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> |
39f0247d | 48 | #include <linux/generic_acl.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> |
304dbdb7 | 69 | |
1da177e4 | 70 | #include <asm/uaccess.h> |
1da177e4 LT |
71 | #include <asm/pgtable.h> |
72 | ||
caefba17 | 73 | #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) |
1da177e4 LT |
74 | #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) |
75 | ||
1da177e4 LT |
76 | /* Pretend that each entry is of this size in directory's i_size */ |
77 | #define BOGO_DIRENT_SIZE 20 | |
78 | ||
69f07ec9 HD |
79 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
80 | #define SHORT_SYMLINK_LEN 128 | |
81 | ||
1aac1400 | 82 | /* |
1ccc3ffa HD |
83 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
84 | * inode->i_private (with i_mutex making sure that it has only one user at | |
85 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
86 | */ |
87 | struct shmem_falloc { | |
887675c9 | 88 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
89 | pgoff_t start; /* start of range currently being fallocated */ |
90 | pgoff_t next; /* the next page offset to be fallocated */ | |
91 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
92 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
93 | }; | |
94 | ||
285b2c4f | 95 | /* Flag allocation requirements to shmem_getpage */ |
1da177e4 | 96 | enum sgp_type { |
1da177e4 LT |
97 | SGP_READ, /* don't exceed i_size, don't allocate page */ |
98 | SGP_CACHE, /* don't exceed i_size, may allocate page */ | |
a0ee5ec5 | 99 | SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ |
1635f6a7 HD |
100 | SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */ |
101 | SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */ | |
1da177e4 LT |
102 | }; |
103 | ||
b76db735 | 104 | #ifdef CONFIG_TMPFS |
680d794b | 105 | static unsigned long shmem_default_max_blocks(void) |
106 | { | |
107 | return totalram_pages / 2; | |
108 | } | |
109 | ||
110 | static unsigned long shmem_default_max_inodes(void) | |
111 | { | |
112 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
113 | } | |
b76db735 | 114 | #endif |
680d794b | 115 | |
bde05d1c HD |
116 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
117 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
118 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 HD |
119 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
120 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type); | |
121 | ||
122 | static inline int shmem_getpage(struct inode *inode, pgoff_t index, | |
123 | struct page **pagep, enum sgp_type sgp, int *fault_type) | |
124 | { | |
125 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
126 | mapping_gfp_mask(inode->i_mapping), fault_type); | |
127 | } | |
1da177e4 | 128 | |
1da177e4 LT |
129 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
130 | { | |
131 | return sb->s_fs_info; | |
132 | } | |
133 | ||
134 | /* | |
135 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
136 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
137 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
138 | * consistent with the pre-accounting of private mappings ... | |
139 | */ | |
140 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
141 | { | |
0b0a0806 | 142 | return (flags & VM_NORESERVE) ? |
191c5424 | 143 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
144 | } |
145 | ||
146 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
147 | { | |
0b0a0806 | 148 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
149 | vm_unacct_memory(VM_ACCT(size)); |
150 | } | |
151 | ||
152 | /* | |
153 | * ... whereas tmpfs objects are accounted incrementally as | |
154 | * pages are allocated, in order to allow huge sparse files. | |
155 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, | |
156 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
157 | */ | |
158 | static inline int shmem_acct_block(unsigned long flags) | |
159 | { | |
0b0a0806 | 160 | return (flags & VM_NORESERVE) ? |
191c5424 | 161 | security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0; |
1da177e4 LT |
162 | } |
163 | ||
164 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
165 | { | |
0b0a0806 | 166 | if (flags & VM_NORESERVE) |
1da177e4 LT |
167 | vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); |
168 | } | |
169 | ||
759b9775 | 170 | static const struct super_operations shmem_ops; |
f5e54d6e | 171 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 172 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
173 | static const struct inode_operations shmem_inode_operations; |
174 | static const struct inode_operations shmem_dir_inode_operations; | |
175 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 176 | static const struct vm_operations_struct shmem_vm_ops; |
1da177e4 | 177 | |
6c231b7b | 178 | static struct backing_dev_info shmem_backing_dev_info __read_mostly = { |
1da177e4 | 179 | .ra_pages = 0, /* No readahead */ |
4f98a2fe | 180 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED, |
1da177e4 LT |
181 | }; |
182 | ||
183 | static LIST_HEAD(shmem_swaplist); | |
cb5f7b9a | 184 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 185 | |
5b04c689 PE |
186 | static int shmem_reserve_inode(struct super_block *sb) |
187 | { | |
188 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
189 | if (sbinfo->max_inodes) { | |
190 | spin_lock(&sbinfo->stat_lock); | |
191 | if (!sbinfo->free_inodes) { | |
192 | spin_unlock(&sbinfo->stat_lock); | |
193 | return -ENOSPC; | |
194 | } | |
195 | sbinfo->free_inodes--; | |
196 | spin_unlock(&sbinfo->stat_lock); | |
197 | } | |
198 | return 0; | |
199 | } | |
200 | ||
201 | static void shmem_free_inode(struct super_block *sb) | |
202 | { | |
203 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
204 | if (sbinfo->max_inodes) { | |
205 | spin_lock(&sbinfo->stat_lock); | |
206 | sbinfo->free_inodes++; | |
207 | spin_unlock(&sbinfo->stat_lock); | |
208 | } | |
209 | } | |
210 | ||
46711810 | 211 | /** |
41ffe5d5 | 212 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
213 | * @inode: inode to recalc |
214 | * | |
215 | * We have to calculate the free blocks since the mm can drop | |
216 | * undirtied hole pages behind our back. | |
217 | * | |
218 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
219 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
220 | * | |
221 | * It has to be called with the spinlock held. | |
222 | */ | |
223 | static void shmem_recalc_inode(struct inode *inode) | |
224 | { | |
225 | struct shmem_inode_info *info = SHMEM_I(inode); | |
226 | long freed; | |
227 | ||
228 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
229 | if (freed > 0) { | |
54af6042 HD |
230 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
231 | if (sbinfo->max_blocks) | |
232 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 233 | info->alloced -= freed; |
54af6042 | 234 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 235 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
236 | } |
237 | } | |
238 | ||
7a5d0fbb HD |
239 | /* |
240 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
241 | */ | |
242 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
243 | pgoff_t index, void *expected, void *replacement) | |
244 | { | |
245 | void **pslot; | |
246 | void *item = NULL; | |
247 | ||
248 | VM_BUG_ON(!expected); | |
249 | pslot = radix_tree_lookup_slot(&mapping->page_tree, index); | |
250 | if (pslot) | |
251 | item = radix_tree_deref_slot_protected(pslot, | |
252 | &mapping->tree_lock); | |
253 | if (item != expected) | |
254 | return -ENOENT; | |
255 | if (replacement) | |
256 | radix_tree_replace_slot(pslot, replacement); | |
257 | else | |
258 | radix_tree_delete(&mapping->page_tree, index); | |
259 | return 0; | |
260 | } | |
261 | ||
d1899228 HD |
262 | /* |
263 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
264 | * that an entry was not already brought back from swap by a racing thread. | |
265 | * | |
266 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
267 | * might be reused, and again be SwapCache, using the same swap as before. | |
268 | */ | |
269 | static bool shmem_confirm_swap(struct address_space *mapping, | |
270 | pgoff_t index, swp_entry_t swap) | |
271 | { | |
272 | void *item; | |
273 | ||
274 | rcu_read_lock(); | |
275 | item = radix_tree_lookup(&mapping->page_tree, index); | |
276 | rcu_read_unlock(); | |
277 | return item == swp_to_radix_entry(swap); | |
278 | } | |
279 | ||
46f65ec1 HD |
280 | /* |
281 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
282 | */ | |
283 | static int shmem_add_to_page_cache(struct page *page, | |
284 | struct address_space *mapping, | |
285 | pgoff_t index, gfp_t gfp, void *expected) | |
286 | { | |
b065b432 | 287 | int error; |
46f65ec1 HD |
288 | |
289 | VM_BUG_ON(!PageLocked(page)); | |
290 | VM_BUG_ON(!PageSwapBacked(page)); | |
291 | ||
b065b432 HD |
292 | page_cache_get(page); |
293 | page->mapping = mapping; | |
294 | page->index = index; | |
295 | ||
296 | spin_lock_irq(&mapping->tree_lock); | |
46f65ec1 | 297 | if (!expected) |
b065b432 HD |
298 | error = radix_tree_insert(&mapping->page_tree, index, page); |
299 | else | |
300 | error = shmem_radix_tree_replace(mapping, index, expected, | |
301 | page); | |
46f65ec1 | 302 | if (!error) { |
b065b432 HD |
303 | mapping->nrpages++; |
304 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
305 | __inc_zone_page_state(page, NR_SHMEM); | |
306 | spin_unlock_irq(&mapping->tree_lock); | |
307 | } else { | |
308 | page->mapping = NULL; | |
309 | spin_unlock_irq(&mapping->tree_lock); | |
310 | page_cache_release(page); | |
46f65ec1 | 311 | } |
46f65ec1 HD |
312 | return error; |
313 | } | |
314 | ||
6922c0c7 HD |
315 | /* |
316 | * Like delete_from_page_cache, but substitutes swap for page. | |
317 | */ | |
318 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
319 | { | |
320 | struct address_space *mapping = page->mapping; | |
321 | int error; | |
322 | ||
323 | spin_lock_irq(&mapping->tree_lock); | |
324 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
325 | page->mapping = NULL; | |
326 | mapping->nrpages--; | |
327 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
328 | __dec_zone_page_state(page, NR_SHMEM); | |
329 | spin_unlock_irq(&mapping->tree_lock); | |
330 | page_cache_release(page); | |
331 | BUG_ON(error); | |
332 | } | |
333 | ||
7a5d0fbb HD |
334 | /* |
335 | * Like find_get_pages, but collecting swap entries as well as pages. | |
336 | */ | |
337 | static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping, | |
338 | pgoff_t start, unsigned int nr_pages, | |
339 | struct page **pages, pgoff_t *indices) | |
340 | { | |
860f2759 JW |
341 | void **slot; |
342 | unsigned int ret = 0; | |
343 | struct radix_tree_iter iter; | |
344 | ||
345 | if (!nr_pages) | |
346 | return 0; | |
7a5d0fbb HD |
347 | |
348 | rcu_read_lock(); | |
349 | restart: | |
860f2759 | 350 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
7a5d0fbb HD |
351 | struct page *page; |
352 | repeat: | |
860f2759 | 353 | page = radix_tree_deref_slot(slot); |
7a5d0fbb HD |
354 | if (unlikely(!page)) |
355 | continue; | |
356 | if (radix_tree_exception(page)) { | |
8079b1c8 HD |
357 | if (radix_tree_deref_retry(page)) |
358 | goto restart; | |
359 | /* | |
360 | * Otherwise, we must be storing a swap entry | |
361 | * here as an exceptional entry: so return it | |
362 | * without attempting to raise page count. | |
363 | */ | |
364 | goto export; | |
7a5d0fbb HD |
365 | } |
366 | if (!page_cache_get_speculative(page)) | |
367 | goto repeat; | |
368 | ||
369 | /* Has the page moved? */ | |
860f2759 | 370 | if (unlikely(page != *slot)) { |
7a5d0fbb HD |
371 | page_cache_release(page); |
372 | goto repeat; | |
373 | } | |
374 | export: | |
860f2759 | 375 | indices[ret] = iter.index; |
7a5d0fbb | 376 | pages[ret] = page; |
860f2759 JW |
377 | if (++ret == nr_pages) |
378 | break; | |
7a5d0fbb | 379 | } |
7a5d0fbb HD |
380 | rcu_read_unlock(); |
381 | return ret; | |
382 | } | |
383 | ||
384 | /* | |
385 | * Remove swap entry from radix tree, free the swap and its page cache. | |
386 | */ | |
387 | static int shmem_free_swap(struct address_space *mapping, | |
388 | pgoff_t index, void *radswap) | |
389 | { | |
390 | int error; | |
391 | ||
392 | spin_lock_irq(&mapping->tree_lock); | |
393 | error = shmem_radix_tree_replace(mapping, index, radswap, NULL); | |
394 | spin_unlock_irq(&mapping->tree_lock); | |
395 | if (!error) | |
396 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
397 | return error; | |
398 | } | |
399 | ||
400 | /* | |
401 | * Pagevec may contain swap entries, so shuffle up pages before releasing. | |
402 | */ | |
24513264 | 403 | static void shmem_deswap_pagevec(struct pagevec *pvec) |
7a5d0fbb HD |
404 | { |
405 | int i, j; | |
406 | ||
407 | for (i = 0, j = 0; i < pagevec_count(pvec); i++) { | |
408 | struct page *page = pvec->pages[i]; | |
409 | if (!radix_tree_exceptional_entry(page)) | |
410 | pvec->pages[j++] = page; | |
411 | } | |
412 | pvec->nr = j; | |
24513264 HD |
413 | } |
414 | ||
415 | /* | |
416 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
417 | */ | |
418 | void shmem_unlock_mapping(struct address_space *mapping) | |
419 | { | |
420 | struct pagevec pvec; | |
421 | pgoff_t indices[PAGEVEC_SIZE]; | |
422 | pgoff_t index = 0; | |
423 | ||
424 | pagevec_init(&pvec, 0); | |
425 | /* | |
426 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
427 | */ | |
428 | while (!mapping_unevictable(mapping)) { | |
429 | /* | |
430 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
431 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
432 | */ | |
433 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, | |
434 | PAGEVEC_SIZE, pvec.pages, indices); | |
435 | if (!pvec.nr) | |
436 | break; | |
437 | index = indices[pvec.nr - 1] + 1; | |
438 | shmem_deswap_pagevec(&pvec); | |
439 | check_move_unevictable_pages(pvec.pages, pvec.nr); | |
440 | pagevec_release(&pvec); | |
441 | cond_resched(); | |
442 | } | |
7a5d0fbb HD |
443 | } |
444 | ||
445 | /* | |
446 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 447 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 448 | */ |
1635f6a7 HD |
449 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
450 | bool unfalloc) | |
1da177e4 | 451 | { |
285b2c4f | 452 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 453 | struct shmem_inode_info *info = SHMEM_I(inode); |
285b2c4f | 454 | pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
83e4fa9c HD |
455 | pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT; |
456 | unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1); | |
457 | unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); | |
bda97eab | 458 | struct pagevec pvec; |
7a5d0fbb HD |
459 | pgoff_t indices[PAGEVEC_SIZE]; |
460 | long nr_swaps_freed = 0; | |
285b2c4f | 461 | pgoff_t index; |
bda97eab HD |
462 | int i; |
463 | ||
83e4fa9c HD |
464 | if (lend == -1) |
465 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
466 | |
467 | pagevec_init(&pvec, 0); | |
468 | index = start; | |
83e4fa9c | 469 | while (index < end) { |
7a5d0fbb | 470 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, |
83e4fa9c | 471 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
7a5d0fbb HD |
472 | pvec.pages, indices); |
473 | if (!pvec.nr) | |
474 | break; | |
bda97eab HD |
475 | mem_cgroup_uncharge_start(); |
476 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
477 | struct page *page = pvec.pages[i]; | |
478 | ||
7a5d0fbb | 479 | index = indices[i]; |
83e4fa9c | 480 | if (index >= end) |
bda97eab HD |
481 | break; |
482 | ||
7a5d0fbb | 483 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
484 | if (unfalloc) |
485 | continue; | |
7a5d0fbb HD |
486 | nr_swaps_freed += !shmem_free_swap(mapping, |
487 | index, page); | |
bda97eab | 488 | continue; |
7a5d0fbb HD |
489 | } |
490 | ||
491 | if (!trylock_page(page)) | |
bda97eab | 492 | continue; |
1635f6a7 HD |
493 | if (!unfalloc || !PageUptodate(page)) { |
494 | if (page->mapping == mapping) { | |
495 | VM_BUG_ON(PageWriteback(page)); | |
496 | truncate_inode_page(mapping, page); | |
497 | } | |
bda97eab | 498 | } |
bda97eab HD |
499 | unlock_page(page); |
500 | } | |
24513264 HD |
501 | shmem_deswap_pagevec(&pvec); |
502 | pagevec_release(&pvec); | |
bda97eab HD |
503 | mem_cgroup_uncharge_end(); |
504 | cond_resched(); | |
505 | index++; | |
506 | } | |
1da177e4 | 507 | |
83e4fa9c | 508 | if (partial_start) { |
bda97eab HD |
509 | struct page *page = NULL; |
510 | shmem_getpage(inode, start - 1, &page, SGP_READ, NULL); | |
511 | if (page) { | |
83e4fa9c HD |
512 | unsigned int top = PAGE_CACHE_SIZE; |
513 | if (start > end) { | |
514 | top = partial_end; | |
515 | partial_end = 0; | |
516 | } | |
517 | zero_user_segment(page, partial_start, top); | |
518 | set_page_dirty(page); | |
519 | unlock_page(page); | |
520 | page_cache_release(page); | |
521 | } | |
522 | } | |
523 | if (partial_end) { | |
524 | struct page *page = NULL; | |
525 | shmem_getpage(inode, end, &page, SGP_READ, NULL); | |
526 | if (page) { | |
527 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
528 | set_page_dirty(page); |
529 | unlock_page(page); | |
530 | page_cache_release(page); | |
531 | } | |
532 | } | |
83e4fa9c HD |
533 | if (start >= end) |
534 | return; | |
bda97eab HD |
535 | |
536 | index = start; | |
7dc7fb43 | 537 | while (index < end) { |
bda97eab | 538 | cond_resched(); |
7a5d0fbb | 539 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, |
83e4fa9c | 540 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
7a5d0fbb HD |
541 | pvec.pages, indices); |
542 | if (!pvec.nr) { | |
7dc7fb43 HD |
543 | /* If all gone or hole-punch or unfalloc, we're done */ |
544 | if (index == start || end != -1) | |
bda97eab | 545 | break; |
7dc7fb43 | 546 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
547 | index = start; |
548 | continue; | |
549 | } | |
bda97eab HD |
550 | mem_cgroup_uncharge_start(); |
551 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
552 | struct page *page = pvec.pages[i]; | |
553 | ||
7a5d0fbb | 554 | index = indices[i]; |
83e4fa9c | 555 | if (index >= end) |
bda97eab HD |
556 | break; |
557 | ||
7a5d0fbb | 558 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
559 | if (unfalloc) |
560 | continue; | |
7dc7fb43 HD |
561 | if (shmem_free_swap(mapping, index, page)) { |
562 | /* Swap was replaced by page: retry */ | |
563 | index--; | |
564 | break; | |
565 | } | |
566 | nr_swaps_freed++; | |
7a5d0fbb HD |
567 | continue; |
568 | } | |
569 | ||
bda97eab | 570 | lock_page(page); |
1635f6a7 HD |
571 | if (!unfalloc || !PageUptodate(page)) { |
572 | if (page->mapping == mapping) { | |
573 | VM_BUG_ON(PageWriteback(page)); | |
574 | truncate_inode_page(mapping, page); | |
7dc7fb43 HD |
575 | } else { |
576 | /* Page was replaced by swap: retry */ | |
577 | unlock_page(page); | |
578 | index--; | |
579 | break; | |
1635f6a7 | 580 | } |
7a5d0fbb | 581 | } |
bda97eab HD |
582 | unlock_page(page); |
583 | } | |
24513264 HD |
584 | shmem_deswap_pagevec(&pvec); |
585 | pagevec_release(&pvec); | |
bda97eab HD |
586 | mem_cgroup_uncharge_end(); |
587 | index++; | |
588 | } | |
94c1e62d | 589 | |
1da177e4 | 590 | spin_lock(&info->lock); |
7a5d0fbb | 591 | info->swapped -= nr_swaps_freed; |
1da177e4 LT |
592 | shmem_recalc_inode(inode); |
593 | spin_unlock(&info->lock); | |
1635f6a7 | 594 | } |
1da177e4 | 595 | |
1635f6a7 HD |
596 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
597 | { | |
598 | shmem_undo_range(inode, lstart, lend, false); | |
285b2c4f | 599 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1da177e4 | 600 | } |
94c1e62d | 601 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 602 | |
94c1e62d | 603 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 LT |
604 | { |
605 | struct inode *inode = dentry->d_inode; | |
1da177e4 LT |
606 | int error; |
607 | ||
db78b877 CH |
608 | error = inode_change_ok(inode, attr); |
609 | if (error) | |
610 | return error; | |
611 | ||
94c1e62d HD |
612 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
613 | loff_t oldsize = inode->i_size; | |
614 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 615 | |
94c1e62d HD |
616 | if (newsize != oldsize) { |
617 | i_size_write(inode, newsize); | |
618 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
619 | } | |
620 | if (newsize < oldsize) { | |
621 | loff_t holebegin = round_up(newsize, PAGE_SIZE); | |
622 | unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); | |
623 | shmem_truncate_range(inode, newsize, (loff_t)-1); | |
624 | /* unmap again to remove racily COWed private pages */ | |
625 | unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); | |
626 | } | |
1da177e4 LT |
627 | } |
628 | ||
db78b877 | 629 | setattr_copy(inode, attr); |
39f0247d | 630 | #ifdef CONFIG_TMPFS_POSIX_ACL |
db78b877 | 631 | if (attr->ia_valid & ATTR_MODE) |
1c7c474c | 632 | error = generic_acl_chmod(inode); |
39f0247d | 633 | #endif |
1da177e4 LT |
634 | return error; |
635 | } | |
636 | ||
1f895f75 | 637 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 638 | { |
1da177e4 LT |
639 | struct shmem_inode_info *info = SHMEM_I(inode); |
640 | ||
3889e6e7 | 641 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
642 | shmem_unacct_size(info->flags, inode->i_size); |
643 | inode->i_size = 0; | |
3889e6e7 | 644 | shmem_truncate_range(inode, 0, (loff_t)-1); |
1da177e4 | 645 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 646 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 647 | list_del_init(&info->swaplist); |
cb5f7b9a | 648 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 649 | } |
69f07ec9 HD |
650 | } else |
651 | kfree(info->symlink); | |
b09e0fa4 | 652 | |
38f38657 | 653 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 654 | WARN_ON(inode->i_blocks); |
5b04c689 | 655 | shmem_free_inode(inode->i_sb); |
dbd5768f | 656 | clear_inode(inode); |
1da177e4 LT |
657 | } |
658 | ||
46f65ec1 HD |
659 | /* |
660 | * If swap found in inode, free it and move page from swapcache to filecache. | |
661 | */ | |
41ffe5d5 | 662 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 663 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 664 | { |
285b2c4f | 665 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 666 | void *radswap; |
41ffe5d5 | 667 | pgoff_t index; |
bde05d1c HD |
668 | gfp_t gfp; |
669 | int error = 0; | |
1da177e4 | 670 | |
46f65ec1 | 671 | radswap = swp_to_radix_entry(swap); |
e504f3fd | 672 | index = radix_tree_locate_item(&mapping->page_tree, radswap); |
46f65ec1 | 673 | if (index == -1) |
285b2c4f | 674 | return 0; |
2e0e26c7 | 675 | |
1b1b32f2 HD |
676 | /* |
677 | * Move _head_ to start search for next from here. | |
1f895f75 | 678 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 679 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 680 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
681 | */ |
682 | if (shmem_swaplist.next != &info->swaplist) | |
683 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 684 | |
bde05d1c HD |
685 | gfp = mapping_gfp_mask(mapping); |
686 | if (shmem_should_replace_page(*pagep, gfp)) { | |
687 | mutex_unlock(&shmem_swaplist_mutex); | |
688 | error = shmem_replace_page(pagep, gfp, info, index); | |
689 | mutex_lock(&shmem_swaplist_mutex); | |
690 | /* | |
691 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
692 | * allocation, but the inode might have been freed while we |
693 | * dropped it: although a racing shmem_evict_inode() cannot | |
694 | * complete without emptying the radix_tree, our page lock | |
695 | * on this swapcache page is not enough to prevent that - | |
696 | * free_swap_and_cache() of our swap entry will only | |
697 | * trylock_page(), removing swap from radix_tree whatever. | |
698 | * | |
699 | * We must not proceed to shmem_add_to_page_cache() if the | |
700 | * inode has been freed, but of course we cannot rely on | |
701 | * inode or mapping or info to check that. However, we can | |
702 | * safely check if our swap entry is still in use (and here | |
703 | * it can't have got reused for another page): if it's still | |
704 | * in use, then the inode cannot have been freed yet, and we | |
705 | * can safely proceed (if it's no longer in use, that tells | |
706 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
707 | */ |
708 | if (!page_swapcount(*pagep)) | |
709 | error = -ENOENT; | |
710 | } | |
711 | ||
d13d1443 | 712 | /* |
778dd893 HD |
713 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
714 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
715 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 716 | */ |
bde05d1c HD |
717 | if (!error) |
718 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
46f65ec1 | 719 | GFP_NOWAIT, radswap); |
48f170fb | 720 | if (error != -ENOMEM) { |
46f65ec1 HD |
721 | /* |
722 | * Truncation and eviction use free_swap_and_cache(), which | |
723 | * only does trylock page: if we raced, best clean up here. | |
724 | */ | |
bde05d1c HD |
725 | delete_from_swap_cache(*pagep); |
726 | set_page_dirty(*pagep); | |
46f65ec1 HD |
727 | if (!error) { |
728 | spin_lock(&info->lock); | |
729 | info->swapped--; | |
730 | spin_unlock(&info->lock); | |
731 | swap_free(swap); | |
732 | } | |
2e0e26c7 | 733 | error = 1; /* not an error, but entry was found */ |
1da177e4 | 734 | } |
2e0e26c7 | 735 | return error; |
1da177e4 LT |
736 | } |
737 | ||
738 | /* | |
46f65ec1 | 739 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 740 | */ |
41ffe5d5 | 741 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 742 | { |
41ffe5d5 | 743 | struct list_head *this, *next; |
1da177e4 LT |
744 | struct shmem_inode_info *info; |
745 | int found = 0; | |
bde05d1c HD |
746 | int error = 0; |
747 | ||
748 | /* | |
749 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 750 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 751 | */ |
0142ef6c | 752 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 753 | goto out; |
778dd893 HD |
754 | |
755 | /* | |
756 | * Charge page using GFP_KERNEL while we can wait, before taking | |
757 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
758 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 HD |
759 | */ |
760 | error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); | |
761 | if (error) | |
762 | goto out; | |
46f65ec1 | 763 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
1da177e4 | 764 | |
cb5f7b9a | 765 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
766 | list_for_each_safe(this, next, &shmem_swaplist) { |
767 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 768 | if (info->swapped) |
bde05d1c | 769 | found = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
770 | else |
771 | list_del_init(&info->swaplist); | |
cb5f7b9a | 772 | cond_resched(); |
2e0e26c7 | 773 | if (found) |
778dd893 | 774 | break; |
1da177e4 | 775 | } |
cb5f7b9a | 776 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 777 | |
778dd893 HD |
778 | if (found < 0) |
779 | error = found; | |
780 | out: | |
aaa46865 HD |
781 | unlock_page(page); |
782 | page_cache_release(page); | |
778dd893 | 783 | return error; |
1da177e4 LT |
784 | } |
785 | ||
786 | /* | |
787 | * Move the page from the page cache to the swap cache. | |
788 | */ | |
789 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
790 | { | |
791 | struct shmem_inode_info *info; | |
1da177e4 | 792 | struct address_space *mapping; |
1da177e4 | 793 | struct inode *inode; |
6922c0c7 HD |
794 | swp_entry_t swap; |
795 | pgoff_t index; | |
1da177e4 LT |
796 | |
797 | BUG_ON(!PageLocked(page)); | |
1da177e4 LT |
798 | mapping = page->mapping; |
799 | index = page->index; | |
800 | inode = mapping->host; | |
801 | info = SHMEM_I(inode); | |
802 | if (info->flags & VM_LOCKED) | |
803 | goto redirty; | |
d9fe526a | 804 | if (!total_swap_pages) |
1da177e4 LT |
805 | goto redirty; |
806 | ||
d9fe526a HD |
807 | /* |
808 | * shmem_backing_dev_info's capabilities prevent regular writeback or | |
809 | * sync from ever calling shmem_writepage; but a stacking filesystem | |
48f170fb | 810 | * might use ->writepage of its underlying filesystem, in which case |
d9fe526a | 811 | * tmpfs should write out to swap only in response to memory pressure, |
48f170fb | 812 | * and not for the writeback threads or sync. |
d9fe526a | 813 | */ |
48f170fb HD |
814 | if (!wbc->for_reclaim) { |
815 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
816 | goto redirty; | |
817 | } | |
1635f6a7 HD |
818 | |
819 | /* | |
820 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
821 | * value into swapfile.c, the only way we can correctly account for a | |
822 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
823 | * |
824 | * That's okay for a page already fallocated earlier, but if we have | |
825 | * not yet completed the fallocation, then (a) we want to keep track | |
826 | * of this page in case we have to undo it, and (b) it may not be a | |
827 | * good idea to continue anyway, once we're pushing into swap. So | |
828 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
829 | */ |
830 | if (!PageUptodate(page)) { | |
1aac1400 HD |
831 | if (inode->i_private) { |
832 | struct shmem_falloc *shmem_falloc; | |
833 | spin_lock(&inode->i_lock); | |
834 | shmem_falloc = inode->i_private; | |
835 | if (shmem_falloc && | |
887675c9 | 836 | !shmem_falloc->waitq && |
1aac1400 HD |
837 | index >= shmem_falloc->start && |
838 | index < shmem_falloc->next) | |
839 | shmem_falloc->nr_unswapped++; | |
840 | else | |
841 | shmem_falloc = NULL; | |
842 | spin_unlock(&inode->i_lock); | |
843 | if (shmem_falloc) | |
844 | goto redirty; | |
845 | } | |
1635f6a7 HD |
846 | clear_highpage(page); |
847 | flush_dcache_page(page); | |
848 | SetPageUptodate(page); | |
849 | } | |
850 | ||
48f170fb HD |
851 | swap = get_swap_page(); |
852 | if (!swap.val) | |
853 | goto redirty; | |
d9fe526a | 854 | |
b1dea800 HD |
855 | /* |
856 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
857 | * if it's not already there. Do it now before the page is |
858 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 859 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
860 | * we've incremented swapped, because shmem_unuse_inode() will |
861 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 862 | */ |
48f170fb HD |
863 | mutex_lock(&shmem_swaplist_mutex); |
864 | if (list_empty(&info->swaplist)) | |
865 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 866 | |
48f170fb | 867 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
aaa46865 | 868 | swap_shmem_alloc(swap); |
6922c0c7 HD |
869 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); |
870 | ||
871 | spin_lock(&info->lock); | |
872 | info->swapped++; | |
873 | shmem_recalc_inode(inode); | |
826267cf | 874 | spin_unlock(&info->lock); |
6922c0c7 HD |
875 | |
876 | mutex_unlock(&shmem_swaplist_mutex); | |
d9fe526a | 877 | BUG_ON(page_mapped(page)); |
9fab5619 | 878 | swap_writepage(page, wbc); |
1da177e4 LT |
879 | return 0; |
880 | } | |
881 | ||
6922c0c7 | 882 | mutex_unlock(&shmem_swaplist_mutex); |
cb4b86ba | 883 | swapcache_free(swap, NULL); |
1da177e4 LT |
884 | redirty: |
885 | set_page_dirty(page); | |
d9fe526a HD |
886 | if (wbc->for_reclaim) |
887 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
888 | unlock_page(page); | |
889 | return 0; | |
1da177e4 LT |
890 | } |
891 | ||
892 | #ifdef CONFIG_NUMA | |
680d794b | 893 | #ifdef CONFIG_TMPFS |
71fe804b | 894 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 895 | { |
095f1fc4 | 896 | char buffer[64]; |
680d794b | 897 | |
71fe804b | 898 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 899 | return; /* show nothing */ |
680d794b | 900 | |
a7a88b23 | 901 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
902 | |
903 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 904 | } |
71fe804b LS |
905 | |
906 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
907 | { | |
908 | struct mempolicy *mpol = NULL; | |
909 | if (sbinfo->mpol) { | |
910 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
911 | mpol = sbinfo->mpol; | |
912 | mpol_get(mpol); | |
913 | spin_unlock(&sbinfo->stat_lock); | |
914 | } | |
915 | return mpol; | |
916 | } | |
680d794b | 917 | #endif /* CONFIG_TMPFS */ |
918 | ||
41ffe5d5 HD |
919 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
920 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 921 | { |
1da177e4 | 922 | struct vm_area_struct pvma; |
18a2f371 | 923 | struct page *page; |
52cd3b07 | 924 | |
1da177e4 | 925 | /* Create a pseudo vma that just contains the policy */ |
c4cc6d07 | 926 | pvma.vm_start = 0; |
09c231cb NZ |
927 | /* Bias interleave by inode number to distribute better across nodes */ |
928 | pvma.vm_pgoff = index + info->vfs_inode.i_ino; | |
c4cc6d07 | 929 | pvma.vm_ops = NULL; |
18a2f371 MG |
930 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
931 | ||
932 | page = swapin_readahead(swap, gfp, &pvma, 0); | |
933 | ||
934 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
935 | mpol_cond_put(pvma.vm_policy); | |
936 | ||
937 | return page; | |
1da177e4 LT |
938 | } |
939 | ||
02098fea | 940 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 941 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
942 | { |
943 | struct vm_area_struct pvma; | |
18a2f371 | 944 | struct page *page; |
1da177e4 | 945 | |
c4cc6d07 HD |
946 | /* Create a pseudo vma that just contains the policy */ |
947 | pvma.vm_start = 0; | |
09c231cb NZ |
948 | /* Bias interleave by inode number to distribute better across nodes */ |
949 | pvma.vm_pgoff = index + info->vfs_inode.i_ino; | |
c4cc6d07 | 950 | pvma.vm_ops = NULL; |
41ffe5d5 | 951 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
52cd3b07 | 952 | |
18a2f371 MG |
953 | page = alloc_page_vma(gfp, &pvma, 0); |
954 | ||
955 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
956 | mpol_cond_put(pvma.vm_policy); | |
957 | ||
958 | return page; | |
1da177e4 | 959 | } |
680d794b | 960 | #else /* !CONFIG_NUMA */ |
961 | #ifdef CONFIG_TMPFS | |
41ffe5d5 | 962 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 963 | { |
964 | } | |
965 | #endif /* CONFIG_TMPFS */ | |
966 | ||
41ffe5d5 HD |
967 | static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
968 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 969 | { |
41ffe5d5 | 970 | return swapin_readahead(swap, gfp, NULL, 0); |
1da177e4 LT |
971 | } |
972 | ||
02098fea | 973 | static inline struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 974 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 | 975 | { |
e84e2e13 | 976 | return alloc_page(gfp); |
1da177e4 | 977 | } |
680d794b | 978 | #endif /* CONFIG_NUMA */ |
1da177e4 | 979 | |
71fe804b LS |
980 | #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) |
981 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
982 | { | |
983 | return NULL; | |
984 | } | |
985 | #endif | |
986 | ||
bde05d1c HD |
987 | /* |
988 | * When a page is moved from swapcache to shmem filecache (either by the | |
989 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
990 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
991 | * ignorance of the mapping it belongs to. If that mapping has special | |
992 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
993 | * we may need to copy to a suitable page before moving to filecache. | |
994 | * | |
995 | * In a future release, this may well be extended to respect cpuset and | |
996 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
997 | * but for now it is a simple matter of zone. | |
998 | */ | |
999 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1000 | { | |
1001 | return page_zonenum(page) > gfp_zone(gfp); | |
1002 | } | |
1003 | ||
1004 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1005 | struct shmem_inode_info *info, pgoff_t index) | |
1006 | { | |
1007 | struct page *oldpage, *newpage; | |
1008 | struct address_space *swap_mapping; | |
1009 | pgoff_t swap_index; | |
1010 | int error; | |
1011 | ||
1012 | oldpage = *pagep; | |
1013 | swap_index = page_private(oldpage); | |
1014 | swap_mapping = page_mapping(oldpage); | |
1015 | ||
1016 | /* | |
1017 | * We have arrived here because our zones are constrained, so don't | |
1018 | * limit chance of success by further cpuset and node constraints. | |
1019 | */ | |
1020 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1021 | newpage = shmem_alloc_page(gfp, info, index); | |
1022 | if (!newpage) | |
1023 | return -ENOMEM; | |
bde05d1c | 1024 | |
bde05d1c HD |
1025 | page_cache_get(newpage); |
1026 | copy_highpage(newpage, oldpage); | |
0142ef6c | 1027 | flush_dcache_page(newpage); |
bde05d1c | 1028 | |
bde05d1c | 1029 | __set_page_locked(newpage); |
bde05d1c | 1030 | SetPageUptodate(newpage); |
bde05d1c | 1031 | SetPageSwapBacked(newpage); |
bde05d1c | 1032 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1033 | SetPageSwapCache(newpage); |
1034 | ||
1035 | /* | |
1036 | * Our caller will very soon move newpage out of swapcache, but it's | |
1037 | * a nice clean interface for us to replace oldpage by newpage there. | |
1038 | */ | |
1039 | spin_lock_irq(&swap_mapping->tree_lock); | |
1040 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
1041 | newpage); | |
0142ef6c HD |
1042 | if (!error) { |
1043 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
1044 | __dec_zone_page_state(oldpage, NR_FILE_PAGES); | |
1045 | } | |
bde05d1c | 1046 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1047 | |
0142ef6c HD |
1048 | if (unlikely(error)) { |
1049 | /* | |
1050 | * Is this possible? I think not, now that our callers check | |
1051 | * both PageSwapCache and page_private after getting page lock; | |
1052 | * but be defensive. Reverse old to newpage for clear and free. | |
1053 | */ | |
1054 | oldpage = newpage; | |
1055 | } else { | |
1056 | mem_cgroup_replace_page_cache(oldpage, newpage); | |
1057 | lru_cache_add_anon(newpage); | |
1058 | *pagep = newpage; | |
1059 | } | |
bde05d1c HD |
1060 | |
1061 | ClearPageSwapCache(oldpage); | |
1062 | set_page_private(oldpage, 0); | |
1063 | ||
1064 | unlock_page(oldpage); | |
1065 | page_cache_release(oldpage); | |
1066 | page_cache_release(oldpage); | |
0142ef6c | 1067 | return error; |
bde05d1c HD |
1068 | } |
1069 | ||
1da177e4 | 1070 | /* |
68da9f05 | 1071 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1072 | * |
1073 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1074 | * vm. If we swap it in we mark it dirty since we also free the swap | |
1075 | * entry since a page cannot live in both the swap and page cache | |
1076 | */ | |
41ffe5d5 | 1077 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
68da9f05 | 1078 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type) |
1da177e4 LT |
1079 | { |
1080 | struct address_space *mapping = inode->i_mapping; | |
54af6042 | 1081 | struct shmem_inode_info *info; |
1da177e4 | 1082 | struct shmem_sb_info *sbinfo; |
27ab7006 | 1083 | struct page *page; |
1da177e4 LT |
1084 | swp_entry_t swap; |
1085 | int error; | |
54af6042 | 1086 | int once = 0; |
1635f6a7 | 1087 | int alloced = 0; |
1da177e4 | 1088 | |
41ffe5d5 | 1089 | if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT)) |
1da177e4 | 1090 | return -EFBIG; |
1da177e4 | 1091 | repeat: |
54af6042 | 1092 | swap.val = 0; |
41ffe5d5 | 1093 | page = find_lock_page(mapping, index); |
54af6042 HD |
1094 | if (radix_tree_exceptional_entry(page)) { |
1095 | swap = radix_to_swp_entry(page); | |
1096 | page = NULL; | |
1097 | } | |
1098 | ||
1635f6a7 | 1099 | if (sgp != SGP_WRITE && sgp != SGP_FALLOC && |
54af6042 HD |
1100 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { |
1101 | error = -EINVAL; | |
1102 | goto failed; | |
1103 | } | |
1104 | ||
1635f6a7 HD |
1105 | /* fallocated page? */ |
1106 | if (page && !PageUptodate(page)) { | |
1107 | if (sgp != SGP_READ) | |
1108 | goto clear; | |
1109 | unlock_page(page); | |
1110 | page_cache_release(page); | |
1111 | page = NULL; | |
1112 | } | |
54af6042 | 1113 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1114 | *pagep = page; |
1115 | return 0; | |
27ab7006 HD |
1116 | } |
1117 | ||
1118 | /* | |
54af6042 HD |
1119 | * Fast cache lookup did not find it: |
1120 | * bring it back from swap or allocate. | |
27ab7006 | 1121 | */ |
54af6042 HD |
1122 | info = SHMEM_I(inode); |
1123 | sbinfo = SHMEM_SB(inode->i_sb); | |
1da177e4 | 1124 | |
1da177e4 LT |
1125 | if (swap.val) { |
1126 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1127 | page = lookup_swap_cache(swap); |
1128 | if (!page) { | |
1da177e4 | 1129 | /* here we actually do the io */ |
68da9f05 HD |
1130 | if (fault_type) |
1131 | *fault_type |= VM_FAULT_MAJOR; | |
41ffe5d5 | 1132 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1133 | if (!page) { |
54af6042 HD |
1134 | error = -ENOMEM; |
1135 | goto failed; | |
1da177e4 | 1136 | } |
1da177e4 LT |
1137 | } |
1138 | ||
1139 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1140 | lock_page(page); |
0142ef6c | 1141 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1142 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1143 | error = -EEXIST; /* try again */ |
d1899228 | 1144 | goto unlock; |
bde05d1c | 1145 | } |
27ab7006 | 1146 | if (!PageUptodate(page)) { |
1da177e4 | 1147 | error = -EIO; |
54af6042 | 1148 | goto failed; |
1da177e4 | 1149 | } |
54af6042 HD |
1150 | wait_on_page_writeback(page); |
1151 | ||
bde05d1c HD |
1152 | if (shmem_should_replace_page(page, gfp)) { |
1153 | error = shmem_replace_page(&page, gfp, info, index); | |
1154 | if (error) | |
1155 | goto failed; | |
1da177e4 | 1156 | } |
27ab7006 | 1157 | |
aa3b1895 HD |
1158 | error = mem_cgroup_cache_charge(page, current->mm, |
1159 | gfp & GFP_RECLAIM_MASK); | |
d1899228 | 1160 | if (!error) { |
aa3b1895 HD |
1161 | error = shmem_add_to_page_cache(page, mapping, index, |
1162 | gfp, swp_to_radix_entry(swap)); | |
215c02bc HD |
1163 | /* |
1164 | * We already confirmed swap under page lock, and make | |
1165 | * no memory allocation here, so usually no possibility | |
1166 | * of error; but free_swap_and_cache() only trylocks a | |
1167 | * page, so it is just possible that the entry has been | |
1168 | * truncated or holepunched since swap was confirmed. | |
1169 | * shmem_undo_range() will have done some of the | |
1170 | * unaccounting, now delete_from_swap_cache() will do | |
1171 | * the rest (including mem_cgroup_uncharge_swapcache). | |
1172 | * Reset swap.val? No, leave it so "failed" goes back to | |
1173 | * "repeat": reading a hole and writing should succeed. | |
1174 | */ | |
1175 | if (error) | |
1176 | delete_from_swap_cache(page); | |
d1899228 | 1177 | } |
54af6042 HD |
1178 | if (error) |
1179 | goto failed; | |
1180 | ||
1181 | spin_lock(&info->lock); | |
285b2c4f | 1182 | info->swapped--; |
54af6042 | 1183 | shmem_recalc_inode(inode); |
27ab7006 | 1184 | spin_unlock(&info->lock); |
54af6042 HD |
1185 | |
1186 | delete_from_swap_cache(page); | |
27ab7006 HD |
1187 | set_page_dirty(page); |
1188 | swap_free(swap); | |
1189 | ||
54af6042 HD |
1190 | } else { |
1191 | if (shmem_acct_block(info->flags)) { | |
1192 | error = -ENOSPC; | |
1193 | goto failed; | |
1da177e4 | 1194 | } |
0edd73b3 | 1195 | if (sbinfo->max_blocks) { |
fc5da22a | 1196 | if (percpu_counter_compare(&sbinfo->used_blocks, |
54af6042 HD |
1197 | sbinfo->max_blocks) >= 0) { |
1198 | error = -ENOSPC; | |
1199 | goto unacct; | |
1200 | } | |
7e496299 | 1201 | percpu_counter_inc(&sbinfo->used_blocks); |
54af6042 | 1202 | } |
1da177e4 | 1203 | |
54af6042 HD |
1204 | page = shmem_alloc_page(gfp, info, index); |
1205 | if (!page) { | |
1206 | error = -ENOMEM; | |
1207 | goto decused; | |
1da177e4 LT |
1208 | } |
1209 | ||
54af6042 HD |
1210 | SetPageSwapBacked(page); |
1211 | __set_page_locked(page); | |
aa3b1895 HD |
1212 | error = mem_cgroup_cache_charge(page, current->mm, |
1213 | gfp & GFP_RECLAIM_MASK); | |
54af6042 HD |
1214 | if (error) |
1215 | goto decused; | |
b065b432 HD |
1216 | error = radix_tree_preload(gfp & GFP_RECLAIM_MASK); |
1217 | if (!error) { | |
1218 | error = shmem_add_to_page_cache(page, mapping, index, | |
1219 | gfp, NULL); | |
1220 | radix_tree_preload_end(); | |
1221 | } | |
1222 | if (error) { | |
1223 | mem_cgroup_uncharge_cache_page(page); | |
1224 | goto decused; | |
1225 | } | |
54af6042 HD |
1226 | lru_cache_add_anon(page); |
1227 | ||
1228 | spin_lock(&info->lock); | |
1da177e4 | 1229 | info->alloced++; |
54af6042 HD |
1230 | inode->i_blocks += BLOCKS_PER_PAGE; |
1231 | shmem_recalc_inode(inode); | |
1da177e4 | 1232 | spin_unlock(&info->lock); |
1635f6a7 | 1233 | alloced = true; |
54af6042 | 1234 | |
ec9516fb | 1235 | /* |
1635f6a7 HD |
1236 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1237 | */ | |
1238 | if (sgp == SGP_FALLOC) | |
1239 | sgp = SGP_WRITE; | |
1240 | clear: | |
1241 | /* | |
1242 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1243 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1244 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb HD |
1245 | */ |
1246 | if (sgp != SGP_WRITE) { | |
1247 | clear_highpage(page); | |
1248 | flush_dcache_page(page); | |
1249 | SetPageUptodate(page); | |
1250 | } | |
a0ee5ec5 | 1251 | if (sgp == SGP_DIRTY) |
27ab7006 | 1252 | set_page_dirty(page); |
1da177e4 | 1253 | } |
bde05d1c | 1254 | |
54af6042 | 1255 | /* Perhaps the file has been truncated since we checked */ |
1635f6a7 | 1256 | if (sgp != SGP_WRITE && sgp != SGP_FALLOC && |
54af6042 HD |
1257 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { |
1258 | error = -EINVAL; | |
1635f6a7 HD |
1259 | if (alloced) |
1260 | goto trunc; | |
1261 | else | |
1262 | goto failed; | |
e83c32e8 | 1263 | } |
54af6042 HD |
1264 | *pagep = page; |
1265 | return 0; | |
1da177e4 | 1266 | |
59a16ead | 1267 | /* |
54af6042 | 1268 | * Error recovery. |
59a16ead | 1269 | */ |
54af6042 | 1270 | trunc: |
1635f6a7 | 1271 | info = SHMEM_I(inode); |
54af6042 HD |
1272 | ClearPageDirty(page); |
1273 | delete_from_page_cache(page); | |
1274 | spin_lock(&info->lock); | |
1275 | info->alloced--; | |
1276 | inode->i_blocks -= BLOCKS_PER_PAGE; | |
59a16ead | 1277 | spin_unlock(&info->lock); |
54af6042 | 1278 | decused: |
1635f6a7 | 1279 | sbinfo = SHMEM_SB(inode->i_sb); |
54af6042 HD |
1280 | if (sbinfo->max_blocks) |
1281 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
1282 | unacct: | |
1283 | shmem_unacct_blocks(info->flags, 1); | |
1284 | failed: | |
d1899228 HD |
1285 | if (swap.val && error != -EINVAL && |
1286 | !shmem_confirm_swap(mapping, index, swap)) | |
1287 | error = -EEXIST; | |
1288 | unlock: | |
27ab7006 | 1289 | if (page) { |
54af6042 | 1290 | unlock_page(page); |
27ab7006 | 1291 | page_cache_release(page); |
54af6042 HD |
1292 | } |
1293 | if (error == -ENOSPC && !once++) { | |
1294 | info = SHMEM_I(inode); | |
1295 | spin_lock(&info->lock); | |
1296 | shmem_recalc_inode(inode); | |
1297 | spin_unlock(&info->lock); | |
27ab7006 | 1298 | goto repeat; |
ff36b801 | 1299 | } |
d1899228 | 1300 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1301 | goto repeat; |
1302 | return error; | |
1da177e4 LT |
1303 | } |
1304 | ||
d0217ac0 | 1305 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1306 | { |
496ad9aa | 1307 | struct inode *inode = file_inode(vma->vm_file); |
1da177e4 | 1308 | int error; |
68da9f05 | 1309 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1310 | |
1ccc3ffa HD |
1311 | /* |
1312 | * Trinity finds that probing a hole which tmpfs is punching can | |
1313 | * prevent the hole-punch from ever completing: which in turn | |
1314 | * locks writers out with its hold on i_mutex. So refrain from | |
887675c9 HD |
1315 | * faulting pages into the hole while it's being punched. Although |
1316 | * shmem_undo_range() does remove the additions, it may be unable to | |
1317 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1318 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1319 | * | |
1320 | * It does not matter if we sometimes reach this check just before the | |
1321 | * hole-punch begins, so that one fault then races with the punch: | |
1322 | * we just need to make racing faults a rare case. | |
1323 | * | |
1324 | * The implementation below would be much simpler if we just used a | |
1325 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1326 | * and bloating every shmem inode for this unlikely case would be sad. | |
1ccc3ffa HD |
1327 | */ |
1328 | if (unlikely(inode->i_private)) { | |
1329 | struct shmem_falloc *shmem_falloc; | |
1330 | ||
1331 | spin_lock(&inode->i_lock); | |
1332 | shmem_falloc = inode->i_private; | |
887675c9 HD |
1333 | if (shmem_falloc && |
1334 | shmem_falloc->waitq && | |
1335 | vmf->pgoff >= shmem_falloc->start && | |
1336 | vmf->pgoff < shmem_falloc->next) { | |
1337 | wait_queue_head_t *shmem_falloc_waitq; | |
1338 | DEFINE_WAIT(shmem_fault_wait); | |
1339 | ||
1340 | ret = VM_FAULT_NOPAGE; | |
1ccc3ffa HD |
1341 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1342 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
887675c9 | 1343 | /* It's polite to up mmap_sem if we can */ |
1ccc3ffa | 1344 | up_read(&vma->vm_mm->mmap_sem); |
887675c9 | 1345 | ret = VM_FAULT_RETRY; |
1ccc3ffa | 1346 | } |
887675c9 HD |
1347 | |
1348 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1349 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1350 | TASK_UNINTERRUPTIBLE); | |
1351 | spin_unlock(&inode->i_lock); | |
1352 | schedule(); | |
1353 | ||
1354 | /* | |
1355 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1356 | * stack of the hole-punching task: shmem_falloc_waitq | |
1357 | * is usually invalid by the time we reach here, but | |
1358 | * finish_wait() does not dereference it in that case; | |
1359 | * though i_lock needed lest racing with wake_up_all(). | |
1360 | */ | |
1361 | spin_lock(&inode->i_lock); | |
1362 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1363 | spin_unlock(&inode->i_lock); | |
1364 | return ret; | |
1ccc3ffa | 1365 | } |
887675c9 | 1366 | spin_unlock(&inode->i_lock); |
1ccc3ffa HD |
1367 | } |
1368 | ||
27d54b39 | 1369 | error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); |
d0217ac0 NP |
1370 | if (error) |
1371 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1372 | |
456f998e YH |
1373 | if (ret & VM_FAULT_MAJOR) { |
1374 | count_vm_event(PGMAJFAULT); | |
1375 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); | |
1376 | } | |
68da9f05 | 1377 | return ret; |
1da177e4 LT |
1378 | } |
1379 | ||
1da177e4 | 1380 | #ifdef CONFIG_NUMA |
41ffe5d5 | 1381 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 1382 | { |
496ad9aa | 1383 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1384 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
1385 | } |
1386 | ||
d8dc74f2 AB |
1387 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
1388 | unsigned long addr) | |
1da177e4 | 1389 | { |
496ad9aa | 1390 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1391 | pgoff_t index; |
1da177e4 | 1392 | |
41ffe5d5 HD |
1393 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
1394 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
1395 | } |
1396 | #endif | |
1397 | ||
1398 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
1399 | { | |
496ad9aa | 1400 | struct inode *inode = file_inode(file); |
1da177e4 LT |
1401 | struct shmem_inode_info *info = SHMEM_I(inode); |
1402 | int retval = -ENOMEM; | |
1403 | ||
1404 | spin_lock(&info->lock); | |
1405 | if (lock && !(info->flags & VM_LOCKED)) { | |
1406 | if (!user_shm_lock(inode->i_size, user)) | |
1407 | goto out_nomem; | |
1408 | info->flags |= VM_LOCKED; | |
89e004ea | 1409 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
1410 | } |
1411 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
1412 | user_shm_unlock(inode->i_size, user); | |
1413 | info->flags &= ~VM_LOCKED; | |
89e004ea | 1414 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
1415 | } |
1416 | retval = 0; | |
89e004ea | 1417 | |
1da177e4 LT |
1418 | out_nomem: |
1419 | spin_unlock(&info->lock); | |
1420 | return retval; | |
1421 | } | |
1422 | ||
9b83a6a8 | 1423 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
1424 | { |
1425 | file_accessed(file); | |
1426 | vma->vm_ops = &shmem_vm_ops; | |
1427 | return 0; | |
1428 | } | |
1429 | ||
454abafe | 1430 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 1431 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
1432 | { |
1433 | struct inode *inode; | |
1434 | struct shmem_inode_info *info; | |
1435 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
1436 | ||
5b04c689 PE |
1437 | if (shmem_reserve_inode(sb)) |
1438 | return NULL; | |
1da177e4 LT |
1439 | |
1440 | inode = new_inode(sb); | |
1441 | if (inode) { | |
85fe4025 | 1442 | inode->i_ino = get_next_ino(); |
454abafe | 1443 | inode_init_owner(inode, dir, mode); |
1da177e4 | 1444 | inode->i_blocks = 0; |
1da177e4 LT |
1445 | inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; |
1446 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
91828a40 | 1447 | inode->i_generation = get_seconds(); |
1da177e4 LT |
1448 | info = SHMEM_I(inode); |
1449 | memset(info, 0, (char *)inode - (char *)info); | |
1450 | spin_lock_init(&info->lock); | |
0b0a0806 | 1451 | info->flags = flags & VM_NORESERVE; |
1da177e4 | 1452 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 1453 | simple_xattrs_init(&info->xattrs); |
72c04902 | 1454 | cache_no_acl(inode); |
1da177e4 LT |
1455 | |
1456 | switch (mode & S_IFMT) { | |
1457 | default: | |
39f0247d | 1458 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
1459 | init_special_inode(inode, mode, dev); |
1460 | break; | |
1461 | case S_IFREG: | |
14fcc23f | 1462 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
1463 | inode->i_op = &shmem_inode_operations; |
1464 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
1465 | mpol_shared_policy_init(&info->policy, |
1466 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
1467 | break; |
1468 | case S_IFDIR: | |
d8c76e6f | 1469 | inc_nlink(inode); |
1da177e4 LT |
1470 | /* Some things misbehave if size == 0 on a directory */ |
1471 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
1472 | inode->i_op = &shmem_dir_inode_operations; | |
1473 | inode->i_fop = &simple_dir_operations; | |
1474 | break; | |
1475 | case S_IFLNK: | |
1476 | /* | |
1477 | * Must not load anything in the rbtree, | |
1478 | * mpol_free_shared_policy will not be called. | |
1479 | */ | |
71fe804b | 1480 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
1481 | break; |
1482 | } | |
5b04c689 PE |
1483 | } else |
1484 | shmem_free_inode(sb); | |
1da177e4 LT |
1485 | return inode; |
1486 | } | |
1487 | ||
1488 | #ifdef CONFIG_TMPFS | |
92e1d5be | 1489 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 1490 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 1491 | |
6d9d88d0 JS |
1492 | #ifdef CONFIG_TMPFS_XATTR |
1493 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
1494 | #else | |
1495 | #define shmem_initxattrs NULL | |
1496 | #endif | |
1497 | ||
1da177e4 | 1498 | static int |
800d15a5 NP |
1499 | shmem_write_begin(struct file *file, struct address_space *mapping, |
1500 | loff_t pos, unsigned len, unsigned flags, | |
1501 | struct page **pagep, void **fsdata) | |
1da177e4 | 1502 | { |
800d15a5 NP |
1503 | struct inode *inode = mapping->host; |
1504 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
800d15a5 NP |
1505 | return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); |
1506 | } | |
1507 | ||
1508 | static int | |
1509 | shmem_write_end(struct file *file, struct address_space *mapping, | |
1510 | loff_t pos, unsigned len, unsigned copied, | |
1511 | struct page *page, void *fsdata) | |
1512 | { | |
1513 | struct inode *inode = mapping->host; | |
1514 | ||
d3602444 HD |
1515 | if (pos + copied > inode->i_size) |
1516 | i_size_write(inode, pos + copied); | |
1517 | ||
ec9516fb HD |
1518 | if (!PageUptodate(page)) { |
1519 | if (copied < PAGE_CACHE_SIZE) { | |
1520 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
1521 | zero_user_segments(page, 0, from, | |
1522 | from + copied, PAGE_CACHE_SIZE); | |
1523 | } | |
1524 | SetPageUptodate(page); | |
1525 | } | |
800d15a5 | 1526 | set_page_dirty(page); |
6746aff7 | 1527 | unlock_page(page); |
800d15a5 NP |
1528 | page_cache_release(page); |
1529 | ||
800d15a5 | 1530 | return copied; |
1da177e4 LT |
1531 | } |
1532 | ||
1da177e4 LT |
1533 | static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) |
1534 | { | |
496ad9aa | 1535 | struct inode *inode = file_inode(filp); |
1da177e4 | 1536 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
1537 | pgoff_t index; |
1538 | unsigned long offset; | |
a0ee5ec5 HD |
1539 | enum sgp_type sgp = SGP_READ; |
1540 | ||
1541 | /* | |
1542 | * Might this read be for a stacking filesystem? Then when reading | |
1543 | * holes of a sparse file, we actually need to allocate those pages, | |
1544 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
1545 | */ | |
1546 | if (segment_eq(get_fs(), KERNEL_DS)) | |
1547 | sgp = SGP_DIRTY; | |
1da177e4 LT |
1548 | |
1549 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1550 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1551 | ||
1552 | for (;;) { | |
1553 | struct page *page = NULL; | |
41ffe5d5 HD |
1554 | pgoff_t end_index; |
1555 | unsigned long nr, ret; | |
1da177e4 LT |
1556 | loff_t i_size = i_size_read(inode); |
1557 | ||
1558 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1559 | if (index > end_index) | |
1560 | break; | |
1561 | if (index == end_index) { | |
1562 | nr = i_size & ~PAGE_CACHE_MASK; | |
1563 | if (nr <= offset) | |
1564 | break; | |
1565 | } | |
1566 | ||
a0ee5ec5 | 1567 | desc->error = shmem_getpage(inode, index, &page, sgp, NULL); |
1da177e4 LT |
1568 | if (desc->error) { |
1569 | if (desc->error == -EINVAL) | |
1570 | desc->error = 0; | |
1571 | break; | |
1572 | } | |
d3602444 HD |
1573 | if (page) |
1574 | unlock_page(page); | |
1da177e4 LT |
1575 | |
1576 | /* | |
1577 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 1578 | * are called without i_mutex protection against truncate |
1da177e4 LT |
1579 | */ |
1580 | nr = PAGE_CACHE_SIZE; | |
1581 | i_size = i_size_read(inode); | |
1582 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1583 | if (index == end_index) { | |
1584 | nr = i_size & ~PAGE_CACHE_MASK; | |
1585 | if (nr <= offset) { | |
1586 | if (page) | |
1587 | page_cache_release(page); | |
1588 | break; | |
1589 | } | |
1590 | } | |
1591 | nr -= offset; | |
1592 | ||
1593 | if (page) { | |
1594 | /* | |
1595 | * If users can be writing to this page using arbitrary | |
1596 | * virtual addresses, take care about potential aliasing | |
1597 | * before reading the page on the kernel side. | |
1598 | */ | |
1599 | if (mapping_writably_mapped(mapping)) | |
1600 | flush_dcache_page(page); | |
1601 | /* | |
1602 | * Mark the page accessed if we read the beginning. | |
1603 | */ | |
1604 | if (!offset) | |
1605 | mark_page_accessed(page); | |
b5810039 | 1606 | } else { |
1da177e4 | 1607 | page = ZERO_PAGE(0); |
b5810039 NP |
1608 | page_cache_get(page); |
1609 | } | |
1da177e4 LT |
1610 | |
1611 | /* | |
1612 | * Ok, we have the page, and it's up-to-date, so | |
1613 | * now we can copy it to user space... | |
1614 | * | |
1615 | * The actor routine returns how many bytes were actually used.. | |
1616 | * NOTE! This may not be the same as how much of a user buffer | |
1617 | * we filled up (we may be padding etc), so we can only update | |
1618 | * "pos" here (the actor routine has to update the user buffer | |
1619 | * pointers and the remaining count). | |
1620 | */ | |
1621 | ret = actor(desc, page, offset, nr); | |
1622 | offset += ret; | |
1623 | index += offset >> PAGE_CACHE_SHIFT; | |
1624 | offset &= ~PAGE_CACHE_MASK; | |
1625 | ||
1626 | page_cache_release(page); | |
1627 | if (ret != nr || !desc->count) | |
1628 | break; | |
1629 | ||
1630 | cond_resched(); | |
1631 | } | |
1632 | ||
1633 | *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; | |
1634 | file_accessed(filp); | |
1635 | } | |
1636 | ||
bcd78e49 HD |
1637 | static ssize_t shmem_file_aio_read(struct kiocb *iocb, |
1638 | const struct iovec *iov, unsigned long nr_segs, loff_t pos) | |
1639 | { | |
1640 | struct file *filp = iocb->ki_filp; | |
1641 | ssize_t retval; | |
1642 | unsigned long seg; | |
1643 | size_t count; | |
1644 | loff_t *ppos = &iocb->ki_pos; | |
1645 | ||
1646 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); | |
1647 | if (retval) | |
1648 | return retval; | |
1649 | ||
1650 | for (seg = 0; seg < nr_segs; seg++) { | |
1651 | read_descriptor_t desc; | |
1652 | ||
1653 | desc.written = 0; | |
1654 | desc.arg.buf = iov[seg].iov_base; | |
1655 | desc.count = iov[seg].iov_len; | |
1656 | if (desc.count == 0) | |
1657 | continue; | |
1658 | desc.error = 0; | |
1659 | do_shmem_file_read(filp, ppos, &desc, file_read_actor); | |
1660 | retval += desc.written; | |
1661 | if (desc.error) { | |
1662 | retval = retval ?: desc.error; | |
1663 | break; | |
1664 | } | |
1665 | if (desc.count > 0) | |
1666 | break; | |
1667 | } | |
1668 | return retval; | |
1da177e4 LT |
1669 | } |
1670 | ||
708e3508 HD |
1671 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, |
1672 | struct pipe_inode_info *pipe, size_t len, | |
1673 | unsigned int flags) | |
1674 | { | |
1675 | struct address_space *mapping = in->f_mapping; | |
71f0e07a | 1676 | struct inode *inode = mapping->host; |
708e3508 HD |
1677 | unsigned int loff, nr_pages, req_pages; |
1678 | struct page *pages[PIPE_DEF_BUFFERS]; | |
1679 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
1680 | struct page *page; | |
1681 | pgoff_t index, end_index; | |
1682 | loff_t isize, left; | |
1683 | int error, page_nr; | |
1684 | struct splice_pipe_desc spd = { | |
1685 | .pages = pages, | |
1686 | .partial = partial, | |
047fe360 | 1687 | .nr_pages_max = PIPE_DEF_BUFFERS, |
708e3508 HD |
1688 | .flags = flags, |
1689 | .ops = &page_cache_pipe_buf_ops, | |
1690 | .spd_release = spd_release_page, | |
1691 | }; | |
1692 | ||
71f0e07a | 1693 | isize = i_size_read(inode); |
708e3508 HD |
1694 | if (unlikely(*ppos >= isize)) |
1695 | return 0; | |
1696 | ||
1697 | left = isize - *ppos; | |
1698 | if (unlikely(left < len)) | |
1699 | len = left; | |
1700 | ||
1701 | if (splice_grow_spd(pipe, &spd)) | |
1702 | return -ENOMEM; | |
1703 | ||
1704 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1705 | loff = *ppos & ~PAGE_CACHE_MASK; | |
1706 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
1707 | nr_pages = min(req_pages, pipe->buffers); | |
1708 | ||
708e3508 HD |
1709 | spd.nr_pages = find_get_pages_contig(mapping, index, |
1710 | nr_pages, spd.pages); | |
1711 | index += spd.nr_pages; | |
708e3508 | 1712 | error = 0; |
708e3508 | 1713 | |
71f0e07a | 1714 | while (spd.nr_pages < nr_pages) { |
71f0e07a HD |
1715 | error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL); |
1716 | if (error) | |
1717 | break; | |
1718 | unlock_page(page); | |
708e3508 HD |
1719 | spd.pages[spd.nr_pages++] = page; |
1720 | index++; | |
1721 | } | |
1722 | ||
708e3508 HD |
1723 | index = *ppos >> PAGE_CACHE_SHIFT; |
1724 | nr_pages = spd.nr_pages; | |
1725 | spd.nr_pages = 0; | |
71f0e07a | 1726 | |
708e3508 HD |
1727 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { |
1728 | unsigned int this_len; | |
1729 | ||
1730 | if (!len) | |
1731 | break; | |
1732 | ||
708e3508 HD |
1733 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); |
1734 | page = spd.pages[page_nr]; | |
1735 | ||
71f0e07a | 1736 | if (!PageUptodate(page) || page->mapping != mapping) { |
71f0e07a HD |
1737 | error = shmem_getpage(inode, index, &page, |
1738 | SGP_CACHE, NULL); | |
1739 | if (error) | |
708e3508 | 1740 | break; |
71f0e07a HD |
1741 | unlock_page(page); |
1742 | page_cache_release(spd.pages[page_nr]); | |
1743 | spd.pages[page_nr] = page; | |
708e3508 | 1744 | } |
71f0e07a HD |
1745 | |
1746 | isize = i_size_read(inode); | |
708e3508 HD |
1747 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; |
1748 | if (unlikely(!isize || index > end_index)) | |
1749 | break; | |
1750 | ||
708e3508 HD |
1751 | if (end_index == index) { |
1752 | unsigned int plen; | |
1753 | ||
708e3508 HD |
1754 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
1755 | if (plen <= loff) | |
1756 | break; | |
1757 | ||
708e3508 HD |
1758 | this_len = min(this_len, plen - loff); |
1759 | len = this_len; | |
1760 | } | |
1761 | ||
1762 | spd.partial[page_nr].offset = loff; | |
1763 | spd.partial[page_nr].len = this_len; | |
1764 | len -= this_len; | |
1765 | loff = 0; | |
1766 | spd.nr_pages++; | |
1767 | index++; | |
1768 | } | |
1769 | ||
708e3508 HD |
1770 | while (page_nr < nr_pages) |
1771 | page_cache_release(spd.pages[page_nr++]); | |
708e3508 HD |
1772 | |
1773 | if (spd.nr_pages) | |
1774 | error = splice_to_pipe(pipe, &spd); | |
1775 | ||
047fe360 | 1776 | splice_shrink_spd(&spd); |
708e3508 HD |
1777 | |
1778 | if (error > 0) { | |
1779 | *ppos += error; | |
1780 | file_accessed(in); | |
1781 | } | |
1782 | return error; | |
1783 | } | |
1784 | ||
220f2ac9 HD |
1785 | /* |
1786 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
1787 | */ | |
1788 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 1789 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
1790 | { |
1791 | struct page *page; | |
1792 | struct pagevec pvec; | |
1793 | pgoff_t indices[PAGEVEC_SIZE]; | |
1794 | bool done = false; | |
1795 | int i; | |
1796 | ||
1797 | pagevec_init(&pvec, 0); | |
1798 | pvec.nr = 1; /* start small: we may be there already */ | |
1799 | while (!done) { | |
1800 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, | |
1801 | pvec.nr, pvec.pages, indices); | |
1802 | if (!pvec.nr) { | |
965c8e59 | 1803 | if (whence == SEEK_DATA) |
220f2ac9 HD |
1804 | index = end; |
1805 | break; | |
1806 | } | |
1807 | for (i = 0; i < pvec.nr; i++, index++) { | |
1808 | if (index < indices[i]) { | |
965c8e59 | 1809 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
1810 | done = true; |
1811 | break; | |
1812 | } | |
1813 | index = indices[i]; | |
1814 | } | |
1815 | page = pvec.pages[i]; | |
1816 | if (page && !radix_tree_exceptional_entry(page)) { | |
1817 | if (!PageUptodate(page)) | |
1818 | page = NULL; | |
1819 | } | |
1820 | if (index >= end || | |
965c8e59 AM |
1821 | (page && whence == SEEK_DATA) || |
1822 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
1823 | done = true; |
1824 | break; | |
1825 | } | |
1826 | } | |
1827 | shmem_deswap_pagevec(&pvec); | |
1828 | pagevec_release(&pvec); | |
1829 | pvec.nr = PAGEVEC_SIZE; | |
1830 | cond_resched(); | |
1831 | } | |
1832 | return index; | |
1833 | } | |
1834 | ||
965c8e59 | 1835 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
1836 | { |
1837 | struct address_space *mapping = file->f_mapping; | |
1838 | struct inode *inode = mapping->host; | |
1839 | pgoff_t start, end; | |
1840 | loff_t new_offset; | |
1841 | ||
965c8e59 AM |
1842 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
1843 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 HD |
1844 | MAX_LFS_FILESIZE, i_size_read(inode)); |
1845 | mutex_lock(&inode->i_mutex); | |
1846 | /* We're holding i_mutex so we can access i_size directly */ | |
1847 | ||
1848 | if (offset < 0) | |
1849 | offset = -EINVAL; | |
1850 | else if (offset >= inode->i_size) | |
1851 | offset = -ENXIO; | |
1852 | else { | |
1853 | start = offset >> PAGE_CACHE_SHIFT; | |
1854 | end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
965c8e59 | 1855 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
220f2ac9 HD |
1856 | new_offset <<= PAGE_CACHE_SHIFT; |
1857 | if (new_offset > offset) { | |
1858 | if (new_offset < inode->i_size) | |
1859 | offset = new_offset; | |
965c8e59 | 1860 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
1861 | offset = -ENXIO; |
1862 | else | |
1863 | offset = inode->i_size; | |
1864 | } | |
1865 | } | |
1866 | ||
1867 | if (offset >= 0 && offset != file->f_pos) { | |
1868 | file->f_pos = offset; | |
1869 | file->f_version = 0; | |
1870 | } | |
1871 | mutex_unlock(&inode->i_mutex); | |
1872 | return offset; | |
1873 | } | |
1874 | ||
83e4fa9c HD |
1875 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
1876 | loff_t len) | |
1877 | { | |
496ad9aa | 1878 | struct inode *inode = file_inode(file); |
e2d12e22 | 1879 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1aac1400 | 1880 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
1881 | pgoff_t start, index, end; |
1882 | int error; | |
83e4fa9c HD |
1883 | |
1884 | mutex_lock(&inode->i_mutex); | |
1885 | ||
1886 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1887 | struct address_space *mapping = file->f_mapping; | |
1888 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
1889 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
887675c9 | 1890 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 1891 | |
887675c9 | 1892 | shmem_falloc.waitq = &shmem_falloc_waitq; |
1ccc3ffa HD |
1893 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
1894 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
1895 | spin_lock(&inode->i_lock); | |
1896 | inode->i_private = &shmem_falloc; | |
1897 | spin_unlock(&inode->i_lock); | |
1898 | ||
83e4fa9c HD |
1899 | if ((u64)unmap_end > (u64)unmap_start) |
1900 | unmap_mapping_range(mapping, unmap_start, | |
1901 | 1 + unmap_end - unmap_start, 0); | |
1902 | shmem_truncate_range(inode, offset, offset + len - 1); | |
1903 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
887675c9 HD |
1904 | |
1905 | spin_lock(&inode->i_lock); | |
1906 | inode->i_private = NULL; | |
1907 | wake_up_all(&shmem_falloc_waitq); | |
1908 | spin_unlock(&inode->i_lock); | |
83e4fa9c | 1909 | error = 0; |
887675c9 | 1910 | goto out; |
e2d12e22 HD |
1911 | } |
1912 | ||
1913 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
1914 | error = inode_newsize_ok(inode, offset + len); | |
1915 | if (error) | |
1916 | goto out; | |
1917 | ||
1918 | start = offset >> PAGE_CACHE_SHIFT; | |
1919 | end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
1920 | /* Try to avoid a swapstorm if len is impossible to satisfy */ | |
1921 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
1922 | error = -ENOSPC; | |
1923 | goto out; | |
83e4fa9c HD |
1924 | } |
1925 | ||
887675c9 | 1926 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
1927 | shmem_falloc.start = start; |
1928 | shmem_falloc.next = start; | |
1929 | shmem_falloc.nr_falloced = 0; | |
1930 | shmem_falloc.nr_unswapped = 0; | |
1931 | spin_lock(&inode->i_lock); | |
1932 | inode->i_private = &shmem_falloc; | |
1933 | spin_unlock(&inode->i_lock); | |
1934 | ||
e2d12e22 HD |
1935 | for (index = start; index < end; index++) { |
1936 | struct page *page; | |
1937 | ||
1938 | /* | |
1939 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
1940 | * been interrupted because we are using up too much memory. | |
1941 | */ | |
1942 | if (signal_pending(current)) | |
1943 | error = -EINTR; | |
1aac1400 HD |
1944 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
1945 | error = -ENOMEM; | |
e2d12e22 | 1946 | else |
1635f6a7 | 1947 | error = shmem_getpage(inode, index, &page, SGP_FALLOC, |
e2d12e22 HD |
1948 | NULL); |
1949 | if (error) { | |
1635f6a7 HD |
1950 | /* Remove the !PageUptodate pages we added */ |
1951 | shmem_undo_range(inode, | |
1952 | (loff_t)start << PAGE_CACHE_SHIFT, | |
6d2eb0f1 | 1953 | ((loff_t)index << PAGE_CACHE_SHIFT) - 1, true); |
1aac1400 | 1954 | goto undone; |
e2d12e22 HD |
1955 | } |
1956 | ||
1aac1400 HD |
1957 | /* |
1958 | * Inform shmem_writepage() how far we have reached. | |
1959 | * No need for lock or barrier: we have the page lock. | |
1960 | */ | |
1961 | shmem_falloc.next++; | |
1962 | if (!PageUptodate(page)) | |
1963 | shmem_falloc.nr_falloced++; | |
1964 | ||
e2d12e22 | 1965 | /* |
1635f6a7 HD |
1966 | * If !PageUptodate, leave it that way so that freeable pages |
1967 | * can be recognized if we need to rollback on error later. | |
1968 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
1969 | * than free the pages we are allocating (and SGP_CACHE pages |
1970 | * might still be clean: we now need to mark those dirty too). | |
1971 | */ | |
1972 | set_page_dirty(page); | |
1973 | unlock_page(page); | |
1974 | page_cache_release(page); | |
1975 | cond_resched(); | |
1976 | } | |
1977 | ||
1978 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
1979 | i_size_write(inode, offset + len); | |
e2d12e22 | 1980 | inode->i_ctime = CURRENT_TIME; |
1aac1400 HD |
1981 | undone: |
1982 | spin_lock(&inode->i_lock); | |
1983 | inode->i_private = NULL; | |
1984 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 1985 | out: |
83e4fa9c HD |
1986 | mutex_unlock(&inode->i_mutex); |
1987 | return error; | |
1988 | } | |
1989 | ||
726c3342 | 1990 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 1991 | { |
726c3342 | 1992 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
1993 | |
1994 | buf->f_type = TMPFS_MAGIC; | |
1995 | buf->f_bsize = PAGE_CACHE_SIZE; | |
1996 | buf->f_namelen = NAME_MAX; | |
0edd73b3 | 1997 | if (sbinfo->max_blocks) { |
1da177e4 | 1998 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
1999 | buf->f_bavail = |
2000 | buf->f_bfree = sbinfo->max_blocks - | |
2001 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2002 | } |
2003 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2004 | buf->f_files = sbinfo->max_inodes; |
2005 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2006 | } |
2007 | /* else leave those fields 0 like simple_statfs */ | |
2008 | return 0; | |
2009 | } | |
2010 | ||
2011 | /* | |
2012 | * File creation. Allocate an inode, and we're done.. | |
2013 | */ | |
2014 | static int | |
1a67aafb | 2015 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2016 | { |
0b0a0806 | 2017 | struct inode *inode; |
1da177e4 LT |
2018 | int error = -ENOSPC; |
2019 | ||
454abafe | 2020 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2021 | if (inode) { |
2a7dba39 | 2022 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2023 | &dentry->d_name, |
6d9d88d0 | 2024 | shmem_initxattrs, NULL); |
570bc1c2 SS |
2025 | if (error) { |
2026 | if (error != -EOPNOTSUPP) { | |
2027 | iput(inode); | |
2028 | return error; | |
2029 | } | |
39f0247d | 2030 | } |
1c7c474c CH |
2031 | #ifdef CONFIG_TMPFS_POSIX_ACL |
2032 | error = generic_acl_init(inode, dir); | |
39f0247d AG |
2033 | if (error) { |
2034 | iput(inode); | |
2035 | return error; | |
570bc1c2 | 2036 | } |
718deb6b AV |
2037 | #else |
2038 | error = 0; | |
1c7c474c | 2039 | #endif |
1da177e4 LT |
2040 | dir->i_size += BOGO_DIRENT_SIZE; |
2041 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2042 | d_instantiate(dentry, inode); | |
2043 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2044 | } |
2045 | return error; | |
2046 | } | |
2047 | ||
18bb1db3 | 2048 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
2049 | { |
2050 | int error; | |
2051 | ||
2052 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
2053 | return error; | |
d8c76e6f | 2054 | inc_nlink(dir); |
1da177e4 LT |
2055 | return 0; |
2056 | } | |
2057 | ||
4acdaf27 | 2058 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2059 | bool excl) |
1da177e4 LT |
2060 | { |
2061 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
2062 | } | |
2063 | ||
2064 | /* | |
2065 | * Link a file.. | |
2066 | */ | |
2067 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
2068 | { | |
2069 | struct inode *inode = old_dentry->d_inode; | |
5b04c689 | 2070 | int ret; |
1da177e4 LT |
2071 | |
2072 | /* | |
2073 | * No ordinary (disk based) filesystem counts links as inodes; | |
2074 | * but each new link needs a new dentry, pinning lowmem, and | |
2075 | * tmpfs dentries cannot be pruned until they are unlinked. | |
2076 | */ | |
5b04c689 PE |
2077 | ret = shmem_reserve_inode(inode->i_sb); |
2078 | if (ret) | |
2079 | goto out; | |
1da177e4 LT |
2080 | |
2081 | dir->i_size += BOGO_DIRENT_SIZE; | |
2082 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 2083 | inc_nlink(inode); |
7de9c6ee | 2084 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
2085 | dget(dentry); /* Extra pinning count for the created dentry */ |
2086 | d_instantiate(dentry, inode); | |
5b04c689 PE |
2087 | out: |
2088 | return ret; | |
1da177e4 LT |
2089 | } |
2090 | ||
2091 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
2092 | { | |
2093 | struct inode *inode = dentry->d_inode; | |
2094 | ||
5b04c689 PE |
2095 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
2096 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
2097 | |
2098 | dir->i_size -= BOGO_DIRENT_SIZE; | |
2099 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 2100 | drop_nlink(inode); |
1da177e4 LT |
2101 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
2102 | return 0; | |
2103 | } | |
2104 | ||
2105 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
2106 | { | |
2107 | if (!simple_empty(dentry)) | |
2108 | return -ENOTEMPTY; | |
2109 | ||
9a53c3a7 DH |
2110 | drop_nlink(dentry->d_inode); |
2111 | drop_nlink(dir); | |
1da177e4 LT |
2112 | return shmem_unlink(dir, dentry); |
2113 | } | |
2114 | ||
2115 | /* | |
2116 | * The VFS layer already does all the dentry stuff for rename, | |
2117 | * we just have to decrement the usage count for the target if | |
2118 | * it exists so that the VFS layer correctly free's it when it | |
2119 | * gets overwritten. | |
2120 | */ | |
2121 | static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) | |
2122 | { | |
2123 | struct inode *inode = old_dentry->d_inode; | |
2124 | int they_are_dirs = S_ISDIR(inode->i_mode); | |
2125 | ||
2126 | if (!simple_empty(new_dentry)) | |
2127 | return -ENOTEMPTY; | |
2128 | ||
2129 | if (new_dentry->d_inode) { | |
2130 | (void) shmem_unlink(new_dir, new_dentry); | |
474740b9 MS |
2131 | if (they_are_dirs) { |
2132 | drop_nlink(new_dentry->d_inode); | |
9a53c3a7 | 2133 | drop_nlink(old_dir); |
474740b9 | 2134 | } |
1da177e4 | 2135 | } else if (they_are_dirs) { |
9a53c3a7 | 2136 | drop_nlink(old_dir); |
d8c76e6f | 2137 | inc_nlink(new_dir); |
1da177e4 LT |
2138 | } |
2139 | ||
2140 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
2141 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
2142 | old_dir->i_ctime = old_dir->i_mtime = | |
2143 | new_dir->i_ctime = new_dir->i_mtime = | |
2144 | inode->i_ctime = CURRENT_TIME; | |
2145 | return 0; | |
2146 | } | |
2147 | ||
2148 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
2149 | { | |
2150 | int error; | |
2151 | int len; | |
2152 | struct inode *inode; | |
9276aad6 | 2153 | struct page *page; |
1da177e4 LT |
2154 | char *kaddr; |
2155 | struct shmem_inode_info *info; | |
2156 | ||
2157 | len = strlen(symname) + 1; | |
2158 | if (len > PAGE_CACHE_SIZE) | |
2159 | return -ENAMETOOLONG; | |
2160 | ||
454abafe | 2161 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
2162 | if (!inode) |
2163 | return -ENOSPC; | |
2164 | ||
9d8f13ba | 2165 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 2166 | shmem_initxattrs, NULL); |
570bc1c2 SS |
2167 | if (error) { |
2168 | if (error != -EOPNOTSUPP) { | |
2169 | iput(inode); | |
2170 | return error; | |
2171 | } | |
2172 | error = 0; | |
2173 | } | |
2174 | ||
1da177e4 LT |
2175 | info = SHMEM_I(inode); |
2176 | inode->i_size = len-1; | |
69f07ec9 HD |
2177 | if (len <= SHORT_SYMLINK_LEN) { |
2178 | info->symlink = kmemdup(symname, len, GFP_KERNEL); | |
2179 | if (!info->symlink) { | |
2180 | iput(inode); | |
2181 | return -ENOMEM; | |
2182 | } | |
2183 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 LT |
2184 | } else { |
2185 | error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); | |
2186 | if (error) { | |
2187 | iput(inode); | |
2188 | return error; | |
2189 | } | |
14fcc23f | 2190 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 2191 | inode->i_op = &shmem_symlink_inode_operations; |
9b04c5fe | 2192 | kaddr = kmap_atomic(page); |
1da177e4 | 2193 | memcpy(kaddr, symname, len); |
9b04c5fe | 2194 | kunmap_atomic(kaddr); |
ec9516fb | 2195 | SetPageUptodate(page); |
1da177e4 | 2196 | set_page_dirty(page); |
6746aff7 | 2197 | unlock_page(page); |
1da177e4 LT |
2198 | page_cache_release(page); |
2199 | } | |
1da177e4 LT |
2200 | dir->i_size += BOGO_DIRENT_SIZE; |
2201 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2202 | d_instantiate(dentry, inode); | |
2203 | dget(dentry); | |
2204 | return 0; | |
2205 | } | |
2206 | ||
69f07ec9 | 2207 | static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 2208 | { |
69f07ec9 | 2209 | nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink); |
cc314eef | 2210 | return NULL; |
1da177e4 LT |
2211 | } |
2212 | ||
cc314eef | 2213 | static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) |
1da177e4 LT |
2214 | { |
2215 | struct page *page = NULL; | |
41ffe5d5 HD |
2216 | int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); |
2217 | nd_set_link(nd, error ? ERR_PTR(error) : kmap(page)); | |
d3602444 HD |
2218 | if (page) |
2219 | unlock_page(page); | |
cc314eef | 2220 | return page; |
1da177e4 LT |
2221 | } |
2222 | ||
cc314eef | 2223 | static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 LT |
2224 | { |
2225 | if (!IS_ERR(nd_get_link(nd))) { | |
cc314eef | 2226 | struct page *page = cookie; |
1da177e4 LT |
2227 | kunmap(page); |
2228 | mark_page_accessed(page); | |
2229 | page_cache_release(page); | |
1da177e4 LT |
2230 | } |
2231 | } | |
2232 | ||
b09e0fa4 | 2233 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 2234 | /* |
b09e0fa4 EP |
2235 | * Superblocks without xattr inode operations may get some security.* xattr |
2236 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
2237 | * like ACLs, we also need to implement the security.* handlers at |
2238 | * filesystem level, though. | |
2239 | */ | |
2240 | ||
6d9d88d0 JS |
2241 | /* |
2242 | * Callback for security_inode_init_security() for acquiring xattrs. | |
2243 | */ | |
2244 | static int shmem_initxattrs(struct inode *inode, | |
2245 | const struct xattr *xattr_array, | |
2246 | void *fs_info) | |
2247 | { | |
2248 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2249 | const struct xattr *xattr; | |
38f38657 | 2250 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
2251 | size_t len; |
2252 | ||
2253 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 2254 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
2255 | if (!new_xattr) |
2256 | return -ENOMEM; | |
2257 | ||
2258 | len = strlen(xattr->name) + 1; | |
2259 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
2260 | GFP_KERNEL); | |
2261 | if (!new_xattr->name) { | |
2262 | kfree(new_xattr); | |
2263 | return -ENOMEM; | |
2264 | } | |
2265 | ||
2266 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
2267 | XATTR_SECURITY_PREFIX_LEN); | |
2268 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
2269 | xattr->name, len); | |
2270 | ||
38f38657 | 2271 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
2272 | } |
2273 | ||
2274 | return 0; | |
2275 | } | |
2276 | ||
bb435453 | 2277 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
b09e0fa4 | 2278 | #ifdef CONFIG_TMPFS_POSIX_ACL |
1c7c474c CH |
2279 | &generic_acl_access_handler, |
2280 | &generic_acl_default_handler, | |
b09e0fa4 | 2281 | #endif |
39f0247d AG |
2282 | NULL |
2283 | }; | |
b09e0fa4 EP |
2284 | |
2285 | static int shmem_xattr_validate(const char *name) | |
2286 | { | |
2287 | struct { const char *prefix; size_t len; } arr[] = { | |
2288 | { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN }, | |
2289 | { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN } | |
2290 | }; | |
2291 | int i; | |
2292 | ||
2293 | for (i = 0; i < ARRAY_SIZE(arr); i++) { | |
2294 | size_t preflen = arr[i].len; | |
2295 | if (strncmp(name, arr[i].prefix, preflen) == 0) { | |
2296 | if (!name[preflen]) | |
2297 | return -EINVAL; | |
2298 | return 0; | |
2299 | } | |
2300 | } | |
2301 | return -EOPNOTSUPP; | |
2302 | } | |
2303 | ||
2304 | static ssize_t shmem_getxattr(struct dentry *dentry, const char *name, | |
2305 | void *buffer, size_t size) | |
2306 | { | |
38f38657 | 2307 | struct shmem_inode_info *info = SHMEM_I(dentry->d_inode); |
b09e0fa4 EP |
2308 | int err; |
2309 | ||
2310 | /* | |
2311 | * If this is a request for a synthetic attribute in the system.* | |
2312 | * namespace use the generic infrastructure to resolve a handler | |
2313 | * for it via sb->s_xattr. | |
2314 | */ | |
2315 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2316 | return generic_getxattr(dentry, name, buffer, size); | |
2317 | ||
2318 | err = shmem_xattr_validate(name); | |
2319 | if (err) | |
2320 | return err; | |
2321 | ||
38f38657 | 2322 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
2323 | } |
2324 | ||
2325 | static int shmem_setxattr(struct dentry *dentry, const char *name, | |
2326 | const void *value, size_t size, int flags) | |
2327 | { | |
38f38657 | 2328 | struct shmem_inode_info *info = SHMEM_I(dentry->d_inode); |
b09e0fa4 EP |
2329 | int err; |
2330 | ||
2331 | /* | |
2332 | * If this is a request for a synthetic attribute in the system.* | |
2333 | * namespace use the generic infrastructure to resolve a handler | |
2334 | * for it via sb->s_xattr. | |
2335 | */ | |
2336 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2337 | return generic_setxattr(dentry, name, value, size, flags); | |
2338 | ||
2339 | err = shmem_xattr_validate(name); | |
2340 | if (err) | |
2341 | return err; | |
2342 | ||
38f38657 | 2343 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
2344 | } |
2345 | ||
2346 | static int shmem_removexattr(struct dentry *dentry, const char *name) | |
2347 | { | |
38f38657 | 2348 | struct shmem_inode_info *info = SHMEM_I(dentry->d_inode); |
b09e0fa4 EP |
2349 | int err; |
2350 | ||
2351 | /* | |
2352 | * If this is a request for a synthetic attribute in the system.* | |
2353 | * namespace use the generic infrastructure to resolve a handler | |
2354 | * for it via sb->s_xattr. | |
2355 | */ | |
2356 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2357 | return generic_removexattr(dentry, name); | |
2358 | ||
2359 | err = shmem_xattr_validate(name); | |
2360 | if (err) | |
2361 | return err; | |
2362 | ||
38f38657 | 2363 | return simple_xattr_remove(&info->xattrs, name); |
b09e0fa4 EP |
2364 | } |
2365 | ||
2366 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
2367 | { | |
38f38657 AR |
2368 | struct shmem_inode_info *info = SHMEM_I(dentry->d_inode); |
2369 | return simple_xattr_list(&info->xattrs, buffer, size); | |
b09e0fa4 EP |
2370 | } |
2371 | #endif /* CONFIG_TMPFS_XATTR */ | |
2372 | ||
69f07ec9 | 2373 | static const struct inode_operations shmem_short_symlink_operations = { |
b09e0fa4 | 2374 | .readlink = generic_readlink, |
69f07ec9 | 2375 | .follow_link = shmem_follow_short_symlink, |
b09e0fa4 EP |
2376 | #ifdef CONFIG_TMPFS_XATTR |
2377 | .setxattr = shmem_setxattr, | |
2378 | .getxattr = shmem_getxattr, | |
2379 | .listxattr = shmem_listxattr, | |
2380 | .removexattr = shmem_removexattr, | |
2381 | #endif | |
2382 | }; | |
2383 | ||
2384 | static const struct inode_operations shmem_symlink_inode_operations = { | |
2385 | .readlink = generic_readlink, | |
2386 | .follow_link = shmem_follow_link, | |
2387 | .put_link = shmem_put_link, | |
2388 | #ifdef CONFIG_TMPFS_XATTR | |
2389 | .setxattr = shmem_setxattr, | |
2390 | .getxattr = shmem_getxattr, | |
2391 | .listxattr = shmem_listxattr, | |
2392 | .removexattr = shmem_removexattr, | |
39f0247d | 2393 | #endif |
b09e0fa4 | 2394 | }; |
39f0247d | 2395 | |
91828a40 DG |
2396 | static struct dentry *shmem_get_parent(struct dentry *child) |
2397 | { | |
2398 | return ERR_PTR(-ESTALE); | |
2399 | } | |
2400 | ||
2401 | static int shmem_match(struct inode *ino, void *vfh) | |
2402 | { | |
2403 | __u32 *fh = vfh; | |
2404 | __u64 inum = fh[2]; | |
2405 | inum = (inum << 32) | fh[1]; | |
2406 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
2407 | } | |
2408 | ||
480b116c CH |
2409 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
2410 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 2411 | { |
91828a40 | 2412 | struct inode *inode; |
480b116c | 2413 | struct dentry *dentry = NULL; |
35c2a7f4 | 2414 | u64 inum; |
480b116c CH |
2415 | |
2416 | if (fh_len < 3) | |
2417 | return NULL; | |
91828a40 | 2418 | |
35c2a7f4 HD |
2419 | inum = fid->raw[2]; |
2420 | inum = (inum << 32) | fid->raw[1]; | |
2421 | ||
480b116c CH |
2422 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
2423 | shmem_match, fid->raw); | |
91828a40 | 2424 | if (inode) { |
480b116c | 2425 | dentry = d_find_alias(inode); |
91828a40 DG |
2426 | iput(inode); |
2427 | } | |
2428 | ||
480b116c | 2429 | return dentry; |
91828a40 DG |
2430 | } |
2431 | ||
b0b0382b AV |
2432 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
2433 | struct inode *parent) | |
91828a40 | 2434 | { |
5fe0c237 AK |
2435 | if (*len < 3) { |
2436 | *len = 3; | |
94e07a75 | 2437 | return FILEID_INVALID; |
5fe0c237 | 2438 | } |
91828a40 | 2439 | |
1d3382cb | 2440 | if (inode_unhashed(inode)) { |
91828a40 DG |
2441 | /* Unfortunately insert_inode_hash is not idempotent, |
2442 | * so as we hash inodes here rather than at creation | |
2443 | * time, we need a lock to ensure we only try | |
2444 | * to do it once | |
2445 | */ | |
2446 | static DEFINE_SPINLOCK(lock); | |
2447 | spin_lock(&lock); | |
1d3382cb | 2448 | if (inode_unhashed(inode)) |
91828a40 DG |
2449 | __insert_inode_hash(inode, |
2450 | inode->i_ino + inode->i_generation); | |
2451 | spin_unlock(&lock); | |
2452 | } | |
2453 | ||
2454 | fh[0] = inode->i_generation; | |
2455 | fh[1] = inode->i_ino; | |
2456 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
2457 | ||
2458 | *len = 3; | |
2459 | return 1; | |
2460 | } | |
2461 | ||
39655164 | 2462 | static const struct export_operations shmem_export_ops = { |
91828a40 | 2463 | .get_parent = shmem_get_parent, |
91828a40 | 2464 | .encode_fh = shmem_encode_fh, |
480b116c | 2465 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
2466 | }; |
2467 | ||
680d794b | 2468 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
2469 | bool remount) | |
1da177e4 LT |
2470 | { |
2471 | char *this_char, *value, *rest; | |
49cd0a5c | 2472 | struct mempolicy *mpol = NULL; |
8751e039 EB |
2473 | uid_t uid; |
2474 | gid_t gid; | |
1da177e4 | 2475 | |
b00dc3ad HD |
2476 | while (options != NULL) { |
2477 | this_char = options; | |
2478 | for (;;) { | |
2479 | /* | |
2480 | * NUL-terminate this option: unfortunately, | |
2481 | * mount options form a comma-separated list, | |
2482 | * but mpol's nodelist may also contain commas. | |
2483 | */ | |
2484 | options = strchr(options, ','); | |
2485 | if (options == NULL) | |
2486 | break; | |
2487 | options++; | |
2488 | if (!isdigit(*options)) { | |
2489 | options[-1] = '\0'; | |
2490 | break; | |
2491 | } | |
2492 | } | |
1da177e4 LT |
2493 | if (!*this_char) |
2494 | continue; | |
2495 | if ((value = strchr(this_char,'=')) != NULL) { | |
2496 | *value++ = 0; | |
2497 | } else { | |
2498 | printk(KERN_ERR | |
2499 | "tmpfs: No value for mount option '%s'\n", | |
2500 | this_char); | |
49cd0a5c | 2501 | goto error; |
1da177e4 LT |
2502 | } |
2503 | ||
2504 | if (!strcmp(this_char,"size")) { | |
2505 | unsigned long long size; | |
2506 | size = memparse(value,&rest); | |
2507 | if (*rest == '%') { | |
2508 | size <<= PAGE_SHIFT; | |
2509 | size *= totalram_pages; | |
2510 | do_div(size, 100); | |
2511 | rest++; | |
2512 | } | |
2513 | if (*rest) | |
2514 | goto bad_val; | |
680d794b | 2515 | sbinfo->max_blocks = |
2516 | DIV_ROUND_UP(size, PAGE_CACHE_SIZE); | |
1da177e4 | 2517 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 2518 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
2519 | if (*rest) |
2520 | goto bad_val; | |
2521 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 2522 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
2523 | if (*rest) |
2524 | goto bad_val; | |
2525 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 2526 | if (remount) |
1da177e4 | 2527 | continue; |
680d794b | 2528 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
2529 | if (*rest) |
2530 | goto bad_val; | |
2531 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 2532 | if (remount) |
1da177e4 | 2533 | continue; |
8751e039 | 2534 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2535 | if (*rest) |
2536 | goto bad_val; | |
8751e039 EB |
2537 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
2538 | if (!uid_valid(sbinfo->uid)) | |
2539 | goto bad_val; | |
1da177e4 | 2540 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 2541 | if (remount) |
1da177e4 | 2542 | continue; |
8751e039 | 2543 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2544 | if (*rest) |
2545 | goto bad_val; | |
8751e039 EB |
2546 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
2547 | if (!gid_valid(sbinfo->gid)) | |
2548 | goto bad_val; | |
7339ff83 | 2549 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
2550 | mpol_put(mpol); |
2551 | mpol = NULL; | |
2552 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 2553 | goto bad_val; |
1da177e4 LT |
2554 | } else { |
2555 | printk(KERN_ERR "tmpfs: Bad mount option %s\n", | |
2556 | this_char); | |
49cd0a5c | 2557 | goto error; |
1da177e4 LT |
2558 | } |
2559 | } | |
49cd0a5c | 2560 | sbinfo->mpol = mpol; |
1da177e4 LT |
2561 | return 0; |
2562 | ||
2563 | bad_val: | |
2564 | printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", | |
2565 | value, this_char); | |
49cd0a5c GT |
2566 | error: |
2567 | mpol_put(mpol); | |
1da177e4 LT |
2568 | return 1; |
2569 | ||
2570 | } | |
2571 | ||
2572 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
2573 | { | |
2574 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 2575 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
2576 | unsigned long inodes; |
2577 | int error = -EINVAL; | |
2578 | ||
5f00110f | 2579 | config.mpol = NULL; |
680d794b | 2580 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 2581 | return error; |
1da177e4 | 2582 | |
0edd73b3 | 2583 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 2584 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 2585 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 2586 | goto out; |
680d794b | 2587 | if (config.max_inodes < inodes) |
0edd73b3 HD |
2588 | goto out; |
2589 | /* | |
54af6042 | 2590 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
2591 | * but we must separately disallow unlimited->limited, because |
2592 | * in that case we have no record of how much is already in use. | |
2593 | */ | |
680d794b | 2594 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 2595 | goto out; |
680d794b | 2596 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
2597 | goto out; |
2598 | ||
2599 | error = 0; | |
680d794b | 2600 | sbinfo->max_blocks = config.max_blocks; |
680d794b | 2601 | sbinfo->max_inodes = config.max_inodes; |
2602 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 2603 | |
5f00110f GT |
2604 | /* |
2605 | * Preserve previous mempolicy unless mpol remount option was specified. | |
2606 | */ | |
2607 | if (config.mpol) { | |
2608 | mpol_put(sbinfo->mpol); | |
2609 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
2610 | } | |
0edd73b3 HD |
2611 | out: |
2612 | spin_unlock(&sbinfo->stat_lock); | |
2613 | return error; | |
1da177e4 | 2614 | } |
680d794b | 2615 | |
34c80b1d | 2616 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 2617 | { |
34c80b1d | 2618 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b | 2619 | |
2620 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
2621 | seq_printf(seq, ",size=%luk", | |
2622 | sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); | |
2623 | if (sbinfo->max_inodes != shmem_default_max_inodes()) | |
2624 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
2625 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 2626 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
2627 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
2628 | seq_printf(seq, ",uid=%u", | |
2629 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
2630 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
2631 | seq_printf(seq, ",gid=%u", | |
2632 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
71fe804b | 2633 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b | 2634 | return 0; |
2635 | } | |
2636 | #endif /* CONFIG_TMPFS */ | |
1da177e4 LT |
2637 | |
2638 | static void shmem_put_super(struct super_block *sb) | |
2639 | { | |
602586a8 HD |
2640 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
2641 | ||
2642 | percpu_counter_destroy(&sbinfo->used_blocks); | |
49cd0a5c | 2643 | mpol_put(sbinfo->mpol); |
602586a8 | 2644 | kfree(sbinfo); |
1da177e4 LT |
2645 | sb->s_fs_info = NULL; |
2646 | } | |
2647 | ||
2b2af54a | 2648 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
2649 | { |
2650 | struct inode *inode; | |
0edd73b3 | 2651 | struct shmem_sb_info *sbinfo; |
680d794b | 2652 | int err = -ENOMEM; |
2653 | ||
2654 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 2655 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b | 2656 | L1_CACHE_BYTES), GFP_KERNEL); |
2657 | if (!sbinfo) | |
2658 | return -ENOMEM; | |
2659 | ||
680d794b | 2660 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
2661 | sbinfo->uid = current_fsuid(); |
2662 | sbinfo->gid = current_fsgid(); | |
680d794b | 2663 | sb->s_fs_info = sbinfo; |
1da177e4 | 2664 | |
0edd73b3 | 2665 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
2666 | /* |
2667 | * Per default we only allow half of the physical ram per | |
2668 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
2669 | * but the internal instance is left unlimited. | |
2670 | */ | |
2671 | if (!(sb->s_flags & MS_NOUSER)) { | |
680d794b | 2672 | sbinfo->max_blocks = shmem_default_max_blocks(); |
2673 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
2674 | if (shmem_parse_options(data, sbinfo, false)) { | |
2675 | err = -EINVAL; | |
2676 | goto failed; | |
2677 | } | |
1da177e4 | 2678 | } |
91828a40 | 2679 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 2680 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
2681 | #else |
2682 | sb->s_flags |= MS_NOUSER; | |
2683 | #endif | |
2684 | ||
0edd73b3 | 2685 | spin_lock_init(&sbinfo->stat_lock); |
602586a8 HD |
2686 | if (percpu_counter_init(&sbinfo->used_blocks, 0)) |
2687 | goto failed; | |
680d794b | 2688 | sbinfo->free_inodes = sbinfo->max_inodes; |
0edd73b3 | 2689 | |
285b2c4f | 2690 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
1da177e4 LT |
2691 | sb->s_blocksize = PAGE_CACHE_SIZE; |
2692 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
2693 | sb->s_magic = TMPFS_MAGIC; | |
2694 | sb->s_op = &shmem_ops; | |
cfd95a9c | 2695 | sb->s_time_gran = 1; |
b09e0fa4 | 2696 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 2697 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
2698 | #endif |
2699 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
2700 | sb->s_flags |= MS_POSIXACL; |
2701 | #endif | |
0edd73b3 | 2702 | |
454abafe | 2703 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
2704 | if (!inode) |
2705 | goto failed; | |
680d794b | 2706 | inode->i_uid = sbinfo->uid; |
2707 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
2708 | sb->s_root = d_make_root(inode); |
2709 | if (!sb->s_root) | |
48fde701 | 2710 | goto failed; |
1da177e4 LT |
2711 | return 0; |
2712 | ||
1da177e4 LT |
2713 | failed: |
2714 | shmem_put_super(sb); | |
2715 | return err; | |
2716 | } | |
2717 | ||
fcc234f8 | 2718 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
2719 | |
2720 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
2721 | { | |
41ffe5d5 HD |
2722 | struct shmem_inode_info *info; |
2723 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
2724 | if (!info) | |
1da177e4 | 2725 | return NULL; |
41ffe5d5 | 2726 | return &info->vfs_inode; |
1da177e4 LT |
2727 | } |
2728 | ||
41ffe5d5 | 2729 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
2730 | { |
2731 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
2732 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
2733 | } | |
2734 | ||
1da177e4 LT |
2735 | static void shmem_destroy_inode(struct inode *inode) |
2736 | { | |
09208d15 | 2737 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 2738 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 2739 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
2740 | } |
2741 | ||
41ffe5d5 | 2742 | static void shmem_init_inode(void *foo) |
1da177e4 | 2743 | { |
41ffe5d5 HD |
2744 | struct shmem_inode_info *info = foo; |
2745 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
2746 | } |
2747 | ||
41ffe5d5 | 2748 | static int shmem_init_inodecache(void) |
1da177e4 LT |
2749 | { |
2750 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
2751 | sizeof(struct shmem_inode_info), | |
41ffe5d5 | 2752 | 0, SLAB_PANIC, shmem_init_inode); |
1da177e4 LT |
2753 | return 0; |
2754 | } | |
2755 | ||
41ffe5d5 | 2756 | static void shmem_destroy_inodecache(void) |
1da177e4 | 2757 | { |
1a1d92c1 | 2758 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
2759 | } |
2760 | ||
f5e54d6e | 2761 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 2762 | .writepage = shmem_writepage, |
76719325 | 2763 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 2764 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
2765 | .write_begin = shmem_write_begin, |
2766 | .write_end = shmem_write_end, | |
1da177e4 | 2767 | #endif |
304dbdb7 | 2768 | .migratepage = migrate_page, |
aa261f54 | 2769 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
2770 | }; |
2771 | ||
15ad7cdc | 2772 | static const struct file_operations shmem_file_operations = { |
1da177e4 LT |
2773 | .mmap = shmem_mmap, |
2774 | #ifdef CONFIG_TMPFS | |
220f2ac9 | 2775 | .llseek = shmem_file_llseek, |
bcd78e49 | 2776 | .read = do_sync_read, |
5402b976 | 2777 | .write = do_sync_write, |
bcd78e49 | 2778 | .aio_read = shmem_file_aio_read, |
5402b976 | 2779 | .aio_write = generic_file_aio_write, |
1b061d92 | 2780 | .fsync = noop_fsync, |
708e3508 | 2781 | .splice_read = shmem_file_splice_read, |
ae976416 | 2782 | .splice_write = generic_file_splice_write, |
83e4fa9c | 2783 | .fallocate = shmem_fallocate, |
1da177e4 LT |
2784 | #endif |
2785 | }; | |
2786 | ||
92e1d5be | 2787 | static const struct inode_operations shmem_inode_operations = { |
94c1e62d | 2788 | .setattr = shmem_setattr, |
b09e0fa4 EP |
2789 | #ifdef CONFIG_TMPFS_XATTR |
2790 | .setxattr = shmem_setxattr, | |
2791 | .getxattr = shmem_getxattr, | |
2792 | .listxattr = shmem_listxattr, | |
2793 | .removexattr = shmem_removexattr, | |
2794 | #endif | |
1da177e4 LT |
2795 | }; |
2796 | ||
92e1d5be | 2797 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
2798 | #ifdef CONFIG_TMPFS |
2799 | .create = shmem_create, | |
2800 | .lookup = simple_lookup, | |
2801 | .link = shmem_link, | |
2802 | .unlink = shmem_unlink, | |
2803 | .symlink = shmem_symlink, | |
2804 | .mkdir = shmem_mkdir, | |
2805 | .rmdir = shmem_rmdir, | |
2806 | .mknod = shmem_mknod, | |
2807 | .rename = shmem_rename, | |
1da177e4 | 2808 | #endif |
b09e0fa4 EP |
2809 | #ifdef CONFIG_TMPFS_XATTR |
2810 | .setxattr = shmem_setxattr, | |
2811 | .getxattr = shmem_getxattr, | |
2812 | .listxattr = shmem_listxattr, | |
2813 | .removexattr = shmem_removexattr, | |
2814 | #endif | |
39f0247d | 2815 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 2816 | .setattr = shmem_setattr, |
39f0247d AG |
2817 | #endif |
2818 | }; | |
2819 | ||
92e1d5be | 2820 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 EP |
2821 | #ifdef CONFIG_TMPFS_XATTR |
2822 | .setxattr = shmem_setxattr, | |
2823 | .getxattr = shmem_getxattr, | |
2824 | .listxattr = shmem_listxattr, | |
2825 | .removexattr = shmem_removexattr, | |
2826 | #endif | |
39f0247d | 2827 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 2828 | .setattr = shmem_setattr, |
39f0247d | 2829 | #endif |
1da177e4 LT |
2830 | }; |
2831 | ||
759b9775 | 2832 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
2833 | .alloc_inode = shmem_alloc_inode, |
2834 | .destroy_inode = shmem_destroy_inode, | |
2835 | #ifdef CONFIG_TMPFS | |
2836 | .statfs = shmem_statfs, | |
2837 | .remount_fs = shmem_remount_fs, | |
680d794b | 2838 | .show_options = shmem_show_options, |
1da177e4 | 2839 | #endif |
1f895f75 | 2840 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
2841 | .drop_inode = generic_delete_inode, |
2842 | .put_super = shmem_put_super, | |
2843 | }; | |
2844 | ||
f0f37e2f | 2845 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 2846 | .fault = shmem_fault, |
1da177e4 LT |
2847 | #ifdef CONFIG_NUMA |
2848 | .set_policy = shmem_set_policy, | |
2849 | .get_policy = shmem_get_policy, | |
2850 | #endif | |
0b173bc4 | 2851 | .remap_pages = generic_file_remap_pages, |
1da177e4 LT |
2852 | }; |
2853 | ||
3c26ff6e AV |
2854 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
2855 | int flags, const char *dev_name, void *data) | |
1da177e4 | 2856 | { |
3c26ff6e | 2857 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
2858 | } |
2859 | ||
41ffe5d5 | 2860 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
2861 | .owner = THIS_MODULE, |
2862 | .name = "tmpfs", | |
3c26ff6e | 2863 | .mount = shmem_mount, |
1da177e4 | 2864 | .kill_sb = kill_litter_super, |
2b8576cb | 2865 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 2866 | }; |
1da177e4 | 2867 | |
41ffe5d5 | 2868 | int __init shmem_init(void) |
1da177e4 LT |
2869 | { |
2870 | int error; | |
2871 | ||
e0bf68dd PZ |
2872 | error = bdi_init(&shmem_backing_dev_info); |
2873 | if (error) | |
2874 | goto out4; | |
2875 | ||
41ffe5d5 | 2876 | error = shmem_init_inodecache(); |
1da177e4 LT |
2877 | if (error) |
2878 | goto out3; | |
2879 | ||
41ffe5d5 | 2880 | error = register_filesystem(&shmem_fs_type); |
1da177e4 LT |
2881 | if (error) { |
2882 | printk(KERN_ERR "Could not register tmpfs\n"); | |
2883 | goto out2; | |
2884 | } | |
95dc112a | 2885 | |
41ffe5d5 HD |
2886 | shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER, |
2887 | shmem_fs_type.name, NULL); | |
1da177e4 LT |
2888 | if (IS_ERR(shm_mnt)) { |
2889 | error = PTR_ERR(shm_mnt); | |
2890 | printk(KERN_ERR "Could not kern_mount tmpfs\n"); | |
2891 | goto out1; | |
2892 | } | |
2893 | return 0; | |
2894 | ||
2895 | out1: | |
41ffe5d5 | 2896 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 2897 | out2: |
41ffe5d5 | 2898 | shmem_destroy_inodecache(); |
1da177e4 | 2899 | out3: |
e0bf68dd PZ |
2900 | bdi_destroy(&shmem_backing_dev_info); |
2901 | out4: | |
1da177e4 LT |
2902 | shm_mnt = ERR_PTR(error); |
2903 | return error; | |
2904 | } | |
853ac43a MM |
2905 | |
2906 | #else /* !CONFIG_SHMEM */ | |
2907 | ||
2908 | /* | |
2909 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
2910 | * | |
2911 | * This is intended for small system where the benefits of the full | |
2912 | * shmem code (swap-backed and resource-limited) are outweighed by | |
2913 | * their complexity. On systems without swap this code should be | |
2914 | * effectively equivalent, but much lighter weight. | |
2915 | */ | |
2916 | ||
41ffe5d5 | 2917 | static struct file_system_type shmem_fs_type = { |
853ac43a | 2918 | .name = "tmpfs", |
3c26ff6e | 2919 | .mount = ramfs_mount, |
853ac43a | 2920 | .kill_sb = kill_litter_super, |
2b8576cb | 2921 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
2922 | }; |
2923 | ||
41ffe5d5 | 2924 | int __init shmem_init(void) |
853ac43a | 2925 | { |
41ffe5d5 | 2926 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 2927 | |
41ffe5d5 | 2928 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
2929 | BUG_ON(IS_ERR(shm_mnt)); |
2930 | ||
2931 | return 0; | |
2932 | } | |
2933 | ||
41ffe5d5 | 2934 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
2935 | { |
2936 | return 0; | |
2937 | } | |
2938 | ||
3f96b79a HD |
2939 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
2940 | { | |
2941 | return 0; | |
2942 | } | |
2943 | ||
24513264 HD |
2944 | void shmem_unlock_mapping(struct address_space *mapping) |
2945 | { | |
2946 | } | |
2947 | ||
41ffe5d5 | 2948 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 2949 | { |
41ffe5d5 | 2950 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
2951 | } |
2952 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
2953 | ||
0b0a0806 HD |
2954 | #define shmem_vm_ops generic_file_vm_ops |
2955 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 2956 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
2957 | #define shmem_acct_size(flags, size) 0 |
2958 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
2959 | |
2960 | #endif /* CONFIG_SHMEM */ | |
2961 | ||
2962 | /* common code */ | |
1da177e4 | 2963 | |
3451538a | 2964 | static struct dentry_operations anon_ops = { |
ad4c3cc4 | 2965 | .d_dname = simple_dname |
3451538a AV |
2966 | }; |
2967 | ||
46711810 | 2968 | /** |
1da177e4 | 2969 | * shmem_file_setup - get an unlinked file living in tmpfs |
1da177e4 LT |
2970 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
2971 | * @size: size to be set for the file | |
0b0a0806 | 2972 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size |
1da177e4 | 2973 | */ |
168f5ac6 | 2974 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) |
1da177e4 | 2975 | { |
6b4d0b27 | 2976 | struct file *res; |
1da177e4 | 2977 | struct inode *inode; |
2c48b9c4 | 2978 | struct path path; |
3451538a | 2979 | struct super_block *sb; |
1da177e4 LT |
2980 | struct qstr this; |
2981 | ||
2982 | if (IS_ERR(shm_mnt)) | |
6b4d0b27 | 2983 | return ERR_CAST(shm_mnt); |
1da177e4 | 2984 | |
285b2c4f | 2985 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
2986 | return ERR_PTR(-EINVAL); |
2987 | ||
2988 | if (shmem_acct_size(flags, size)) | |
2989 | return ERR_PTR(-ENOMEM); | |
2990 | ||
6b4d0b27 | 2991 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
2992 | this.name = name; |
2993 | this.len = strlen(name); | |
2994 | this.hash = 0; /* will go */ | |
3451538a AV |
2995 | sb = shm_mnt->mnt_sb; |
2996 | path.dentry = d_alloc_pseudo(sb, &this); | |
2c48b9c4 | 2997 | if (!path.dentry) |
1da177e4 | 2998 | goto put_memory; |
3451538a | 2999 | d_set_d_op(path.dentry, &anon_ops); |
2c48b9c4 | 3000 | path.mnt = mntget(shm_mnt); |
1da177e4 | 3001 | |
6b4d0b27 | 3002 | res = ERR_PTR(-ENOSPC); |
3451538a | 3003 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 3004 | if (!inode) |
4b42af81 | 3005 | goto put_dentry; |
1da177e4 | 3006 | |
2c48b9c4 | 3007 | d_instantiate(path.dentry, inode); |
1da177e4 | 3008 | inode->i_size = size; |
6d6b77f1 | 3009 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
3010 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
3011 | if (IS_ERR(res)) | |
4b42af81 | 3012 | goto put_dentry; |
4b42af81 | 3013 | |
6b4d0b27 | 3014 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 3015 | &shmem_file_operations); |
6b4d0b27 | 3016 | if (IS_ERR(res)) |
4b42af81 AV |
3017 | goto put_dentry; |
3018 | ||
6b4d0b27 | 3019 | return res; |
1da177e4 | 3020 | |
1da177e4 | 3021 | put_dentry: |
2c48b9c4 | 3022 | path_put(&path); |
1da177e4 LT |
3023 | put_memory: |
3024 | shmem_unacct_size(flags, size); | |
6b4d0b27 | 3025 | return res; |
1da177e4 | 3026 | } |
395e0ddc | 3027 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 3028 | |
46711810 | 3029 | /** |
1da177e4 | 3030 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
3031 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
3032 | */ | |
3033 | int shmem_zero_setup(struct vm_area_struct *vma) | |
3034 | { | |
3035 | struct file *file; | |
3036 | loff_t size = vma->vm_end - vma->vm_start; | |
3037 | ||
3038 | file = shmem_file_setup("dev/zero", size, vma->vm_flags); | |
3039 | if (IS_ERR(file)) | |
3040 | return PTR_ERR(file); | |
3041 | ||
3042 | if (vma->vm_file) | |
3043 | fput(vma->vm_file); | |
3044 | vma->vm_file = file; | |
3045 | vma->vm_ops = &shmem_vm_ops; | |
3046 | return 0; | |
3047 | } | |
d9d90e5e HD |
3048 | |
3049 | /** | |
3050 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
3051 | * @mapping: the page's address_space | |
3052 | * @index: the page index | |
3053 | * @gfp: the page allocator flags to use if allocating | |
3054 | * | |
3055 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
3056 | * with any new page allocations done using the specified allocation flags. | |
3057 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
3058 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
3059 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
3060 | * | |
68da9f05 HD |
3061 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
3062 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
3063 | */ |
3064 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
3065 | pgoff_t index, gfp_t gfp) | |
3066 | { | |
68da9f05 HD |
3067 | #ifdef CONFIG_SHMEM |
3068 | struct inode *inode = mapping->host; | |
9276aad6 | 3069 | struct page *page; |
68da9f05 HD |
3070 | int error; |
3071 | ||
3072 | BUG_ON(mapping->a_ops != &shmem_aops); | |
3073 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL); | |
3074 | if (error) | |
3075 | page = ERR_PTR(error); | |
3076 | else | |
3077 | unlock_page(page); | |
3078 | return page; | |
3079 | #else | |
3080 | /* | |
3081 | * The tiny !SHMEM case uses ramfs without swap | |
3082 | */ | |
d9d90e5e | 3083 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 3084 | #endif |
d9d90e5e HD |
3085 | } |
3086 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |