]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Resizable virtual memory filesystem for Linux. | |
3 | * | |
4 | * Copyright (C) 2000 Linus Torvalds. | |
5 | * 2000 Transmeta Corp. | |
6 | * 2000-2001 Christoph Rohland | |
7 | * 2000-2001 SAP AG | |
8 | * 2002 Red Hat Inc. | |
6922c0c7 HD |
9 | * Copyright (C) 2002-2011 Hugh Dickins. |
10 | * Copyright (C) 2011 Google Inc. | |
0edd73b3 | 11 | * Copyright (C) 2002-2005 VERITAS Software Corporation. |
1da177e4 LT |
12 | * Copyright (C) 2004 Andi Kleen, SuSE Labs |
13 | * | |
14 | * Extended attribute support for tmpfs: | |
15 | * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net> | |
16 | * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com> | |
17 | * | |
853ac43a MM |
18 | * tiny-shmem: |
19 | * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com> | |
20 | * | |
1da177e4 LT |
21 | * This file is released under the GPL. |
22 | */ | |
23 | ||
853ac43a MM |
24 | #include <linux/fs.h> |
25 | #include <linux/init.h> | |
26 | #include <linux/vfs.h> | |
27 | #include <linux/mount.h> | |
caefba17 | 28 | #include <linux/pagemap.h> |
853ac43a MM |
29 | #include <linux/file.h> |
30 | #include <linux/mm.h> | |
b95f1b31 | 31 | #include <linux/export.h> |
853ac43a MM |
32 | #include <linux/swap.h> |
33 | ||
34 | static struct vfsmount *shm_mnt; | |
35 | ||
36 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
37 | /* |
38 | * This virtual memory filesystem is heavily based on the ramfs. It | |
39 | * extends ramfs by the ability to use swap and honor resource limits | |
40 | * which makes it a completely usable filesystem. | |
41 | */ | |
42 | ||
39f0247d | 43 | #include <linux/xattr.h> |
a5694255 | 44 | #include <linux/exportfs.h> |
1c7c474c | 45 | #include <linux/posix_acl.h> |
39f0247d | 46 | #include <linux/generic_acl.h> |
1da177e4 | 47 | #include <linux/mman.h> |
1da177e4 LT |
48 | #include <linux/string.h> |
49 | #include <linux/slab.h> | |
50 | #include <linux/backing-dev.h> | |
51 | #include <linux/shmem_fs.h> | |
1da177e4 | 52 | #include <linux/writeback.h> |
1da177e4 | 53 | #include <linux/blkdev.h> |
bda97eab | 54 | #include <linux/pagevec.h> |
41ffe5d5 | 55 | #include <linux/percpu_counter.h> |
83e4fa9c | 56 | #include <linux/falloc.h> |
708e3508 | 57 | #include <linux/splice.h> |
1da177e4 LT |
58 | #include <linux/security.h> |
59 | #include <linux/swapops.h> | |
60 | #include <linux/mempolicy.h> | |
61 | #include <linux/namei.h> | |
b00dc3ad | 62 | #include <linux/ctype.h> |
304dbdb7 | 63 | #include <linux/migrate.h> |
c1f60a5a | 64 | #include <linux/highmem.h> |
680d794b | 65 | #include <linux/seq_file.h> |
92562927 | 66 | #include <linux/magic.h> |
304dbdb7 | 67 | |
1da177e4 | 68 | #include <asm/uaccess.h> |
1da177e4 LT |
69 | #include <asm/pgtable.h> |
70 | ||
caefba17 | 71 | #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) |
1da177e4 LT |
72 | #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) |
73 | ||
1da177e4 LT |
74 | /* Pretend that each entry is of this size in directory's i_size */ |
75 | #define BOGO_DIRENT_SIZE 20 | |
76 | ||
69f07ec9 HD |
77 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
78 | #define SHORT_SYMLINK_LEN 128 | |
79 | ||
b09e0fa4 EP |
80 | struct shmem_xattr { |
81 | struct list_head list; /* anchored by shmem_inode_info->xattr_list */ | |
82 | char *name; /* xattr name */ | |
83 | size_t size; | |
84 | char value[0]; | |
85 | }; | |
86 | ||
1aac1400 HD |
87 | /* |
88 | * shmem_fallocate and shmem_writepage communicate via inode->i_private | |
89 | * (with i_mutex making sure that it has only one user at a time): | |
90 | * we would prefer not to enlarge the shmem inode just for that. | |
91 | */ | |
92 | struct shmem_falloc { | |
93 | pgoff_t start; /* start of range currently being fallocated */ | |
94 | pgoff_t next; /* the next page offset to be fallocated */ | |
95 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
96 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
97 | }; | |
98 | ||
285b2c4f | 99 | /* Flag allocation requirements to shmem_getpage */ |
1da177e4 | 100 | enum sgp_type { |
1da177e4 LT |
101 | SGP_READ, /* don't exceed i_size, don't allocate page */ |
102 | SGP_CACHE, /* don't exceed i_size, may allocate page */ | |
a0ee5ec5 | 103 | SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ |
1635f6a7 HD |
104 | SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */ |
105 | SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */ | |
1da177e4 LT |
106 | }; |
107 | ||
b76db735 | 108 | #ifdef CONFIG_TMPFS |
680d794b AM |
109 | static unsigned long shmem_default_max_blocks(void) |
110 | { | |
111 | return totalram_pages / 2; | |
112 | } | |
113 | ||
114 | static unsigned long shmem_default_max_inodes(void) | |
115 | { | |
116 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
117 | } | |
b76db735 | 118 | #endif |
680d794b | 119 | |
bde05d1c HD |
120 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
121 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
122 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 HD |
123 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
124 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type); | |
125 | ||
126 | static inline int shmem_getpage(struct inode *inode, pgoff_t index, | |
127 | struct page **pagep, enum sgp_type sgp, int *fault_type) | |
128 | { | |
129 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
130 | mapping_gfp_mask(inode->i_mapping), fault_type); | |
131 | } | |
1da177e4 | 132 | |
1da177e4 LT |
133 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
134 | { | |
135 | return sb->s_fs_info; | |
136 | } | |
137 | ||
138 | /* | |
139 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
140 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
141 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
142 | * consistent with the pre-accounting of private mappings ... | |
143 | */ | |
144 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
145 | { | |
0b0a0806 | 146 | return (flags & VM_NORESERVE) ? |
191c5424 | 147 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
148 | } |
149 | ||
150 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
151 | { | |
0b0a0806 | 152 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
153 | vm_unacct_memory(VM_ACCT(size)); |
154 | } | |
155 | ||
156 | /* | |
157 | * ... whereas tmpfs objects are accounted incrementally as | |
158 | * pages are allocated, in order to allow huge sparse files. | |
159 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, | |
160 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
161 | */ | |
162 | static inline int shmem_acct_block(unsigned long flags) | |
163 | { | |
0b0a0806 | 164 | return (flags & VM_NORESERVE) ? |
191c5424 | 165 | security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0; |
1da177e4 LT |
166 | } |
167 | ||
168 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
169 | { | |
0b0a0806 | 170 | if (flags & VM_NORESERVE) |
1da177e4 LT |
171 | vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); |
172 | } | |
173 | ||
759b9775 | 174 | static const struct super_operations shmem_ops; |
f5e54d6e | 175 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 176 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
177 | static const struct inode_operations shmem_inode_operations; |
178 | static const struct inode_operations shmem_dir_inode_operations; | |
179 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 180 | static const struct vm_operations_struct shmem_vm_ops; |
1da177e4 | 181 | |
6c231b7b | 182 | static struct backing_dev_info shmem_backing_dev_info __read_mostly = { |
1da177e4 | 183 | .ra_pages = 0, /* No readahead */ |
4f98a2fe | 184 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED, |
1da177e4 LT |
185 | }; |
186 | ||
187 | static LIST_HEAD(shmem_swaplist); | |
cb5f7b9a | 188 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 189 | |
5b04c689 PE |
190 | static int shmem_reserve_inode(struct super_block *sb) |
191 | { | |
192 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
193 | if (sbinfo->max_inodes) { | |
194 | spin_lock(&sbinfo->stat_lock); | |
195 | if (!sbinfo->free_inodes) { | |
196 | spin_unlock(&sbinfo->stat_lock); | |
197 | return -ENOSPC; | |
198 | } | |
199 | sbinfo->free_inodes--; | |
200 | spin_unlock(&sbinfo->stat_lock); | |
201 | } | |
202 | return 0; | |
203 | } | |
204 | ||
205 | static void shmem_free_inode(struct super_block *sb) | |
206 | { | |
207 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
208 | if (sbinfo->max_inodes) { | |
209 | spin_lock(&sbinfo->stat_lock); | |
210 | sbinfo->free_inodes++; | |
211 | spin_unlock(&sbinfo->stat_lock); | |
212 | } | |
213 | } | |
214 | ||
46711810 | 215 | /** |
41ffe5d5 | 216 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
217 | * @inode: inode to recalc |
218 | * | |
219 | * We have to calculate the free blocks since the mm can drop | |
220 | * undirtied hole pages behind our back. | |
221 | * | |
222 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
223 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
224 | * | |
225 | * It has to be called with the spinlock held. | |
226 | */ | |
227 | static void shmem_recalc_inode(struct inode *inode) | |
228 | { | |
229 | struct shmem_inode_info *info = SHMEM_I(inode); | |
230 | long freed; | |
231 | ||
232 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
233 | if (freed > 0) { | |
54af6042 HD |
234 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
235 | if (sbinfo->max_blocks) | |
236 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 237 | info->alloced -= freed; |
54af6042 | 238 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 239 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
240 | } |
241 | } | |
242 | ||
7a5d0fbb HD |
243 | /* |
244 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
245 | */ | |
246 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
247 | pgoff_t index, void *expected, void *replacement) | |
248 | { | |
249 | void **pslot; | |
250 | void *item = NULL; | |
251 | ||
252 | VM_BUG_ON(!expected); | |
253 | pslot = radix_tree_lookup_slot(&mapping->page_tree, index); | |
254 | if (pslot) | |
255 | item = radix_tree_deref_slot_protected(pslot, | |
256 | &mapping->tree_lock); | |
257 | if (item != expected) | |
258 | return -ENOENT; | |
259 | if (replacement) | |
260 | radix_tree_replace_slot(pslot, replacement); | |
261 | else | |
262 | radix_tree_delete(&mapping->page_tree, index); | |
263 | return 0; | |
264 | } | |
265 | ||
46f65ec1 HD |
266 | /* |
267 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
268 | */ | |
269 | static int shmem_add_to_page_cache(struct page *page, | |
270 | struct address_space *mapping, | |
271 | pgoff_t index, gfp_t gfp, void *expected) | |
272 | { | |
aa3b1895 | 273 | int error = 0; |
46f65ec1 HD |
274 | |
275 | VM_BUG_ON(!PageLocked(page)); | |
276 | VM_BUG_ON(!PageSwapBacked(page)); | |
277 | ||
46f65ec1 HD |
278 | if (!expected) |
279 | error = radix_tree_preload(gfp & GFP_RECLAIM_MASK); | |
280 | if (!error) { | |
281 | page_cache_get(page); | |
282 | page->mapping = mapping; | |
283 | page->index = index; | |
284 | ||
285 | spin_lock_irq(&mapping->tree_lock); | |
286 | if (!expected) | |
287 | error = radix_tree_insert(&mapping->page_tree, | |
288 | index, page); | |
289 | else | |
290 | error = shmem_radix_tree_replace(mapping, index, | |
291 | expected, page); | |
292 | if (!error) { | |
293 | mapping->nrpages++; | |
294 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
295 | __inc_zone_page_state(page, NR_SHMEM); | |
296 | spin_unlock_irq(&mapping->tree_lock); | |
297 | } else { | |
298 | page->mapping = NULL; | |
299 | spin_unlock_irq(&mapping->tree_lock); | |
300 | page_cache_release(page); | |
301 | } | |
302 | if (!expected) | |
303 | radix_tree_preload_end(); | |
304 | } | |
305 | if (error) | |
306 | mem_cgroup_uncharge_cache_page(page); | |
46f65ec1 HD |
307 | return error; |
308 | } | |
309 | ||
6922c0c7 HD |
310 | /* |
311 | * Like delete_from_page_cache, but substitutes swap for page. | |
312 | */ | |
313 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
314 | { | |
315 | struct address_space *mapping = page->mapping; | |
316 | int error; | |
317 | ||
318 | spin_lock_irq(&mapping->tree_lock); | |
319 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
320 | page->mapping = NULL; | |
321 | mapping->nrpages--; | |
322 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
323 | __dec_zone_page_state(page, NR_SHMEM); | |
324 | spin_unlock_irq(&mapping->tree_lock); | |
325 | page_cache_release(page); | |
326 | BUG_ON(error); | |
327 | } | |
328 | ||
7a5d0fbb HD |
329 | /* |
330 | * Like find_get_pages, but collecting swap entries as well as pages. | |
331 | */ | |
332 | static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping, | |
333 | pgoff_t start, unsigned int nr_pages, | |
334 | struct page **pages, pgoff_t *indices) | |
335 | { | |
336 | unsigned int i; | |
337 | unsigned int ret; | |
338 | unsigned int nr_found; | |
339 | ||
340 | rcu_read_lock(); | |
341 | restart: | |
342 | nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, | |
343 | (void ***)pages, indices, start, nr_pages); | |
344 | ret = 0; | |
345 | for (i = 0; i < nr_found; i++) { | |
346 | struct page *page; | |
347 | repeat: | |
348 | page = radix_tree_deref_slot((void **)pages[i]); | |
349 | if (unlikely(!page)) | |
350 | continue; | |
351 | if (radix_tree_exception(page)) { | |
8079b1c8 HD |
352 | if (radix_tree_deref_retry(page)) |
353 | goto restart; | |
354 | /* | |
355 | * Otherwise, we must be storing a swap entry | |
356 | * here as an exceptional entry: so return it | |
357 | * without attempting to raise page count. | |
358 | */ | |
359 | goto export; | |
7a5d0fbb HD |
360 | } |
361 | if (!page_cache_get_speculative(page)) | |
362 | goto repeat; | |
363 | ||
364 | /* Has the page moved? */ | |
365 | if (unlikely(page != *((void **)pages[i]))) { | |
366 | page_cache_release(page); | |
367 | goto repeat; | |
368 | } | |
369 | export: | |
370 | indices[ret] = indices[i]; | |
371 | pages[ret] = page; | |
372 | ret++; | |
373 | } | |
374 | if (unlikely(!ret && nr_found)) | |
375 | goto restart; | |
376 | rcu_read_unlock(); | |
377 | return ret; | |
378 | } | |
379 | ||
380 | /* | |
381 | * Remove swap entry from radix tree, free the swap and its page cache. | |
382 | */ | |
383 | static int shmem_free_swap(struct address_space *mapping, | |
384 | pgoff_t index, void *radswap) | |
385 | { | |
386 | int error; | |
387 | ||
388 | spin_lock_irq(&mapping->tree_lock); | |
389 | error = shmem_radix_tree_replace(mapping, index, radswap, NULL); | |
390 | spin_unlock_irq(&mapping->tree_lock); | |
391 | if (!error) | |
392 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
393 | return error; | |
394 | } | |
395 | ||
396 | /* | |
397 | * Pagevec may contain swap entries, so shuffle up pages before releasing. | |
398 | */ | |
24513264 | 399 | static void shmem_deswap_pagevec(struct pagevec *pvec) |
7a5d0fbb HD |
400 | { |
401 | int i, j; | |
402 | ||
403 | for (i = 0, j = 0; i < pagevec_count(pvec); i++) { | |
404 | struct page *page = pvec->pages[i]; | |
405 | if (!radix_tree_exceptional_entry(page)) | |
406 | pvec->pages[j++] = page; | |
407 | } | |
408 | pvec->nr = j; | |
24513264 HD |
409 | } |
410 | ||
411 | /* | |
412 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
413 | */ | |
414 | void shmem_unlock_mapping(struct address_space *mapping) | |
415 | { | |
416 | struct pagevec pvec; | |
417 | pgoff_t indices[PAGEVEC_SIZE]; | |
418 | pgoff_t index = 0; | |
419 | ||
420 | pagevec_init(&pvec, 0); | |
421 | /* | |
422 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
423 | */ | |
424 | while (!mapping_unevictable(mapping)) { | |
425 | /* | |
426 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
427 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
428 | */ | |
429 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, | |
430 | PAGEVEC_SIZE, pvec.pages, indices); | |
431 | if (!pvec.nr) | |
432 | break; | |
433 | index = indices[pvec.nr - 1] + 1; | |
434 | shmem_deswap_pagevec(&pvec); | |
435 | check_move_unevictable_pages(pvec.pages, pvec.nr); | |
436 | pagevec_release(&pvec); | |
437 | cond_resched(); | |
438 | } | |
7a5d0fbb HD |
439 | } |
440 | ||
441 | /* | |
442 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 443 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 444 | */ |
1635f6a7 HD |
445 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
446 | bool unfalloc) | |
1da177e4 | 447 | { |
285b2c4f | 448 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 449 | struct shmem_inode_info *info = SHMEM_I(inode); |
285b2c4f | 450 | pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
83e4fa9c HD |
451 | pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT; |
452 | unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1); | |
453 | unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); | |
bda97eab | 454 | struct pagevec pvec; |
7a5d0fbb HD |
455 | pgoff_t indices[PAGEVEC_SIZE]; |
456 | long nr_swaps_freed = 0; | |
285b2c4f | 457 | pgoff_t index; |
bda97eab HD |
458 | int i; |
459 | ||
83e4fa9c HD |
460 | if (lend == -1) |
461 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
462 | |
463 | pagevec_init(&pvec, 0); | |
464 | index = start; | |
83e4fa9c | 465 | while (index < end) { |
7a5d0fbb | 466 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, |
83e4fa9c | 467 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
7a5d0fbb HD |
468 | pvec.pages, indices); |
469 | if (!pvec.nr) | |
470 | break; | |
bda97eab HD |
471 | mem_cgroup_uncharge_start(); |
472 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
473 | struct page *page = pvec.pages[i]; | |
474 | ||
7a5d0fbb | 475 | index = indices[i]; |
83e4fa9c | 476 | if (index >= end) |
bda97eab HD |
477 | break; |
478 | ||
7a5d0fbb | 479 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
480 | if (unfalloc) |
481 | continue; | |
7a5d0fbb HD |
482 | nr_swaps_freed += !shmem_free_swap(mapping, |
483 | index, page); | |
bda97eab | 484 | continue; |
7a5d0fbb HD |
485 | } |
486 | ||
487 | if (!trylock_page(page)) | |
bda97eab | 488 | continue; |
1635f6a7 HD |
489 | if (!unfalloc || !PageUptodate(page)) { |
490 | if (page->mapping == mapping) { | |
491 | VM_BUG_ON(PageWriteback(page)); | |
492 | truncate_inode_page(mapping, page); | |
493 | } | |
bda97eab | 494 | } |
bda97eab HD |
495 | unlock_page(page); |
496 | } | |
24513264 HD |
497 | shmem_deswap_pagevec(&pvec); |
498 | pagevec_release(&pvec); | |
bda97eab HD |
499 | mem_cgroup_uncharge_end(); |
500 | cond_resched(); | |
501 | index++; | |
502 | } | |
1da177e4 | 503 | |
83e4fa9c | 504 | if (partial_start) { |
bda97eab HD |
505 | struct page *page = NULL; |
506 | shmem_getpage(inode, start - 1, &page, SGP_READ, NULL); | |
507 | if (page) { | |
83e4fa9c HD |
508 | unsigned int top = PAGE_CACHE_SIZE; |
509 | if (start > end) { | |
510 | top = partial_end; | |
511 | partial_end = 0; | |
512 | } | |
513 | zero_user_segment(page, partial_start, top); | |
514 | set_page_dirty(page); | |
515 | unlock_page(page); | |
516 | page_cache_release(page); | |
517 | } | |
518 | } | |
519 | if (partial_end) { | |
520 | struct page *page = NULL; | |
521 | shmem_getpage(inode, end, &page, SGP_READ, NULL); | |
522 | if (page) { | |
523 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
524 | set_page_dirty(page); |
525 | unlock_page(page); | |
526 | page_cache_release(page); | |
527 | } | |
528 | } | |
83e4fa9c HD |
529 | if (start >= end) |
530 | return; | |
bda97eab HD |
531 | |
532 | index = start; | |
533 | for ( ; ; ) { | |
534 | cond_resched(); | |
7a5d0fbb | 535 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, |
83e4fa9c | 536 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
7a5d0fbb HD |
537 | pvec.pages, indices); |
538 | if (!pvec.nr) { | |
1635f6a7 | 539 | if (index == start || unfalloc) |
bda97eab HD |
540 | break; |
541 | index = start; | |
542 | continue; | |
543 | } | |
1635f6a7 | 544 | if ((index == start || unfalloc) && indices[0] >= end) { |
24513264 HD |
545 | shmem_deswap_pagevec(&pvec); |
546 | pagevec_release(&pvec); | |
bda97eab HD |
547 | break; |
548 | } | |
549 | mem_cgroup_uncharge_start(); | |
550 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
551 | struct page *page = pvec.pages[i]; | |
552 | ||
7a5d0fbb | 553 | index = indices[i]; |
83e4fa9c | 554 | if (index >= end) |
bda97eab HD |
555 | break; |
556 | ||
7a5d0fbb | 557 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
558 | if (unfalloc) |
559 | continue; | |
7a5d0fbb HD |
560 | nr_swaps_freed += !shmem_free_swap(mapping, |
561 | index, page); | |
562 | continue; | |
563 | } | |
564 | ||
bda97eab | 565 | lock_page(page); |
1635f6a7 HD |
566 | if (!unfalloc || !PageUptodate(page)) { |
567 | if (page->mapping == mapping) { | |
568 | VM_BUG_ON(PageWriteback(page)); | |
569 | truncate_inode_page(mapping, page); | |
570 | } | |
7a5d0fbb | 571 | } |
bda97eab HD |
572 | unlock_page(page); |
573 | } | |
24513264 HD |
574 | shmem_deswap_pagevec(&pvec); |
575 | pagevec_release(&pvec); | |
bda97eab HD |
576 | mem_cgroup_uncharge_end(); |
577 | index++; | |
578 | } | |
94c1e62d | 579 | |
1da177e4 | 580 | spin_lock(&info->lock); |
7a5d0fbb | 581 | info->swapped -= nr_swaps_freed; |
1da177e4 LT |
582 | shmem_recalc_inode(inode); |
583 | spin_unlock(&info->lock); | |
1635f6a7 | 584 | } |
1da177e4 | 585 | |
1635f6a7 HD |
586 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
587 | { | |
588 | shmem_undo_range(inode, lstart, lend, false); | |
285b2c4f | 589 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1da177e4 | 590 | } |
94c1e62d | 591 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 592 | |
94c1e62d | 593 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 LT |
594 | { |
595 | struct inode *inode = dentry->d_inode; | |
1da177e4 LT |
596 | int error; |
597 | ||
db78b877 CH |
598 | error = inode_change_ok(inode, attr); |
599 | if (error) | |
600 | return error; | |
601 | ||
94c1e62d HD |
602 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
603 | loff_t oldsize = inode->i_size; | |
604 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 605 | |
94c1e62d HD |
606 | if (newsize != oldsize) { |
607 | i_size_write(inode, newsize); | |
608 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
609 | } | |
610 | if (newsize < oldsize) { | |
611 | loff_t holebegin = round_up(newsize, PAGE_SIZE); | |
612 | unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); | |
613 | shmem_truncate_range(inode, newsize, (loff_t)-1); | |
614 | /* unmap again to remove racily COWed private pages */ | |
615 | unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); | |
616 | } | |
1da177e4 LT |
617 | } |
618 | ||
db78b877 | 619 | setattr_copy(inode, attr); |
39f0247d | 620 | #ifdef CONFIG_TMPFS_POSIX_ACL |
db78b877 | 621 | if (attr->ia_valid & ATTR_MODE) |
1c7c474c | 622 | error = generic_acl_chmod(inode); |
39f0247d | 623 | #endif |
1da177e4 LT |
624 | return error; |
625 | } | |
626 | ||
1f895f75 | 627 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 628 | { |
1da177e4 | 629 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 630 | struct shmem_xattr *xattr, *nxattr; |
1da177e4 | 631 | |
3889e6e7 | 632 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
633 | shmem_unacct_size(info->flags, inode->i_size); |
634 | inode->i_size = 0; | |
3889e6e7 | 635 | shmem_truncate_range(inode, 0, (loff_t)-1); |
1da177e4 | 636 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 637 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 638 | list_del_init(&info->swaplist); |
cb5f7b9a | 639 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 640 | } |
69f07ec9 HD |
641 | } else |
642 | kfree(info->symlink); | |
b09e0fa4 EP |
643 | |
644 | list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) { | |
645 | kfree(xattr->name); | |
646 | kfree(xattr); | |
647 | } | |
0edd73b3 | 648 | BUG_ON(inode->i_blocks); |
5b04c689 | 649 | shmem_free_inode(inode->i_sb); |
dbd5768f | 650 | clear_inode(inode); |
1da177e4 LT |
651 | } |
652 | ||
46f65ec1 HD |
653 | /* |
654 | * If swap found in inode, free it and move page from swapcache to filecache. | |
655 | */ | |
41ffe5d5 | 656 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 657 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 658 | { |
285b2c4f | 659 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 660 | void *radswap; |
41ffe5d5 | 661 | pgoff_t index; |
bde05d1c HD |
662 | gfp_t gfp; |
663 | int error = 0; | |
1da177e4 | 664 | |
46f65ec1 | 665 | radswap = swp_to_radix_entry(swap); |
e504f3fd | 666 | index = radix_tree_locate_item(&mapping->page_tree, radswap); |
46f65ec1 | 667 | if (index == -1) |
285b2c4f | 668 | return 0; |
2e0e26c7 | 669 | |
1b1b32f2 HD |
670 | /* |
671 | * Move _head_ to start search for next from here. | |
1f895f75 | 672 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 673 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 674 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
675 | */ |
676 | if (shmem_swaplist.next != &info->swaplist) | |
677 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 678 | |
bde05d1c HD |
679 | gfp = mapping_gfp_mask(mapping); |
680 | if (shmem_should_replace_page(*pagep, gfp)) { | |
681 | mutex_unlock(&shmem_swaplist_mutex); | |
682 | error = shmem_replace_page(pagep, gfp, info, index); | |
683 | mutex_lock(&shmem_swaplist_mutex); | |
684 | /* | |
685 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
686 | * allocation, but the inode might have been freed while we |
687 | * dropped it: although a racing shmem_evict_inode() cannot | |
688 | * complete without emptying the radix_tree, our page lock | |
689 | * on this swapcache page is not enough to prevent that - | |
690 | * free_swap_and_cache() of our swap entry will only | |
691 | * trylock_page(), removing swap from radix_tree whatever. | |
692 | * | |
693 | * We must not proceed to shmem_add_to_page_cache() if the | |
694 | * inode has been freed, but of course we cannot rely on | |
695 | * inode or mapping or info to check that. However, we can | |
696 | * safely check if our swap entry is still in use (and here | |
697 | * it can't have got reused for another page): if it's still | |
698 | * in use, then the inode cannot have been freed yet, and we | |
699 | * can safely proceed (if it's no longer in use, that tells | |
700 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
701 | */ |
702 | if (!page_swapcount(*pagep)) | |
703 | error = -ENOENT; | |
704 | } | |
705 | ||
d13d1443 | 706 | /* |
778dd893 HD |
707 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
708 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
709 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 710 | */ |
bde05d1c HD |
711 | if (!error) |
712 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
46f65ec1 | 713 | GFP_NOWAIT, radswap); |
48f170fb | 714 | if (error != -ENOMEM) { |
46f65ec1 HD |
715 | /* |
716 | * Truncation and eviction use free_swap_and_cache(), which | |
717 | * only does trylock page: if we raced, best clean up here. | |
718 | */ | |
bde05d1c HD |
719 | delete_from_swap_cache(*pagep); |
720 | set_page_dirty(*pagep); | |
46f65ec1 HD |
721 | if (!error) { |
722 | spin_lock(&info->lock); | |
723 | info->swapped--; | |
724 | spin_unlock(&info->lock); | |
725 | swap_free(swap); | |
726 | } | |
2e0e26c7 | 727 | error = 1; /* not an error, but entry was found */ |
1da177e4 | 728 | } |
2e0e26c7 | 729 | return error; |
1da177e4 LT |
730 | } |
731 | ||
732 | /* | |
46f65ec1 | 733 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 734 | */ |
41ffe5d5 | 735 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 736 | { |
41ffe5d5 | 737 | struct list_head *this, *next; |
1da177e4 LT |
738 | struct shmem_inode_info *info; |
739 | int found = 0; | |
bde05d1c HD |
740 | int error = 0; |
741 | ||
742 | /* | |
743 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 744 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 745 | */ |
0142ef6c | 746 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 747 | goto out; |
778dd893 HD |
748 | |
749 | /* | |
750 | * Charge page using GFP_KERNEL while we can wait, before taking | |
751 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
752 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 HD |
753 | */ |
754 | error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); | |
755 | if (error) | |
756 | goto out; | |
46f65ec1 | 757 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
1da177e4 | 758 | |
cb5f7b9a | 759 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
760 | list_for_each_safe(this, next, &shmem_swaplist) { |
761 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 762 | if (info->swapped) |
bde05d1c | 763 | found = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
764 | else |
765 | list_del_init(&info->swaplist); | |
cb5f7b9a | 766 | cond_resched(); |
2e0e26c7 | 767 | if (found) |
778dd893 | 768 | break; |
1da177e4 | 769 | } |
cb5f7b9a | 770 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 771 | |
778dd893 HD |
772 | if (found < 0) |
773 | error = found; | |
774 | out: | |
aaa46865 HD |
775 | unlock_page(page); |
776 | page_cache_release(page); | |
778dd893 | 777 | return error; |
1da177e4 LT |
778 | } |
779 | ||
780 | /* | |
781 | * Move the page from the page cache to the swap cache. | |
782 | */ | |
783 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
784 | { | |
785 | struct shmem_inode_info *info; | |
1da177e4 | 786 | struct address_space *mapping; |
1da177e4 | 787 | struct inode *inode; |
6922c0c7 HD |
788 | swp_entry_t swap; |
789 | pgoff_t index; | |
1da177e4 LT |
790 | |
791 | BUG_ON(!PageLocked(page)); | |
1da177e4 LT |
792 | mapping = page->mapping; |
793 | index = page->index; | |
794 | inode = mapping->host; | |
795 | info = SHMEM_I(inode); | |
796 | if (info->flags & VM_LOCKED) | |
797 | goto redirty; | |
d9fe526a | 798 | if (!total_swap_pages) |
1da177e4 LT |
799 | goto redirty; |
800 | ||
d9fe526a HD |
801 | /* |
802 | * shmem_backing_dev_info's capabilities prevent regular writeback or | |
803 | * sync from ever calling shmem_writepage; but a stacking filesystem | |
48f170fb | 804 | * might use ->writepage of its underlying filesystem, in which case |
d9fe526a | 805 | * tmpfs should write out to swap only in response to memory pressure, |
48f170fb | 806 | * and not for the writeback threads or sync. |
d9fe526a | 807 | */ |
48f170fb HD |
808 | if (!wbc->for_reclaim) { |
809 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
810 | goto redirty; | |
811 | } | |
1635f6a7 HD |
812 | |
813 | /* | |
814 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
815 | * value into swapfile.c, the only way we can correctly account for a | |
816 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
817 | * |
818 | * That's okay for a page already fallocated earlier, but if we have | |
819 | * not yet completed the fallocation, then (a) we want to keep track | |
820 | * of this page in case we have to undo it, and (b) it may not be a | |
821 | * good idea to continue anyway, once we're pushing into swap. So | |
822 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
823 | */ |
824 | if (!PageUptodate(page)) { | |
1aac1400 HD |
825 | if (inode->i_private) { |
826 | struct shmem_falloc *shmem_falloc; | |
827 | spin_lock(&inode->i_lock); | |
828 | shmem_falloc = inode->i_private; | |
829 | if (shmem_falloc && | |
830 | index >= shmem_falloc->start && | |
831 | index < shmem_falloc->next) | |
832 | shmem_falloc->nr_unswapped++; | |
833 | else | |
834 | shmem_falloc = NULL; | |
835 | spin_unlock(&inode->i_lock); | |
836 | if (shmem_falloc) | |
837 | goto redirty; | |
838 | } | |
1635f6a7 HD |
839 | clear_highpage(page); |
840 | flush_dcache_page(page); | |
841 | SetPageUptodate(page); | |
842 | } | |
843 | ||
48f170fb HD |
844 | swap = get_swap_page(); |
845 | if (!swap.val) | |
846 | goto redirty; | |
d9fe526a | 847 | |
b1dea800 HD |
848 | /* |
849 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
850 | * if it's not already there. Do it now before the page is |
851 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 852 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
853 | * we've incremented swapped, because shmem_unuse_inode() will |
854 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 855 | */ |
48f170fb HD |
856 | mutex_lock(&shmem_swaplist_mutex); |
857 | if (list_empty(&info->swaplist)) | |
858 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 859 | |
48f170fb | 860 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
aaa46865 | 861 | swap_shmem_alloc(swap); |
6922c0c7 HD |
862 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); |
863 | ||
864 | spin_lock(&info->lock); | |
865 | info->swapped++; | |
866 | shmem_recalc_inode(inode); | |
826267cf | 867 | spin_unlock(&info->lock); |
6922c0c7 HD |
868 | |
869 | mutex_unlock(&shmem_swaplist_mutex); | |
d9fe526a | 870 | BUG_ON(page_mapped(page)); |
9fab5619 | 871 | swap_writepage(page, wbc); |
1da177e4 LT |
872 | return 0; |
873 | } | |
874 | ||
6922c0c7 | 875 | mutex_unlock(&shmem_swaplist_mutex); |
cb4b86ba | 876 | swapcache_free(swap, NULL); |
1da177e4 LT |
877 | redirty: |
878 | set_page_dirty(page); | |
d9fe526a HD |
879 | if (wbc->for_reclaim) |
880 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
881 | unlock_page(page); | |
882 | return 0; | |
1da177e4 LT |
883 | } |
884 | ||
885 | #ifdef CONFIG_NUMA | |
680d794b | 886 | #ifdef CONFIG_TMPFS |
71fe804b | 887 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 888 | { |
095f1fc4 | 889 | char buffer[64]; |
680d794b | 890 | |
71fe804b | 891 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 892 | return; /* show nothing */ |
680d794b | 893 | |
71fe804b | 894 | mpol_to_str(buffer, sizeof(buffer), mpol, 1); |
095f1fc4 LS |
895 | |
896 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 897 | } |
71fe804b LS |
898 | |
899 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
900 | { | |
901 | struct mempolicy *mpol = NULL; | |
902 | if (sbinfo->mpol) { | |
903 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
904 | mpol = sbinfo->mpol; | |
905 | mpol_get(mpol); | |
906 | spin_unlock(&sbinfo->stat_lock); | |
907 | } | |
908 | return mpol; | |
909 | } | |
680d794b AM |
910 | #endif /* CONFIG_TMPFS */ |
911 | ||
41ffe5d5 HD |
912 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
913 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 914 | { |
52cd3b07 | 915 | struct mempolicy mpol, *spol; |
1da177e4 LT |
916 | struct vm_area_struct pvma; |
917 | ||
52cd3b07 | 918 | spol = mpol_cond_copy(&mpol, |
41ffe5d5 | 919 | mpol_shared_policy_lookup(&info->policy, index)); |
52cd3b07 | 920 | |
1da177e4 | 921 | /* Create a pseudo vma that just contains the policy */ |
c4cc6d07 | 922 | pvma.vm_start = 0; |
41ffe5d5 | 923 | pvma.vm_pgoff = index; |
c4cc6d07 | 924 | pvma.vm_ops = NULL; |
52cd3b07 | 925 | pvma.vm_policy = spol; |
41ffe5d5 | 926 | return swapin_readahead(swap, gfp, &pvma, 0); |
1da177e4 LT |
927 | } |
928 | ||
02098fea | 929 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 930 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
931 | { |
932 | struct vm_area_struct pvma; | |
1da177e4 | 933 | |
c4cc6d07 HD |
934 | /* Create a pseudo vma that just contains the policy */ |
935 | pvma.vm_start = 0; | |
41ffe5d5 | 936 | pvma.vm_pgoff = index; |
c4cc6d07 | 937 | pvma.vm_ops = NULL; |
41ffe5d5 | 938 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
52cd3b07 LS |
939 | |
940 | /* | |
941 | * alloc_page_vma() will drop the shared policy reference | |
942 | */ | |
943 | return alloc_page_vma(gfp, &pvma, 0); | |
1da177e4 | 944 | } |
680d794b AM |
945 | #else /* !CONFIG_NUMA */ |
946 | #ifdef CONFIG_TMPFS | |
41ffe5d5 | 947 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b AM |
948 | { |
949 | } | |
950 | #endif /* CONFIG_TMPFS */ | |
951 | ||
41ffe5d5 HD |
952 | static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
953 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 954 | { |
41ffe5d5 | 955 | return swapin_readahead(swap, gfp, NULL, 0); |
1da177e4 LT |
956 | } |
957 | ||
02098fea | 958 | static inline struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 959 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 | 960 | { |
e84e2e13 | 961 | return alloc_page(gfp); |
1da177e4 | 962 | } |
680d794b | 963 | #endif /* CONFIG_NUMA */ |
1da177e4 | 964 | |
71fe804b LS |
965 | #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) |
966 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
967 | { | |
968 | return NULL; | |
969 | } | |
970 | #endif | |
971 | ||
bde05d1c HD |
972 | /* |
973 | * When a page is moved from swapcache to shmem filecache (either by the | |
974 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
975 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
976 | * ignorance of the mapping it belongs to. If that mapping has special | |
977 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
978 | * we may need to copy to a suitable page before moving to filecache. | |
979 | * | |
980 | * In a future release, this may well be extended to respect cpuset and | |
981 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
982 | * but for now it is a simple matter of zone. | |
983 | */ | |
984 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
985 | { | |
986 | return page_zonenum(page) > gfp_zone(gfp); | |
987 | } | |
988 | ||
989 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
990 | struct shmem_inode_info *info, pgoff_t index) | |
991 | { | |
992 | struct page *oldpage, *newpage; | |
993 | struct address_space *swap_mapping; | |
994 | pgoff_t swap_index; | |
995 | int error; | |
996 | ||
997 | oldpage = *pagep; | |
998 | swap_index = page_private(oldpage); | |
999 | swap_mapping = page_mapping(oldpage); | |
1000 | ||
1001 | /* | |
1002 | * We have arrived here because our zones are constrained, so don't | |
1003 | * limit chance of success by further cpuset and node constraints. | |
1004 | */ | |
1005 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1006 | newpage = shmem_alloc_page(gfp, info, index); | |
1007 | if (!newpage) | |
1008 | return -ENOMEM; | |
bde05d1c | 1009 | |
bde05d1c HD |
1010 | page_cache_get(newpage); |
1011 | copy_highpage(newpage, oldpage); | |
0142ef6c | 1012 | flush_dcache_page(newpage); |
bde05d1c | 1013 | |
bde05d1c | 1014 | __set_page_locked(newpage); |
bde05d1c | 1015 | SetPageUptodate(newpage); |
bde05d1c | 1016 | SetPageSwapBacked(newpage); |
bde05d1c | 1017 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1018 | SetPageSwapCache(newpage); |
1019 | ||
1020 | /* | |
1021 | * Our caller will very soon move newpage out of swapcache, but it's | |
1022 | * a nice clean interface for us to replace oldpage by newpage there. | |
1023 | */ | |
1024 | spin_lock_irq(&swap_mapping->tree_lock); | |
1025 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
1026 | newpage); | |
0142ef6c HD |
1027 | if (!error) { |
1028 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
1029 | __dec_zone_page_state(oldpage, NR_FILE_PAGES); | |
1030 | } | |
bde05d1c | 1031 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1032 | |
0142ef6c HD |
1033 | if (unlikely(error)) { |
1034 | /* | |
1035 | * Is this possible? I think not, now that our callers check | |
1036 | * both PageSwapCache and page_private after getting page lock; | |
1037 | * but be defensive. Reverse old to newpage for clear and free. | |
1038 | */ | |
1039 | oldpage = newpage; | |
1040 | } else { | |
1041 | mem_cgroup_replace_page_cache(oldpage, newpage); | |
1042 | lru_cache_add_anon(newpage); | |
1043 | *pagep = newpage; | |
1044 | } | |
bde05d1c HD |
1045 | |
1046 | ClearPageSwapCache(oldpage); | |
1047 | set_page_private(oldpage, 0); | |
1048 | ||
1049 | unlock_page(oldpage); | |
1050 | page_cache_release(oldpage); | |
1051 | page_cache_release(oldpage); | |
0142ef6c | 1052 | return error; |
bde05d1c HD |
1053 | } |
1054 | ||
1da177e4 | 1055 | /* |
68da9f05 | 1056 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1057 | * |
1058 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1059 | * vm. If we swap it in we mark it dirty since we also free the swap | |
1060 | * entry since a page cannot live in both the swap and page cache | |
1061 | */ | |
41ffe5d5 | 1062 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
68da9f05 | 1063 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type) |
1da177e4 LT |
1064 | { |
1065 | struct address_space *mapping = inode->i_mapping; | |
54af6042 | 1066 | struct shmem_inode_info *info; |
1da177e4 | 1067 | struct shmem_sb_info *sbinfo; |
27ab7006 | 1068 | struct page *page; |
1da177e4 LT |
1069 | swp_entry_t swap; |
1070 | int error; | |
54af6042 | 1071 | int once = 0; |
1635f6a7 | 1072 | int alloced = 0; |
1da177e4 | 1073 | |
41ffe5d5 | 1074 | if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT)) |
1da177e4 | 1075 | return -EFBIG; |
1da177e4 | 1076 | repeat: |
54af6042 | 1077 | swap.val = 0; |
41ffe5d5 | 1078 | page = find_lock_page(mapping, index); |
54af6042 HD |
1079 | if (radix_tree_exceptional_entry(page)) { |
1080 | swap = radix_to_swp_entry(page); | |
1081 | page = NULL; | |
1082 | } | |
1083 | ||
1635f6a7 | 1084 | if (sgp != SGP_WRITE && sgp != SGP_FALLOC && |
54af6042 HD |
1085 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { |
1086 | error = -EINVAL; | |
1087 | goto failed; | |
1088 | } | |
1089 | ||
1635f6a7 HD |
1090 | /* fallocated page? */ |
1091 | if (page && !PageUptodate(page)) { | |
1092 | if (sgp != SGP_READ) | |
1093 | goto clear; | |
1094 | unlock_page(page); | |
1095 | page_cache_release(page); | |
1096 | page = NULL; | |
1097 | } | |
54af6042 | 1098 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1099 | *pagep = page; |
1100 | return 0; | |
27ab7006 HD |
1101 | } |
1102 | ||
1103 | /* | |
54af6042 HD |
1104 | * Fast cache lookup did not find it: |
1105 | * bring it back from swap or allocate. | |
27ab7006 | 1106 | */ |
54af6042 HD |
1107 | info = SHMEM_I(inode); |
1108 | sbinfo = SHMEM_SB(inode->i_sb); | |
1da177e4 | 1109 | |
1da177e4 LT |
1110 | if (swap.val) { |
1111 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1112 | page = lookup_swap_cache(swap); |
1113 | if (!page) { | |
1da177e4 | 1114 | /* here we actually do the io */ |
68da9f05 HD |
1115 | if (fault_type) |
1116 | *fault_type |= VM_FAULT_MAJOR; | |
41ffe5d5 | 1117 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1118 | if (!page) { |
54af6042 HD |
1119 | error = -ENOMEM; |
1120 | goto failed; | |
1da177e4 | 1121 | } |
1da177e4 LT |
1122 | } |
1123 | ||
1124 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1125 | lock_page(page); |
0142ef6c HD |
1126 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
1127 | page->mapping) { | |
bde05d1c HD |
1128 | error = -EEXIST; /* try again */ |
1129 | goto failed; | |
1130 | } | |
27ab7006 | 1131 | if (!PageUptodate(page)) { |
1da177e4 | 1132 | error = -EIO; |
54af6042 | 1133 | goto failed; |
1da177e4 | 1134 | } |
54af6042 HD |
1135 | wait_on_page_writeback(page); |
1136 | ||
bde05d1c HD |
1137 | if (shmem_should_replace_page(page, gfp)) { |
1138 | error = shmem_replace_page(&page, gfp, info, index); | |
1139 | if (error) | |
1140 | goto failed; | |
1da177e4 | 1141 | } |
27ab7006 | 1142 | |
aa3b1895 HD |
1143 | error = mem_cgroup_cache_charge(page, current->mm, |
1144 | gfp & GFP_RECLAIM_MASK); | |
1145 | if (!error) | |
1146 | error = shmem_add_to_page_cache(page, mapping, index, | |
1147 | gfp, swp_to_radix_entry(swap)); | |
54af6042 HD |
1148 | if (error) |
1149 | goto failed; | |
1150 | ||
1151 | spin_lock(&info->lock); | |
285b2c4f | 1152 | info->swapped--; |
54af6042 | 1153 | shmem_recalc_inode(inode); |
27ab7006 | 1154 | spin_unlock(&info->lock); |
54af6042 HD |
1155 | |
1156 | delete_from_swap_cache(page); | |
27ab7006 HD |
1157 | set_page_dirty(page); |
1158 | swap_free(swap); | |
1159 | ||
54af6042 HD |
1160 | } else { |
1161 | if (shmem_acct_block(info->flags)) { | |
1162 | error = -ENOSPC; | |
1163 | goto failed; | |
1da177e4 | 1164 | } |
0edd73b3 | 1165 | if (sbinfo->max_blocks) { |
fc5da22a | 1166 | if (percpu_counter_compare(&sbinfo->used_blocks, |
54af6042 HD |
1167 | sbinfo->max_blocks) >= 0) { |
1168 | error = -ENOSPC; | |
1169 | goto unacct; | |
1170 | } | |
7e496299 | 1171 | percpu_counter_inc(&sbinfo->used_blocks); |
54af6042 | 1172 | } |
1da177e4 | 1173 | |
54af6042 HD |
1174 | page = shmem_alloc_page(gfp, info, index); |
1175 | if (!page) { | |
1176 | error = -ENOMEM; | |
1177 | goto decused; | |
1da177e4 LT |
1178 | } |
1179 | ||
54af6042 HD |
1180 | SetPageSwapBacked(page); |
1181 | __set_page_locked(page); | |
aa3b1895 HD |
1182 | error = mem_cgroup_cache_charge(page, current->mm, |
1183 | gfp & GFP_RECLAIM_MASK); | |
1184 | if (!error) | |
1185 | error = shmem_add_to_page_cache(page, mapping, index, | |
1186 | gfp, NULL); | |
54af6042 HD |
1187 | if (error) |
1188 | goto decused; | |
1189 | lru_cache_add_anon(page); | |
1190 | ||
1191 | spin_lock(&info->lock); | |
1da177e4 | 1192 | info->alloced++; |
54af6042 HD |
1193 | inode->i_blocks += BLOCKS_PER_PAGE; |
1194 | shmem_recalc_inode(inode); | |
1da177e4 | 1195 | spin_unlock(&info->lock); |
1635f6a7 | 1196 | alloced = true; |
54af6042 | 1197 | |
ec9516fb | 1198 | /* |
1635f6a7 HD |
1199 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1200 | */ | |
1201 | if (sgp == SGP_FALLOC) | |
1202 | sgp = SGP_WRITE; | |
1203 | clear: | |
1204 | /* | |
1205 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1206 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1207 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb HD |
1208 | */ |
1209 | if (sgp != SGP_WRITE) { | |
1210 | clear_highpage(page); | |
1211 | flush_dcache_page(page); | |
1212 | SetPageUptodate(page); | |
1213 | } | |
a0ee5ec5 | 1214 | if (sgp == SGP_DIRTY) |
27ab7006 | 1215 | set_page_dirty(page); |
1da177e4 | 1216 | } |
bde05d1c | 1217 | |
54af6042 | 1218 | /* Perhaps the file has been truncated since we checked */ |
1635f6a7 | 1219 | if (sgp != SGP_WRITE && sgp != SGP_FALLOC && |
54af6042 HD |
1220 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { |
1221 | error = -EINVAL; | |
1635f6a7 HD |
1222 | if (alloced) |
1223 | goto trunc; | |
1224 | else | |
1225 | goto failed; | |
e83c32e8 | 1226 | } |
54af6042 HD |
1227 | *pagep = page; |
1228 | return 0; | |
1da177e4 | 1229 | |
59a16ead | 1230 | /* |
54af6042 | 1231 | * Error recovery. |
59a16ead | 1232 | */ |
54af6042 | 1233 | trunc: |
1635f6a7 | 1234 | info = SHMEM_I(inode); |
54af6042 HD |
1235 | ClearPageDirty(page); |
1236 | delete_from_page_cache(page); | |
1237 | spin_lock(&info->lock); | |
1238 | info->alloced--; | |
1239 | inode->i_blocks -= BLOCKS_PER_PAGE; | |
59a16ead | 1240 | spin_unlock(&info->lock); |
54af6042 | 1241 | decused: |
1635f6a7 | 1242 | sbinfo = SHMEM_SB(inode->i_sb); |
54af6042 HD |
1243 | if (sbinfo->max_blocks) |
1244 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
1245 | unacct: | |
1246 | shmem_unacct_blocks(info->flags, 1); | |
1247 | failed: | |
1248 | if (swap.val && error != -EINVAL) { | |
1249 | struct page *test = find_get_page(mapping, index); | |
1250 | if (test && !radix_tree_exceptional_entry(test)) | |
1251 | page_cache_release(test); | |
1252 | /* Have another try if the entry has changed */ | |
1253 | if (test != swp_to_radix_entry(swap)) | |
1254 | error = -EEXIST; | |
1255 | } | |
27ab7006 | 1256 | if (page) { |
54af6042 | 1257 | unlock_page(page); |
27ab7006 | 1258 | page_cache_release(page); |
54af6042 HD |
1259 | } |
1260 | if (error == -ENOSPC && !once++) { | |
1261 | info = SHMEM_I(inode); | |
1262 | spin_lock(&info->lock); | |
1263 | shmem_recalc_inode(inode); | |
1264 | spin_unlock(&info->lock); | |
27ab7006 | 1265 | goto repeat; |
ff36b801 | 1266 | } |
54af6042 HD |
1267 | if (error == -EEXIST) |
1268 | goto repeat; | |
1269 | return error; | |
1da177e4 LT |
1270 | } |
1271 | ||
d0217ac0 | 1272 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1273 | { |
d3ac7f89 | 1274 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 | 1275 | int error; |
68da9f05 | 1276 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1277 | |
27d54b39 | 1278 | error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); |
d0217ac0 NP |
1279 | if (error) |
1280 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1281 | |
456f998e YH |
1282 | if (ret & VM_FAULT_MAJOR) { |
1283 | count_vm_event(PGMAJFAULT); | |
1284 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); | |
1285 | } | |
68da9f05 | 1286 | return ret; |
1da177e4 LT |
1287 | } |
1288 | ||
1da177e4 | 1289 | #ifdef CONFIG_NUMA |
41ffe5d5 | 1290 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 1291 | { |
41ffe5d5 HD |
1292 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1293 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); | |
1da177e4 LT |
1294 | } |
1295 | ||
d8dc74f2 AB |
1296 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
1297 | unsigned long addr) | |
1da177e4 | 1298 | { |
41ffe5d5 HD |
1299 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1300 | pgoff_t index; | |
1da177e4 | 1301 | |
41ffe5d5 HD |
1302 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
1303 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
1304 | } |
1305 | #endif | |
1306 | ||
1307 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
1308 | { | |
d3ac7f89 | 1309 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
1310 | struct shmem_inode_info *info = SHMEM_I(inode); |
1311 | int retval = -ENOMEM; | |
1312 | ||
1313 | spin_lock(&info->lock); | |
1314 | if (lock && !(info->flags & VM_LOCKED)) { | |
1315 | if (!user_shm_lock(inode->i_size, user)) | |
1316 | goto out_nomem; | |
1317 | info->flags |= VM_LOCKED; | |
89e004ea | 1318 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
1319 | } |
1320 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
1321 | user_shm_unlock(inode->i_size, user); | |
1322 | info->flags &= ~VM_LOCKED; | |
89e004ea | 1323 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
1324 | } |
1325 | retval = 0; | |
89e004ea | 1326 | |
1da177e4 LT |
1327 | out_nomem: |
1328 | spin_unlock(&info->lock); | |
1329 | return retval; | |
1330 | } | |
1331 | ||
9b83a6a8 | 1332 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
1333 | { |
1334 | file_accessed(file); | |
1335 | vma->vm_ops = &shmem_vm_ops; | |
d0217ac0 | 1336 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
1337 | return 0; |
1338 | } | |
1339 | ||
454abafe | 1340 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 1341 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
1342 | { |
1343 | struct inode *inode; | |
1344 | struct shmem_inode_info *info; | |
1345 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
1346 | ||
5b04c689 PE |
1347 | if (shmem_reserve_inode(sb)) |
1348 | return NULL; | |
1da177e4 LT |
1349 | |
1350 | inode = new_inode(sb); | |
1351 | if (inode) { | |
85fe4025 | 1352 | inode->i_ino = get_next_ino(); |
454abafe | 1353 | inode_init_owner(inode, dir, mode); |
1da177e4 | 1354 | inode->i_blocks = 0; |
1da177e4 LT |
1355 | inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; |
1356 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
91828a40 | 1357 | inode->i_generation = get_seconds(); |
1da177e4 LT |
1358 | info = SHMEM_I(inode); |
1359 | memset(info, 0, (char *)inode - (char *)info); | |
1360 | spin_lock_init(&info->lock); | |
0b0a0806 | 1361 | info->flags = flags & VM_NORESERVE; |
1da177e4 | 1362 | INIT_LIST_HEAD(&info->swaplist); |
b09e0fa4 | 1363 | INIT_LIST_HEAD(&info->xattr_list); |
72c04902 | 1364 | cache_no_acl(inode); |
1da177e4 LT |
1365 | |
1366 | switch (mode & S_IFMT) { | |
1367 | default: | |
39f0247d | 1368 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
1369 | init_special_inode(inode, mode, dev); |
1370 | break; | |
1371 | case S_IFREG: | |
14fcc23f | 1372 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
1373 | inode->i_op = &shmem_inode_operations; |
1374 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
1375 | mpol_shared_policy_init(&info->policy, |
1376 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
1377 | break; |
1378 | case S_IFDIR: | |
d8c76e6f | 1379 | inc_nlink(inode); |
1da177e4 LT |
1380 | /* Some things misbehave if size == 0 on a directory */ |
1381 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
1382 | inode->i_op = &shmem_dir_inode_operations; | |
1383 | inode->i_fop = &simple_dir_operations; | |
1384 | break; | |
1385 | case S_IFLNK: | |
1386 | /* | |
1387 | * Must not load anything in the rbtree, | |
1388 | * mpol_free_shared_policy will not be called. | |
1389 | */ | |
71fe804b | 1390 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
1391 | break; |
1392 | } | |
5b04c689 PE |
1393 | } else |
1394 | shmem_free_inode(sb); | |
1da177e4 LT |
1395 | return inode; |
1396 | } | |
1397 | ||
1398 | #ifdef CONFIG_TMPFS | |
92e1d5be | 1399 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 1400 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 1401 | |
6d9d88d0 JS |
1402 | #ifdef CONFIG_TMPFS_XATTR |
1403 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
1404 | #else | |
1405 | #define shmem_initxattrs NULL | |
1406 | #endif | |
1407 | ||
1da177e4 | 1408 | static int |
800d15a5 NP |
1409 | shmem_write_begin(struct file *file, struct address_space *mapping, |
1410 | loff_t pos, unsigned len, unsigned flags, | |
1411 | struct page **pagep, void **fsdata) | |
1da177e4 | 1412 | { |
800d15a5 NP |
1413 | struct inode *inode = mapping->host; |
1414 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
800d15a5 NP |
1415 | return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); |
1416 | } | |
1417 | ||
1418 | static int | |
1419 | shmem_write_end(struct file *file, struct address_space *mapping, | |
1420 | loff_t pos, unsigned len, unsigned copied, | |
1421 | struct page *page, void *fsdata) | |
1422 | { | |
1423 | struct inode *inode = mapping->host; | |
1424 | ||
d3602444 HD |
1425 | if (pos + copied > inode->i_size) |
1426 | i_size_write(inode, pos + copied); | |
1427 | ||
ec9516fb HD |
1428 | if (!PageUptodate(page)) { |
1429 | if (copied < PAGE_CACHE_SIZE) { | |
1430 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
1431 | zero_user_segments(page, 0, from, | |
1432 | from + copied, PAGE_CACHE_SIZE); | |
1433 | } | |
1434 | SetPageUptodate(page); | |
1435 | } | |
800d15a5 | 1436 | set_page_dirty(page); |
6746aff7 | 1437 | unlock_page(page); |
800d15a5 NP |
1438 | page_cache_release(page); |
1439 | ||
800d15a5 | 1440 | return copied; |
1da177e4 LT |
1441 | } |
1442 | ||
1da177e4 LT |
1443 | static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) |
1444 | { | |
d3ac7f89 | 1445 | struct inode *inode = filp->f_path.dentry->d_inode; |
1da177e4 | 1446 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
1447 | pgoff_t index; |
1448 | unsigned long offset; | |
a0ee5ec5 HD |
1449 | enum sgp_type sgp = SGP_READ; |
1450 | ||
1451 | /* | |
1452 | * Might this read be for a stacking filesystem? Then when reading | |
1453 | * holes of a sparse file, we actually need to allocate those pages, | |
1454 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
1455 | */ | |
1456 | if (segment_eq(get_fs(), KERNEL_DS)) | |
1457 | sgp = SGP_DIRTY; | |
1da177e4 LT |
1458 | |
1459 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1460 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1461 | ||
1462 | for (;;) { | |
1463 | struct page *page = NULL; | |
41ffe5d5 HD |
1464 | pgoff_t end_index; |
1465 | unsigned long nr, ret; | |
1da177e4 LT |
1466 | loff_t i_size = i_size_read(inode); |
1467 | ||
1468 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1469 | if (index > end_index) | |
1470 | break; | |
1471 | if (index == end_index) { | |
1472 | nr = i_size & ~PAGE_CACHE_MASK; | |
1473 | if (nr <= offset) | |
1474 | break; | |
1475 | } | |
1476 | ||
a0ee5ec5 | 1477 | desc->error = shmem_getpage(inode, index, &page, sgp, NULL); |
1da177e4 LT |
1478 | if (desc->error) { |
1479 | if (desc->error == -EINVAL) | |
1480 | desc->error = 0; | |
1481 | break; | |
1482 | } | |
d3602444 HD |
1483 | if (page) |
1484 | unlock_page(page); | |
1da177e4 LT |
1485 | |
1486 | /* | |
1487 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 1488 | * are called without i_mutex protection against truncate |
1da177e4 LT |
1489 | */ |
1490 | nr = PAGE_CACHE_SIZE; | |
1491 | i_size = i_size_read(inode); | |
1492 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1493 | if (index == end_index) { | |
1494 | nr = i_size & ~PAGE_CACHE_MASK; | |
1495 | if (nr <= offset) { | |
1496 | if (page) | |
1497 | page_cache_release(page); | |
1498 | break; | |
1499 | } | |
1500 | } | |
1501 | nr -= offset; | |
1502 | ||
1503 | if (page) { | |
1504 | /* | |
1505 | * If users can be writing to this page using arbitrary | |
1506 | * virtual addresses, take care about potential aliasing | |
1507 | * before reading the page on the kernel side. | |
1508 | */ | |
1509 | if (mapping_writably_mapped(mapping)) | |
1510 | flush_dcache_page(page); | |
1511 | /* | |
1512 | * Mark the page accessed if we read the beginning. | |
1513 | */ | |
1514 | if (!offset) | |
1515 | mark_page_accessed(page); | |
b5810039 | 1516 | } else { |
1da177e4 | 1517 | page = ZERO_PAGE(0); |
b5810039 NP |
1518 | page_cache_get(page); |
1519 | } | |
1da177e4 LT |
1520 | |
1521 | /* | |
1522 | * Ok, we have the page, and it's up-to-date, so | |
1523 | * now we can copy it to user space... | |
1524 | * | |
1525 | * The actor routine returns how many bytes were actually used.. | |
1526 | * NOTE! This may not be the same as how much of a user buffer | |
1527 | * we filled up (we may be padding etc), so we can only update | |
1528 | * "pos" here (the actor routine has to update the user buffer | |
1529 | * pointers and the remaining count). | |
1530 | */ | |
1531 | ret = actor(desc, page, offset, nr); | |
1532 | offset += ret; | |
1533 | index += offset >> PAGE_CACHE_SHIFT; | |
1534 | offset &= ~PAGE_CACHE_MASK; | |
1535 | ||
1536 | page_cache_release(page); | |
1537 | if (ret != nr || !desc->count) | |
1538 | break; | |
1539 | ||
1540 | cond_resched(); | |
1541 | } | |
1542 | ||
1543 | *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; | |
1544 | file_accessed(filp); | |
1545 | } | |
1546 | ||
bcd78e49 HD |
1547 | static ssize_t shmem_file_aio_read(struct kiocb *iocb, |
1548 | const struct iovec *iov, unsigned long nr_segs, loff_t pos) | |
1549 | { | |
1550 | struct file *filp = iocb->ki_filp; | |
1551 | ssize_t retval; | |
1552 | unsigned long seg; | |
1553 | size_t count; | |
1554 | loff_t *ppos = &iocb->ki_pos; | |
1555 | ||
1556 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); | |
1557 | if (retval) | |
1558 | return retval; | |
1559 | ||
1560 | for (seg = 0; seg < nr_segs; seg++) { | |
1561 | read_descriptor_t desc; | |
1562 | ||
1563 | desc.written = 0; | |
1564 | desc.arg.buf = iov[seg].iov_base; | |
1565 | desc.count = iov[seg].iov_len; | |
1566 | if (desc.count == 0) | |
1567 | continue; | |
1568 | desc.error = 0; | |
1569 | do_shmem_file_read(filp, ppos, &desc, file_read_actor); | |
1570 | retval += desc.written; | |
1571 | if (desc.error) { | |
1572 | retval = retval ?: desc.error; | |
1573 | break; | |
1574 | } | |
1575 | if (desc.count > 0) | |
1576 | break; | |
1577 | } | |
1578 | return retval; | |
1da177e4 LT |
1579 | } |
1580 | ||
708e3508 HD |
1581 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, |
1582 | struct pipe_inode_info *pipe, size_t len, | |
1583 | unsigned int flags) | |
1584 | { | |
1585 | struct address_space *mapping = in->f_mapping; | |
71f0e07a | 1586 | struct inode *inode = mapping->host; |
708e3508 HD |
1587 | unsigned int loff, nr_pages, req_pages; |
1588 | struct page *pages[PIPE_DEF_BUFFERS]; | |
1589 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
1590 | struct page *page; | |
1591 | pgoff_t index, end_index; | |
1592 | loff_t isize, left; | |
1593 | int error, page_nr; | |
1594 | struct splice_pipe_desc spd = { | |
1595 | .pages = pages, | |
1596 | .partial = partial, | |
047fe360 | 1597 | .nr_pages_max = PIPE_DEF_BUFFERS, |
708e3508 HD |
1598 | .flags = flags, |
1599 | .ops = &page_cache_pipe_buf_ops, | |
1600 | .spd_release = spd_release_page, | |
1601 | }; | |
1602 | ||
71f0e07a | 1603 | isize = i_size_read(inode); |
708e3508 HD |
1604 | if (unlikely(*ppos >= isize)) |
1605 | return 0; | |
1606 | ||
1607 | left = isize - *ppos; | |
1608 | if (unlikely(left < len)) | |
1609 | len = left; | |
1610 | ||
1611 | if (splice_grow_spd(pipe, &spd)) | |
1612 | return -ENOMEM; | |
1613 | ||
1614 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1615 | loff = *ppos & ~PAGE_CACHE_MASK; | |
1616 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
1617 | nr_pages = min(req_pages, pipe->buffers); | |
1618 | ||
708e3508 HD |
1619 | spd.nr_pages = find_get_pages_contig(mapping, index, |
1620 | nr_pages, spd.pages); | |
1621 | index += spd.nr_pages; | |
708e3508 | 1622 | error = 0; |
708e3508 | 1623 | |
71f0e07a | 1624 | while (spd.nr_pages < nr_pages) { |
71f0e07a HD |
1625 | error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL); |
1626 | if (error) | |
1627 | break; | |
1628 | unlock_page(page); | |
708e3508 HD |
1629 | spd.pages[spd.nr_pages++] = page; |
1630 | index++; | |
1631 | } | |
1632 | ||
708e3508 HD |
1633 | index = *ppos >> PAGE_CACHE_SHIFT; |
1634 | nr_pages = spd.nr_pages; | |
1635 | spd.nr_pages = 0; | |
71f0e07a | 1636 | |
708e3508 HD |
1637 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { |
1638 | unsigned int this_len; | |
1639 | ||
1640 | if (!len) | |
1641 | break; | |
1642 | ||
708e3508 HD |
1643 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); |
1644 | page = spd.pages[page_nr]; | |
1645 | ||
71f0e07a | 1646 | if (!PageUptodate(page) || page->mapping != mapping) { |
71f0e07a HD |
1647 | error = shmem_getpage(inode, index, &page, |
1648 | SGP_CACHE, NULL); | |
1649 | if (error) | |
708e3508 | 1650 | break; |
71f0e07a HD |
1651 | unlock_page(page); |
1652 | page_cache_release(spd.pages[page_nr]); | |
1653 | spd.pages[page_nr] = page; | |
708e3508 | 1654 | } |
71f0e07a HD |
1655 | |
1656 | isize = i_size_read(inode); | |
708e3508 HD |
1657 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; |
1658 | if (unlikely(!isize || index > end_index)) | |
1659 | break; | |
1660 | ||
708e3508 HD |
1661 | if (end_index == index) { |
1662 | unsigned int plen; | |
1663 | ||
708e3508 HD |
1664 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
1665 | if (plen <= loff) | |
1666 | break; | |
1667 | ||
708e3508 HD |
1668 | this_len = min(this_len, plen - loff); |
1669 | len = this_len; | |
1670 | } | |
1671 | ||
1672 | spd.partial[page_nr].offset = loff; | |
1673 | spd.partial[page_nr].len = this_len; | |
1674 | len -= this_len; | |
1675 | loff = 0; | |
1676 | spd.nr_pages++; | |
1677 | index++; | |
1678 | } | |
1679 | ||
708e3508 HD |
1680 | while (page_nr < nr_pages) |
1681 | page_cache_release(spd.pages[page_nr++]); | |
708e3508 HD |
1682 | |
1683 | if (spd.nr_pages) | |
1684 | error = splice_to_pipe(pipe, &spd); | |
1685 | ||
047fe360 | 1686 | splice_shrink_spd(&spd); |
708e3508 HD |
1687 | |
1688 | if (error > 0) { | |
1689 | *ppos += error; | |
1690 | file_accessed(in); | |
1691 | } | |
1692 | return error; | |
1693 | } | |
1694 | ||
4fb5ef08 HD |
1695 | /* |
1696 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
1697 | */ | |
1698 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
1699 | pgoff_t index, pgoff_t end, int origin) | |
1700 | { | |
1701 | struct page *page; | |
1702 | struct pagevec pvec; | |
1703 | pgoff_t indices[PAGEVEC_SIZE]; | |
1704 | bool done = false; | |
1705 | int i; | |
1706 | ||
1707 | pagevec_init(&pvec, 0); | |
1708 | pvec.nr = 1; /* start small: we may be there already */ | |
1709 | while (!done) { | |
1710 | pvec.nr = shmem_find_get_pages_and_swap(mapping, index, | |
1711 | pvec.nr, pvec.pages, indices); | |
1712 | if (!pvec.nr) { | |
1713 | if (origin == SEEK_DATA) | |
1714 | index = end; | |
1715 | break; | |
1716 | } | |
1717 | for (i = 0; i < pvec.nr; i++, index++) { | |
1718 | if (index < indices[i]) { | |
1719 | if (origin == SEEK_HOLE) { | |
1720 | done = true; | |
1721 | break; | |
1722 | } | |
1723 | index = indices[i]; | |
1724 | } | |
1725 | page = pvec.pages[i]; | |
1726 | if (page && !radix_tree_exceptional_entry(page)) { | |
1727 | if (!PageUptodate(page)) | |
1728 | page = NULL; | |
1729 | } | |
1730 | if (index >= end || | |
1731 | (page && origin == SEEK_DATA) || | |
1732 | (!page && origin == SEEK_HOLE)) { | |
1733 | done = true; | |
1734 | break; | |
1735 | } | |
1736 | } | |
1737 | shmem_deswap_pagevec(&pvec); | |
1738 | pagevec_release(&pvec); | |
1739 | pvec.nr = PAGEVEC_SIZE; | |
1740 | cond_resched(); | |
1741 | } | |
1742 | return index; | |
1743 | } | |
1744 | ||
1745 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int origin) | |
1746 | { | |
1747 | struct address_space *mapping; | |
1748 | struct inode *inode; | |
1749 | pgoff_t start, end; | |
1750 | loff_t new_offset; | |
1751 | ||
1752 | if (origin != SEEK_DATA && origin != SEEK_HOLE) | |
1753 | return generic_file_llseek_size(file, offset, origin, | |
1754 | MAX_LFS_FILESIZE); | |
1755 | mapping = file->f_mapping; | |
1756 | inode = mapping->host; | |
1757 | mutex_lock(&inode->i_mutex); | |
1758 | /* We're holding i_mutex so we can access i_size directly */ | |
1759 | ||
1760 | if (offset < 0) | |
1761 | offset = -EINVAL; | |
1762 | else if (offset >= inode->i_size) | |
1763 | offset = -ENXIO; | |
1764 | else { | |
1765 | start = offset >> PAGE_CACHE_SHIFT; | |
1766 | end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
1767 | new_offset = shmem_seek_hole_data(mapping, start, end, origin); | |
1768 | new_offset <<= PAGE_CACHE_SHIFT; | |
1769 | if (new_offset > offset) { | |
1770 | if (new_offset < inode->i_size) | |
1771 | offset = new_offset; | |
1772 | else if (origin == SEEK_DATA) | |
1773 | offset = -ENXIO; | |
1774 | else | |
1775 | offset = inode->i_size; | |
1776 | } | |
1777 | } | |
1778 | ||
1779 | if (offset >= 0 && offset != file->f_pos) { | |
1780 | file->f_pos = offset; | |
1781 | file->f_version = 0; | |
1782 | } | |
1783 | mutex_unlock(&inode->i_mutex); | |
1784 | return offset; | |
1785 | } | |
1786 | ||
83e4fa9c HD |
1787 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
1788 | loff_t len) | |
1789 | { | |
1790 | struct inode *inode = file->f_path.dentry->d_inode; | |
e2d12e22 | 1791 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1aac1400 | 1792 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
1793 | pgoff_t start, index, end; |
1794 | int error; | |
83e4fa9c HD |
1795 | |
1796 | mutex_lock(&inode->i_mutex); | |
1797 | ||
1798 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1799 | struct address_space *mapping = file->f_mapping; | |
1800 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
1801 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
1802 | ||
1803 | if ((u64)unmap_end > (u64)unmap_start) | |
1804 | unmap_mapping_range(mapping, unmap_start, | |
1805 | 1 + unmap_end - unmap_start, 0); | |
1806 | shmem_truncate_range(inode, offset, offset + len - 1); | |
1807 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
1808 | error = 0; | |
e2d12e22 HD |
1809 | goto out; |
1810 | } | |
1811 | ||
1812 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
1813 | error = inode_newsize_ok(inode, offset + len); | |
1814 | if (error) | |
1815 | goto out; | |
1816 | ||
1817 | start = offset >> PAGE_CACHE_SHIFT; | |
1818 | end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
1819 | /* Try to avoid a swapstorm if len is impossible to satisfy */ | |
1820 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
1821 | error = -ENOSPC; | |
1822 | goto out; | |
83e4fa9c HD |
1823 | } |
1824 | ||
1aac1400 HD |
1825 | shmem_falloc.start = start; |
1826 | shmem_falloc.next = start; | |
1827 | shmem_falloc.nr_falloced = 0; | |
1828 | shmem_falloc.nr_unswapped = 0; | |
1829 | spin_lock(&inode->i_lock); | |
1830 | inode->i_private = &shmem_falloc; | |
1831 | spin_unlock(&inode->i_lock); | |
1832 | ||
e2d12e22 HD |
1833 | for (index = start; index < end; index++) { |
1834 | struct page *page; | |
1835 | ||
1836 | /* | |
1837 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
1838 | * been interrupted because we are using up too much memory. | |
1839 | */ | |
1840 | if (signal_pending(current)) | |
1841 | error = -EINTR; | |
1aac1400 HD |
1842 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
1843 | error = -ENOMEM; | |
e2d12e22 | 1844 | else |
1635f6a7 | 1845 | error = shmem_getpage(inode, index, &page, SGP_FALLOC, |
e2d12e22 HD |
1846 | NULL); |
1847 | if (error) { | |
1635f6a7 HD |
1848 | /* Remove the !PageUptodate pages we added */ |
1849 | shmem_undo_range(inode, | |
1850 | (loff_t)start << PAGE_CACHE_SHIFT, | |
1851 | (loff_t)index << PAGE_CACHE_SHIFT, true); | |
1aac1400 | 1852 | goto undone; |
e2d12e22 HD |
1853 | } |
1854 | ||
1aac1400 HD |
1855 | /* |
1856 | * Inform shmem_writepage() how far we have reached. | |
1857 | * No need for lock or barrier: we have the page lock. | |
1858 | */ | |
1859 | shmem_falloc.next++; | |
1860 | if (!PageUptodate(page)) | |
1861 | shmem_falloc.nr_falloced++; | |
1862 | ||
e2d12e22 | 1863 | /* |
1635f6a7 HD |
1864 | * If !PageUptodate, leave it that way so that freeable pages |
1865 | * can be recognized if we need to rollback on error later. | |
1866 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
1867 | * than free the pages we are allocating (and SGP_CACHE pages |
1868 | * might still be clean: we now need to mark those dirty too). | |
1869 | */ | |
1870 | set_page_dirty(page); | |
1871 | unlock_page(page); | |
1872 | page_cache_release(page); | |
1873 | cond_resched(); | |
1874 | } | |
1875 | ||
1876 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
1877 | i_size_write(inode, offset + len); | |
e2d12e22 | 1878 | inode->i_ctime = CURRENT_TIME; |
1aac1400 HD |
1879 | undone: |
1880 | spin_lock(&inode->i_lock); | |
1881 | inode->i_private = NULL; | |
1882 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 1883 | out: |
83e4fa9c HD |
1884 | mutex_unlock(&inode->i_mutex); |
1885 | return error; | |
1886 | } | |
1887 | ||
726c3342 | 1888 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 1889 | { |
726c3342 | 1890 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
1891 | |
1892 | buf->f_type = TMPFS_MAGIC; | |
1893 | buf->f_bsize = PAGE_CACHE_SIZE; | |
1894 | buf->f_namelen = NAME_MAX; | |
0edd73b3 | 1895 | if (sbinfo->max_blocks) { |
1da177e4 | 1896 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
1897 | buf->f_bavail = |
1898 | buf->f_bfree = sbinfo->max_blocks - | |
1899 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
1900 | } |
1901 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
1902 | buf->f_files = sbinfo->max_inodes; |
1903 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
1904 | } |
1905 | /* else leave those fields 0 like simple_statfs */ | |
1906 | return 0; | |
1907 | } | |
1908 | ||
1909 | /* | |
1910 | * File creation. Allocate an inode, and we're done.. | |
1911 | */ | |
1912 | static int | |
1a67aafb | 1913 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 1914 | { |
0b0a0806 | 1915 | struct inode *inode; |
1da177e4 LT |
1916 | int error = -ENOSPC; |
1917 | ||
454abafe | 1918 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 1919 | if (inode) { |
2a7dba39 | 1920 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 1921 | &dentry->d_name, |
6d9d88d0 | 1922 | shmem_initxattrs, NULL); |
570bc1c2 SS |
1923 | if (error) { |
1924 | if (error != -EOPNOTSUPP) { | |
1925 | iput(inode); | |
1926 | return error; | |
1927 | } | |
39f0247d | 1928 | } |
1c7c474c CH |
1929 | #ifdef CONFIG_TMPFS_POSIX_ACL |
1930 | error = generic_acl_init(inode, dir); | |
39f0247d AG |
1931 | if (error) { |
1932 | iput(inode); | |
1933 | return error; | |
570bc1c2 | 1934 | } |
718deb6b AV |
1935 | #else |
1936 | error = 0; | |
1c7c474c | 1937 | #endif |
1da177e4 LT |
1938 | dir->i_size += BOGO_DIRENT_SIZE; |
1939 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
1940 | d_instantiate(dentry, inode); | |
1941 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
1942 | } |
1943 | return error; | |
1944 | } | |
1945 | ||
18bb1db3 | 1946 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
1947 | { |
1948 | int error; | |
1949 | ||
1950 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
1951 | return error; | |
d8c76e6f | 1952 | inc_nlink(dir); |
1da177e4 LT |
1953 | return 0; |
1954 | } | |
1955 | ||
4acdaf27 | 1956 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
1da177e4 LT |
1957 | struct nameidata *nd) |
1958 | { | |
1959 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
1960 | } | |
1961 | ||
1962 | /* | |
1963 | * Link a file.. | |
1964 | */ | |
1965 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1966 | { | |
1967 | struct inode *inode = old_dentry->d_inode; | |
5b04c689 | 1968 | int ret; |
1da177e4 LT |
1969 | |
1970 | /* | |
1971 | * No ordinary (disk based) filesystem counts links as inodes; | |
1972 | * but each new link needs a new dentry, pinning lowmem, and | |
1973 | * tmpfs dentries cannot be pruned until they are unlinked. | |
1974 | */ | |
5b04c689 PE |
1975 | ret = shmem_reserve_inode(inode->i_sb); |
1976 | if (ret) | |
1977 | goto out; | |
1da177e4 LT |
1978 | |
1979 | dir->i_size += BOGO_DIRENT_SIZE; | |
1980 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 1981 | inc_nlink(inode); |
7de9c6ee | 1982 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
1983 | dget(dentry); /* Extra pinning count for the created dentry */ |
1984 | d_instantiate(dentry, inode); | |
5b04c689 PE |
1985 | out: |
1986 | return ret; | |
1da177e4 LT |
1987 | } |
1988 | ||
1989 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
1990 | { | |
1991 | struct inode *inode = dentry->d_inode; | |
1992 | ||
5b04c689 PE |
1993 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
1994 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
1995 | |
1996 | dir->i_size -= BOGO_DIRENT_SIZE; | |
1997 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 1998 | drop_nlink(inode); |
1da177e4 LT |
1999 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
2000 | return 0; | |
2001 | } | |
2002 | ||
2003 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
2004 | { | |
2005 | if (!simple_empty(dentry)) | |
2006 | return -ENOTEMPTY; | |
2007 | ||
9a53c3a7 DH |
2008 | drop_nlink(dentry->d_inode); |
2009 | drop_nlink(dir); | |
1da177e4 LT |
2010 | return shmem_unlink(dir, dentry); |
2011 | } | |
2012 | ||
2013 | /* | |
2014 | * The VFS layer already does all the dentry stuff for rename, | |
2015 | * we just have to decrement the usage count for the target if | |
2016 | * it exists so that the VFS layer correctly free's it when it | |
2017 | * gets overwritten. | |
2018 | */ | |
2019 | static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) | |
2020 | { | |
2021 | struct inode *inode = old_dentry->d_inode; | |
2022 | int they_are_dirs = S_ISDIR(inode->i_mode); | |
2023 | ||
2024 | if (!simple_empty(new_dentry)) | |
2025 | return -ENOTEMPTY; | |
2026 | ||
2027 | if (new_dentry->d_inode) { | |
2028 | (void) shmem_unlink(new_dir, new_dentry); | |
2029 | if (they_are_dirs) | |
9a53c3a7 | 2030 | drop_nlink(old_dir); |
1da177e4 | 2031 | } else if (they_are_dirs) { |
9a53c3a7 | 2032 | drop_nlink(old_dir); |
d8c76e6f | 2033 | inc_nlink(new_dir); |
1da177e4 LT |
2034 | } |
2035 | ||
2036 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
2037 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
2038 | old_dir->i_ctime = old_dir->i_mtime = | |
2039 | new_dir->i_ctime = new_dir->i_mtime = | |
2040 | inode->i_ctime = CURRENT_TIME; | |
2041 | return 0; | |
2042 | } | |
2043 | ||
2044 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
2045 | { | |
2046 | int error; | |
2047 | int len; | |
2048 | struct inode *inode; | |
9276aad6 | 2049 | struct page *page; |
1da177e4 LT |
2050 | char *kaddr; |
2051 | struct shmem_inode_info *info; | |
2052 | ||
2053 | len = strlen(symname) + 1; | |
2054 | if (len > PAGE_CACHE_SIZE) | |
2055 | return -ENAMETOOLONG; | |
2056 | ||
454abafe | 2057 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
2058 | if (!inode) |
2059 | return -ENOSPC; | |
2060 | ||
9d8f13ba | 2061 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 2062 | shmem_initxattrs, NULL); |
570bc1c2 SS |
2063 | if (error) { |
2064 | if (error != -EOPNOTSUPP) { | |
2065 | iput(inode); | |
2066 | return error; | |
2067 | } | |
2068 | error = 0; | |
2069 | } | |
2070 | ||
1da177e4 LT |
2071 | info = SHMEM_I(inode); |
2072 | inode->i_size = len-1; | |
69f07ec9 HD |
2073 | if (len <= SHORT_SYMLINK_LEN) { |
2074 | info->symlink = kmemdup(symname, len, GFP_KERNEL); | |
2075 | if (!info->symlink) { | |
2076 | iput(inode); | |
2077 | return -ENOMEM; | |
2078 | } | |
2079 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 LT |
2080 | } else { |
2081 | error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); | |
2082 | if (error) { | |
2083 | iput(inode); | |
2084 | return error; | |
2085 | } | |
14fcc23f | 2086 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 2087 | inode->i_op = &shmem_symlink_inode_operations; |
9b04c5fe | 2088 | kaddr = kmap_atomic(page); |
1da177e4 | 2089 | memcpy(kaddr, symname, len); |
9b04c5fe | 2090 | kunmap_atomic(kaddr); |
ec9516fb | 2091 | SetPageUptodate(page); |
1da177e4 | 2092 | set_page_dirty(page); |
6746aff7 | 2093 | unlock_page(page); |
1da177e4 LT |
2094 | page_cache_release(page); |
2095 | } | |
1da177e4 LT |
2096 | dir->i_size += BOGO_DIRENT_SIZE; |
2097 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2098 | d_instantiate(dentry, inode); | |
2099 | dget(dentry); | |
2100 | return 0; | |
2101 | } | |
2102 | ||
69f07ec9 | 2103 | static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 2104 | { |
69f07ec9 | 2105 | nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink); |
cc314eef | 2106 | return NULL; |
1da177e4 LT |
2107 | } |
2108 | ||
cc314eef | 2109 | static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) |
1da177e4 LT |
2110 | { |
2111 | struct page *page = NULL; | |
41ffe5d5 HD |
2112 | int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); |
2113 | nd_set_link(nd, error ? ERR_PTR(error) : kmap(page)); | |
d3602444 HD |
2114 | if (page) |
2115 | unlock_page(page); | |
cc314eef | 2116 | return page; |
1da177e4 LT |
2117 | } |
2118 | ||
cc314eef | 2119 | static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 LT |
2120 | { |
2121 | if (!IS_ERR(nd_get_link(nd))) { | |
cc314eef | 2122 | struct page *page = cookie; |
1da177e4 LT |
2123 | kunmap(page); |
2124 | mark_page_accessed(page); | |
2125 | page_cache_release(page); | |
1da177e4 LT |
2126 | } |
2127 | } | |
2128 | ||
b09e0fa4 | 2129 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 2130 | /* |
b09e0fa4 EP |
2131 | * Superblocks without xattr inode operations may get some security.* xattr |
2132 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
2133 | * like ACLs, we also need to implement the security.* handlers at |
2134 | * filesystem level, though. | |
2135 | */ | |
2136 | ||
6d9d88d0 JS |
2137 | /* |
2138 | * Allocate new xattr and copy in the value; but leave the name to callers. | |
2139 | */ | |
2140 | static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size) | |
2141 | { | |
2142 | struct shmem_xattr *new_xattr; | |
2143 | size_t len; | |
2144 | ||
2145 | /* wrap around? */ | |
2146 | len = sizeof(*new_xattr) + size; | |
2147 | if (len <= sizeof(*new_xattr)) | |
2148 | return NULL; | |
2149 | ||
2150 | new_xattr = kmalloc(len, GFP_KERNEL); | |
2151 | if (!new_xattr) | |
2152 | return NULL; | |
2153 | ||
2154 | new_xattr->size = size; | |
2155 | memcpy(new_xattr->value, value, size); | |
2156 | return new_xattr; | |
2157 | } | |
2158 | ||
2159 | /* | |
2160 | * Callback for security_inode_init_security() for acquiring xattrs. | |
2161 | */ | |
2162 | static int shmem_initxattrs(struct inode *inode, | |
2163 | const struct xattr *xattr_array, | |
2164 | void *fs_info) | |
2165 | { | |
2166 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2167 | const struct xattr *xattr; | |
2168 | struct shmem_xattr *new_xattr; | |
2169 | size_t len; | |
2170 | ||
2171 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
2172 | new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len); | |
2173 | if (!new_xattr) | |
2174 | return -ENOMEM; | |
2175 | ||
2176 | len = strlen(xattr->name) + 1; | |
2177 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
2178 | GFP_KERNEL); | |
2179 | if (!new_xattr->name) { | |
2180 | kfree(new_xattr); | |
2181 | return -ENOMEM; | |
2182 | } | |
2183 | ||
2184 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
2185 | XATTR_SECURITY_PREFIX_LEN); | |
2186 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
2187 | xattr->name, len); | |
2188 | ||
2189 | spin_lock(&info->lock); | |
2190 | list_add(&new_xattr->list, &info->xattr_list); | |
2191 | spin_unlock(&info->lock); | |
2192 | } | |
2193 | ||
2194 | return 0; | |
2195 | } | |
2196 | ||
b09e0fa4 EP |
2197 | static int shmem_xattr_get(struct dentry *dentry, const char *name, |
2198 | void *buffer, size_t size) | |
39f0247d | 2199 | { |
b09e0fa4 EP |
2200 | struct shmem_inode_info *info; |
2201 | struct shmem_xattr *xattr; | |
2202 | int ret = -ENODATA; | |
39f0247d | 2203 | |
b09e0fa4 EP |
2204 | info = SHMEM_I(dentry->d_inode); |
2205 | ||
2206 | spin_lock(&info->lock); | |
2207 | list_for_each_entry(xattr, &info->xattr_list, list) { | |
2208 | if (strcmp(name, xattr->name)) | |
2209 | continue; | |
2210 | ||
2211 | ret = xattr->size; | |
2212 | if (buffer) { | |
2213 | if (size < xattr->size) | |
2214 | ret = -ERANGE; | |
2215 | else | |
2216 | memcpy(buffer, xattr->value, xattr->size); | |
2217 | } | |
2218 | break; | |
2219 | } | |
2220 | spin_unlock(&info->lock); | |
2221 | return ret; | |
39f0247d AG |
2222 | } |
2223 | ||
6d9d88d0 | 2224 | static int shmem_xattr_set(struct inode *inode, const char *name, |
b09e0fa4 | 2225 | const void *value, size_t size, int flags) |
39f0247d | 2226 | { |
b09e0fa4 EP |
2227 | struct shmem_inode_info *info = SHMEM_I(inode); |
2228 | struct shmem_xattr *xattr; | |
2229 | struct shmem_xattr *new_xattr = NULL; | |
b09e0fa4 EP |
2230 | int err = 0; |
2231 | ||
2232 | /* value == NULL means remove */ | |
2233 | if (value) { | |
6d9d88d0 | 2234 | new_xattr = shmem_xattr_alloc(value, size); |
b09e0fa4 EP |
2235 | if (!new_xattr) |
2236 | return -ENOMEM; | |
2237 | ||
2238 | new_xattr->name = kstrdup(name, GFP_KERNEL); | |
2239 | if (!new_xattr->name) { | |
2240 | kfree(new_xattr); | |
2241 | return -ENOMEM; | |
2242 | } | |
b09e0fa4 EP |
2243 | } |
2244 | ||
2245 | spin_lock(&info->lock); | |
2246 | list_for_each_entry(xattr, &info->xattr_list, list) { | |
2247 | if (!strcmp(name, xattr->name)) { | |
2248 | if (flags & XATTR_CREATE) { | |
2249 | xattr = new_xattr; | |
2250 | err = -EEXIST; | |
2251 | } else if (new_xattr) { | |
2252 | list_replace(&xattr->list, &new_xattr->list); | |
2253 | } else { | |
2254 | list_del(&xattr->list); | |
2255 | } | |
2256 | goto out; | |
2257 | } | |
2258 | } | |
2259 | if (flags & XATTR_REPLACE) { | |
2260 | xattr = new_xattr; | |
2261 | err = -ENODATA; | |
2262 | } else { | |
2263 | list_add(&new_xattr->list, &info->xattr_list); | |
2264 | xattr = NULL; | |
2265 | } | |
2266 | out: | |
2267 | spin_unlock(&info->lock); | |
2268 | if (xattr) | |
2269 | kfree(xattr->name); | |
2270 | kfree(xattr); | |
2271 | return err; | |
39f0247d AG |
2272 | } |
2273 | ||
bb435453 | 2274 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
b09e0fa4 | 2275 | #ifdef CONFIG_TMPFS_POSIX_ACL |
1c7c474c CH |
2276 | &generic_acl_access_handler, |
2277 | &generic_acl_default_handler, | |
b09e0fa4 | 2278 | #endif |
39f0247d AG |
2279 | NULL |
2280 | }; | |
b09e0fa4 EP |
2281 | |
2282 | static int shmem_xattr_validate(const char *name) | |
2283 | { | |
2284 | struct { const char *prefix; size_t len; } arr[] = { | |
2285 | { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN }, | |
2286 | { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN } | |
2287 | }; | |
2288 | int i; | |
2289 | ||
2290 | for (i = 0; i < ARRAY_SIZE(arr); i++) { | |
2291 | size_t preflen = arr[i].len; | |
2292 | if (strncmp(name, arr[i].prefix, preflen) == 0) { | |
2293 | if (!name[preflen]) | |
2294 | return -EINVAL; | |
2295 | return 0; | |
2296 | } | |
2297 | } | |
2298 | return -EOPNOTSUPP; | |
2299 | } | |
2300 | ||
2301 | static ssize_t shmem_getxattr(struct dentry *dentry, const char *name, | |
2302 | void *buffer, size_t size) | |
2303 | { | |
2304 | int err; | |
2305 | ||
2306 | /* | |
2307 | * If this is a request for a synthetic attribute in the system.* | |
2308 | * namespace use the generic infrastructure to resolve a handler | |
2309 | * for it via sb->s_xattr. | |
2310 | */ | |
2311 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2312 | return generic_getxattr(dentry, name, buffer, size); | |
2313 | ||
2314 | err = shmem_xattr_validate(name); | |
2315 | if (err) | |
2316 | return err; | |
2317 | ||
2318 | return shmem_xattr_get(dentry, name, buffer, size); | |
2319 | } | |
2320 | ||
2321 | static int shmem_setxattr(struct dentry *dentry, const char *name, | |
2322 | const void *value, size_t size, int flags) | |
2323 | { | |
2324 | int err; | |
2325 | ||
2326 | /* | |
2327 | * If this is a request for a synthetic attribute in the system.* | |
2328 | * namespace use the generic infrastructure to resolve a handler | |
2329 | * for it via sb->s_xattr. | |
2330 | */ | |
2331 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2332 | return generic_setxattr(dentry, name, value, size, flags); | |
2333 | ||
2334 | err = shmem_xattr_validate(name); | |
2335 | if (err) | |
2336 | return err; | |
2337 | ||
2338 | if (size == 0) | |
2339 | value = ""; /* empty EA, do not remove */ | |
2340 | ||
6d9d88d0 | 2341 | return shmem_xattr_set(dentry->d_inode, name, value, size, flags); |
b09e0fa4 EP |
2342 | |
2343 | } | |
2344 | ||
2345 | static int shmem_removexattr(struct dentry *dentry, const char *name) | |
2346 | { | |
2347 | int err; | |
2348 | ||
2349 | /* | |
2350 | * If this is a request for a synthetic attribute in the system.* | |
2351 | * namespace use the generic infrastructure to resolve a handler | |
2352 | * for it via sb->s_xattr. | |
2353 | */ | |
2354 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2355 | return generic_removexattr(dentry, name); | |
2356 | ||
2357 | err = shmem_xattr_validate(name); | |
2358 | if (err) | |
2359 | return err; | |
2360 | ||
6d9d88d0 | 2361 | return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE); |
b09e0fa4 EP |
2362 | } |
2363 | ||
2364 | static bool xattr_is_trusted(const char *name) | |
2365 | { | |
2366 | return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN); | |
2367 | } | |
2368 | ||
2369 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
2370 | { | |
2371 | bool trusted = capable(CAP_SYS_ADMIN); | |
2372 | struct shmem_xattr *xattr; | |
2373 | struct shmem_inode_info *info; | |
2374 | size_t used = 0; | |
2375 | ||
2376 | info = SHMEM_I(dentry->d_inode); | |
2377 | ||
2378 | spin_lock(&info->lock); | |
2379 | list_for_each_entry(xattr, &info->xattr_list, list) { | |
2380 | size_t len; | |
2381 | ||
2382 | /* skip "trusted." attributes for unprivileged callers */ | |
2383 | if (!trusted && xattr_is_trusted(xattr->name)) | |
2384 | continue; | |
2385 | ||
2386 | len = strlen(xattr->name) + 1; | |
2387 | used += len; | |
2388 | if (buffer) { | |
2389 | if (size < used) { | |
2390 | used = -ERANGE; | |
2391 | break; | |
2392 | } | |
2393 | memcpy(buffer, xattr->name, len); | |
2394 | buffer += len; | |
2395 | } | |
2396 | } | |
2397 | spin_unlock(&info->lock); | |
2398 | ||
2399 | return used; | |
2400 | } | |
2401 | #endif /* CONFIG_TMPFS_XATTR */ | |
2402 | ||
69f07ec9 | 2403 | static const struct inode_operations shmem_short_symlink_operations = { |
b09e0fa4 | 2404 | .readlink = generic_readlink, |
69f07ec9 | 2405 | .follow_link = shmem_follow_short_symlink, |
b09e0fa4 EP |
2406 | #ifdef CONFIG_TMPFS_XATTR |
2407 | .setxattr = shmem_setxattr, | |
2408 | .getxattr = shmem_getxattr, | |
2409 | .listxattr = shmem_listxattr, | |
2410 | .removexattr = shmem_removexattr, | |
2411 | #endif | |
2412 | }; | |
2413 | ||
2414 | static const struct inode_operations shmem_symlink_inode_operations = { | |
2415 | .readlink = generic_readlink, | |
2416 | .follow_link = shmem_follow_link, | |
2417 | .put_link = shmem_put_link, | |
2418 | #ifdef CONFIG_TMPFS_XATTR | |
2419 | .setxattr = shmem_setxattr, | |
2420 | .getxattr = shmem_getxattr, | |
2421 | .listxattr = shmem_listxattr, | |
2422 | .removexattr = shmem_removexattr, | |
39f0247d | 2423 | #endif |
b09e0fa4 | 2424 | }; |
39f0247d | 2425 | |
91828a40 DG |
2426 | static struct dentry *shmem_get_parent(struct dentry *child) |
2427 | { | |
2428 | return ERR_PTR(-ESTALE); | |
2429 | } | |
2430 | ||
2431 | static int shmem_match(struct inode *ino, void *vfh) | |
2432 | { | |
2433 | __u32 *fh = vfh; | |
2434 | __u64 inum = fh[2]; | |
2435 | inum = (inum << 32) | fh[1]; | |
2436 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
2437 | } | |
2438 | ||
480b116c CH |
2439 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
2440 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 2441 | { |
91828a40 | 2442 | struct inode *inode; |
480b116c CH |
2443 | struct dentry *dentry = NULL; |
2444 | u64 inum = fid->raw[2]; | |
2445 | inum = (inum << 32) | fid->raw[1]; | |
2446 | ||
2447 | if (fh_len < 3) | |
2448 | return NULL; | |
91828a40 | 2449 | |
480b116c CH |
2450 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
2451 | shmem_match, fid->raw); | |
91828a40 | 2452 | if (inode) { |
480b116c | 2453 | dentry = d_find_alias(inode); |
91828a40 DG |
2454 | iput(inode); |
2455 | } | |
2456 | ||
480b116c | 2457 | return dentry; |
91828a40 DG |
2458 | } |
2459 | ||
b0b0382b AV |
2460 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
2461 | struct inode *parent) | |
91828a40 | 2462 | { |
5fe0c237 AK |
2463 | if (*len < 3) { |
2464 | *len = 3; | |
91828a40 | 2465 | return 255; |
5fe0c237 | 2466 | } |
91828a40 | 2467 | |
1d3382cb | 2468 | if (inode_unhashed(inode)) { |
91828a40 DG |
2469 | /* Unfortunately insert_inode_hash is not idempotent, |
2470 | * so as we hash inodes here rather than at creation | |
2471 | * time, we need a lock to ensure we only try | |
2472 | * to do it once | |
2473 | */ | |
2474 | static DEFINE_SPINLOCK(lock); | |
2475 | spin_lock(&lock); | |
1d3382cb | 2476 | if (inode_unhashed(inode)) |
91828a40 DG |
2477 | __insert_inode_hash(inode, |
2478 | inode->i_ino + inode->i_generation); | |
2479 | spin_unlock(&lock); | |
2480 | } | |
2481 | ||
2482 | fh[0] = inode->i_generation; | |
2483 | fh[1] = inode->i_ino; | |
2484 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
2485 | ||
2486 | *len = 3; | |
2487 | return 1; | |
2488 | } | |
2489 | ||
39655164 | 2490 | static const struct export_operations shmem_export_ops = { |
91828a40 | 2491 | .get_parent = shmem_get_parent, |
91828a40 | 2492 | .encode_fh = shmem_encode_fh, |
480b116c | 2493 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
2494 | }; |
2495 | ||
680d794b AM |
2496 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
2497 | bool remount) | |
1da177e4 LT |
2498 | { |
2499 | char *this_char, *value, *rest; | |
8751e039 EB |
2500 | uid_t uid; |
2501 | gid_t gid; | |
1da177e4 | 2502 | |
b00dc3ad HD |
2503 | while (options != NULL) { |
2504 | this_char = options; | |
2505 | for (;;) { | |
2506 | /* | |
2507 | * NUL-terminate this option: unfortunately, | |
2508 | * mount options form a comma-separated list, | |
2509 | * but mpol's nodelist may also contain commas. | |
2510 | */ | |
2511 | options = strchr(options, ','); | |
2512 | if (options == NULL) | |
2513 | break; | |
2514 | options++; | |
2515 | if (!isdigit(*options)) { | |
2516 | options[-1] = '\0'; | |
2517 | break; | |
2518 | } | |
2519 | } | |
1da177e4 LT |
2520 | if (!*this_char) |
2521 | continue; | |
2522 | if ((value = strchr(this_char,'=')) != NULL) { | |
2523 | *value++ = 0; | |
2524 | } else { | |
2525 | printk(KERN_ERR | |
2526 | "tmpfs: No value for mount option '%s'\n", | |
2527 | this_char); | |
2528 | return 1; | |
2529 | } | |
2530 | ||
2531 | if (!strcmp(this_char,"size")) { | |
2532 | unsigned long long size; | |
2533 | size = memparse(value,&rest); | |
2534 | if (*rest == '%') { | |
2535 | size <<= PAGE_SHIFT; | |
2536 | size *= totalram_pages; | |
2537 | do_div(size, 100); | |
2538 | rest++; | |
2539 | } | |
2540 | if (*rest) | |
2541 | goto bad_val; | |
680d794b AM |
2542 | sbinfo->max_blocks = |
2543 | DIV_ROUND_UP(size, PAGE_CACHE_SIZE); | |
1da177e4 | 2544 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 2545 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
2546 | if (*rest) |
2547 | goto bad_val; | |
2548 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 2549 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
2550 | if (*rest) |
2551 | goto bad_val; | |
2552 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 2553 | if (remount) |
1da177e4 | 2554 | continue; |
680d794b | 2555 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
2556 | if (*rest) |
2557 | goto bad_val; | |
2558 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 2559 | if (remount) |
1da177e4 | 2560 | continue; |
8751e039 | 2561 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2562 | if (*rest) |
2563 | goto bad_val; | |
8751e039 EB |
2564 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
2565 | if (!uid_valid(sbinfo->uid)) | |
2566 | goto bad_val; | |
1da177e4 | 2567 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 2568 | if (remount) |
1da177e4 | 2569 | continue; |
8751e039 | 2570 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2571 | if (*rest) |
2572 | goto bad_val; | |
8751e039 EB |
2573 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
2574 | if (!gid_valid(sbinfo->gid)) | |
2575 | goto bad_val; | |
7339ff83 | 2576 | } else if (!strcmp(this_char,"mpol")) { |
71fe804b | 2577 | if (mpol_parse_str(value, &sbinfo->mpol, 1)) |
7339ff83 | 2578 | goto bad_val; |
1da177e4 LT |
2579 | } else { |
2580 | printk(KERN_ERR "tmpfs: Bad mount option %s\n", | |
2581 | this_char); | |
2582 | return 1; | |
2583 | } | |
2584 | } | |
2585 | return 0; | |
2586 | ||
2587 | bad_val: | |
2588 | printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", | |
2589 | value, this_char); | |
2590 | return 1; | |
2591 | ||
2592 | } | |
2593 | ||
2594 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
2595 | { | |
2596 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 2597 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
2598 | unsigned long inodes; |
2599 | int error = -EINVAL; | |
2600 | ||
680d794b | 2601 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 2602 | return error; |
1da177e4 | 2603 | |
0edd73b3 | 2604 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 2605 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 2606 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 2607 | goto out; |
680d794b | 2608 | if (config.max_inodes < inodes) |
0edd73b3 HD |
2609 | goto out; |
2610 | /* | |
54af6042 | 2611 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
2612 | * but we must separately disallow unlimited->limited, because |
2613 | * in that case we have no record of how much is already in use. | |
2614 | */ | |
680d794b | 2615 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 2616 | goto out; |
680d794b | 2617 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
2618 | goto out; |
2619 | ||
2620 | error = 0; | |
680d794b | 2621 | sbinfo->max_blocks = config.max_blocks; |
680d794b AM |
2622 | sbinfo->max_inodes = config.max_inodes; |
2623 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b LS |
2624 | |
2625 | mpol_put(sbinfo->mpol); | |
2626 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
0edd73b3 HD |
2627 | out: |
2628 | spin_unlock(&sbinfo->stat_lock); | |
2629 | return error; | |
1da177e4 | 2630 | } |
680d794b | 2631 | |
34c80b1d | 2632 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 2633 | { |
34c80b1d | 2634 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b AM |
2635 | |
2636 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
2637 | seq_printf(seq, ",size=%luk", | |
2638 | sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); | |
2639 | if (sbinfo->max_inodes != shmem_default_max_inodes()) | |
2640 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
2641 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 2642 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
2643 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
2644 | seq_printf(seq, ",uid=%u", | |
2645 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
2646 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
2647 | seq_printf(seq, ",gid=%u", | |
2648 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
71fe804b | 2649 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b AM |
2650 | return 0; |
2651 | } | |
2652 | #endif /* CONFIG_TMPFS */ | |
1da177e4 LT |
2653 | |
2654 | static void shmem_put_super(struct super_block *sb) | |
2655 | { | |
602586a8 HD |
2656 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
2657 | ||
2658 | percpu_counter_destroy(&sbinfo->used_blocks); | |
2659 | kfree(sbinfo); | |
1da177e4 LT |
2660 | sb->s_fs_info = NULL; |
2661 | } | |
2662 | ||
2b2af54a | 2663 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
2664 | { |
2665 | struct inode *inode; | |
0edd73b3 | 2666 | struct shmem_sb_info *sbinfo; |
680d794b AM |
2667 | int err = -ENOMEM; |
2668 | ||
2669 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 2670 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b AM |
2671 | L1_CACHE_BYTES), GFP_KERNEL); |
2672 | if (!sbinfo) | |
2673 | return -ENOMEM; | |
2674 | ||
680d794b | 2675 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
2676 | sbinfo->uid = current_fsuid(); |
2677 | sbinfo->gid = current_fsgid(); | |
680d794b | 2678 | sb->s_fs_info = sbinfo; |
1da177e4 | 2679 | |
0edd73b3 | 2680 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
2681 | /* |
2682 | * Per default we only allow half of the physical ram per | |
2683 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
2684 | * but the internal instance is left unlimited. | |
2685 | */ | |
2686 | if (!(sb->s_flags & MS_NOUSER)) { | |
680d794b AM |
2687 | sbinfo->max_blocks = shmem_default_max_blocks(); |
2688 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
2689 | if (shmem_parse_options(data, sbinfo, false)) { | |
2690 | err = -EINVAL; | |
2691 | goto failed; | |
2692 | } | |
1da177e4 | 2693 | } |
91828a40 | 2694 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 2695 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
2696 | #else |
2697 | sb->s_flags |= MS_NOUSER; | |
2698 | #endif | |
2699 | ||
0edd73b3 | 2700 | spin_lock_init(&sbinfo->stat_lock); |
602586a8 HD |
2701 | if (percpu_counter_init(&sbinfo->used_blocks, 0)) |
2702 | goto failed; | |
680d794b | 2703 | sbinfo->free_inodes = sbinfo->max_inodes; |
0edd73b3 | 2704 | |
285b2c4f | 2705 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
1da177e4 LT |
2706 | sb->s_blocksize = PAGE_CACHE_SIZE; |
2707 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
2708 | sb->s_magic = TMPFS_MAGIC; | |
2709 | sb->s_op = &shmem_ops; | |
cfd95a9c | 2710 | sb->s_time_gran = 1; |
b09e0fa4 | 2711 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 2712 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
2713 | #endif |
2714 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
2715 | sb->s_flags |= MS_POSIXACL; |
2716 | #endif | |
0edd73b3 | 2717 | |
454abafe | 2718 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
2719 | if (!inode) |
2720 | goto failed; | |
680d794b AM |
2721 | inode->i_uid = sbinfo->uid; |
2722 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
2723 | sb->s_root = d_make_root(inode); |
2724 | if (!sb->s_root) | |
48fde701 | 2725 | goto failed; |
1da177e4 LT |
2726 | return 0; |
2727 | ||
1da177e4 LT |
2728 | failed: |
2729 | shmem_put_super(sb); | |
2730 | return err; | |
2731 | } | |
2732 | ||
fcc234f8 | 2733 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
2734 | |
2735 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
2736 | { | |
41ffe5d5 HD |
2737 | struct shmem_inode_info *info; |
2738 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
2739 | if (!info) | |
1da177e4 | 2740 | return NULL; |
41ffe5d5 | 2741 | return &info->vfs_inode; |
1da177e4 LT |
2742 | } |
2743 | ||
41ffe5d5 | 2744 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
2745 | { |
2746 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
2747 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
2748 | } | |
2749 | ||
1da177e4 LT |
2750 | static void shmem_destroy_inode(struct inode *inode) |
2751 | { | |
09208d15 | 2752 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 2753 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 2754 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
2755 | } |
2756 | ||
41ffe5d5 | 2757 | static void shmem_init_inode(void *foo) |
1da177e4 | 2758 | { |
41ffe5d5 HD |
2759 | struct shmem_inode_info *info = foo; |
2760 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
2761 | } |
2762 | ||
41ffe5d5 | 2763 | static int shmem_init_inodecache(void) |
1da177e4 LT |
2764 | { |
2765 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
2766 | sizeof(struct shmem_inode_info), | |
41ffe5d5 | 2767 | 0, SLAB_PANIC, shmem_init_inode); |
1da177e4 LT |
2768 | return 0; |
2769 | } | |
2770 | ||
41ffe5d5 | 2771 | static void shmem_destroy_inodecache(void) |
1da177e4 | 2772 | { |
1a1d92c1 | 2773 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
2774 | } |
2775 | ||
f5e54d6e | 2776 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 2777 | .writepage = shmem_writepage, |
76719325 | 2778 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 2779 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
2780 | .write_begin = shmem_write_begin, |
2781 | .write_end = shmem_write_end, | |
1da177e4 | 2782 | #endif |
304dbdb7 | 2783 | .migratepage = migrate_page, |
aa261f54 | 2784 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
2785 | }; |
2786 | ||
15ad7cdc | 2787 | static const struct file_operations shmem_file_operations = { |
1da177e4 LT |
2788 | .mmap = shmem_mmap, |
2789 | #ifdef CONFIG_TMPFS | |
4fb5ef08 | 2790 | .llseek = shmem_file_llseek, |
bcd78e49 | 2791 | .read = do_sync_read, |
5402b976 | 2792 | .write = do_sync_write, |
bcd78e49 | 2793 | .aio_read = shmem_file_aio_read, |
5402b976 | 2794 | .aio_write = generic_file_aio_write, |
1b061d92 | 2795 | .fsync = noop_fsync, |
708e3508 | 2796 | .splice_read = shmem_file_splice_read, |
ae976416 | 2797 | .splice_write = generic_file_splice_write, |
83e4fa9c | 2798 | .fallocate = shmem_fallocate, |
1da177e4 LT |
2799 | #endif |
2800 | }; | |
2801 | ||
92e1d5be | 2802 | static const struct inode_operations shmem_inode_operations = { |
94c1e62d | 2803 | .setattr = shmem_setattr, |
b09e0fa4 EP |
2804 | #ifdef CONFIG_TMPFS_XATTR |
2805 | .setxattr = shmem_setxattr, | |
2806 | .getxattr = shmem_getxattr, | |
2807 | .listxattr = shmem_listxattr, | |
2808 | .removexattr = shmem_removexattr, | |
2809 | #endif | |
1da177e4 LT |
2810 | }; |
2811 | ||
92e1d5be | 2812 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
2813 | #ifdef CONFIG_TMPFS |
2814 | .create = shmem_create, | |
2815 | .lookup = simple_lookup, | |
2816 | .link = shmem_link, | |
2817 | .unlink = shmem_unlink, | |
2818 | .symlink = shmem_symlink, | |
2819 | .mkdir = shmem_mkdir, | |
2820 | .rmdir = shmem_rmdir, | |
2821 | .mknod = shmem_mknod, | |
2822 | .rename = shmem_rename, | |
1da177e4 | 2823 | #endif |
b09e0fa4 EP |
2824 | #ifdef CONFIG_TMPFS_XATTR |
2825 | .setxattr = shmem_setxattr, | |
2826 | .getxattr = shmem_getxattr, | |
2827 | .listxattr = shmem_listxattr, | |
2828 | .removexattr = shmem_removexattr, | |
2829 | #endif | |
39f0247d | 2830 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 2831 | .setattr = shmem_setattr, |
39f0247d AG |
2832 | #endif |
2833 | }; | |
2834 | ||
92e1d5be | 2835 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 EP |
2836 | #ifdef CONFIG_TMPFS_XATTR |
2837 | .setxattr = shmem_setxattr, | |
2838 | .getxattr = shmem_getxattr, | |
2839 | .listxattr = shmem_listxattr, | |
2840 | .removexattr = shmem_removexattr, | |
2841 | #endif | |
39f0247d | 2842 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 2843 | .setattr = shmem_setattr, |
39f0247d | 2844 | #endif |
1da177e4 LT |
2845 | }; |
2846 | ||
759b9775 | 2847 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
2848 | .alloc_inode = shmem_alloc_inode, |
2849 | .destroy_inode = shmem_destroy_inode, | |
2850 | #ifdef CONFIG_TMPFS | |
2851 | .statfs = shmem_statfs, | |
2852 | .remount_fs = shmem_remount_fs, | |
680d794b | 2853 | .show_options = shmem_show_options, |
1da177e4 | 2854 | #endif |
1f895f75 | 2855 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
2856 | .drop_inode = generic_delete_inode, |
2857 | .put_super = shmem_put_super, | |
2858 | }; | |
2859 | ||
f0f37e2f | 2860 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 2861 | .fault = shmem_fault, |
1da177e4 LT |
2862 | #ifdef CONFIG_NUMA |
2863 | .set_policy = shmem_set_policy, | |
2864 | .get_policy = shmem_get_policy, | |
2865 | #endif | |
2866 | }; | |
2867 | ||
3c26ff6e AV |
2868 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
2869 | int flags, const char *dev_name, void *data) | |
1da177e4 | 2870 | { |
3c26ff6e | 2871 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
2872 | } |
2873 | ||
41ffe5d5 | 2874 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
2875 | .owner = THIS_MODULE, |
2876 | .name = "tmpfs", | |
3c26ff6e | 2877 | .mount = shmem_mount, |
1da177e4 LT |
2878 | .kill_sb = kill_litter_super, |
2879 | }; | |
1da177e4 | 2880 | |
41ffe5d5 | 2881 | int __init shmem_init(void) |
1da177e4 LT |
2882 | { |
2883 | int error; | |
2884 | ||
e0bf68dd PZ |
2885 | error = bdi_init(&shmem_backing_dev_info); |
2886 | if (error) | |
2887 | goto out4; | |
2888 | ||
41ffe5d5 | 2889 | error = shmem_init_inodecache(); |
1da177e4 LT |
2890 | if (error) |
2891 | goto out3; | |
2892 | ||
41ffe5d5 | 2893 | error = register_filesystem(&shmem_fs_type); |
1da177e4 LT |
2894 | if (error) { |
2895 | printk(KERN_ERR "Could not register tmpfs\n"); | |
2896 | goto out2; | |
2897 | } | |
95dc112a | 2898 | |
41ffe5d5 HD |
2899 | shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER, |
2900 | shmem_fs_type.name, NULL); | |
1da177e4 LT |
2901 | if (IS_ERR(shm_mnt)) { |
2902 | error = PTR_ERR(shm_mnt); | |
2903 | printk(KERN_ERR "Could not kern_mount tmpfs\n"); | |
2904 | goto out1; | |
2905 | } | |
2906 | return 0; | |
2907 | ||
2908 | out1: | |
41ffe5d5 | 2909 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 2910 | out2: |
41ffe5d5 | 2911 | shmem_destroy_inodecache(); |
1da177e4 | 2912 | out3: |
e0bf68dd PZ |
2913 | bdi_destroy(&shmem_backing_dev_info); |
2914 | out4: | |
1da177e4 LT |
2915 | shm_mnt = ERR_PTR(error); |
2916 | return error; | |
2917 | } | |
853ac43a MM |
2918 | |
2919 | #else /* !CONFIG_SHMEM */ | |
2920 | ||
2921 | /* | |
2922 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
2923 | * | |
2924 | * This is intended for small system where the benefits of the full | |
2925 | * shmem code (swap-backed and resource-limited) are outweighed by | |
2926 | * their complexity. On systems without swap this code should be | |
2927 | * effectively equivalent, but much lighter weight. | |
2928 | */ | |
2929 | ||
2930 | #include <linux/ramfs.h> | |
2931 | ||
41ffe5d5 | 2932 | static struct file_system_type shmem_fs_type = { |
853ac43a | 2933 | .name = "tmpfs", |
3c26ff6e | 2934 | .mount = ramfs_mount, |
853ac43a MM |
2935 | .kill_sb = kill_litter_super, |
2936 | }; | |
2937 | ||
41ffe5d5 | 2938 | int __init shmem_init(void) |
853ac43a | 2939 | { |
41ffe5d5 | 2940 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 2941 | |
41ffe5d5 | 2942 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
2943 | BUG_ON(IS_ERR(shm_mnt)); |
2944 | ||
2945 | return 0; | |
2946 | } | |
2947 | ||
41ffe5d5 | 2948 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
2949 | { |
2950 | return 0; | |
2951 | } | |
2952 | ||
3f96b79a HD |
2953 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
2954 | { | |
2955 | return 0; | |
2956 | } | |
2957 | ||
24513264 HD |
2958 | void shmem_unlock_mapping(struct address_space *mapping) |
2959 | { | |
2960 | } | |
2961 | ||
41ffe5d5 | 2962 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 2963 | { |
41ffe5d5 | 2964 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
2965 | } |
2966 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
2967 | ||
0b0a0806 HD |
2968 | #define shmem_vm_ops generic_file_vm_ops |
2969 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 2970 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
2971 | #define shmem_acct_size(flags, size) 0 |
2972 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
2973 | |
2974 | #endif /* CONFIG_SHMEM */ | |
2975 | ||
2976 | /* common code */ | |
1da177e4 | 2977 | |
46711810 | 2978 | /** |
1da177e4 | 2979 | * shmem_file_setup - get an unlinked file living in tmpfs |
1da177e4 LT |
2980 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
2981 | * @size: size to be set for the file | |
0b0a0806 | 2982 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size |
1da177e4 | 2983 | */ |
168f5ac6 | 2984 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) |
1da177e4 LT |
2985 | { |
2986 | int error; | |
2987 | struct file *file; | |
2988 | struct inode *inode; | |
2c48b9c4 AV |
2989 | struct path path; |
2990 | struct dentry *root; | |
1da177e4 LT |
2991 | struct qstr this; |
2992 | ||
2993 | if (IS_ERR(shm_mnt)) | |
2994 | return (void *)shm_mnt; | |
2995 | ||
285b2c4f | 2996 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
2997 | return ERR_PTR(-EINVAL); |
2998 | ||
2999 | if (shmem_acct_size(flags, size)) | |
3000 | return ERR_PTR(-ENOMEM); | |
3001 | ||
3002 | error = -ENOMEM; | |
3003 | this.name = name; | |
3004 | this.len = strlen(name); | |
3005 | this.hash = 0; /* will go */ | |
3006 | root = shm_mnt->mnt_root; | |
2c48b9c4 AV |
3007 | path.dentry = d_alloc(root, &this); |
3008 | if (!path.dentry) | |
1da177e4 | 3009 | goto put_memory; |
2c48b9c4 | 3010 | path.mnt = mntget(shm_mnt); |
1da177e4 | 3011 | |
1da177e4 | 3012 | error = -ENOSPC; |
454abafe | 3013 | inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 3014 | if (!inode) |
4b42af81 | 3015 | goto put_dentry; |
1da177e4 | 3016 | |
2c48b9c4 | 3017 | d_instantiate(path.dentry, inode); |
1da177e4 | 3018 | inode->i_size = size; |
6d6b77f1 | 3019 | clear_nlink(inode); /* It is unlinked */ |
853ac43a MM |
3020 | #ifndef CONFIG_MMU |
3021 | error = ramfs_nommu_expand_for_mapping(inode, size); | |
3022 | if (error) | |
4b42af81 | 3023 | goto put_dentry; |
853ac43a | 3024 | #endif |
4b42af81 AV |
3025 | |
3026 | error = -ENFILE; | |
2c48b9c4 | 3027 | file = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 AV |
3028 | &shmem_file_operations); |
3029 | if (!file) | |
3030 | goto put_dentry; | |
3031 | ||
1da177e4 LT |
3032 | return file; |
3033 | ||
1da177e4 | 3034 | put_dentry: |
2c48b9c4 | 3035 | path_put(&path); |
1da177e4 LT |
3036 | put_memory: |
3037 | shmem_unacct_size(flags, size); | |
3038 | return ERR_PTR(error); | |
3039 | } | |
395e0ddc | 3040 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 3041 | |
46711810 | 3042 | /** |
1da177e4 | 3043 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
3044 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
3045 | */ | |
3046 | int shmem_zero_setup(struct vm_area_struct *vma) | |
3047 | { | |
3048 | struct file *file; | |
3049 | loff_t size = vma->vm_end - vma->vm_start; | |
3050 | ||
3051 | file = shmem_file_setup("dev/zero", size, vma->vm_flags); | |
3052 | if (IS_ERR(file)) | |
3053 | return PTR_ERR(file); | |
3054 | ||
3055 | if (vma->vm_file) | |
3056 | fput(vma->vm_file); | |
3057 | vma->vm_file = file; | |
3058 | vma->vm_ops = &shmem_vm_ops; | |
bee4c36a | 3059 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
3060 | return 0; |
3061 | } | |
d9d90e5e HD |
3062 | |
3063 | /** | |
3064 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
3065 | * @mapping: the page's address_space | |
3066 | * @index: the page index | |
3067 | * @gfp: the page allocator flags to use if allocating | |
3068 | * | |
3069 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
3070 | * with any new page allocations done using the specified allocation flags. | |
3071 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
3072 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
3073 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
3074 | * | |
68da9f05 HD |
3075 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
3076 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
3077 | */ |
3078 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
3079 | pgoff_t index, gfp_t gfp) | |
3080 | { | |
68da9f05 HD |
3081 | #ifdef CONFIG_SHMEM |
3082 | struct inode *inode = mapping->host; | |
9276aad6 | 3083 | struct page *page; |
68da9f05 HD |
3084 | int error; |
3085 | ||
3086 | BUG_ON(mapping->a_ops != &shmem_aops); | |
3087 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL); | |
3088 | if (error) | |
3089 | page = ERR_PTR(error); | |
3090 | else | |
3091 | unlock_page(page); | |
3092 | return page; | |
3093 | #else | |
3094 | /* | |
3095 | * The tiny !SHMEM case uses ramfs without swap | |
3096 | */ | |
d9d90e5e | 3097 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 3098 | #endif |
d9d90e5e HD |
3099 | } |
3100 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |