260 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 2261 | return error; |
1da177e4 | 2262 | |
0edd73b3 HD |
2263 | spin_lock(&sbinfo->stat_lock); |
2264 | blocks = sbinfo->max_blocks - sbinfo->free_blocks; | |
2265 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; | |
680d794b | 2266 | if (config.max_blocks < blocks) |
0edd73b3 | 2267 | goto out; |
680d794b | 2268 | if (config.max_inodes < inodes) |
0edd73b3 HD |
2269 | goto out; |
2270 | /* | |
2271 | * Those tests also disallow limited->unlimited while any are in | |
2272 | * use, so i_blocks will always be zero when max_blocks is zero; | |
2273 | * but we must separately disallow unlimited->limited, because | |
2274 | * in that case we have no record of how much is already in use. | |
2275 | */ | |
680d794b | 2276 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 2277 | goto out; |
680d794b | 2278 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
2279 | goto out; |
2280 | ||
2281 | error = 0; | |
680d794b | 2282 | sbinfo->max_blocks = config.max_blocks; |
2283 | sbinfo->free_blocks = config.max_blocks - blocks; | |
2284 | sbinfo->max_inodes = config.max_inodes; | |
2285 | sbinfo->free_inodes = config.max_inodes - inodes; | |
2286 | sbinfo->policy = config.policy; | |
028fec41 | 2287 | sbinfo->flags = config.flags; |
680d794b | 2288 | sbinfo->policy_nodes = config.policy_nodes; |
0edd73b3 HD |
2289 | out: |
2290 | spin_unlock(&sbinfo->stat_lock); | |
2291 | return error; | |
1da177e4 | 2292 | } |
680d794b | 2293 | |
2294 | static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) | |
2295 | { | |
2296 | struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); | |
2297 | ||
2298 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
2299 | seq_printf(seq, ",size=%luk", | |
2300 | sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); | |
2301 | if (sbinfo->max_inodes != shmem_default_max_inodes()) | |
2302 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
2303 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
2304 | seq_printf(seq, ",mode=%03o", sbinfo->mode); | |
2305 | if (sbinfo->uid != 0) | |
2306 | seq_printf(seq, ",uid=%u", sbinfo->uid); | |
2307 | if (sbinfo->gid != 0) | |
2308 | seq_printf(seq, ",gid=%u", sbinfo->gid); | |
028fec41 DR |
2309 | shmem_show_mpol(seq, sbinfo->policy, sbinfo->flags, |
2310 | sbinfo->policy_nodes); | |
680d794b | 2311 | return 0; |
2312 | } | |
2313 | #endif /* CONFIG_TMPFS */ | |
1da177e4 LT |
2314 | |
2315 | static void shmem_put_super(struct super_block *sb) | |
2316 | { | |
2317 | kfree(sb->s_fs_info); | |
2318 | sb->s_fs_info = NULL; | |
2319 | } | |
2320 | ||
1da177e4 LT |
2321 | static int shmem_fill_super(struct super_block *sb, |
2322 | void *data, int silent) | |
2323 | { | |
2324 | struct inode *inode; | |
2325 | struct dentry *root; | |
0edd73b3 | 2326 | struct shmem_sb_info *sbinfo; |
680d794b | 2327 | int err = -ENOMEM; |
2328 | ||
2329 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
2330 | sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), | |
2331 | L1_CACHE_BYTES), GFP_KERNEL); | |
2332 | if (!sbinfo) | |
2333 | return -ENOMEM; | |
2334 | ||
2335 | sbinfo->max_blocks = 0; | |
2336 | sbinfo->max_inodes = 0; | |
2337 | sbinfo->mode = S_IRWXUGO | S_ISVTX; | |
2338 | sbinfo->uid = current->fsuid; | |
2339 | sbinfo->gid = current->fsgid; | |
2340 | sbinfo->policy = MPOL_DEFAULT; | |
028fec41 | 2341 | sbinfo->flags = 0; |
680d794b | 2342 | sbinfo->policy_nodes = node_states[N_HIGH_MEMORY]; |
2343 | sb->s_fs_info = sbinfo; | |
1da177e4 | 2344 | |
0edd73b3 | 2345 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
2346 | /* |
2347 | * Per default we only allow half of the physical ram per | |
2348 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
2349 | * but the internal instance is left unlimited. | |
2350 | */ | |
2351 | if (!(sb->s_flags & MS_NOUSER)) { | |
680d794b | 2352 | sbinfo->max_blocks = shmem_default_max_blocks(); |
2353 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
2354 | if (shmem_parse_options(data, sbinfo, false)) { | |
2355 | err = -EINVAL; | |
2356 | goto failed; | |
2357 | } | |
1da177e4 | 2358 | } |
91828a40 | 2359 | sb->s_export_op = &shmem_export_ops; |
1da177e4 LT |
2360 | #else |
2361 | sb->s_flags |= MS_NOUSER; | |
2362 | #endif | |
2363 | ||
0edd73b3 | 2364 | spin_lock_init(&sbinfo->stat_lock); |
680d794b | 2365 | sbinfo->free_blocks = sbinfo->max_blocks; |
2366 | sbinfo->free_inodes = sbinfo->max_inodes; | |
0edd73b3 | 2367 | |
1da177e4 LT |
2368 | sb->s_maxbytes = SHMEM_MAX_BYTES; |
2369 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
2370 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
2371 | sb->s_magic = TMPFS_MAGIC; | |
2372 | sb->s_op = &shmem_ops; | |
cfd95a9c | 2373 | sb->s_time_gran = 1; |
39f0247d AG |
2374 | #ifdef CONFIG_TMPFS_POSIX_ACL |
2375 | sb->s_xattr = shmem_xattr_handlers; | |
2376 | sb->s_flags |= MS_POSIXACL; | |
2377 | #endif | |
0edd73b3 | 2378 | |
680d794b | 2379 | inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0); |
1da177e4 LT |
2380 | if (!inode) |
2381 | goto failed; | |
680d794b | 2382 | inode->i_uid = sbinfo->uid; |
2383 | inode->i_gid = sbinfo->gid; | |
1da177e4 LT |
2384 | root = d_alloc_root(inode); |
2385 | if (!root) | |
2386 | goto failed_iput; | |
2387 | sb->s_root = root; | |
2388 | return 0; | |
2389 | ||
2390 | failed_iput: | |
2391 | iput(inode); | |
2392 | failed: | |
2393 | shmem_put_super(sb); | |
2394 | return err; | |
2395 | } | |
2396 | ||
fcc234f8 | 2397 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
2398 | |
2399 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
2400 | { | |
2401 | struct shmem_inode_info *p; | |
e94b1766 | 2402 | p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); |
1da177e4 LT |
2403 | if (!p) |
2404 | return NULL; | |
2405 | return &p->vfs_inode; | |
2406 | } | |
2407 | ||
2408 | static void shmem_destroy_inode(struct inode *inode) | |
2409 | { | |
2410 | if ((inode->i_mode & S_IFMT) == S_IFREG) { | |
2411 | /* only struct inode is valid if it's an inline symlink */ | |
2412 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); | |
2413 | } | |
39f0247d | 2414 | shmem_acl_destroy_inode(inode); |
1da177e4 LT |
2415 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
2416 | } | |
2417 | ||
4ba9b9d0 | 2418 | static void init_once(struct kmem_cache *cachep, void *foo) |
1da177e4 LT |
2419 | { |
2420 | struct shmem_inode_info *p = (struct shmem_inode_info *) foo; | |
2421 | ||
a35afb83 | 2422 | inode_init_once(&p->vfs_inode); |
39f0247d | 2423 | #ifdef CONFIG_TMPFS_POSIX_ACL |
a35afb83 CL |
2424 | p->i_acl = NULL; |
2425 | p->i_default_acl = NULL; | |
39f0247d | 2426 | #endif |
1da177e4 LT |
2427 | } |
2428 | ||
2429 | static int init_inodecache(void) | |
2430 | { | |
2431 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
2432 | sizeof(struct shmem_inode_info), | |
040b5c6f | 2433 | 0, SLAB_PANIC, init_once); |
1da177e4 LT |
2434 | return 0; |
2435 | } | |
2436 | ||
2437 | static void destroy_inodecache(void) | |
2438 | { | |
1a1d92c1 | 2439 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
2440 | } |
2441 | ||
f5e54d6e | 2442 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 2443 | .writepage = shmem_writepage, |
76719325 | 2444 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 2445 | #ifdef CONFIG_TMPFS |
ae976416 | 2446 | .readpage = shmem_readpage, |
800d15a5 NP |
2447 | .write_begin = shmem_write_begin, |
2448 | .write_end = shmem_write_end, | |
1da177e4 | 2449 | #endif |
304dbdb7 | 2450 | .migratepage = migrate_page, |
1da177e4 LT |
2451 | }; |
2452 | ||
15ad7cdc | 2453 | static const struct file_operations shmem_file_operations = { |
1da177e4 LT |
2454 | .mmap = shmem_mmap, |
2455 | #ifdef CONFIG_TMPFS | |
2456 | .llseek = generic_file_llseek, | |
2457 | .read = shmem_file_read, | |
5402b976 HD |
2458 | .write = do_sync_write, |
2459 | .aio_write = generic_file_aio_write, | |
1da177e4 | 2460 | .fsync = simple_sync_file, |
ae976416 HD |
2461 | .splice_read = generic_file_splice_read, |
2462 | .splice_write = generic_file_splice_write, | |
1da177e4 LT |
2463 | #endif |
2464 | }; | |
2465 | ||
92e1d5be | 2466 | static const struct inode_operations shmem_inode_operations = { |
1da177e4 LT |
2467 | .truncate = shmem_truncate, |
2468 | .setattr = shmem_notify_change, | |
f6b3ec23 | 2469 | .truncate_range = shmem_truncate_range, |
39f0247d AG |
2470 | #ifdef CONFIG_TMPFS_POSIX_ACL |
2471 | .setxattr = generic_setxattr, | |
2472 | .getxattr = generic_getxattr, | |
2473 | .listxattr = generic_listxattr, | |
2474 | .removexattr = generic_removexattr, | |
2475 | .permission = shmem_permission, | |
2476 | #endif | |
2477 | ||
1da177e4 LT |
2478 | }; |
2479 | ||
92e1d5be | 2480 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
2481 | #ifdef CONFIG_TMPFS |
2482 | .create = shmem_create, | |
2483 | .lookup = simple_lookup, | |
2484 | .link = shmem_link, | |
2485 | .unlink = shmem_unlink, | |
2486 | .symlink = shmem_symlink, | |
2487 | .mkdir = shmem_mkdir, | |
2488 | .rmdir = shmem_rmdir, | |
2489 | .mknod = shmem_mknod, | |
2490 | .rename = shmem_rename, | |
1da177e4 | 2491 | #endif |
39f0247d AG |
2492 | #ifdef CONFIG_TMPFS_POSIX_ACL |
2493 | .setattr = shmem_notify_change, | |
2494 | .setxattr = generic_setxattr, | |
2495 | .getxattr = generic_getxattr, | |
2496 | .listxattr = generic_listxattr, | |
2497 | .removexattr = generic_removexattr, | |
2498 | .permission = shmem_permission, | |
2499 | #endif | |
2500 | }; | |
2501 | ||
92e1d5be | 2502 | static const struct inode_operations shmem_special_inode_operations = { |
39f0247d AG |
2503 | #ifdef CONFIG_TMPFS_POSIX_ACL |
2504 | .setattr = shmem_notify_change, | |
2505 | .setxattr = generic_setxattr, | |
2506 | .getxattr = generic_getxattr, | |
2507 | .listxattr = generic_listxattr, | |
2508 | .removexattr = generic_removexattr, | |
2509 | .permission = shmem_permission, | |
2510 | #endif | |
1da177e4 LT |
2511 | }; |
2512 | ||
759b9775 | 2513 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
2514 | .alloc_inode = shmem_alloc_inode, |
2515 | .destroy_inode = shmem_destroy_inode, | |
2516 | #ifdef CONFIG_TMPFS | |
2517 | .statfs = shmem_statfs, | |
2518 | .remount_fs = shmem_remount_fs, | |
680d794b | 2519 | .show_options = shmem_show_options, |
1da177e4 LT |
2520 | #endif |
2521 | .delete_inode = shmem_delete_inode, | |
2522 | .drop_inode = generic_delete_inode, | |
2523 | .put_super = shmem_put_super, | |
2524 | }; | |
2525 | ||
2526 | static struct vm_operations_struct shmem_vm_ops = { | |
54cb8821 | 2527 | .fault = shmem_fault, |
1da177e4 LT |
2528 | #ifdef CONFIG_NUMA |
2529 | .set_policy = shmem_set_policy, | |
2530 | .get_policy = shmem_get_policy, | |
2531 | #endif | |
2532 | }; | |
2533 | ||
2534 | ||
454e2398 DH |
2535 | static int shmem_get_sb(struct file_system_type *fs_type, |
2536 | int flags, const char *dev_name, void *data, struct vfsmount *mnt) | |
1da177e4 | 2537 | { |
454e2398 | 2538 | return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); |
1da177e4 LT |
2539 | } |
2540 | ||
2541 | static struct file_system_type tmpfs_fs_type = { | |
2542 | .owner = THIS_MODULE, | |
2543 | .name = "tmpfs", | |
2544 | .get_sb = shmem_get_sb, | |
2545 | .kill_sb = kill_litter_super, | |
2546 | }; | |
2547 | static struct vfsmount *shm_mnt; | |
2548 | ||
2549 | static int __init init_tmpfs(void) | |
2550 | { | |
2551 | int error; | |
2552 | ||
e0bf68dd PZ |
2553 | error = bdi_init(&shmem_backing_dev_info); |
2554 | if (error) | |
2555 | goto out4; | |
2556 | ||
1da177e4 LT |
2557 | error = init_inodecache(); |
2558 | if (error) | |
2559 | goto out3; | |
2560 | ||
2561 | error = register_filesystem(&tmpfs_fs_type); | |
2562 | if (error) { | |
2563 | printk(KERN_ERR "Could not register tmpfs\n"); | |
2564 | goto out2; | |
2565 | } | |
95dc112a | 2566 | |
1f5ce9e9 | 2567 | shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, |
1da177e4 LT |
2568 | tmpfs_fs_type.name, NULL); |
2569 | if (IS_ERR(shm_mnt)) { | |
2570 | error = PTR_ERR(shm_mnt); | |
2571 | printk(KERN_ERR "Could not kern_mount tmpfs\n"); | |
2572 | goto out1; | |
2573 | } | |
2574 | return 0; | |
2575 | ||
2576 | out1: | |
2577 | unregister_filesystem(&tmpfs_fs_type); | |
2578 | out2: | |
2579 | destroy_inodecache(); | |
2580 | out3: | |
e0bf68dd PZ |
2581 | bdi_destroy(&shmem_backing_dev_info); |
2582 | out4: | |
1da177e4 LT |
2583 | shm_mnt = ERR_PTR(error); |
2584 | return error; | |
2585 | } | |
2586 | module_init(init_tmpfs) | |
2587 | ||
46711810 | 2588 | /** |
1da177e4 | 2589 | * shmem_file_setup - get an unlinked file living in tmpfs |
1da177e4 LT |
2590 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
2591 | * @size: size to be set for the file | |
46711810 | 2592 | * @flags: vm_flags |
1da177e4 LT |
2593 | */ |
2594 | struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) | |
2595 | { | |
2596 | int error; | |
2597 | struct file *file; | |
2598 | struct inode *inode; | |
2599 | struct dentry *dentry, *root; | |
2600 | struct qstr this; | |
2601 | ||
2602 | if (IS_ERR(shm_mnt)) | |
2603 | return (void *)shm_mnt; | |
2604 | ||
2605 | if (size < 0 || size > SHMEM_MAX_BYTES) | |
2606 | return ERR_PTR(-EINVAL); | |
2607 | ||
2608 | if (shmem_acct_size(flags, size)) | |
2609 | return ERR_PTR(-ENOMEM); | |
2610 | ||
2611 | error = -ENOMEM; | |
2612 | this.name = name; | |
2613 | this.len = strlen(name); | |
2614 | this.hash = 0; /* will go */ | |
2615 | root = shm_mnt->mnt_root; | |
2616 | dentry = d_alloc(root, &this); | |
2617 | if (!dentry) | |
2618 | goto put_memory; | |
2619 | ||
2620 | error = -ENFILE; | |
2621 | file = get_empty_filp(); | |
2622 | if (!file) | |
2623 | goto put_dentry; | |
2624 | ||
2625 | error = -ENOSPC; | |
2626 | inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); | |
2627 | if (!inode) | |
2628 | goto close_file; | |
2629 | ||
2630 | SHMEM_I(inode)->flags = flags & VM_ACCOUNT; | |
2631 | d_instantiate(dentry, inode); | |
2632 | inode->i_size = size; | |
2633 | inode->i_nlink = 0; /* It is unlinked */ | |
ce8d2cdf DH |
2634 | init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ, |
2635 | &shmem_file_operations); | |
1da177e4 LT |
2636 | return file; |
2637 | ||
2638 | close_file: | |
2639 | put_filp(file); | |
2640 | put_dentry: | |
2641 | dput(dentry); | |
2642 | put_memory: | |
2643 | shmem_unacct_size(flags, size); | |
2644 | return ERR_PTR(error); | |
2645 | } | |
2646 | ||
46711810 | 2647 | /** |
1da177e4 | 2648 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
2649 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
2650 | */ | |
2651 | int shmem_zero_setup(struct vm_area_struct *vma) | |
2652 | { | |
2653 | struct file *file; | |
2654 | loff_t size = vma->vm_end - vma->vm_start; | |
2655 | ||
2656 | file = shmem_file_setup("dev/zero", size, vma->vm_flags); | |
2657 | if (IS_ERR(file)) | |
2658 | return PTR_ERR(file); | |
2659 | ||
2660 | if (vma->vm_file) | |
2661 | fput(vma->vm_file); | |
2662 | vma->vm_file = file; | |
2663 | vma->vm_ops = &shmem_vm_ops; | |
2664 | return 0; | |
2665 | } |