]>
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> | |
250297ed | 28 | #include <linux/ramfs.h> |
caefba17 | 29 | #include <linux/pagemap.h> |
853ac43a MM |
30 | #include <linux/file.h> |
31 | #include <linux/mm.h> | |
b95f1b31 | 32 | #include <linux/export.h> |
853ac43a | 33 | #include <linux/swap.h> |
a27bb332 | 34 | #include <linux/aio.h> |
853ac43a MM |
35 | |
36 | static struct vfsmount *shm_mnt; | |
37 | ||
38 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
39 | /* |
40 | * This virtual memory filesystem is heavily based on the ramfs. It | |
41 | * extends ramfs by the ability to use swap and honor resource limits | |
42 | * which makes it a completely usable filesystem. | |
43 | */ | |
44 | ||
39f0247d | 45 | #include <linux/xattr.h> |
a5694255 | 46 | #include <linux/exportfs.h> |
1c7c474c | 47 | #include <linux/posix_acl.h> |
feda821e | 48 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 49 | #include <linux/mman.h> |
1da177e4 LT |
50 | #include <linux/string.h> |
51 | #include <linux/slab.h> | |
52 | #include <linux/backing-dev.h> | |
53 | #include <linux/shmem_fs.h> | |
1da177e4 | 54 | #include <linux/writeback.h> |
1da177e4 | 55 | #include <linux/blkdev.h> |
bda97eab | 56 | #include <linux/pagevec.h> |
41ffe5d5 | 57 | #include <linux/percpu_counter.h> |
83e4fa9c | 58 | #include <linux/falloc.h> |
708e3508 | 59 | #include <linux/splice.h> |
1da177e4 LT |
60 | #include <linux/security.h> |
61 | #include <linux/swapops.h> | |
62 | #include <linux/mempolicy.h> | |
63 | #include <linux/namei.h> | |
b00dc3ad | 64 | #include <linux/ctype.h> |
304dbdb7 | 65 | #include <linux/migrate.h> |
c1f60a5a | 66 | #include <linux/highmem.h> |
680d794b | 67 | #include <linux/seq_file.h> |
92562927 | 68 | #include <linux/magic.h> |
40e041a2 | 69 | #include <linux/fcntl.h> |
304dbdb7 | 70 | |
1da177e4 | 71 | #include <asm/uaccess.h> |
1da177e4 LT |
72 | #include <asm/pgtable.h> |
73 | ||
caefba17 | 74 | #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) |
1da177e4 LT |
75 | #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) |
76 | ||
1da177e4 LT |
77 | /* Pretend that each entry is of this size in directory's i_size */ |
78 | #define BOGO_DIRENT_SIZE 20 | |
79 | ||
69f07ec9 HD |
80 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
81 | #define SHORT_SYMLINK_LEN 128 | |
82 | ||
1aac1400 | 83 | /* |
f00cdc6d HD |
84 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
85 | * inode->i_private (with i_mutex making sure that it has only one user at | |
86 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
87 | */ |
88 | struct shmem_falloc { | |
8e205f77 | 89 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
90 | pgoff_t start; /* start of range currently being fallocated */ |
91 | pgoff_t next; /* the next page offset to be fallocated */ | |
92 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
93 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
94 | }; | |
95 | ||
285b2c4f | 96 | /* Flag allocation requirements to shmem_getpage */ |
1da177e4 | 97 | enum sgp_type { |
1da177e4 LT |
98 | SGP_READ, /* don't exceed i_size, don't allocate page */ |
99 | SGP_CACHE, /* don't exceed i_size, may allocate page */ | |
a0ee5ec5 | 100 | SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ |
1635f6a7 HD |
101 | SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */ |
102 | SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */ | |
1da177e4 LT |
103 | }; |
104 | ||
b76db735 | 105 | #ifdef CONFIG_TMPFS |
680d794b AM |
106 | static unsigned long shmem_default_max_blocks(void) |
107 | { | |
108 | return totalram_pages / 2; | |
109 | } | |
110 | ||
111 | static unsigned long shmem_default_max_inodes(void) | |
112 | { | |
113 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
114 | } | |
b76db735 | 115 | #endif |
680d794b | 116 | |
bde05d1c HD |
117 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
118 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
119 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 HD |
120 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
121 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type); | |
122 | ||
123 | static inline int shmem_getpage(struct inode *inode, pgoff_t index, | |
124 | struct page **pagep, enum sgp_type sgp, int *fault_type) | |
125 | { | |
126 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
127 | mapping_gfp_mask(inode->i_mapping), fault_type); | |
128 | } | |
1da177e4 | 129 | |
1da177e4 LT |
130 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
131 | { | |
132 | return sb->s_fs_info; | |
133 | } | |
134 | ||
135 | /* | |
136 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
137 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
138 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
139 | * consistent with the pre-accounting of private mappings ... | |
140 | */ | |
141 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
142 | { | |
0b0a0806 | 143 | return (flags & VM_NORESERVE) ? |
191c5424 | 144 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
145 | } |
146 | ||
147 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
148 | { | |
0b0a0806 | 149 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
150 | vm_unacct_memory(VM_ACCT(size)); |
151 | } | |
152 | ||
77142517 KK |
153 | static inline int shmem_reacct_size(unsigned long flags, |
154 | loff_t oldsize, loff_t newsize) | |
155 | { | |
156 | if (!(flags & VM_NORESERVE)) { | |
157 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
158 | return security_vm_enough_memory_mm(current->mm, | |
159 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
160 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
161 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
162 | } | |
163 | return 0; | |
164 | } | |
165 | ||
1da177e4 LT |
166 | /* |
167 | * ... whereas tmpfs objects are accounted incrementally as | |
168 | * pages are allocated, in order to allow huge sparse files. | |
169 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, | |
170 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
171 | */ | |
172 | static inline int shmem_acct_block(unsigned long flags) | |
173 | { | |
0b0a0806 | 174 | return (flags & VM_NORESERVE) ? |
191c5424 | 175 | security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0; |
1da177e4 LT |
176 | } |
177 | ||
178 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
179 | { | |
0b0a0806 | 180 | if (flags & VM_NORESERVE) |
1da177e4 LT |
181 | vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); |
182 | } | |
183 | ||
759b9775 | 184 | static const struct super_operations shmem_ops; |
f5e54d6e | 185 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 186 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
187 | static const struct inode_operations shmem_inode_operations; |
188 | static const struct inode_operations shmem_dir_inode_operations; | |
189 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 190 | static const struct vm_operations_struct shmem_vm_ops; |
1da177e4 | 191 | |
6c231b7b | 192 | static struct backing_dev_info shmem_backing_dev_info __read_mostly = { |
1da177e4 | 193 | .ra_pages = 0, /* No readahead */ |
4f98a2fe | 194 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED, |
1da177e4 LT |
195 | }; |
196 | ||
197 | static LIST_HEAD(shmem_swaplist); | |
cb5f7b9a | 198 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 199 | |
5b04c689 PE |
200 | static int shmem_reserve_inode(struct super_block *sb) |
201 | { | |
202 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
203 | if (sbinfo->max_inodes) { | |
204 | spin_lock(&sbinfo->stat_lock); | |
205 | if (!sbinfo->free_inodes) { | |
206 | spin_unlock(&sbinfo->stat_lock); | |
207 | return -ENOSPC; | |
208 | } | |
209 | sbinfo->free_inodes--; | |
210 | spin_unlock(&sbinfo->stat_lock); | |
211 | } | |
212 | return 0; | |
213 | } | |
214 | ||
215 | static void shmem_free_inode(struct super_block *sb) | |
216 | { | |
217 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
218 | if (sbinfo->max_inodes) { | |
219 | spin_lock(&sbinfo->stat_lock); | |
220 | sbinfo->free_inodes++; | |
221 | spin_unlock(&sbinfo->stat_lock); | |
222 | } | |
223 | } | |
224 | ||
46711810 | 225 | /** |
41ffe5d5 | 226 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
227 | * @inode: inode to recalc |
228 | * | |
229 | * We have to calculate the free blocks since the mm can drop | |
230 | * undirtied hole pages behind our back. | |
231 | * | |
232 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
233 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
234 | * | |
235 | * It has to be called with the spinlock held. | |
236 | */ | |
237 | static void shmem_recalc_inode(struct inode *inode) | |
238 | { | |
239 | struct shmem_inode_info *info = SHMEM_I(inode); | |
240 | long freed; | |
241 | ||
242 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
243 | if (freed > 0) { | |
54af6042 HD |
244 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
245 | if (sbinfo->max_blocks) | |
246 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 247 | info->alloced -= freed; |
54af6042 | 248 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 249 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
250 | } |
251 | } | |
252 | ||
7a5d0fbb HD |
253 | /* |
254 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
255 | */ | |
256 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
257 | pgoff_t index, void *expected, void *replacement) | |
258 | { | |
259 | void **pslot; | |
6dbaf22c | 260 | void *item; |
7a5d0fbb HD |
261 | |
262 | VM_BUG_ON(!expected); | |
6dbaf22c | 263 | VM_BUG_ON(!replacement); |
7a5d0fbb | 264 | pslot = radix_tree_lookup_slot(&mapping->page_tree, index); |
6dbaf22c JW |
265 | if (!pslot) |
266 | return -ENOENT; | |
267 | item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock); | |
7a5d0fbb HD |
268 | if (item != expected) |
269 | return -ENOENT; | |
6dbaf22c | 270 | radix_tree_replace_slot(pslot, replacement); |
7a5d0fbb HD |
271 | return 0; |
272 | } | |
273 | ||
d1899228 HD |
274 | /* |
275 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
276 | * that an entry was not already brought back from swap by a racing thread. | |
277 | * | |
278 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
279 | * might be reused, and again be SwapCache, using the same swap as before. | |
280 | */ | |
281 | static bool shmem_confirm_swap(struct address_space *mapping, | |
282 | pgoff_t index, swp_entry_t swap) | |
283 | { | |
284 | void *item; | |
285 | ||
286 | rcu_read_lock(); | |
287 | item = radix_tree_lookup(&mapping->page_tree, index); | |
288 | rcu_read_unlock(); | |
289 | return item == swp_to_radix_entry(swap); | |
290 | } | |
291 | ||
46f65ec1 HD |
292 | /* |
293 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
294 | */ | |
295 | static int shmem_add_to_page_cache(struct page *page, | |
296 | struct address_space *mapping, | |
fed400a1 | 297 | pgoff_t index, void *expected) |
46f65ec1 | 298 | { |
b065b432 | 299 | int error; |
46f65ec1 | 300 | |
309381fe SL |
301 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
302 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
46f65ec1 | 303 | |
b065b432 HD |
304 | page_cache_get(page); |
305 | page->mapping = mapping; | |
306 | page->index = index; | |
307 | ||
308 | spin_lock_irq(&mapping->tree_lock); | |
46f65ec1 | 309 | if (!expected) |
b065b432 HD |
310 | error = radix_tree_insert(&mapping->page_tree, index, page); |
311 | else | |
312 | error = shmem_radix_tree_replace(mapping, index, expected, | |
313 | page); | |
46f65ec1 | 314 | if (!error) { |
b065b432 HD |
315 | mapping->nrpages++; |
316 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
317 | __inc_zone_page_state(page, NR_SHMEM); | |
318 | spin_unlock_irq(&mapping->tree_lock); | |
319 | } else { | |
320 | page->mapping = NULL; | |
321 | spin_unlock_irq(&mapping->tree_lock); | |
322 | page_cache_release(page); | |
46f65ec1 | 323 | } |
46f65ec1 HD |
324 | return error; |
325 | } | |
326 | ||
6922c0c7 HD |
327 | /* |
328 | * Like delete_from_page_cache, but substitutes swap for page. | |
329 | */ | |
330 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
331 | { | |
332 | struct address_space *mapping = page->mapping; | |
333 | int error; | |
334 | ||
335 | spin_lock_irq(&mapping->tree_lock); | |
336 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
337 | page->mapping = NULL; | |
338 | mapping->nrpages--; | |
339 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
340 | __dec_zone_page_state(page, NR_SHMEM); | |
341 | spin_unlock_irq(&mapping->tree_lock); | |
342 | page_cache_release(page); | |
343 | BUG_ON(error); | |
344 | } | |
345 | ||
7a5d0fbb HD |
346 | /* |
347 | * Remove swap entry from radix tree, free the swap and its page cache. | |
348 | */ | |
349 | static int shmem_free_swap(struct address_space *mapping, | |
350 | pgoff_t index, void *radswap) | |
351 | { | |
6dbaf22c | 352 | void *old; |
7a5d0fbb HD |
353 | |
354 | spin_lock_irq(&mapping->tree_lock); | |
6dbaf22c | 355 | old = radix_tree_delete_item(&mapping->page_tree, index, radswap); |
7a5d0fbb | 356 | spin_unlock_irq(&mapping->tree_lock); |
6dbaf22c JW |
357 | if (old != radswap) |
358 | return -ENOENT; | |
359 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
360 | return 0; | |
7a5d0fbb HD |
361 | } |
362 | ||
24513264 HD |
363 | /* |
364 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
365 | */ | |
366 | void shmem_unlock_mapping(struct address_space *mapping) | |
367 | { | |
368 | struct pagevec pvec; | |
369 | pgoff_t indices[PAGEVEC_SIZE]; | |
370 | pgoff_t index = 0; | |
371 | ||
372 | pagevec_init(&pvec, 0); | |
373 | /* | |
374 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
375 | */ | |
376 | while (!mapping_unevictable(mapping)) { | |
377 | /* | |
378 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
379 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
380 | */ | |
0cd6144a JW |
381 | pvec.nr = find_get_entries(mapping, index, |
382 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
383 | if (!pvec.nr) |
384 | break; | |
385 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 386 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
387 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
388 | pagevec_release(&pvec); | |
389 | cond_resched(); | |
390 | } | |
7a5d0fbb HD |
391 | } |
392 | ||
393 | /* | |
394 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 395 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 396 | */ |
1635f6a7 HD |
397 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
398 | bool unfalloc) | |
1da177e4 | 399 | { |
285b2c4f | 400 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 401 | struct shmem_inode_info *info = SHMEM_I(inode); |
285b2c4f | 402 | pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
83e4fa9c HD |
403 | pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT; |
404 | unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1); | |
405 | unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); | |
bda97eab | 406 | struct pagevec pvec; |
7a5d0fbb HD |
407 | pgoff_t indices[PAGEVEC_SIZE]; |
408 | long nr_swaps_freed = 0; | |
285b2c4f | 409 | pgoff_t index; |
bda97eab HD |
410 | int i; |
411 | ||
83e4fa9c HD |
412 | if (lend == -1) |
413 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
414 | |
415 | pagevec_init(&pvec, 0); | |
416 | index = start; | |
83e4fa9c | 417 | while (index < end) { |
0cd6144a JW |
418 | pvec.nr = find_get_entries(mapping, index, |
419 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
420 | pvec.pages, indices); | |
7a5d0fbb HD |
421 | if (!pvec.nr) |
422 | break; | |
bda97eab HD |
423 | for (i = 0; i < pagevec_count(&pvec); i++) { |
424 | struct page *page = pvec.pages[i]; | |
425 | ||
7a5d0fbb | 426 | index = indices[i]; |
83e4fa9c | 427 | if (index >= end) |
bda97eab HD |
428 | break; |
429 | ||
7a5d0fbb | 430 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
431 | if (unfalloc) |
432 | continue; | |
7a5d0fbb HD |
433 | nr_swaps_freed += !shmem_free_swap(mapping, |
434 | index, page); | |
bda97eab | 435 | continue; |
7a5d0fbb HD |
436 | } |
437 | ||
438 | if (!trylock_page(page)) | |
bda97eab | 439 | continue; |
1635f6a7 HD |
440 | if (!unfalloc || !PageUptodate(page)) { |
441 | if (page->mapping == mapping) { | |
309381fe | 442 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
443 | truncate_inode_page(mapping, page); |
444 | } | |
bda97eab | 445 | } |
bda97eab HD |
446 | unlock_page(page); |
447 | } | |
0cd6144a | 448 | pagevec_remove_exceptionals(&pvec); |
24513264 | 449 | pagevec_release(&pvec); |
bda97eab HD |
450 | cond_resched(); |
451 | index++; | |
452 | } | |
1da177e4 | 453 | |
83e4fa9c | 454 | if (partial_start) { |
bda97eab HD |
455 | struct page *page = NULL; |
456 | shmem_getpage(inode, start - 1, &page, SGP_READ, NULL); | |
457 | if (page) { | |
83e4fa9c HD |
458 | unsigned int top = PAGE_CACHE_SIZE; |
459 | if (start > end) { | |
460 | top = partial_end; | |
461 | partial_end = 0; | |
462 | } | |
463 | zero_user_segment(page, partial_start, top); | |
464 | set_page_dirty(page); | |
465 | unlock_page(page); | |
466 | page_cache_release(page); | |
467 | } | |
468 | } | |
469 | if (partial_end) { | |
470 | struct page *page = NULL; | |
471 | shmem_getpage(inode, end, &page, SGP_READ, NULL); | |
472 | if (page) { | |
473 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
474 | set_page_dirty(page); |
475 | unlock_page(page); | |
476 | page_cache_release(page); | |
477 | } | |
478 | } | |
83e4fa9c HD |
479 | if (start >= end) |
480 | return; | |
bda97eab HD |
481 | |
482 | index = start; | |
b1a36650 | 483 | while (index < end) { |
bda97eab | 484 | cond_resched(); |
0cd6144a JW |
485 | |
486 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 487 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 488 | pvec.pages, indices); |
7a5d0fbb | 489 | if (!pvec.nr) { |
b1a36650 HD |
490 | /* If all gone or hole-punch or unfalloc, we're done */ |
491 | if (index == start || end != -1) | |
bda97eab | 492 | break; |
b1a36650 | 493 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
494 | index = start; |
495 | continue; | |
496 | } | |
bda97eab HD |
497 | for (i = 0; i < pagevec_count(&pvec); i++) { |
498 | struct page *page = pvec.pages[i]; | |
499 | ||
7a5d0fbb | 500 | index = indices[i]; |
83e4fa9c | 501 | if (index >= end) |
bda97eab HD |
502 | break; |
503 | ||
7a5d0fbb | 504 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
505 | if (unfalloc) |
506 | continue; | |
b1a36650 HD |
507 | if (shmem_free_swap(mapping, index, page)) { |
508 | /* Swap was replaced by page: retry */ | |
509 | index--; | |
510 | break; | |
511 | } | |
512 | nr_swaps_freed++; | |
7a5d0fbb HD |
513 | continue; |
514 | } | |
515 | ||
bda97eab | 516 | lock_page(page); |
1635f6a7 HD |
517 | if (!unfalloc || !PageUptodate(page)) { |
518 | if (page->mapping == mapping) { | |
309381fe | 519 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 520 | truncate_inode_page(mapping, page); |
b1a36650 HD |
521 | } else { |
522 | /* Page was replaced by swap: retry */ | |
523 | unlock_page(page); | |
524 | index--; | |
525 | break; | |
1635f6a7 | 526 | } |
7a5d0fbb | 527 | } |
bda97eab HD |
528 | unlock_page(page); |
529 | } | |
0cd6144a | 530 | pagevec_remove_exceptionals(&pvec); |
24513264 | 531 | pagevec_release(&pvec); |
bda97eab HD |
532 | index++; |
533 | } | |
94c1e62d | 534 | |
1da177e4 | 535 | spin_lock(&info->lock); |
7a5d0fbb | 536 | info->swapped -= nr_swaps_freed; |
1da177e4 LT |
537 | shmem_recalc_inode(inode); |
538 | spin_unlock(&info->lock); | |
1635f6a7 | 539 | } |
1da177e4 | 540 | |
1635f6a7 HD |
541 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
542 | { | |
543 | shmem_undo_range(inode, lstart, lend, false); | |
285b2c4f | 544 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1da177e4 | 545 | } |
94c1e62d | 546 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 547 | |
94c1e62d | 548 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 LT |
549 | { |
550 | struct inode *inode = dentry->d_inode; | |
40e041a2 | 551 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 LT |
552 | int error; |
553 | ||
db78b877 CH |
554 | error = inode_change_ok(inode, attr); |
555 | if (error) | |
556 | return error; | |
557 | ||
94c1e62d HD |
558 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
559 | loff_t oldsize = inode->i_size; | |
560 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 561 | |
40e041a2 DH |
562 | /* protected by i_mutex */ |
563 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
564 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
565 | return -EPERM; | |
566 | ||
94c1e62d | 567 | if (newsize != oldsize) { |
77142517 KK |
568 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
569 | oldsize, newsize); | |
570 | if (error) | |
571 | return error; | |
94c1e62d HD |
572 | i_size_write(inode, newsize); |
573 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
574 | } | |
575 | if (newsize < oldsize) { | |
576 | loff_t holebegin = round_up(newsize, PAGE_SIZE); | |
577 | unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); | |
578 | shmem_truncate_range(inode, newsize, (loff_t)-1); | |
579 | /* unmap again to remove racily COWed private pages */ | |
580 | unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); | |
581 | } | |
1da177e4 LT |
582 | } |
583 | ||
db78b877 | 584 | setattr_copy(inode, attr); |
db78b877 | 585 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 586 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
587 | return error; |
588 | } | |
589 | ||
1f895f75 | 590 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 591 | { |
1da177e4 LT |
592 | struct shmem_inode_info *info = SHMEM_I(inode); |
593 | ||
3889e6e7 | 594 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
595 | shmem_unacct_size(info->flags, inode->i_size); |
596 | inode->i_size = 0; | |
3889e6e7 | 597 | shmem_truncate_range(inode, 0, (loff_t)-1); |
1da177e4 | 598 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 599 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 600 | list_del_init(&info->swaplist); |
cb5f7b9a | 601 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 602 | } |
69f07ec9 HD |
603 | } else |
604 | kfree(info->symlink); | |
b09e0fa4 | 605 | |
38f38657 | 606 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 607 | WARN_ON(inode->i_blocks); |
5b04c689 | 608 | shmem_free_inode(inode->i_sb); |
dbd5768f | 609 | clear_inode(inode); |
1da177e4 LT |
610 | } |
611 | ||
46f65ec1 HD |
612 | /* |
613 | * If swap found in inode, free it and move page from swapcache to filecache. | |
614 | */ | |
41ffe5d5 | 615 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 616 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 617 | { |
285b2c4f | 618 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 619 | void *radswap; |
41ffe5d5 | 620 | pgoff_t index; |
bde05d1c HD |
621 | gfp_t gfp; |
622 | int error = 0; | |
1da177e4 | 623 | |
46f65ec1 | 624 | radswap = swp_to_radix_entry(swap); |
e504f3fd | 625 | index = radix_tree_locate_item(&mapping->page_tree, radswap); |
46f65ec1 | 626 | if (index == -1) |
00501b53 | 627 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 628 | |
1b1b32f2 HD |
629 | /* |
630 | * Move _head_ to start search for next from here. | |
1f895f75 | 631 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 632 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 633 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
634 | */ |
635 | if (shmem_swaplist.next != &info->swaplist) | |
636 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 637 | |
bde05d1c HD |
638 | gfp = mapping_gfp_mask(mapping); |
639 | if (shmem_should_replace_page(*pagep, gfp)) { | |
640 | mutex_unlock(&shmem_swaplist_mutex); | |
641 | error = shmem_replace_page(pagep, gfp, info, index); | |
642 | mutex_lock(&shmem_swaplist_mutex); | |
643 | /* | |
644 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
645 | * allocation, but the inode might have been freed while we |
646 | * dropped it: although a racing shmem_evict_inode() cannot | |
647 | * complete without emptying the radix_tree, our page lock | |
648 | * on this swapcache page is not enough to prevent that - | |
649 | * free_swap_and_cache() of our swap entry will only | |
650 | * trylock_page(), removing swap from radix_tree whatever. | |
651 | * | |
652 | * We must not proceed to shmem_add_to_page_cache() if the | |
653 | * inode has been freed, but of course we cannot rely on | |
654 | * inode or mapping or info to check that. However, we can | |
655 | * safely check if our swap entry is still in use (and here | |
656 | * it can't have got reused for another page): if it's still | |
657 | * in use, then the inode cannot have been freed yet, and we | |
658 | * can safely proceed (if it's no longer in use, that tells | |
659 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
660 | */ |
661 | if (!page_swapcount(*pagep)) | |
662 | error = -ENOENT; | |
663 | } | |
664 | ||
d13d1443 | 665 | /* |
778dd893 HD |
666 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
667 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
668 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 669 | */ |
bde05d1c HD |
670 | if (!error) |
671 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 672 | radswap); |
48f170fb | 673 | if (error != -ENOMEM) { |
46f65ec1 HD |
674 | /* |
675 | * Truncation and eviction use free_swap_and_cache(), which | |
676 | * only does trylock page: if we raced, best clean up here. | |
677 | */ | |
bde05d1c HD |
678 | delete_from_swap_cache(*pagep); |
679 | set_page_dirty(*pagep); | |
46f65ec1 HD |
680 | if (!error) { |
681 | spin_lock(&info->lock); | |
682 | info->swapped--; | |
683 | spin_unlock(&info->lock); | |
684 | swap_free(swap); | |
685 | } | |
1da177e4 | 686 | } |
2e0e26c7 | 687 | return error; |
1da177e4 LT |
688 | } |
689 | ||
690 | /* | |
46f65ec1 | 691 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 692 | */ |
41ffe5d5 | 693 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 694 | { |
41ffe5d5 | 695 | struct list_head *this, *next; |
1da177e4 | 696 | struct shmem_inode_info *info; |
00501b53 | 697 | struct mem_cgroup *memcg; |
bde05d1c HD |
698 | int error = 0; |
699 | ||
700 | /* | |
701 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 702 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 703 | */ |
0142ef6c | 704 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 705 | goto out; |
778dd893 HD |
706 | |
707 | /* | |
708 | * Charge page using GFP_KERNEL while we can wait, before taking | |
709 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
710 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 711 | */ |
00501b53 | 712 | error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg); |
778dd893 HD |
713 | if (error) |
714 | goto out; | |
46f65ec1 | 715 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 716 | error = -EAGAIN; |
1da177e4 | 717 | |
cb5f7b9a | 718 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
719 | list_for_each_safe(this, next, &shmem_swaplist) { |
720 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 721 | if (info->swapped) |
00501b53 | 722 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
723 | else |
724 | list_del_init(&info->swaplist); | |
cb5f7b9a | 725 | cond_resched(); |
00501b53 | 726 | if (error != -EAGAIN) |
778dd893 | 727 | break; |
00501b53 | 728 | /* found nothing in this: move on to search the next */ |
1da177e4 | 729 | } |
cb5f7b9a | 730 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 731 | |
00501b53 JW |
732 | if (error) { |
733 | if (error != -ENOMEM) | |
734 | error = 0; | |
735 | mem_cgroup_cancel_charge(page, memcg); | |
736 | } else | |
737 | mem_cgroup_commit_charge(page, memcg, true); | |
778dd893 | 738 | out: |
aaa46865 HD |
739 | unlock_page(page); |
740 | page_cache_release(page); | |
778dd893 | 741 | return error; |
1da177e4 LT |
742 | } |
743 | ||
744 | /* | |
745 | * Move the page from the page cache to the swap cache. | |
746 | */ | |
747 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
748 | { | |
749 | struct shmem_inode_info *info; | |
1da177e4 | 750 | struct address_space *mapping; |
1da177e4 | 751 | struct inode *inode; |
6922c0c7 HD |
752 | swp_entry_t swap; |
753 | pgoff_t index; | |
1da177e4 LT |
754 | |
755 | BUG_ON(!PageLocked(page)); | |
1da177e4 LT |
756 | mapping = page->mapping; |
757 | index = page->index; | |
758 | inode = mapping->host; | |
759 | info = SHMEM_I(inode); | |
760 | if (info->flags & VM_LOCKED) | |
761 | goto redirty; | |
d9fe526a | 762 | if (!total_swap_pages) |
1da177e4 LT |
763 | goto redirty; |
764 | ||
d9fe526a HD |
765 | /* |
766 | * shmem_backing_dev_info's capabilities prevent regular writeback or | |
767 | * sync from ever calling shmem_writepage; but a stacking filesystem | |
48f170fb | 768 | * might use ->writepage of its underlying filesystem, in which case |
d9fe526a | 769 | * tmpfs should write out to swap only in response to memory pressure, |
48f170fb | 770 | * and not for the writeback threads or sync. |
d9fe526a | 771 | */ |
48f170fb HD |
772 | if (!wbc->for_reclaim) { |
773 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
774 | goto redirty; | |
775 | } | |
1635f6a7 HD |
776 | |
777 | /* | |
778 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
779 | * value into swapfile.c, the only way we can correctly account for a | |
780 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
781 | * |
782 | * That's okay for a page already fallocated earlier, but if we have | |
783 | * not yet completed the fallocation, then (a) we want to keep track | |
784 | * of this page in case we have to undo it, and (b) it may not be a | |
785 | * good idea to continue anyway, once we're pushing into swap. So | |
786 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
787 | */ |
788 | if (!PageUptodate(page)) { | |
1aac1400 HD |
789 | if (inode->i_private) { |
790 | struct shmem_falloc *shmem_falloc; | |
791 | spin_lock(&inode->i_lock); | |
792 | shmem_falloc = inode->i_private; | |
793 | if (shmem_falloc && | |
8e205f77 | 794 | !shmem_falloc->waitq && |
1aac1400 HD |
795 | index >= shmem_falloc->start && |
796 | index < shmem_falloc->next) | |
797 | shmem_falloc->nr_unswapped++; | |
798 | else | |
799 | shmem_falloc = NULL; | |
800 | spin_unlock(&inode->i_lock); | |
801 | if (shmem_falloc) | |
802 | goto redirty; | |
803 | } | |
1635f6a7 HD |
804 | clear_highpage(page); |
805 | flush_dcache_page(page); | |
806 | SetPageUptodate(page); | |
807 | } | |
808 | ||
48f170fb HD |
809 | swap = get_swap_page(); |
810 | if (!swap.val) | |
811 | goto redirty; | |
d9fe526a | 812 | |
b1dea800 HD |
813 | /* |
814 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
815 | * if it's not already there. Do it now before the page is |
816 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 817 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
818 | * we've incremented swapped, because shmem_unuse_inode() will |
819 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 820 | */ |
48f170fb HD |
821 | mutex_lock(&shmem_swaplist_mutex); |
822 | if (list_empty(&info->swaplist)) | |
823 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 824 | |
48f170fb | 825 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
aaa46865 | 826 | swap_shmem_alloc(swap); |
6922c0c7 HD |
827 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); |
828 | ||
829 | spin_lock(&info->lock); | |
830 | info->swapped++; | |
831 | shmem_recalc_inode(inode); | |
826267cf | 832 | spin_unlock(&info->lock); |
6922c0c7 HD |
833 | |
834 | mutex_unlock(&shmem_swaplist_mutex); | |
d9fe526a | 835 | BUG_ON(page_mapped(page)); |
9fab5619 | 836 | swap_writepage(page, wbc); |
1da177e4 LT |
837 | return 0; |
838 | } | |
839 | ||
6922c0c7 | 840 | mutex_unlock(&shmem_swaplist_mutex); |
0a31bc97 | 841 | swapcache_free(swap); |
1da177e4 LT |
842 | redirty: |
843 | set_page_dirty(page); | |
d9fe526a HD |
844 | if (wbc->for_reclaim) |
845 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
846 | unlock_page(page); | |
847 | return 0; | |
1da177e4 LT |
848 | } |
849 | ||
850 | #ifdef CONFIG_NUMA | |
680d794b | 851 | #ifdef CONFIG_TMPFS |
71fe804b | 852 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 853 | { |
095f1fc4 | 854 | char buffer[64]; |
680d794b | 855 | |
71fe804b | 856 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 857 | return; /* show nothing */ |
680d794b | 858 | |
a7a88b23 | 859 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
860 | |
861 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 862 | } |
71fe804b LS |
863 | |
864 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
865 | { | |
866 | struct mempolicy *mpol = NULL; | |
867 | if (sbinfo->mpol) { | |
868 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
869 | mpol = sbinfo->mpol; | |
870 | mpol_get(mpol); | |
871 | spin_unlock(&sbinfo->stat_lock); | |
872 | } | |
873 | return mpol; | |
874 | } | |
680d794b AM |
875 | #endif /* CONFIG_TMPFS */ |
876 | ||
41ffe5d5 HD |
877 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
878 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 879 | { |
1da177e4 | 880 | struct vm_area_struct pvma; |
18a2f371 | 881 | struct page *page; |
52cd3b07 | 882 | |
1da177e4 | 883 | /* Create a pseudo vma that just contains the policy */ |
c4cc6d07 | 884 | pvma.vm_start = 0; |
09c231cb NZ |
885 | /* Bias interleave by inode number to distribute better across nodes */ |
886 | pvma.vm_pgoff = index + info->vfs_inode.i_ino; | |
c4cc6d07 | 887 | pvma.vm_ops = NULL; |
18a2f371 MG |
888 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
889 | ||
890 | page = swapin_readahead(swap, gfp, &pvma, 0); | |
891 | ||
892 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
893 | mpol_cond_put(pvma.vm_policy); | |
894 | ||
895 | return page; | |
1da177e4 LT |
896 | } |
897 | ||
02098fea | 898 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 899 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
900 | { |
901 | struct vm_area_struct pvma; | |
18a2f371 | 902 | struct page *page; |
1da177e4 | 903 | |
c4cc6d07 HD |
904 | /* Create a pseudo vma that just contains the policy */ |
905 | pvma.vm_start = 0; | |
09c231cb NZ |
906 | /* Bias interleave by inode number to distribute better across nodes */ |
907 | pvma.vm_pgoff = index + info->vfs_inode.i_ino; | |
c4cc6d07 | 908 | pvma.vm_ops = NULL; |
41ffe5d5 | 909 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
52cd3b07 | 910 | |
18a2f371 MG |
911 | page = alloc_page_vma(gfp, &pvma, 0); |
912 | ||
913 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
914 | mpol_cond_put(pvma.vm_policy); | |
915 | ||
916 | return page; | |
1da177e4 | 917 | } |
680d794b AM |
918 | #else /* !CONFIG_NUMA */ |
919 | #ifdef CONFIG_TMPFS | |
41ffe5d5 | 920 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b AM |
921 | { |
922 | } | |
923 | #endif /* CONFIG_TMPFS */ | |
924 | ||
41ffe5d5 HD |
925 | static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
926 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 927 | { |
41ffe5d5 | 928 | return swapin_readahead(swap, gfp, NULL, 0); |
1da177e4 LT |
929 | } |
930 | ||
02098fea | 931 | static inline struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 932 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 | 933 | { |
e84e2e13 | 934 | return alloc_page(gfp); |
1da177e4 | 935 | } |
680d794b | 936 | #endif /* CONFIG_NUMA */ |
1da177e4 | 937 | |
71fe804b LS |
938 | #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) |
939 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
940 | { | |
941 | return NULL; | |
942 | } | |
943 | #endif | |
944 | ||
bde05d1c HD |
945 | /* |
946 | * When a page is moved from swapcache to shmem filecache (either by the | |
947 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
948 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
949 | * ignorance of the mapping it belongs to. If that mapping has special | |
950 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
951 | * we may need to copy to a suitable page before moving to filecache. | |
952 | * | |
953 | * In a future release, this may well be extended to respect cpuset and | |
954 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
955 | * but for now it is a simple matter of zone. | |
956 | */ | |
957 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
958 | { | |
959 | return page_zonenum(page) > gfp_zone(gfp); | |
960 | } | |
961 | ||
962 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
963 | struct shmem_inode_info *info, pgoff_t index) | |
964 | { | |
965 | struct page *oldpage, *newpage; | |
966 | struct address_space *swap_mapping; | |
967 | pgoff_t swap_index; | |
968 | int error; | |
969 | ||
970 | oldpage = *pagep; | |
971 | swap_index = page_private(oldpage); | |
972 | swap_mapping = page_mapping(oldpage); | |
973 | ||
974 | /* | |
975 | * We have arrived here because our zones are constrained, so don't | |
976 | * limit chance of success by further cpuset and node constraints. | |
977 | */ | |
978 | gfp &= ~GFP_CONSTRAINT_MASK; | |
979 | newpage = shmem_alloc_page(gfp, info, index); | |
980 | if (!newpage) | |
981 | return -ENOMEM; | |
bde05d1c | 982 | |
bde05d1c HD |
983 | page_cache_get(newpage); |
984 | copy_highpage(newpage, oldpage); | |
0142ef6c | 985 | flush_dcache_page(newpage); |
bde05d1c | 986 | |
bde05d1c | 987 | __set_page_locked(newpage); |
bde05d1c | 988 | SetPageUptodate(newpage); |
bde05d1c | 989 | SetPageSwapBacked(newpage); |
bde05d1c | 990 | set_page_private(newpage, swap_index); |
bde05d1c HD |
991 | SetPageSwapCache(newpage); |
992 | ||
993 | /* | |
994 | * Our caller will very soon move newpage out of swapcache, but it's | |
995 | * a nice clean interface for us to replace oldpage by newpage there. | |
996 | */ | |
997 | spin_lock_irq(&swap_mapping->tree_lock); | |
998 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
999 | newpage); | |
0142ef6c HD |
1000 | if (!error) { |
1001 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
1002 | __dec_zone_page_state(oldpage, NR_FILE_PAGES); | |
1003 | } | |
bde05d1c | 1004 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1005 | |
0142ef6c HD |
1006 | if (unlikely(error)) { |
1007 | /* | |
1008 | * Is this possible? I think not, now that our callers check | |
1009 | * both PageSwapCache and page_private after getting page lock; | |
1010 | * but be defensive. Reverse old to newpage for clear and free. | |
1011 | */ | |
1012 | oldpage = newpage; | |
1013 | } else { | |
0a31bc97 | 1014 | mem_cgroup_migrate(oldpage, newpage, false); |
0142ef6c HD |
1015 | lru_cache_add_anon(newpage); |
1016 | *pagep = newpage; | |
1017 | } | |
bde05d1c HD |
1018 | |
1019 | ClearPageSwapCache(oldpage); | |
1020 | set_page_private(oldpage, 0); | |
1021 | ||
1022 | unlock_page(oldpage); | |
1023 | page_cache_release(oldpage); | |
1024 | page_cache_release(oldpage); | |
0142ef6c | 1025 | return error; |
bde05d1c HD |
1026 | } |
1027 | ||
1da177e4 | 1028 | /* |
68da9f05 | 1029 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1030 | * |
1031 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1032 | * vm. If we swap it in we mark it dirty since we also free the swap | |
1033 | * entry since a page cannot live in both the swap and page cache | |
1034 | */ | |
41ffe5d5 | 1035 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
68da9f05 | 1036 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type) |
1da177e4 LT |
1037 | { |
1038 | struct address_space *mapping = inode->i_mapping; | |
54af6042 | 1039 | struct shmem_inode_info *info; |
1da177e4 | 1040 | struct shmem_sb_info *sbinfo; |
00501b53 | 1041 | struct mem_cgroup *memcg; |
27ab7006 | 1042 | struct page *page; |
1da177e4 LT |
1043 | swp_entry_t swap; |
1044 | int error; | |
54af6042 | 1045 | int once = 0; |
1635f6a7 | 1046 | int alloced = 0; |
1da177e4 | 1047 | |
41ffe5d5 | 1048 | if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT)) |
1da177e4 | 1049 | return -EFBIG; |
1da177e4 | 1050 | repeat: |
54af6042 | 1051 | swap.val = 0; |
0cd6144a | 1052 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1053 | if (radix_tree_exceptional_entry(page)) { |
1054 | swap = radix_to_swp_entry(page); | |
1055 | page = NULL; | |
1056 | } | |
1057 | ||
1635f6a7 | 1058 | if (sgp != SGP_WRITE && sgp != SGP_FALLOC && |
54af6042 HD |
1059 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { |
1060 | error = -EINVAL; | |
1061 | goto failed; | |
1062 | } | |
1063 | ||
66d2f4d2 HD |
1064 | if (page && sgp == SGP_WRITE) |
1065 | mark_page_accessed(page); | |
1066 | ||
1635f6a7 HD |
1067 | /* fallocated page? */ |
1068 | if (page && !PageUptodate(page)) { | |
1069 | if (sgp != SGP_READ) | |
1070 | goto clear; | |
1071 | unlock_page(page); | |
1072 | page_cache_release(page); | |
1073 | page = NULL; | |
1074 | } | |
54af6042 | 1075 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1076 | *pagep = page; |
1077 | return 0; | |
27ab7006 HD |
1078 | } |
1079 | ||
1080 | /* | |
54af6042 HD |
1081 | * Fast cache lookup did not find it: |
1082 | * bring it back from swap or allocate. | |
27ab7006 | 1083 | */ |
54af6042 HD |
1084 | info = SHMEM_I(inode); |
1085 | sbinfo = SHMEM_SB(inode->i_sb); | |
1da177e4 | 1086 | |
1da177e4 LT |
1087 | if (swap.val) { |
1088 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1089 | page = lookup_swap_cache(swap); |
1090 | if (!page) { | |
1da177e4 | 1091 | /* here we actually do the io */ |
68da9f05 HD |
1092 | if (fault_type) |
1093 | *fault_type |= VM_FAULT_MAJOR; | |
41ffe5d5 | 1094 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1095 | if (!page) { |
54af6042 HD |
1096 | error = -ENOMEM; |
1097 | goto failed; | |
1da177e4 | 1098 | } |
1da177e4 LT |
1099 | } |
1100 | ||
1101 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1102 | lock_page(page); |
0142ef6c | 1103 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1104 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1105 | error = -EEXIST; /* try again */ |
d1899228 | 1106 | goto unlock; |
bde05d1c | 1107 | } |
27ab7006 | 1108 | if (!PageUptodate(page)) { |
1da177e4 | 1109 | error = -EIO; |
54af6042 | 1110 | goto failed; |
1da177e4 | 1111 | } |
54af6042 HD |
1112 | wait_on_page_writeback(page); |
1113 | ||
bde05d1c HD |
1114 | if (shmem_should_replace_page(page, gfp)) { |
1115 | error = shmem_replace_page(&page, gfp, info, index); | |
1116 | if (error) | |
1117 | goto failed; | |
1da177e4 | 1118 | } |
27ab7006 | 1119 | |
00501b53 | 1120 | error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg); |
d1899228 | 1121 | if (!error) { |
aa3b1895 | 1122 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1123 | swp_to_radix_entry(swap)); |
215c02bc HD |
1124 | /* |
1125 | * We already confirmed swap under page lock, and make | |
1126 | * no memory allocation here, so usually no possibility | |
1127 | * of error; but free_swap_and_cache() only trylocks a | |
1128 | * page, so it is just possible that the entry has been | |
1129 | * truncated or holepunched since swap was confirmed. | |
1130 | * shmem_undo_range() will have done some of the | |
1131 | * unaccounting, now delete_from_swap_cache() will do | |
1132 | * the rest (including mem_cgroup_uncharge_swapcache). | |
1133 | * Reset swap.val? No, leave it so "failed" goes back to | |
1134 | * "repeat": reading a hole and writing should succeed. | |
1135 | */ | |
00501b53 JW |
1136 | if (error) { |
1137 | mem_cgroup_cancel_charge(page, memcg); | |
215c02bc | 1138 | delete_from_swap_cache(page); |
00501b53 | 1139 | } |
d1899228 | 1140 | } |
54af6042 HD |
1141 | if (error) |
1142 | goto failed; | |
1143 | ||
00501b53 JW |
1144 | mem_cgroup_commit_charge(page, memcg, true); |
1145 | ||
54af6042 | 1146 | spin_lock(&info->lock); |
285b2c4f | 1147 | info->swapped--; |
54af6042 | 1148 | shmem_recalc_inode(inode); |
27ab7006 | 1149 | spin_unlock(&info->lock); |
54af6042 | 1150 | |
66d2f4d2 HD |
1151 | if (sgp == SGP_WRITE) |
1152 | mark_page_accessed(page); | |
1153 | ||
54af6042 | 1154 | delete_from_swap_cache(page); |
27ab7006 HD |
1155 | set_page_dirty(page); |
1156 | swap_free(swap); | |
1157 | ||
54af6042 HD |
1158 | } else { |
1159 | if (shmem_acct_block(info->flags)) { | |
1160 | error = -ENOSPC; | |
1161 | goto failed; | |
1da177e4 | 1162 | } |
0edd73b3 | 1163 | if (sbinfo->max_blocks) { |
fc5da22a | 1164 | if (percpu_counter_compare(&sbinfo->used_blocks, |
54af6042 HD |
1165 | sbinfo->max_blocks) >= 0) { |
1166 | error = -ENOSPC; | |
1167 | goto unacct; | |
1168 | } | |
7e496299 | 1169 | percpu_counter_inc(&sbinfo->used_blocks); |
54af6042 | 1170 | } |
1da177e4 | 1171 | |
54af6042 HD |
1172 | page = shmem_alloc_page(gfp, info, index); |
1173 | if (!page) { | |
1174 | error = -ENOMEM; | |
1175 | goto decused; | |
1da177e4 LT |
1176 | } |
1177 | ||
07a42788 | 1178 | __SetPageSwapBacked(page); |
54af6042 | 1179 | __set_page_locked(page); |
66d2f4d2 | 1180 | if (sgp == SGP_WRITE) |
eb39d618 | 1181 | __SetPageReferenced(page); |
66d2f4d2 | 1182 | |
00501b53 | 1183 | error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg); |
54af6042 HD |
1184 | if (error) |
1185 | goto decused; | |
5e4c0d97 | 1186 | error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK); |
b065b432 HD |
1187 | if (!error) { |
1188 | error = shmem_add_to_page_cache(page, mapping, index, | |
fed400a1 | 1189 | NULL); |
b065b432 HD |
1190 | radix_tree_preload_end(); |
1191 | } | |
1192 | if (error) { | |
00501b53 | 1193 | mem_cgroup_cancel_charge(page, memcg); |
b065b432 HD |
1194 | goto decused; |
1195 | } | |
00501b53 | 1196 | mem_cgroup_commit_charge(page, memcg, false); |
54af6042 HD |
1197 | lru_cache_add_anon(page); |
1198 | ||
1199 | spin_lock(&info->lock); | |
1da177e4 | 1200 | info->alloced++; |
54af6042 HD |
1201 | inode->i_blocks += BLOCKS_PER_PAGE; |
1202 | shmem_recalc_inode(inode); | |
1da177e4 | 1203 | spin_unlock(&info->lock); |
1635f6a7 | 1204 | alloced = true; |
54af6042 | 1205 | |
ec9516fb | 1206 | /* |
1635f6a7 HD |
1207 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1208 | */ | |
1209 | if (sgp == SGP_FALLOC) | |
1210 | sgp = SGP_WRITE; | |
1211 | clear: | |
1212 | /* | |
1213 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1214 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1215 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb HD |
1216 | */ |
1217 | if (sgp != SGP_WRITE) { | |
1218 | clear_highpage(page); | |
1219 | flush_dcache_page(page); | |
1220 | SetPageUptodate(page); | |
1221 | } | |
a0ee5ec5 | 1222 | if (sgp == SGP_DIRTY) |
27ab7006 | 1223 | set_page_dirty(page); |
1da177e4 | 1224 | } |
bde05d1c | 1225 | |
54af6042 | 1226 | /* Perhaps the file has been truncated since we checked */ |
1635f6a7 | 1227 | if (sgp != SGP_WRITE && sgp != SGP_FALLOC && |
54af6042 HD |
1228 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { |
1229 | error = -EINVAL; | |
1635f6a7 HD |
1230 | if (alloced) |
1231 | goto trunc; | |
1232 | else | |
1233 | goto failed; | |
e83c32e8 | 1234 | } |
54af6042 HD |
1235 | *pagep = page; |
1236 | return 0; | |
1da177e4 | 1237 | |
59a16ead | 1238 | /* |
54af6042 | 1239 | * Error recovery. |
59a16ead | 1240 | */ |
54af6042 | 1241 | trunc: |
1635f6a7 | 1242 | info = SHMEM_I(inode); |
54af6042 HD |
1243 | ClearPageDirty(page); |
1244 | delete_from_page_cache(page); | |
1245 | spin_lock(&info->lock); | |
1246 | info->alloced--; | |
1247 | inode->i_blocks -= BLOCKS_PER_PAGE; | |
59a16ead | 1248 | spin_unlock(&info->lock); |
54af6042 | 1249 | decused: |
1635f6a7 | 1250 | sbinfo = SHMEM_SB(inode->i_sb); |
54af6042 HD |
1251 | if (sbinfo->max_blocks) |
1252 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
1253 | unacct: | |
1254 | shmem_unacct_blocks(info->flags, 1); | |
1255 | failed: | |
d1899228 HD |
1256 | if (swap.val && error != -EINVAL && |
1257 | !shmem_confirm_swap(mapping, index, swap)) | |
1258 | error = -EEXIST; | |
1259 | unlock: | |
27ab7006 | 1260 | if (page) { |
54af6042 | 1261 | unlock_page(page); |
27ab7006 | 1262 | page_cache_release(page); |
54af6042 HD |
1263 | } |
1264 | if (error == -ENOSPC && !once++) { | |
1265 | info = SHMEM_I(inode); | |
1266 | spin_lock(&info->lock); | |
1267 | shmem_recalc_inode(inode); | |
1268 | spin_unlock(&info->lock); | |
27ab7006 | 1269 | goto repeat; |
ff36b801 | 1270 | } |
d1899228 | 1271 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1272 | goto repeat; |
1273 | return error; | |
1da177e4 LT |
1274 | } |
1275 | ||
d0217ac0 | 1276 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1277 | { |
496ad9aa | 1278 | struct inode *inode = file_inode(vma->vm_file); |
1da177e4 | 1279 | int error; |
68da9f05 | 1280 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1281 | |
f00cdc6d HD |
1282 | /* |
1283 | * Trinity finds that probing a hole which tmpfs is punching can | |
1284 | * prevent the hole-punch from ever completing: which in turn | |
1285 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1286 | * faulting pages into the hole while it's being punched. Although |
1287 | * shmem_undo_range() does remove the additions, it may be unable to | |
1288 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1289 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1290 | * | |
1291 | * It does not matter if we sometimes reach this check just before the | |
1292 | * hole-punch begins, so that one fault then races with the punch: | |
1293 | * we just need to make racing faults a rare case. | |
1294 | * | |
1295 | * The implementation below would be much simpler if we just used a | |
1296 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1297 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1298 | */ |
1299 | if (unlikely(inode->i_private)) { | |
1300 | struct shmem_falloc *shmem_falloc; | |
1301 | ||
1302 | spin_lock(&inode->i_lock); | |
1303 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1304 | if (shmem_falloc && |
1305 | shmem_falloc->waitq && | |
1306 | vmf->pgoff >= shmem_falloc->start && | |
1307 | vmf->pgoff < shmem_falloc->next) { | |
1308 | wait_queue_head_t *shmem_falloc_waitq; | |
1309 | DEFINE_WAIT(shmem_fault_wait); | |
1310 | ||
1311 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1312 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1313 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1314 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1315 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1316 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1317 | } |
8e205f77 HD |
1318 | |
1319 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1320 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1321 | TASK_UNINTERRUPTIBLE); | |
1322 | spin_unlock(&inode->i_lock); | |
1323 | schedule(); | |
1324 | ||
1325 | /* | |
1326 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1327 | * stack of the hole-punching task: shmem_falloc_waitq | |
1328 | * is usually invalid by the time we reach here, but | |
1329 | * finish_wait() does not dereference it in that case; | |
1330 | * though i_lock needed lest racing with wake_up_all(). | |
1331 | */ | |
1332 | spin_lock(&inode->i_lock); | |
1333 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1334 | spin_unlock(&inode->i_lock); | |
1335 | return ret; | |
f00cdc6d | 1336 | } |
8e205f77 | 1337 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
1338 | } |
1339 | ||
27d54b39 | 1340 | error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); |
d0217ac0 NP |
1341 | if (error) |
1342 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1343 | |
456f998e YH |
1344 | if (ret & VM_FAULT_MAJOR) { |
1345 | count_vm_event(PGMAJFAULT); | |
1346 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); | |
1347 | } | |
68da9f05 | 1348 | return ret; |
1da177e4 LT |
1349 | } |
1350 | ||
1da177e4 | 1351 | #ifdef CONFIG_NUMA |
41ffe5d5 | 1352 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 1353 | { |
496ad9aa | 1354 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1355 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
1356 | } |
1357 | ||
d8dc74f2 AB |
1358 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
1359 | unsigned long addr) | |
1da177e4 | 1360 | { |
496ad9aa | 1361 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1362 | pgoff_t index; |
1da177e4 | 1363 | |
41ffe5d5 HD |
1364 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
1365 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
1366 | } |
1367 | #endif | |
1368 | ||
1369 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
1370 | { | |
496ad9aa | 1371 | struct inode *inode = file_inode(file); |
1da177e4 LT |
1372 | struct shmem_inode_info *info = SHMEM_I(inode); |
1373 | int retval = -ENOMEM; | |
1374 | ||
1375 | spin_lock(&info->lock); | |
1376 | if (lock && !(info->flags & VM_LOCKED)) { | |
1377 | if (!user_shm_lock(inode->i_size, user)) | |
1378 | goto out_nomem; | |
1379 | info->flags |= VM_LOCKED; | |
89e004ea | 1380 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
1381 | } |
1382 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
1383 | user_shm_unlock(inode->i_size, user); | |
1384 | info->flags &= ~VM_LOCKED; | |
89e004ea | 1385 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
1386 | } |
1387 | retval = 0; | |
89e004ea | 1388 | |
1da177e4 LT |
1389 | out_nomem: |
1390 | spin_unlock(&info->lock); | |
1391 | return retval; | |
1392 | } | |
1393 | ||
9b83a6a8 | 1394 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
1395 | { |
1396 | file_accessed(file); | |
1397 | vma->vm_ops = &shmem_vm_ops; | |
1398 | return 0; | |
1399 | } | |
1400 | ||
454abafe | 1401 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 1402 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
1403 | { |
1404 | struct inode *inode; | |
1405 | struct shmem_inode_info *info; | |
1406 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
1407 | ||
5b04c689 PE |
1408 | if (shmem_reserve_inode(sb)) |
1409 | return NULL; | |
1da177e4 LT |
1410 | |
1411 | inode = new_inode(sb); | |
1412 | if (inode) { | |
85fe4025 | 1413 | inode->i_ino = get_next_ino(); |
454abafe | 1414 | inode_init_owner(inode, dir, mode); |
1da177e4 | 1415 | inode->i_blocks = 0; |
1da177e4 LT |
1416 | inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; |
1417 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
91828a40 | 1418 | inode->i_generation = get_seconds(); |
1da177e4 LT |
1419 | info = SHMEM_I(inode); |
1420 | memset(info, 0, (char *)inode - (char *)info); | |
1421 | spin_lock_init(&info->lock); | |
40e041a2 | 1422 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 1423 | info->flags = flags & VM_NORESERVE; |
1da177e4 | 1424 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 1425 | simple_xattrs_init(&info->xattrs); |
72c04902 | 1426 | cache_no_acl(inode); |
1da177e4 LT |
1427 | |
1428 | switch (mode & S_IFMT) { | |
1429 | default: | |
39f0247d | 1430 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
1431 | init_special_inode(inode, mode, dev); |
1432 | break; | |
1433 | case S_IFREG: | |
14fcc23f | 1434 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
1435 | inode->i_op = &shmem_inode_operations; |
1436 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
1437 | mpol_shared_policy_init(&info->policy, |
1438 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
1439 | break; |
1440 | case S_IFDIR: | |
d8c76e6f | 1441 | inc_nlink(inode); |
1da177e4 LT |
1442 | /* Some things misbehave if size == 0 on a directory */ |
1443 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
1444 | inode->i_op = &shmem_dir_inode_operations; | |
1445 | inode->i_fop = &simple_dir_operations; | |
1446 | break; | |
1447 | case S_IFLNK: | |
1448 | /* | |
1449 | * Must not load anything in the rbtree, | |
1450 | * mpol_free_shared_policy will not be called. | |
1451 | */ | |
71fe804b | 1452 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
1453 | break; |
1454 | } | |
5b04c689 PE |
1455 | } else |
1456 | shmem_free_inode(sb); | |
1da177e4 LT |
1457 | return inode; |
1458 | } | |
1459 | ||
0cd6144a JW |
1460 | bool shmem_mapping(struct address_space *mapping) |
1461 | { | |
1462 | return mapping->backing_dev_info == &shmem_backing_dev_info; | |
1463 | } | |
1464 | ||
1da177e4 | 1465 | #ifdef CONFIG_TMPFS |
92e1d5be | 1466 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 1467 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 1468 | |
6d9d88d0 JS |
1469 | #ifdef CONFIG_TMPFS_XATTR |
1470 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
1471 | #else | |
1472 | #define shmem_initxattrs NULL | |
1473 | #endif | |
1474 | ||
1da177e4 | 1475 | static int |
800d15a5 NP |
1476 | shmem_write_begin(struct file *file, struct address_space *mapping, |
1477 | loff_t pos, unsigned len, unsigned flags, | |
1478 | struct page **pagep, void **fsdata) | |
1da177e4 | 1479 | { |
800d15a5 | 1480 | struct inode *inode = mapping->host; |
40e041a2 | 1481 | struct shmem_inode_info *info = SHMEM_I(inode); |
800d15a5 | 1482 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
40e041a2 DH |
1483 | |
1484 | /* i_mutex is held by caller */ | |
1485 | if (unlikely(info->seals)) { | |
1486 | if (info->seals & F_SEAL_WRITE) | |
1487 | return -EPERM; | |
1488 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
1489 | return -EPERM; | |
1490 | } | |
1491 | ||
66d2f4d2 | 1492 | return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); |
800d15a5 NP |
1493 | } |
1494 | ||
1495 | static int | |
1496 | shmem_write_end(struct file *file, struct address_space *mapping, | |
1497 | loff_t pos, unsigned len, unsigned copied, | |
1498 | struct page *page, void *fsdata) | |
1499 | { | |
1500 | struct inode *inode = mapping->host; | |
1501 | ||
d3602444 HD |
1502 | if (pos + copied > inode->i_size) |
1503 | i_size_write(inode, pos + copied); | |
1504 | ||
ec9516fb HD |
1505 | if (!PageUptodate(page)) { |
1506 | if (copied < PAGE_CACHE_SIZE) { | |
1507 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
1508 | zero_user_segments(page, 0, from, | |
1509 | from + copied, PAGE_CACHE_SIZE); | |
1510 | } | |
1511 | SetPageUptodate(page); | |
1512 | } | |
800d15a5 | 1513 | set_page_dirty(page); |
6746aff7 | 1514 | unlock_page(page); |
800d15a5 NP |
1515 | page_cache_release(page); |
1516 | ||
800d15a5 | 1517 | return copied; |
1da177e4 LT |
1518 | } |
1519 | ||
2ba5bbed | 1520 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 1521 | { |
6e58e79d AV |
1522 | struct file *file = iocb->ki_filp; |
1523 | struct inode *inode = file_inode(file); | |
1da177e4 | 1524 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
1525 | pgoff_t index; |
1526 | unsigned long offset; | |
a0ee5ec5 | 1527 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 1528 | int error = 0; |
cb66a7a1 | 1529 | ssize_t retval = 0; |
6e58e79d | 1530 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
1531 | |
1532 | /* | |
1533 | * Might this read be for a stacking filesystem? Then when reading | |
1534 | * holes of a sparse file, we actually need to allocate those pages, | |
1535 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
1536 | */ | |
1537 | if (segment_eq(get_fs(), KERNEL_DS)) | |
1538 | sgp = SGP_DIRTY; | |
1da177e4 LT |
1539 | |
1540 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1541 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1542 | ||
1543 | for (;;) { | |
1544 | struct page *page = NULL; | |
41ffe5d5 HD |
1545 | pgoff_t end_index; |
1546 | unsigned long nr, ret; | |
1da177e4 LT |
1547 | loff_t i_size = i_size_read(inode); |
1548 | ||
1549 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1550 | if (index > end_index) | |
1551 | break; | |
1552 | if (index == end_index) { | |
1553 | nr = i_size & ~PAGE_CACHE_MASK; | |
1554 | if (nr <= offset) | |
1555 | break; | |
1556 | } | |
1557 | ||
6e58e79d AV |
1558 | error = shmem_getpage(inode, index, &page, sgp, NULL); |
1559 | if (error) { | |
1560 | if (error == -EINVAL) | |
1561 | error = 0; | |
1da177e4 LT |
1562 | break; |
1563 | } | |
d3602444 HD |
1564 | if (page) |
1565 | unlock_page(page); | |
1da177e4 LT |
1566 | |
1567 | /* | |
1568 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 1569 | * are called without i_mutex protection against truncate |
1da177e4 LT |
1570 | */ |
1571 | nr = PAGE_CACHE_SIZE; | |
1572 | i_size = i_size_read(inode); | |
1573 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1574 | if (index == end_index) { | |
1575 | nr = i_size & ~PAGE_CACHE_MASK; | |
1576 | if (nr <= offset) { | |
1577 | if (page) | |
1578 | page_cache_release(page); | |
1579 | break; | |
1580 | } | |
1581 | } | |
1582 | nr -= offset; | |
1583 | ||
1584 | if (page) { | |
1585 | /* | |
1586 | * If users can be writing to this page using arbitrary | |
1587 | * virtual addresses, take care about potential aliasing | |
1588 | * before reading the page on the kernel side. | |
1589 | */ | |
1590 | if (mapping_writably_mapped(mapping)) | |
1591 | flush_dcache_page(page); | |
1592 | /* | |
1593 | * Mark the page accessed if we read the beginning. | |
1594 | */ | |
1595 | if (!offset) | |
1596 | mark_page_accessed(page); | |
b5810039 | 1597 | } else { |
1da177e4 | 1598 | page = ZERO_PAGE(0); |
b5810039 NP |
1599 | page_cache_get(page); |
1600 | } | |
1da177e4 LT |
1601 | |
1602 | /* | |
1603 | * Ok, we have the page, and it's up-to-date, so | |
1604 | * now we can copy it to user space... | |
1da177e4 | 1605 | */ |
2ba5bbed | 1606 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 1607 | retval += ret; |
1da177e4 LT |
1608 | offset += ret; |
1609 | index += offset >> PAGE_CACHE_SHIFT; | |
1610 | offset &= ~PAGE_CACHE_MASK; | |
1611 | ||
1612 | page_cache_release(page); | |
2ba5bbed | 1613 | if (!iov_iter_count(to)) |
1da177e4 | 1614 | break; |
6e58e79d AV |
1615 | if (ret < nr) { |
1616 | error = -EFAULT; | |
1617 | break; | |
1618 | } | |
1da177e4 LT |
1619 | cond_resched(); |
1620 | } | |
1621 | ||
1622 | *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; | |
6e58e79d AV |
1623 | file_accessed(file); |
1624 | return retval ? retval : error; | |
1da177e4 LT |
1625 | } |
1626 | ||
708e3508 HD |
1627 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, |
1628 | struct pipe_inode_info *pipe, size_t len, | |
1629 | unsigned int flags) | |
1630 | { | |
1631 | struct address_space *mapping = in->f_mapping; | |
71f0e07a | 1632 | struct inode *inode = mapping->host; |
708e3508 HD |
1633 | unsigned int loff, nr_pages, req_pages; |
1634 | struct page *pages[PIPE_DEF_BUFFERS]; | |
1635 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
1636 | struct page *page; | |
1637 | pgoff_t index, end_index; | |
1638 | loff_t isize, left; | |
1639 | int error, page_nr; | |
1640 | struct splice_pipe_desc spd = { | |
1641 | .pages = pages, | |
1642 | .partial = partial, | |
047fe360 | 1643 | .nr_pages_max = PIPE_DEF_BUFFERS, |
708e3508 HD |
1644 | .flags = flags, |
1645 | .ops = &page_cache_pipe_buf_ops, | |
1646 | .spd_release = spd_release_page, | |
1647 | }; | |
1648 | ||
71f0e07a | 1649 | isize = i_size_read(inode); |
708e3508 HD |
1650 | if (unlikely(*ppos >= isize)) |
1651 | return 0; | |
1652 | ||
1653 | left = isize - *ppos; | |
1654 | if (unlikely(left < len)) | |
1655 | len = left; | |
1656 | ||
1657 | if (splice_grow_spd(pipe, &spd)) | |
1658 | return -ENOMEM; | |
1659 | ||
1660 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1661 | loff = *ppos & ~PAGE_CACHE_MASK; | |
1662 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
a786c06d | 1663 | nr_pages = min(req_pages, spd.nr_pages_max); |
708e3508 | 1664 | |
708e3508 HD |
1665 | spd.nr_pages = find_get_pages_contig(mapping, index, |
1666 | nr_pages, spd.pages); | |
1667 | index += spd.nr_pages; | |
708e3508 | 1668 | error = 0; |
708e3508 | 1669 | |
71f0e07a | 1670 | while (spd.nr_pages < nr_pages) { |
71f0e07a HD |
1671 | error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL); |
1672 | if (error) | |
1673 | break; | |
1674 | unlock_page(page); | |
708e3508 HD |
1675 | spd.pages[spd.nr_pages++] = page; |
1676 | index++; | |
1677 | } | |
1678 | ||
708e3508 HD |
1679 | index = *ppos >> PAGE_CACHE_SHIFT; |
1680 | nr_pages = spd.nr_pages; | |
1681 | spd.nr_pages = 0; | |
71f0e07a | 1682 | |
708e3508 HD |
1683 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { |
1684 | unsigned int this_len; | |
1685 | ||
1686 | if (!len) | |
1687 | break; | |
1688 | ||
708e3508 HD |
1689 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); |
1690 | page = spd.pages[page_nr]; | |
1691 | ||
71f0e07a | 1692 | if (!PageUptodate(page) || page->mapping != mapping) { |
71f0e07a HD |
1693 | error = shmem_getpage(inode, index, &page, |
1694 | SGP_CACHE, NULL); | |
1695 | if (error) | |
708e3508 | 1696 | break; |
71f0e07a HD |
1697 | unlock_page(page); |
1698 | page_cache_release(spd.pages[page_nr]); | |
1699 | spd.pages[page_nr] = page; | |
708e3508 | 1700 | } |
71f0e07a HD |
1701 | |
1702 | isize = i_size_read(inode); | |
708e3508 HD |
1703 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; |
1704 | if (unlikely(!isize || index > end_index)) | |
1705 | break; | |
1706 | ||
708e3508 HD |
1707 | if (end_index == index) { |
1708 | unsigned int plen; | |
1709 | ||
708e3508 HD |
1710 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
1711 | if (plen <= loff) | |
1712 | break; | |
1713 | ||
708e3508 HD |
1714 | this_len = min(this_len, plen - loff); |
1715 | len = this_len; | |
1716 | } | |
1717 | ||
1718 | spd.partial[page_nr].offset = loff; | |
1719 | spd.partial[page_nr].len = this_len; | |
1720 | len -= this_len; | |
1721 | loff = 0; | |
1722 | spd.nr_pages++; | |
1723 | index++; | |
1724 | } | |
1725 | ||
708e3508 HD |
1726 | while (page_nr < nr_pages) |
1727 | page_cache_release(spd.pages[page_nr++]); | |
708e3508 HD |
1728 | |
1729 | if (spd.nr_pages) | |
1730 | error = splice_to_pipe(pipe, &spd); | |
1731 | ||
047fe360 | 1732 | splice_shrink_spd(&spd); |
708e3508 HD |
1733 | |
1734 | if (error > 0) { | |
1735 | *ppos += error; | |
1736 | file_accessed(in); | |
1737 | } | |
1738 | return error; | |
1739 | } | |
1740 | ||
220f2ac9 HD |
1741 | /* |
1742 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
1743 | */ | |
1744 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 1745 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
1746 | { |
1747 | struct page *page; | |
1748 | struct pagevec pvec; | |
1749 | pgoff_t indices[PAGEVEC_SIZE]; | |
1750 | bool done = false; | |
1751 | int i; | |
1752 | ||
1753 | pagevec_init(&pvec, 0); | |
1754 | pvec.nr = 1; /* start small: we may be there already */ | |
1755 | while (!done) { | |
0cd6144a | 1756 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
1757 | pvec.nr, pvec.pages, indices); |
1758 | if (!pvec.nr) { | |
965c8e59 | 1759 | if (whence == SEEK_DATA) |
220f2ac9 HD |
1760 | index = end; |
1761 | break; | |
1762 | } | |
1763 | for (i = 0; i < pvec.nr; i++, index++) { | |
1764 | if (index < indices[i]) { | |
965c8e59 | 1765 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
1766 | done = true; |
1767 | break; | |
1768 | } | |
1769 | index = indices[i]; | |
1770 | } | |
1771 | page = pvec.pages[i]; | |
1772 | if (page && !radix_tree_exceptional_entry(page)) { | |
1773 | if (!PageUptodate(page)) | |
1774 | page = NULL; | |
1775 | } | |
1776 | if (index >= end || | |
965c8e59 AM |
1777 | (page && whence == SEEK_DATA) || |
1778 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
1779 | done = true; |
1780 | break; | |
1781 | } | |
1782 | } | |
0cd6144a | 1783 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
1784 | pagevec_release(&pvec); |
1785 | pvec.nr = PAGEVEC_SIZE; | |
1786 | cond_resched(); | |
1787 | } | |
1788 | return index; | |
1789 | } | |
1790 | ||
965c8e59 | 1791 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
1792 | { |
1793 | struct address_space *mapping = file->f_mapping; | |
1794 | struct inode *inode = mapping->host; | |
1795 | pgoff_t start, end; | |
1796 | loff_t new_offset; | |
1797 | ||
965c8e59 AM |
1798 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
1799 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 HD |
1800 | MAX_LFS_FILESIZE, i_size_read(inode)); |
1801 | mutex_lock(&inode->i_mutex); | |
1802 | /* We're holding i_mutex so we can access i_size directly */ | |
1803 | ||
1804 | if (offset < 0) | |
1805 | offset = -EINVAL; | |
1806 | else if (offset >= inode->i_size) | |
1807 | offset = -ENXIO; | |
1808 | else { | |
1809 | start = offset >> PAGE_CACHE_SHIFT; | |
1810 | end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
965c8e59 | 1811 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
220f2ac9 HD |
1812 | new_offset <<= PAGE_CACHE_SHIFT; |
1813 | if (new_offset > offset) { | |
1814 | if (new_offset < inode->i_size) | |
1815 | offset = new_offset; | |
965c8e59 | 1816 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
1817 | offset = -ENXIO; |
1818 | else | |
1819 | offset = inode->i_size; | |
1820 | } | |
1821 | } | |
1822 | ||
387aae6f HD |
1823 | if (offset >= 0) |
1824 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
220f2ac9 HD |
1825 | mutex_unlock(&inode->i_mutex); |
1826 | return offset; | |
1827 | } | |
1828 | ||
40e041a2 DH |
1829 | static int shmem_wait_for_pins(struct address_space *mapping) |
1830 | { | |
1831 | return 0; | |
1832 | } | |
1833 | ||
1834 | #define F_ALL_SEALS (F_SEAL_SEAL | \ | |
1835 | F_SEAL_SHRINK | \ | |
1836 | F_SEAL_GROW | \ | |
1837 | F_SEAL_WRITE) | |
1838 | ||
1839 | int shmem_add_seals(struct file *file, unsigned int seals) | |
1840 | { | |
1841 | struct inode *inode = file_inode(file); | |
1842 | struct shmem_inode_info *info = SHMEM_I(inode); | |
1843 | int error; | |
1844 | ||
1845 | /* | |
1846 | * SEALING | |
1847 | * Sealing allows multiple parties to share a shmem-file but restrict | |
1848 | * access to a specific subset of file operations. Seals can only be | |
1849 | * added, but never removed. This way, mutually untrusted parties can | |
1850 | * share common memory regions with a well-defined policy. A malicious | |
1851 | * peer can thus never perform unwanted operations on a shared object. | |
1852 | * | |
1853 | * Seals are only supported on special shmem-files and always affect | |
1854 | * the whole underlying inode. Once a seal is set, it may prevent some | |
1855 | * kinds of access to the file. Currently, the following seals are | |
1856 | * defined: | |
1857 | * SEAL_SEAL: Prevent further seals from being set on this file | |
1858 | * SEAL_SHRINK: Prevent the file from shrinking | |
1859 | * SEAL_GROW: Prevent the file from growing | |
1860 | * SEAL_WRITE: Prevent write access to the file | |
1861 | * | |
1862 | * As we don't require any trust relationship between two parties, we | |
1863 | * must prevent seals from being removed. Therefore, sealing a file | |
1864 | * only adds a given set of seals to the file, it never touches | |
1865 | * existing seals. Furthermore, the "setting seals"-operation can be | |
1866 | * sealed itself, which basically prevents any further seal from being | |
1867 | * added. | |
1868 | * | |
1869 | * Semantics of sealing are only defined on volatile files. Only | |
1870 | * anonymous shmem files support sealing. More importantly, seals are | |
1871 | * never written to disk. Therefore, there's no plan to support it on | |
1872 | * other file types. | |
1873 | */ | |
1874 | ||
1875 | if (file->f_op != &shmem_file_operations) | |
1876 | return -EINVAL; | |
1877 | if (!(file->f_mode & FMODE_WRITE)) | |
1878 | return -EPERM; | |
1879 | if (seals & ~(unsigned int)F_ALL_SEALS) | |
1880 | return -EINVAL; | |
1881 | ||
1882 | mutex_lock(&inode->i_mutex); | |
1883 | ||
1884 | if (info->seals & F_SEAL_SEAL) { | |
1885 | error = -EPERM; | |
1886 | goto unlock; | |
1887 | } | |
1888 | ||
1889 | if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) { | |
1890 | error = mapping_deny_writable(file->f_mapping); | |
1891 | if (error) | |
1892 | goto unlock; | |
1893 | ||
1894 | error = shmem_wait_for_pins(file->f_mapping); | |
1895 | if (error) { | |
1896 | mapping_allow_writable(file->f_mapping); | |
1897 | goto unlock; | |
1898 | } | |
1899 | } | |
1900 | ||
1901 | info->seals |= seals; | |
1902 | error = 0; | |
1903 | ||
1904 | unlock: | |
1905 | mutex_unlock(&inode->i_mutex); | |
1906 | return error; | |
1907 | } | |
1908 | EXPORT_SYMBOL_GPL(shmem_add_seals); | |
1909 | ||
1910 | int shmem_get_seals(struct file *file) | |
1911 | { | |
1912 | if (file->f_op != &shmem_file_operations) | |
1913 | return -EINVAL; | |
1914 | ||
1915 | return SHMEM_I(file_inode(file))->seals; | |
1916 | } | |
1917 | EXPORT_SYMBOL_GPL(shmem_get_seals); | |
1918 | ||
1919 | long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg) | |
1920 | { | |
1921 | long error; | |
1922 | ||
1923 | switch (cmd) { | |
1924 | case F_ADD_SEALS: | |
1925 | /* disallow upper 32bit */ | |
1926 | if (arg > UINT_MAX) | |
1927 | return -EINVAL; | |
1928 | ||
1929 | error = shmem_add_seals(file, arg); | |
1930 | break; | |
1931 | case F_GET_SEALS: | |
1932 | error = shmem_get_seals(file); | |
1933 | break; | |
1934 | default: | |
1935 | error = -EINVAL; | |
1936 | break; | |
1937 | } | |
1938 | ||
1939 | return error; | |
1940 | } | |
1941 | ||
83e4fa9c HD |
1942 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
1943 | loff_t len) | |
1944 | { | |
496ad9aa | 1945 | struct inode *inode = file_inode(file); |
e2d12e22 | 1946 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 1947 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 1948 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
1949 | pgoff_t start, index, end; |
1950 | int error; | |
83e4fa9c | 1951 | |
13ace4d0 HD |
1952 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
1953 | return -EOPNOTSUPP; | |
1954 | ||
83e4fa9c HD |
1955 | mutex_lock(&inode->i_mutex); |
1956 | ||
1957 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1958 | struct address_space *mapping = file->f_mapping; | |
1959 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
1960 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 1961 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 1962 | |
40e041a2 DH |
1963 | /* protected by i_mutex */ |
1964 | if (info->seals & F_SEAL_WRITE) { | |
1965 | error = -EPERM; | |
1966 | goto out; | |
1967 | } | |
1968 | ||
8e205f77 | 1969 | shmem_falloc.waitq = &shmem_falloc_waitq; |
f00cdc6d HD |
1970 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
1971 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
1972 | spin_lock(&inode->i_lock); | |
1973 | inode->i_private = &shmem_falloc; | |
1974 | spin_unlock(&inode->i_lock); | |
1975 | ||
83e4fa9c HD |
1976 | if ((u64)unmap_end > (u64)unmap_start) |
1977 | unmap_mapping_range(mapping, unmap_start, | |
1978 | 1 + unmap_end - unmap_start, 0); | |
1979 | shmem_truncate_range(inode, offset, offset + len - 1); | |
1980 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
1981 | |
1982 | spin_lock(&inode->i_lock); | |
1983 | inode->i_private = NULL; | |
1984 | wake_up_all(&shmem_falloc_waitq); | |
1985 | spin_unlock(&inode->i_lock); | |
83e4fa9c | 1986 | error = 0; |
8e205f77 | 1987 | goto out; |
e2d12e22 HD |
1988 | } |
1989 | ||
1990 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
1991 | error = inode_newsize_ok(inode, offset + len); | |
1992 | if (error) | |
1993 | goto out; | |
1994 | ||
40e041a2 DH |
1995 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
1996 | error = -EPERM; | |
1997 | goto out; | |
1998 | } | |
1999 | ||
e2d12e22 HD |
2000 | start = offset >> PAGE_CACHE_SHIFT; |
2001 | end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
2002 | /* Try to avoid a swapstorm if len is impossible to satisfy */ | |
2003 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2004 | error = -ENOSPC; | |
2005 | goto out; | |
83e4fa9c HD |
2006 | } |
2007 | ||
8e205f77 | 2008 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2009 | shmem_falloc.start = start; |
2010 | shmem_falloc.next = start; | |
2011 | shmem_falloc.nr_falloced = 0; | |
2012 | shmem_falloc.nr_unswapped = 0; | |
2013 | spin_lock(&inode->i_lock); | |
2014 | inode->i_private = &shmem_falloc; | |
2015 | spin_unlock(&inode->i_lock); | |
2016 | ||
e2d12e22 HD |
2017 | for (index = start; index < end; index++) { |
2018 | struct page *page; | |
2019 | ||
2020 | /* | |
2021 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2022 | * been interrupted because we are using up too much memory. | |
2023 | */ | |
2024 | if (signal_pending(current)) | |
2025 | error = -EINTR; | |
1aac1400 HD |
2026 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2027 | error = -ENOMEM; | |
e2d12e22 | 2028 | else |
1635f6a7 | 2029 | error = shmem_getpage(inode, index, &page, SGP_FALLOC, |
e2d12e22 HD |
2030 | NULL); |
2031 | if (error) { | |
1635f6a7 HD |
2032 | /* Remove the !PageUptodate pages we added */ |
2033 | shmem_undo_range(inode, | |
2034 | (loff_t)start << PAGE_CACHE_SHIFT, | |
2035 | (loff_t)index << PAGE_CACHE_SHIFT, true); | |
1aac1400 | 2036 | goto undone; |
e2d12e22 HD |
2037 | } |
2038 | ||
1aac1400 HD |
2039 | /* |
2040 | * Inform shmem_writepage() how far we have reached. | |
2041 | * No need for lock or barrier: we have the page lock. | |
2042 | */ | |
2043 | shmem_falloc.next++; | |
2044 | if (!PageUptodate(page)) | |
2045 | shmem_falloc.nr_falloced++; | |
2046 | ||
e2d12e22 | 2047 | /* |
1635f6a7 HD |
2048 | * If !PageUptodate, leave it that way so that freeable pages |
2049 | * can be recognized if we need to rollback on error later. | |
2050 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2051 | * than free the pages we are allocating (and SGP_CACHE pages |
2052 | * might still be clean: we now need to mark those dirty too). | |
2053 | */ | |
2054 | set_page_dirty(page); | |
2055 | unlock_page(page); | |
2056 | page_cache_release(page); | |
2057 | cond_resched(); | |
2058 | } | |
2059 | ||
2060 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2061 | i_size_write(inode, offset + len); | |
e2d12e22 | 2062 | inode->i_ctime = CURRENT_TIME; |
1aac1400 HD |
2063 | undone: |
2064 | spin_lock(&inode->i_lock); | |
2065 | inode->i_private = NULL; | |
2066 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2067 | out: |
83e4fa9c HD |
2068 | mutex_unlock(&inode->i_mutex); |
2069 | return error; | |
2070 | } | |
2071 | ||
726c3342 | 2072 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2073 | { |
726c3342 | 2074 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2075 | |
2076 | buf->f_type = TMPFS_MAGIC; | |
2077 | buf->f_bsize = PAGE_CACHE_SIZE; | |
2078 | buf->f_namelen = NAME_MAX; | |
0edd73b3 | 2079 | if (sbinfo->max_blocks) { |
1da177e4 | 2080 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2081 | buf->f_bavail = |
2082 | buf->f_bfree = sbinfo->max_blocks - | |
2083 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2084 | } |
2085 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2086 | buf->f_files = sbinfo->max_inodes; |
2087 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2088 | } |
2089 | /* else leave those fields 0 like simple_statfs */ | |
2090 | return 0; | |
2091 | } | |
2092 | ||
2093 | /* | |
2094 | * File creation. Allocate an inode, and we're done.. | |
2095 | */ | |
2096 | static int | |
1a67aafb | 2097 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2098 | { |
0b0a0806 | 2099 | struct inode *inode; |
1da177e4 LT |
2100 | int error = -ENOSPC; |
2101 | ||
454abafe | 2102 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2103 | if (inode) { |
feda821e CH |
2104 | error = simple_acl_create(dir, inode); |
2105 | if (error) | |
2106 | goto out_iput; | |
2a7dba39 | 2107 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2108 | &dentry->d_name, |
6d9d88d0 | 2109 | shmem_initxattrs, NULL); |
feda821e CH |
2110 | if (error && error != -EOPNOTSUPP) |
2111 | goto out_iput; | |
37ec43cd | 2112 | |
718deb6b | 2113 | error = 0; |
1da177e4 LT |
2114 | dir->i_size += BOGO_DIRENT_SIZE; |
2115 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2116 | d_instantiate(dentry, inode); | |
2117 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2118 | } |
2119 | return error; | |
feda821e CH |
2120 | out_iput: |
2121 | iput(inode); | |
2122 | return error; | |
1da177e4 LT |
2123 | } |
2124 | ||
60545d0d AV |
2125 | static int |
2126 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2127 | { | |
2128 | struct inode *inode; | |
2129 | int error = -ENOSPC; | |
2130 | ||
2131 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
2132 | if (inode) { | |
2133 | error = security_inode_init_security(inode, dir, | |
2134 | NULL, | |
2135 | shmem_initxattrs, NULL); | |
feda821e CH |
2136 | if (error && error != -EOPNOTSUPP) |
2137 | goto out_iput; | |
2138 | error = simple_acl_create(dir, inode); | |
2139 | if (error) | |
2140 | goto out_iput; | |
60545d0d AV |
2141 | d_tmpfile(dentry, inode); |
2142 | } | |
2143 | return error; | |
feda821e CH |
2144 | out_iput: |
2145 | iput(inode); | |
2146 | return error; | |
60545d0d AV |
2147 | } |
2148 | ||
18bb1db3 | 2149 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
2150 | { |
2151 | int error; | |
2152 | ||
2153 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
2154 | return error; | |
d8c76e6f | 2155 | inc_nlink(dir); |
1da177e4 LT |
2156 | return 0; |
2157 | } | |
2158 | ||
4acdaf27 | 2159 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2160 | bool excl) |
1da177e4 LT |
2161 | { |
2162 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
2163 | } | |
2164 | ||
2165 | /* | |
2166 | * Link a file.. | |
2167 | */ | |
2168 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
2169 | { | |
2170 | struct inode *inode = old_dentry->d_inode; | |
5b04c689 | 2171 | int ret; |
1da177e4 LT |
2172 | |
2173 | /* | |
2174 | * No ordinary (disk based) filesystem counts links as inodes; | |
2175 | * but each new link needs a new dentry, pinning lowmem, and | |
2176 | * tmpfs dentries cannot be pruned until they are unlinked. | |
2177 | */ | |
5b04c689 PE |
2178 | ret = shmem_reserve_inode(inode->i_sb); |
2179 | if (ret) | |
2180 | goto out; | |
1da177e4 LT |
2181 | |
2182 | dir->i_size += BOGO_DIRENT_SIZE; | |
2183 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 2184 | inc_nlink(inode); |
7de9c6ee | 2185 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
2186 | dget(dentry); /* Extra pinning count for the created dentry */ |
2187 | d_instantiate(dentry, inode); | |
5b04c689 PE |
2188 | out: |
2189 | return ret; | |
1da177e4 LT |
2190 | } |
2191 | ||
2192 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
2193 | { | |
2194 | struct inode *inode = dentry->d_inode; | |
2195 | ||
5b04c689 PE |
2196 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
2197 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
2198 | |
2199 | dir->i_size -= BOGO_DIRENT_SIZE; | |
2200 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 2201 | drop_nlink(inode); |
1da177e4 LT |
2202 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
2203 | return 0; | |
2204 | } | |
2205 | ||
2206 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
2207 | { | |
2208 | if (!simple_empty(dentry)) | |
2209 | return -ENOTEMPTY; | |
2210 | ||
9a53c3a7 DH |
2211 | drop_nlink(dentry->d_inode); |
2212 | drop_nlink(dir); | |
1da177e4 LT |
2213 | return shmem_unlink(dir, dentry); |
2214 | } | |
2215 | ||
2216 | /* | |
2217 | * The VFS layer already does all the dentry stuff for rename, | |
2218 | * we just have to decrement the usage count for the target if | |
2219 | * it exists so that the VFS layer correctly free's it when it | |
2220 | * gets overwritten. | |
2221 | */ | |
2222 | static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) | |
2223 | { | |
2224 | struct inode *inode = old_dentry->d_inode; | |
2225 | int they_are_dirs = S_ISDIR(inode->i_mode); | |
2226 | ||
2227 | if (!simple_empty(new_dentry)) | |
2228 | return -ENOTEMPTY; | |
2229 | ||
2230 | if (new_dentry->d_inode) { | |
2231 | (void) shmem_unlink(new_dir, new_dentry); | |
2232 | if (they_are_dirs) | |
9a53c3a7 | 2233 | drop_nlink(old_dir); |
1da177e4 | 2234 | } else if (they_are_dirs) { |
9a53c3a7 | 2235 | drop_nlink(old_dir); |
d8c76e6f | 2236 | inc_nlink(new_dir); |
1da177e4 LT |
2237 | } |
2238 | ||
2239 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
2240 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
2241 | old_dir->i_ctime = old_dir->i_mtime = | |
2242 | new_dir->i_ctime = new_dir->i_mtime = | |
2243 | inode->i_ctime = CURRENT_TIME; | |
2244 | return 0; | |
2245 | } | |
2246 | ||
2247 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
2248 | { | |
2249 | int error; | |
2250 | int len; | |
2251 | struct inode *inode; | |
9276aad6 | 2252 | struct page *page; |
1da177e4 LT |
2253 | char *kaddr; |
2254 | struct shmem_inode_info *info; | |
2255 | ||
2256 | len = strlen(symname) + 1; | |
2257 | if (len > PAGE_CACHE_SIZE) | |
2258 | return -ENAMETOOLONG; | |
2259 | ||
454abafe | 2260 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
2261 | if (!inode) |
2262 | return -ENOSPC; | |
2263 | ||
9d8f13ba | 2264 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 2265 | shmem_initxattrs, NULL); |
570bc1c2 SS |
2266 | if (error) { |
2267 | if (error != -EOPNOTSUPP) { | |
2268 | iput(inode); | |
2269 | return error; | |
2270 | } | |
2271 | error = 0; | |
2272 | } | |
2273 | ||
1da177e4 LT |
2274 | info = SHMEM_I(inode); |
2275 | inode->i_size = len-1; | |
69f07ec9 HD |
2276 | if (len <= SHORT_SYMLINK_LEN) { |
2277 | info->symlink = kmemdup(symname, len, GFP_KERNEL); | |
2278 | if (!info->symlink) { | |
2279 | iput(inode); | |
2280 | return -ENOMEM; | |
2281 | } | |
2282 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 LT |
2283 | } else { |
2284 | error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); | |
2285 | if (error) { | |
2286 | iput(inode); | |
2287 | return error; | |
2288 | } | |
14fcc23f | 2289 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 2290 | inode->i_op = &shmem_symlink_inode_operations; |
9b04c5fe | 2291 | kaddr = kmap_atomic(page); |
1da177e4 | 2292 | memcpy(kaddr, symname, len); |
9b04c5fe | 2293 | kunmap_atomic(kaddr); |
ec9516fb | 2294 | SetPageUptodate(page); |
1da177e4 | 2295 | set_page_dirty(page); |
6746aff7 | 2296 | unlock_page(page); |
1da177e4 LT |
2297 | page_cache_release(page); |
2298 | } | |
1da177e4 LT |
2299 | dir->i_size += BOGO_DIRENT_SIZE; |
2300 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2301 | d_instantiate(dentry, inode); | |
2302 | dget(dentry); | |
2303 | return 0; | |
2304 | } | |
2305 | ||
69f07ec9 | 2306 | static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 2307 | { |
69f07ec9 | 2308 | nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink); |
cc314eef | 2309 | return NULL; |
1da177e4 LT |
2310 | } |
2311 | ||
cc314eef | 2312 | static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) |
1da177e4 LT |
2313 | { |
2314 | struct page *page = NULL; | |
41ffe5d5 HD |
2315 | int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); |
2316 | nd_set_link(nd, error ? ERR_PTR(error) : kmap(page)); | |
d3602444 HD |
2317 | if (page) |
2318 | unlock_page(page); | |
cc314eef | 2319 | return page; |
1da177e4 LT |
2320 | } |
2321 | ||
cc314eef | 2322 | static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 LT |
2323 | { |
2324 | if (!IS_ERR(nd_get_link(nd))) { | |
cc314eef | 2325 | struct page *page = cookie; |
1da177e4 LT |
2326 | kunmap(page); |
2327 | mark_page_accessed(page); | |
2328 | page_cache_release(page); | |
1da177e4 LT |
2329 | } |
2330 | } | |
2331 | ||
b09e0fa4 | 2332 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 2333 | /* |
b09e0fa4 EP |
2334 | * Superblocks without xattr inode operations may get some security.* xattr |
2335 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
2336 | * like ACLs, we also need to implement the security.* handlers at |
2337 | * filesystem level, though. | |
2338 | */ | |
2339 | ||
6d9d88d0 JS |
2340 | /* |
2341 | * Callback for security_inode_init_security() for acquiring xattrs. | |
2342 | */ | |
2343 | static int shmem_initxattrs(struct inode *inode, | |
2344 | const struct xattr *xattr_array, | |
2345 | void *fs_info) | |
2346 | { | |
2347 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2348 | const struct xattr *xattr; | |
38f38657 | 2349 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
2350 | size_t len; |
2351 | ||
2352 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 2353 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
2354 | if (!new_xattr) |
2355 | return -ENOMEM; | |
2356 | ||
2357 | len = strlen(xattr->name) + 1; | |
2358 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
2359 | GFP_KERNEL); | |
2360 | if (!new_xattr->name) { | |
2361 | kfree(new_xattr); | |
2362 | return -ENOMEM; | |
2363 | } | |
2364 | ||
2365 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
2366 | XATTR_SECURITY_PREFIX_LEN); | |
2367 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
2368 | xattr->name, len); | |
2369 | ||
38f38657 | 2370 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
2371 | } |
2372 | ||
2373 | return 0; | |
2374 | } | |
2375 | ||
bb435453 | 2376 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
b09e0fa4 | 2377 | #ifdef CONFIG_TMPFS_POSIX_ACL |
feda821e CH |
2378 | &posix_acl_access_xattr_handler, |
2379 | &posix_acl_default_xattr_handler, | |
b09e0fa4 | 2380 | #endif |
39f0247d AG |
2381 | NULL |
2382 | }; | |
b09e0fa4 EP |
2383 | |
2384 | static int shmem_xattr_validate(const char *name) | |
2385 | { | |
2386 | struct { const char *prefix; size_t len; } arr[] = { | |
2387 | { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN }, | |
2388 | { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN } | |
2389 | }; | |
2390 | int i; | |
2391 | ||
2392 | for (i = 0; i < ARRAY_SIZE(arr); i++) { | |
2393 | size_t preflen = arr[i].len; | |
2394 | if (strncmp(name, arr[i].prefix, preflen) == 0) { | |
2395 | if (!name[preflen]) | |
2396 | return -EINVAL; | |
2397 | return 0; | |
2398 | } | |
2399 | } | |
2400 | return -EOPNOTSUPP; | |
2401 | } | |
2402 | ||
2403 | static ssize_t shmem_getxattr(struct dentry *dentry, const char *name, | |
2404 | void *buffer, size_t size) | |
2405 | { | |
38f38657 | 2406 | struct shmem_inode_info *info = SHMEM_I(dentry->d_inode); |
b09e0fa4 EP |
2407 | int err; |
2408 | ||
2409 | /* | |
2410 | * If this is a request for a synthetic attribute in the system.* | |
2411 | * namespace use the generic infrastructure to resolve a handler | |
2412 | * for it via sb->s_xattr. | |
2413 | */ | |
2414 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2415 | return generic_getxattr(dentry, name, buffer, size); | |
2416 | ||
2417 | err = shmem_xattr_validate(name); | |
2418 | if (err) | |
2419 | return err; | |
2420 | ||
38f38657 | 2421 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
2422 | } |
2423 | ||
2424 | static int shmem_setxattr(struct dentry *dentry, const char *name, | |
2425 | const void *value, size_t size, int flags) | |
2426 | { | |
38f38657 | 2427 | struct shmem_inode_info *info = SHMEM_I(dentry->d_inode); |
b09e0fa4 EP |
2428 | int err; |
2429 | ||
2430 | /* | |
2431 | * If this is a request for a synthetic attribute in the system.* | |
2432 | * namespace use the generic infrastructure to resolve a handler | |
2433 | * for it via sb->s_xattr. | |
2434 | */ | |
2435 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2436 | return generic_setxattr(dentry, name, value, size, flags); | |
2437 | ||
2438 | err = shmem_xattr_validate(name); | |
2439 | if (err) | |
2440 | return err; | |
2441 | ||
38f38657 | 2442 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
2443 | } |
2444 | ||
2445 | static int shmem_removexattr(struct dentry *dentry, const char *name) | |
2446 | { | |
38f38657 | 2447 | struct shmem_inode_info *info = SHMEM_I(dentry->d_inode); |
b09e0fa4 EP |
2448 | int err; |
2449 | ||
2450 | /* | |
2451 | * If this is a request for a synthetic attribute in the system.* | |
2452 | * namespace use the generic infrastructure to resolve a handler | |
2453 | * for it via sb->s_xattr. | |
2454 | */ | |
2455 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2456 | return generic_removexattr(dentry, name); | |
2457 | ||
2458 | err = shmem_xattr_validate(name); | |
2459 | if (err) | |
2460 | return err; | |
2461 | ||
38f38657 | 2462 | return simple_xattr_remove(&info->xattrs, name); |
b09e0fa4 EP |
2463 | } |
2464 | ||
2465 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
2466 | { | |
38f38657 AR |
2467 | struct shmem_inode_info *info = SHMEM_I(dentry->d_inode); |
2468 | return simple_xattr_list(&info->xattrs, buffer, size); | |
b09e0fa4 EP |
2469 | } |
2470 | #endif /* CONFIG_TMPFS_XATTR */ | |
2471 | ||
69f07ec9 | 2472 | static const struct inode_operations shmem_short_symlink_operations = { |
b09e0fa4 | 2473 | .readlink = generic_readlink, |
69f07ec9 | 2474 | .follow_link = shmem_follow_short_symlink, |
b09e0fa4 EP |
2475 | #ifdef CONFIG_TMPFS_XATTR |
2476 | .setxattr = shmem_setxattr, | |
2477 | .getxattr = shmem_getxattr, | |
2478 | .listxattr = shmem_listxattr, | |
2479 | .removexattr = shmem_removexattr, | |
2480 | #endif | |
2481 | }; | |
2482 | ||
2483 | static const struct inode_operations shmem_symlink_inode_operations = { | |
2484 | .readlink = generic_readlink, | |
2485 | .follow_link = shmem_follow_link, | |
2486 | .put_link = shmem_put_link, | |
2487 | #ifdef CONFIG_TMPFS_XATTR | |
2488 | .setxattr = shmem_setxattr, | |
2489 | .getxattr = shmem_getxattr, | |
2490 | .listxattr = shmem_listxattr, | |
2491 | .removexattr = shmem_removexattr, | |
39f0247d | 2492 | #endif |
b09e0fa4 | 2493 | }; |
39f0247d | 2494 | |
91828a40 DG |
2495 | static struct dentry *shmem_get_parent(struct dentry *child) |
2496 | { | |
2497 | return ERR_PTR(-ESTALE); | |
2498 | } | |
2499 | ||
2500 | static int shmem_match(struct inode *ino, void *vfh) | |
2501 | { | |
2502 | __u32 *fh = vfh; | |
2503 | __u64 inum = fh[2]; | |
2504 | inum = (inum << 32) | fh[1]; | |
2505 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
2506 | } | |
2507 | ||
480b116c CH |
2508 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
2509 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 2510 | { |
91828a40 | 2511 | struct inode *inode; |
480b116c | 2512 | struct dentry *dentry = NULL; |
35c2a7f4 | 2513 | u64 inum; |
480b116c CH |
2514 | |
2515 | if (fh_len < 3) | |
2516 | return NULL; | |
91828a40 | 2517 | |
35c2a7f4 HD |
2518 | inum = fid->raw[2]; |
2519 | inum = (inum << 32) | fid->raw[1]; | |
2520 | ||
480b116c CH |
2521 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
2522 | shmem_match, fid->raw); | |
91828a40 | 2523 | if (inode) { |
480b116c | 2524 | dentry = d_find_alias(inode); |
91828a40 DG |
2525 | iput(inode); |
2526 | } | |
2527 | ||
480b116c | 2528 | return dentry; |
91828a40 DG |
2529 | } |
2530 | ||
b0b0382b AV |
2531 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
2532 | struct inode *parent) | |
91828a40 | 2533 | { |
5fe0c237 AK |
2534 | if (*len < 3) { |
2535 | *len = 3; | |
94e07a75 | 2536 | return FILEID_INVALID; |
5fe0c237 | 2537 | } |
91828a40 | 2538 | |
1d3382cb | 2539 | if (inode_unhashed(inode)) { |
91828a40 DG |
2540 | /* Unfortunately insert_inode_hash is not idempotent, |
2541 | * so as we hash inodes here rather than at creation | |
2542 | * time, we need a lock to ensure we only try | |
2543 | * to do it once | |
2544 | */ | |
2545 | static DEFINE_SPINLOCK(lock); | |
2546 | spin_lock(&lock); | |
1d3382cb | 2547 | if (inode_unhashed(inode)) |
91828a40 DG |
2548 | __insert_inode_hash(inode, |
2549 | inode->i_ino + inode->i_generation); | |
2550 | spin_unlock(&lock); | |
2551 | } | |
2552 | ||
2553 | fh[0] = inode->i_generation; | |
2554 | fh[1] = inode->i_ino; | |
2555 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
2556 | ||
2557 | *len = 3; | |
2558 | return 1; | |
2559 | } | |
2560 | ||
39655164 | 2561 | static const struct export_operations shmem_export_ops = { |
91828a40 | 2562 | .get_parent = shmem_get_parent, |
91828a40 | 2563 | .encode_fh = shmem_encode_fh, |
480b116c | 2564 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
2565 | }; |
2566 | ||
680d794b AM |
2567 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
2568 | bool remount) | |
1da177e4 LT |
2569 | { |
2570 | char *this_char, *value, *rest; | |
49cd0a5c | 2571 | struct mempolicy *mpol = NULL; |
8751e039 EB |
2572 | uid_t uid; |
2573 | gid_t gid; | |
1da177e4 | 2574 | |
b00dc3ad HD |
2575 | while (options != NULL) { |
2576 | this_char = options; | |
2577 | for (;;) { | |
2578 | /* | |
2579 | * NUL-terminate this option: unfortunately, | |
2580 | * mount options form a comma-separated list, | |
2581 | * but mpol's nodelist may also contain commas. | |
2582 | */ | |
2583 | options = strchr(options, ','); | |
2584 | if (options == NULL) | |
2585 | break; | |
2586 | options++; | |
2587 | if (!isdigit(*options)) { | |
2588 | options[-1] = '\0'; | |
2589 | break; | |
2590 | } | |
2591 | } | |
1da177e4 LT |
2592 | if (!*this_char) |
2593 | continue; | |
2594 | if ((value = strchr(this_char,'=')) != NULL) { | |
2595 | *value++ = 0; | |
2596 | } else { | |
2597 | printk(KERN_ERR | |
2598 | "tmpfs: No value for mount option '%s'\n", | |
2599 | this_char); | |
49cd0a5c | 2600 | goto error; |
1da177e4 LT |
2601 | } |
2602 | ||
2603 | if (!strcmp(this_char,"size")) { | |
2604 | unsigned long long size; | |
2605 | size = memparse(value,&rest); | |
2606 | if (*rest == '%') { | |
2607 | size <<= PAGE_SHIFT; | |
2608 | size *= totalram_pages; | |
2609 | do_div(size, 100); | |
2610 | rest++; | |
2611 | } | |
2612 | if (*rest) | |
2613 | goto bad_val; | |
680d794b AM |
2614 | sbinfo->max_blocks = |
2615 | DIV_ROUND_UP(size, PAGE_CACHE_SIZE); | |
1da177e4 | 2616 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 2617 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
2618 | if (*rest) |
2619 | goto bad_val; | |
2620 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 2621 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
2622 | if (*rest) |
2623 | goto bad_val; | |
2624 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 2625 | if (remount) |
1da177e4 | 2626 | continue; |
680d794b | 2627 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
2628 | if (*rest) |
2629 | goto bad_val; | |
2630 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 2631 | if (remount) |
1da177e4 | 2632 | continue; |
8751e039 | 2633 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2634 | if (*rest) |
2635 | goto bad_val; | |
8751e039 EB |
2636 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
2637 | if (!uid_valid(sbinfo->uid)) | |
2638 | goto bad_val; | |
1da177e4 | 2639 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 2640 | if (remount) |
1da177e4 | 2641 | continue; |
8751e039 | 2642 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2643 | if (*rest) |
2644 | goto bad_val; | |
8751e039 EB |
2645 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
2646 | if (!gid_valid(sbinfo->gid)) | |
2647 | goto bad_val; | |
7339ff83 | 2648 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
2649 | mpol_put(mpol); |
2650 | mpol = NULL; | |
2651 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 2652 | goto bad_val; |
1da177e4 LT |
2653 | } else { |
2654 | printk(KERN_ERR "tmpfs: Bad mount option %s\n", | |
2655 | this_char); | |
49cd0a5c | 2656 | goto error; |
1da177e4 LT |
2657 | } |
2658 | } | |
49cd0a5c | 2659 | sbinfo->mpol = mpol; |
1da177e4 LT |
2660 | return 0; |
2661 | ||
2662 | bad_val: | |
2663 | printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", | |
2664 | value, this_char); | |
49cd0a5c GT |
2665 | error: |
2666 | mpol_put(mpol); | |
1da177e4 LT |
2667 | return 1; |
2668 | ||
2669 | } | |
2670 | ||
2671 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
2672 | { | |
2673 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 2674 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
2675 | unsigned long inodes; |
2676 | int error = -EINVAL; | |
2677 | ||
5f00110f | 2678 | config.mpol = NULL; |
680d794b | 2679 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 2680 | return error; |
1da177e4 | 2681 | |
0edd73b3 | 2682 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 2683 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 2684 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 2685 | goto out; |
680d794b | 2686 | if (config.max_inodes < inodes) |
0edd73b3 HD |
2687 | goto out; |
2688 | /* | |
54af6042 | 2689 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
2690 | * but we must separately disallow unlimited->limited, because |
2691 | * in that case we have no record of how much is already in use. | |
2692 | */ | |
680d794b | 2693 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 2694 | goto out; |
680d794b | 2695 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
2696 | goto out; |
2697 | ||
2698 | error = 0; | |
680d794b | 2699 | sbinfo->max_blocks = config.max_blocks; |
680d794b AM |
2700 | sbinfo->max_inodes = config.max_inodes; |
2701 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 2702 | |
5f00110f GT |
2703 | /* |
2704 | * Preserve previous mempolicy unless mpol remount option was specified. | |
2705 | */ | |
2706 | if (config.mpol) { | |
2707 | mpol_put(sbinfo->mpol); | |
2708 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
2709 | } | |
0edd73b3 HD |
2710 | out: |
2711 | spin_unlock(&sbinfo->stat_lock); | |
2712 | return error; | |
1da177e4 | 2713 | } |
680d794b | 2714 | |
34c80b1d | 2715 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 2716 | { |
34c80b1d | 2717 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b AM |
2718 | |
2719 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
2720 | seq_printf(seq, ",size=%luk", | |
2721 | sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); | |
2722 | if (sbinfo->max_inodes != shmem_default_max_inodes()) | |
2723 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
2724 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 2725 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
2726 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
2727 | seq_printf(seq, ",uid=%u", | |
2728 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
2729 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
2730 | seq_printf(seq, ",gid=%u", | |
2731 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
71fe804b | 2732 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b AM |
2733 | return 0; |
2734 | } | |
2735 | #endif /* CONFIG_TMPFS */ | |
1da177e4 LT |
2736 | |
2737 | static void shmem_put_super(struct super_block *sb) | |
2738 | { | |
602586a8 HD |
2739 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
2740 | ||
2741 | percpu_counter_destroy(&sbinfo->used_blocks); | |
49cd0a5c | 2742 | mpol_put(sbinfo->mpol); |
602586a8 | 2743 | kfree(sbinfo); |
1da177e4 LT |
2744 | sb->s_fs_info = NULL; |
2745 | } | |
2746 | ||
2b2af54a | 2747 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
2748 | { |
2749 | struct inode *inode; | |
0edd73b3 | 2750 | struct shmem_sb_info *sbinfo; |
680d794b AM |
2751 | int err = -ENOMEM; |
2752 | ||
2753 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 2754 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b AM |
2755 | L1_CACHE_BYTES), GFP_KERNEL); |
2756 | if (!sbinfo) | |
2757 | return -ENOMEM; | |
2758 | ||
680d794b | 2759 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
2760 | sbinfo->uid = current_fsuid(); |
2761 | sbinfo->gid = current_fsgid(); | |
680d794b | 2762 | sb->s_fs_info = sbinfo; |
1da177e4 | 2763 | |
0edd73b3 | 2764 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
2765 | /* |
2766 | * Per default we only allow half of the physical ram per | |
2767 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
2768 | * but the internal instance is left unlimited. | |
2769 | */ | |
ca4e0519 | 2770 | if (!(sb->s_flags & MS_KERNMOUNT)) { |
680d794b AM |
2771 | sbinfo->max_blocks = shmem_default_max_blocks(); |
2772 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
2773 | if (shmem_parse_options(data, sbinfo, false)) { | |
2774 | err = -EINVAL; | |
2775 | goto failed; | |
2776 | } | |
ca4e0519 AV |
2777 | } else { |
2778 | sb->s_flags |= MS_NOUSER; | |
1da177e4 | 2779 | } |
91828a40 | 2780 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 2781 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
2782 | #else |
2783 | sb->s_flags |= MS_NOUSER; | |
2784 | #endif | |
2785 | ||
0edd73b3 | 2786 | spin_lock_init(&sbinfo->stat_lock); |
602586a8 HD |
2787 | if (percpu_counter_init(&sbinfo->used_blocks, 0)) |
2788 | goto failed; | |
680d794b | 2789 | sbinfo->free_inodes = sbinfo->max_inodes; |
0edd73b3 | 2790 | |
285b2c4f | 2791 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
1da177e4 LT |
2792 | sb->s_blocksize = PAGE_CACHE_SIZE; |
2793 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
2794 | sb->s_magic = TMPFS_MAGIC; | |
2795 | sb->s_op = &shmem_ops; | |
cfd95a9c | 2796 | sb->s_time_gran = 1; |
b09e0fa4 | 2797 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 2798 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
2799 | #endif |
2800 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
2801 | sb->s_flags |= MS_POSIXACL; |
2802 | #endif | |
0edd73b3 | 2803 | |
454abafe | 2804 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
2805 | if (!inode) |
2806 | goto failed; | |
680d794b AM |
2807 | inode->i_uid = sbinfo->uid; |
2808 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
2809 | sb->s_root = d_make_root(inode); |
2810 | if (!sb->s_root) | |
48fde701 | 2811 | goto failed; |
1da177e4 LT |
2812 | return 0; |
2813 | ||
1da177e4 LT |
2814 | failed: |
2815 | shmem_put_super(sb); | |
2816 | return err; | |
2817 | } | |
2818 | ||
fcc234f8 | 2819 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
2820 | |
2821 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
2822 | { | |
41ffe5d5 HD |
2823 | struct shmem_inode_info *info; |
2824 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
2825 | if (!info) | |
1da177e4 | 2826 | return NULL; |
41ffe5d5 | 2827 | return &info->vfs_inode; |
1da177e4 LT |
2828 | } |
2829 | ||
41ffe5d5 | 2830 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
2831 | { |
2832 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
2833 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
2834 | } | |
2835 | ||
1da177e4 LT |
2836 | static void shmem_destroy_inode(struct inode *inode) |
2837 | { | |
09208d15 | 2838 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 2839 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 2840 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
2841 | } |
2842 | ||
41ffe5d5 | 2843 | static void shmem_init_inode(void *foo) |
1da177e4 | 2844 | { |
41ffe5d5 HD |
2845 | struct shmem_inode_info *info = foo; |
2846 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
2847 | } |
2848 | ||
41ffe5d5 | 2849 | static int shmem_init_inodecache(void) |
1da177e4 LT |
2850 | { |
2851 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
2852 | sizeof(struct shmem_inode_info), | |
41ffe5d5 | 2853 | 0, SLAB_PANIC, shmem_init_inode); |
1da177e4 LT |
2854 | return 0; |
2855 | } | |
2856 | ||
41ffe5d5 | 2857 | static void shmem_destroy_inodecache(void) |
1da177e4 | 2858 | { |
1a1d92c1 | 2859 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
2860 | } |
2861 | ||
f5e54d6e | 2862 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 2863 | .writepage = shmem_writepage, |
76719325 | 2864 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 2865 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
2866 | .write_begin = shmem_write_begin, |
2867 | .write_end = shmem_write_end, | |
1da177e4 | 2868 | #endif |
304dbdb7 | 2869 | .migratepage = migrate_page, |
aa261f54 | 2870 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
2871 | }; |
2872 | ||
15ad7cdc | 2873 | static const struct file_operations shmem_file_operations = { |
1da177e4 LT |
2874 | .mmap = shmem_mmap, |
2875 | #ifdef CONFIG_TMPFS | |
220f2ac9 | 2876 | .llseek = shmem_file_llseek, |
2ba5bbed | 2877 | .read = new_sync_read, |
8174202b | 2878 | .write = new_sync_write, |
2ba5bbed | 2879 | .read_iter = shmem_file_read_iter, |
8174202b | 2880 | .write_iter = generic_file_write_iter, |
1b061d92 | 2881 | .fsync = noop_fsync, |
708e3508 | 2882 | .splice_read = shmem_file_splice_read, |
f6cb85d0 | 2883 | .splice_write = iter_file_splice_write, |
83e4fa9c | 2884 | .fallocate = shmem_fallocate, |
1da177e4 LT |
2885 | #endif |
2886 | }; | |
2887 | ||
92e1d5be | 2888 | static const struct inode_operations shmem_inode_operations = { |
94c1e62d | 2889 | .setattr = shmem_setattr, |
b09e0fa4 EP |
2890 | #ifdef CONFIG_TMPFS_XATTR |
2891 | .setxattr = shmem_setxattr, | |
2892 | .getxattr = shmem_getxattr, | |
2893 | .listxattr = shmem_listxattr, | |
2894 | .removexattr = shmem_removexattr, | |
feda821e | 2895 | .set_acl = simple_set_acl, |
b09e0fa4 | 2896 | #endif |
1da177e4 LT |
2897 | }; |
2898 | ||
92e1d5be | 2899 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
2900 | #ifdef CONFIG_TMPFS |
2901 | .create = shmem_create, | |
2902 | .lookup = simple_lookup, | |
2903 | .link = shmem_link, | |
2904 | .unlink = shmem_unlink, | |
2905 | .symlink = shmem_symlink, | |
2906 | .mkdir = shmem_mkdir, | |
2907 | .rmdir = shmem_rmdir, | |
2908 | .mknod = shmem_mknod, | |
2909 | .rename = shmem_rename, | |
60545d0d | 2910 | .tmpfile = shmem_tmpfile, |
1da177e4 | 2911 | #endif |
b09e0fa4 EP |
2912 | #ifdef CONFIG_TMPFS_XATTR |
2913 | .setxattr = shmem_setxattr, | |
2914 | .getxattr = shmem_getxattr, | |
2915 | .listxattr = shmem_listxattr, | |
2916 | .removexattr = shmem_removexattr, | |
2917 | #endif | |
39f0247d | 2918 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 2919 | .setattr = shmem_setattr, |
feda821e | 2920 | .set_acl = simple_set_acl, |
39f0247d AG |
2921 | #endif |
2922 | }; | |
2923 | ||
92e1d5be | 2924 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 EP |
2925 | #ifdef CONFIG_TMPFS_XATTR |
2926 | .setxattr = shmem_setxattr, | |
2927 | .getxattr = shmem_getxattr, | |
2928 | .listxattr = shmem_listxattr, | |
2929 | .removexattr = shmem_removexattr, | |
2930 | #endif | |
39f0247d | 2931 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 2932 | .setattr = shmem_setattr, |
feda821e | 2933 | .set_acl = simple_set_acl, |
39f0247d | 2934 | #endif |
1da177e4 LT |
2935 | }; |
2936 | ||
759b9775 | 2937 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
2938 | .alloc_inode = shmem_alloc_inode, |
2939 | .destroy_inode = shmem_destroy_inode, | |
2940 | #ifdef CONFIG_TMPFS | |
2941 | .statfs = shmem_statfs, | |
2942 | .remount_fs = shmem_remount_fs, | |
680d794b | 2943 | .show_options = shmem_show_options, |
1da177e4 | 2944 | #endif |
1f895f75 | 2945 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
2946 | .drop_inode = generic_delete_inode, |
2947 | .put_super = shmem_put_super, | |
2948 | }; | |
2949 | ||
f0f37e2f | 2950 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 2951 | .fault = shmem_fault, |
d7c17551 | 2952 | .map_pages = filemap_map_pages, |
1da177e4 LT |
2953 | #ifdef CONFIG_NUMA |
2954 | .set_policy = shmem_set_policy, | |
2955 | .get_policy = shmem_get_policy, | |
2956 | #endif | |
0b173bc4 | 2957 | .remap_pages = generic_file_remap_pages, |
1da177e4 LT |
2958 | }; |
2959 | ||
3c26ff6e AV |
2960 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
2961 | int flags, const char *dev_name, void *data) | |
1da177e4 | 2962 | { |
3c26ff6e | 2963 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
2964 | } |
2965 | ||
41ffe5d5 | 2966 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
2967 | .owner = THIS_MODULE, |
2968 | .name = "tmpfs", | |
3c26ff6e | 2969 | .mount = shmem_mount, |
1da177e4 | 2970 | .kill_sb = kill_litter_super, |
2b8576cb | 2971 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 2972 | }; |
1da177e4 | 2973 | |
41ffe5d5 | 2974 | int __init shmem_init(void) |
1da177e4 LT |
2975 | { |
2976 | int error; | |
2977 | ||
16203a7a RL |
2978 | /* If rootfs called this, don't re-init */ |
2979 | if (shmem_inode_cachep) | |
2980 | return 0; | |
2981 | ||
e0bf68dd PZ |
2982 | error = bdi_init(&shmem_backing_dev_info); |
2983 | if (error) | |
2984 | goto out4; | |
2985 | ||
41ffe5d5 | 2986 | error = shmem_init_inodecache(); |
1da177e4 LT |
2987 | if (error) |
2988 | goto out3; | |
2989 | ||
41ffe5d5 | 2990 | error = register_filesystem(&shmem_fs_type); |
1da177e4 LT |
2991 | if (error) { |
2992 | printk(KERN_ERR "Could not register tmpfs\n"); | |
2993 | goto out2; | |
2994 | } | |
95dc112a | 2995 | |
ca4e0519 | 2996 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
2997 | if (IS_ERR(shm_mnt)) { |
2998 | error = PTR_ERR(shm_mnt); | |
2999 | printk(KERN_ERR "Could not kern_mount tmpfs\n"); | |
3000 | goto out1; | |
3001 | } | |
3002 | return 0; | |
3003 | ||
3004 | out1: | |
41ffe5d5 | 3005 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 3006 | out2: |
41ffe5d5 | 3007 | shmem_destroy_inodecache(); |
1da177e4 | 3008 | out3: |
e0bf68dd PZ |
3009 | bdi_destroy(&shmem_backing_dev_info); |
3010 | out4: | |
1da177e4 LT |
3011 | shm_mnt = ERR_PTR(error); |
3012 | return error; | |
3013 | } | |
853ac43a MM |
3014 | |
3015 | #else /* !CONFIG_SHMEM */ | |
3016 | ||
3017 | /* | |
3018 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
3019 | * | |
3020 | * This is intended for small system where the benefits of the full | |
3021 | * shmem code (swap-backed and resource-limited) are outweighed by | |
3022 | * their complexity. On systems without swap this code should be | |
3023 | * effectively equivalent, but much lighter weight. | |
3024 | */ | |
3025 | ||
41ffe5d5 | 3026 | static struct file_system_type shmem_fs_type = { |
853ac43a | 3027 | .name = "tmpfs", |
3c26ff6e | 3028 | .mount = ramfs_mount, |
853ac43a | 3029 | .kill_sb = kill_litter_super, |
2b8576cb | 3030 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
3031 | }; |
3032 | ||
41ffe5d5 | 3033 | int __init shmem_init(void) |
853ac43a | 3034 | { |
41ffe5d5 | 3035 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 3036 | |
41ffe5d5 | 3037 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
3038 | BUG_ON(IS_ERR(shm_mnt)); |
3039 | ||
3040 | return 0; | |
3041 | } | |
3042 | ||
41ffe5d5 | 3043 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
3044 | { |
3045 | return 0; | |
3046 | } | |
3047 | ||
3f96b79a HD |
3048 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
3049 | { | |
3050 | return 0; | |
3051 | } | |
3052 | ||
24513264 HD |
3053 | void shmem_unlock_mapping(struct address_space *mapping) |
3054 | { | |
3055 | } | |
3056 | ||
41ffe5d5 | 3057 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 3058 | { |
41ffe5d5 | 3059 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
3060 | } |
3061 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
3062 | ||
0b0a0806 HD |
3063 | #define shmem_vm_ops generic_file_vm_ops |
3064 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 3065 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
3066 | #define shmem_acct_size(flags, size) 0 |
3067 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
3068 | |
3069 | #endif /* CONFIG_SHMEM */ | |
3070 | ||
3071 | /* common code */ | |
1da177e4 | 3072 | |
3451538a | 3073 | static struct dentry_operations anon_ops = { |
118b2302 | 3074 | .d_dname = simple_dname |
3451538a AV |
3075 | }; |
3076 | ||
c7277090 EP |
3077 | static struct file *__shmem_file_setup(const char *name, loff_t size, |
3078 | unsigned long flags, unsigned int i_flags) | |
1da177e4 | 3079 | { |
6b4d0b27 | 3080 | struct file *res; |
1da177e4 | 3081 | struct inode *inode; |
2c48b9c4 | 3082 | struct path path; |
3451538a | 3083 | struct super_block *sb; |
1da177e4 LT |
3084 | struct qstr this; |
3085 | ||
3086 | if (IS_ERR(shm_mnt)) | |
6b4d0b27 | 3087 | return ERR_CAST(shm_mnt); |
1da177e4 | 3088 | |
285b2c4f | 3089 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
3090 | return ERR_PTR(-EINVAL); |
3091 | ||
3092 | if (shmem_acct_size(flags, size)) | |
3093 | return ERR_PTR(-ENOMEM); | |
3094 | ||
6b4d0b27 | 3095 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
3096 | this.name = name; |
3097 | this.len = strlen(name); | |
3098 | this.hash = 0; /* will go */ | |
3451538a | 3099 | sb = shm_mnt->mnt_sb; |
66ee4b88 | 3100 | path.mnt = mntget(shm_mnt); |
3451538a | 3101 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 3102 | if (!path.dentry) |
1da177e4 | 3103 | goto put_memory; |
3451538a | 3104 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 3105 | |
6b4d0b27 | 3106 | res = ERR_PTR(-ENOSPC); |
3451538a | 3107 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 3108 | if (!inode) |
66ee4b88 | 3109 | goto put_memory; |
1da177e4 | 3110 | |
c7277090 | 3111 | inode->i_flags |= i_flags; |
2c48b9c4 | 3112 | d_instantiate(path.dentry, inode); |
1da177e4 | 3113 | inode->i_size = size; |
6d6b77f1 | 3114 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
3115 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
3116 | if (IS_ERR(res)) | |
66ee4b88 | 3117 | goto put_path; |
4b42af81 | 3118 | |
6b4d0b27 | 3119 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 3120 | &shmem_file_operations); |
6b4d0b27 | 3121 | if (IS_ERR(res)) |
66ee4b88 | 3122 | goto put_path; |
4b42af81 | 3123 | |
6b4d0b27 | 3124 | return res; |
1da177e4 | 3125 | |
1da177e4 LT |
3126 | put_memory: |
3127 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
3128 | put_path: |
3129 | path_put(&path); | |
6b4d0b27 | 3130 | return res; |
1da177e4 | 3131 | } |
c7277090 EP |
3132 | |
3133 | /** | |
3134 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
3135 | * kernel internal. There will be NO LSM permission checks against the | |
3136 | * underlying inode. So users of this interface must do LSM checks at a | |
3137 | * higher layer. The one user is the big_key implementation. LSM checks | |
3138 | * are provided at the key level rather than the inode level. | |
3139 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
3140 | * @size: size to be set for the file | |
3141 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3142 | */ | |
3143 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
3144 | { | |
3145 | return __shmem_file_setup(name, size, flags, S_PRIVATE); | |
3146 | } | |
3147 | ||
3148 | /** | |
3149 | * shmem_file_setup - get an unlinked file living in tmpfs | |
3150 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
3151 | * @size: size to be set for the file | |
3152 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3153 | */ | |
3154 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
3155 | { | |
3156 | return __shmem_file_setup(name, size, flags, 0); | |
3157 | } | |
395e0ddc | 3158 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 3159 | |
46711810 | 3160 | /** |
1da177e4 | 3161 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
3162 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
3163 | */ | |
3164 | int shmem_zero_setup(struct vm_area_struct *vma) | |
3165 | { | |
3166 | struct file *file; | |
3167 | loff_t size = vma->vm_end - vma->vm_start; | |
3168 | ||
3169 | file = shmem_file_setup("dev/zero", size, vma->vm_flags); | |
3170 | if (IS_ERR(file)) | |
3171 | return PTR_ERR(file); | |
3172 | ||
3173 | if (vma->vm_file) | |
3174 | fput(vma->vm_file); | |
3175 | vma->vm_file = file; | |
3176 | vma->vm_ops = &shmem_vm_ops; | |
3177 | return 0; | |
3178 | } | |
d9d90e5e HD |
3179 | |
3180 | /** | |
3181 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
3182 | * @mapping: the page's address_space | |
3183 | * @index: the page index | |
3184 | * @gfp: the page allocator flags to use if allocating | |
3185 | * | |
3186 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
3187 | * with any new page allocations done using the specified allocation flags. | |
3188 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
3189 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
3190 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
3191 | * | |
68da9f05 HD |
3192 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
3193 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
3194 | */ |
3195 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
3196 | pgoff_t index, gfp_t gfp) | |
3197 | { | |
68da9f05 HD |
3198 | #ifdef CONFIG_SHMEM |
3199 | struct inode *inode = mapping->host; | |
9276aad6 | 3200 | struct page *page; |
68da9f05 HD |
3201 | int error; |
3202 | ||
3203 | BUG_ON(mapping->a_ops != &shmem_aops); | |
3204 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL); | |
3205 | if (error) | |
3206 | page = ERR_PTR(error); | |
3207 | else | |
3208 | unlock_page(page); | |
3209 | return page; | |
3210 | #else | |
3211 | /* | |
3212 | * The tiny !SHMEM case uses ramfs without swap | |
3213 | */ | |
d9d90e5e | 3214 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 3215 | #endif |
d9d90e5e HD |
3216 | } |
3217 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |