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> |
83e4fa9c | 56 | #include <linux/falloc.h> |
708e3508 | 57 | #include <linux/splice.h> |
1da177e4 LT |
58 | #include <linux/security.h> |
59 | #include <linux/swapops.h> | |
60 | #include <linux/mempolicy.h> | |
61 | #include <linux/namei.h> | |
b00dc3ad | 62 | #include <linux/ctype.h> |
304dbdb7 | 63 | #include <linux/migrate.h> |
c1f60a5a | 64 | #include <linux/highmem.h> |
680d794b | 65 | #include <linux/seq_file.h> |
92562927 | 66 | #include <linux/magic.h> |
304dbdb7 | 67 | |
1da177e4 | 68 | #include <asm/uaccess.h> |
1da177e4 LT |
69 | #include <asm/pgtable.h> |
70 | ||
caefba17 | 71 | #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) |
1da177e4 LT |
72 | #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) |
73 | ||
1da177e4 LT |
74 | /* Pretend that each entry is of this size in directory's i_size */ |
75 | #define BOGO_DIRENT_SIZE 20 | |
76 | ||
69f07ec9 HD |
77 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
78 | #define SHORT_SYMLINK_LEN 128 | |
79 | ||
b09e0fa4 EP |
80 | struct shmem_xattr { |
81 | struct list_head list; /* anchored by shmem_inode_info->xattr_list */ | |
82 | char *name; /* xattr name */ | |
83 | size_t size; | |
84 | char value[0]; | |
85 | }; | |
86 | ||
285b2c4f | 87 | /* Flag allocation requirements to shmem_getpage */ |
1da177e4 | 88 | enum sgp_type { |
1da177e4 LT |
89 | SGP_READ, /* don't exceed i_size, don't allocate page */ |
90 | SGP_CACHE, /* don't exceed i_size, may allocate page */ | |
a0ee5ec5 | 91 | SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ |
1635f6a7 HD |
92 | SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */ |
93 | SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */ | |
1da177e4 LT |
94 | }; |
95 | ||
b76db735 | 96 | #ifdef CONFIG_TMPFS |
680d794b | 97 | static unsigned long shmem_default_max_blocks(void) |
98 | { | |
99 | return totalram_pages / 2; | |
100 | } | |
101 | ||
102 | static unsigned long shmem_default_max_inodes(void) | |
103 | { | |
104 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
105 | } | |
b76db735 | 106 | #endif |
680d794b | 107 | |
bde05d1c HD |
108 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
109 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
110 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 HD |
111 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
112 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type); | |
113 | ||
114 | static inline int shmem_getpage(struct inode *inode, pgoff_t index, | |
115 | struct page **pagep, enum sgp_type sgp, int *fault_type) | |
116 | { | |
117 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
118 | mapping_gfp_mask(inode->i_mapping), fault_type); | |
119 | } | |
1da177e4 | 120 | |
1da177e4 LT |
121 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
122 | { | |
123 | return sb->s_fs_info; | |
124 | } | |
125 | ||
126 | /* | |
127 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
128 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
129 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
130 | * consistent with the pre-accounting of private mappings ... | |
131 | */ | |
132 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
133 | { | |
0b0a0806 | 134 | return (flags & VM_NORESERVE) ? |
191c5424 | 135 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
136 | } |
137 | ||
138 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
139 | { | |
0b0a0806 | 140 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
141 | vm_unacct_memory(VM_ACCT(size)); |
142 | } | |
143 | ||
144 | /* | |
145 | * ... whereas tmpfs objects are accounted incrementally as | |
146 | * pages are allocated, in order to allow huge sparse files. | |
147 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, | |
148 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
149 | */ | |
150 | static inline int shmem_acct_block(unsigned long flags) | |
151 | { | |
0b0a0806 | 152 | return (flags & VM_NORESERVE) ? |
191c5424 | 153 | security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0; |
1da177e4 LT |
154 | } |
155 | ||
156 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
157 | { | |
0b0a0806 | 158 | if (flags & VM_NORESERVE) |
1da177e4 LT |
159 | vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); |
160 | } | |
161 | ||
759b9775 | 162 | static const struct super_operations shmem_ops; |
f5e54d6e | 163 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 164 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
165 | static const struct inode_operations shmem_inode_operations; |
166 | static const struct inode_operations shmem_dir_inode_operations; | |
167 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 168 | static const struct vm_operations_struct shmem_vm_ops; |
1da177e4 | 169 | |
6c231b7b | 170 | static struct backing_dev_info shmem_backing_dev_info __read_mostly = { |
1da177e4 | 171 | .ra_pages = 0, /* No readahead */ |
4f98a2fe | 172 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED, |
1da177e4 LT |
173 | }; |
174 | ||
175 | static LIST_HEAD(shmem_swaplist); | |
cb5f7b9a | 176 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 177 | |
5b04c689 PE |
178 | static int shmem_reserve_inode(struct super_block *sb) |
179 | { | |
180 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
181 | if (sbinfo->max_inodes) { | |
182 | spin_lock(&sbinfo->stat_lock); | |
183 | if (!sbinfo->free_inodes) { | |
184 | spin_unlock(&sbinfo->stat_lock); | |
185 | return -ENOSPC; | |
186 | } | |
187 | sbinfo->free_inodes--; | |
188 | spin_unlock(&sbinfo->stat_lock); | |
189 | } | |
190 | return 0; | |
191 | } | |
192 | ||
193 | static void shmem_free_inode(struct super_block *sb) | |
194 | { | |
195 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
196 | if (sbinfo->max_inodes) { | |
197 | spin_lock(&sbinfo->stat_lock); | |
198 | sbinfo->free_inodes++; | |
199 | spin_unlock(&sbinfo->stat_lock); | |
200 | } | |
201 | } | |
202 | ||
46711810 | 203 | /** |
41ffe5d5 | 204 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
205 | * @inode: inode to recalc |
206 | * | |
207 | * We have to calculate the free blocks since the mm can drop | |
208 | * undirtied hole pages behind our back. | |
209 | * | |
210 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
211 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
212 | * | |
213 | * It has to be called with the spinlock held. | |
214 | */ | |
215 | static void shmem_recalc_inode(struct inode *inode) | |
216 | { | |
217 | struct shmem_inode_info *info = SHMEM_I(inode); | |
218 | long freed; | |
219 | ||
220 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
221 | if (freed > 0) { | |
54af6042 HD |
222 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
223 | if (sbinfo->max_blocks) | |
224 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 225 | info->alloced -= freed; |
54af6042 | 226 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 227 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
228 | } |
229 | } | |
230 | ||
7a5d0fbb HD |
231 | /* |
232 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
233 | */ | |
234 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
235 | pgoff_t index, void *expected, void *replacement) | |
236 | { | |
237 | void **pslot; | |
238 | void *item = NULL; | |
239 | ||
240 | VM_BUG_ON(!expected); | |
241 | pslot = radix_tree_lookup_slot(&mapping->page_tree, index); | |
242 | if (pslot) | |
243 | item = radix_tree_deref_slot_protected(pslot, | |
244 | &mapping->tree_lock); | |
245 | if (item != expected) | |
246 | return -ENOENT; | |
247 | if (replacement) | |
248 | radix_tree_replace_slot(pslot, replacement); | |
249 | else | |
250 | radix_tree_delete(&mapping->page_tree, index); | |
251 | return 0; | |
252 | } | |
253 | ||
46f65ec1 HD |
254 | /* |
255 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
256 | */ | |
257 | static int shmem_add_to_page_cache(struct page *page, | |
258 | struct address_space *mapping, | |
259 | pgoff_t index, gfp_t gfp, void *expected) | |
260 | { | |
aa3b1895 | 261 | int error = 0; |
46f65ec1 HD |
262 | |
263 | VM_BUG_ON(!PageLocked(page)); | |
264 | VM_BUG_ON(!PageSwapBacked(page)); | |
265 | ||
46f65ec1 HD |
266 | if (!expected) |
267 | error = radix_tree_preload(gfp & GFP_RECLAIM_MASK); | |
268 | if (!error) { | |
269 | page_cache_get(page); | |
270 | page->mapping = mapping; | |
271 | page->index = index; | |
272 | ||
273 | spin_lock_irq(&mapping->tree_lock); | |
274 | if (!expected) | |
275 | error = radix_tree_insert(&mapping->page_tree, | |
276 | index, page); | |
277 | else | |
278 | error = shmem_radix_tree_replace(mapping, index, | |
279 | expected, page); | |
280 | if (!error) { | |
281 | mapping->nrpages++; | |
282 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
283 | __inc_zone_page_state(page, NR_SHMEM); | |
284 | spin_unlock_irq(&mapping->tree_lock); | |
285 | } else { | |
286 | page->mapping = NULL; | |
287 | spin_unlock_irq(&mapping->tree_lock); | |
288 | page_cache_release(page); | |
289 | } | |
290 | if (!expected) | |
291 | radix_tree_preload_end(); | |
292 | } | |
293 | if (error) | |
294 | mem_cgroup_uncharge_cache_page(page); | |
46f65ec1 HD |
295 | return error; |
296 | } | |
297 | ||
6922c0c7 HD |
298 | /* |
299 | * Like delete_from_page_cache, but substitutes swap for page. | |
300 | */ | |
301 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
302 | { | |
303 | struct address_space *mapping = page->mapping; | |
304 | int error; | |
305 | ||
306 | spin_lock_irq(&mapping->tree_lock); | |
307 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
308 | page->mapping = NULL; | |
309 | mapping->nrpages--; | |
310 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
311 | __dec_zone_page_state(page, NR_SHMEM); | |
312 | spin_unlock_irq(&mapping->tree_lock); | |
313 | page_cache_release(page); | |
314 | BUG_ON(error); | |
315 | } | |
316 | ||
7a5d0fbb HD |
317 | /* |
318 | * Like find_get_pages, but collecting swap entries as well as pages. | |
319 | */ | |
320 | static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping, | |
321 | pgoff_t start, unsigned int nr_pages, | |
322 | struct page **pages, pgoff_t *indices) | |
323 | { | |
324 | unsigned int i; | |
325 | unsigned int ret; | |
326 | unsigned int nr_found; | |
327 | ||
328 | rcu_read_lock(); | |
329 | restart: | |
330 | nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, | |
331 | (void ***)pages, indices, start, nr_pages); | |
332 | ret = 0; | |
333 | for (i = 0; i < nr_found; i++) { | |
334 | struct page *page; | |
335 | repeat: | |
336 | page = radix_tree_deref_slot((void **)pages[i]); | |
337 | if (unlikely(!page)) | |
338 | continue; | |
339 | if (radix_tree_exception(page)) { | |
8079b1c8 HD |
340 | if (radix_tree_deref_retry(page)) |
341 | goto restart; | |
342 | /* | |
343 | * Otherwise, we must be storing a swap entry | |
344 | * here as an exceptional entry: so return it | |
345 | * without attempting to raise page count. | |
346 | */ | |
347 | goto export; | |
7a5d0fbb HD |
348 | } |
349 | if (!page_cache_get_speculative(page)) | |
350 | goto repeat; | |
351 | ||
352 | /* Has the page moved? */ | |
353 | if (unlikely(page != *((void **)pages[i]))) { | |
354 | page_cache_release(page); | |
355 | goto repeat; | |
356 | } | |
357 | export: | |
358 | indices[ret] = indices[i]; | |
359 | pages[ret] = page; | |
360 | ret++; | |
361 | } | |
362 | if (unlikely(!ret && nr_found)) | |
363 | goto restart; | |
364 | rcu_read_unlock(); | |
365 | return ret; | |
366 | } | |
367 | ||
368 | /* | |
369 | * Remove swap entry from radix tree, free the swap and its page cache. | |
370 | */ | |
371 | static int shmem_free_swap(struct address_space *mapping, | |
372 | pgoff_t index, void *radswap) | |
373 | { | |
374 | int error; | |
375 | ||
376 | spin_lock_irq(&mapping->tree_lock); | |
377 | error = shmem_radix_tree_replace(mapping, index, radswap, NULL); | |
378 | spin_unlock_irq(&mapping->tree_lock); | |
379 | if (!error) | |
380 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
381 | return error; | |
382 | } | |
383 | ||
384 | /* | |
385 | * Pagevec may contain swap entries, so shuffle up pages before releasing. | |
386 | */ | |
24513264 | 387 | static void shmem_deswap_pagevec(struct pagevec *pvec) |
7a5d0fbb HD |
388 | { |
389 | int i, j; | |
390 | ||
391 | for (i = 0, j = 0; i < pagevec_count(pvec); i++) { | |
392 | struct page *page = pvec->pages[i]; | |
393 | if (!radix_tree_exceptional_entry(page)) | |
394 | pvec->pages[j++] = page; | |
395 | } | |
396 | pvec->nr = j; | |
24513264 HD |
397 | } |
398 | ||
399 | /* | |
400 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
401 | */ | |
402 | void shmem_unlock_mapping(struct address_space *mapping) | |
403 | { | |
404 | struct pagevec pvec; | |
405 | pgoff_t indices[PAGEVEC_SIZE]; | |
406 | pgoff_t index = 0; | |
407 | ||
408 | pagevec_init(&pvec, 0); | |
409 | /* | |
410 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
411 | */ | |
412 | while (!mapping_unevictable(mapping)) { | |
413 | /* | |
414 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
415 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
416 | */ | |
417 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, | |
418 | PAGEVEC_SIZE, pvec.pages, indices); | |
419 | if (!pvec.nr) | |
420 | break; | |
421 | index = indices[pvec.nr - 1] + 1; | |
422 | shmem_deswap_pagevec(&pvec); | |
423 | check_move_unevictable_pages(pvec.pages, pvec.nr); | |
424 | pagevec_release(&pvec); | |
425 | cond_resched(); | |
426 | } | |
7a5d0fbb HD |
427 | } |
428 | ||
429 | /* | |
430 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 431 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 432 | */ |
1635f6a7 HD |
433 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
434 | bool unfalloc) | |
1da177e4 | 435 | { |
285b2c4f | 436 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 437 | struct shmem_inode_info *info = SHMEM_I(inode); |
285b2c4f | 438 | pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
83e4fa9c HD |
439 | pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT; |
440 | unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1); | |
441 | unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); | |
bda97eab | 442 | struct pagevec pvec; |
7a5d0fbb HD |
443 | pgoff_t indices[PAGEVEC_SIZE]; |
444 | long nr_swaps_freed = 0; | |
285b2c4f | 445 | pgoff_t index; |
bda97eab HD |
446 | int i; |
447 | ||
83e4fa9c HD |
448 | if (lend == -1) |
449 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
450 | |
451 | pagevec_init(&pvec, 0); | |
452 | index = start; | |
83e4fa9c | 453 | while (index < end) { |
7a5d0fbb | 454 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, |
83e4fa9c | 455 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
7a5d0fbb HD |
456 | pvec.pages, indices); |
457 | if (!pvec.nr) | |
458 | break; | |
bda97eab HD |
459 | mem_cgroup_uncharge_start(); |
460 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
461 | struct page *page = pvec.pages[i]; | |
462 | ||
7a5d0fbb | 463 | index = indices[i]; |
83e4fa9c | 464 | if (index >= end) |
bda97eab HD |
465 | break; |
466 | ||
7a5d0fbb | 467 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
468 | if (unfalloc) |
469 | continue; | |
7a5d0fbb HD |
470 | nr_swaps_freed += !shmem_free_swap(mapping, |
471 | index, page); | |
bda97eab | 472 | continue; |
7a5d0fbb HD |
473 | } |
474 | ||
475 | if (!trylock_page(page)) | |
bda97eab | 476 | continue; |
1635f6a7 HD |
477 | if (!unfalloc || !PageUptodate(page)) { |
478 | if (page->mapping == mapping) { | |
479 | VM_BUG_ON(PageWriteback(page)); | |
480 | truncate_inode_page(mapping, page); | |
481 | } | |
bda97eab | 482 | } |
bda97eab HD |
483 | unlock_page(page); |
484 | } | |
24513264 HD |
485 | shmem_deswap_pagevec(&pvec); |
486 | pagevec_release(&pvec); | |
bda97eab HD |
487 | mem_cgroup_uncharge_end(); |
488 | cond_resched(); | |
489 | index++; | |
490 | } | |
1da177e4 | 491 | |
83e4fa9c | 492 | if (partial_start) { |
bda97eab HD |
493 | struct page *page = NULL; |
494 | shmem_getpage(inode, start - 1, &page, SGP_READ, NULL); | |
495 | if (page) { | |
83e4fa9c HD |
496 | unsigned int top = PAGE_CACHE_SIZE; |
497 | if (start > end) { | |
498 | top = partial_end; | |
499 | partial_end = 0; | |
500 | } | |
501 | zero_user_segment(page, partial_start, top); | |
502 | set_page_dirty(page); | |
503 | unlock_page(page); | |
504 | page_cache_release(page); | |
505 | } | |
506 | } | |
507 | if (partial_end) { | |
508 | struct page *page = NULL; | |
509 | shmem_getpage(inode, end, &page, SGP_READ, NULL); | |
510 | if (page) { | |
511 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
512 | set_page_dirty(page); |
513 | unlock_page(page); | |
514 | page_cache_release(page); | |
515 | } | |
516 | } | |
83e4fa9c HD |
517 | if (start >= end) |
518 | return; | |
bda97eab HD |
519 | |
520 | index = start; | |
521 | for ( ; ; ) { | |
522 | cond_resched(); | |
7a5d0fbb | 523 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, |
83e4fa9c | 524 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
7a5d0fbb HD |
525 | pvec.pages, indices); |
526 | if (!pvec.nr) { | |
1635f6a7 | 527 | if (index == start || unfalloc) |
bda97eab HD |
528 | break; |
529 | index = start; | |
530 | continue; | |
531 | } | |
1635f6a7 | 532 | if ((index == start || unfalloc) && indices[0] >= end) { |
24513264 HD |
533 | shmem_deswap_pagevec(&pvec); |
534 | pagevec_release(&pvec); | |
bda97eab HD |
535 | break; |
536 | } | |
537 | mem_cgroup_uncharge_start(); | |
538 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
539 | struct page *page = pvec.pages[i]; | |
540 | ||
7a5d0fbb | 541 | index = indices[i]; |
83e4fa9c | 542 | if (index >= end) |
bda97eab HD |
543 | break; |
544 | ||
7a5d0fbb | 545 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
546 | if (unfalloc) |
547 | continue; | |
7a5d0fbb HD |
548 | nr_swaps_freed += !shmem_free_swap(mapping, |
549 | index, page); | |
550 | continue; | |
551 | } | |
552 | ||
bda97eab | 553 | lock_page(page); |
1635f6a7 HD |
554 | if (!unfalloc || !PageUptodate(page)) { |
555 | if (page->mapping == mapping) { | |
556 | VM_BUG_ON(PageWriteback(page)); | |
557 | truncate_inode_page(mapping, page); | |
558 | } | |
7a5d0fbb | 559 | } |
bda97eab HD |
560 | unlock_page(page); |
561 | } | |
24513264 HD |
562 | shmem_deswap_pagevec(&pvec); |
563 | pagevec_release(&pvec); | |
bda97eab HD |
564 | mem_cgroup_uncharge_end(); |
565 | index++; | |
566 | } | |
94c1e62d | 567 | |
1da177e4 | 568 | spin_lock(&info->lock); |
7a5d0fbb | 569 | info->swapped -= nr_swaps_freed; |
1da177e4 LT |
570 | shmem_recalc_inode(inode); |
571 | spin_unlock(&info->lock); | |
1635f6a7 | 572 | } |
1da177e4 | 573 | |
1635f6a7 HD |
574 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
575 | { | |
576 | shmem_undo_range(inode, lstart, lend, false); | |
285b2c4f | 577 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1da177e4 | 578 | } |
94c1e62d | 579 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 580 | |
94c1e62d | 581 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 LT |
582 | { |
583 | struct inode *inode = dentry->d_inode; | |
1da177e4 LT |
584 | int error; |
585 | ||
db78b877 CH |
586 | error = inode_change_ok(inode, attr); |
587 | if (error) | |
588 | return error; | |
589 | ||
94c1e62d HD |
590 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
591 | loff_t oldsize = inode->i_size; | |
592 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 593 | |
94c1e62d HD |
594 | if (newsize != oldsize) { |
595 | i_size_write(inode, newsize); | |
596 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
597 | } | |
598 | if (newsize < oldsize) { | |
599 | loff_t holebegin = round_up(newsize, PAGE_SIZE); | |
600 | unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); | |
601 | shmem_truncate_range(inode, newsize, (loff_t)-1); | |
602 | /* unmap again to remove racily COWed private pages */ | |
603 | unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); | |
604 | } | |
1da177e4 LT |
605 | } |
606 | ||
db78b877 | 607 | setattr_copy(inode, attr); |
39f0247d | 608 | #ifdef CONFIG_TMPFS_POSIX_ACL |
db78b877 | 609 | if (attr->ia_valid & ATTR_MODE) |
1c7c474c | 610 | error = generic_acl_chmod(inode); |
39f0247d | 611 | #endif |
1da177e4 LT |
612 | return error; |
613 | } | |
614 | ||
1f895f75 | 615 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 616 | { |
1da177e4 | 617 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 618 | struct shmem_xattr *xattr, *nxattr; |
1da177e4 | 619 | |
3889e6e7 | 620 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
621 | shmem_unacct_size(info->flags, inode->i_size); |
622 | inode->i_size = 0; | |
3889e6e7 | 623 | shmem_truncate_range(inode, 0, (loff_t)-1); |
1da177e4 | 624 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 625 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 626 | list_del_init(&info->swaplist); |
cb5f7b9a | 627 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 628 | } |
69f07ec9 HD |
629 | } else |
630 | kfree(info->symlink); | |
b09e0fa4 EP |
631 | |
632 | list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) { | |
633 | kfree(xattr->name); | |
634 | kfree(xattr); | |
635 | } | |
0edd73b3 | 636 | BUG_ON(inode->i_blocks); |
5b04c689 | 637 | shmem_free_inode(inode->i_sb); |
dbd5768f | 638 | clear_inode(inode); |
1da177e4 LT |
639 | } |
640 | ||
46f65ec1 HD |
641 | /* |
642 | * If swap found in inode, free it and move page from swapcache to filecache. | |
643 | */ | |
41ffe5d5 | 644 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 645 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 646 | { |
285b2c4f | 647 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 648 | void *radswap; |
41ffe5d5 | 649 | pgoff_t index; |
bde05d1c HD |
650 | gfp_t gfp; |
651 | int error = 0; | |
1da177e4 | 652 | |
46f65ec1 | 653 | radswap = swp_to_radix_entry(swap); |
e504f3fd | 654 | index = radix_tree_locate_item(&mapping->page_tree, radswap); |
46f65ec1 | 655 | if (index == -1) |
285b2c4f | 656 | return 0; |
2e0e26c7 | 657 | |
1b1b32f2 HD |
658 | /* |
659 | * Move _head_ to start search for next from here. | |
1f895f75 | 660 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 661 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 662 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
663 | */ |
664 | if (shmem_swaplist.next != &info->swaplist) | |
665 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 666 | |
bde05d1c HD |
667 | gfp = mapping_gfp_mask(mapping); |
668 | if (shmem_should_replace_page(*pagep, gfp)) { | |
669 | mutex_unlock(&shmem_swaplist_mutex); | |
670 | error = shmem_replace_page(pagep, gfp, info, index); | |
671 | mutex_lock(&shmem_swaplist_mutex); | |
672 | /* | |
673 | * We needed to drop mutex to make that restrictive page | |
674 | * allocation; but the inode might already be freed by now, | |
675 | * and we cannot refer to inode or mapping or info to check. | |
676 | * However, we do hold page lock on the PageSwapCache page, | |
677 | * so can check if that still has our reference remaining. | |
678 | */ | |
679 | if (!page_swapcount(*pagep)) | |
680 | error = -ENOENT; | |
681 | } | |
682 | ||
d13d1443 | 683 | /* |
778dd893 HD |
684 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
685 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
686 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 687 | */ |
bde05d1c HD |
688 | if (!error) |
689 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
46f65ec1 | 690 | GFP_NOWAIT, radswap); |
48f170fb | 691 | if (error != -ENOMEM) { |
46f65ec1 HD |
692 | /* |
693 | * Truncation and eviction use free_swap_and_cache(), which | |
694 | * only does trylock page: if we raced, best clean up here. | |
695 | */ | |
bde05d1c HD |
696 | delete_from_swap_cache(*pagep); |
697 | set_page_dirty(*pagep); | |
46f65ec1 HD |
698 | if (!error) { |
699 | spin_lock(&info->lock); | |
700 | info->swapped--; | |
701 | spin_unlock(&info->lock); | |
702 | swap_free(swap); | |
703 | } | |
2e0e26c7 | 704 | error = 1; /* not an error, but entry was found */ |
1da177e4 | 705 | } |
2e0e26c7 | 706 | return error; |
1da177e4 LT |
707 | } |
708 | ||
709 | /* | |
46f65ec1 | 710 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 711 | */ |
41ffe5d5 | 712 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 713 | { |
41ffe5d5 | 714 | struct list_head *this, *next; |
1da177e4 LT |
715 | struct shmem_inode_info *info; |
716 | int found = 0; | |
bde05d1c HD |
717 | int error = 0; |
718 | ||
719 | /* | |
720 | * There's a faint possibility that swap page was replaced before | |
721 | * caller locked it: it will come back later with the right page. | |
722 | */ | |
723 | if (unlikely(!PageSwapCache(page))) | |
724 | goto out; | |
778dd893 HD |
725 | |
726 | /* | |
727 | * Charge page using GFP_KERNEL while we can wait, before taking | |
728 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
729 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 HD |
730 | */ |
731 | error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); | |
732 | if (error) | |
733 | goto out; | |
46f65ec1 | 734 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
1da177e4 | 735 | |
cb5f7b9a | 736 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
737 | list_for_each_safe(this, next, &shmem_swaplist) { |
738 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 739 | if (info->swapped) |
bde05d1c | 740 | found = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
741 | else |
742 | list_del_init(&info->swaplist); | |
cb5f7b9a | 743 | cond_resched(); |
2e0e26c7 | 744 | if (found) |
778dd893 | 745 | break; |
1da177e4 | 746 | } |
cb5f7b9a | 747 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 748 | |
778dd893 HD |
749 | if (found < 0) |
750 | error = found; | |
751 | out: | |
aaa46865 HD |
752 | unlock_page(page); |
753 | page_cache_release(page); | |
778dd893 | 754 | return error; |
1da177e4 LT |
755 | } |
756 | ||
757 | /* | |
758 | * Move the page from the page cache to the swap cache. | |
759 | */ | |
760 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
761 | { | |
762 | struct shmem_inode_info *info; | |
1da177e4 | 763 | struct address_space *mapping; |
1da177e4 | 764 | struct inode *inode; |
6922c0c7 HD |
765 | swp_entry_t swap; |
766 | pgoff_t index; | |
1da177e4 LT |
767 | |
768 | BUG_ON(!PageLocked(page)); | |
1da177e4 LT |
769 | mapping = page->mapping; |
770 | index = page->index; | |
771 | inode = mapping->host; | |
772 | info = SHMEM_I(inode); | |
773 | if (info->flags & VM_LOCKED) | |
774 | goto redirty; | |
d9fe526a | 775 | if (!total_swap_pages) |
1da177e4 LT |
776 | goto redirty; |
777 | ||
d9fe526a HD |
778 | /* |
779 | * shmem_backing_dev_info's capabilities prevent regular writeback or | |
780 | * sync from ever calling shmem_writepage; but a stacking filesystem | |
48f170fb | 781 | * might use ->writepage of its underlying filesystem, in which case |
d9fe526a | 782 | * tmpfs should write out to swap only in response to memory pressure, |
48f170fb | 783 | * and not for the writeback threads or sync. |
d9fe526a | 784 | */ |
48f170fb HD |
785 | if (!wbc->for_reclaim) { |
786 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
787 | goto redirty; | |
788 | } | |
1635f6a7 HD |
789 | |
790 | /* | |
791 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
792 | * value into swapfile.c, the only way we can correctly account for a | |
793 | * fallocated page arriving here is now to initialize it and write it. | |
794 | */ | |
795 | if (!PageUptodate(page)) { | |
796 | clear_highpage(page); | |
797 | flush_dcache_page(page); | |
798 | SetPageUptodate(page); | |
799 | } | |
800 | ||
48f170fb HD |
801 | swap = get_swap_page(); |
802 | if (!swap.val) | |
803 | goto redirty; | |
d9fe526a | 804 | |
b1dea800 HD |
805 | /* |
806 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
807 | * if it's not already there. Do it now before the page is |
808 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 809 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
810 | * we've incremented swapped, because shmem_unuse_inode() will |
811 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 812 | */ |
48f170fb HD |
813 | mutex_lock(&shmem_swaplist_mutex); |
814 | if (list_empty(&info->swaplist)) | |
815 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 816 | |
48f170fb | 817 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
aaa46865 | 818 | swap_shmem_alloc(swap); |
6922c0c7 HD |
819 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); |
820 | ||
821 | spin_lock(&info->lock); | |
822 | info->swapped++; | |
823 | shmem_recalc_inode(inode); | |
826267cf | 824 | spin_unlock(&info->lock); |
6922c0c7 HD |
825 | |
826 | mutex_unlock(&shmem_swaplist_mutex); | |
d9fe526a | 827 | BUG_ON(page_mapped(page)); |
9fab5619 | 828 | swap_writepage(page, wbc); |
1da177e4 LT |
829 | return 0; |
830 | } | |
831 | ||
6922c0c7 | 832 | mutex_unlock(&shmem_swaplist_mutex); |
cb4b86ba | 833 | swapcache_free(swap, NULL); |
1da177e4 LT |
834 | redirty: |
835 | set_page_dirty(page); | |
d9fe526a HD |
836 | if (wbc->for_reclaim) |
837 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
838 | unlock_page(page); | |
839 | return 0; | |
1da177e4 LT |
840 | } |
841 | ||
842 | #ifdef CONFIG_NUMA | |
680d794b | 843 | #ifdef CONFIG_TMPFS |
71fe804b | 844 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 845 | { |
095f1fc4 | 846 | char buffer[64]; |
680d794b | 847 | |
71fe804b | 848 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 849 | return; /* show nothing */ |
680d794b | 850 | |
71fe804b | 851 | mpol_to_str(buffer, sizeof(buffer), mpol, 1); |
095f1fc4 LS |
852 | |
853 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 854 | } |
71fe804b LS |
855 | |
856 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
857 | { | |
858 | struct mempolicy *mpol = NULL; | |
859 | if (sbinfo->mpol) { | |
860 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
861 | mpol = sbinfo->mpol; | |
862 | mpol_get(mpol); | |
863 | spin_unlock(&sbinfo->stat_lock); | |
864 | } | |
865 | return mpol; | |
866 | } | |
680d794b | 867 | #endif /* CONFIG_TMPFS */ |
868 | ||
41ffe5d5 HD |
869 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
870 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 871 | { |
52cd3b07 | 872 | struct mempolicy mpol, *spol; |
1da177e4 LT |
873 | struct vm_area_struct pvma; |
874 | ||
52cd3b07 | 875 | spol = mpol_cond_copy(&mpol, |
41ffe5d5 | 876 | mpol_shared_policy_lookup(&info->policy, index)); |
52cd3b07 | 877 | |
1da177e4 | 878 | /* Create a pseudo vma that just contains the policy */ |
c4cc6d07 | 879 | pvma.vm_start = 0; |
41ffe5d5 | 880 | pvma.vm_pgoff = index; |
c4cc6d07 | 881 | pvma.vm_ops = NULL; |
52cd3b07 | 882 | pvma.vm_policy = spol; |
41ffe5d5 | 883 | return swapin_readahead(swap, gfp, &pvma, 0); |
1da177e4 LT |
884 | } |
885 | ||
02098fea | 886 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 887 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
888 | { |
889 | struct vm_area_struct pvma; | |
1da177e4 | 890 | |
c4cc6d07 HD |
891 | /* Create a pseudo vma that just contains the policy */ |
892 | pvma.vm_start = 0; | |
41ffe5d5 | 893 | pvma.vm_pgoff = index; |
c4cc6d07 | 894 | pvma.vm_ops = NULL; |
41ffe5d5 | 895 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
52cd3b07 LS |
896 | |
897 | /* | |
898 | * alloc_page_vma() will drop the shared policy reference | |
899 | */ | |
900 | return alloc_page_vma(gfp, &pvma, 0); | |
1da177e4 | 901 | } |
680d794b | 902 | #else /* !CONFIG_NUMA */ |
903 | #ifdef CONFIG_TMPFS | |
41ffe5d5 | 904 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 905 | { |
906 | } | |
907 | #endif /* CONFIG_TMPFS */ | |
908 | ||
41ffe5d5 HD |
909 | static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
910 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 911 | { |
41ffe5d5 | 912 | return swapin_readahead(swap, gfp, NULL, 0); |
1da177e4 LT |
913 | } |
914 | ||
02098fea | 915 | static inline struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 916 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 | 917 | { |
e84e2e13 | 918 | return alloc_page(gfp); |
1da177e4 | 919 | } |
680d794b | 920 | #endif /* CONFIG_NUMA */ |
1da177e4 | 921 | |
71fe804b LS |
922 | #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) |
923 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
924 | { | |
925 | return NULL; | |
926 | } | |
927 | #endif | |
928 | ||
bde05d1c HD |
929 | /* |
930 | * When a page is moved from swapcache to shmem filecache (either by the | |
931 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
932 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
933 | * ignorance of the mapping it belongs to. If that mapping has special | |
934 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
935 | * we may need to copy to a suitable page before moving to filecache. | |
936 | * | |
937 | * In a future release, this may well be extended to respect cpuset and | |
938 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
939 | * but for now it is a simple matter of zone. | |
940 | */ | |
941 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
942 | { | |
943 | return page_zonenum(page) > gfp_zone(gfp); | |
944 | } | |
945 | ||
946 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
947 | struct shmem_inode_info *info, pgoff_t index) | |
948 | { | |
949 | struct page *oldpage, *newpage; | |
950 | struct address_space *swap_mapping; | |
951 | pgoff_t swap_index; | |
952 | int error; | |
953 | ||
954 | oldpage = *pagep; | |
955 | swap_index = page_private(oldpage); | |
956 | swap_mapping = page_mapping(oldpage); | |
957 | ||
958 | /* | |
959 | * We have arrived here because our zones are constrained, so don't | |
960 | * limit chance of success by further cpuset and node constraints. | |
961 | */ | |
962 | gfp &= ~GFP_CONSTRAINT_MASK; | |
963 | newpage = shmem_alloc_page(gfp, info, index); | |
964 | if (!newpage) | |
965 | return -ENOMEM; | |
966 | VM_BUG_ON(shmem_should_replace_page(newpage, gfp)); | |
967 | ||
968 | *pagep = newpage; | |
969 | page_cache_get(newpage); | |
970 | copy_highpage(newpage, oldpage); | |
971 | ||
972 | VM_BUG_ON(!PageLocked(oldpage)); | |
973 | __set_page_locked(newpage); | |
974 | VM_BUG_ON(!PageUptodate(oldpage)); | |
975 | SetPageUptodate(newpage); | |
976 | VM_BUG_ON(!PageSwapBacked(oldpage)); | |
977 | SetPageSwapBacked(newpage); | |
978 | VM_BUG_ON(!swap_index); | |
979 | set_page_private(newpage, swap_index); | |
980 | VM_BUG_ON(!PageSwapCache(oldpage)); | |
981 | SetPageSwapCache(newpage); | |
982 | ||
983 | /* | |
984 | * Our caller will very soon move newpage out of swapcache, but it's | |
985 | * a nice clean interface for us to replace oldpage by newpage there. | |
986 | */ | |
987 | spin_lock_irq(&swap_mapping->tree_lock); | |
988 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
989 | newpage); | |
990 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
991 | __dec_zone_page_state(oldpage, NR_FILE_PAGES); | |
992 | spin_unlock_irq(&swap_mapping->tree_lock); | |
993 | BUG_ON(error); | |
994 | ||
995 | mem_cgroup_replace_page_cache(oldpage, newpage); | |
996 | lru_cache_add_anon(newpage); | |
997 | ||
998 | ClearPageSwapCache(oldpage); | |
999 | set_page_private(oldpage, 0); | |
1000 | ||
1001 | unlock_page(oldpage); | |
1002 | page_cache_release(oldpage); | |
1003 | page_cache_release(oldpage); | |
1004 | return 0; | |
1005 | } | |
1006 | ||
1da177e4 | 1007 | /* |
68da9f05 | 1008 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1009 | * |
1010 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1011 | * vm. If we swap it in we mark it dirty since we also free the swap | |
1012 | * entry since a page cannot live in both the swap and page cache | |
1013 | */ | |
41ffe5d5 | 1014 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
68da9f05 | 1015 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type) |
1da177e4 LT |
1016 | { |
1017 | struct address_space *mapping = inode->i_mapping; | |
54af6042 | 1018 | struct shmem_inode_info *info; |
1da177e4 | 1019 | struct shmem_sb_info *sbinfo; |
27ab7006 | 1020 | struct page *page; |
1da177e4 LT |
1021 | swp_entry_t swap; |
1022 | int error; | |
54af6042 | 1023 | int once = 0; |
1635f6a7 | 1024 | int alloced = 0; |
1da177e4 | 1025 | |
41ffe5d5 | 1026 | if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT)) |
1da177e4 | 1027 | return -EFBIG; |
1da177e4 | 1028 | repeat: |
54af6042 | 1029 | swap.val = 0; |
41ffe5d5 | 1030 | page = find_lock_page(mapping, index); |
54af6042 HD |
1031 | if (radix_tree_exceptional_entry(page)) { |
1032 | swap = radix_to_swp_entry(page); | |
1033 | page = NULL; | |
1034 | } | |
1035 | ||
1635f6a7 | 1036 | if (sgp != SGP_WRITE && sgp != SGP_FALLOC && |
54af6042 HD |
1037 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { |
1038 | error = -EINVAL; | |
1039 | goto failed; | |
1040 | } | |
1041 | ||
1635f6a7 HD |
1042 | /* fallocated page? */ |
1043 | if (page && !PageUptodate(page)) { | |
1044 | if (sgp != SGP_READ) | |
1045 | goto clear; | |
1046 | unlock_page(page); | |
1047 | page_cache_release(page); | |
1048 | page = NULL; | |
1049 | } | |
54af6042 | 1050 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1051 | *pagep = page; |
1052 | return 0; | |
27ab7006 HD |
1053 | } |
1054 | ||
1055 | /* | |
54af6042 HD |
1056 | * Fast cache lookup did not find it: |
1057 | * bring it back from swap or allocate. | |
27ab7006 | 1058 | */ |
54af6042 HD |
1059 | info = SHMEM_I(inode); |
1060 | sbinfo = SHMEM_SB(inode->i_sb); | |
1da177e4 | 1061 | |
1da177e4 LT |
1062 | if (swap.val) { |
1063 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1064 | page = lookup_swap_cache(swap); |
1065 | if (!page) { | |
1da177e4 | 1066 | /* here we actually do the io */ |
68da9f05 HD |
1067 | if (fault_type) |
1068 | *fault_type |= VM_FAULT_MAJOR; | |
41ffe5d5 | 1069 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1070 | if (!page) { |
54af6042 HD |
1071 | error = -ENOMEM; |
1072 | goto failed; | |
1da177e4 | 1073 | } |
1da177e4 LT |
1074 | } |
1075 | ||
1076 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1077 | lock_page(page); |
bde05d1c HD |
1078 | if (!PageSwapCache(page) || page->mapping) { |
1079 | error = -EEXIST; /* try again */ | |
1080 | goto failed; | |
1081 | } | |
27ab7006 | 1082 | if (!PageUptodate(page)) { |
1da177e4 | 1083 | error = -EIO; |
54af6042 | 1084 | goto failed; |
1da177e4 | 1085 | } |
54af6042 HD |
1086 | wait_on_page_writeback(page); |
1087 | ||
bde05d1c HD |
1088 | if (shmem_should_replace_page(page, gfp)) { |
1089 | error = shmem_replace_page(&page, gfp, info, index); | |
1090 | if (error) | |
1091 | goto failed; | |
1da177e4 | 1092 | } |
27ab7006 | 1093 | |
aa3b1895 HD |
1094 | error = mem_cgroup_cache_charge(page, current->mm, |
1095 | gfp & GFP_RECLAIM_MASK); | |
1096 | if (!error) | |
1097 | error = shmem_add_to_page_cache(page, mapping, index, | |
1098 | gfp, swp_to_radix_entry(swap)); | |
54af6042 HD |
1099 | if (error) |
1100 | goto failed; | |
1101 | ||
1102 | spin_lock(&info->lock); | |
285b2c4f | 1103 | info->swapped--; |
54af6042 | 1104 | shmem_recalc_inode(inode); |
27ab7006 | 1105 | spin_unlock(&info->lock); |
54af6042 HD |
1106 | |
1107 | delete_from_swap_cache(page); | |
27ab7006 HD |
1108 | set_page_dirty(page); |
1109 | swap_free(swap); | |
1110 | ||
54af6042 HD |
1111 | } else { |
1112 | if (shmem_acct_block(info->flags)) { | |
1113 | error = -ENOSPC; | |
1114 | goto failed; | |
1da177e4 | 1115 | } |
0edd73b3 | 1116 | if (sbinfo->max_blocks) { |
fc5da22a | 1117 | if (percpu_counter_compare(&sbinfo->used_blocks, |
54af6042 HD |
1118 | sbinfo->max_blocks) >= 0) { |
1119 | error = -ENOSPC; | |
1120 | goto unacct; | |
1121 | } | |
7e496299 | 1122 | percpu_counter_inc(&sbinfo->used_blocks); |
54af6042 | 1123 | } |
1da177e4 | 1124 | |
54af6042 HD |
1125 | page = shmem_alloc_page(gfp, info, index); |
1126 | if (!page) { | |
1127 | error = -ENOMEM; | |
1128 | goto decused; | |
1da177e4 LT |
1129 | } |
1130 | ||
54af6042 HD |
1131 | SetPageSwapBacked(page); |
1132 | __set_page_locked(page); | |
aa3b1895 HD |
1133 | error = mem_cgroup_cache_charge(page, current->mm, |
1134 | gfp & GFP_RECLAIM_MASK); | |
1135 | if (!error) | |
1136 | error = shmem_add_to_page_cache(page, mapping, index, | |
1137 | gfp, NULL); | |
54af6042 HD |
1138 | if (error) |
1139 | goto decused; | |
1140 | lru_cache_add_anon(page); | |
1141 | ||
1142 | spin_lock(&info->lock); | |
1da177e4 | 1143 | info->alloced++; |
54af6042 HD |
1144 | inode->i_blocks += BLOCKS_PER_PAGE; |
1145 | shmem_recalc_inode(inode); | |
1da177e4 | 1146 | spin_unlock(&info->lock); |
1635f6a7 | 1147 | alloced = true; |
54af6042 | 1148 | |
ec9516fb | 1149 | /* |
1635f6a7 HD |
1150 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1151 | */ | |
1152 | if (sgp == SGP_FALLOC) | |
1153 | sgp = SGP_WRITE; | |
1154 | clear: | |
1155 | /* | |
1156 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1157 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1158 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb HD |
1159 | */ |
1160 | if (sgp != SGP_WRITE) { | |
1161 | clear_highpage(page); | |
1162 | flush_dcache_page(page); | |
1163 | SetPageUptodate(page); | |
1164 | } | |
a0ee5ec5 | 1165 | if (sgp == SGP_DIRTY) |
27ab7006 | 1166 | set_page_dirty(page); |
1da177e4 | 1167 | } |
bde05d1c | 1168 | |
54af6042 | 1169 | /* Perhaps the file has been truncated since we checked */ |
1635f6a7 | 1170 | if (sgp != SGP_WRITE && sgp != SGP_FALLOC && |
54af6042 HD |
1171 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { |
1172 | error = -EINVAL; | |
1635f6a7 HD |
1173 | if (alloced) |
1174 | goto trunc; | |
1175 | else | |
1176 | goto failed; | |
e83c32e8 | 1177 | } |
54af6042 HD |
1178 | *pagep = page; |
1179 | return 0; | |
1da177e4 | 1180 | |
59a16ead | 1181 | /* |
54af6042 | 1182 | * Error recovery. |
59a16ead | 1183 | */ |
54af6042 | 1184 | trunc: |
1635f6a7 | 1185 | info = SHMEM_I(inode); |
54af6042 HD |
1186 | ClearPageDirty(page); |
1187 | delete_from_page_cache(page); | |
1188 | spin_lock(&info->lock); | |
1189 | info->alloced--; | |
1190 | inode->i_blocks -= BLOCKS_PER_PAGE; | |
59a16ead | 1191 | spin_unlock(&info->lock); |
54af6042 | 1192 | decused: |
1635f6a7 | 1193 | sbinfo = SHMEM_SB(inode->i_sb); |
54af6042 HD |
1194 | if (sbinfo->max_blocks) |
1195 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
1196 | unacct: | |
1197 | shmem_unacct_blocks(info->flags, 1); | |
1198 | failed: | |
1199 | if (swap.val && error != -EINVAL) { | |
1200 | struct page *test = find_get_page(mapping, index); | |
1201 | if (test && !radix_tree_exceptional_entry(test)) | |
1202 | page_cache_release(test); | |
1203 | /* Have another try if the entry has changed */ | |
1204 | if (test != swp_to_radix_entry(swap)) | |
1205 | error = -EEXIST; | |
1206 | } | |
27ab7006 | 1207 | if (page) { |
54af6042 | 1208 | unlock_page(page); |
27ab7006 | 1209 | page_cache_release(page); |
54af6042 HD |
1210 | } |
1211 | if (error == -ENOSPC && !once++) { | |
1212 | info = SHMEM_I(inode); | |
1213 | spin_lock(&info->lock); | |
1214 | shmem_recalc_inode(inode); | |
1215 | spin_unlock(&info->lock); | |
27ab7006 | 1216 | goto repeat; |
ff36b801 | 1217 | } |
54af6042 HD |
1218 | if (error == -EEXIST) |
1219 | goto repeat; | |
1220 | return error; | |
1da177e4 LT |
1221 | } |
1222 | ||
d0217ac0 | 1223 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1224 | { |
d3ac7f89 | 1225 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 | 1226 | int error; |
68da9f05 | 1227 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1228 | |
27d54b39 | 1229 | error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); |
d0217ac0 NP |
1230 | if (error) |
1231 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1232 | |
456f998e YH |
1233 | if (ret & VM_FAULT_MAJOR) { |
1234 | count_vm_event(PGMAJFAULT); | |
1235 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); | |
1236 | } | |
68da9f05 | 1237 | return ret; |
1da177e4 LT |
1238 | } |
1239 | ||
1da177e4 | 1240 | #ifdef CONFIG_NUMA |
41ffe5d5 | 1241 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 1242 | { |
41ffe5d5 HD |
1243 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1244 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); | |
1da177e4 LT |
1245 | } |
1246 | ||
d8dc74f2 AB |
1247 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
1248 | unsigned long addr) | |
1da177e4 | 1249 | { |
41ffe5d5 HD |
1250 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1251 | pgoff_t index; | |
1da177e4 | 1252 | |
41ffe5d5 HD |
1253 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
1254 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
1255 | } |
1256 | #endif | |
1257 | ||
1258 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
1259 | { | |
d3ac7f89 | 1260 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
1261 | struct shmem_inode_info *info = SHMEM_I(inode); |
1262 | int retval = -ENOMEM; | |
1263 | ||
1264 | spin_lock(&info->lock); | |
1265 | if (lock && !(info->flags & VM_LOCKED)) { | |
1266 | if (!user_shm_lock(inode->i_size, user)) | |
1267 | goto out_nomem; | |
1268 | info->flags |= VM_LOCKED; | |
89e004ea | 1269 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
1270 | } |
1271 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
1272 | user_shm_unlock(inode->i_size, user); | |
1273 | info->flags &= ~VM_LOCKED; | |
89e004ea | 1274 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
1275 | } |
1276 | retval = 0; | |
89e004ea | 1277 | |
1da177e4 LT |
1278 | out_nomem: |
1279 | spin_unlock(&info->lock); | |
1280 | return retval; | |
1281 | } | |
1282 | ||
9b83a6a8 | 1283 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
1284 | { |
1285 | file_accessed(file); | |
1286 | vma->vm_ops = &shmem_vm_ops; | |
d0217ac0 | 1287 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
1288 | return 0; |
1289 | } | |
1290 | ||
454abafe | 1291 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 1292 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
1293 | { |
1294 | struct inode *inode; | |
1295 | struct shmem_inode_info *info; | |
1296 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
1297 | ||
5b04c689 PE |
1298 | if (shmem_reserve_inode(sb)) |
1299 | return NULL; | |
1da177e4 LT |
1300 | |
1301 | inode = new_inode(sb); | |
1302 | if (inode) { | |
85fe4025 | 1303 | inode->i_ino = get_next_ino(); |
454abafe | 1304 | inode_init_owner(inode, dir, mode); |
1da177e4 | 1305 | inode->i_blocks = 0; |
1da177e4 LT |
1306 | inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; |
1307 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
91828a40 | 1308 | inode->i_generation = get_seconds(); |
1da177e4 LT |
1309 | info = SHMEM_I(inode); |
1310 | memset(info, 0, (char *)inode - (char *)info); | |
1311 | spin_lock_init(&info->lock); | |
0b0a0806 | 1312 | info->flags = flags & VM_NORESERVE; |
1da177e4 | 1313 | INIT_LIST_HEAD(&info->swaplist); |
b09e0fa4 | 1314 | INIT_LIST_HEAD(&info->xattr_list); |
72c04902 | 1315 | cache_no_acl(inode); |
1da177e4 LT |
1316 | |
1317 | switch (mode & S_IFMT) { | |
1318 | default: | |
39f0247d | 1319 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
1320 | init_special_inode(inode, mode, dev); |
1321 | break; | |
1322 | case S_IFREG: | |
14fcc23f | 1323 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
1324 | inode->i_op = &shmem_inode_operations; |
1325 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
1326 | mpol_shared_policy_init(&info->policy, |
1327 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
1328 | break; |
1329 | case S_IFDIR: | |
d8c76e6f | 1330 | inc_nlink(inode); |
1da177e4 LT |
1331 | /* Some things misbehave if size == 0 on a directory */ |
1332 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
1333 | inode->i_op = &shmem_dir_inode_operations; | |
1334 | inode->i_fop = &simple_dir_operations; | |
1335 | break; | |
1336 | case S_IFLNK: | |
1337 | /* | |
1338 | * Must not load anything in the rbtree, | |
1339 | * mpol_free_shared_policy will not be called. | |
1340 | */ | |
71fe804b | 1341 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
1342 | break; |
1343 | } | |
5b04c689 PE |
1344 | } else |
1345 | shmem_free_inode(sb); | |
1da177e4 LT |
1346 | return inode; |
1347 | } | |
1348 | ||
1349 | #ifdef CONFIG_TMPFS | |
92e1d5be | 1350 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 1351 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 1352 | |
6d9d88d0 JS |
1353 | #ifdef CONFIG_TMPFS_XATTR |
1354 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
1355 | #else | |
1356 | #define shmem_initxattrs NULL | |
1357 | #endif | |
1358 | ||
1da177e4 | 1359 | static int |
800d15a5 NP |
1360 | shmem_write_begin(struct file *file, struct address_space *mapping, |
1361 | loff_t pos, unsigned len, unsigned flags, | |
1362 | struct page **pagep, void **fsdata) | |
1da177e4 | 1363 | { |
800d15a5 NP |
1364 | struct inode *inode = mapping->host; |
1365 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
800d15a5 NP |
1366 | return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); |
1367 | } | |
1368 | ||
1369 | static int | |
1370 | shmem_write_end(struct file *file, struct address_space *mapping, | |
1371 | loff_t pos, unsigned len, unsigned copied, | |
1372 | struct page *page, void *fsdata) | |
1373 | { | |
1374 | struct inode *inode = mapping->host; | |
1375 | ||
d3602444 HD |
1376 | if (pos + copied > inode->i_size) |
1377 | i_size_write(inode, pos + copied); | |
1378 | ||
ec9516fb HD |
1379 | if (!PageUptodate(page)) { |
1380 | if (copied < PAGE_CACHE_SIZE) { | |
1381 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
1382 | zero_user_segments(page, 0, from, | |
1383 | from + copied, PAGE_CACHE_SIZE); | |
1384 | } | |
1385 | SetPageUptodate(page); | |
1386 | } | |
800d15a5 | 1387 | set_page_dirty(page); |
6746aff7 | 1388 | unlock_page(page); |
800d15a5 NP |
1389 | page_cache_release(page); |
1390 | ||
800d15a5 | 1391 | return copied; |
1da177e4 LT |
1392 | } |
1393 | ||
1da177e4 LT |
1394 | static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) |
1395 | { | |
d3ac7f89 | 1396 | struct inode *inode = filp->f_path.dentry->d_inode; |
1da177e4 | 1397 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
1398 | pgoff_t index; |
1399 | unsigned long offset; | |
a0ee5ec5 HD |
1400 | enum sgp_type sgp = SGP_READ; |
1401 | ||
1402 | /* | |
1403 | * Might this read be for a stacking filesystem? Then when reading | |
1404 | * holes of a sparse file, we actually need to allocate those pages, | |
1405 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
1406 | */ | |
1407 | if (segment_eq(get_fs(), KERNEL_DS)) | |
1408 | sgp = SGP_DIRTY; | |
1da177e4 LT |
1409 | |
1410 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1411 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1412 | ||
1413 | for (;;) { | |
1414 | struct page *page = NULL; | |
41ffe5d5 HD |
1415 | pgoff_t end_index; |
1416 | unsigned long nr, ret; | |
1da177e4 LT |
1417 | loff_t i_size = i_size_read(inode); |
1418 | ||
1419 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1420 | if (index > end_index) | |
1421 | break; | |
1422 | if (index == end_index) { | |
1423 | nr = i_size & ~PAGE_CACHE_MASK; | |
1424 | if (nr <= offset) | |
1425 | break; | |
1426 | } | |
1427 | ||
a0ee5ec5 | 1428 | desc->error = shmem_getpage(inode, index, &page, sgp, NULL); |
1da177e4 LT |
1429 | if (desc->error) { |
1430 | if (desc->error == -EINVAL) | |
1431 | desc->error = 0; | |
1432 | break; | |
1433 | } | |
d3602444 HD |
1434 | if (page) |
1435 | unlock_page(page); | |
1da177e4 LT |
1436 | |
1437 | /* | |
1438 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 1439 | * are called without i_mutex protection against truncate |
1da177e4 LT |
1440 | */ |
1441 | nr = PAGE_CACHE_SIZE; | |
1442 | i_size = i_size_read(inode); | |
1443 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1444 | if (index == end_index) { | |
1445 | nr = i_size & ~PAGE_CACHE_MASK; | |
1446 | if (nr <= offset) { | |
1447 | if (page) | |
1448 | page_cache_release(page); | |
1449 | break; | |
1450 | } | |
1451 | } | |
1452 | nr -= offset; | |
1453 | ||
1454 | if (page) { | |
1455 | /* | |
1456 | * If users can be writing to this page using arbitrary | |
1457 | * virtual addresses, take care about potential aliasing | |
1458 | * before reading the page on the kernel side. | |
1459 | */ | |
1460 | if (mapping_writably_mapped(mapping)) | |
1461 | flush_dcache_page(page); | |
1462 | /* | |
1463 | * Mark the page accessed if we read the beginning. | |
1464 | */ | |
1465 | if (!offset) | |
1466 | mark_page_accessed(page); | |
b5810039 | 1467 | } else { |
1da177e4 | 1468 | page = ZERO_PAGE(0); |
b5810039 NP |
1469 | page_cache_get(page); |
1470 | } | |
1da177e4 LT |
1471 | |
1472 | /* | |
1473 | * Ok, we have the page, and it's up-to-date, so | |
1474 | * now we can copy it to user space... | |
1475 | * | |
1476 | * The actor routine returns how many bytes were actually used.. | |
1477 | * NOTE! This may not be the same as how much of a user buffer | |
1478 | * we filled up (we may be padding etc), so we can only update | |
1479 | * "pos" here (the actor routine has to update the user buffer | |
1480 | * pointers and the remaining count). | |
1481 | */ | |
1482 | ret = actor(desc, page, offset, nr); | |
1483 | offset += ret; | |
1484 | index += offset >> PAGE_CACHE_SHIFT; | |
1485 | offset &= ~PAGE_CACHE_MASK; | |
1486 | ||
1487 | page_cache_release(page); | |
1488 | if (ret != nr || !desc->count) | |
1489 | break; | |
1490 | ||
1491 | cond_resched(); | |
1492 | } | |
1493 | ||
1494 | *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; | |
1495 | file_accessed(filp); | |
1496 | } | |
1497 | ||
bcd78e49 HD |
1498 | static ssize_t shmem_file_aio_read(struct kiocb *iocb, |
1499 | const struct iovec *iov, unsigned long nr_segs, loff_t pos) | |
1500 | { | |
1501 | struct file *filp = iocb->ki_filp; | |
1502 | ssize_t retval; | |
1503 | unsigned long seg; | |
1504 | size_t count; | |
1505 | loff_t *ppos = &iocb->ki_pos; | |
1506 | ||
1507 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); | |
1508 | if (retval) | |
1509 | return retval; | |
1510 | ||
1511 | for (seg = 0; seg < nr_segs; seg++) { | |
1512 | read_descriptor_t desc; | |
1513 | ||
1514 | desc.written = 0; | |
1515 | desc.arg.buf = iov[seg].iov_base; | |
1516 | desc.count = iov[seg].iov_len; | |
1517 | if (desc.count == 0) | |
1518 | continue; | |
1519 | desc.error = 0; | |
1520 | do_shmem_file_read(filp, ppos, &desc, file_read_actor); | |
1521 | retval += desc.written; | |
1522 | if (desc.error) { | |
1523 | retval = retval ?: desc.error; | |
1524 | break; | |
1525 | } | |
1526 | if (desc.count > 0) | |
1527 | break; | |
1528 | } | |
1529 | return retval; | |
1da177e4 LT |
1530 | } |
1531 | ||
708e3508 HD |
1532 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, |
1533 | struct pipe_inode_info *pipe, size_t len, | |
1534 | unsigned int flags) | |
1535 | { | |
1536 | struct address_space *mapping = in->f_mapping; | |
71f0e07a | 1537 | struct inode *inode = mapping->host; |
708e3508 HD |
1538 | unsigned int loff, nr_pages, req_pages; |
1539 | struct page *pages[PIPE_DEF_BUFFERS]; | |
1540 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
1541 | struct page *page; | |
1542 | pgoff_t index, end_index; | |
1543 | loff_t isize, left; | |
1544 | int error, page_nr; | |
1545 | struct splice_pipe_desc spd = { | |
1546 | .pages = pages, | |
1547 | .partial = partial, | |
1548 | .flags = flags, | |
1549 | .ops = &page_cache_pipe_buf_ops, | |
1550 | .spd_release = spd_release_page, | |
1551 | }; | |
1552 | ||
71f0e07a | 1553 | isize = i_size_read(inode); |
708e3508 HD |
1554 | if (unlikely(*ppos >= isize)) |
1555 | return 0; | |
1556 | ||
1557 | left = isize - *ppos; | |
1558 | if (unlikely(left < len)) | |
1559 | len = left; | |
1560 | ||
1561 | if (splice_grow_spd(pipe, &spd)) | |
1562 | return -ENOMEM; | |
1563 | ||
1564 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1565 | loff = *ppos & ~PAGE_CACHE_MASK; | |
1566 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
1567 | nr_pages = min(req_pages, pipe->buffers); | |
1568 | ||
708e3508 HD |
1569 | spd.nr_pages = find_get_pages_contig(mapping, index, |
1570 | nr_pages, spd.pages); | |
1571 | index += spd.nr_pages; | |
708e3508 | 1572 | error = 0; |
708e3508 | 1573 | |
71f0e07a | 1574 | while (spd.nr_pages < nr_pages) { |
71f0e07a HD |
1575 | error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL); |
1576 | if (error) | |
1577 | break; | |
1578 | unlock_page(page); | |
708e3508 HD |
1579 | spd.pages[spd.nr_pages++] = page; |
1580 | index++; | |
1581 | } | |
1582 | ||
708e3508 HD |
1583 | index = *ppos >> PAGE_CACHE_SHIFT; |
1584 | nr_pages = spd.nr_pages; | |
1585 | spd.nr_pages = 0; | |
71f0e07a | 1586 | |
708e3508 HD |
1587 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { |
1588 | unsigned int this_len; | |
1589 | ||
1590 | if (!len) | |
1591 | break; | |
1592 | ||
708e3508 HD |
1593 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); |
1594 | page = spd.pages[page_nr]; | |
1595 | ||
71f0e07a | 1596 | if (!PageUptodate(page) || page->mapping != mapping) { |
71f0e07a HD |
1597 | error = shmem_getpage(inode, index, &page, |
1598 | SGP_CACHE, NULL); | |
1599 | if (error) | |
708e3508 | 1600 | break; |
71f0e07a HD |
1601 | unlock_page(page); |
1602 | page_cache_release(spd.pages[page_nr]); | |
1603 | spd.pages[page_nr] = page; | |
708e3508 | 1604 | } |
71f0e07a HD |
1605 | |
1606 | isize = i_size_read(inode); | |
708e3508 HD |
1607 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; |
1608 | if (unlikely(!isize || index > end_index)) | |
1609 | break; | |
1610 | ||
708e3508 HD |
1611 | if (end_index == index) { |
1612 | unsigned int plen; | |
1613 | ||
708e3508 HD |
1614 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
1615 | if (plen <= loff) | |
1616 | break; | |
1617 | ||
708e3508 HD |
1618 | this_len = min(this_len, plen - loff); |
1619 | len = this_len; | |
1620 | } | |
1621 | ||
1622 | spd.partial[page_nr].offset = loff; | |
1623 | spd.partial[page_nr].len = this_len; | |
1624 | len -= this_len; | |
1625 | loff = 0; | |
1626 | spd.nr_pages++; | |
1627 | index++; | |
1628 | } | |
1629 | ||
708e3508 HD |
1630 | while (page_nr < nr_pages) |
1631 | page_cache_release(spd.pages[page_nr++]); | |
708e3508 HD |
1632 | |
1633 | if (spd.nr_pages) | |
1634 | error = splice_to_pipe(pipe, &spd); | |
1635 | ||
1636 | splice_shrink_spd(pipe, &spd); | |
1637 | ||
1638 | if (error > 0) { | |
1639 | *ppos += error; | |
1640 | file_accessed(in); | |
1641 | } | |
1642 | return error; | |
1643 | } | |
1644 | ||
83e4fa9c HD |
1645 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
1646 | loff_t len) | |
1647 | { | |
1648 | struct inode *inode = file->f_path.dentry->d_inode; | |
e2d12e22 HD |
1649 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1650 | pgoff_t start, index, end; | |
1651 | int error; | |
83e4fa9c HD |
1652 | |
1653 | mutex_lock(&inode->i_mutex); | |
1654 | ||
1655 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1656 | struct address_space *mapping = file->f_mapping; | |
1657 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
1658 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
1659 | ||
1660 | if ((u64)unmap_end > (u64)unmap_start) | |
1661 | unmap_mapping_range(mapping, unmap_start, | |
1662 | 1 + unmap_end - unmap_start, 0); | |
1663 | shmem_truncate_range(inode, offset, offset + len - 1); | |
1664 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
1665 | error = 0; | |
e2d12e22 HD |
1666 | goto out; |
1667 | } | |
1668 | ||
1669 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
1670 | error = inode_newsize_ok(inode, offset + len); | |
1671 | if (error) | |
1672 | goto out; | |
1673 | ||
1674 | start = offset >> PAGE_CACHE_SHIFT; | |
1675 | end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
1676 | /* Try to avoid a swapstorm if len is impossible to satisfy */ | |
1677 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
1678 | error = -ENOSPC; | |
1679 | goto out; | |
83e4fa9c HD |
1680 | } |
1681 | ||
e2d12e22 HD |
1682 | for (index = start; index < end; index++) { |
1683 | struct page *page; | |
1684 | ||
1685 | /* | |
1686 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
1687 | * been interrupted because we are using up too much memory. | |
1688 | */ | |
1689 | if (signal_pending(current)) | |
1690 | error = -EINTR; | |
1691 | else | |
1635f6a7 | 1692 | error = shmem_getpage(inode, index, &page, SGP_FALLOC, |
e2d12e22 HD |
1693 | NULL); |
1694 | if (error) { | |
1635f6a7 HD |
1695 | /* Remove the !PageUptodate pages we added */ |
1696 | shmem_undo_range(inode, | |
1697 | (loff_t)start << PAGE_CACHE_SHIFT, | |
1698 | (loff_t)index << PAGE_CACHE_SHIFT, true); | |
e2d12e22 HD |
1699 | goto ctime; |
1700 | } | |
1701 | ||
e2d12e22 | 1702 | /* |
1635f6a7 HD |
1703 | * If !PageUptodate, leave it that way so that freeable pages |
1704 | * can be recognized if we need to rollback on error later. | |
1705 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
1706 | * than free the pages we are allocating (and SGP_CACHE pages |
1707 | * might still be clean: we now need to mark those dirty too). | |
1708 | */ | |
1709 | set_page_dirty(page); | |
1710 | unlock_page(page); | |
1711 | page_cache_release(page); | |
1712 | cond_resched(); | |
1713 | } | |
1714 | ||
1715 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
1716 | i_size_write(inode, offset + len); | |
1717 | ctime: | |
1718 | inode->i_ctime = CURRENT_TIME; | |
1719 | out: | |
83e4fa9c HD |
1720 | mutex_unlock(&inode->i_mutex); |
1721 | return error; | |
1722 | } | |
1723 | ||
726c3342 | 1724 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 1725 | { |
726c3342 | 1726 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
1727 | |
1728 | buf->f_type = TMPFS_MAGIC; | |
1729 | buf->f_bsize = PAGE_CACHE_SIZE; | |
1730 | buf->f_namelen = NAME_MAX; | |
0edd73b3 | 1731 | if (sbinfo->max_blocks) { |
1da177e4 | 1732 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
1733 | buf->f_bavail = |
1734 | buf->f_bfree = sbinfo->max_blocks - | |
1735 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
1736 | } |
1737 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
1738 | buf->f_files = sbinfo->max_inodes; |
1739 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
1740 | } |
1741 | /* else leave those fields 0 like simple_statfs */ | |
1742 | return 0; | |
1743 | } | |
1744 | ||
1745 | /* | |
1746 | * File creation. Allocate an inode, and we're done.. | |
1747 | */ | |
1748 | static int | |
1a67aafb | 1749 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 1750 | { |
0b0a0806 | 1751 | struct inode *inode; |
1da177e4 LT |
1752 | int error = -ENOSPC; |
1753 | ||
454abafe | 1754 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 1755 | if (inode) { |
2a7dba39 | 1756 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 1757 | &dentry->d_name, |
6d9d88d0 | 1758 | shmem_initxattrs, NULL); |
570bc1c2 SS |
1759 | if (error) { |
1760 | if (error != -EOPNOTSUPP) { | |
1761 | iput(inode); | |
1762 | return error; | |
1763 | } | |
39f0247d | 1764 | } |
1c7c474c CH |
1765 | #ifdef CONFIG_TMPFS_POSIX_ACL |
1766 | error = generic_acl_init(inode, dir); | |
39f0247d AG |
1767 | if (error) { |
1768 | iput(inode); | |
1769 | return error; | |
570bc1c2 | 1770 | } |
718deb6b AV |
1771 | #else |
1772 | error = 0; | |
1c7c474c | 1773 | #endif |
1da177e4 LT |
1774 | dir->i_size += BOGO_DIRENT_SIZE; |
1775 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
1776 | d_instantiate(dentry, inode); | |
1777 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
1778 | } |
1779 | return error; | |
1780 | } | |
1781 | ||
18bb1db3 | 1782 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
1783 | { |
1784 | int error; | |
1785 | ||
1786 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
1787 | return error; | |
d8c76e6f | 1788 | inc_nlink(dir); |
1da177e4 LT |
1789 | return 0; |
1790 | } | |
1791 | ||
4acdaf27 | 1792 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
1da177e4 LT |
1793 | struct nameidata *nd) |
1794 | { | |
1795 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
1796 | } | |
1797 | ||
1798 | /* | |
1799 | * Link a file.. | |
1800 | */ | |
1801 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1802 | { | |
1803 | struct inode *inode = old_dentry->d_inode; | |
5b04c689 | 1804 | int ret; |
1da177e4 LT |
1805 | |
1806 | /* | |
1807 | * No ordinary (disk based) filesystem counts links as inodes; | |
1808 | * but each new link needs a new dentry, pinning lowmem, and | |
1809 | * tmpfs dentries cannot be pruned until they are unlinked. | |
1810 | */ | |
5b04c689 PE |
1811 | ret = shmem_reserve_inode(inode->i_sb); |
1812 | if (ret) | |
1813 | goto out; | |
1da177e4 LT |
1814 | |
1815 | dir->i_size += BOGO_DIRENT_SIZE; | |
1816 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 1817 | inc_nlink(inode); |
7de9c6ee | 1818 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
1819 | dget(dentry); /* Extra pinning count for the created dentry */ |
1820 | d_instantiate(dentry, inode); | |
5b04c689 PE |
1821 | out: |
1822 | return ret; | |
1da177e4 LT |
1823 | } |
1824 | ||
1825 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
1826 | { | |
1827 | struct inode *inode = dentry->d_inode; | |
1828 | ||
5b04c689 PE |
1829 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
1830 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
1831 | |
1832 | dir->i_size -= BOGO_DIRENT_SIZE; | |
1833 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 1834 | drop_nlink(inode); |
1da177e4 LT |
1835 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
1836 | return 0; | |
1837 | } | |
1838 | ||
1839 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
1840 | { | |
1841 | if (!simple_empty(dentry)) | |
1842 | return -ENOTEMPTY; | |
1843 | ||
9a53c3a7 DH |
1844 | drop_nlink(dentry->d_inode); |
1845 | drop_nlink(dir); | |
1da177e4 LT |
1846 | return shmem_unlink(dir, dentry); |
1847 | } | |
1848 | ||
1849 | /* | |
1850 | * The VFS layer already does all the dentry stuff for rename, | |
1851 | * we just have to decrement the usage count for the target if | |
1852 | * it exists so that the VFS layer correctly free's it when it | |
1853 | * gets overwritten. | |
1854 | */ | |
1855 | static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) | |
1856 | { | |
1857 | struct inode *inode = old_dentry->d_inode; | |
1858 | int they_are_dirs = S_ISDIR(inode->i_mode); | |
1859 | ||
1860 | if (!simple_empty(new_dentry)) | |
1861 | return -ENOTEMPTY; | |
1862 | ||
1863 | if (new_dentry->d_inode) { | |
1864 | (void) shmem_unlink(new_dir, new_dentry); | |
1865 | if (they_are_dirs) | |
9a53c3a7 | 1866 | drop_nlink(old_dir); |
1da177e4 | 1867 | } else if (they_are_dirs) { |
9a53c3a7 | 1868 | drop_nlink(old_dir); |
d8c76e6f | 1869 | inc_nlink(new_dir); |
1da177e4 LT |
1870 | } |
1871 | ||
1872 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
1873 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
1874 | old_dir->i_ctime = old_dir->i_mtime = | |
1875 | new_dir->i_ctime = new_dir->i_mtime = | |
1876 | inode->i_ctime = CURRENT_TIME; | |
1877 | return 0; | |
1878 | } | |
1879 | ||
1880 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1881 | { | |
1882 | int error; | |
1883 | int len; | |
1884 | struct inode *inode; | |
9276aad6 | 1885 | struct page *page; |
1da177e4 LT |
1886 | char *kaddr; |
1887 | struct shmem_inode_info *info; | |
1888 | ||
1889 | len = strlen(symname) + 1; | |
1890 | if (len > PAGE_CACHE_SIZE) | |
1891 | return -ENAMETOOLONG; | |
1892 | ||
454abafe | 1893 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
1894 | if (!inode) |
1895 | return -ENOSPC; | |
1896 | ||
9d8f13ba | 1897 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 1898 | shmem_initxattrs, NULL); |
570bc1c2 SS |
1899 | if (error) { |
1900 | if (error != -EOPNOTSUPP) { | |
1901 | iput(inode); | |
1902 | return error; | |
1903 | } | |
1904 | error = 0; | |
1905 | } | |
1906 | ||
1da177e4 LT |
1907 | info = SHMEM_I(inode); |
1908 | inode->i_size = len-1; | |
69f07ec9 HD |
1909 | if (len <= SHORT_SYMLINK_LEN) { |
1910 | info->symlink = kmemdup(symname, len, GFP_KERNEL); | |
1911 | if (!info->symlink) { | |
1912 | iput(inode); | |
1913 | return -ENOMEM; | |
1914 | } | |
1915 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 LT |
1916 | } else { |
1917 | error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); | |
1918 | if (error) { | |
1919 | iput(inode); | |
1920 | return error; | |
1921 | } | |
14fcc23f | 1922 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 1923 | inode->i_op = &shmem_symlink_inode_operations; |
9b04c5fe | 1924 | kaddr = kmap_atomic(page); |
1da177e4 | 1925 | memcpy(kaddr, symname, len); |
9b04c5fe | 1926 | kunmap_atomic(kaddr); |
ec9516fb | 1927 | SetPageUptodate(page); |
1da177e4 | 1928 | set_page_dirty(page); |
6746aff7 | 1929 | unlock_page(page); |
1da177e4 LT |
1930 | page_cache_release(page); |
1931 | } | |
1da177e4 LT |
1932 | dir->i_size += BOGO_DIRENT_SIZE; |
1933 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
1934 | d_instantiate(dentry, inode); | |
1935 | dget(dentry); | |
1936 | return 0; | |
1937 | } | |
1938 | ||
69f07ec9 | 1939 | static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 1940 | { |
69f07ec9 | 1941 | nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink); |
cc314eef | 1942 | return NULL; |
1da177e4 LT |
1943 | } |
1944 | ||
cc314eef | 1945 | static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) |
1da177e4 LT |
1946 | { |
1947 | struct page *page = NULL; | |
41ffe5d5 HD |
1948 | int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); |
1949 | nd_set_link(nd, error ? ERR_PTR(error) : kmap(page)); | |
d3602444 HD |
1950 | if (page) |
1951 | unlock_page(page); | |
cc314eef | 1952 | return page; |
1da177e4 LT |
1953 | } |
1954 | ||
cc314eef | 1955 | static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 LT |
1956 | { |
1957 | if (!IS_ERR(nd_get_link(nd))) { | |
cc314eef | 1958 | struct page *page = cookie; |
1da177e4 LT |
1959 | kunmap(page); |
1960 | mark_page_accessed(page); | |
1961 | page_cache_release(page); | |
1da177e4 LT |
1962 | } |
1963 | } | |
1964 | ||
b09e0fa4 | 1965 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 1966 | /* |
b09e0fa4 EP |
1967 | * Superblocks without xattr inode operations may get some security.* xattr |
1968 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
1969 | * like ACLs, we also need to implement the security.* handlers at |
1970 | * filesystem level, though. | |
1971 | */ | |
1972 | ||
6d9d88d0 JS |
1973 | /* |
1974 | * Allocate new xattr and copy in the value; but leave the name to callers. | |
1975 | */ | |
1976 | static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size) | |
1977 | { | |
1978 | struct shmem_xattr *new_xattr; | |
1979 | size_t len; | |
1980 | ||
1981 | /* wrap around? */ | |
1982 | len = sizeof(*new_xattr) + size; | |
1983 | if (len <= sizeof(*new_xattr)) | |
1984 | return NULL; | |
1985 | ||
1986 | new_xattr = kmalloc(len, GFP_KERNEL); | |
1987 | if (!new_xattr) | |
1988 | return NULL; | |
1989 | ||
1990 | new_xattr->size = size; | |
1991 | memcpy(new_xattr->value, value, size); | |
1992 | return new_xattr; | |
1993 | } | |
1994 | ||
1995 | /* | |
1996 | * Callback for security_inode_init_security() for acquiring xattrs. | |
1997 | */ | |
1998 | static int shmem_initxattrs(struct inode *inode, | |
1999 | const struct xattr *xattr_array, | |
2000 | void *fs_info) | |
2001 | { | |
2002 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2003 | const struct xattr *xattr; | |
2004 | struct shmem_xattr *new_xattr; | |
2005 | size_t len; | |
2006 | ||
2007 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
2008 | new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len); | |
2009 | if (!new_xattr) | |
2010 | return -ENOMEM; | |
2011 | ||
2012 | len = strlen(xattr->name) + 1; | |
2013 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
2014 | GFP_KERNEL); | |
2015 | if (!new_xattr->name) { | |
2016 | kfree(new_xattr); | |
2017 | return -ENOMEM; | |
2018 | } | |
2019 | ||
2020 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
2021 | XATTR_SECURITY_PREFIX_LEN); | |
2022 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
2023 | xattr->name, len); | |
2024 | ||
2025 | spin_lock(&info->lock); | |
2026 | list_add(&new_xattr->list, &info->xattr_list); | |
2027 | spin_unlock(&info->lock); | |
2028 | } | |
2029 | ||
2030 | return 0; | |
2031 | } | |
2032 | ||
b09e0fa4 EP |
2033 | static int shmem_xattr_get(struct dentry *dentry, const char *name, |
2034 | void *buffer, size_t size) | |
39f0247d | 2035 | { |
b09e0fa4 EP |
2036 | struct shmem_inode_info *info; |
2037 | struct shmem_xattr *xattr; | |
2038 | int ret = -ENODATA; | |
39f0247d | 2039 | |
b09e0fa4 EP |
2040 | info = SHMEM_I(dentry->d_inode); |
2041 | ||
2042 | spin_lock(&info->lock); | |
2043 | list_for_each_entry(xattr, &info->xattr_list, list) { | |
2044 | if (strcmp(name, xattr->name)) | |
2045 | continue; | |
2046 | ||
2047 | ret = xattr->size; | |
2048 | if (buffer) { | |
2049 | if (size < xattr->size) | |
2050 | ret = -ERANGE; | |
2051 | else | |
2052 | memcpy(buffer, xattr->value, xattr->size); | |
2053 | } | |
2054 | break; | |
2055 | } | |
2056 | spin_unlock(&info->lock); | |
2057 | return ret; | |
39f0247d AG |
2058 | } |
2059 | ||
6d9d88d0 | 2060 | static int shmem_xattr_set(struct inode *inode, const char *name, |
b09e0fa4 | 2061 | const void *value, size_t size, int flags) |
39f0247d | 2062 | { |
b09e0fa4 EP |
2063 | struct shmem_inode_info *info = SHMEM_I(inode); |
2064 | struct shmem_xattr *xattr; | |
2065 | struct shmem_xattr *new_xattr = NULL; | |
b09e0fa4 EP |
2066 | int err = 0; |
2067 | ||
2068 | /* value == NULL means remove */ | |
2069 | if (value) { | |
6d9d88d0 | 2070 | new_xattr = shmem_xattr_alloc(value, size); |
b09e0fa4 EP |
2071 | if (!new_xattr) |
2072 | return -ENOMEM; | |
2073 | ||
2074 | new_xattr->name = kstrdup(name, GFP_KERNEL); | |
2075 | if (!new_xattr->name) { | |
2076 | kfree(new_xattr); | |
2077 | return -ENOMEM; | |
2078 | } | |
b09e0fa4 EP |
2079 | } |
2080 | ||
2081 | spin_lock(&info->lock); | |
2082 | list_for_each_entry(xattr, &info->xattr_list, list) { | |
2083 | if (!strcmp(name, xattr->name)) { | |
2084 | if (flags & XATTR_CREATE) { | |
2085 | xattr = new_xattr; | |
2086 | err = -EEXIST; | |
2087 | } else if (new_xattr) { | |
2088 | list_replace(&xattr->list, &new_xattr->list); | |
2089 | } else { | |
2090 | list_del(&xattr->list); | |
2091 | } | |
2092 | goto out; | |
2093 | } | |
2094 | } | |
2095 | if (flags & XATTR_REPLACE) { | |
2096 | xattr = new_xattr; | |
2097 | err = -ENODATA; | |
2098 | } else { | |
2099 | list_add(&new_xattr->list, &info->xattr_list); | |
2100 | xattr = NULL; | |
2101 | } | |
2102 | out: | |
2103 | spin_unlock(&info->lock); | |
2104 | if (xattr) | |
2105 | kfree(xattr->name); | |
2106 | kfree(xattr); | |
2107 | return err; | |
39f0247d AG |
2108 | } |
2109 | ||
bb435453 | 2110 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
b09e0fa4 | 2111 | #ifdef CONFIG_TMPFS_POSIX_ACL |
1c7c474c CH |
2112 | &generic_acl_access_handler, |
2113 | &generic_acl_default_handler, | |
b09e0fa4 | 2114 | #endif |
39f0247d AG |
2115 | NULL |
2116 | }; | |
b09e0fa4 EP |
2117 | |
2118 | static int shmem_xattr_validate(const char *name) | |
2119 | { | |
2120 | struct { const char *prefix; size_t len; } arr[] = { | |
2121 | { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN }, | |
2122 | { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN } | |
2123 | }; | |
2124 | int i; | |
2125 | ||
2126 | for (i = 0; i < ARRAY_SIZE(arr); i++) { | |
2127 | size_t preflen = arr[i].len; | |
2128 | if (strncmp(name, arr[i].prefix, preflen) == 0) { | |
2129 | if (!name[preflen]) | |
2130 | return -EINVAL; | |
2131 | return 0; | |
2132 | } | |
2133 | } | |
2134 | return -EOPNOTSUPP; | |
2135 | } | |
2136 | ||
2137 | static ssize_t shmem_getxattr(struct dentry *dentry, const char *name, | |
2138 | void *buffer, size_t size) | |
2139 | { | |
2140 | int err; | |
2141 | ||
2142 | /* | |
2143 | * If this is a request for a synthetic attribute in the system.* | |
2144 | * namespace use the generic infrastructure to resolve a handler | |
2145 | * for it via sb->s_xattr. | |
2146 | */ | |
2147 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2148 | return generic_getxattr(dentry, name, buffer, size); | |
2149 | ||
2150 | err = shmem_xattr_validate(name); | |
2151 | if (err) | |
2152 | return err; | |
2153 | ||
2154 | return shmem_xattr_get(dentry, name, buffer, size); | |
2155 | } | |
2156 | ||
2157 | static int shmem_setxattr(struct dentry *dentry, const char *name, | |
2158 | const void *value, size_t size, int flags) | |
2159 | { | |
2160 | int err; | |
2161 | ||
2162 | /* | |
2163 | * If this is a request for a synthetic attribute in the system.* | |
2164 | * namespace use the generic infrastructure to resolve a handler | |
2165 | * for it via sb->s_xattr. | |
2166 | */ | |
2167 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2168 | return generic_setxattr(dentry, name, value, size, flags); | |
2169 | ||
2170 | err = shmem_xattr_validate(name); | |
2171 | if (err) | |
2172 | return err; | |
2173 | ||
2174 | if (size == 0) | |
2175 | value = ""; /* empty EA, do not remove */ | |
2176 | ||
6d9d88d0 | 2177 | return shmem_xattr_set(dentry->d_inode, name, value, size, flags); |
b09e0fa4 EP |
2178 | |
2179 | } | |
2180 | ||
2181 | static int shmem_removexattr(struct dentry *dentry, const char *name) | |
2182 | { | |
2183 | int err; | |
2184 | ||
2185 | /* | |
2186 | * If this is a request for a synthetic attribute in the system.* | |
2187 | * namespace use the generic infrastructure to resolve a handler | |
2188 | * for it via sb->s_xattr. | |
2189 | */ | |
2190 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2191 | return generic_removexattr(dentry, name); | |
2192 | ||
2193 | err = shmem_xattr_validate(name); | |
2194 | if (err) | |
2195 | return err; | |
2196 | ||
6d9d88d0 | 2197 | return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE); |
b09e0fa4 EP |
2198 | } |
2199 | ||
2200 | static bool xattr_is_trusted(const char *name) | |
2201 | { | |
2202 | return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN); | |
2203 | } | |
2204 | ||
2205 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
2206 | { | |
2207 | bool trusted = capable(CAP_SYS_ADMIN); | |
2208 | struct shmem_xattr *xattr; | |
2209 | struct shmem_inode_info *info; | |
2210 | size_t used = 0; | |
2211 | ||
2212 | info = SHMEM_I(dentry->d_inode); | |
2213 | ||
2214 | spin_lock(&info->lock); | |
2215 | list_for_each_entry(xattr, &info->xattr_list, list) { | |
2216 | size_t len; | |
2217 | ||
2218 | /* skip "trusted." attributes for unprivileged callers */ | |
2219 | if (!trusted && xattr_is_trusted(xattr->name)) | |
2220 | continue; | |
2221 | ||
2222 | len = strlen(xattr->name) + 1; | |
2223 | used += len; | |
2224 | if (buffer) { | |
2225 | if (size < used) { | |
2226 | used = -ERANGE; | |
2227 | break; | |
2228 | } | |
2229 | memcpy(buffer, xattr->name, len); | |
2230 | buffer += len; | |
2231 | } | |
2232 | } | |
2233 | spin_unlock(&info->lock); | |
2234 | ||
2235 | return used; | |
2236 | } | |
2237 | #endif /* CONFIG_TMPFS_XATTR */ | |
2238 | ||
69f07ec9 | 2239 | static const struct inode_operations shmem_short_symlink_operations = { |
b09e0fa4 | 2240 | .readlink = generic_readlink, |
69f07ec9 | 2241 | .follow_link = shmem_follow_short_symlink, |
b09e0fa4 EP |
2242 | #ifdef CONFIG_TMPFS_XATTR |
2243 | .setxattr = shmem_setxattr, | |
2244 | .getxattr = shmem_getxattr, | |
2245 | .listxattr = shmem_listxattr, | |
2246 | .removexattr = shmem_removexattr, | |
2247 | #endif | |
2248 | }; | |
2249 | ||
2250 | static const struct inode_operations shmem_symlink_inode_operations = { | |
2251 | .readlink = generic_readlink, | |
2252 | .follow_link = shmem_follow_link, | |
2253 | .put_link = shmem_put_link, | |
2254 | #ifdef CONFIG_TMPFS_XATTR | |
2255 | .setxattr = shmem_setxattr, | |
2256 | .getxattr = shmem_getxattr, | |
2257 | .listxattr = shmem_listxattr, | |
2258 | .removexattr = shmem_removexattr, | |
39f0247d | 2259 | #endif |
b09e0fa4 | 2260 | }; |
39f0247d | 2261 | |
91828a40 DG |
2262 | static struct dentry *shmem_get_parent(struct dentry *child) |
2263 | { | |
2264 | return ERR_PTR(-ESTALE); | |
2265 | } | |
2266 | ||
2267 | static int shmem_match(struct inode *ino, void *vfh) | |
2268 | { | |
2269 | __u32 *fh = vfh; | |
2270 | __u64 inum = fh[2]; | |
2271 | inum = (inum << 32) | fh[1]; | |
2272 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
2273 | } | |
2274 | ||
480b116c CH |
2275 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
2276 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 2277 | { |
91828a40 | 2278 | struct inode *inode; |
480b116c CH |
2279 | struct dentry *dentry = NULL; |
2280 | u64 inum = fid->raw[2]; | |
2281 | inum = (inum << 32) | fid->raw[1]; | |
2282 | ||
2283 | if (fh_len < 3) | |
2284 | return NULL; | |
91828a40 | 2285 | |
480b116c CH |
2286 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
2287 | shmem_match, fid->raw); | |
91828a40 | 2288 | if (inode) { |
480b116c | 2289 | dentry = d_find_alias(inode); |
91828a40 DG |
2290 | iput(inode); |
2291 | } | |
2292 | ||
480b116c | 2293 | return dentry; |
91828a40 DG |
2294 | } |
2295 | ||
2296 | static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, | |
2297 | int connectable) | |
2298 | { | |
2299 | struct inode *inode = dentry->d_inode; | |
2300 | ||
5fe0c237 AK |
2301 | if (*len < 3) { |
2302 | *len = 3; | |
91828a40 | 2303 | return 255; |
5fe0c237 | 2304 | } |
91828a40 | 2305 | |
1d3382cb | 2306 | if (inode_unhashed(inode)) { |
91828a40 DG |
2307 | /* Unfortunately insert_inode_hash is not idempotent, |
2308 | * so as we hash inodes here rather than at creation | |
2309 | * time, we need a lock to ensure we only try | |
2310 | * to do it once | |
2311 | */ | |
2312 | static DEFINE_SPINLOCK(lock); | |
2313 | spin_lock(&lock); | |
1d3382cb | 2314 | if (inode_unhashed(inode)) |
91828a40 DG |
2315 | __insert_inode_hash(inode, |
2316 | inode->i_ino + inode->i_generation); | |
2317 | spin_unlock(&lock); | |
2318 | } | |
2319 | ||
2320 | fh[0] = inode->i_generation; | |
2321 | fh[1] = inode->i_ino; | |
2322 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
2323 | ||
2324 | *len = 3; | |
2325 | return 1; | |
2326 | } | |
2327 | ||
39655164 | 2328 | static const struct export_operations shmem_export_ops = { |
91828a40 | 2329 | .get_parent = shmem_get_parent, |
91828a40 | 2330 | .encode_fh = shmem_encode_fh, |
480b116c | 2331 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
2332 | }; |
2333 | ||
680d794b | 2334 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
2335 | bool remount) | |
1da177e4 LT |
2336 | { |
2337 | char *this_char, *value, *rest; | |
8751e039 EB |
2338 | uid_t uid; |
2339 | gid_t gid; | |
1da177e4 | 2340 | |
b00dc3ad HD |
2341 | while (options != NULL) { |
2342 | this_char = options; | |
2343 | for (;;) { | |
2344 | /* | |
2345 | * NUL-terminate this option: unfortunately, | |
2346 | * mount options form a comma-separated list, | |
2347 | * but mpol's nodelist may also contain commas. | |
2348 | */ | |
2349 | options = strchr(options, ','); | |
2350 | if (options == NULL) | |
2351 | break; | |
2352 | options++; | |
2353 | if (!isdigit(*options)) { | |
2354 | options[-1] = '\0'; | |
2355 | break; | |
2356 | } | |
2357 | } | |
1da177e4 LT |
2358 | if (!*this_char) |
2359 | continue; | |
2360 | if ((value = strchr(this_char,'=')) != NULL) { | |
2361 | *value++ = 0; | |
2362 | } else { | |
2363 | printk(KERN_ERR | |
2364 | "tmpfs: No value for mount option '%s'\n", | |
2365 | this_char); | |
2366 | return 1; | |
2367 | } | |
2368 | ||
2369 | if (!strcmp(this_char,"size")) { | |
2370 | unsigned long long size; | |
2371 | size = memparse(value,&rest); | |
2372 | if (*rest == '%') { | |
2373 | size <<= PAGE_SHIFT; | |
2374 | size *= totalram_pages; | |
2375 | do_div(size, 100); | |
2376 | rest++; | |
2377 | } | |
2378 | if (*rest) | |
2379 | goto bad_val; | |
680d794b | 2380 | sbinfo->max_blocks = |
2381 | DIV_ROUND_UP(size, PAGE_CACHE_SIZE); | |
1da177e4 | 2382 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 2383 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
2384 | if (*rest) |
2385 | goto bad_val; | |
2386 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 2387 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
2388 | if (*rest) |
2389 | goto bad_val; | |
2390 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 2391 | if (remount) |
1da177e4 | 2392 | continue; |
680d794b | 2393 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
2394 | if (*rest) |
2395 | goto bad_val; | |
2396 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 2397 | if (remount) |
1da177e4 | 2398 | continue; |
8751e039 | 2399 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2400 | if (*rest) |
2401 | goto bad_val; | |
8751e039 EB |
2402 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
2403 | if (!uid_valid(sbinfo->uid)) | |
2404 | goto bad_val; | |
1da177e4 | 2405 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 2406 | if (remount) |
1da177e4 | 2407 | continue; |
8751e039 | 2408 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2409 | if (*rest) |
2410 | goto bad_val; | |
8751e039 EB |
2411 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
2412 | if (!gid_valid(sbinfo->gid)) | |
2413 | goto bad_val; | |
7339ff83 | 2414 | } else if (!strcmp(this_char,"mpol")) { |
71fe804b | 2415 | if (mpol_parse_str(value, &sbinfo->mpol, 1)) |
7339ff83 | 2416 | goto bad_val; |
1da177e4 LT |
2417 | } else { |
2418 | printk(KERN_ERR "tmpfs: Bad mount option %s\n", | |
2419 | this_char); | |
2420 | return 1; | |
2421 | } | |
2422 | } | |
2423 | return 0; | |
2424 | ||
2425 | bad_val: | |
2426 | printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", | |
2427 | value, this_char); | |
2428 | return 1; | |
2429 | ||
2430 | } | |
2431 | ||
2432 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
2433 | { | |
2434 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 2435 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
2436 | unsigned long inodes; |
2437 | int error = -EINVAL; | |
2438 | ||
680d794b | 2439 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 2440 | return error; |
1da177e4 | 2441 | |
0edd73b3 | 2442 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 2443 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 2444 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 2445 | goto out; |
680d794b | 2446 | if (config.max_inodes < inodes) |
0edd73b3 HD |
2447 | goto out; |
2448 | /* | |
54af6042 | 2449 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
2450 | * but we must separately disallow unlimited->limited, because |
2451 | * in that case we have no record of how much is already in use. | |
2452 | */ | |
680d794b | 2453 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 2454 | goto out; |
680d794b | 2455 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
2456 | goto out; |
2457 | ||
2458 | error = 0; | |
680d794b | 2459 | sbinfo->max_blocks = config.max_blocks; |
680d794b | 2460 | sbinfo->max_inodes = config.max_inodes; |
2461 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b LS |
2462 | |
2463 | mpol_put(sbinfo->mpol); | |
2464 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
0edd73b3 HD |
2465 | out: |
2466 | spin_unlock(&sbinfo->stat_lock); | |
2467 | return error; | |
1da177e4 | 2468 | } |
680d794b | 2469 | |
34c80b1d | 2470 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 2471 | { |
34c80b1d | 2472 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b | 2473 | |
2474 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
2475 | seq_printf(seq, ",size=%luk", | |
2476 | sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); | |
2477 | if (sbinfo->max_inodes != shmem_default_max_inodes()) | |
2478 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
2479 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 2480 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
2481 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
2482 | seq_printf(seq, ",uid=%u", | |
2483 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
2484 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
2485 | seq_printf(seq, ",gid=%u", | |
2486 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
71fe804b | 2487 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b | 2488 | return 0; |
2489 | } | |
2490 | #endif /* CONFIG_TMPFS */ | |
1da177e4 LT |
2491 | |
2492 | static void shmem_put_super(struct super_block *sb) | |
2493 | { | |
602586a8 HD |
2494 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
2495 | ||
2496 | percpu_counter_destroy(&sbinfo->used_blocks); | |
2497 | kfree(sbinfo); | |
1da177e4 LT |
2498 | sb->s_fs_info = NULL; |
2499 | } | |
2500 | ||
2b2af54a | 2501 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
2502 | { |
2503 | struct inode *inode; | |
0edd73b3 | 2504 | struct shmem_sb_info *sbinfo; |
680d794b | 2505 | int err = -ENOMEM; |
2506 | ||
2507 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 2508 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b | 2509 | L1_CACHE_BYTES), GFP_KERNEL); |
2510 | if (!sbinfo) | |
2511 | return -ENOMEM; | |
2512 | ||
680d794b | 2513 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
2514 | sbinfo->uid = current_fsuid(); |
2515 | sbinfo->gid = current_fsgid(); | |
680d794b | 2516 | sb->s_fs_info = sbinfo; |
1da177e4 | 2517 | |
0edd73b3 | 2518 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
2519 | /* |
2520 | * Per default we only allow half of the physical ram per | |
2521 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
2522 | * but the internal instance is left unlimited. | |
2523 | */ | |
2524 | if (!(sb->s_flags & MS_NOUSER)) { | |
680d794b | 2525 | sbinfo->max_blocks = shmem_default_max_blocks(); |
2526 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
2527 | if (shmem_parse_options(data, sbinfo, false)) { | |
2528 | err = -EINVAL; | |
2529 | goto failed; | |
2530 | } | |
1da177e4 | 2531 | } |
91828a40 | 2532 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 2533 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
2534 | #else |
2535 | sb->s_flags |= MS_NOUSER; | |
2536 | #endif | |
2537 | ||
0edd73b3 | 2538 | spin_lock_init(&sbinfo->stat_lock); |
602586a8 HD |
2539 | if (percpu_counter_init(&sbinfo->used_blocks, 0)) |
2540 | goto failed; | |
680d794b | 2541 | sbinfo->free_inodes = sbinfo->max_inodes; |
0edd73b3 | 2542 | |
285b2c4f | 2543 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
1da177e4 LT |
2544 | sb->s_blocksize = PAGE_CACHE_SIZE; |
2545 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
2546 | sb->s_magic = TMPFS_MAGIC; | |
2547 | sb->s_op = &shmem_ops; | |
cfd95a9c | 2548 | sb->s_time_gran = 1; |
b09e0fa4 | 2549 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 2550 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
2551 | #endif |
2552 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
2553 | sb->s_flags |= MS_POSIXACL; |
2554 | #endif | |
0edd73b3 | 2555 | |
454abafe | 2556 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
2557 | if (!inode) |
2558 | goto failed; | |
680d794b | 2559 | inode->i_uid = sbinfo->uid; |
2560 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
2561 | sb->s_root = d_make_root(inode); |
2562 | if (!sb->s_root) | |
48fde701 | 2563 | goto failed; |
1da177e4 LT |
2564 | return 0; |
2565 | ||
1da177e4 LT |
2566 | failed: |
2567 | shmem_put_super(sb); | |
2568 | return err; | |
2569 | } | |
2570 | ||
fcc234f8 | 2571 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
2572 | |
2573 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
2574 | { | |
41ffe5d5 HD |
2575 | struct shmem_inode_info *info; |
2576 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
2577 | if (!info) | |
1da177e4 | 2578 | return NULL; |
41ffe5d5 | 2579 | return &info->vfs_inode; |
1da177e4 LT |
2580 | } |
2581 | ||
41ffe5d5 | 2582 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
2583 | { |
2584 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
2585 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
2586 | } | |
2587 | ||
1da177e4 LT |
2588 | static void shmem_destroy_inode(struct inode *inode) |
2589 | { | |
09208d15 | 2590 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 2591 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 2592 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
2593 | } |
2594 | ||
41ffe5d5 | 2595 | static void shmem_init_inode(void *foo) |
1da177e4 | 2596 | { |
41ffe5d5 HD |
2597 | struct shmem_inode_info *info = foo; |
2598 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
2599 | } |
2600 | ||
41ffe5d5 | 2601 | static int shmem_init_inodecache(void) |
1da177e4 LT |
2602 | { |
2603 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
2604 | sizeof(struct shmem_inode_info), | |
41ffe5d5 | 2605 | 0, SLAB_PANIC, shmem_init_inode); |
1da177e4 LT |
2606 | return 0; |
2607 | } | |
2608 | ||
41ffe5d5 | 2609 | static void shmem_destroy_inodecache(void) |
1da177e4 | 2610 | { |
1a1d92c1 | 2611 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
2612 | } |
2613 | ||
f5e54d6e | 2614 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 2615 | .writepage = shmem_writepage, |
76719325 | 2616 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 2617 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
2618 | .write_begin = shmem_write_begin, |
2619 | .write_end = shmem_write_end, | |
1da177e4 | 2620 | #endif |
304dbdb7 | 2621 | .migratepage = migrate_page, |
aa261f54 | 2622 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
2623 | }; |
2624 | ||
15ad7cdc | 2625 | static const struct file_operations shmem_file_operations = { |
1da177e4 LT |
2626 | .mmap = shmem_mmap, |
2627 | #ifdef CONFIG_TMPFS | |
2628 | .llseek = generic_file_llseek, | |
bcd78e49 | 2629 | .read = do_sync_read, |
5402b976 | 2630 | .write = do_sync_write, |
bcd78e49 | 2631 | .aio_read = shmem_file_aio_read, |
5402b976 | 2632 | .aio_write = generic_file_aio_write, |
1b061d92 | 2633 | .fsync = noop_fsync, |
708e3508 | 2634 | .splice_read = shmem_file_splice_read, |
ae976416 | 2635 | .splice_write = generic_file_splice_write, |
83e4fa9c | 2636 | .fallocate = shmem_fallocate, |
1da177e4 LT |
2637 | #endif |
2638 | }; | |
2639 | ||
92e1d5be | 2640 | static const struct inode_operations shmem_inode_operations = { |
94c1e62d | 2641 | .setattr = shmem_setattr, |
b09e0fa4 EP |
2642 | #ifdef CONFIG_TMPFS_XATTR |
2643 | .setxattr = shmem_setxattr, | |
2644 | .getxattr = shmem_getxattr, | |
2645 | .listxattr = shmem_listxattr, | |
2646 | .removexattr = shmem_removexattr, | |
2647 | #endif | |
1da177e4 LT |
2648 | }; |
2649 | ||
92e1d5be | 2650 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
2651 | #ifdef CONFIG_TMPFS |
2652 | .create = shmem_create, | |
2653 | .lookup = simple_lookup, | |
2654 | .link = shmem_link, | |
2655 | .unlink = shmem_unlink, | |
2656 | .symlink = shmem_symlink, | |
2657 | .mkdir = shmem_mkdir, | |
2658 | .rmdir = shmem_rmdir, | |
2659 | .mknod = shmem_mknod, | |
2660 | .rename = shmem_rename, | |
1da177e4 | 2661 | #endif |
b09e0fa4 EP |
2662 | #ifdef CONFIG_TMPFS_XATTR |
2663 | .setxattr = shmem_setxattr, | |
2664 | .getxattr = shmem_getxattr, | |
2665 | .listxattr = shmem_listxattr, | |
2666 | .removexattr = shmem_removexattr, | |
2667 | #endif | |
39f0247d | 2668 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 2669 | .setattr = shmem_setattr, |
39f0247d AG |
2670 | #endif |
2671 | }; | |
2672 | ||
92e1d5be | 2673 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 EP |
2674 | #ifdef CONFIG_TMPFS_XATTR |
2675 | .setxattr = shmem_setxattr, | |
2676 | .getxattr = shmem_getxattr, | |
2677 | .listxattr = shmem_listxattr, | |
2678 | .removexattr = shmem_removexattr, | |
2679 | #endif | |
39f0247d | 2680 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 2681 | .setattr = shmem_setattr, |
39f0247d | 2682 | #endif |
1da177e4 LT |
2683 | }; |
2684 | ||
759b9775 | 2685 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
2686 | .alloc_inode = shmem_alloc_inode, |
2687 | .destroy_inode = shmem_destroy_inode, | |
2688 | #ifdef CONFIG_TMPFS | |
2689 | .statfs = shmem_statfs, | |
2690 | .remount_fs = shmem_remount_fs, | |
680d794b | 2691 | .show_options = shmem_show_options, |
1da177e4 | 2692 | #endif |
1f895f75 | 2693 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
2694 | .drop_inode = generic_delete_inode, |
2695 | .put_super = shmem_put_super, | |
2696 | }; | |
2697 | ||
f0f37e2f | 2698 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 2699 | .fault = shmem_fault, |
1da177e4 LT |
2700 | #ifdef CONFIG_NUMA |
2701 | .set_policy = shmem_set_policy, | |
2702 | .get_policy = shmem_get_policy, | |
2703 | #endif | |
2704 | }; | |
2705 | ||
3c26ff6e AV |
2706 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
2707 | int flags, const char *dev_name, void *data) | |
1da177e4 | 2708 | { |
3c26ff6e | 2709 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
2710 | } |
2711 | ||
41ffe5d5 | 2712 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
2713 | .owner = THIS_MODULE, |
2714 | .name = "tmpfs", | |
3c26ff6e | 2715 | .mount = shmem_mount, |
1da177e4 LT |
2716 | .kill_sb = kill_litter_super, |
2717 | }; | |
1da177e4 | 2718 | |
41ffe5d5 | 2719 | int __init shmem_init(void) |
1da177e4 LT |
2720 | { |
2721 | int error; | |
2722 | ||
e0bf68dd PZ |
2723 | error = bdi_init(&shmem_backing_dev_info); |
2724 | if (error) | |
2725 | goto out4; | |
2726 | ||
41ffe5d5 | 2727 | error = shmem_init_inodecache(); |
1da177e4 LT |
2728 | if (error) |
2729 | goto out3; | |
2730 | ||
41ffe5d5 | 2731 | error = register_filesystem(&shmem_fs_type); |
1da177e4 LT |
2732 | if (error) { |
2733 | printk(KERN_ERR "Could not register tmpfs\n"); | |
2734 | goto out2; | |
2735 | } | |
95dc112a | 2736 | |
41ffe5d5 HD |
2737 | shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER, |
2738 | shmem_fs_type.name, NULL); | |
1da177e4 LT |
2739 | if (IS_ERR(shm_mnt)) { |
2740 | error = PTR_ERR(shm_mnt); | |
2741 | printk(KERN_ERR "Could not kern_mount tmpfs\n"); | |
2742 | goto out1; | |
2743 | } | |
2744 | return 0; | |
2745 | ||
2746 | out1: | |
41ffe5d5 | 2747 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 2748 | out2: |
41ffe5d5 | 2749 | shmem_destroy_inodecache(); |
1da177e4 | 2750 | out3: |
e0bf68dd PZ |
2751 | bdi_destroy(&shmem_backing_dev_info); |
2752 | out4: | |
1da177e4 LT |
2753 | shm_mnt = ERR_PTR(error); |
2754 | return error; | |
2755 | } | |
853ac43a MM |
2756 | |
2757 | #else /* !CONFIG_SHMEM */ | |
2758 | ||
2759 | /* | |
2760 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
2761 | * | |
2762 | * This is intended for small system where the benefits of the full | |
2763 | * shmem code (swap-backed and resource-limited) are outweighed by | |
2764 | * their complexity. On systems without swap this code should be | |
2765 | * effectively equivalent, but much lighter weight. | |
2766 | */ | |
2767 | ||
2768 | #include <linux/ramfs.h> | |
2769 | ||
41ffe5d5 | 2770 | static struct file_system_type shmem_fs_type = { |
853ac43a | 2771 | .name = "tmpfs", |
3c26ff6e | 2772 | .mount = ramfs_mount, |
853ac43a MM |
2773 | .kill_sb = kill_litter_super, |
2774 | }; | |
2775 | ||
41ffe5d5 | 2776 | int __init shmem_init(void) |
853ac43a | 2777 | { |
41ffe5d5 | 2778 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 2779 | |
41ffe5d5 | 2780 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
2781 | BUG_ON(IS_ERR(shm_mnt)); |
2782 | ||
2783 | return 0; | |
2784 | } | |
2785 | ||
41ffe5d5 | 2786 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
2787 | { |
2788 | return 0; | |
2789 | } | |
2790 | ||
3f96b79a HD |
2791 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
2792 | { | |
2793 | return 0; | |
2794 | } | |
2795 | ||
24513264 HD |
2796 | void shmem_unlock_mapping(struct address_space *mapping) |
2797 | { | |
2798 | } | |
2799 | ||
41ffe5d5 | 2800 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 2801 | { |
41ffe5d5 | 2802 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
2803 | } |
2804 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
2805 | ||
0b0a0806 HD |
2806 | #define shmem_vm_ops generic_file_vm_ops |
2807 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 2808 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
2809 | #define shmem_acct_size(flags, size) 0 |
2810 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
2811 | |
2812 | #endif /* CONFIG_SHMEM */ | |
2813 | ||
2814 | /* common code */ | |
1da177e4 | 2815 | |
46711810 | 2816 | /** |
1da177e4 | 2817 | * shmem_file_setup - get an unlinked file living in tmpfs |
1da177e4 LT |
2818 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
2819 | * @size: size to be set for the file | |
0b0a0806 | 2820 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size |
1da177e4 | 2821 | */ |
168f5ac6 | 2822 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) |
1da177e4 LT |
2823 | { |
2824 | int error; | |
2825 | struct file *file; | |
2826 | struct inode *inode; | |
2c48b9c4 AV |
2827 | struct path path; |
2828 | struct dentry *root; | |
1da177e4 LT |
2829 | struct qstr this; |
2830 | ||
2831 | if (IS_ERR(shm_mnt)) | |
2832 | return (void *)shm_mnt; | |
2833 | ||
285b2c4f | 2834 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
2835 | return ERR_PTR(-EINVAL); |
2836 | ||
2837 | if (shmem_acct_size(flags, size)) | |
2838 | return ERR_PTR(-ENOMEM); | |
2839 | ||
2840 | error = -ENOMEM; | |
2841 | this.name = name; | |
2842 | this.len = strlen(name); | |
2843 | this.hash = 0; /* will go */ | |
2844 | root = shm_mnt->mnt_root; | |
2c48b9c4 AV |
2845 | path.dentry = d_alloc(root, &this); |
2846 | if (!path.dentry) | |
1da177e4 | 2847 | goto put_memory; |
2c48b9c4 | 2848 | path.mnt = mntget(shm_mnt); |
1da177e4 | 2849 | |
1da177e4 | 2850 | error = -ENOSPC; |
454abafe | 2851 | inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 2852 | if (!inode) |
4b42af81 | 2853 | goto put_dentry; |
1da177e4 | 2854 | |
2c48b9c4 | 2855 | d_instantiate(path.dentry, inode); |
1da177e4 | 2856 | inode->i_size = size; |
6d6b77f1 | 2857 | clear_nlink(inode); /* It is unlinked */ |
853ac43a MM |
2858 | #ifndef CONFIG_MMU |
2859 | error = ramfs_nommu_expand_for_mapping(inode, size); | |
2860 | if (error) | |
4b42af81 | 2861 | goto put_dentry; |
853ac43a | 2862 | #endif |
4b42af81 AV |
2863 | |
2864 | error = -ENFILE; | |
2c48b9c4 | 2865 | file = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 AV |
2866 | &shmem_file_operations); |
2867 | if (!file) | |
2868 | goto put_dentry; | |
2869 | ||
1da177e4 LT |
2870 | return file; |
2871 | ||
1da177e4 | 2872 | put_dentry: |
2c48b9c4 | 2873 | path_put(&path); |
1da177e4 LT |
2874 | put_memory: |
2875 | shmem_unacct_size(flags, size); | |
2876 | return ERR_PTR(error); | |
2877 | } | |
395e0ddc | 2878 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 2879 | |
46711810 | 2880 | /** |
1da177e4 | 2881 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
2882 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
2883 | */ | |
2884 | int shmem_zero_setup(struct vm_area_struct *vma) | |
2885 | { | |
2886 | struct file *file; | |
2887 | loff_t size = vma->vm_end - vma->vm_start; | |
2888 | ||
2889 | file = shmem_file_setup("dev/zero", size, vma->vm_flags); | |
2890 | if (IS_ERR(file)) | |
2891 | return PTR_ERR(file); | |
2892 | ||
2893 | if (vma->vm_file) | |
2894 | fput(vma->vm_file); | |
2895 | vma->vm_file = file; | |
2896 | vma->vm_ops = &shmem_vm_ops; | |
bee4c36a | 2897 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
2898 | return 0; |
2899 | } | |
d9d90e5e HD |
2900 | |
2901 | /** | |
2902 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
2903 | * @mapping: the page's address_space | |
2904 | * @index: the page index | |
2905 | * @gfp: the page allocator flags to use if allocating | |
2906 | * | |
2907 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
2908 | * with any new page allocations done using the specified allocation flags. | |
2909 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
2910 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
2911 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
2912 | * | |
68da9f05 HD |
2913 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
2914 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
2915 | */ |
2916 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
2917 | pgoff_t index, gfp_t gfp) | |
2918 | { | |
68da9f05 HD |
2919 | #ifdef CONFIG_SHMEM |
2920 | struct inode *inode = mapping->host; | |
9276aad6 | 2921 | struct page *page; |
68da9f05 HD |
2922 | int error; |
2923 | ||
2924 | BUG_ON(mapping->a_ops != &shmem_aops); | |
2925 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL); | |
2926 | if (error) | |
2927 | page = ERR_PTR(error); | |
2928 | else | |
2929 | unlock_page(page); | |
2930 | return page; | |
2931 | #else | |
2932 | /* | |
2933 | * The tiny !SHMEM case uses ramfs without swap | |
2934 | */ | |
d9d90e5e | 2935 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 2936 | #endif |
d9d90e5e HD |
2937 | } |
2938 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |