]>
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. | |
6922c0c7 HD |
9 | * Copyright (C) 2002-2011 Hugh Dickins. |
10 | * Copyright (C) 2011 Google Inc. | |
0edd73b3 | 11 | * Copyright (C) 2002-2005 VERITAS Software Corporation. |
1da177e4 LT |
12 | * Copyright (C) 2004 Andi Kleen, SuSE Labs |
13 | * | |
14 | * Extended attribute support for tmpfs: | |
15 | * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net> | |
16 | * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com> | |
17 | * | |
853ac43a MM |
18 | * tiny-shmem: |
19 | * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com> | |
20 | * | |
1da177e4 LT |
21 | * This file is released under the GPL. |
22 | */ | |
23 | ||
853ac43a MM |
24 | #include <linux/fs.h> |
25 | #include <linux/init.h> | |
26 | #include <linux/vfs.h> | |
27 | #include <linux/mount.h> | |
caefba17 | 28 | #include <linux/pagemap.h> |
853ac43a MM |
29 | #include <linux/file.h> |
30 | #include <linux/mm.h> | |
b95f1b31 | 31 | #include <linux/export.h> |
853ac43a MM |
32 | #include <linux/swap.h> |
33 | ||
34 | static struct vfsmount *shm_mnt; | |
35 | ||
36 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
37 | /* |
38 | * This virtual memory filesystem is heavily based on the ramfs. It | |
39 | * extends ramfs by the ability to use swap and honor resource limits | |
40 | * which makes it a completely usable filesystem. | |
41 | */ | |
42 | ||
39f0247d | 43 | #include <linux/xattr.h> |
a5694255 | 44 | #include <linux/exportfs.h> |
1c7c474c | 45 | #include <linux/posix_acl.h> |
39f0247d | 46 | #include <linux/generic_acl.h> |
1da177e4 | 47 | #include <linux/mman.h> |
1da177e4 LT |
48 | #include <linux/string.h> |
49 | #include <linux/slab.h> | |
50 | #include <linux/backing-dev.h> | |
51 | #include <linux/shmem_fs.h> | |
1da177e4 | 52 | #include <linux/writeback.h> |
1da177e4 | 53 | #include <linux/blkdev.h> |
bda97eab | 54 | #include <linux/pagevec.h> |
41ffe5d5 | 55 | #include <linux/percpu_counter.h> |
708e3508 | 56 | #include <linux/splice.h> |
1da177e4 LT |
57 | #include <linux/security.h> |
58 | #include <linux/swapops.h> | |
59 | #include <linux/mempolicy.h> | |
60 | #include <linux/namei.h> | |
b00dc3ad | 61 | #include <linux/ctype.h> |
304dbdb7 | 62 | #include <linux/migrate.h> |
c1f60a5a | 63 | #include <linux/highmem.h> |
680d794b | 64 | #include <linux/seq_file.h> |
92562927 | 65 | #include <linux/magic.h> |
304dbdb7 | 66 | |
1da177e4 | 67 | #include <asm/uaccess.h> |
1da177e4 LT |
68 | #include <asm/pgtable.h> |
69 | ||
caefba17 | 70 | #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) |
1da177e4 LT |
71 | #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) |
72 | ||
1da177e4 LT |
73 | /* Pretend that each entry is of this size in directory's i_size */ |
74 | #define BOGO_DIRENT_SIZE 20 | |
75 | ||
69f07ec9 HD |
76 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
77 | #define SHORT_SYMLINK_LEN 128 | |
78 | ||
b09e0fa4 EP |
79 | struct shmem_xattr { |
80 | struct list_head list; /* anchored by shmem_inode_info->xattr_list */ | |
81 | char *name; /* xattr name */ | |
82 | size_t size; | |
83 | char value[0]; | |
84 | }; | |
85 | ||
285b2c4f | 86 | /* Flag allocation requirements to shmem_getpage */ |
1da177e4 | 87 | enum sgp_type { |
1da177e4 LT |
88 | SGP_READ, /* don't exceed i_size, don't allocate page */ |
89 | SGP_CACHE, /* don't exceed i_size, may allocate page */ | |
a0ee5ec5 | 90 | SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ |
1da177e4 LT |
91 | SGP_WRITE, /* may exceed i_size, may allocate page */ |
92 | }; | |
93 | ||
b76db735 | 94 | #ifdef CONFIG_TMPFS |
680d794b AM |
95 | static unsigned long shmem_default_max_blocks(void) |
96 | { | |
97 | return totalram_pages / 2; | |
98 | } | |
99 | ||
100 | static unsigned long shmem_default_max_inodes(void) | |
101 | { | |
102 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
103 | } | |
b76db735 | 104 | #endif |
680d794b | 105 | |
bde05d1c HD |
106 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
107 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
108 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 HD |
109 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
110 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type); | |
111 | ||
112 | static inline int shmem_getpage(struct inode *inode, pgoff_t index, | |
113 | struct page **pagep, enum sgp_type sgp, int *fault_type) | |
114 | { | |
115 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
116 | mapping_gfp_mask(inode->i_mapping), fault_type); | |
117 | } | |
1da177e4 | 118 | |
1da177e4 LT |
119 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
120 | { | |
121 | return sb->s_fs_info; | |
122 | } | |
123 | ||
124 | /* | |
125 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
126 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
127 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
128 | * consistent with the pre-accounting of private mappings ... | |
129 | */ | |
130 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
131 | { | |
0b0a0806 | 132 | return (flags & VM_NORESERVE) ? |
191c5424 | 133 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
134 | } |
135 | ||
136 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
137 | { | |
0b0a0806 | 138 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
139 | vm_unacct_memory(VM_ACCT(size)); |
140 | } | |
141 | ||
142 | /* | |
143 | * ... whereas tmpfs objects are accounted incrementally as | |
144 | * pages are allocated, in order to allow huge sparse files. | |
145 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, | |
146 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
147 | */ | |
148 | static inline int shmem_acct_block(unsigned long flags) | |
149 | { | |
0b0a0806 | 150 | return (flags & VM_NORESERVE) ? |
191c5424 | 151 | security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0; |
1da177e4 LT |
152 | } |
153 | ||
154 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
155 | { | |
0b0a0806 | 156 | if (flags & VM_NORESERVE) |
1da177e4 LT |
157 | vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); |
158 | } | |
159 | ||
759b9775 | 160 | static const struct super_operations shmem_ops; |
f5e54d6e | 161 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 162 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
163 | static const struct inode_operations shmem_inode_operations; |
164 | static const struct inode_operations shmem_dir_inode_operations; | |
165 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 166 | static const struct vm_operations_struct shmem_vm_ops; |
1da177e4 | 167 | |
6c231b7b | 168 | static struct backing_dev_info shmem_backing_dev_info __read_mostly = { |
1da177e4 | 169 | .ra_pages = 0, /* No readahead */ |
4f98a2fe | 170 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED, |
1da177e4 LT |
171 | }; |
172 | ||
173 | static LIST_HEAD(shmem_swaplist); | |
cb5f7b9a | 174 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 175 | |
5b04c689 PE |
176 | static int shmem_reserve_inode(struct super_block *sb) |
177 | { | |
178 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
179 | if (sbinfo->max_inodes) { | |
180 | spin_lock(&sbinfo->stat_lock); | |
181 | if (!sbinfo->free_inodes) { | |
182 | spin_unlock(&sbinfo->stat_lock); | |
183 | return -ENOSPC; | |
184 | } | |
185 | sbinfo->free_inodes--; | |
186 | spin_unlock(&sbinfo->stat_lock); | |
187 | } | |
188 | return 0; | |
189 | } | |
190 | ||
191 | static void shmem_free_inode(struct super_block *sb) | |
192 | { | |
193 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
194 | if (sbinfo->max_inodes) { | |
195 | spin_lock(&sbinfo->stat_lock); | |
196 | sbinfo->free_inodes++; | |
197 | spin_unlock(&sbinfo->stat_lock); | |
198 | } | |
199 | } | |
200 | ||
46711810 | 201 | /** |
41ffe5d5 | 202 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
203 | * @inode: inode to recalc |
204 | * | |
205 | * We have to calculate the free blocks since the mm can drop | |
206 | * undirtied hole pages behind our back. | |
207 | * | |
208 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
209 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
210 | * | |
211 | * It has to be called with the spinlock held. | |
212 | */ | |
213 | static void shmem_recalc_inode(struct inode *inode) | |
214 | { | |
215 | struct shmem_inode_info *info = SHMEM_I(inode); | |
216 | long freed; | |
217 | ||
218 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
219 | if (freed > 0) { | |
54af6042 HD |
220 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
221 | if (sbinfo->max_blocks) | |
222 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 223 | info->alloced -= freed; |
54af6042 | 224 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 225 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
226 | } |
227 | } | |
228 | ||
7a5d0fbb HD |
229 | /* |
230 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
231 | */ | |
232 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
233 | pgoff_t index, void *expected, void *replacement) | |
234 | { | |
235 | void **pslot; | |
236 | void *item = NULL; | |
237 | ||
238 | VM_BUG_ON(!expected); | |
239 | pslot = radix_tree_lookup_slot(&mapping->page_tree, index); | |
240 | if (pslot) | |
241 | item = radix_tree_deref_slot_protected(pslot, | |
242 | &mapping->tree_lock); | |
243 | if (item != expected) | |
244 | return -ENOENT; | |
245 | if (replacement) | |
246 | radix_tree_replace_slot(pslot, replacement); | |
247 | else | |
248 | radix_tree_delete(&mapping->page_tree, index); | |
249 | return 0; | |
250 | } | |
251 | ||
46f65ec1 HD |
252 | /* |
253 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
254 | */ | |
255 | static int shmem_add_to_page_cache(struct page *page, | |
256 | struct address_space *mapping, | |
257 | pgoff_t index, gfp_t gfp, void *expected) | |
258 | { | |
aa3b1895 | 259 | int error = 0; |
46f65ec1 HD |
260 | |
261 | VM_BUG_ON(!PageLocked(page)); | |
262 | VM_BUG_ON(!PageSwapBacked(page)); | |
263 | ||
46f65ec1 HD |
264 | if (!expected) |
265 | error = radix_tree_preload(gfp & GFP_RECLAIM_MASK); | |
266 | if (!error) { | |
267 | page_cache_get(page); | |
268 | page->mapping = mapping; | |
269 | page->index = index; | |
270 | ||
271 | spin_lock_irq(&mapping->tree_lock); | |
272 | if (!expected) | |
273 | error = radix_tree_insert(&mapping->page_tree, | |
274 | index, page); | |
275 | else | |
276 | error = shmem_radix_tree_replace(mapping, index, | |
277 | expected, page); | |
278 | if (!error) { | |
279 | mapping->nrpages++; | |
280 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
281 | __inc_zone_page_state(page, NR_SHMEM); | |
282 | spin_unlock_irq(&mapping->tree_lock); | |
283 | } else { | |
284 | page->mapping = NULL; | |
285 | spin_unlock_irq(&mapping->tree_lock); | |
286 | page_cache_release(page); | |
287 | } | |
288 | if (!expected) | |
289 | radix_tree_preload_end(); | |
290 | } | |
291 | if (error) | |
292 | mem_cgroup_uncharge_cache_page(page); | |
46f65ec1 HD |
293 | return error; |
294 | } | |
295 | ||
6922c0c7 HD |
296 | /* |
297 | * Like delete_from_page_cache, but substitutes swap for page. | |
298 | */ | |
299 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
300 | { | |
301 | struct address_space *mapping = page->mapping; | |
302 | int error; | |
303 | ||
304 | spin_lock_irq(&mapping->tree_lock); | |
305 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
306 | page->mapping = NULL; | |
307 | mapping->nrpages--; | |
308 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
309 | __dec_zone_page_state(page, NR_SHMEM); | |
310 | spin_unlock_irq(&mapping->tree_lock); | |
311 | page_cache_release(page); | |
312 | BUG_ON(error); | |
313 | } | |
314 | ||
7a5d0fbb HD |
315 | /* |
316 | * Like find_get_pages, but collecting swap entries as well as pages. | |
317 | */ | |
318 | static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping, | |
319 | pgoff_t start, unsigned int nr_pages, | |
320 | struct page **pages, pgoff_t *indices) | |
321 | { | |
322 | unsigned int i; | |
323 | unsigned int ret; | |
324 | unsigned int nr_found; | |
325 | ||
326 | rcu_read_lock(); | |
327 | restart: | |
328 | nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, | |
329 | (void ***)pages, indices, start, nr_pages); | |
330 | ret = 0; | |
331 | for (i = 0; i < nr_found; i++) { | |
332 | struct page *page; | |
333 | repeat: | |
334 | page = radix_tree_deref_slot((void **)pages[i]); | |
335 | if (unlikely(!page)) | |
336 | continue; | |
337 | if (radix_tree_exception(page)) { | |
8079b1c8 HD |
338 | if (radix_tree_deref_retry(page)) |
339 | goto restart; | |
340 | /* | |
341 | * Otherwise, we must be storing a swap entry | |
342 | * here as an exceptional entry: so return it | |
343 | * without attempting to raise page count. | |
344 | */ | |
345 | goto export; | |
7a5d0fbb HD |
346 | } |
347 | if (!page_cache_get_speculative(page)) | |
348 | goto repeat; | |
349 | ||
350 | /* Has the page moved? */ | |
351 | if (unlikely(page != *((void **)pages[i]))) { | |
352 | page_cache_release(page); | |
353 | goto repeat; | |
354 | } | |
355 | export: | |
356 | indices[ret] = indices[i]; | |
357 | pages[ret] = page; | |
358 | ret++; | |
359 | } | |
360 | if (unlikely(!ret && nr_found)) | |
361 | goto restart; | |
362 | rcu_read_unlock(); | |
363 | return ret; | |
364 | } | |
365 | ||
366 | /* | |
367 | * Remove swap entry from radix tree, free the swap and its page cache. | |
368 | */ | |
369 | static int shmem_free_swap(struct address_space *mapping, | |
370 | pgoff_t index, void *radswap) | |
371 | { | |
372 | int error; | |
373 | ||
374 | spin_lock_irq(&mapping->tree_lock); | |
375 | error = shmem_radix_tree_replace(mapping, index, radswap, NULL); | |
376 | spin_unlock_irq(&mapping->tree_lock); | |
377 | if (!error) | |
378 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
379 | return error; | |
380 | } | |
381 | ||
382 | /* | |
383 | * Pagevec may contain swap entries, so shuffle up pages before releasing. | |
384 | */ | |
24513264 | 385 | static void shmem_deswap_pagevec(struct pagevec *pvec) |
7a5d0fbb HD |
386 | { |
387 | int i, j; | |
388 | ||
389 | for (i = 0, j = 0; i < pagevec_count(pvec); i++) { | |
390 | struct page *page = pvec->pages[i]; | |
391 | if (!radix_tree_exceptional_entry(page)) | |
392 | pvec->pages[j++] = page; | |
393 | } | |
394 | pvec->nr = j; | |
24513264 HD |
395 | } |
396 | ||
397 | /* | |
398 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
399 | */ | |
400 | void shmem_unlock_mapping(struct address_space *mapping) | |
401 | { | |
402 | struct pagevec pvec; | |
403 | pgoff_t indices[PAGEVEC_SIZE]; | |
404 | pgoff_t index = 0; | |
405 | ||
406 | pagevec_init(&pvec, 0); | |
407 | /* | |
408 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
409 | */ | |
410 | while (!mapping_unevictable(mapping)) { | |
411 | /* | |
412 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
413 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
414 | */ | |
415 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, | |
416 | PAGEVEC_SIZE, pvec.pages, indices); | |
417 | if (!pvec.nr) | |
418 | break; | |
419 | index = indices[pvec.nr - 1] + 1; | |
420 | shmem_deswap_pagevec(&pvec); | |
421 | check_move_unevictable_pages(pvec.pages, pvec.nr); | |
422 | pagevec_release(&pvec); | |
423 | cond_resched(); | |
424 | } | |
7a5d0fbb HD |
425 | } |
426 | ||
427 | /* | |
428 | * Remove range of pages and swap entries from radix tree, and free them. | |
429 | */ | |
285b2c4f | 430 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
1da177e4 | 431 | { |
285b2c4f | 432 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 433 | struct shmem_inode_info *info = SHMEM_I(inode); |
285b2c4f | 434 | pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
bda97eab | 435 | unsigned partial = lstart & (PAGE_CACHE_SIZE - 1); |
285b2c4f | 436 | pgoff_t end = (lend >> PAGE_CACHE_SHIFT); |
bda97eab | 437 | struct pagevec pvec; |
7a5d0fbb HD |
438 | pgoff_t indices[PAGEVEC_SIZE]; |
439 | long nr_swaps_freed = 0; | |
285b2c4f | 440 | pgoff_t index; |
bda97eab HD |
441 | int i; |
442 | ||
443 | BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1)); | |
444 | ||
445 | pagevec_init(&pvec, 0); | |
446 | index = start; | |
7a5d0fbb HD |
447 | while (index <= end) { |
448 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, | |
449 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
450 | pvec.pages, indices); | |
451 | if (!pvec.nr) | |
452 | break; | |
bda97eab HD |
453 | mem_cgroup_uncharge_start(); |
454 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
455 | struct page *page = pvec.pages[i]; | |
456 | ||
7a5d0fbb | 457 | index = indices[i]; |
bda97eab HD |
458 | if (index > end) |
459 | break; | |
460 | ||
7a5d0fbb HD |
461 | if (radix_tree_exceptional_entry(page)) { |
462 | nr_swaps_freed += !shmem_free_swap(mapping, | |
463 | index, page); | |
bda97eab | 464 | continue; |
7a5d0fbb HD |
465 | } |
466 | ||
467 | if (!trylock_page(page)) | |
bda97eab | 468 | continue; |
7a5d0fbb HD |
469 | if (page->mapping == mapping) { |
470 | VM_BUG_ON(PageWriteback(page)); | |
471 | truncate_inode_page(mapping, page); | |
bda97eab | 472 | } |
bda97eab HD |
473 | unlock_page(page); |
474 | } | |
24513264 HD |
475 | shmem_deswap_pagevec(&pvec); |
476 | pagevec_release(&pvec); | |
bda97eab HD |
477 | mem_cgroup_uncharge_end(); |
478 | cond_resched(); | |
479 | index++; | |
480 | } | |
1da177e4 | 481 | |
bda97eab HD |
482 | if (partial) { |
483 | struct page *page = NULL; | |
484 | shmem_getpage(inode, start - 1, &page, SGP_READ, NULL); | |
485 | if (page) { | |
486 | zero_user_segment(page, partial, PAGE_CACHE_SIZE); | |
487 | set_page_dirty(page); | |
488 | unlock_page(page); | |
489 | page_cache_release(page); | |
490 | } | |
491 | } | |
492 | ||
493 | index = start; | |
494 | for ( ; ; ) { | |
495 | cond_resched(); | |
7a5d0fbb HD |
496 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, |
497 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
498 | pvec.pages, indices); | |
499 | if (!pvec.nr) { | |
bda97eab HD |
500 | if (index == start) |
501 | break; | |
502 | index = start; | |
503 | continue; | |
504 | } | |
7a5d0fbb | 505 | if (index == start && indices[0] > end) { |
24513264 HD |
506 | shmem_deswap_pagevec(&pvec); |
507 | pagevec_release(&pvec); | |
bda97eab HD |
508 | break; |
509 | } | |
510 | mem_cgroup_uncharge_start(); | |
511 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
512 | struct page *page = pvec.pages[i]; | |
513 | ||
7a5d0fbb | 514 | index = indices[i]; |
bda97eab HD |
515 | if (index > end) |
516 | break; | |
517 | ||
7a5d0fbb HD |
518 | if (radix_tree_exceptional_entry(page)) { |
519 | nr_swaps_freed += !shmem_free_swap(mapping, | |
520 | index, page); | |
521 | continue; | |
522 | } | |
523 | ||
bda97eab | 524 | lock_page(page); |
7a5d0fbb HD |
525 | if (page->mapping == mapping) { |
526 | VM_BUG_ON(PageWriteback(page)); | |
527 | truncate_inode_page(mapping, page); | |
528 | } | |
bda97eab HD |
529 | unlock_page(page); |
530 | } | |
24513264 HD |
531 | shmem_deswap_pagevec(&pvec); |
532 | pagevec_release(&pvec); | |
bda97eab HD |
533 | mem_cgroup_uncharge_end(); |
534 | index++; | |
535 | } | |
94c1e62d | 536 | |
1da177e4 | 537 | spin_lock(&info->lock); |
7a5d0fbb | 538 | info->swapped -= nr_swaps_freed; |
1da177e4 LT |
539 | shmem_recalc_inode(inode); |
540 | spin_unlock(&info->lock); | |
541 | ||
285b2c4f | 542 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1da177e4 | 543 | } |
94c1e62d | 544 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 545 | |
94c1e62d | 546 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 LT |
547 | { |
548 | struct inode *inode = dentry->d_inode; | |
1da177e4 LT |
549 | int error; |
550 | ||
db78b877 CH |
551 | error = inode_change_ok(inode, attr); |
552 | if (error) | |
553 | return error; | |
554 | ||
94c1e62d HD |
555 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
556 | loff_t oldsize = inode->i_size; | |
557 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 558 | |
94c1e62d HD |
559 | if (newsize != oldsize) { |
560 | i_size_write(inode, newsize); | |
561 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
562 | } | |
563 | if (newsize < oldsize) { | |
564 | loff_t holebegin = round_up(newsize, PAGE_SIZE); | |
565 | unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); | |
566 | shmem_truncate_range(inode, newsize, (loff_t)-1); | |
567 | /* unmap again to remove racily COWed private pages */ | |
568 | unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); | |
569 | } | |
1da177e4 LT |
570 | } |
571 | ||
db78b877 | 572 | setattr_copy(inode, attr); |
39f0247d | 573 | #ifdef CONFIG_TMPFS_POSIX_ACL |
db78b877 | 574 | if (attr->ia_valid & ATTR_MODE) |
1c7c474c | 575 | error = generic_acl_chmod(inode); |
39f0247d | 576 | #endif |
1da177e4 LT |
577 | return error; |
578 | } | |
579 | ||
1f895f75 | 580 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 581 | { |
1da177e4 | 582 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 583 | struct shmem_xattr *xattr, *nxattr; |
1da177e4 | 584 | |
3889e6e7 | 585 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
586 | shmem_unacct_size(info->flags, inode->i_size); |
587 | inode->i_size = 0; | |
3889e6e7 | 588 | shmem_truncate_range(inode, 0, (loff_t)-1); |
1da177e4 | 589 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 590 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 591 | list_del_init(&info->swaplist); |
cb5f7b9a | 592 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 593 | } |
69f07ec9 HD |
594 | } else |
595 | kfree(info->symlink); | |
b09e0fa4 EP |
596 | |
597 | list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) { | |
598 | kfree(xattr->name); | |
599 | kfree(xattr); | |
600 | } | |
0edd73b3 | 601 | BUG_ON(inode->i_blocks); |
5b04c689 | 602 | shmem_free_inode(inode->i_sb); |
dbd5768f | 603 | clear_inode(inode); |
1da177e4 LT |
604 | } |
605 | ||
46f65ec1 HD |
606 | /* |
607 | * If swap found in inode, free it and move page from swapcache to filecache. | |
608 | */ | |
41ffe5d5 | 609 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 610 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 611 | { |
285b2c4f | 612 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 613 | void *radswap; |
41ffe5d5 | 614 | pgoff_t index; |
bde05d1c HD |
615 | gfp_t gfp; |
616 | int error = 0; | |
1da177e4 | 617 | |
46f65ec1 | 618 | radswap = swp_to_radix_entry(swap); |
e504f3fd | 619 | index = radix_tree_locate_item(&mapping->page_tree, radswap); |
46f65ec1 | 620 | if (index == -1) |
285b2c4f | 621 | return 0; |
2e0e26c7 | 622 | |
1b1b32f2 HD |
623 | /* |
624 | * Move _head_ to start search for next from here. | |
1f895f75 | 625 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 626 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 627 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
628 | */ |
629 | if (shmem_swaplist.next != &info->swaplist) | |
630 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 631 | |
bde05d1c HD |
632 | gfp = mapping_gfp_mask(mapping); |
633 | if (shmem_should_replace_page(*pagep, gfp)) { | |
634 | mutex_unlock(&shmem_swaplist_mutex); | |
635 | error = shmem_replace_page(pagep, gfp, info, index); | |
636 | mutex_lock(&shmem_swaplist_mutex); | |
637 | /* | |
638 | * We needed to drop mutex to make that restrictive page | |
639 | * allocation; but the inode might already be freed by now, | |
640 | * and we cannot refer to inode or mapping or info to check. | |
641 | * However, we do hold page lock on the PageSwapCache page, | |
642 | * so can check if that still has our reference remaining. | |
643 | */ | |
644 | if (!page_swapcount(*pagep)) | |
645 | error = -ENOENT; | |
646 | } | |
647 | ||
d13d1443 | 648 | /* |
778dd893 HD |
649 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
650 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
651 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 652 | */ |
bde05d1c HD |
653 | if (!error) |
654 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
46f65ec1 | 655 | GFP_NOWAIT, radswap); |
48f170fb | 656 | if (error != -ENOMEM) { |
46f65ec1 HD |
657 | /* |
658 | * Truncation and eviction use free_swap_and_cache(), which | |
659 | * only does trylock page: if we raced, best clean up here. | |
660 | */ | |
bde05d1c HD |
661 | delete_from_swap_cache(*pagep); |
662 | set_page_dirty(*pagep); | |
46f65ec1 HD |
663 | if (!error) { |
664 | spin_lock(&info->lock); | |
665 | info->swapped--; | |
666 | spin_unlock(&info->lock); | |
667 | swap_free(swap); | |
668 | } | |
2e0e26c7 | 669 | error = 1; /* not an error, but entry was found */ |
1da177e4 | 670 | } |
2e0e26c7 | 671 | return error; |
1da177e4 LT |
672 | } |
673 | ||
674 | /* | |
46f65ec1 | 675 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 676 | */ |
41ffe5d5 | 677 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 678 | { |
41ffe5d5 | 679 | struct list_head *this, *next; |
1da177e4 LT |
680 | struct shmem_inode_info *info; |
681 | int found = 0; | |
bde05d1c HD |
682 | int error = 0; |
683 | ||
684 | /* | |
685 | * There's a faint possibility that swap page was replaced before | |
686 | * caller locked it: it will come back later with the right page. | |
687 | */ | |
688 | if (unlikely(!PageSwapCache(page))) | |
689 | goto out; | |
778dd893 HD |
690 | |
691 | /* | |
692 | * Charge page using GFP_KERNEL while we can wait, before taking | |
693 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
694 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 HD |
695 | */ |
696 | error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); | |
697 | if (error) | |
698 | goto out; | |
46f65ec1 | 699 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
1da177e4 | 700 | |
cb5f7b9a | 701 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
702 | list_for_each_safe(this, next, &shmem_swaplist) { |
703 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 704 | if (info->swapped) |
bde05d1c | 705 | found = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
706 | else |
707 | list_del_init(&info->swaplist); | |
cb5f7b9a | 708 | cond_resched(); |
2e0e26c7 | 709 | if (found) |
778dd893 | 710 | break; |
1da177e4 | 711 | } |
cb5f7b9a | 712 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 713 | |
778dd893 HD |
714 | if (found < 0) |
715 | error = found; | |
716 | out: | |
aaa46865 HD |
717 | unlock_page(page); |
718 | page_cache_release(page); | |
778dd893 | 719 | return error; |
1da177e4 LT |
720 | } |
721 | ||
722 | /* | |
723 | * Move the page from the page cache to the swap cache. | |
724 | */ | |
725 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
726 | { | |
727 | struct shmem_inode_info *info; | |
1da177e4 | 728 | struct address_space *mapping; |
1da177e4 | 729 | struct inode *inode; |
6922c0c7 HD |
730 | swp_entry_t swap; |
731 | pgoff_t index; | |
1da177e4 LT |
732 | |
733 | BUG_ON(!PageLocked(page)); | |
1da177e4 LT |
734 | mapping = page->mapping; |
735 | index = page->index; | |
736 | inode = mapping->host; | |
737 | info = SHMEM_I(inode); | |
738 | if (info->flags & VM_LOCKED) | |
739 | goto redirty; | |
d9fe526a | 740 | if (!total_swap_pages) |
1da177e4 LT |
741 | goto redirty; |
742 | ||
d9fe526a HD |
743 | /* |
744 | * shmem_backing_dev_info's capabilities prevent regular writeback or | |
745 | * sync from ever calling shmem_writepage; but a stacking filesystem | |
48f170fb | 746 | * might use ->writepage of its underlying filesystem, in which case |
d9fe526a | 747 | * tmpfs should write out to swap only in response to memory pressure, |
48f170fb | 748 | * and not for the writeback threads or sync. |
d9fe526a | 749 | */ |
48f170fb HD |
750 | if (!wbc->for_reclaim) { |
751 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
752 | goto redirty; | |
753 | } | |
754 | swap = get_swap_page(); | |
755 | if (!swap.val) | |
756 | goto redirty; | |
d9fe526a | 757 | |
b1dea800 HD |
758 | /* |
759 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
760 | * if it's not already there. Do it now before the page is |
761 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 762 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
763 | * we've incremented swapped, because shmem_unuse_inode() will |
764 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 765 | */ |
48f170fb HD |
766 | mutex_lock(&shmem_swaplist_mutex); |
767 | if (list_empty(&info->swaplist)) | |
768 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 769 | |
48f170fb | 770 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
aaa46865 | 771 | swap_shmem_alloc(swap); |
6922c0c7 HD |
772 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); |
773 | ||
774 | spin_lock(&info->lock); | |
775 | info->swapped++; | |
776 | shmem_recalc_inode(inode); | |
826267cf | 777 | spin_unlock(&info->lock); |
6922c0c7 HD |
778 | |
779 | mutex_unlock(&shmem_swaplist_mutex); | |
d9fe526a | 780 | BUG_ON(page_mapped(page)); |
9fab5619 | 781 | swap_writepage(page, wbc); |
1da177e4 LT |
782 | return 0; |
783 | } | |
784 | ||
6922c0c7 | 785 | mutex_unlock(&shmem_swaplist_mutex); |
cb4b86ba | 786 | swapcache_free(swap, NULL); |
1da177e4 LT |
787 | redirty: |
788 | set_page_dirty(page); | |
d9fe526a HD |
789 | if (wbc->for_reclaim) |
790 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
791 | unlock_page(page); | |
792 | return 0; | |
1da177e4 LT |
793 | } |
794 | ||
795 | #ifdef CONFIG_NUMA | |
680d794b | 796 | #ifdef CONFIG_TMPFS |
71fe804b | 797 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 798 | { |
095f1fc4 | 799 | char buffer[64]; |
680d794b | 800 | |
71fe804b | 801 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 802 | return; /* show nothing */ |
680d794b | 803 | |
71fe804b | 804 | mpol_to_str(buffer, sizeof(buffer), mpol, 1); |
095f1fc4 LS |
805 | |
806 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 807 | } |
71fe804b LS |
808 | |
809 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
810 | { | |
811 | struct mempolicy *mpol = NULL; | |
812 | if (sbinfo->mpol) { | |
813 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
814 | mpol = sbinfo->mpol; | |
815 | mpol_get(mpol); | |
816 | spin_unlock(&sbinfo->stat_lock); | |
817 | } | |
818 | return mpol; | |
819 | } | |
680d794b AM |
820 | #endif /* CONFIG_TMPFS */ |
821 | ||
41ffe5d5 HD |
822 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
823 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 824 | { |
52cd3b07 | 825 | struct mempolicy mpol, *spol; |
1da177e4 LT |
826 | struct vm_area_struct pvma; |
827 | ||
52cd3b07 | 828 | spol = mpol_cond_copy(&mpol, |
41ffe5d5 | 829 | mpol_shared_policy_lookup(&info->policy, index)); |
52cd3b07 | 830 | |
1da177e4 | 831 | /* Create a pseudo vma that just contains the policy */ |
c4cc6d07 | 832 | pvma.vm_start = 0; |
41ffe5d5 | 833 | pvma.vm_pgoff = index; |
c4cc6d07 | 834 | pvma.vm_ops = NULL; |
52cd3b07 | 835 | pvma.vm_policy = spol; |
41ffe5d5 | 836 | return swapin_readahead(swap, gfp, &pvma, 0); |
1da177e4 LT |
837 | } |
838 | ||
02098fea | 839 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 840 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
841 | { |
842 | struct vm_area_struct pvma; | |
1da177e4 | 843 | |
c4cc6d07 HD |
844 | /* Create a pseudo vma that just contains the policy */ |
845 | pvma.vm_start = 0; | |
41ffe5d5 | 846 | pvma.vm_pgoff = index; |
c4cc6d07 | 847 | pvma.vm_ops = NULL; |
41ffe5d5 | 848 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
52cd3b07 LS |
849 | |
850 | /* | |
851 | * alloc_page_vma() will drop the shared policy reference | |
852 | */ | |
853 | return alloc_page_vma(gfp, &pvma, 0); | |
1da177e4 | 854 | } |
680d794b AM |
855 | #else /* !CONFIG_NUMA */ |
856 | #ifdef CONFIG_TMPFS | |
41ffe5d5 | 857 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b AM |
858 | { |
859 | } | |
860 | #endif /* CONFIG_TMPFS */ | |
861 | ||
41ffe5d5 HD |
862 | static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
863 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 864 | { |
41ffe5d5 | 865 | return swapin_readahead(swap, gfp, NULL, 0); |
1da177e4 LT |
866 | } |
867 | ||
02098fea | 868 | static inline struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 869 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 | 870 | { |
e84e2e13 | 871 | return alloc_page(gfp); |
1da177e4 | 872 | } |
680d794b | 873 | #endif /* CONFIG_NUMA */ |
1da177e4 | 874 | |
71fe804b LS |
875 | #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) |
876 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
877 | { | |
878 | return NULL; | |
879 | } | |
880 | #endif | |
881 | ||
bde05d1c HD |
882 | /* |
883 | * When a page is moved from swapcache to shmem filecache (either by the | |
884 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
885 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
886 | * ignorance of the mapping it belongs to. If that mapping has special | |
887 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
888 | * we may need to copy to a suitable page before moving to filecache. | |
889 | * | |
890 | * In a future release, this may well be extended to respect cpuset and | |
891 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
892 | * but for now it is a simple matter of zone. | |
893 | */ | |
894 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
895 | { | |
896 | return page_zonenum(page) > gfp_zone(gfp); | |
897 | } | |
898 | ||
899 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
900 | struct shmem_inode_info *info, pgoff_t index) | |
901 | { | |
902 | struct page *oldpage, *newpage; | |
903 | struct address_space *swap_mapping; | |
904 | pgoff_t swap_index; | |
905 | int error; | |
906 | ||
907 | oldpage = *pagep; | |
908 | swap_index = page_private(oldpage); | |
909 | swap_mapping = page_mapping(oldpage); | |
910 | ||
911 | /* | |
912 | * We have arrived here because our zones are constrained, so don't | |
913 | * limit chance of success by further cpuset and node constraints. | |
914 | */ | |
915 | gfp &= ~GFP_CONSTRAINT_MASK; | |
916 | newpage = shmem_alloc_page(gfp, info, index); | |
917 | if (!newpage) | |
918 | return -ENOMEM; | |
919 | VM_BUG_ON(shmem_should_replace_page(newpage, gfp)); | |
920 | ||
921 | *pagep = newpage; | |
922 | page_cache_get(newpage); | |
923 | copy_highpage(newpage, oldpage); | |
924 | ||
925 | VM_BUG_ON(!PageLocked(oldpage)); | |
926 | __set_page_locked(newpage); | |
927 | VM_BUG_ON(!PageUptodate(oldpage)); | |
928 | SetPageUptodate(newpage); | |
929 | VM_BUG_ON(!PageSwapBacked(oldpage)); | |
930 | SetPageSwapBacked(newpage); | |
931 | VM_BUG_ON(!swap_index); | |
932 | set_page_private(newpage, swap_index); | |
933 | VM_BUG_ON(!PageSwapCache(oldpage)); | |
934 | SetPageSwapCache(newpage); | |
935 | ||
936 | /* | |
937 | * Our caller will very soon move newpage out of swapcache, but it's | |
938 | * a nice clean interface for us to replace oldpage by newpage there. | |
939 | */ | |
940 | spin_lock_irq(&swap_mapping->tree_lock); | |
941 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
942 | newpage); | |
943 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
944 | __dec_zone_page_state(oldpage, NR_FILE_PAGES); | |
945 | spin_unlock_irq(&swap_mapping->tree_lock); | |
946 | BUG_ON(error); | |
947 | ||
948 | mem_cgroup_replace_page_cache(oldpage, newpage); | |
949 | lru_cache_add_anon(newpage); | |
950 | ||
951 | ClearPageSwapCache(oldpage); | |
952 | set_page_private(oldpage, 0); | |
953 | ||
954 | unlock_page(oldpage); | |
955 | page_cache_release(oldpage); | |
956 | page_cache_release(oldpage); | |
957 | return 0; | |
958 | } | |
959 | ||
1da177e4 | 960 | /* |
68da9f05 | 961 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
962 | * |
963 | * If we allocate a new one we do not mark it dirty. That's up to the | |
964 | * vm. If we swap it in we mark it dirty since we also free the swap | |
965 | * entry since a page cannot live in both the swap and page cache | |
966 | */ | |
41ffe5d5 | 967 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
68da9f05 | 968 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type) |
1da177e4 LT |
969 | { |
970 | struct address_space *mapping = inode->i_mapping; | |
54af6042 | 971 | struct shmem_inode_info *info; |
1da177e4 | 972 | struct shmem_sb_info *sbinfo; |
27ab7006 | 973 | struct page *page; |
1da177e4 LT |
974 | swp_entry_t swap; |
975 | int error; | |
54af6042 | 976 | int once = 0; |
1da177e4 | 977 | |
41ffe5d5 | 978 | if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT)) |
1da177e4 | 979 | return -EFBIG; |
1da177e4 | 980 | repeat: |
54af6042 | 981 | swap.val = 0; |
41ffe5d5 | 982 | page = find_lock_page(mapping, index); |
54af6042 HD |
983 | if (radix_tree_exceptional_entry(page)) { |
984 | swap = radix_to_swp_entry(page); | |
985 | page = NULL; | |
986 | } | |
987 | ||
988 | if (sgp != SGP_WRITE && | |
989 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { | |
990 | error = -EINVAL; | |
991 | goto failed; | |
992 | } | |
993 | ||
994 | if (page || (sgp == SGP_READ && !swap.val)) { | |
b409f9fc | 995 | /* |
27ab7006 HD |
996 | * Once we can get the page lock, it must be uptodate: |
997 | * if there were an error in reading back from swap, | |
998 | * the page would not be inserted into the filecache. | |
b409f9fc | 999 | */ |
54af6042 HD |
1000 | BUG_ON(page && !PageUptodate(page)); |
1001 | *pagep = page; | |
1002 | return 0; | |
27ab7006 HD |
1003 | } |
1004 | ||
1005 | /* | |
54af6042 HD |
1006 | * Fast cache lookup did not find it: |
1007 | * bring it back from swap or allocate. | |
27ab7006 | 1008 | */ |
54af6042 HD |
1009 | info = SHMEM_I(inode); |
1010 | sbinfo = SHMEM_SB(inode->i_sb); | |
1da177e4 | 1011 | |
1da177e4 LT |
1012 | if (swap.val) { |
1013 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1014 | page = lookup_swap_cache(swap); |
1015 | if (!page) { | |
1da177e4 | 1016 | /* here we actually do the io */ |
68da9f05 HD |
1017 | if (fault_type) |
1018 | *fault_type |= VM_FAULT_MAJOR; | |
41ffe5d5 | 1019 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1020 | if (!page) { |
54af6042 HD |
1021 | error = -ENOMEM; |
1022 | goto failed; | |
1da177e4 | 1023 | } |
1da177e4 LT |
1024 | } |
1025 | ||
1026 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1027 | lock_page(page); |
bde05d1c HD |
1028 | if (!PageSwapCache(page) || page->mapping) { |
1029 | error = -EEXIST; /* try again */ | |
1030 | goto failed; | |
1031 | } | |
27ab7006 | 1032 | if (!PageUptodate(page)) { |
1da177e4 | 1033 | error = -EIO; |
54af6042 | 1034 | goto failed; |
1da177e4 | 1035 | } |
54af6042 HD |
1036 | wait_on_page_writeback(page); |
1037 | ||
bde05d1c HD |
1038 | if (shmem_should_replace_page(page, gfp)) { |
1039 | error = shmem_replace_page(&page, gfp, info, index); | |
1040 | if (error) | |
1041 | goto failed; | |
1da177e4 | 1042 | } |
27ab7006 | 1043 | |
aa3b1895 HD |
1044 | error = mem_cgroup_cache_charge(page, current->mm, |
1045 | gfp & GFP_RECLAIM_MASK); | |
1046 | if (!error) | |
1047 | error = shmem_add_to_page_cache(page, mapping, index, | |
1048 | gfp, swp_to_radix_entry(swap)); | |
54af6042 HD |
1049 | if (error) |
1050 | goto failed; | |
1051 | ||
1052 | spin_lock(&info->lock); | |
285b2c4f | 1053 | info->swapped--; |
54af6042 | 1054 | shmem_recalc_inode(inode); |
27ab7006 | 1055 | spin_unlock(&info->lock); |
54af6042 HD |
1056 | |
1057 | delete_from_swap_cache(page); | |
27ab7006 HD |
1058 | set_page_dirty(page); |
1059 | swap_free(swap); | |
1060 | ||
54af6042 HD |
1061 | } else { |
1062 | if (shmem_acct_block(info->flags)) { | |
1063 | error = -ENOSPC; | |
1064 | goto failed; | |
1da177e4 | 1065 | } |
0edd73b3 | 1066 | if (sbinfo->max_blocks) { |
fc5da22a | 1067 | if (percpu_counter_compare(&sbinfo->used_blocks, |
54af6042 HD |
1068 | sbinfo->max_blocks) >= 0) { |
1069 | error = -ENOSPC; | |
1070 | goto unacct; | |
1071 | } | |
7e496299 | 1072 | percpu_counter_inc(&sbinfo->used_blocks); |
54af6042 | 1073 | } |
1da177e4 | 1074 | |
54af6042 HD |
1075 | page = shmem_alloc_page(gfp, info, index); |
1076 | if (!page) { | |
1077 | error = -ENOMEM; | |
1078 | goto decused; | |
1da177e4 LT |
1079 | } |
1080 | ||
54af6042 HD |
1081 | SetPageSwapBacked(page); |
1082 | __set_page_locked(page); | |
aa3b1895 HD |
1083 | error = mem_cgroup_cache_charge(page, current->mm, |
1084 | gfp & GFP_RECLAIM_MASK); | |
1085 | if (!error) | |
1086 | error = shmem_add_to_page_cache(page, mapping, index, | |
1087 | gfp, NULL); | |
54af6042 HD |
1088 | if (error) |
1089 | goto decused; | |
1090 | lru_cache_add_anon(page); | |
1091 | ||
1092 | spin_lock(&info->lock); | |
1da177e4 | 1093 | info->alloced++; |
54af6042 HD |
1094 | inode->i_blocks += BLOCKS_PER_PAGE; |
1095 | shmem_recalc_inode(inode); | |
1da177e4 | 1096 | spin_unlock(&info->lock); |
54af6042 | 1097 | |
ec9516fb HD |
1098 | /* |
1099 | * Let SGP_WRITE caller clear ends if write does not fill page | |
1100 | */ | |
1101 | if (sgp != SGP_WRITE) { | |
1102 | clear_highpage(page); | |
1103 | flush_dcache_page(page); | |
1104 | SetPageUptodate(page); | |
1105 | } | |
a0ee5ec5 | 1106 | if (sgp == SGP_DIRTY) |
27ab7006 | 1107 | set_page_dirty(page); |
1da177e4 | 1108 | } |
bde05d1c | 1109 | |
54af6042 HD |
1110 | /* Perhaps the file has been truncated since we checked */ |
1111 | if (sgp != SGP_WRITE && | |
1112 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { | |
1113 | error = -EINVAL; | |
1114 | goto trunc; | |
e83c32e8 | 1115 | } |
54af6042 HD |
1116 | *pagep = page; |
1117 | return 0; | |
1da177e4 | 1118 | |
59a16ead | 1119 | /* |
54af6042 | 1120 | * Error recovery. |
59a16ead | 1121 | */ |
54af6042 HD |
1122 | trunc: |
1123 | ClearPageDirty(page); | |
1124 | delete_from_page_cache(page); | |
1125 | spin_lock(&info->lock); | |
1126 | info->alloced--; | |
1127 | inode->i_blocks -= BLOCKS_PER_PAGE; | |
59a16ead | 1128 | spin_unlock(&info->lock); |
54af6042 HD |
1129 | decused: |
1130 | if (sbinfo->max_blocks) | |
1131 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
1132 | unacct: | |
1133 | shmem_unacct_blocks(info->flags, 1); | |
1134 | failed: | |
1135 | if (swap.val && error != -EINVAL) { | |
1136 | struct page *test = find_get_page(mapping, index); | |
1137 | if (test && !radix_tree_exceptional_entry(test)) | |
1138 | page_cache_release(test); | |
1139 | /* Have another try if the entry has changed */ | |
1140 | if (test != swp_to_radix_entry(swap)) | |
1141 | error = -EEXIST; | |
1142 | } | |
27ab7006 | 1143 | if (page) { |
54af6042 | 1144 | unlock_page(page); |
27ab7006 | 1145 | page_cache_release(page); |
54af6042 HD |
1146 | } |
1147 | if (error == -ENOSPC && !once++) { | |
1148 | info = SHMEM_I(inode); | |
1149 | spin_lock(&info->lock); | |
1150 | shmem_recalc_inode(inode); | |
1151 | spin_unlock(&info->lock); | |
27ab7006 | 1152 | goto repeat; |
ff36b801 | 1153 | } |
54af6042 HD |
1154 | if (error == -EEXIST) |
1155 | goto repeat; | |
1156 | return error; | |
1da177e4 LT |
1157 | } |
1158 | ||
d0217ac0 | 1159 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1160 | { |
d3ac7f89 | 1161 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 | 1162 | int error; |
68da9f05 | 1163 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1164 | |
27d54b39 | 1165 | error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); |
d0217ac0 NP |
1166 | if (error) |
1167 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1168 | |
456f998e YH |
1169 | if (ret & VM_FAULT_MAJOR) { |
1170 | count_vm_event(PGMAJFAULT); | |
1171 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); | |
1172 | } | |
68da9f05 | 1173 | return ret; |
1da177e4 LT |
1174 | } |
1175 | ||
1da177e4 | 1176 | #ifdef CONFIG_NUMA |
41ffe5d5 | 1177 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 1178 | { |
41ffe5d5 HD |
1179 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1180 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); | |
1da177e4 LT |
1181 | } |
1182 | ||
d8dc74f2 AB |
1183 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
1184 | unsigned long addr) | |
1da177e4 | 1185 | { |
41ffe5d5 HD |
1186 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1187 | pgoff_t index; | |
1da177e4 | 1188 | |
41ffe5d5 HD |
1189 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
1190 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
1191 | } |
1192 | #endif | |
1193 | ||
1194 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
1195 | { | |
d3ac7f89 | 1196 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
1197 | struct shmem_inode_info *info = SHMEM_I(inode); |
1198 | int retval = -ENOMEM; | |
1199 | ||
1200 | spin_lock(&info->lock); | |
1201 | if (lock && !(info->flags & VM_LOCKED)) { | |
1202 | if (!user_shm_lock(inode->i_size, user)) | |
1203 | goto out_nomem; | |
1204 | info->flags |= VM_LOCKED; | |
89e004ea | 1205 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
1206 | } |
1207 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
1208 | user_shm_unlock(inode->i_size, user); | |
1209 | info->flags &= ~VM_LOCKED; | |
89e004ea | 1210 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
1211 | } |
1212 | retval = 0; | |
89e004ea | 1213 | |
1da177e4 LT |
1214 | out_nomem: |
1215 | spin_unlock(&info->lock); | |
1216 | return retval; | |
1217 | } | |
1218 | ||
9b83a6a8 | 1219 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
1220 | { |
1221 | file_accessed(file); | |
1222 | vma->vm_ops = &shmem_vm_ops; | |
d0217ac0 | 1223 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
1224 | return 0; |
1225 | } | |
1226 | ||
454abafe | 1227 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 1228 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
1229 | { |
1230 | struct inode *inode; | |
1231 | struct shmem_inode_info *info; | |
1232 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
1233 | ||
5b04c689 PE |
1234 | if (shmem_reserve_inode(sb)) |
1235 | return NULL; | |
1da177e4 LT |
1236 | |
1237 | inode = new_inode(sb); | |
1238 | if (inode) { | |
85fe4025 | 1239 | inode->i_ino = get_next_ino(); |
454abafe | 1240 | inode_init_owner(inode, dir, mode); |
1da177e4 | 1241 | inode->i_blocks = 0; |
1da177e4 LT |
1242 | inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; |
1243 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
91828a40 | 1244 | inode->i_generation = get_seconds(); |
1da177e4 LT |
1245 | info = SHMEM_I(inode); |
1246 | memset(info, 0, (char *)inode - (char *)info); | |
1247 | spin_lock_init(&info->lock); | |
0b0a0806 | 1248 | info->flags = flags & VM_NORESERVE; |
1da177e4 | 1249 | INIT_LIST_HEAD(&info->swaplist); |
b09e0fa4 | 1250 | INIT_LIST_HEAD(&info->xattr_list); |
72c04902 | 1251 | cache_no_acl(inode); |
1da177e4 LT |
1252 | |
1253 | switch (mode & S_IFMT) { | |
1254 | default: | |
39f0247d | 1255 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
1256 | init_special_inode(inode, mode, dev); |
1257 | break; | |
1258 | case S_IFREG: | |
14fcc23f | 1259 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
1260 | inode->i_op = &shmem_inode_operations; |
1261 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
1262 | mpol_shared_policy_init(&info->policy, |
1263 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
1264 | break; |
1265 | case S_IFDIR: | |
d8c76e6f | 1266 | inc_nlink(inode); |
1da177e4 LT |
1267 | /* Some things misbehave if size == 0 on a directory */ |
1268 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
1269 | inode->i_op = &shmem_dir_inode_operations; | |
1270 | inode->i_fop = &simple_dir_operations; | |
1271 | break; | |
1272 | case S_IFLNK: | |
1273 | /* | |
1274 | * Must not load anything in the rbtree, | |
1275 | * mpol_free_shared_policy will not be called. | |
1276 | */ | |
71fe804b | 1277 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
1278 | break; |
1279 | } | |
5b04c689 PE |
1280 | } else |
1281 | shmem_free_inode(sb); | |
1da177e4 LT |
1282 | return inode; |
1283 | } | |
1284 | ||
1285 | #ifdef CONFIG_TMPFS | |
92e1d5be | 1286 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 1287 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 1288 | |
6d9d88d0 JS |
1289 | #ifdef CONFIG_TMPFS_XATTR |
1290 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
1291 | #else | |
1292 | #define shmem_initxattrs NULL | |
1293 | #endif | |
1294 | ||
1da177e4 | 1295 | static int |
800d15a5 NP |
1296 | shmem_write_begin(struct file *file, struct address_space *mapping, |
1297 | loff_t pos, unsigned len, unsigned flags, | |
1298 | struct page **pagep, void **fsdata) | |
1da177e4 | 1299 | { |
800d15a5 NP |
1300 | struct inode *inode = mapping->host; |
1301 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
800d15a5 NP |
1302 | return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); |
1303 | } | |
1304 | ||
1305 | static int | |
1306 | shmem_write_end(struct file *file, struct address_space *mapping, | |
1307 | loff_t pos, unsigned len, unsigned copied, | |
1308 | struct page *page, void *fsdata) | |
1309 | { | |
1310 | struct inode *inode = mapping->host; | |
1311 | ||
d3602444 HD |
1312 | if (pos + copied > inode->i_size) |
1313 | i_size_write(inode, pos + copied); | |
1314 | ||
ec9516fb HD |
1315 | if (!PageUptodate(page)) { |
1316 | if (copied < PAGE_CACHE_SIZE) { | |
1317 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
1318 | zero_user_segments(page, 0, from, | |
1319 | from + copied, PAGE_CACHE_SIZE); | |
1320 | } | |
1321 | SetPageUptodate(page); | |
1322 | } | |
800d15a5 | 1323 | set_page_dirty(page); |
6746aff7 | 1324 | unlock_page(page); |
800d15a5 NP |
1325 | page_cache_release(page); |
1326 | ||
800d15a5 | 1327 | return copied; |
1da177e4 LT |
1328 | } |
1329 | ||
1da177e4 LT |
1330 | static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) |
1331 | { | |
d3ac7f89 | 1332 | struct inode *inode = filp->f_path.dentry->d_inode; |
1da177e4 | 1333 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
1334 | pgoff_t index; |
1335 | unsigned long offset; | |
a0ee5ec5 HD |
1336 | enum sgp_type sgp = SGP_READ; |
1337 | ||
1338 | /* | |
1339 | * Might this read be for a stacking filesystem? Then when reading | |
1340 | * holes of a sparse file, we actually need to allocate those pages, | |
1341 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
1342 | */ | |
1343 | if (segment_eq(get_fs(), KERNEL_DS)) | |
1344 | sgp = SGP_DIRTY; | |
1da177e4 LT |
1345 | |
1346 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1347 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1348 | ||
1349 | for (;;) { | |
1350 | struct page *page = NULL; | |
41ffe5d5 HD |
1351 | pgoff_t end_index; |
1352 | unsigned long nr, ret; | |
1da177e4 LT |
1353 | loff_t i_size = i_size_read(inode); |
1354 | ||
1355 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1356 | if (index > end_index) | |
1357 | break; | |
1358 | if (index == end_index) { | |
1359 | nr = i_size & ~PAGE_CACHE_MASK; | |
1360 | if (nr <= offset) | |
1361 | break; | |
1362 | } | |
1363 | ||
a0ee5ec5 | 1364 | desc->error = shmem_getpage(inode, index, &page, sgp, NULL); |
1da177e4 LT |
1365 | if (desc->error) { |
1366 | if (desc->error == -EINVAL) | |
1367 | desc->error = 0; | |
1368 | break; | |
1369 | } | |
d3602444 HD |
1370 | if (page) |
1371 | unlock_page(page); | |
1da177e4 LT |
1372 | |
1373 | /* | |
1374 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 1375 | * are called without i_mutex protection against truncate |
1da177e4 LT |
1376 | */ |
1377 | nr = PAGE_CACHE_SIZE; | |
1378 | i_size = i_size_read(inode); | |
1379 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1380 | if (index == end_index) { | |
1381 | nr = i_size & ~PAGE_CACHE_MASK; | |
1382 | if (nr <= offset) { | |
1383 | if (page) | |
1384 | page_cache_release(page); | |
1385 | break; | |
1386 | } | |
1387 | } | |
1388 | nr -= offset; | |
1389 | ||
1390 | if (page) { | |
1391 | /* | |
1392 | * If users can be writing to this page using arbitrary | |
1393 | * virtual addresses, take care about potential aliasing | |
1394 | * before reading the page on the kernel side. | |
1395 | */ | |
1396 | if (mapping_writably_mapped(mapping)) | |
1397 | flush_dcache_page(page); | |
1398 | /* | |
1399 | * Mark the page accessed if we read the beginning. | |
1400 | */ | |
1401 | if (!offset) | |
1402 | mark_page_accessed(page); | |
b5810039 | 1403 | } else { |
1da177e4 | 1404 | page = ZERO_PAGE(0); |
b5810039 NP |
1405 | page_cache_get(page); |
1406 | } | |
1da177e4 LT |
1407 | |
1408 | /* | |
1409 | * Ok, we have the page, and it's up-to-date, so | |
1410 | * now we can copy it to user space... | |
1411 | * | |
1412 | * The actor routine returns how many bytes were actually used.. | |
1413 | * NOTE! This may not be the same as how much of a user buffer | |
1414 | * we filled up (we may be padding etc), so we can only update | |
1415 | * "pos" here (the actor routine has to update the user buffer | |
1416 | * pointers and the remaining count). | |
1417 | */ | |
1418 | ret = actor(desc, page, offset, nr); | |
1419 | offset += ret; | |
1420 | index += offset >> PAGE_CACHE_SHIFT; | |
1421 | offset &= ~PAGE_CACHE_MASK; | |
1422 | ||
1423 | page_cache_release(page); | |
1424 | if (ret != nr || !desc->count) | |
1425 | break; | |
1426 | ||
1427 | cond_resched(); | |
1428 | } | |
1429 | ||
1430 | *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; | |
1431 | file_accessed(filp); | |
1432 | } | |
1433 | ||
bcd78e49 HD |
1434 | static ssize_t shmem_file_aio_read(struct kiocb *iocb, |
1435 | const struct iovec *iov, unsigned long nr_segs, loff_t pos) | |
1436 | { | |
1437 | struct file *filp = iocb->ki_filp; | |
1438 | ssize_t retval; | |
1439 | unsigned long seg; | |
1440 | size_t count; | |
1441 | loff_t *ppos = &iocb->ki_pos; | |
1442 | ||
1443 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); | |
1444 | if (retval) | |
1445 | return retval; | |
1446 | ||
1447 | for (seg = 0; seg < nr_segs; seg++) { | |
1448 | read_descriptor_t desc; | |
1449 | ||
1450 | desc.written = 0; | |
1451 | desc.arg.buf = iov[seg].iov_base; | |
1452 | desc.count = iov[seg].iov_len; | |
1453 | if (desc.count == 0) | |
1454 | continue; | |
1455 | desc.error = 0; | |
1456 | do_shmem_file_read(filp, ppos, &desc, file_read_actor); | |
1457 | retval += desc.written; | |
1458 | if (desc.error) { | |
1459 | retval = retval ?: desc.error; | |
1460 | break; | |
1461 | } | |
1462 | if (desc.count > 0) | |
1463 | break; | |
1464 | } | |
1465 | return retval; | |
1da177e4 LT |
1466 | } |
1467 | ||
708e3508 HD |
1468 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, |
1469 | struct pipe_inode_info *pipe, size_t len, | |
1470 | unsigned int flags) | |
1471 | { | |
1472 | struct address_space *mapping = in->f_mapping; | |
71f0e07a | 1473 | struct inode *inode = mapping->host; |
708e3508 HD |
1474 | unsigned int loff, nr_pages, req_pages; |
1475 | struct page *pages[PIPE_DEF_BUFFERS]; | |
1476 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
1477 | struct page *page; | |
1478 | pgoff_t index, end_index; | |
1479 | loff_t isize, left; | |
1480 | int error, page_nr; | |
1481 | struct splice_pipe_desc spd = { | |
1482 | .pages = pages, | |
1483 | .partial = partial, | |
1484 | .flags = flags, | |
1485 | .ops = &page_cache_pipe_buf_ops, | |
1486 | .spd_release = spd_release_page, | |
1487 | }; | |
1488 | ||
71f0e07a | 1489 | isize = i_size_read(inode); |
708e3508 HD |
1490 | if (unlikely(*ppos >= isize)) |
1491 | return 0; | |
1492 | ||
1493 | left = isize - *ppos; | |
1494 | if (unlikely(left < len)) | |
1495 | len = left; | |
1496 | ||
1497 | if (splice_grow_spd(pipe, &spd)) | |
1498 | return -ENOMEM; | |
1499 | ||
1500 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1501 | loff = *ppos & ~PAGE_CACHE_MASK; | |
1502 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
1503 | nr_pages = min(req_pages, pipe->buffers); | |
1504 | ||
708e3508 HD |
1505 | spd.nr_pages = find_get_pages_contig(mapping, index, |
1506 | nr_pages, spd.pages); | |
1507 | index += spd.nr_pages; | |
708e3508 | 1508 | error = 0; |
708e3508 | 1509 | |
71f0e07a | 1510 | while (spd.nr_pages < nr_pages) { |
71f0e07a HD |
1511 | error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL); |
1512 | if (error) | |
1513 | break; | |
1514 | unlock_page(page); | |
708e3508 HD |
1515 | spd.pages[spd.nr_pages++] = page; |
1516 | index++; | |
1517 | } | |
1518 | ||
708e3508 HD |
1519 | index = *ppos >> PAGE_CACHE_SHIFT; |
1520 | nr_pages = spd.nr_pages; | |
1521 | spd.nr_pages = 0; | |
71f0e07a | 1522 | |
708e3508 HD |
1523 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { |
1524 | unsigned int this_len; | |
1525 | ||
1526 | if (!len) | |
1527 | break; | |
1528 | ||
708e3508 HD |
1529 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); |
1530 | page = spd.pages[page_nr]; | |
1531 | ||
71f0e07a | 1532 | if (!PageUptodate(page) || page->mapping != mapping) { |
71f0e07a HD |
1533 | error = shmem_getpage(inode, index, &page, |
1534 | SGP_CACHE, NULL); | |
1535 | if (error) | |
708e3508 | 1536 | break; |
71f0e07a HD |
1537 | unlock_page(page); |
1538 | page_cache_release(spd.pages[page_nr]); | |
1539 | spd.pages[page_nr] = page; | |
708e3508 | 1540 | } |
71f0e07a HD |
1541 | |
1542 | isize = i_size_read(inode); | |
708e3508 HD |
1543 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; |
1544 | if (unlikely(!isize || index > end_index)) | |
1545 | break; | |
1546 | ||
708e3508 HD |
1547 | if (end_index == index) { |
1548 | unsigned int plen; | |
1549 | ||
708e3508 HD |
1550 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
1551 | if (plen <= loff) | |
1552 | break; | |
1553 | ||
708e3508 HD |
1554 | this_len = min(this_len, plen - loff); |
1555 | len = this_len; | |
1556 | } | |
1557 | ||
1558 | spd.partial[page_nr].offset = loff; | |
1559 | spd.partial[page_nr].len = this_len; | |
1560 | len -= this_len; | |
1561 | loff = 0; | |
1562 | spd.nr_pages++; | |
1563 | index++; | |
1564 | } | |
1565 | ||
708e3508 HD |
1566 | while (page_nr < nr_pages) |
1567 | page_cache_release(spd.pages[page_nr++]); | |
708e3508 HD |
1568 | |
1569 | if (spd.nr_pages) | |
1570 | error = splice_to_pipe(pipe, &spd); | |
1571 | ||
1572 | splice_shrink_spd(pipe, &spd); | |
1573 | ||
1574 | if (error > 0) { | |
1575 | *ppos += error; | |
1576 | file_accessed(in); | |
1577 | } | |
1578 | return error; | |
1579 | } | |
1580 | ||
726c3342 | 1581 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 1582 | { |
726c3342 | 1583 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
1584 | |
1585 | buf->f_type = TMPFS_MAGIC; | |
1586 | buf->f_bsize = PAGE_CACHE_SIZE; | |
1587 | buf->f_namelen = NAME_MAX; | |
0edd73b3 | 1588 | if (sbinfo->max_blocks) { |
1da177e4 | 1589 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
1590 | buf->f_bavail = |
1591 | buf->f_bfree = sbinfo->max_blocks - | |
1592 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
1593 | } |
1594 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
1595 | buf->f_files = sbinfo->max_inodes; |
1596 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
1597 | } |
1598 | /* else leave those fields 0 like simple_statfs */ | |
1599 | return 0; | |
1600 | } | |
1601 | ||
1602 | /* | |
1603 | * File creation. Allocate an inode, and we're done.. | |
1604 | */ | |
1605 | static int | |
1a67aafb | 1606 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 1607 | { |
0b0a0806 | 1608 | struct inode *inode; |
1da177e4 LT |
1609 | int error = -ENOSPC; |
1610 | ||
454abafe | 1611 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 1612 | if (inode) { |
2a7dba39 | 1613 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 1614 | &dentry->d_name, |
6d9d88d0 | 1615 | shmem_initxattrs, NULL); |
570bc1c2 SS |
1616 | if (error) { |
1617 | if (error != -EOPNOTSUPP) { | |
1618 | iput(inode); | |
1619 | return error; | |
1620 | } | |
39f0247d | 1621 | } |
1c7c474c CH |
1622 | #ifdef CONFIG_TMPFS_POSIX_ACL |
1623 | error = generic_acl_init(inode, dir); | |
39f0247d AG |
1624 | if (error) { |
1625 | iput(inode); | |
1626 | return error; | |
570bc1c2 | 1627 | } |
718deb6b AV |
1628 | #else |
1629 | error = 0; | |
1c7c474c | 1630 | #endif |
1da177e4 LT |
1631 | dir->i_size += BOGO_DIRENT_SIZE; |
1632 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
1633 | d_instantiate(dentry, inode); | |
1634 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
1635 | } |
1636 | return error; | |
1637 | } | |
1638 | ||
18bb1db3 | 1639 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
1640 | { |
1641 | int error; | |
1642 | ||
1643 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
1644 | return error; | |
d8c76e6f | 1645 | inc_nlink(dir); |
1da177e4 LT |
1646 | return 0; |
1647 | } | |
1648 | ||
4acdaf27 | 1649 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
1da177e4 LT |
1650 | struct nameidata *nd) |
1651 | { | |
1652 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
1653 | } | |
1654 | ||
1655 | /* | |
1656 | * Link a file.. | |
1657 | */ | |
1658 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1659 | { | |
1660 | struct inode *inode = old_dentry->d_inode; | |
5b04c689 | 1661 | int ret; |
1da177e4 LT |
1662 | |
1663 | /* | |
1664 | * No ordinary (disk based) filesystem counts links as inodes; | |
1665 | * but each new link needs a new dentry, pinning lowmem, and | |
1666 | * tmpfs dentries cannot be pruned until they are unlinked. | |
1667 | */ | |
5b04c689 PE |
1668 | ret = shmem_reserve_inode(inode->i_sb); |
1669 | if (ret) | |
1670 | goto out; | |
1da177e4 LT |
1671 | |
1672 | dir->i_size += BOGO_DIRENT_SIZE; | |
1673 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 1674 | inc_nlink(inode); |
7de9c6ee | 1675 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
1676 | dget(dentry); /* Extra pinning count for the created dentry */ |
1677 | d_instantiate(dentry, inode); | |
5b04c689 PE |
1678 | out: |
1679 | return ret; | |
1da177e4 LT |
1680 | } |
1681 | ||
1682 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
1683 | { | |
1684 | struct inode *inode = dentry->d_inode; | |
1685 | ||
5b04c689 PE |
1686 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
1687 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
1688 | |
1689 | dir->i_size -= BOGO_DIRENT_SIZE; | |
1690 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 1691 | drop_nlink(inode); |
1da177e4 LT |
1692 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
1693 | return 0; | |
1694 | } | |
1695 | ||
1696 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
1697 | { | |
1698 | if (!simple_empty(dentry)) | |
1699 | return -ENOTEMPTY; | |
1700 | ||
9a53c3a7 DH |
1701 | drop_nlink(dentry->d_inode); |
1702 | drop_nlink(dir); | |
1da177e4 LT |
1703 | return shmem_unlink(dir, dentry); |
1704 | } | |
1705 | ||
1706 | /* | |
1707 | * The VFS layer already does all the dentry stuff for rename, | |
1708 | * we just have to decrement the usage count for the target if | |
1709 | * it exists so that the VFS layer correctly free's it when it | |
1710 | * gets overwritten. | |
1711 | */ | |
1712 | static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) | |
1713 | { | |
1714 | struct inode *inode = old_dentry->d_inode; | |
1715 | int they_are_dirs = S_ISDIR(inode->i_mode); | |
1716 | ||
1717 | if (!simple_empty(new_dentry)) | |
1718 | return -ENOTEMPTY; | |
1719 | ||
1720 | if (new_dentry->d_inode) { | |
1721 | (void) shmem_unlink(new_dir, new_dentry); | |
1722 | if (they_are_dirs) | |
9a53c3a7 | 1723 | drop_nlink(old_dir); |
1da177e4 | 1724 | } else if (they_are_dirs) { |
9a53c3a7 | 1725 | drop_nlink(old_dir); |
d8c76e6f | 1726 | inc_nlink(new_dir); |
1da177e4 LT |
1727 | } |
1728 | ||
1729 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
1730 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
1731 | old_dir->i_ctime = old_dir->i_mtime = | |
1732 | new_dir->i_ctime = new_dir->i_mtime = | |
1733 | inode->i_ctime = CURRENT_TIME; | |
1734 | return 0; | |
1735 | } | |
1736 | ||
1737 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1738 | { | |
1739 | int error; | |
1740 | int len; | |
1741 | struct inode *inode; | |
9276aad6 | 1742 | struct page *page; |
1da177e4 LT |
1743 | char *kaddr; |
1744 | struct shmem_inode_info *info; | |
1745 | ||
1746 | len = strlen(symname) + 1; | |
1747 | if (len > PAGE_CACHE_SIZE) | |
1748 | return -ENAMETOOLONG; | |
1749 | ||
454abafe | 1750 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
1751 | if (!inode) |
1752 | return -ENOSPC; | |
1753 | ||
9d8f13ba | 1754 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 1755 | shmem_initxattrs, NULL); |
570bc1c2 SS |
1756 | if (error) { |
1757 | if (error != -EOPNOTSUPP) { | |
1758 | iput(inode); | |
1759 | return error; | |
1760 | } | |
1761 | error = 0; | |
1762 | } | |
1763 | ||
1da177e4 LT |
1764 | info = SHMEM_I(inode); |
1765 | inode->i_size = len-1; | |
69f07ec9 HD |
1766 | if (len <= SHORT_SYMLINK_LEN) { |
1767 | info->symlink = kmemdup(symname, len, GFP_KERNEL); | |
1768 | if (!info->symlink) { | |
1769 | iput(inode); | |
1770 | return -ENOMEM; | |
1771 | } | |
1772 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 LT |
1773 | } else { |
1774 | error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); | |
1775 | if (error) { | |
1776 | iput(inode); | |
1777 | return error; | |
1778 | } | |
14fcc23f | 1779 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 1780 | inode->i_op = &shmem_symlink_inode_operations; |
9b04c5fe | 1781 | kaddr = kmap_atomic(page); |
1da177e4 | 1782 | memcpy(kaddr, symname, len); |
9b04c5fe | 1783 | kunmap_atomic(kaddr); |
ec9516fb | 1784 | SetPageUptodate(page); |
1da177e4 | 1785 | set_page_dirty(page); |
6746aff7 | 1786 | unlock_page(page); |
1da177e4 LT |
1787 | page_cache_release(page); |
1788 | } | |
1da177e4 LT |
1789 | dir->i_size += BOGO_DIRENT_SIZE; |
1790 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
1791 | d_instantiate(dentry, inode); | |
1792 | dget(dentry); | |
1793 | return 0; | |
1794 | } | |
1795 | ||
69f07ec9 | 1796 | static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 1797 | { |
69f07ec9 | 1798 | nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink); |
cc314eef | 1799 | return NULL; |
1da177e4 LT |
1800 | } |
1801 | ||
cc314eef | 1802 | static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) |
1da177e4 LT |
1803 | { |
1804 | struct page *page = NULL; | |
41ffe5d5 HD |
1805 | int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); |
1806 | nd_set_link(nd, error ? ERR_PTR(error) : kmap(page)); | |
d3602444 HD |
1807 | if (page) |
1808 | unlock_page(page); | |
cc314eef | 1809 | return page; |
1da177e4 LT |
1810 | } |
1811 | ||
cc314eef | 1812 | static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 LT |
1813 | { |
1814 | if (!IS_ERR(nd_get_link(nd))) { | |
cc314eef | 1815 | struct page *page = cookie; |
1da177e4 LT |
1816 | kunmap(page); |
1817 | mark_page_accessed(page); | |
1818 | page_cache_release(page); | |
1da177e4 LT |
1819 | } |
1820 | } | |
1821 | ||
b09e0fa4 | 1822 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 1823 | /* |
b09e0fa4 EP |
1824 | * Superblocks without xattr inode operations may get some security.* xattr |
1825 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
1826 | * like ACLs, we also need to implement the security.* handlers at |
1827 | * filesystem level, though. | |
1828 | */ | |
1829 | ||
6d9d88d0 JS |
1830 | /* |
1831 | * Allocate new xattr and copy in the value; but leave the name to callers. | |
1832 | */ | |
1833 | static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size) | |
1834 | { | |
1835 | struct shmem_xattr *new_xattr; | |
1836 | size_t len; | |
1837 | ||
1838 | /* wrap around? */ | |
1839 | len = sizeof(*new_xattr) + size; | |
1840 | if (len <= sizeof(*new_xattr)) | |
1841 | return NULL; | |
1842 | ||
1843 | new_xattr = kmalloc(len, GFP_KERNEL); | |
1844 | if (!new_xattr) | |
1845 | return NULL; | |
1846 | ||
1847 | new_xattr->size = size; | |
1848 | memcpy(new_xattr->value, value, size); | |
1849 | return new_xattr; | |
1850 | } | |
1851 | ||
1852 | /* | |
1853 | * Callback for security_inode_init_security() for acquiring xattrs. | |
1854 | */ | |
1855 | static int shmem_initxattrs(struct inode *inode, | |
1856 | const struct xattr *xattr_array, | |
1857 | void *fs_info) | |
1858 | { | |
1859 | struct shmem_inode_info *info = SHMEM_I(inode); | |
1860 | const struct xattr *xattr; | |
1861 | struct shmem_xattr *new_xattr; | |
1862 | size_t len; | |
1863 | ||
1864 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
1865 | new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len); | |
1866 | if (!new_xattr) | |
1867 | return -ENOMEM; | |
1868 | ||
1869 | len = strlen(xattr->name) + 1; | |
1870 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
1871 | GFP_KERNEL); | |
1872 | if (!new_xattr->name) { | |
1873 | kfree(new_xattr); | |
1874 | return -ENOMEM; | |
1875 | } | |
1876 | ||
1877 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
1878 | XATTR_SECURITY_PREFIX_LEN); | |
1879 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
1880 | xattr->name, len); | |
1881 | ||
1882 | spin_lock(&info->lock); | |
1883 | list_add(&new_xattr->list, &info->xattr_list); | |
1884 | spin_unlock(&info->lock); | |
1885 | } | |
1886 | ||
1887 | return 0; | |
1888 | } | |
1889 | ||
b09e0fa4 EP |
1890 | static int shmem_xattr_get(struct dentry *dentry, const char *name, |
1891 | void *buffer, size_t size) | |
39f0247d | 1892 | { |
b09e0fa4 EP |
1893 | struct shmem_inode_info *info; |
1894 | struct shmem_xattr *xattr; | |
1895 | int ret = -ENODATA; | |
39f0247d | 1896 | |
b09e0fa4 EP |
1897 | info = SHMEM_I(dentry->d_inode); |
1898 | ||
1899 | spin_lock(&info->lock); | |
1900 | list_for_each_entry(xattr, &info->xattr_list, list) { | |
1901 | if (strcmp(name, xattr->name)) | |
1902 | continue; | |
1903 | ||
1904 | ret = xattr->size; | |
1905 | if (buffer) { | |
1906 | if (size < xattr->size) | |
1907 | ret = -ERANGE; | |
1908 | else | |
1909 | memcpy(buffer, xattr->value, xattr->size); | |
1910 | } | |
1911 | break; | |
1912 | } | |
1913 | spin_unlock(&info->lock); | |
1914 | return ret; | |
39f0247d AG |
1915 | } |
1916 | ||
6d9d88d0 | 1917 | static int shmem_xattr_set(struct inode *inode, const char *name, |
b09e0fa4 | 1918 | const void *value, size_t size, int flags) |
39f0247d | 1919 | { |
b09e0fa4 EP |
1920 | struct shmem_inode_info *info = SHMEM_I(inode); |
1921 | struct shmem_xattr *xattr; | |
1922 | struct shmem_xattr *new_xattr = NULL; | |
b09e0fa4 EP |
1923 | int err = 0; |
1924 | ||
1925 | /* value == NULL means remove */ | |
1926 | if (value) { | |
6d9d88d0 | 1927 | new_xattr = shmem_xattr_alloc(value, size); |
b09e0fa4 EP |
1928 | if (!new_xattr) |
1929 | return -ENOMEM; | |
1930 | ||
1931 | new_xattr->name = kstrdup(name, GFP_KERNEL); | |
1932 | if (!new_xattr->name) { | |
1933 | kfree(new_xattr); | |
1934 | return -ENOMEM; | |
1935 | } | |
b09e0fa4 EP |
1936 | } |
1937 | ||
1938 | spin_lock(&info->lock); | |
1939 | list_for_each_entry(xattr, &info->xattr_list, list) { | |
1940 | if (!strcmp(name, xattr->name)) { | |
1941 | if (flags & XATTR_CREATE) { | |
1942 | xattr = new_xattr; | |
1943 | err = -EEXIST; | |
1944 | } else if (new_xattr) { | |
1945 | list_replace(&xattr->list, &new_xattr->list); | |
1946 | } else { | |
1947 | list_del(&xattr->list); | |
1948 | } | |
1949 | goto out; | |
1950 | } | |
1951 | } | |
1952 | if (flags & XATTR_REPLACE) { | |
1953 | xattr = new_xattr; | |
1954 | err = -ENODATA; | |
1955 | } else { | |
1956 | list_add(&new_xattr->list, &info->xattr_list); | |
1957 | xattr = NULL; | |
1958 | } | |
1959 | out: | |
1960 | spin_unlock(&info->lock); | |
1961 | if (xattr) | |
1962 | kfree(xattr->name); | |
1963 | kfree(xattr); | |
1964 | return err; | |
39f0247d AG |
1965 | } |
1966 | ||
bb435453 | 1967 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
b09e0fa4 | 1968 | #ifdef CONFIG_TMPFS_POSIX_ACL |
1c7c474c CH |
1969 | &generic_acl_access_handler, |
1970 | &generic_acl_default_handler, | |
b09e0fa4 | 1971 | #endif |
39f0247d AG |
1972 | NULL |
1973 | }; | |
b09e0fa4 EP |
1974 | |
1975 | static int shmem_xattr_validate(const char *name) | |
1976 | { | |
1977 | struct { const char *prefix; size_t len; } arr[] = { | |
1978 | { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN }, | |
1979 | { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN } | |
1980 | }; | |
1981 | int i; | |
1982 | ||
1983 | for (i = 0; i < ARRAY_SIZE(arr); i++) { | |
1984 | size_t preflen = arr[i].len; | |
1985 | if (strncmp(name, arr[i].prefix, preflen) == 0) { | |
1986 | if (!name[preflen]) | |
1987 | return -EINVAL; | |
1988 | return 0; | |
1989 | } | |
1990 | } | |
1991 | return -EOPNOTSUPP; | |
1992 | } | |
1993 | ||
1994 | static ssize_t shmem_getxattr(struct dentry *dentry, const char *name, | |
1995 | void *buffer, size_t size) | |
1996 | { | |
1997 | int err; | |
1998 | ||
1999 | /* | |
2000 | * If this is a request for a synthetic attribute in the system.* | |
2001 | * namespace use the generic infrastructure to resolve a handler | |
2002 | * for it via sb->s_xattr. | |
2003 | */ | |
2004 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2005 | return generic_getxattr(dentry, name, buffer, size); | |
2006 | ||
2007 | err = shmem_xattr_validate(name); | |
2008 | if (err) | |
2009 | return err; | |
2010 | ||
2011 | return shmem_xattr_get(dentry, name, buffer, size); | |
2012 | } | |
2013 | ||
2014 | static int shmem_setxattr(struct dentry *dentry, const char *name, | |
2015 | const void *value, size_t size, int flags) | |
2016 | { | |
2017 | int err; | |
2018 | ||
2019 | /* | |
2020 | * If this is a request for a synthetic attribute in the system.* | |
2021 | * namespace use the generic infrastructure to resolve a handler | |
2022 | * for it via sb->s_xattr. | |
2023 | */ | |
2024 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2025 | return generic_setxattr(dentry, name, value, size, flags); | |
2026 | ||
2027 | err = shmem_xattr_validate(name); | |
2028 | if (err) | |
2029 | return err; | |
2030 | ||
2031 | if (size == 0) | |
2032 | value = ""; /* empty EA, do not remove */ | |
2033 | ||
6d9d88d0 | 2034 | return shmem_xattr_set(dentry->d_inode, name, value, size, flags); |
b09e0fa4 EP |
2035 | |
2036 | } | |
2037 | ||
2038 | static int shmem_removexattr(struct dentry *dentry, const char *name) | |
2039 | { | |
2040 | int err; | |
2041 | ||
2042 | /* | |
2043 | * If this is a request for a synthetic attribute in the system.* | |
2044 | * namespace use the generic infrastructure to resolve a handler | |
2045 | * for it via sb->s_xattr. | |
2046 | */ | |
2047 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2048 | return generic_removexattr(dentry, name); | |
2049 | ||
2050 | err = shmem_xattr_validate(name); | |
2051 | if (err) | |
2052 | return err; | |
2053 | ||
6d9d88d0 | 2054 | return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE); |
b09e0fa4 EP |
2055 | } |
2056 | ||
2057 | static bool xattr_is_trusted(const char *name) | |
2058 | { | |
2059 | return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN); | |
2060 | } | |
2061 | ||
2062 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
2063 | { | |
2064 | bool trusted = capable(CAP_SYS_ADMIN); | |
2065 | struct shmem_xattr *xattr; | |
2066 | struct shmem_inode_info *info; | |
2067 | size_t used = 0; | |
2068 | ||
2069 | info = SHMEM_I(dentry->d_inode); | |
2070 | ||
2071 | spin_lock(&info->lock); | |
2072 | list_for_each_entry(xattr, &info->xattr_list, list) { | |
2073 | size_t len; | |
2074 | ||
2075 | /* skip "trusted." attributes for unprivileged callers */ | |
2076 | if (!trusted && xattr_is_trusted(xattr->name)) | |
2077 | continue; | |
2078 | ||
2079 | len = strlen(xattr->name) + 1; | |
2080 | used += len; | |
2081 | if (buffer) { | |
2082 | if (size < used) { | |
2083 | used = -ERANGE; | |
2084 | break; | |
2085 | } | |
2086 | memcpy(buffer, xattr->name, len); | |
2087 | buffer += len; | |
2088 | } | |
2089 | } | |
2090 | spin_unlock(&info->lock); | |
2091 | ||
2092 | return used; | |
2093 | } | |
2094 | #endif /* CONFIG_TMPFS_XATTR */ | |
2095 | ||
69f07ec9 | 2096 | static const struct inode_operations shmem_short_symlink_operations = { |
b09e0fa4 | 2097 | .readlink = generic_readlink, |
69f07ec9 | 2098 | .follow_link = shmem_follow_short_symlink, |
b09e0fa4 EP |
2099 | #ifdef CONFIG_TMPFS_XATTR |
2100 | .setxattr = shmem_setxattr, | |
2101 | .getxattr = shmem_getxattr, | |
2102 | .listxattr = shmem_listxattr, | |
2103 | .removexattr = shmem_removexattr, | |
2104 | #endif | |
2105 | }; | |
2106 | ||
2107 | static const struct inode_operations shmem_symlink_inode_operations = { | |
2108 | .readlink = generic_readlink, | |
2109 | .follow_link = shmem_follow_link, | |
2110 | .put_link = shmem_put_link, | |
2111 | #ifdef CONFIG_TMPFS_XATTR | |
2112 | .setxattr = shmem_setxattr, | |
2113 | .getxattr = shmem_getxattr, | |
2114 | .listxattr = shmem_listxattr, | |
2115 | .removexattr = shmem_removexattr, | |
39f0247d | 2116 | #endif |
b09e0fa4 | 2117 | }; |
39f0247d | 2118 | |
91828a40 DG |
2119 | static struct dentry *shmem_get_parent(struct dentry *child) |
2120 | { | |
2121 | return ERR_PTR(-ESTALE); | |
2122 | } | |
2123 | ||
2124 | static int shmem_match(struct inode *ino, void *vfh) | |
2125 | { | |
2126 | __u32 *fh = vfh; | |
2127 | __u64 inum = fh[2]; | |
2128 | inum = (inum << 32) | fh[1]; | |
2129 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
2130 | } | |
2131 | ||
480b116c CH |
2132 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
2133 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 2134 | { |
91828a40 | 2135 | struct inode *inode; |
480b116c CH |
2136 | struct dentry *dentry = NULL; |
2137 | u64 inum = fid->raw[2]; | |
2138 | inum = (inum << 32) | fid->raw[1]; | |
2139 | ||
2140 | if (fh_len < 3) | |
2141 | return NULL; | |
91828a40 | 2142 | |
480b116c CH |
2143 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
2144 | shmem_match, fid->raw); | |
91828a40 | 2145 | if (inode) { |
480b116c | 2146 | dentry = d_find_alias(inode); |
91828a40 DG |
2147 | iput(inode); |
2148 | } | |
2149 | ||
480b116c | 2150 | return dentry; |
91828a40 DG |
2151 | } |
2152 | ||
2153 | static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, | |
2154 | int connectable) | |
2155 | { | |
2156 | struct inode *inode = dentry->d_inode; | |
2157 | ||
5fe0c237 AK |
2158 | if (*len < 3) { |
2159 | *len = 3; | |
91828a40 | 2160 | return 255; |
5fe0c237 | 2161 | } |
91828a40 | 2162 | |
1d3382cb | 2163 | if (inode_unhashed(inode)) { |
91828a40 DG |
2164 | /* Unfortunately insert_inode_hash is not idempotent, |
2165 | * so as we hash inodes here rather than at creation | |
2166 | * time, we need a lock to ensure we only try | |
2167 | * to do it once | |
2168 | */ | |
2169 | static DEFINE_SPINLOCK(lock); | |
2170 | spin_lock(&lock); | |
1d3382cb | 2171 | if (inode_unhashed(inode)) |
91828a40 DG |
2172 | __insert_inode_hash(inode, |
2173 | inode->i_ino + inode->i_generation); | |
2174 | spin_unlock(&lock); | |
2175 | } | |
2176 | ||
2177 | fh[0] = inode->i_generation; | |
2178 | fh[1] = inode->i_ino; | |
2179 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
2180 | ||
2181 | *len = 3; | |
2182 | return 1; | |
2183 | } | |
2184 | ||
39655164 | 2185 | static const struct export_operations shmem_export_ops = { |
91828a40 | 2186 | .get_parent = shmem_get_parent, |
91828a40 | 2187 | .encode_fh = shmem_encode_fh, |
480b116c | 2188 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
2189 | }; |
2190 | ||
680d794b AM |
2191 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
2192 | bool remount) | |
1da177e4 LT |
2193 | { |
2194 | char *this_char, *value, *rest; | |
8751e039 EB |
2195 | uid_t uid; |
2196 | gid_t gid; | |
1da177e4 | 2197 | |
b00dc3ad HD |
2198 | while (options != NULL) { |
2199 | this_char = options; | |
2200 | for (;;) { | |
2201 | /* | |
2202 | * NUL-terminate this option: unfortunately, | |
2203 | * mount options form a comma-separated list, | |
2204 | * but mpol's nodelist may also contain commas. | |
2205 | */ | |
2206 | options = strchr(options, ','); | |
2207 | if (options == NULL) | |
2208 | break; | |
2209 | options++; | |
2210 | if (!isdigit(*options)) { | |
2211 | options[-1] = '\0'; | |
2212 | break; | |
2213 | } | |
2214 | } | |
1da177e4 LT |
2215 | if (!*this_char) |
2216 | continue; | |
2217 | if ((value = strchr(this_char,'=')) != NULL) { | |
2218 | *value++ = 0; | |
2219 | } else { | |
2220 | printk(KERN_ERR | |
2221 | "tmpfs: No value for mount option '%s'\n", | |
2222 | this_char); | |
2223 | return 1; | |
2224 | } | |
2225 | ||
2226 | if (!strcmp(this_char,"size")) { | |
2227 | unsigned long long size; | |
2228 | size = memparse(value,&rest); | |
2229 | if (*rest == '%') { | |
2230 | size <<= PAGE_SHIFT; | |
2231 | size *= totalram_pages; | |
2232 | do_div(size, 100); | |
2233 | rest++; | |
2234 | } | |
2235 | if (*rest) | |
2236 | goto bad_val; | |
680d794b AM |
2237 | sbinfo->max_blocks = |
2238 | DIV_ROUND_UP(size, PAGE_CACHE_SIZE); | |
1da177e4 | 2239 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 2240 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
2241 | if (*rest) |
2242 | goto bad_val; | |
2243 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 2244 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
2245 | if (*rest) |
2246 | goto bad_val; | |
2247 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 2248 | if (remount) |
1da177e4 | 2249 | continue; |
680d794b | 2250 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
2251 | if (*rest) |
2252 | goto bad_val; | |
2253 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 2254 | if (remount) |
1da177e4 | 2255 | continue; |
8751e039 | 2256 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2257 | if (*rest) |
2258 | goto bad_val; | |
8751e039 EB |
2259 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
2260 | if (!uid_valid(sbinfo->uid)) | |
2261 | goto bad_val; | |
1da177e4 | 2262 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 2263 | if (remount) |
1da177e4 | 2264 | continue; |
8751e039 | 2265 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2266 | if (*rest) |
2267 | goto bad_val; | |
8751e039 EB |
2268 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
2269 | if (!gid_valid(sbinfo->gid)) | |
2270 | goto bad_val; | |
7339ff83 | 2271 | } else if (!strcmp(this_char,"mpol")) { |
71fe804b | 2272 | if (mpol_parse_str(value, &sbinfo->mpol, 1)) |
7339ff83 | 2273 | goto bad_val; |
1da177e4 LT |
2274 | } else { |
2275 | printk(KERN_ERR "tmpfs: Bad mount option %s\n", | |
2276 | this_char); | |
2277 | return 1; | |
2278 | } | |
2279 | } | |
2280 | return 0; | |
2281 | ||
2282 | bad_val: | |
2283 | printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", | |
2284 | value, this_char); | |
2285 | return 1; | |
2286 | ||
2287 | } | |
2288 | ||
2289 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
2290 | { | |
2291 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 2292 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
2293 | unsigned long inodes; |
2294 | int error = -EINVAL; | |
2295 | ||
680d794b | 2296 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 2297 | return error; |
1da177e4 | 2298 | |
0edd73b3 | 2299 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 2300 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 2301 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 2302 | goto out; |
680d794b | 2303 | if (config.max_inodes < inodes) |
0edd73b3 HD |
2304 | goto out; |
2305 | /* | |
54af6042 | 2306 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
2307 | * but we must separately disallow unlimited->limited, because |
2308 | * in that case we have no record of how much is already in use. | |
2309 | */ | |
680d794b | 2310 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 2311 | goto out; |
680d794b | 2312 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
2313 | goto out; |
2314 | ||
2315 | error = 0; | |
680d794b | 2316 | sbinfo->max_blocks = config.max_blocks; |
680d794b AM |
2317 | sbinfo->max_inodes = config.max_inodes; |
2318 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b LS |
2319 | |
2320 | mpol_put(sbinfo->mpol); | |
2321 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
0edd73b3 HD |
2322 | out: |
2323 | spin_unlock(&sbinfo->stat_lock); | |
2324 | return error; | |
1da177e4 | 2325 | } |
680d794b | 2326 | |
34c80b1d | 2327 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 2328 | { |
34c80b1d | 2329 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b AM |
2330 | |
2331 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
2332 | seq_printf(seq, ",size=%luk", | |
2333 | sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); | |
2334 | if (sbinfo->max_inodes != shmem_default_max_inodes()) | |
2335 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
2336 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 2337 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
2338 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
2339 | seq_printf(seq, ",uid=%u", | |
2340 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
2341 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
2342 | seq_printf(seq, ",gid=%u", | |
2343 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
71fe804b | 2344 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b AM |
2345 | return 0; |
2346 | } | |
2347 | #endif /* CONFIG_TMPFS */ | |
1da177e4 LT |
2348 | |
2349 | static void shmem_put_super(struct super_block *sb) | |
2350 | { | |
602586a8 HD |
2351 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
2352 | ||
2353 | percpu_counter_destroy(&sbinfo->used_blocks); | |
2354 | kfree(sbinfo); | |
1da177e4 LT |
2355 | sb->s_fs_info = NULL; |
2356 | } | |
2357 | ||
2b2af54a | 2358 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
2359 | { |
2360 | struct inode *inode; | |
0edd73b3 | 2361 | struct shmem_sb_info *sbinfo; |
680d794b AM |
2362 | int err = -ENOMEM; |
2363 | ||
2364 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 2365 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b AM |
2366 | L1_CACHE_BYTES), GFP_KERNEL); |
2367 | if (!sbinfo) | |
2368 | return -ENOMEM; | |
2369 | ||
680d794b | 2370 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
2371 | sbinfo->uid = current_fsuid(); |
2372 | sbinfo->gid = current_fsgid(); | |
680d794b | 2373 | sb->s_fs_info = sbinfo; |
1da177e4 | 2374 | |
0edd73b3 | 2375 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
2376 | /* |
2377 | * Per default we only allow half of the physical ram per | |
2378 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
2379 | * but the internal instance is left unlimited. | |
2380 | */ | |
2381 | if (!(sb->s_flags & MS_NOUSER)) { | |
680d794b AM |
2382 | sbinfo->max_blocks = shmem_default_max_blocks(); |
2383 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
2384 | if (shmem_parse_options(data, sbinfo, false)) { | |
2385 | err = -EINVAL; | |
2386 | goto failed; | |
2387 | } | |
1da177e4 | 2388 | } |
91828a40 | 2389 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 2390 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
2391 | #else |
2392 | sb->s_flags |= MS_NOUSER; | |
2393 | #endif | |
2394 | ||
0edd73b3 | 2395 | spin_lock_init(&sbinfo->stat_lock); |
602586a8 HD |
2396 | if (percpu_counter_init(&sbinfo->used_blocks, 0)) |
2397 | goto failed; | |
680d794b | 2398 | sbinfo->free_inodes = sbinfo->max_inodes; |
0edd73b3 | 2399 | |
285b2c4f | 2400 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
1da177e4 LT |
2401 | sb->s_blocksize = PAGE_CACHE_SIZE; |
2402 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
2403 | sb->s_magic = TMPFS_MAGIC; | |
2404 | sb->s_op = &shmem_ops; | |
cfd95a9c | 2405 | sb->s_time_gran = 1; |
b09e0fa4 | 2406 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 2407 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
2408 | #endif |
2409 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
2410 | sb->s_flags |= MS_POSIXACL; |
2411 | #endif | |
0edd73b3 | 2412 | |
454abafe | 2413 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
2414 | if (!inode) |
2415 | goto failed; | |
680d794b AM |
2416 | inode->i_uid = sbinfo->uid; |
2417 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
2418 | sb->s_root = d_make_root(inode); |
2419 | if (!sb->s_root) | |
48fde701 | 2420 | goto failed; |
1da177e4 LT |
2421 | return 0; |
2422 | ||
1da177e4 LT |
2423 | failed: |
2424 | shmem_put_super(sb); | |
2425 | return err; | |
2426 | } | |
2427 | ||
fcc234f8 | 2428 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
2429 | |
2430 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
2431 | { | |
41ffe5d5 HD |
2432 | struct shmem_inode_info *info; |
2433 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
2434 | if (!info) | |
1da177e4 | 2435 | return NULL; |
41ffe5d5 | 2436 | return &info->vfs_inode; |
1da177e4 LT |
2437 | } |
2438 | ||
41ffe5d5 | 2439 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
2440 | { |
2441 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
2442 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
2443 | } | |
2444 | ||
1da177e4 LT |
2445 | static void shmem_destroy_inode(struct inode *inode) |
2446 | { | |
09208d15 | 2447 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 2448 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 2449 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
2450 | } |
2451 | ||
41ffe5d5 | 2452 | static void shmem_init_inode(void *foo) |
1da177e4 | 2453 | { |
41ffe5d5 HD |
2454 | struct shmem_inode_info *info = foo; |
2455 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
2456 | } |
2457 | ||
41ffe5d5 | 2458 | static int shmem_init_inodecache(void) |
1da177e4 LT |
2459 | { |
2460 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
2461 | sizeof(struct shmem_inode_info), | |
41ffe5d5 | 2462 | 0, SLAB_PANIC, shmem_init_inode); |
1da177e4 LT |
2463 | return 0; |
2464 | } | |
2465 | ||
41ffe5d5 | 2466 | static void shmem_destroy_inodecache(void) |
1da177e4 | 2467 | { |
1a1d92c1 | 2468 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
2469 | } |
2470 | ||
f5e54d6e | 2471 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 2472 | .writepage = shmem_writepage, |
76719325 | 2473 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 2474 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
2475 | .write_begin = shmem_write_begin, |
2476 | .write_end = shmem_write_end, | |
1da177e4 | 2477 | #endif |
304dbdb7 | 2478 | .migratepage = migrate_page, |
aa261f54 | 2479 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
2480 | }; |
2481 | ||
15ad7cdc | 2482 | static const struct file_operations shmem_file_operations = { |
1da177e4 LT |
2483 | .mmap = shmem_mmap, |
2484 | #ifdef CONFIG_TMPFS | |
2485 | .llseek = generic_file_llseek, | |
bcd78e49 | 2486 | .read = do_sync_read, |
5402b976 | 2487 | .write = do_sync_write, |
bcd78e49 | 2488 | .aio_read = shmem_file_aio_read, |
5402b976 | 2489 | .aio_write = generic_file_aio_write, |
1b061d92 | 2490 | .fsync = noop_fsync, |
708e3508 | 2491 | .splice_read = shmem_file_splice_read, |
ae976416 | 2492 | .splice_write = generic_file_splice_write, |
1da177e4 LT |
2493 | #endif |
2494 | }; | |
2495 | ||
92e1d5be | 2496 | static const struct inode_operations shmem_inode_operations = { |
94c1e62d | 2497 | .setattr = shmem_setattr, |
f6b3ec23 | 2498 | .truncate_range = shmem_truncate_range, |
b09e0fa4 EP |
2499 | #ifdef CONFIG_TMPFS_XATTR |
2500 | .setxattr = shmem_setxattr, | |
2501 | .getxattr = shmem_getxattr, | |
2502 | .listxattr = shmem_listxattr, | |
2503 | .removexattr = shmem_removexattr, | |
2504 | #endif | |
1da177e4 LT |
2505 | }; |
2506 | ||
92e1d5be | 2507 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
2508 | #ifdef CONFIG_TMPFS |
2509 | .create = shmem_create, | |
2510 | .lookup = simple_lookup, | |
2511 | .link = shmem_link, | |
2512 | .unlink = shmem_unlink, | |
2513 | .symlink = shmem_symlink, | |
2514 | .mkdir = shmem_mkdir, | |
2515 | .rmdir = shmem_rmdir, | |
2516 | .mknod = shmem_mknod, | |
2517 | .rename = shmem_rename, | |
1da177e4 | 2518 | #endif |
b09e0fa4 EP |
2519 | #ifdef CONFIG_TMPFS_XATTR |
2520 | .setxattr = shmem_setxattr, | |
2521 | .getxattr = shmem_getxattr, | |
2522 | .listxattr = shmem_listxattr, | |
2523 | .removexattr = shmem_removexattr, | |
2524 | #endif | |
39f0247d | 2525 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 2526 | .setattr = shmem_setattr, |
39f0247d AG |
2527 | #endif |
2528 | }; | |
2529 | ||
92e1d5be | 2530 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 EP |
2531 | #ifdef CONFIG_TMPFS_XATTR |
2532 | .setxattr = shmem_setxattr, | |
2533 | .getxattr = shmem_getxattr, | |
2534 | .listxattr = shmem_listxattr, | |
2535 | .removexattr = shmem_removexattr, | |
2536 | #endif | |
39f0247d | 2537 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 2538 | .setattr = shmem_setattr, |
39f0247d | 2539 | #endif |
1da177e4 LT |
2540 | }; |
2541 | ||
759b9775 | 2542 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
2543 | .alloc_inode = shmem_alloc_inode, |
2544 | .destroy_inode = shmem_destroy_inode, | |
2545 | #ifdef CONFIG_TMPFS | |
2546 | .statfs = shmem_statfs, | |
2547 | .remount_fs = shmem_remount_fs, | |
680d794b | 2548 | .show_options = shmem_show_options, |
1da177e4 | 2549 | #endif |
1f895f75 | 2550 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
2551 | .drop_inode = generic_delete_inode, |
2552 | .put_super = shmem_put_super, | |
2553 | }; | |
2554 | ||
f0f37e2f | 2555 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 2556 | .fault = shmem_fault, |
1da177e4 LT |
2557 | #ifdef CONFIG_NUMA |
2558 | .set_policy = shmem_set_policy, | |
2559 | .get_policy = shmem_get_policy, | |
2560 | #endif | |
2561 | }; | |
2562 | ||
3c26ff6e AV |
2563 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
2564 | int flags, const char *dev_name, void *data) | |
1da177e4 | 2565 | { |
3c26ff6e | 2566 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
2567 | } |
2568 | ||
41ffe5d5 | 2569 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
2570 | .owner = THIS_MODULE, |
2571 | .name = "tmpfs", | |
3c26ff6e | 2572 | .mount = shmem_mount, |
1da177e4 LT |
2573 | .kill_sb = kill_litter_super, |
2574 | }; | |
1da177e4 | 2575 | |
41ffe5d5 | 2576 | int __init shmem_init(void) |
1da177e4 LT |
2577 | { |
2578 | int error; | |
2579 | ||
e0bf68dd PZ |
2580 | error = bdi_init(&shmem_backing_dev_info); |
2581 | if (error) | |
2582 | goto out4; | |
2583 | ||
41ffe5d5 | 2584 | error = shmem_init_inodecache(); |
1da177e4 LT |
2585 | if (error) |
2586 | goto out3; | |
2587 | ||
41ffe5d5 | 2588 | error = register_filesystem(&shmem_fs_type); |
1da177e4 LT |
2589 | if (error) { |
2590 | printk(KERN_ERR "Could not register tmpfs\n"); | |
2591 | goto out2; | |
2592 | } | |
95dc112a | 2593 | |
41ffe5d5 HD |
2594 | shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER, |
2595 | shmem_fs_type.name, NULL); | |
1da177e4 LT |
2596 | if (IS_ERR(shm_mnt)) { |
2597 | error = PTR_ERR(shm_mnt); | |
2598 | printk(KERN_ERR "Could not kern_mount tmpfs\n"); | |
2599 | goto out1; | |
2600 | } | |
2601 | return 0; | |
2602 | ||
2603 | out1: | |
41ffe5d5 | 2604 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 2605 | out2: |
41ffe5d5 | 2606 | shmem_destroy_inodecache(); |
1da177e4 | 2607 | out3: |
e0bf68dd PZ |
2608 | bdi_destroy(&shmem_backing_dev_info); |
2609 | out4: | |
1da177e4 LT |
2610 | shm_mnt = ERR_PTR(error); |
2611 | return error; | |
2612 | } | |
853ac43a MM |
2613 | |
2614 | #else /* !CONFIG_SHMEM */ | |
2615 | ||
2616 | /* | |
2617 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
2618 | * | |
2619 | * This is intended for small system where the benefits of the full | |
2620 | * shmem code (swap-backed and resource-limited) are outweighed by | |
2621 | * their complexity. On systems without swap this code should be | |
2622 | * effectively equivalent, but much lighter weight. | |
2623 | */ | |
2624 | ||
2625 | #include <linux/ramfs.h> | |
2626 | ||
41ffe5d5 | 2627 | static struct file_system_type shmem_fs_type = { |
853ac43a | 2628 | .name = "tmpfs", |
3c26ff6e | 2629 | .mount = ramfs_mount, |
853ac43a MM |
2630 | .kill_sb = kill_litter_super, |
2631 | }; | |
2632 | ||
41ffe5d5 | 2633 | int __init shmem_init(void) |
853ac43a | 2634 | { |
41ffe5d5 | 2635 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 2636 | |
41ffe5d5 | 2637 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
2638 | BUG_ON(IS_ERR(shm_mnt)); |
2639 | ||
2640 | return 0; | |
2641 | } | |
2642 | ||
41ffe5d5 | 2643 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
2644 | { |
2645 | return 0; | |
2646 | } | |
2647 | ||
3f96b79a HD |
2648 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
2649 | { | |
2650 | return 0; | |
2651 | } | |
2652 | ||
24513264 HD |
2653 | void shmem_unlock_mapping(struct address_space *mapping) |
2654 | { | |
2655 | } | |
2656 | ||
41ffe5d5 | 2657 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 2658 | { |
41ffe5d5 | 2659 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
2660 | } |
2661 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
2662 | ||
0b0a0806 HD |
2663 | #define shmem_vm_ops generic_file_vm_ops |
2664 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 2665 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
2666 | #define shmem_acct_size(flags, size) 0 |
2667 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
2668 | |
2669 | #endif /* CONFIG_SHMEM */ | |
2670 | ||
2671 | /* common code */ | |
1da177e4 | 2672 | |
46711810 | 2673 | /** |
1da177e4 | 2674 | * shmem_file_setup - get an unlinked file living in tmpfs |
1da177e4 LT |
2675 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
2676 | * @size: size to be set for the file | |
0b0a0806 | 2677 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size |
1da177e4 | 2678 | */ |
168f5ac6 | 2679 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) |
1da177e4 LT |
2680 | { |
2681 | int error; | |
2682 | struct file *file; | |
2683 | struct inode *inode; | |
2c48b9c4 AV |
2684 | struct path path; |
2685 | struct dentry *root; | |
1da177e4 LT |
2686 | struct qstr this; |
2687 | ||
2688 | if (IS_ERR(shm_mnt)) | |
2689 | return (void *)shm_mnt; | |
2690 | ||
285b2c4f | 2691 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
2692 | return ERR_PTR(-EINVAL); |
2693 | ||
2694 | if (shmem_acct_size(flags, size)) | |
2695 | return ERR_PTR(-ENOMEM); | |
2696 | ||
2697 | error = -ENOMEM; | |
2698 | this.name = name; | |
2699 | this.len = strlen(name); | |
2700 | this.hash = 0; /* will go */ | |
2701 | root = shm_mnt->mnt_root; | |
2c48b9c4 AV |
2702 | path.dentry = d_alloc(root, &this); |
2703 | if (!path.dentry) | |
1da177e4 | 2704 | goto put_memory; |
2c48b9c4 | 2705 | path.mnt = mntget(shm_mnt); |
1da177e4 | 2706 | |
1da177e4 | 2707 | error = -ENOSPC; |
454abafe | 2708 | inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 2709 | if (!inode) |
4b42af81 | 2710 | goto put_dentry; |
1da177e4 | 2711 | |
2c48b9c4 | 2712 | d_instantiate(path.dentry, inode); |
1da177e4 | 2713 | inode->i_size = size; |
6d6b77f1 | 2714 | clear_nlink(inode); /* It is unlinked */ |
853ac43a MM |
2715 | #ifndef CONFIG_MMU |
2716 | error = ramfs_nommu_expand_for_mapping(inode, size); | |
2717 | if (error) | |
4b42af81 | 2718 | goto put_dentry; |
853ac43a | 2719 | #endif |
4b42af81 AV |
2720 | |
2721 | error = -ENFILE; | |
2c48b9c4 | 2722 | file = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 AV |
2723 | &shmem_file_operations); |
2724 | if (!file) | |
2725 | goto put_dentry; | |
2726 | ||
1da177e4 LT |
2727 | return file; |
2728 | ||
1da177e4 | 2729 | put_dentry: |
2c48b9c4 | 2730 | path_put(&path); |
1da177e4 LT |
2731 | put_memory: |
2732 | shmem_unacct_size(flags, size); | |
2733 | return ERR_PTR(error); | |
2734 | } | |
395e0ddc | 2735 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 2736 | |
46711810 | 2737 | /** |
1da177e4 | 2738 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
2739 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
2740 | */ | |
2741 | int shmem_zero_setup(struct vm_area_struct *vma) | |
2742 | { | |
2743 | struct file *file; | |
2744 | loff_t size = vma->vm_end - vma->vm_start; | |
2745 | ||
2746 | file = shmem_file_setup("dev/zero", size, vma->vm_flags); | |
2747 | if (IS_ERR(file)) | |
2748 | return PTR_ERR(file); | |
2749 | ||
2750 | if (vma->vm_file) | |
2751 | fput(vma->vm_file); | |
2752 | vma->vm_file = file; | |
2753 | vma->vm_ops = &shmem_vm_ops; | |
bee4c36a | 2754 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
2755 | return 0; |
2756 | } | |
d9d90e5e HD |
2757 | |
2758 | /** | |
2759 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
2760 | * @mapping: the page's address_space | |
2761 | * @index: the page index | |
2762 | * @gfp: the page allocator flags to use if allocating | |
2763 | * | |
2764 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
2765 | * with any new page allocations done using the specified allocation flags. | |
2766 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
2767 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
2768 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
2769 | * | |
68da9f05 HD |
2770 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
2771 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
2772 | */ |
2773 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
2774 | pgoff_t index, gfp_t gfp) | |
2775 | { | |
68da9f05 HD |
2776 | #ifdef CONFIG_SHMEM |
2777 | struct inode *inode = mapping->host; | |
9276aad6 | 2778 | struct page *page; |
68da9f05 HD |
2779 | int error; |
2780 | ||
2781 | BUG_ON(mapping->a_ops != &shmem_aops); | |
2782 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL); | |
2783 | if (error) | |
2784 | page = ERR_PTR(error); | |
2785 | else | |
2786 | unlock_page(page); | |
2787 | return page; | |
2788 | #else | |
2789 | /* | |
2790 | * The tiny !SHMEM case uses ramfs without swap | |
2791 | */ | |
d9d90e5e | 2792 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 2793 | #endif |
d9d90e5e HD |
2794 | } |
2795 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |