]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Resizable virtual memory filesystem for Linux. | |
3 | * | |
4 | * Copyright (C) 2000 Linus Torvalds. | |
5 | * 2000 Transmeta Corp. | |
6 | * 2000-2001 Christoph Rohland | |
7 | * 2000-2001 SAP AG | |
8 | * 2002 Red Hat Inc. | |
6922c0c7 HD |
9 | * Copyright (C) 2002-2011 Hugh Dickins. |
10 | * Copyright (C) 2011 Google Inc. | |
0edd73b3 | 11 | * Copyright (C) 2002-2005 VERITAS Software Corporation. |
1da177e4 LT |
12 | * Copyright (C) 2004 Andi Kleen, SuSE Labs |
13 | * | |
14 | * Extended attribute support for tmpfs: | |
15 | * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net> | |
16 | * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com> | |
17 | * | |
853ac43a MM |
18 | * tiny-shmem: |
19 | * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com> | |
20 | * | |
1da177e4 LT |
21 | * This file is released under the GPL. |
22 | */ | |
23 | ||
853ac43a MM |
24 | #include <linux/fs.h> |
25 | #include <linux/init.h> | |
26 | #include <linux/vfs.h> | |
27 | #include <linux/mount.h> | |
250297ed | 28 | #include <linux/ramfs.h> |
caefba17 | 29 | #include <linux/pagemap.h> |
853ac43a MM |
30 | #include <linux/file.h> |
31 | #include <linux/mm.h> | |
b95f1b31 | 32 | #include <linux/export.h> |
853ac43a | 33 | #include <linux/swap.h> |
a27bb332 | 34 | #include <linux/aio.h> |
853ac43a MM |
35 | |
36 | static struct vfsmount *shm_mnt; | |
37 | ||
38 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
39 | /* |
40 | * This virtual memory filesystem is heavily based on the ramfs. It | |
41 | * extends ramfs by the ability to use swap and honor resource limits | |
42 | * which makes it a completely usable filesystem. | |
43 | */ | |
44 | ||
39f0247d | 45 | #include <linux/xattr.h> |
a5694255 | 46 | #include <linux/exportfs.h> |
1c7c474c | 47 | #include <linux/posix_acl.h> |
feda821e | 48 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 49 | #include <linux/mman.h> |
1da177e4 LT |
50 | #include <linux/string.h> |
51 | #include <linux/slab.h> | |
52 | #include <linux/backing-dev.h> | |
53 | #include <linux/shmem_fs.h> | |
1da177e4 | 54 | #include <linux/writeback.h> |
1da177e4 | 55 | #include <linux/blkdev.h> |
bda97eab | 56 | #include <linux/pagevec.h> |
41ffe5d5 | 57 | #include <linux/percpu_counter.h> |
83e4fa9c | 58 | #include <linux/falloc.h> |
708e3508 | 59 | #include <linux/splice.h> |
1da177e4 LT |
60 | #include <linux/security.h> |
61 | #include <linux/swapops.h> | |
62 | #include <linux/mempolicy.h> | |
63 | #include <linux/namei.h> | |
b00dc3ad | 64 | #include <linux/ctype.h> |
304dbdb7 | 65 | #include <linux/migrate.h> |
c1f60a5a | 66 | #include <linux/highmem.h> |
680d794b | 67 | #include <linux/seq_file.h> |
92562927 | 68 | #include <linux/magic.h> |
304dbdb7 | 69 | |
1da177e4 | 70 | #include <asm/uaccess.h> |
1da177e4 LT |
71 | #include <asm/pgtable.h> |
72 | ||
caefba17 | 73 | #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) |
1da177e4 LT |
74 | #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) |
75 | ||
1da177e4 LT |
76 | /* Pretend that each entry is of this size in directory's i_size */ |
77 | #define BOGO_DIRENT_SIZE 20 | |
78 | ||
69f07ec9 HD |
79 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
80 | #define SHORT_SYMLINK_LEN 128 | |
81 | ||
1aac1400 | 82 | /* |
f00cdc6d HD |
83 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
84 | * inode->i_private (with i_mutex making sure that it has only one user at | |
85 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
86 | */ |
87 | struct shmem_falloc { | |
8e205f77 | 88 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
89 | pgoff_t start; /* start of range currently being fallocated */ |
90 | pgoff_t next; /* the next page offset to be fallocated */ | |
91 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
92 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
93 | }; | |
94 | ||
285b2c4f | 95 | /* Flag allocation requirements to shmem_getpage */ |
1da177e4 | 96 | enum sgp_type { |
1da177e4 LT |
97 | SGP_READ, /* don't exceed i_size, don't allocate page */ |
98 | SGP_CACHE, /* don't exceed i_size, may allocate page */ | |
a0ee5ec5 | 99 | SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ |
1635f6a7 HD |
100 | SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */ |
101 | SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */ | |
1da177e4 LT |
102 | }; |
103 | ||
b76db735 | 104 | #ifdef CONFIG_TMPFS |
680d794b AM |
105 | static unsigned long shmem_default_max_blocks(void) |
106 | { | |
107 | return totalram_pages / 2; | |
108 | } | |
109 | ||
110 | static unsigned long shmem_default_max_inodes(void) | |
111 | { | |
112 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
113 | } | |
b76db735 | 114 | #endif |
680d794b | 115 | |
bde05d1c HD |
116 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
117 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
118 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 HD |
119 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
120 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type); | |
121 | ||
122 | static inline int shmem_getpage(struct inode *inode, pgoff_t index, | |
123 | struct page **pagep, enum sgp_type sgp, int *fault_type) | |
124 | { | |
125 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
126 | mapping_gfp_mask(inode->i_mapping), fault_type); | |
127 | } | |
1da177e4 | 128 | |
1da177e4 LT |
129 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
130 | { | |
131 | return sb->s_fs_info; | |
132 | } | |
133 | ||
134 | /* | |
135 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
136 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
137 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
138 | * consistent with the pre-accounting of private mappings ... | |
139 | */ | |
140 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
141 | { | |
0b0a0806 | 142 | return (flags & VM_NORESERVE) ? |
191c5424 | 143 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
144 | } |
145 | ||
146 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
147 | { | |
0b0a0806 | 148 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
149 | vm_unacct_memory(VM_ACCT(size)); |
150 | } | |
151 | ||
77142517 KK |
152 | static inline int shmem_reacct_size(unsigned long flags, |
153 | loff_t oldsize, loff_t newsize) | |
154 | { | |
155 | if (!(flags & VM_NORESERVE)) { | |
156 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
157 | return security_vm_enough_memory_mm(current->mm, | |
158 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
159 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
160 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
161 | } | |
162 | return 0; | |
163 | } | |
164 | ||
1da177e4 LT |
165 | /* |
166 | * ... whereas tmpfs objects are accounted incrementally as | |
167 | * pages are allocated, in order to allow huge sparse files. | |
168 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, | |
169 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
170 | */ | |
171 | static inline int shmem_acct_block(unsigned long flags) | |
172 | { | |
0b0a0806 | 173 | return (flags & VM_NORESERVE) ? |
191c5424 | 174 | security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0; |
1da177e4 LT |
175 | } |
176 | ||
177 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
178 | { | |
0b0a0806 | 179 | if (flags & VM_NORESERVE) |
1da177e4 LT |
180 | vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); |
181 | } | |
182 | ||
759b9775 | 183 | static const struct super_operations shmem_ops; |
f5e54d6e | 184 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 185 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
186 | static const struct inode_operations shmem_inode_operations; |
187 | static const struct inode_operations shmem_dir_inode_operations; | |
188 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 189 | static const struct vm_operations_struct shmem_vm_ops; |
1da177e4 | 190 | |
6c231b7b | 191 | static struct backing_dev_info shmem_backing_dev_info __read_mostly = { |
1da177e4 | 192 | .ra_pages = 0, /* No readahead */ |
4f98a2fe | 193 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED, |
1da177e4 LT |
194 | }; |
195 | ||
196 | static LIST_HEAD(shmem_swaplist); | |
cb5f7b9a | 197 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 198 | |
5b04c689 PE |
199 | static int shmem_reserve_inode(struct super_block *sb) |
200 | { | |
201 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
202 | if (sbinfo->max_inodes) { | |
203 | spin_lock(&sbinfo->stat_lock); | |
204 | if (!sbinfo->free_inodes) { | |
205 | spin_unlock(&sbinfo->stat_lock); | |
206 | return -ENOSPC; | |
207 | } | |
208 | sbinfo->free_inodes--; | |
209 | spin_unlock(&sbinfo->stat_lock); | |
210 | } | |
211 | return 0; | |
212 | } | |
213 | ||
214 | static void shmem_free_inode(struct super_block *sb) | |
215 | { | |
216 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
217 | if (sbinfo->max_inodes) { | |
218 | spin_lock(&sbinfo->stat_lock); | |
219 | sbinfo->free_inodes++; | |
220 | spin_unlock(&sbinfo->stat_lock); | |
221 | } | |
222 | } | |
223 | ||
46711810 | 224 | /** |
41ffe5d5 | 225 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
226 | * @inode: inode to recalc |
227 | * | |
228 | * We have to calculate the free blocks since the mm can drop | |
229 | * undirtied hole pages behind our back. | |
230 | * | |
231 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
232 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
233 | * | |
234 | * It has to be called with the spinlock held. | |
235 | */ | |
236 | static void shmem_recalc_inode(struct inode *inode) | |
237 | { | |
238 | struct shmem_inode_info *info = SHMEM_I(inode); | |
239 | long freed; | |
240 | ||
241 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
242 | if (freed > 0) { | |
54af6042 HD |
243 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
244 | if (sbinfo->max_blocks) | |
245 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 246 | info->alloced -= freed; |
54af6042 | 247 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 248 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
249 | } |
250 | } | |
251 | ||
7a5d0fbb HD |
252 | /* |
253 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
254 | */ | |
255 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
256 | pgoff_t index, void *expected, void *replacement) | |
257 | { | |
258 | void **pslot; | |
6dbaf22c | 259 | void *item; |
7a5d0fbb HD |
260 | |
261 | VM_BUG_ON(!expected); | |
6dbaf22c | 262 | VM_BUG_ON(!replacement); |
7a5d0fbb | 263 | pslot = radix_tree_lookup_slot(&mapping->page_tree, index); |
6dbaf22c JW |
264 | if (!pslot) |
265 | return -ENOENT; | |
266 | item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock); | |
7a5d0fbb HD |
267 | if (item != expected) |
268 | return -ENOENT; | |
6dbaf22c | 269 | radix_tree_replace_slot(pslot, replacement); |
7a5d0fbb HD |
270 | return 0; |
271 | } | |
272 | ||
d1899228 HD |
273 | /* |
274 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
275 | * that an entry was not already brought back from swap by a racing thread. | |
276 | * | |
277 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
278 | * might be reused, and again be SwapCache, using the same swap as before. | |
279 | */ | |
280 | static bool shmem_confirm_swap(struct address_space *mapping, | |
281 | pgoff_t index, swp_entry_t swap) | |
282 | { | |
283 | void *item; | |
284 | ||
285 | rcu_read_lock(); | |
286 | item = radix_tree_lookup(&mapping->page_tree, index); | |
287 | rcu_read_unlock(); | |
288 | return item == swp_to_radix_entry(swap); | |
289 | } | |
290 | ||
46f65ec1 HD |
291 | /* |
292 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
293 | */ | |
294 | static int shmem_add_to_page_cache(struct page *page, | |
295 | struct address_space *mapping, | |
fed400a1 | 296 | pgoff_t index, void *expected) |
46f65ec1 | 297 | { |
b065b432 | 298 | int error; |
46f65ec1 | 299 | |
309381fe SL |
300 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
301 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
46f65ec1 | 302 | |
b065b432 HD |
303 | page_cache_get(page); |
304 | page->mapping = mapping; | |
305 | page->index = index; | |
306 | ||
307 | spin_lock_irq(&mapping->tree_lock); | |
46f65ec1 | 308 | if (!expected) |
b065b432 HD |
309 | error = radix_tree_insert(&mapping->page_tree, index, page); |
310 | else | |
311 | error = shmem_radix_tree_replace(mapping, index, expected, | |
312 | page); | |
46f65ec1 | 313 | if (!error) { |
b065b432 HD |
314 | mapping->nrpages++; |
315 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
316 | __inc_zone_page_state(page, NR_SHMEM); | |
317 | spin_unlock_irq(&mapping->tree_lock); | |
318 | } else { | |
319 | page->mapping = NULL; | |
320 | spin_unlock_irq(&mapping->tree_lock); | |
321 | page_cache_release(page); | |
46f65ec1 | 322 | } |
46f65ec1 HD |
323 | return error; |
324 | } | |
325 | ||
6922c0c7 HD |
326 | /* |
327 | * Like delete_from_page_cache, but substitutes swap for page. | |
328 | */ | |
329 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
330 | { | |
331 | struct address_space *mapping = page->mapping; | |
332 | int error; | |
333 | ||
334 | spin_lock_irq(&mapping->tree_lock); | |
335 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
336 | page->mapping = NULL; | |
337 | mapping->nrpages--; | |
338 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
339 | __dec_zone_page_state(page, NR_SHMEM); | |
340 | spin_unlock_irq(&mapping->tree_lock); | |
341 | page_cache_release(page); | |
342 | BUG_ON(error); | |
343 | } | |
344 | ||
7a5d0fbb HD |
345 | /* |
346 | * Remove swap entry from radix tree, free the swap and its page cache. | |
347 | */ | |
348 | static int shmem_free_swap(struct address_space *mapping, | |
349 | pgoff_t index, void *radswap) | |
350 | { | |
6dbaf22c | 351 | void *old; |
7a5d0fbb HD |
352 | |
353 | spin_lock_irq(&mapping->tree_lock); | |
6dbaf22c | 354 | old = radix_tree_delete_item(&mapping->page_tree, index, radswap); |
7a5d0fbb | 355 | spin_unlock_irq(&mapping->tree_lock); |
6dbaf22c JW |
356 | if (old != radswap) |
357 | return -ENOENT; | |
358 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
359 | return 0; | |
7a5d0fbb HD |
360 | } |
361 | ||
24513264 HD |
362 | /* |
363 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
364 | */ | |
365 | void shmem_unlock_mapping(struct address_space *mapping) | |
366 | { | |
367 | struct pagevec pvec; | |
368 | pgoff_t indices[PAGEVEC_SIZE]; | |
369 | pgoff_t index = 0; | |
370 | ||
371 | pagevec_init(&pvec, 0); | |
372 | /* | |
373 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
374 | */ | |
375 | while (!mapping_unevictable(mapping)) { | |
376 | /* | |
377 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
378 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
379 | */ | |
0cd6144a JW |
380 | pvec.nr = find_get_entries(mapping, index, |
381 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
382 | if (!pvec.nr) |
383 | break; | |
384 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 385 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
386 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
387 | pagevec_release(&pvec); | |
388 | cond_resched(); | |
389 | } | |
7a5d0fbb HD |
390 | } |
391 | ||
392 | /* | |
393 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 394 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 395 | */ |
1635f6a7 HD |
396 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
397 | bool unfalloc) | |
1da177e4 | 398 | { |
285b2c4f | 399 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 400 | struct shmem_inode_info *info = SHMEM_I(inode); |
285b2c4f | 401 | pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
83e4fa9c HD |
402 | pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT; |
403 | unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1); | |
404 | unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); | |
bda97eab | 405 | struct pagevec pvec; |
7a5d0fbb HD |
406 | pgoff_t indices[PAGEVEC_SIZE]; |
407 | long nr_swaps_freed = 0; | |
285b2c4f | 408 | pgoff_t index; |
bda97eab HD |
409 | int i; |
410 | ||
83e4fa9c HD |
411 | if (lend == -1) |
412 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
413 | |
414 | pagevec_init(&pvec, 0); | |
415 | index = start; | |
83e4fa9c | 416 | while (index < end) { |
0cd6144a JW |
417 | pvec.nr = find_get_entries(mapping, index, |
418 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
419 | pvec.pages, indices); | |
7a5d0fbb HD |
420 | if (!pvec.nr) |
421 | break; | |
bda97eab HD |
422 | mem_cgroup_uncharge_start(); |
423 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
424 | struct page *page = pvec.pages[i]; | |
425 | ||
7a5d0fbb | 426 | index = indices[i]; |
83e4fa9c | 427 | if (index >= end) |
bda97eab HD |
428 | break; |
429 | ||
7a5d0fbb | 430 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
431 | if (unfalloc) |
432 | continue; | |
7a5d0fbb HD |
433 | nr_swaps_freed += !shmem_free_swap(mapping, |
434 | index, page); | |
bda97eab | 435 | continue; |
7a5d0fbb HD |
436 | } |
437 | ||
438 | if (!trylock_page(page)) | |
bda97eab | 439 | continue; |
1635f6a7 HD |
440 | if (!unfalloc || !PageUptodate(page)) { |
441 | if (page->mapping == mapping) { | |
309381fe | 442 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
443 | truncate_inode_page(mapping, page); |
444 | } | |
bda97eab | 445 | } |
bda97eab HD |
446 | unlock_page(page); |
447 | } | |
0cd6144a | 448 | pagevec_remove_exceptionals(&pvec); |
24513264 | 449 | pagevec_release(&pvec); |
bda97eab HD |
450 | mem_cgroup_uncharge_end(); |
451 | cond_resched(); | |
452 | index++; | |
453 | } | |
1da177e4 | 454 | |
83e4fa9c | 455 | if (partial_start) { |
bda97eab HD |
456 | struct page *page = NULL; |
457 | shmem_getpage(inode, start - 1, &page, SGP_READ, NULL); | |
458 | if (page) { | |
83e4fa9c HD |
459 | unsigned int top = PAGE_CACHE_SIZE; |
460 | if (start > end) { | |
461 | top = partial_end; | |
462 | partial_end = 0; | |
463 | } | |
464 | zero_user_segment(page, partial_start, top); | |
465 | set_page_dirty(page); | |
466 | unlock_page(page); | |
467 | page_cache_release(page); | |
468 | } | |
469 | } | |
470 | if (partial_end) { | |
471 | struct page *page = NULL; | |
472 | shmem_getpage(inode, end, &page, SGP_READ, NULL); | |
473 | if (page) { | |
474 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
475 | set_page_dirty(page); |
476 | unlock_page(page); | |
477 | page_cache_release(page); | |
478 | } | |
479 | } | |
83e4fa9c HD |
480 | if (start >= end) |
481 | return; | |
bda97eab HD |
482 | |
483 | index = start; | |
b1a36650 | 484 | while (index < end) { |
bda97eab | 485 | cond_resched(); |
0cd6144a JW |
486 | |
487 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 488 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 489 | pvec.pages, indices); |
7a5d0fbb | 490 | if (!pvec.nr) { |
b1a36650 HD |
491 | /* If all gone or hole-punch or unfalloc, we're done */ |
492 | if (index == start || end != -1) | |
bda97eab | 493 | break; |
b1a36650 | 494 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
495 | index = start; |
496 | continue; | |
497 | } | |
bda97eab HD |
498 | mem_cgroup_uncharge_start(); |
499 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
500 | struct page *page = pvec.pages[i]; | |
501 | ||
7a5d0fbb | 502 | index = indices[i]; |
83e4fa9c | 503 | if (index >= end) |
bda97eab HD |
504 | break; |
505 | ||
7a5d0fbb | 506 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
507 | if (unfalloc) |
508 | continue; | |
b1a36650 HD |
509 | if (shmem_free_swap(mapping, index, page)) { |
510 | /* Swap was replaced by page: retry */ | |
511 | index--; | |
512 | break; | |
513 | } | |
514 | nr_swaps_freed++; | |
7a5d0fbb HD |
515 | continue; |
516 | } | |
517 | ||
bda97eab | 518 | lock_page(page); |
1635f6a7 HD |
519 | if (!unfalloc || !PageUptodate(page)) { |
520 | if (page->mapping == mapping) { | |
309381fe | 521 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 522 | truncate_inode_page(mapping, page); |
b1a36650 HD |
523 | } else { |
524 | /* Page was replaced by swap: retry */ | |
525 | unlock_page(page); | |
526 | index--; | |
527 | break; | |
1635f6a7 | 528 | } |
7a5d0fbb | 529 | } |
bda97eab HD |
530 | unlock_page(page); |
531 | } | |
0cd6144a | 532 | pagevec_remove_exceptionals(&pvec); |
24513264 | 533 | pagevec_release(&pvec); |
bda97eab HD |
534 | mem_cgroup_uncharge_end(); |
535 | index++; | |
536 | } | |
94c1e62d | 537 | |
1da177e4 | 538 | spin_lock(&info->lock); |
7a5d0fbb | 539 | info->swapped -= nr_swaps_freed; |
1da177e4 LT |
540 | shmem_recalc_inode(inode); |
541 | spin_unlock(&info->lock); | |
1635f6a7 | 542 | } |
1da177e4 | 543 | |
1635f6a7 HD |
544 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
545 | { | |
546 | shmem_undo_range(inode, lstart, lend, false); | |
285b2c4f | 547 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1da177e4 | 548 | } |
94c1e62d | 549 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 550 | |
94c1e62d | 551 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 LT |
552 | { |
553 | struct inode *inode = dentry->d_inode; | |
1da177e4 LT |
554 | int error; |
555 | ||
db78b877 CH |
556 | error = inode_change_ok(inode, attr); |
557 | if (error) | |
558 | return error; | |
559 | ||
94c1e62d HD |
560 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
561 | loff_t oldsize = inode->i_size; | |
562 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 563 | |
94c1e62d | 564 | if (newsize != oldsize) { |
77142517 KK |
565 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
566 | oldsize, newsize); | |
567 | if (error) | |
568 | return error; | |
94c1e62d HD |
569 | i_size_write(inode, newsize); |
570 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
571 | } | |
572 | if (newsize < oldsize) { | |
573 | loff_t holebegin = round_up(newsize, PAGE_SIZE); | |
574 | unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); | |
575 | shmem_truncate_range(inode, newsize, (loff_t)-1); | |
576 | /* unmap again to remove racily COWed private pages */ | |
577 | unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); | |
578 | } | |
1da177e4 LT |
579 | } |
580 | ||
db78b877 | 581 | setattr_copy(inode, attr); |
db78b877 | 582 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 583 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
584 | return error; |
585 | } | |
586 | ||
1f895f75 | 587 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 588 | { |
1da177e4 LT |
589 | struct shmem_inode_info *info = SHMEM_I(inode); |
590 | ||
3889e6e7 | 591 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
592 | shmem_unacct_size(info->flags, inode->i_size); |
593 | inode->i_size = 0; | |
3889e6e7 | 594 | shmem_truncate_range(inode, 0, (loff_t)-1); |
1da177e4 | 595 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 596 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 597 | list_del_init(&info->swaplist); |
cb5f7b9a | 598 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 599 | } |
69f07ec9 HD |
600 | } else |
601 | kfree(info->symlink); | |
b09e0fa4 | 602 | |
38f38657 | 603 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 604 | WARN_ON(inode->i_blocks); |
5b04c689 | 605 | shmem_free_inode(inode->i_sb); |
dbd5768f | 606 | clear_inode(inode); |
1da177e4 LT |
607 | } |
608 | ||
46f65ec1 HD |
609 | /* |
610 | * If swap found in inode, free it and move page from swapcache to filecache. | |
611 | */ | |
41ffe5d5 | 612 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 613 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 614 | { |
285b2c4f | 615 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 616 | void *radswap; |
41ffe5d5 | 617 | pgoff_t index; |
bde05d1c HD |
618 | gfp_t gfp; |
619 | int error = 0; | |
1da177e4 | 620 | |
46f65ec1 | 621 | radswap = swp_to_radix_entry(swap); |
e504f3fd | 622 | index = radix_tree_locate_item(&mapping->page_tree, radswap); |
46f65ec1 | 623 | if (index == -1) |
285b2c4f | 624 | return 0; |
2e0e26c7 | 625 | |
1b1b32f2 HD |
626 | /* |
627 | * Move _head_ to start search for next from here. | |
1f895f75 | 628 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 629 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 630 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
631 | */ |
632 | if (shmem_swaplist.next != &info->swaplist) | |
633 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 634 | |
bde05d1c HD |
635 | gfp = mapping_gfp_mask(mapping); |
636 | if (shmem_should_replace_page(*pagep, gfp)) { | |
637 | mutex_unlock(&shmem_swaplist_mutex); | |
638 | error = shmem_replace_page(pagep, gfp, info, index); | |
639 | mutex_lock(&shmem_swaplist_mutex); | |
640 | /* | |
641 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
642 | * allocation, but the inode might have been freed while we |
643 | * dropped it: although a racing shmem_evict_inode() cannot | |
644 | * complete without emptying the radix_tree, our page lock | |
645 | * on this swapcache page is not enough to prevent that - | |
646 | * free_swap_and_cache() of our swap entry will only | |
647 | * trylock_page(), removing swap from radix_tree whatever. | |
648 | * | |
649 | * We must not proceed to shmem_add_to_page_cache() if the | |
650 | * inode has been freed, but of course we cannot rely on | |
651 | * inode or mapping or info to check that. However, we can | |
652 | * safely check if our swap entry is still in use (and here | |
653 | * it can't have got reused for another page): if it's still | |
654 | * in use, then the inode cannot have been freed yet, and we | |
655 | * can safely proceed (if it's no longer in use, that tells | |
656 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
657 | */ |
658 | if (!page_swapcount(*pagep)) | |
659 | error = -ENOENT; | |
660 | } | |
661 | ||
d13d1443 | 662 | /* |
778dd893 HD |
663 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
664 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
665 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 666 | */ |
bde05d1c HD |
667 | if (!error) |
668 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 669 | radswap); |
48f170fb | 670 | if (error != -ENOMEM) { |
46f65ec1 HD |
671 | /* |
672 | * Truncation and eviction use free_swap_and_cache(), which | |
673 | * only does trylock page: if we raced, best clean up here. | |
674 | */ | |
bde05d1c HD |
675 | delete_from_swap_cache(*pagep); |
676 | set_page_dirty(*pagep); | |
46f65ec1 HD |
677 | if (!error) { |
678 | spin_lock(&info->lock); | |
679 | info->swapped--; | |
680 | spin_unlock(&info->lock); | |
681 | swap_free(swap); | |
682 | } | |
2e0e26c7 | 683 | error = 1; /* not an error, but entry was found */ |
1da177e4 | 684 | } |
2e0e26c7 | 685 | return error; |
1da177e4 LT |
686 | } |
687 | ||
688 | /* | |
46f65ec1 | 689 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 690 | */ |
41ffe5d5 | 691 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 692 | { |
41ffe5d5 | 693 | struct list_head *this, *next; |
1da177e4 LT |
694 | struct shmem_inode_info *info; |
695 | int found = 0; | |
bde05d1c HD |
696 | int error = 0; |
697 | ||
698 | /* | |
699 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 700 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 701 | */ |
0142ef6c | 702 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 703 | goto out; |
778dd893 HD |
704 | |
705 | /* | |
706 | * Charge page using GFP_KERNEL while we can wait, before taking | |
707 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
708 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 709 | */ |
d715ae08 | 710 | error = mem_cgroup_charge_file(page, current->mm, GFP_KERNEL); |
778dd893 HD |
711 | if (error) |
712 | goto out; | |
46f65ec1 | 713 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
1da177e4 | 714 | |
cb5f7b9a | 715 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
716 | list_for_each_safe(this, next, &shmem_swaplist) { |
717 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 718 | if (info->swapped) |
bde05d1c | 719 | found = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
720 | else |
721 | list_del_init(&info->swaplist); | |
cb5f7b9a | 722 | cond_resched(); |
2e0e26c7 | 723 | if (found) |
778dd893 | 724 | break; |
1da177e4 | 725 | } |
cb5f7b9a | 726 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 727 | |
778dd893 HD |
728 | if (found < 0) |
729 | error = found; | |
730 | out: | |
aaa46865 HD |
731 | unlock_page(page); |
732 | page_cache_release(page); | |
778dd893 | 733 | return error; |
1da177e4 LT |
734 | } |
735 | ||
736 | /* | |
737 | * Move the page from the page cache to the swap cache. | |
738 | */ | |
739 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
740 | { | |
741 | struct shmem_inode_info *info; | |
1da177e4 | 742 | struct address_space *mapping; |
1da177e4 | 743 | struct inode *inode; |
6922c0c7 HD |
744 | swp_entry_t swap; |
745 | pgoff_t index; | |
1da177e4 LT |
746 | |
747 | BUG_ON(!PageLocked(page)); | |
1da177e4 LT |
748 | mapping = page->mapping; |
749 | index = page->index; | |
750 | inode = mapping->host; | |
751 | info = SHMEM_I(inode); | |
752 | if (info->flags & VM_LOCKED) | |
753 | goto redirty; | |
d9fe526a | 754 | if (!total_swap_pages) |
1da177e4 LT |
755 | goto redirty; |
756 | ||
d9fe526a HD |
757 | /* |
758 | * shmem_backing_dev_info's capabilities prevent regular writeback or | |
759 | * sync from ever calling shmem_writepage; but a stacking filesystem | |
48f170fb | 760 | * might use ->writepage of its underlying filesystem, in which case |
d9fe526a | 761 | * tmpfs should write out to swap only in response to memory pressure, |
48f170fb | 762 | * and not for the writeback threads or sync. |
d9fe526a | 763 | */ |
48f170fb HD |
764 | if (!wbc->for_reclaim) { |
765 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
766 | goto redirty; | |
767 | } | |
1635f6a7 HD |
768 | |
769 | /* | |
770 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
771 | * value into swapfile.c, the only way we can correctly account for a | |
772 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
773 | * |
774 | * That's okay for a page already fallocated earlier, but if we have | |
775 | * not yet completed the fallocation, then (a) we want to keep track | |
776 | * of this page in case we have to undo it, and (b) it may not be a | |
777 | * good idea to continue anyway, once we're pushing into swap. So | |
778 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
779 | */ |
780 | if (!PageUptodate(page)) { | |
1aac1400 HD |
781 | if (inode->i_private) { |
782 | struct shmem_falloc *shmem_falloc; | |
783 | spin_lock(&inode->i_lock); | |
784 | shmem_falloc = inode->i_private; | |
785 | if (shmem_falloc && | |
8e205f77 | 786 | !shmem_falloc->waitq && |
1aac1400 HD |
787 | index >= shmem_falloc->start && |
788 | index < shmem_falloc->next) | |
789 | shmem_falloc->nr_unswapped++; | |
790 | else | |
791 | shmem_falloc = NULL; | |
792 | spin_unlock(&inode->i_lock); | |
793 | if (shmem_falloc) | |
794 | goto redirty; | |
795 | } | |
1635f6a7 HD |
796 | clear_highpage(page); |
797 | flush_dcache_page(page); | |
798 | SetPageUptodate(page); | |
799 | } | |
800 | ||
48f170fb HD |
801 | swap = get_swap_page(); |
802 | if (!swap.val) | |
803 | goto redirty; | |
d9fe526a | 804 | |
b1dea800 HD |
805 | /* |
806 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
807 | * if it's not already there. Do it now before the page is |
808 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 809 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
810 | * we've incremented swapped, because shmem_unuse_inode() will |
811 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 812 | */ |
48f170fb HD |
813 | mutex_lock(&shmem_swaplist_mutex); |
814 | if (list_empty(&info->swaplist)) | |
815 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 816 | |
48f170fb | 817 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
aaa46865 | 818 | swap_shmem_alloc(swap); |
6922c0c7 HD |
819 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); |
820 | ||
821 | spin_lock(&info->lock); | |
822 | info->swapped++; | |
823 | shmem_recalc_inode(inode); | |
826267cf | 824 | spin_unlock(&info->lock); |
6922c0c7 HD |
825 | |
826 | mutex_unlock(&shmem_swaplist_mutex); | |
d9fe526a | 827 | BUG_ON(page_mapped(page)); |
9fab5619 | 828 | swap_writepage(page, wbc); |
1da177e4 LT |
829 | return 0; |
830 | } | |
831 | ||
6922c0c7 | 832 | mutex_unlock(&shmem_swaplist_mutex); |
cb4b86ba | 833 | swapcache_free(swap, NULL); |
1da177e4 LT |
834 | redirty: |
835 | set_page_dirty(page); | |
d9fe526a HD |
836 | if (wbc->for_reclaim) |
837 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
838 | unlock_page(page); | |
839 | return 0; | |
1da177e4 LT |
840 | } |
841 | ||
842 | #ifdef CONFIG_NUMA | |
680d794b | 843 | #ifdef CONFIG_TMPFS |
71fe804b | 844 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 845 | { |
095f1fc4 | 846 | char buffer[64]; |
680d794b | 847 | |
71fe804b | 848 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 849 | return; /* show nothing */ |
680d794b | 850 | |
a7a88b23 | 851 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
852 | |
853 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 854 | } |
71fe804b LS |
855 | |
856 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
857 | { | |
858 | struct mempolicy *mpol = NULL; | |
859 | if (sbinfo->mpol) { | |
860 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
861 | mpol = sbinfo->mpol; | |
862 | mpol_get(mpol); | |
863 | spin_unlock(&sbinfo->stat_lock); | |
864 | } | |
865 | return mpol; | |
866 | } | |
680d794b AM |
867 | #endif /* CONFIG_TMPFS */ |
868 | ||
41ffe5d5 HD |
869 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
870 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 871 | { |
1da177e4 | 872 | struct vm_area_struct pvma; |
18a2f371 | 873 | struct page *page; |
52cd3b07 | 874 | |
1da177e4 | 875 | /* Create a pseudo vma that just contains the policy */ |
c4cc6d07 | 876 | pvma.vm_start = 0; |
09c231cb NZ |
877 | /* Bias interleave by inode number to distribute better across nodes */ |
878 | pvma.vm_pgoff = index + info->vfs_inode.i_ino; | |
c4cc6d07 | 879 | pvma.vm_ops = NULL; |
18a2f371 MG |
880 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
881 | ||
882 | page = swapin_readahead(swap, gfp, &pvma, 0); | |
883 | ||
884 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
885 | mpol_cond_put(pvma.vm_policy); | |
886 | ||
887 | return page; | |
1da177e4 LT |
888 | } |
889 | ||
02098fea | 890 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 891 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
892 | { |
893 | struct vm_area_struct pvma; | |
18a2f371 | 894 | struct page *page; |
1da177e4 | 895 | |
c4cc6d07 HD |
896 | /* Create a pseudo vma that just contains the policy */ |
897 | pvma.vm_start = 0; | |
09c231cb NZ |
898 | /* Bias interleave by inode number to distribute better across nodes */ |
899 | pvma.vm_pgoff = index + info->vfs_inode.i_ino; | |
c4cc6d07 | 900 | pvma.vm_ops = NULL; |
41ffe5d5 | 901 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
52cd3b07 | 902 | |
18a2f371 MG |
903 | page = alloc_page_vma(gfp, &pvma, 0); |
904 | ||
905 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
906 | mpol_cond_put(pvma.vm_policy); | |
907 | ||
908 | return page; | |
1da177e4 | 909 | } |
680d794b AM |
910 | #else /* !CONFIG_NUMA */ |
911 | #ifdef CONFIG_TMPFS | |
41ffe5d5 | 912 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b AM |
913 | { |
914 | } | |
915 | #endif /* CONFIG_TMPFS */ | |
916 | ||
41ffe5d5 HD |
917 | static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
918 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 919 | { |
41ffe5d5 | 920 | return swapin_readahead(swap, gfp, NULL, 0); |
1da177e4 LT |
921 | } |
922 | ||
02098fea | 923 | static inline struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 924 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 | 925 | { |
e84e2e13 | 926 | return alloc_page(gfp); |
1da177e4 | 927 | } |
680d794b | 928 | #endif /* CONFIG_NUMA */ |
1da177e4 | 929 | |
71fe804b LS |
930 | #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) |
931 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
932 | { | |
933 | return NULL; | |
934 | } | |
935 | #endif | |
936 | ||
bde05d1c HD |
937 | /* |
938 | * When a page is moved from swapcache to shmem filecache (either by the | |
939 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
940 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
941 | * ignorance of the mapping it belongs to. If that mapping has special | |
942 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
943 | * we may need to copy to a suitable page before moving to filecache. | |
944 | * | |
945 | * In a future release, this may well be extended to respect cpuset and | |
946 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
947 | * but for now it is a simple matter of zone. | |
948 | */ | |
949 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
950 | { | |
951 | return page_zonenum(page) > gfp_zone(gfp); | |
952 | } | |
953 | ||
954 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
955 | struct shmem_inode_info *info, pgoff_t index) | |
956 | { | |
957 | struct page *oldpage, *newpage; | |
958 | struct address_space *swap_mapping; | |
959 | pgoff_t swap_index; | |
960 | int error; | |
961 | ||
962 | oldpage = *pagep; | |
963 | swap_index = page_private(oldpage); | |
964 | swap_mapping = page_mapping(oldpage); | |
965 | ||
966 | /* | |
967 | * We have arrived here because our zones are constrained, so don't | |
968 | * limit chance of success by further cpuset and node constraints. | |
969 | */ | |
970 | gfp &= ~GFP_CONSTRAINT_MASK; | |
971 | newpage = shmem_alloc_page(gfp, info, index); | |
972 | if (!newpage) | |
973 | return -ENOMEM; | |
bde05d1c | 974 | |
bde05d1c HD |
975 | page_cache_get(newpage); |
976 | copy_highpage(newpage, oldpage); | |
0142ef6c | 977 | flush_dcache_page(newpage); |
bde05d1c | 978 | |
bde05d1c | 979 | __set_page_locked(newpage); |
bde05d1c | 980 | SetPageUptodate(newpage); |
bde05d1c | 981 | SetPageSwapBacked(newpage); |
bde05d1c | 982 | set_page_private(newpage, swap_index); |
bde05d1c HD |
983 | SetPageSwapCache(newpage); |
984 | ||
985 | /* | |
986 | * Our caller will very soon move newpage out of swapcache, but it's | |
987 | * a nice clean interface for us to replace oldpage by newpage there. | |
988 | */ | |
989 | spin_lock_irq(&swap_mapping->tree_lock); | |
990 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
991 | newpage); | |
0142ef6c HD |
992 | if (!error) { |
993 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
994 | __dec_zone_page_state(oldpage, NR_FILE_PAGES); | |
995 | } | |
bde05d1c | 996 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 997 | |
0142ef6c HD |
998 | if (unlikely(error)) { |
999 | /* | |
1000 | * Is this possible? I think not, now that our callers check | |
1001 | * both PageSwapCache and page_private after getting page lock; | |
1002 | * but be defensive. Reverse old to newpage for clear and free. | |
1003 | */ | |
1004 | oldpage = newpage; | |
1005 | } else { | |
1006 | mem_cgroup_replace_page_cache(oldpage, newpage); | |
1007 | lru_cache_add_anon(newpage); | |
1008 | *pagep = newpage; | |
1009 | } | |
bde05d1c HD |
1010 | |
1011 | ClearPageSwapCache(oldpage); | |
1012 | set_page_private(oldpage, 0); | |
1013 | ||
1014 | unlock_page(oldpage); | |
1015 | page_cache_release(oldpage); | |
1016 | page_cache_release(oldpage); | |
0142ef6c | 1017 | return error; |
bde05d1c HD |
1018 | } |
1019 | ||
1da177e4 | 1020 | /* |
68da9f05 | 1021 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1022 | * |
1023 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1024 | * vm. If we swap it in we mark it dirty since we also free the swap | |
1025 | * entry since a page cannot live in both the swap and page cache | |
1026 | */ | |
41ffe5d5 | 1027 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
68da9f05 | 1028 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type) |
1da177e4 LT |
1029 | { |
1030 | struct address_space *mapping = inode->i_mapping; | |
54af6042 | 1031 | struct shmem_inode_info *info; |
1da177e4 | 1032 | struct shmem_sb_info *sbinfo; |
27ab7006 | 1033 | struct page *page; |
1da177e4 LT |
1034 | swp_entry_t swap; |
1035 | int error; | |
54af6042 | 1036 | int once = 0; |
1635f6a7 | 1037 | int alloced = 0; |
1da177e4 | 1038 | |
41ffe5d5 | 1039 | if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT)) |
1da177e4 | 1040 | return -EFBIG; |
1da177e4 | 1041 | repeat: |
54af6042 | 1042 | swap.val = 0; |
0cd6144a | 1043 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1044 | if (radix_tree_exceptional_entry(page)) { |
1045 | swap = radix_to_swp_entry(page); | |
1046 | page = NULL; | |
1047 | } | |
1048 | ||
1635f6a7 | 1049 | if (sgp != SGP_WRITE && sgp != SGP_FALLOC && |
54af6042 HD |
1050 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { |
1051 | error = -EINVAL; | |
1052 | goto failed; | |
1053 | } | |
1054 | ||
66d2f4d2 HD |
1055 | if (page && sgp == SGP_WRITE) |
1056 | mark_page_accessed(page); | |
1057 | ||
1635f6a7 HD |
1058 | /* fallocated page? */ |
1059 | if (page && !PageUptodate(page)) { | |
1060 | if (sgp != SGP_READ) | |
1061 | goto clear; | |
1062 | unlock_page(page); | |
1063 | page_cache_release(page); | |
1064 | page = NULL; | |
1065 | } | |
54af6042 | 1066 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1067 | *pagep = page; |
1068 | return 0; | |
27ab7006 HD |
1069 | } |
1070 | ||
1071 | /* | |
54af6042 HD |
1072 | * Fast cache lookup did not find it: |
1073 | * bring it back from swap or allocate. | |
27ab7006 | 1074 | */ |
54af6042 HD |
1075 | info = SHMEM_I(inode); |
1076 | sbinfo = SHMEM_SB(inode->i_sb); | |
1da177e4 | 1077 | |
1da177e4 LT |
1078 | if (swap.val) { |
1079 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1080 | page = lookup_swap_cache(swap); |
1081 | if (!page) { | |
1da177e4 | 1082 | /* here we actually do the io */ |
68da9f05 HD |
1083 | if (fault_type) |
1084 | *fault_type |= VM_FAULT_MAJOR; | |
41ffe5d5 | 1085 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1086 | if (!page) { |
54af6042 HD |
1087 | error = -ENOMEM; |
1088 | goto failed; | |
1da177e4 | 1089 | } |
1da177e4 LT |
1090 | } |
1091 | ||
1092 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1093 | lock_page(page); |
0142ef6c | 1094 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1095 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1096 | error = -EEXIST; /* try again */ |
d1899228 | 1097 | goto unlock; |
bde05d1c | 1098 | } |
27ab7006 | 1099 | if (!PageUptodate(page)) { |
1da177e4 | 1100 | error = -EIO; |
54af6042 | 1101 | goto failed; |
1da177e4 | 1102 | } |
54af6042 HD |
1103 | wait_on_page_writeback(page); |
1104 | ||
bde05d1c HD |
1105 | if (shmem_should_replace_page(page, gfp)) { |
1106 | error = shmem_replace_page(&page, gfp, info, index); | |
1107 | if (error) | |
1108 | goto failed; | |
1da177e4 | 1109 | } |
27ab7006 | 1110 | |
d715ae08 | 1111 | error = mem_cgroup_charge_file(page, current->mm, |
aa3b1895 | 1112 | gfp & GFP_RECLAIM_MASK); |
d1899228 | 1113 | if (!error) { |
aa3b1895 | 1114 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1115 | swp_to_radix_entry(swap)); |
215c02bc HD |
1116 | /* |
1117 | * We already confirmed swap under page lock, and make | |
1118 | * no memory allocation here, so usually no possibility | |
1119 | * of error; but free_swap_and_cache() only trylocks a | |
1120 | * page, so it is just possible that the entry has been | |
1121 | * truncated or holepunched since swap was confirmed. | |
1122 | * shmem_undo_range() will have done some of the | |
1123 | * unaccounting, now delete_from_swap_cache() will do | |
1124 | * the rest (including mem_cgroup_uncharge_swapcache). | |
1125 | * Reset swap.val? No, leave it so "failed" goes back to | |
1126 | * "repeat": reading a hole and writing should succeed. | |
1127 | */ | |
1128 | if (error) | |
1129 | delete_from_swap_cache(page); | |
d1899228 | 1130 | } |
54af6042 HD |
1131 | if (error) |
1132 | goto failed; | |
1133 | ||
1134 | spin_lock(&info->lock); | |
285b2c4f | 1135 | info->swapped--; |
54af6042 | 1136 | shmem_recalc_inode(inode); |
27ab7006 | 1137 | spin_unlock(&info->lock); |
54af6042 | 1138 | |
66d2f4d2 HD |
1139 | if (sgp == SGP_WRITE) |
1140 | mark_page_accessed(page); | |
1141 | ||
54af6042 | 1142 | delete_from_swap_cache(page); |
27ab7006 HD |
1143 | set_page_dirty(page); |
1144 | swap_free(swap); | |
1145 | ||
54af6042 HD |
1146 | } else { |
1147 | if (shmem_acct_block(info->flags)) { | |
1148 | error = -ENOSPC; | |
1149 | goto failed; | |
1da177e4 | 1150 | } |
0edd73b3 | 1151 | if (sbinfo->max_blocks) { |
fc5da22a | 1152 | if (percpu_counter_compare(&sbinfo->used_blocks, |
54af6042 HD |
1153 | sbinfo->max_blocks) >= 0) { |
1154 | error = -ENOSPC; | |
1155 | goto unacct; | |
1156 | } | |
7e496299 | 1157 | percpu_counter_inc(&sbinfo->used_blocks); |
54af6042 | 1158 | } |
1da177e4 | 1159 | |
54af6042 HD |
1160 | page = shmem_alloc_page(gfp, info, index); |
1161 | if (!page) { | |
1162 | error = -ENOMEM; | |
1163 | goto decused; | |
1da177e4 LT |
1164 | } |
1165 | ||
07a42788 | 1166 | __SetPageSwapBacked(page); |
54af6042 | 1167 | __set_page_locked(page); |
66d2f4d2 | 1168 | if (sgp == SGP_WRITE) |
eb39d618 | 1169 | __SetPageReferenced(page); |
66d2f4d2 | 1170 | |
d715ae08 | 1171 | error = mem_cgroup_charge_file(page, current->mm, |
aa3b1895 | 1172 | gfp & GFP_RECLAIM_MASK); |
54af6042 HD |
1173 | if (error) |
1174 | goto decused; | |
5e4c0d97 | 1175 | error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK); |
b065b432 HD |
1176 | if (!error) { |
1177 | error = shmem_add_to_page_cache(page, mapping, index, | |
fed400a1 | 1178 | NULL); |
b065b432 HD |
1179 | radix_tree_preload_end(); |
1180 | } | |
1181 | if (error) { | |
1182 | mem_cgroup_uncharge_cache_page(page); | |
1183 | goto decused; | |
1184 | } | |
54af6042 HD |
1185 | lru_cache_add_anon(page); |
1186 | ||
1187 | spin_lock(&info->lock); | |
1da177e4 | 1188 | info->alloced++; |
54af6042 HD |
1189 | inode->i_blocks += BLOCKS_PER_PAGE; |
1190 | shmem_recalc_inode(inode); | |
1da177e4 | 1191 | spin_unlock(&info->lock); |
1635f6a7 | 1192 | alloced = true; |
54af6042 | 1193 | |
ec9516fb | 1194 | /* |
1635f6a7 HD |
1195 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1196 | */ | |
1197 | if (sgp == SGP_FALLOC) | |
1198 | sgp = SGP_WRITE; | |
1199 | clear: | |
1200 | /* | |
1201 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1202 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1203 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb HD |
1204 | */ |
1205 | if (sgp != SGP_WRITE) { | |
1206 | clear_highpage(page); | |
1207 | flush_dcache_page(page); | |
1208 | SetPageUptodate(page); | |
1209 | } | |
a0ee5ec5 | 1210 | if (sgp == SGP_DIRTY) |
27ab7006 | 1211 | set_page_dirty(page); |
1da177e4 | 1212 | } |
bde05d1c | 1213 | |
54af6042 | 1214 | /* Perhaps the file has been truncated since we checked */ |
1635f6a7 | 1215 | if (sgp != SGP_WRITE && sgp != SGP_FALLOC && |
54af6042 HD |
1216 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { |
1217 | error = -EINVAL; | |
1635f6a7 HD |
1218 | if (alloced) |
1219 | goto trunc; | |
1220 | else | |
1221 | goto failed; | |
e83c32e8 | 1222 | } |
54af6042 HD |
1223 | *pagep = page; |
1224 | return 0; | |
1da177e4 | 1225 | |
59a16ead | 1226 | /* |
54af6042 | 1227 | * Error recovery. |
59a16ead | 1228 | */ |
54af6042 | 1229 | trunc: |
1635f6a7 | 1230 | info = SHMEM_I(inode); |
54af6042 HD |
1231 | ClearPageDirty(page); |
1232 | delete_from_page_cache(page); | |
1233 | spin_lock(&info->lock); | |
1234 | info->alloced--; | |
1235 | inode->i_blocks -= BLOCKS_PER_PAGE; | |
59a16ead | 1236 | spin_unlock(&info->lock); |
54af6042 | 1237 | decused: |
1635f6a7 | 1238 | sbinfo = SHMEM_SB(inode->i_sb); |
54af6042 HD |
1239 | if (sbinfo->max_blocks) |
1240 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
1241 | unacct: | |
1242 | shmem_unacct_blocks(info->flags, 1); | |
1243 | failed: | |
d1899228 HD |
1244 | if (swap.val && error != -EINVAL && |
1245 | !shmem_confirm_swap(mapping, index, swap)) | |
1246 | error = -EEXIST; | |
1247 | unlock: | |
27ab7006 | 1248 | if (page) { |
54af6042 | 1249 | unlock_page(page); |
27ab7006 | 1250 | page_cache_release(page); |
54af6042 HD |
1251 | } |
1252 | if (error == -ENOSPC && !once++) { | |
1253 | info = SHMEM_I(inode); | |
1254 | spin_lock(&info->lock); | |
1255 | shmem_recalc_inode(inode); | |
1256 | spin_unlock(&info->lock); | |
27ab7006 | 1257 | goto repeat; |
ff36b801 | 1258 | } |
d1899228 | 1259 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1260 | goto repeat; |
1261 | return error; | |
1da177e4 LT |
1262 | } |
1263 | ||
d0217ac0 | 1264 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1265 | { |
496ad9aa | 1266 | struct inode *inode = file_inode(vma->vm_file); |
1da177e4 | 1267 | int error; |
68da9f05 | 1268 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1269 | |
f00cdc6d HD |
1270 | /* |
1271 | * Trinity finds that probing a hole which tmpfs is punching can | |
1272 | * prevent the hole-punch from ever completing: which in turn | |
1273 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1274 | * faulting pages into the hole while it's being punched. Although |
1275 | * shmem_undo_range() does remove the additions, it may be unable to | |
1276 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1277 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1278 | * | |
1279 | * It does not matter if we sometimes reach this check just before the | |
1280 | * hole-punch begins, so that one fault then races with the punch: | |
1281 | * we just need to make racing faults a rare case. | |
1282 | * | |
1283 | * The implementation below would be much simpler if we just used a | |
1284 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1285 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1286 | */ |
1287 | if (unlikely(inode->i_private)) { | |
1288 | struct shmem_falloc *shmem_falloc; | |
1289 | ||
1290 | spin_lock(&inode->i_lock); | |
1291 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1292 | if (shmem_falloc && |
1293 | shmem_falloc->waitq && | |
1294 | vmf->pgoff >= shmem_falloc->start && | |
1295 | vmf->pgoff < shmem_falloc->next) { | |
1296 | wait_queue_head_t *shmem_falloc_waitq; | |
1297 | DEFINE_WAIT(shmem_fault_wait); | |
1298 | ||
1299 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1300 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1301 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1302 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1303 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1304 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1305 | } |
8e205f77 HD |
1306 | |
1307 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1308 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1309 | TASK_UNINTERRUPTIBLE); | |
1310 | spin_unlock(&inode->i_lock); | |
1311 | schedule(); | |
1312 | ||
1313 | /* | |
1314 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1315 | * stack of the hole-punching task: shmem_falloc_waitq | |
1316 | * is usually invalid by the time we reach here, but | |
1317 | * finish_wait() does not dereference it in that case; | |
1318 | * though i_lock needed lest racing with wake_up_all(). | |
1319 | */ | |
1320 | spin_lock(&inode->i_lock); | |
1321 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1322 | spin_unlock(&inode->i_lock); | |
1323 | return ret; | |
f00cdc6d | 1324 | } |
8e205f77 | 1325 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
1326 | } |
1327 | ||
27d54b39 | 1328 | error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); |
d0217ac0 NP |
1329 | if (error) |
1330 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1331 | |
456f998e YH |
1332 | if (ret & VM_FAULT_MAJOR) { |
1333 | count_vm_event(PGMAJFAULT); | |
1334 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); | |
1335 | } | |
68da9f05 | 1336 | return ret; |
1da177e4 LT |
1337 | } |
1338 | ||
1da177e4 | 1339 | #ifdef CONFIG_NUMA |
41ffe5d5 | 1340 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 1341 | { |
496ad9aa | 1342 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1343 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
1344 | } |
1345 | ||
d8dc74f2 AB |
1346 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
1347 | unsigned long addr) | |
1da177e4 | 1348 | { |
496ad9aa | 1349 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1350 | pgoff_t index; |
1da177e4 | 1351 | |
41ffe5d5 HD |
1352 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
1353 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
1354 | } |
1355 | #endif | |
1356 | ||
1357 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
1358 | { | |
496ad9aa | 1359 | struct inode *inode = file_inode(file); |
1da177e4 LT |
1360 | struct shmem_inode_info *info = SHMEM_I(inode); |
1361 | int retval = -ENOMEM; | |
1362 | ||
1363 | spin_lock(&info->lock); | |
1364 | if (lock && !(info->flags & VM_LOCKED)) { | |
1365 | if (!user_shm_lock(inode->i_size, user)) | |
1366 | goto out_nomem; | |
1367 | info->flags |= VM_LOCKED; | |
89e004ea | 1368 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
1369 | } |
1370 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
1371 | user_shm_unlock(inode->i_size, user); | |
1372 | info->flags &= ~VM_LOCKED; | |
89e004ea | 1373 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
1374 | } |
1375 | retval = 0; | |
89e004ea | 1376 | |
1da177e4 LT |
1377 | out_nomem: |
1378 | spin_unlock(&info->lock); | |
1379 | return retval; | |
1380 | } | |
1381 | ||
9b83a6a8 | 1382 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
1383 | { |
1384 | file_accessed(file); | |
1385 | vma->vm_ops = &shmem_vm_ops; | |
1386 | return 0; | |
1387 | } | |
1388 | ||
454abafe | 1389 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 1390 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
1391 | { |
1392 | struct inode *inode; | |
1393 | struct shmem_inode_info *info; | |
1394 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
1395 | ||
5b04c689 PE |
1396 | if (shmem_reserve_inode(sb)) |
1397 | return NULL; | |
1da177e4 LT |
1398 | |
1399 | inode = new_inode(sb); | |
1400 | if (inode) { | |
85fe4025 | 1401 | inode->i_ino = get_next_ino(); |
454abafe | 1402 | inode_init_owner(inode, dir, mode); |
1da177e4 | 1403 | inode->i_blocks = 0; |
1da177e4 LT |
1404 | inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; |
1405 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
91828a40 | 1406 | inode->i_generation = get_seconds(); |
1da177e4 LT |
1407 | info = SHMEM_I(inode); |
1408 | memset(info, 0, (char *)inode - (char *)info); | |
1409 | spin_lock_init(&info->lock); | |
0b0a0806 | 1410 | info->flags = flags & VM_NORESERVE; |
1da177e4 | 1411 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 1412 | simple_xattrs_init(&info->xattrs); |
72c04902 | 1413 | cache_no_acl(inode); |
1da177e4 LT |
1414 | |
1415 | switch (mode & S_IFMT) { | |
1416 | default: | |
39f0247d | 1417 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
1418 | init_special_inode(inode, mode, dev); |
1419 | break; | |
1420 | case S_IFREG: | |
14fcc23f | 1421 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
1422 | inode->i_op = &shmem_inode_operations; |
1423 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
1424 | mpol_shared_policy_init(&info->policy, |
1425 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
1426 | break; |
1427 | case S_IFDIR: | |
d8c76e6f | 1428 | inc_nlink(inode); |
1da177e4 LT |
1429 | /* Some things misbehave if size == 0 on a directory */ |
1430 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
1431 | inode->i_op = &shmem_dir_inode_operations; | |
1432 | inode->i_fop = &simple_dir_operations; | |
1433 | break; | |
1434 | case S_IFLNK: | |
1435 | /* | |
1436 | * Must not load anything in the rbtree, | |
1437 | * mpol_free_shared_policy will not be called. | |
1438 | */ | |
71fe804b | 1439 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
1440 | break; |
1441 | } | |
5b04c689 PE |
1442 | } else |
1443 | shmem_free_inode(sb); | |
1da177e4 LT |
1444 | return inode; |
1445 | } | |
1446 | ||
0cd6144a JW |
1447 | bool shmem_mapping(struct address_space *mapping) |
1448 | { | |
1449 | return mapping->backing_dev_info == &shmem_backing_dev_info; | |
1450 | } | |
1451 | ||
1da177e4 | 1452 | #ifdef CONFIG_TMPFS |
92e1d5be | 1453 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 1454 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 1455 | |
6d9d88d0 JS |
1456 | #ifdef CONFIG_TMPFS_XATTR |
1457 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
1458 | #else | |
1459 | #define shmem_initxattrs NULL | |
1460 | #endif | |
1461 | ||
1da177e4 | 1462 | static int |
800d15a5 NP |
1463 | shmem_write_begin(struct file *file, struct address_space *mapping, |
1464 | loff_t pos, unsigned len, unsigned flags, | |
1465 | struct page **pagep, void **fsdata) | |
1da177e4 | 1466 | { |
800d15a5 NP |
1467 | struct inode *inode = mapping->host; |
1468 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
66d2f4d2 | 1469 | return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); |
800d15a5 NP |
1470 | } |
1471 | ||
1472 | static int | |
1473 | shmem_write_end(struct file *file, struct address_space *mapping, | |
1474 | loff_t pos, unsigned len, unsigned copied, | |
1475 | struct page *page, void *fsdata) | |
1476 | { | |
1477 | struct inode *inode = mapping->host; | |
1478 | ||
d3602444 HD |
1479 | if (pos + copied > inode->i_size) |
1480 | i_size_write(inode, pos + copied); | |
1481 | ||
ec9516fb HD |
1482 | if (!PageUptodate(page)) { |
1483 | if (copied < PAGE_CACHE_SIZE) { | |
1484 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
1485 | zero_user_segments(page, 0, from, | |
1486 | from + copied, PAGE_CACHE_SIZE); | |
1487 | } | |
1488 | SetPageUptodate(page); | |
1489 | } | |
800d15a5 | 1490 | set_page_dirty(page); |
6746aff7 | 1491 | unlock_page(page); |
800d15a5 NP |
1492 | page_cache_release(page); |
1493 | ||
800d15a5 | 1494 | return copied; |
1da177e4 LT |
1495 | } |
1496 | ||
2ba5bbed | 1497 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 1498 | { |
6e58e79d AV |
1499 | struct file *file = iocb->ki_filp; |
1500 | struct inode *inode = file_inode(file); | |
1da177e4 | 1501 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
1502 | pgoff_t index; |
1503 | unsigned long offset; | |
a0ee5ec5 | 1504 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 1505 | int error = 0; |
cb66a7a1 | 1506 | ssize_t retval = 0; |
6e58e79d | 1507 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
1508 | |
1509 | /* | |
1510 | * Might this read be for a stacking filesystem? Then when reading | |
1511 | * holes of a sparse file, we actually need to allocate those pages, | |
1512 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
1513 | */ | |
1514 | if (segment_eq(get_fs(), KERNEL_DS)) | |
1515 | sgp = SGP_DIRTY; | |
1da177e4 LT |
1516 | |
1517 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1518 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1519 | ||
1520 | for (;;) { | |
1521 | struct page *page = NULL; | |
41ffe5d5 HD |
1522 | pgoff_t end_index; |
1523 | unsigned long nr, ret; | |
1da177e4 LT |
1524 | loff_t i_size = i_size_read(inode); |
1525 | ||
1526 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1527 | if (index > end_index) | |
1528 | break; | |
1529 | if (index == end_index) { | |
1530 | nr = i_size & ~PAGE_CACHE_MASK; | |
1531 | if (nr <= offset) | |
1532 | break; | |
1533 | } | |
1534 | ||
6e58e79d AV |
1535 | error = shmem_getpage(inode, index, &page, sgp, NULL); |
1536 | if (error) { | |
1537 | if (error == -EINVAL) | |
1538 | error = 0; | |
1da177e4 LT |
1539 | break; |
1540 | } | |
d3602444 HD |
1541 | if (page) |
1542 | unlock_page(page); | |
1da177e4 LT |
1543 | |
1544 | /* | |
1545 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 1546 | * are called without i_mutex protection against truncate |
1da177e4 LT |
1547 | */ |
1548 | nr = PAGE_CACHE_SIZE; | |
1549 | i_size = i_size_read(inode); | |
1550 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1551 | if (index == end_index) { | |
1552 | nr = i_size & ~PAGE_CACHE_MASK; | |
1553 | if (nr <= offset) { | |
1554 | if (page) | |
1555 | page_cache_release(page); | |
1556 | break; | |
1557 | } | |
1558 | } | |
1559 | nr -= offset; | |
1560 | ||
1561 | if (page) { | |
1562 | /* | |
1563 | * If users can be writing to this page using arbitrary | |
1564 | * virtual addresses, take care about potential aliasing | |
1565 | * before reading the page on the kernel side. | |
1566 | */ | |
1567 | if (mapping_writably_mapped(mapping)) | |
1568 | flush_dcache_page(page); | |
1569 | /* | |
1570 | * Mark the page accessed if we read the beginning. | |
1571 | */ | |
1572 | if (!offset) | |
1573 | mark_page_accessed(page); | |
b5810039 | 1574 | } else { |
1da177e4 | 1575 | page = ZERO_PAGE(0); |
b5810039 NP |
1576 | page_cache_get(page); |
1577 | } | |
1da177e4 LT |
1578 | |
1579 | /* | |
1580 | * Ok, we have the page, and it's up-to-date, so | |
1581 | * now we can copy it to user space... | |
1da177e4 | 1582 | */ |
2ba5bbed | 1583 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 1584 | retval += ret; |
1da177e4 LT |
1585 | offset += ret; |
1586 | index += offset >> PAGE_CACHE_SHIFT; | |
1587 | offset &= ~PAGE_CACHE_MASK; | |
1588 | ||
1589 | page_cache_release(page); | |
2ba5bbed | 1590 | if (!iov_iter_count(to)) |
1da177e4 | 1591 | break; |
6e58e79d AV |
1592 | if (ret < nr) { |
1593 | error = -EFAULT; | |
1594 | break; | |
1595 | } | |
1da177e4 LT |
1596 | cond_resched(); |
1597 | } | |
1598 | ||
1599 | *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; | |
6e58e79d AV |
1600 | file_accessed(file); |
1601 | return retval ? retval : error; | |
1da177e4 LT |
1602 | } |
1603 | ||
708e3508 HD |
1604 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, |
1605 | struct pipe_inode_info *pipe, size_t len, | |
1606 | unsigned int flags) | |
1607 | { | |
1608 | struct address_space *mapping = in->f_mapping; | |
71f0e07a | 1609 | struct inode *inode = mapping->host; |
708e3508 HD |
1610 | unsigned int loff, nr_pages, req_pages; |
1611 | struct page *pages[PIPE_DEF_BUFFERS]; | |
1612 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
1613 | struct page *page; | |
1614 | pgoff_t index, end_index; | |
1615 | loff_t isize, left; | |
1616 | int error, page_nr; | |
1617 | struct splice_pipe_desc spd = { | |
1618 | .pages = pages, | |
1619 | .partial = partial, | |
047fe360 | 1620 | .nr_pages_max = PIPE_DEF_BUFFERS, |
708e3508 HD |
1621 | .flags = flags, |
1622 | .ops = &page_cache_pipe_buf_ops, | |
1623 | .spd_release = spd_release_page, | |
1624 | }; | |
1625 | ||
71f0e07a | 1626 | isize = i_size_read(inode); |
708e3508 HD |
1627 | if (unlikely(*ppos >= isize)) |
1628 | return 0; | |
1629 | ||
1630 | left = isize - *ppos; | |
1631 | if (unlikely(left < len)) | |
1632 | len = left; | |
1633 | ||
1634 | if (splice_grow_spd(pipe, &spd)) | |
1635 | return -ENOMEM; | |
1636 | ||
1637 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1638 | loff = *ppos & ~PAGE_CACHE_MASK; | |
1639 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
a786c06d | 1640 | nr_pages = min(req_pages, spd.nr_pages_max); |
708e3508 | 1641 | |
708e3508 HD |
1642 | spd.nr_pages = find_get_pages_contig(mapping, index, |
1643 | nr_pages, spd.pages); | |
1644 | index += spd.nr_pages; | |
708e3508 | 1645 | error = 0; |
708e3508 | 1646 | |
71f0e07a | 1647 | while (spd.nr_pages < nr_pages) { |
71f0e07a HD |
1648 | error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL); |
1649 | if (error) | |
1650 | break; | |
1651 | unlock_page(page); | |
708e3508 HD |
1652 | spd.pages[spd.nr_pages++] = page; |
1653 | index++; | |
1654 | } | |
1655 | ||
708e3508 HD |
1656 | index = *ppos >> PAGE_CACHE_SHIFT; |
1657 | nr_pages = spd.nr_pages; | |
1658 | spd.nr_pages = 0; | |
71f0e07a | 1659 | |
708e3508 HD |
1660 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { |
1661 | unsigned int this_len; | |
1662 | ||
1663 | if (!len) | |
1664 | break; | |
1665 | ||
708e3508 HD |
1666 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); |
1667 | page = spd.pages[page_nr]; | |
1668 | ||
71f0e07a | 1669 | if (!PageUptodate(page) || page->mapping != mapping) { |
71f0e07a HD |
1670 | error = shmem_getpage(inode, index, &page, |
1671 | SGP_CACHE, NULL); | |
1672 | if (error) | |
708e3508 | 1673 | break; |
71f0e07a HD |
1674 | unlock_page(page); |
1675 | page_cache_release(spd.pages[page_nr]); | |
1676 | spd.pages[page_nr] = page; | |
708e3508 | 1677 | } |
71f0e07a HD |
1678 | |
1679 | isize = i_size_read(inode); | |
708e3508 HD |
1680 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; |
1681 | if (unlikely(!isize || index > end_index)) | |
1682 | break; | |
1683 | ||
708e3508 HD |
1684 | if (end_index == index) { |
1685 | unsigned int plen; | |
1686 | ||
708e3508 HD |
1687 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
1688 | if (plen <= loff) | |
1689 | break; | |
1690 | ||
708e3508 HD |
1691 | this_len = min(this_len, plen - loff); |
1692 | len = this_len; | |
1693 | } | |
1694 | ||
1695 | spd.partial[page_nr].offset = loff; | |
1696 | spd.partial[page_nr].len = this_len; | |
1697 | len -= this_len; | |
1698 | loff = 0; | |
1699 | spd.nr_pages++; | |
1700 | index++; | |
1701 | } | |
1702 | ||
708e3508 HD |
1703 | while (page_nr < nr_pages) |
1704 | page_cache_release(spd.pages[page_nr++]); | |
708e3508 HD |
1705 | |
1706 | if (spd.nr_pages) | |
1707 | error = splice_to_pipe(pipe, &spd); | |
1708 | ||
047fe360 | 1709 | splice_shrink_spd(&spd); |
708e3508 HD |
1710 | |
1711 | if (error > 0) { | |
1712 | *ppos += error; | |
1713 | file_accessed(in); | |
1714 | } | |
1715 | return error; | |
1716 | } | |
1717 | ||
220f2ac9 HD |
1718 | /* |
1719 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
1720 | */ | |
1721 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 1722 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
1723 | { |
1724 | struct page *page; | |
1725 | struct pagevec pvec; | |
1726 | pgoff_t indices[PAGEVEC_SIZE]; | |
1727 | bool done = false; | |
1728 | int i; | |
1729 | ||
1730 | pagevec_init(&pvec, 0); | |
1731 | pvec.nr = 1; /* start small: we may be there already */ | |
1732 | while (!done) { | |
0cd6144a | 1733 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
1734 | pvec.nr, pvec.pages, indices); |
1735 | if (!pvec.nr) { | |
965c8e59 | 1736 | if (whence == SEEK_DATA) |
220f2ac9 HD |
1737 | index = end; |
1738 | break; | |
1739 | } | |
1740 | for (i = 0; i < pvec.nr; i++, index++) { | |
1741 | if (index < indices[i]) { | |
965c8e59 | 1742 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
1743 | done = true; |
1744 | break; | |
1745 | } | |
1746 | index = indices[i]; | |
1747 | } | |
1748 | page = pvec.pages[i]; | |
1749 | if (page && !radix_tree_exceptional_entry(page)) { | |
1750 | if (!PageUptodate(page)) | |
1751 | page = NULL; | |
1752 | } | |
1753 | if (index >= end || | |
965c8e59 AM |
1754 | (page && whence == SEEK_DATA) || |
1755 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
1756 | done = true; |
1757 | break; | |
1758 | } | |
1759 | } | |
0cd6144a | 1760 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
1761 | pagevec_release(&pvec); |
1762 | pvec.nr = PAGEVEC_SIZE; | |
1763 | cond_resched(); | |
1764 | } | |
1765 | return index; | |
1766 | } | |
1767 | ||
965c8e59 | 1768 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
1769 | { |
1770 | struct address_space *mapping = file->f_mapping; | |
1771 | struct inode *inode = mapping->host; | |
1772 | pgoff_t start, end; | |
1773 | loff_t new_offset; | |
1774 | ||
965c8e59 AM |
1775 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
1776 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 HD |
1777 | MAX_LFS_FILESIZE, i_size_read(inode)); |
1778 | mutex_lock(&inode->i_mutex); | |
1779 | /* We're holding i_mutex so we can access i_size directly */ | |
1780 | ||
1781 | if (offset < 0) | |
1782 | offset = -EINVAL; | |
1783 | else if (offset >= inode->i_size) | |
1784 | offset = -ENXIO; | |
1785 | else { | |
1786 | start = offset >> PAGE_CACHE_SHIFT; | |
1787 | end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
965c8e59 | 1788 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
220f2ac9 HD |
1789 | new_offset <<= PAGE_CACHE_SHIFT; |
1790 | if (new_offset > offset) { | |
1791 | if (new_offset < inode->i_size) | |
1792 | offset = new_offset; | |
965c8e59 | 1793 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
1794 | offset = -ENXIO; |
1795 | else | |
1796 | offset = inode->i_size; | |
1797 | } | |
1798 | } | |
1799 | ||
387aae6f HD |
1800 | if (offset >= 0) |
1801 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
220f2ac9 HD |
1802 | mutex_unlock(&inode->i_mutex); |
1803 | return offset; | |
1804 | } | |
1805 | ||
83e4fa9c HD |
1806 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
1807 | loff_t len) | |
1808 | { | |
496ad9aa | 1809 | struct inode *inode = file_inode(file); |
e2d12e22 | 1810 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1aac1400 | 1811 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
1812 | pgoff_t start, index, end; |
1813 | int error; | |
83e4fa9c | 1814 | |
13ace4d0 HD |
1815 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
1816 | return -EOPNOTSUPP; | |
1817 | ||
83e4fa9c HD |
1818 | mutex_lock(&inode->i_mutex); |
1819 | ||
1820 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1821 | struct address_space *mapping = file->f_mapping; | |
1822 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
1823 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 1824 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 1825 | |
8e205f77 | 1826 | shmem_falloc.waitq = &shmem_falloc_waitq; |
f00cdc6d HD |
1827 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
1828 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
1829 | spin_lock(&inode->i_lock); | |
1830 | inode->i_private = &shmem_falloc; | |
1831 | spin_unlock(&inode->i_lock); | |
1832 | ||
83e4fa9c HD |
1833 | if ((u64)unmap_end > (u64)unmap_start) |
1834 | unmap_mapping_range(mapping, unmap_start, | |
1835 | 1 + unmap_end - unmap_start, 0); | |
1836 | shmem_truncate_range(inode, offset, offset + len - 1); | |
1837 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
1838 | |
1839 | spin_lock(&inode->i_lock); | |
1840 | inode->i_private = NULL; | |
1841 | wake_up_all(&shmem_falloc_waitq); | |
1842 | spin_unlock(&inode->i_lock); | |
83e4fa9c | 1843 | error = 0; |
8e205f77 | 1844 | goto out; |
e2d12e22 HD |
1845 | } |
1846 | ||
1847 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
1848 | error = inode_newsize_ok(inode, offset + len); | |
1849 | if (error) | |
1850 | goto out; | |
1851 | ||
1852 | start = offset >> PAGE_CACHE_SHIFT; | |
1853 | end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
1854 | /* Try to avoid a swapstorm if len is impossible to satisfy */ | |
1855 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
1856 | error = -ENOSPC; | |
1857 | goto out; | |
83e4fa9c HD |
1858 | } |
1859 | ||
8e205f77 | 1860 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
1861 | shmem_falloc.start = start; |
1862 | shmem_falloc.next = start; | |
1863 | shmem_falloc.nr_falloced = 0; | |
1864 | shmem_falloc.nr_unswapped = 0; | |
1865 | spin_lock(&inode->i_lock); | |
1866 | inode->i_private = &shmem_falloc; | |
1867 | spin_unlock(&inode->i_lock); | |
1868 | ||
e2d12e22 HD |
1869 | for (index = start; index < end; index++) { |
1870 | struct page *page; | |
1871 | ||
1872 | /* | |
1873 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
1874 | * been interrupted because we are using up too much memory. | |
1875 | */ | |
1876 | if (signal_pending(current)) | |
1877 | error = -EINTR; | |
1aac1400 HD |
1878 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
1879 | error = -ENOMEM; | |
e2d12e22 | 1880 | else |
1635f6a7 | 1881 | error = shmem_getpage(inode, index, &page, SGP_FALLOC, |
e2d12e22 HD |
1882 | NULL); |
1883 | if (error) { | |
1635f6a7 HD |
1884 | /* Remove the !PageUptodate pages we added */ |
1885 | shmem_undo_range(inode, | |
1886 | (loff_t)start << PAGE_CACHE_SHIFT, | |
1887 | (loff_t)index << PAGE_CACHE_SHIFT, true); | |
1aac1400 | 1888 | goto undone; |
e2d12e22 HD |
1889 | } |
1890 | ||
1aac1400 HD |
1891 | /* |
1892 | * Inform shmem_writepage() how far we have reached. | |
1893 | * No need for lock or barrier: we have the page lock. | |
1894 | */ | |
1895 | shmem_falloc.next++; | |
1896 | if (!PageUptodate(page)) | |
1897 | shmem_falloc.nr_falloced++; | |
1898 | ||
e2d12e22 | 1899 | /* |
1635f6a7 HD |
1900 | * If !PageUptodate, leave it that way so that freeable pages |
1901 | * can be recognized if we need to rollback on error later. | |
1902 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
1903 | * than free the pages we are allocating (and SGP_CACHE pages |
1904 | * might still be clean: we now need to mark those dirty too). | |
1905 | */ | |
1906 | set_page_dirty(page); | |
1907 | unlock_page(page); | |
1908 | page_cache_release(page); | |
1909 | cond_resched(); | |
1910 | } | |
1911 | ||
1912 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
1913 | i_size_write(inode, offset + len); | |
e2d12e22 | 1914 | inode->i_ctime = CURRENT_TIME; |
1aac1400 HD |
1915 | undone: |
1916 | spin_lock(&inode->i_lock); | |
1917 | inode->i_private = NULL; | |
1918 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 1919 | out: |
83e4fa9c HD |
1920 | mutex_unlock(&inode->i_mutex); |
1921 | return error; | |
1922 | } | |
1923 | ||
726c3342 | 1924 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 1925 | { |
726c3342 | 1926 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
1927 | |
1928 | buf->f_type = TMPFS_MAGIC; | |
1929 | buf->f_bsize = PAGE_CACHE_SIZE; | |
1930 | buf->f_namelen = NAME_MAX; | |
0edd73b3 | 1931 | if (sbinfo->max_blocks) { |
1da177e4 | 1932 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
1933 | buf->f_bavail = |
1934 | buf->f_bfree = sbinfo->max_blocks - | |
1935 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
1936 | } |
1937 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
1938 | buf->f_files = sbinfo->max_inodes; |
1939 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
1940 | } |
1941 | /* else leave those fields 0 like simple_statfs */ | |
1942 | return 0; | |
1943 | } | |
1944 | ||
1945 | /* | |
1946 | * File creation. Allocate an inode, and we're done.. | |
1947 | */ | |
1948 | static int | |
1a67aafb | 1949 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 1950 | { |
0b0a0806 | 1951 | struct inode *inode; |
1da177e4 LT |
1952 | int error = -ENOSPC; |
1953 | ||
454abafe | 1954 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 1955 | if (inode) { |
feda821e CH |
1956 | error = simple_acl_create(dir, inode); |
1957 | if (error) | |
1958 | goto out_iput; | |
2a7dba39 | 1959 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 1960 | &dentry->d_name, |
6d9d88d0 | 1961 | shmem_initxattrs, NULL); |
feda821e CH |
1962 | if (error && error != -EOPNOTSUPP) |
1963 | goto out_iput; | |
37ec43cd | 1964 | |
718deb6b | 1965 | error = 0; |
1da177e4 LT |
1966 | dir->i_size += BOGO_DIRENT_SIZE; |
1967 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
1968 | d_instantiate(dentry, inode); | |
1969 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
1970 | } |
1971 | return error; | |
feda821e CH |
1972 | out_iput: |
1973 | iput(inode); | |
1974 | return error; | |
1da177e4 LT |
1975 | } |
1976 | ||
60545d0d AV |
1977 | static int |
1978 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
1979 | { | |
1980 | struct inode *inode; | |
1981 | int error = -ENOSPC; | |
1982 | ||
1983 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
1984 | if (inode) { | |
1985 | error = security_inode_init_security(inode, dir, | |
1986 | NULL, | |
1987 | shmem_initxattrs, NULL); | |
feda821e CH |
1988 | if (error && error != -EOPNOTSUPP) |
1989 | goto out_iput; | |
1990 | error = simple_acl_create(dir, inode); | |
1991 | if (error) | |
1992 | goto out_iput; | |
60545d0d AV |
1993 | d_tmpfile(dentry, inode); |
1994 | } | |
1995 | return error; | |
feda821e CH |
1996 | out_iput: |
1997 | iput(inode); | |
1998 | return error; | |
60545d0d AV |
1999 | } |
2000 | ||
18bb1db3 | 2001 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
2002 | { |
2003 | int error; | |
2004 | ||
2005 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
2006 | return error; | |
d8c76e6f | 2007 | inc_nlink(dir); |
1da177e4 LT |
2008 | return 0; |
2009 | } | |
2010 | ||
4acdaf27 | 2011 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2012 | bool excl) |
1da177e4 LT |
2013 | { |
2014 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
2015 | } | |
2016 | ||
2017 | /* | |
2018 | * Link a file.. | |
2019 | */ | |
2020 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
2021 | { | |
2022 | struct inode *inode = old_dentry->d_inode; | |
5b04c689 | 2023 | int ret; |
1da177e4 LT |
2024 | |
2025 | /* | |
2026 | * No ordinary (disk based) filesystem counts links as inodes; | |
2027 | * but each new link needs a new dentry, pinning lowmem, and | |
2028 | * tmpfs dentries cannot be pruned until they are unlinked. | |
2029 | */ | |
5b04c689 PE |
2030 | ret = shmem_reserve_inode(inode->i_sb); |
2031 | if (ret) | |
2032 | goto out; | |
1da177e4 LT |
2033 | |
2034 | dir->i_size += BOGO_DIRENT_SIZE; | |
2035 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 2036 | inc_nlink(inode); |
7de9c6ee | 2037 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
2038 | dget(dentry); /* Extra pinning count for the created dentry */ |
2039 | d_instantiate(dentry, inode); | |
5b04c689 PE |
2040 | out: |
2041 | return ret; | |
1da177e4 LT |
2042 | } |
2043 | ||
2044 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
2045 | { | |
2046 | struct inode *inode = dentry->d_inode; | |
2047 | ||
5b04c689 PE |
2048 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
2049 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
2050 | |
2051 | dir->i_size -= BOGO_DIRENT_SIZE; | |
2052 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 2053 | drop_nlink(inode); |
1da177e4 LT |
2054 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
2055 | return 0; | |
2056 | } | |
2057 | ||
2058 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
2059 | { | |
2060 | if (!simple_empty(dentry)) | |
2061 | return -ENOTEMPTY; | |
2062 | ||
9a53c3a7 DH |
2063 | drop_nlink(dentry->d_inode); |
2064 | drop_nlink(dir); | |
1da177e4 LT |
2065 | return shmem_unlink(dir, dentry); |
2066 | } | |
2067 | ||
2068 | /* | |
2069 | * The VFS layer already does all the dentry stuff for rename, | |
2070 | * we just have to decrement the usage count for the target if | |
2071 | * it exists so that the VFS layer correctly free's it when it | |
2072 | * gets overwritten. | |
2073 | */ | |
2074 | static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) | |
2075 | { | |
2076 | struct inode *inode = old_dentry->d_inode; | |
2077 | int they_are_dirs = S_ISDIR(inode->i_mode); | |
2078 | ||
2079 | if (!simple_empty(new_dentry)) | |
2080 | return -ENOTEMPTY; | |
2081 | ||
2082 | if (new_dentry->d_inode) { | |
2083 | (void) shmem_unlink(new_dir, new_dentry); | |
2084 | if (they_are_dirs) | |
9a53c3a7 | 2085 | drop_nlink(old_dir); |
1da177e4 | 2086 | } else if (they_are_dirs) { |
9a53c3a7 | 2087 | drop_nlink(old_dir); |
d8c76e6f | 2088 | inc_nlink(new_dir); |
1da177e4 LT |
2089 | } |
2090 | ||
2091 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
2092 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
2093 | old_dir->i_ctime = old_dir->i_mtime = | |
2094 | new_dir->i_ctime = new_dir->i_mtime = | |
2095 | inode->i_ctime = CURRENT_TIME; | |
2096 | return 0; | |
2097 | } | |
2098 | ||
2099 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
2100 | { | |
2101 | int error; | |
2102 | int len; | |
2103 | struct inode *inode; | |
9276aad6 | 2104 | struct page *page; |
1da177e4 LT |
2105 | char *kaddr; |
2106 | struct shmem_inode_info *info; | |
2107 | ||
2108 | len = strlen(symname) + 1; | |
2109 | if (len > PAGE_CACHE_SIZE) | |
2110 | return -ENAMETOOLONG; | |
2111 | ||
454abafe | 2112 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
2113 | if (!inode) |
2114 | return -ENOSPC; | |
2115 | ||
9d8f13ba | 2116 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 2117 | shmem_initxattrs, NULL); |
570bc1c2 SS |
2118 | if (error) { |
2119 | if (error != -EOPNOTSUPP) { | |
2120 | iput(inode); | |
2121 | return error; | |
2122 | } | |
2123 | error = 0; | |
2124 | } | |
2125 | ||
1da177e4 LT |
2126 | info = SHMEM_I(inode); |
2127 | inode->i_size = len-1; | |
69f07ec9 HD |
2128 | if (len <= SHORT_SYMLINK_LEN) { |
2129 | info->symlink = kmemdup(symname, len, GFP_KERNEL); | |
2130 | if (!info->symlink) { | |
2131 | iput(inode); | |
2132 | return -ENOMEM; | |
2133 | } | |
2134 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 LT |
2135 | } else { |
2136 | error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); | |
2137 | if (error) { | |
2138 | iput(inode); | |
2139 | return error; | |
2140 | } | |
14fcc23f | 2141 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 2142 | inode->i_op = &shmem_symlink_inode_operations; |
9b04c5fe | 2143 | kaddr = kmap_atomic(page); |
1da177e4 | 2144 | memcpy(kaddr, symname, len); |
9b04c5fe | 2145 | kunmap_atomic(kaddr); |
ec9516fb | 2146 | SetPageUptodate(page); |
1da177e4 | 2147 | set_page_dirty(page); |
6746aff7 | 2148 | unlock_page(page); |
1da177e4 LT |
2149 | page_cache_release(page); |
2150 | } | |
1da177e4 LT |
2151 | dir->i_size += BOGO_DIRENT_SIZE; |
2152 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2153 | d_instantiate(dentry, inode); | |
2154 | dget(dentry); | |
2155 | return 0; | |
2156 | } | |
2157 | ||
69f07ec9 | 2158 | static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 2159 | { |
69f07ec9 | 2160 | nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink); |
cc314eef | 2161 | return NULL; |
1da177e4 LT |
2162 | } |
2163 | ||
cc314eef | 2164 | static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) |
1da177e4 LT |
2165 | { |
2166 | struct page *page = NULL; | |
41ffe5d5 HD |
2167 | int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); |
2168 | nd_set_link(nd, error ? ERR_PTR(error) : kmap(page)); | |
d3602444 HD |
2169 | if (page) |
2170 | unlock_page(page); | |
cc314eef | 2171 | return page; |
1da177e4 LT |
2172 | } |
2173 | ||
cc314eef | 2174 | static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 LT |
2175 | { |
2176 | if (!IS_ERR(nd_get_link(nd))) { | |
cc314eef | 2177 | struct page *page = cookie; |
1da177e4 LT |
2178 | kunmap(page); |
2179 | mark_page_accessed(page); | |
2180 | page_cache_release(page); | |
1da177e4 LT |
2181 | } |
2182 | } | |
2183 | ||
b09e0fa4 | 2184 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 2185 | /* |
b09e0fa4 EP |
2186 | * Superblocks without xattr inode operations may get some security.* xattr |
2187 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
2188 | * like ACLs, we also need to implement the security.* handlers at |
2189 | * filesystem level, though. | |
2190 | */ | |
2191 | ||
6d9d88d0 JS |
2192 | /* |
2193 | * Callback for security_inode_init_security() for acquiring xattrs. | |
2194 | */ | |
2195 | static int shmem_initxattrs(struct inode *inode, | |
2196 | const struct xattr *xattr_array, | |
2197 | void *fs_info) | |
2198 | { | |
2199 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2200 | const struct xattr *xattr; | |
38f38657 | 2201 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
2202 | size_t len; |
2203 | ||
2204 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 2205 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
2206 | if (!new_xattr) |
2207 | return -ENOMEM; | |
2208 | ||
2209 | len = strlen(xattr->name) + 1; | |
2210 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
2211 | GFP_KERNEL); | |
2212 | if (!new_xattr->name) { | |
2213 | kfree(new_xattr); | |
2214 | return -ENOMEM; | |
2215 | } | |
2216 | ||
2217 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
2218 | XATTR_SECURITY_PREFIX_LEN); | |
2219 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
2220 | xattr->name, len); | |
2221 | ||
38f38657 | 2222 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
2223 | } |
2224 | ||
2225 | return 0; | |
2226 | } | |
2227 | ||
bb435453 | 2228 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
b09e0fa4 | 2229 | #ifdef CONFIG_TMPFS_POSIX_ACL |
feda821e CH |
2230 | &posix_acl_access_xattr_handler, |
2231 | &posix_acl_default_xattr_handler, | |
b09e0fa4 | 2232 | #endif |
39f0247d AG |
2233 | NULL |
2234 | }; | |
b09e0fa4 EP |
2235 | |
2236 | static int shmem_xattr_validate(const char *name) | |
2237 | { | |
2238 | struct { const char *prefix; size_t len; } arr[] = { | |
2239 | { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN }, | |
2240 | { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN } | |
2241 | }; | |
2242 | int i; | |
2243 | ||
2244 | for (i = 0; i < ARRAY_SIZE(arr); i++) { | |
2245 | size_t preflen = arr[i].len; | |
2246 | if (strncmp(name, arr[i].prefix, preflen) == 0) { | |
2247 | if (!name[preflen]) | |
2248 | return -EINVAL; | |
2249 | return 0; | |
2250 | } | |
2251 | } | |
2252 | return -EOPNOTSUPP; | |
2253 | } | |
2254 | ||
2255 | static ssize_t shmem_getxattr(struct dentry *dentry, const char *name, | |
2256 | void *buffer, size_t size) | |
2257 | { | |
38f38657 | 2258 | struct shmem_inode_info *info = SHMEM_I(dentry->d_inode); |
b09e0fa4 EP |
2259 | int err; |
2260 | ||
2261 | /* | |
2262 | * If this is a request for a synthetic attribute in the system.* | |
2263 | * namespace use the generic infrastructure to resolve a handler | |
2264 | * for it via sb->s_xattr. | |
2265 | */ | |
2266 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2267 | return generic_getxattr(dentry, name, buffer, size); | |
2268 | ||
2269 | err = shmem_xattr_validate(name); | |
2270 | if (err) | |
2271 | return err; | |
2272 | ||
38f38657 | 2273 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
2274 | } |
2275 | ||
2276 | static int shmem_setxattr(struct dentry *dentry, const char *name, | |
2277 | const void *value, size_t size, int flags) | |
2278 | { | |
38f38657 | 2279 | struct shmem_inode_info *info = SHMEM_I(dentry->d_inode); |
b09e0fa4 EP |
2280 | int err; |
2281 | ||
2282 | /* | |
2283 | * If this is a request for a synthetic attribute in the system.* | |
2284 | * namespace use the generic infrastructure to resolve a handler | |
2285 | * for it via sb->s_xattr. | |
2286 | */ | |
2287 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2288 | return generic_setxattr(dentry, name, value, size, flags); | |
2289 | ||
2290 | err = shmem_xattr_validate(name); | |
2291 | if (err) | |
2292 | return err; | |
2293 | ||
38f38657 | 2294 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
2295 | } |
2296 | ||
2297 | static int shmem_removexattr(struct dentry *dentry, const char *name) | |
2298 | { | |
38f38657 | 2299 | struct shmem_inode_info *info = SHMEM_I(dentry->d_inode); |
b09e0fa4 EP |
2300 | int err; |
2301 | ||
2302 | /* | |
2303 | * If this is a request for a synthetic attribute in the system.* | |
2304 | * namespace use the generic infrastructure to resolve a handler | |
2305 | * for it via sb->s_xattr. | |
2306 | */ | |
2307 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2308 | return generic_removexattr(dentry, name); | |
2309 | ||
2310 | err = shmem_xattr_validate(name); | |
2311 | if (err) | |
2312 | return err; | |
2313 | ||
38f38657 | 2314 | return simple_xattr_remove(&info->xattrs, name); |
b09e0fa4 EP |
2315 | } |
2316 | ||
2317 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
2318 | { | |
38f38657 AR |
2319 | struct shmem_inode_info *info = SHMEM_I(dentry->d_inode); |
2320 | return simple_xattr_list(&info->xattrs, buffer, size); | |
b09e0fa4 EP |
2321 | } |
2322 | #endif /* CONFIG_TMPFS_XATTR */ | |
2323 | ||
69f07ec9 | 2324 | static const struct inode_operations shmem_short_symlink_operations = { |
b09e0fa4 | 2325 | .readlink = generic_readlink, |
69f07ec9 | 2326 | .follow_link = shmem_follow_short_symlink, |
b09e0fa4 EP |
2327 | #ifdef CONFIG_TMPFS_XATTR |
2328 | .setxattr = shmem_setxattr, | |
2329 | .getxattr = shmem_getxattr, | |
2330 | .listxattr = shmem_listxattr, | |
2331 | .removexattr = shmem_removexattr, | |
2332 | #endif | |
2333 | }; | |
2334 | ||
2335 | static const struct inode_operations shmem_symlink_inode_operations = { | |
2336 | .readlink = generic_readlink, | |
2337 | .follow_link = shmem_follow_link, | |
2338 | .put_link = shmem_put_link, | |
2339 | #ifdef CONFIG_TMPFS_XATTR | |
2340 | .setxattr = shmem_setxattr, | |
2341 | .getxattr = shmem_getxattr, | |
2342 | .listxattr = shmem_listxattr, | |
2343 | .removexattr = shmem_removexattr, | |
39f0247d | 2344 | #endif |
b09e0fa4 | 2345 | }; |
39f0247d | 2346 | |
91828a40 DG |
2347 | static struct dentry *shmem_get_parent(struct dentry *child) |
2348 | { | |
2349 | return ERR_PTR(-ESTALE); | |
2350 | } | |
2351 | ||
2352 | static int shmem_match(struct inode *ino, void *vfh) | |
2353 | { | |
2354 | __u32 *fh = vfh; | |
2355 | __u64 inum = fh[2]; | |
2356 | inum = (inum << 32) | fh[1]; | |
2357 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
2358 | } | |
2359 | ||
480b116c CH |
2360 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
2361 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 2362 | { |
91828a40 | 2363 | struct inode *inode; |
480b116c | 2364 | struct dentry *dentry = NULL; |
35c2a7f4 | 2365 | u64 inum; |
480b116c CH |
2366 | |
2367 | if (fh_len < 3) | |
2368 | return NULL; | |
91828a40 | 2369 | |
35c2a7f4 HD |
2370 | inum = fid->raw[2]; |
2371 | inum = (inum << 32) | fid->raw[1]; | |
2372 | ||
480b116c CH |
2373 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
2374 | shmem_match, fid->raw); | |
91828a40 | 2375 | if (inode) { |
480b116c | 2376 | dentry = d_find_alias(inode); |
91828a40 DG |
2377 | iput(inode); |
2378 | } | |
2379 | ||
480b116c | 2380 | return dentry; |
91828a40 DG |
2381 | } |
2382 | ||
b0b0382b AV |
2383 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
2384 | struct inode *parent) | |
91828a40 | 2385 | { |
5fe0c237 AK |
2386 | if (*len < 3) { |
2387 | *len = 3; | |
94e07a75 | 2388 | return FILEID_INVALID; |
5fe0c237 | 2389 | } |
91828a40 | 2390 | |
1d3382cb | 2391 | if (inode_unhashed(inode)) { |
91828a40 DG |
2392 | /* Unfortunately insert_inode_hash is not idempotent, |
2393 | * so as we hash inodes here rather than at creation | |
2394 | * time, we need a lock to ensure we only try | |
2395 | * to do it once | |
2396 | */ | |
2397 | static DEFINE_SPINLOCK(lock); | |
2398 | spin_lock(&lock); | |
1d3382cb | 2399 | if (inode_unhashed(inode)) |
91828a40 DG |
2400 | __insert_inode_hash(inode, |
2401 | inode->i_ino + inode->i_generation); | |
2402 | spin_unlock(&lock); | |
2403 | } | |
2404 | ||
2405 | fh[0] = inode->i_generation; | |
2406 | fh[1] = inode->i_ino; | |
2407 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
2408 | ||
2409 | *len = 3; | |
2410 | return 1; | |
2411 | } | |
2412 | ||
39655164 | 2413 | static const struct export_operations shmem_export_ops = { |
91828a40 | 2414 | .get_parent = shmem_get_parent, |
91828a40 | 2415 | .encode_fh = shmem_encode_fh, |
480b116c | 2416 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
2417 | }; |
2418 | ||
680d794b AM |
2419 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
2420 | bool remount) | |
1da177e4 LT |
2421 | { |
2422 | char *this_char, *value, *rest; | |
49cd0a5c | 2423 | struct mempolicy *mpol = NULL; |
8751e039 EB |
2424 | uid_t uid; |
2425 | gid_t gid; | |
1da177e4 | 2426 | |
b00dc3ad HD |
2427 | while (options != NULL) { |
2428 | this_char = options; | |
2429 | for (;;) { | |
2430 | /* | |
2431 | * NUL-terminate this option: unfortunately, | |
2432 | * mount options form a comma-separated list, | |
2433 | * but mpol's nodelist may also contain commas. | |
2434 | */ | |
2435 | options = strchr(options, ','); | |
2436 | if (options == NULL) | |
2437 | break; | |
2438 | options++; | |
2439 | if (!isdigit(*options)) { | |
2440 | options[-1] = '\0'; | |
2441 | break; | |
2442 | } | |
2443 | } | |
1da177e4 LT |
2444 | if (!*this_char) |
2445 | continue; | |
2446 | if ((value = strchr(this_char,'=')) != NULL) { | |
2447 | *value++ = 0; | |
2448 | } else { | |
2449 | printk(KERN_ERR | |
2450 | "tmpfs: No value for mount option '%s'\n", | |
2451 | this_char); | |
49cd0a5c | 2452 | goto error; |
1da177e4 LT |
2453 | } |
2454 | ||
2455 | if (!strcmp(this_char,"size")) { | |
2456 | unsigned long long size; | |
2457 | size = memparse(value,&rest); | |
2458 | if (*rest == '%') { | |
2459 | size <<= PAGE_SHIFT; | |
2460 | size *= totalram_pages; | |
2461 | do_div(size, 100); | |
2462 | rest++; | |
2463 | } | |
2464 | if (*rest) | |
2465 | goto bad_val; | |
680d794b AM |
2466 | sbinfo->max_blocks = |
2467 | DIV_ROUND_UP(size, PAGE_CACHE_SIZE); | |
1da177e4 | 2468 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 2469 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
2470 | if (*rest) |
2471 | goto bad_val; | |
2472 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 2473 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
2474 | if (*rest) |
2475 | goto bad_val; | |
2476 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 2477 | if (remount) |
1da177e4 | 2478 | continue; |
680d794b | 2479 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
2480 | if (*rest) |
2481 | goto bad_val; | |
2482 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 2483 | if (remount) |
1da177e4 | 2484 | continue; |
8751e039 | 2485 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2486 | if (*rest) |
2487 | goto bad_val; | |
8751e039 EB |
2488 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
2489 | if (!uid_valid(sbinfo->uid)) | |
2490 | goto bad_val; | |
1da177e4 | 2491 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 2492 | if (remount) |
1da177e4 | 2493 | continue; |
8751e039 | 2494 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2495 | if (*rest) |
2496 | goto bad_val; | |
8751e039 EB |
2497 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
2498 | if (!gid_valid(sbinfo->gid)) | |
2499 | goto bad_val; | |
7339ff83 | 2500 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
2501 | mpol_put(mpol); |
2502 | mpol = NULL; | |
2503 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 2504 | goto bad_val; |
1da177e4 LT |
2505 | } else { |
2506 | printk(KERN_ERR "tmpfs: Bad mount option %s\n", | |
2507 | this_char); | |
49cd0a5c | 2508 | goto error; |
1da177e4 LT |
2509 | } |
2510 | } | |
49cd0a5c | 2511 | sbinfo->mpol = mpol; |
1da177e4 LT |
2512 | return 0; |
2513 | ||
2514 | bad_val: | |
2515 | printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", | |
2516 | value, this_char); | |
49cd0a5c GT |
2517 | error: |
2518 | mpol_put(mpol); | |
1da177e4 LT |
2519 | return 1; |
2520 | ||
2521 | } | |
2522 | ||
2523 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
2524 | { | |
2525 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 2526 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
2527 | unsigned long inodes; |
2528 | int error = -EINVAL; | |
2529 | ||
5f00110f | 2530 | config.mpol = NULL; |
680d794b | 2531 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 2532 | return error; |
1da177e4 | 2533 | |
0edd73b3 | 2534 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 2535 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 2536 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 2537 | goto out; |
680d794b | 2538 | if (config.max_inodes < inodes) |
0edd73b3 HD |
2539 | goto out; |
2540 | /* | |
54af6042 | 2541 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
2542 | * but we must separately disallow unlimited->limited, because |
2543 | * in that case we have no record of how much is already in use. | |
2544 | */ | |
680d794b | 2545 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 2546 | goto out; |
680d794b | 2547 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
2548 | goto out; |
2549 | ||
2550 | error = 0; | |
680d794b | 2551 | sbinfo->max_blocks = config.max_blocks; |
680d794b AM |
2552 | sbinfo->max_inodes = config.max_inodes; |
2553 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 2554 | |
5f00110f GT |
2555 | /* |
2556 | * Preserve previous mempolicy unless mpol remount option was specified. | |
2557 | */ | |
2558 | if (config.mpol) { | |
2559 | mpol_put(sbinfo->mpol); | |
2560 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
2561 | } | |
0edd73b3 HD |
2562 | out: |
2563 | spin_unlock(&sbinfo->stat_lock); | |
2564 | return error; | |
1da177e4 | 2565 | } |
680d794b | 2566 | |
34c80b1d | 2567 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 2568 | { |
34c80b1d | 2569 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b AM |
2570 | |
2571 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
2572 | seq_printf(seq, ",size=%luk", | |
2573 | sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); | |
2574 | if (sbinfo->max_inodes != shmem_default_max_inodes()) | |
2575 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
2576 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 2577 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
2578 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
2579 | seq_printf(seq, ",uid=%u", | |
2580 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
2581 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
2582 | seq_printf(seq, ",gid=%u", | |
2583 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
71fe804b | 2584 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b AM |
2585 | return 0; |
2586 | } | |
2587 | #endif /* CONFIG_TMPFS */ | |
1da177e4 LT |
2588 | |
2589 | static void shmem_put_super(struct super_block *sb) | |
2590 | { | |
602586a8 HD |
2591 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
2592 | ||
2593 | percpu_counter_destroy(&sbinfo->used_blocks); | |
49cd0a5c | 2594 | mpol_put(sbinfo->mpol); |
602586a8 | 2595 | kfree(sbinfo); |
1da177e4 LT |
2596 | sb->s_fs_info = NULL; |
2597 | } | |
2598 | ||
2b2af54a | 2599 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
2600 | { |
2601 | struct inode *inode; | |
0edd73b3 | 2602 | struct shmem_sb_info *sbinfo; |
680d794b AM |
2603 | int err = -ENOMEM; |
2604 | ||
2605 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 2606 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b AM |
2607 | L1_CACHE_BYTES), GFP_KERNEL); |
2608 | if (!sbinfo) | |
2609 | return -ENOMEM; | |
2610 | ||
680d794b | 2611 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
2612 | sbinfo->uid = current_fsuid(); |
2613 | sbinfo->gid = current_fsgid(); | |
680d794b | 2614 | sb->s_fs_info = sbinfo; |
1da177e4 | 2615 | |
0edd73b3 | 2616 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
2617 | /* |
2618 | * Per default we only allow half of the physical ram per | |
2619 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
2620 | * but the internal instance is left unlimited. | |
2621 | */ | |
ca4e0519 | 2622 | if (!(sb->s_flags & MS_KERNMOUNT)) { |
680d794b AM |
2623 | sbinfo->max_blocks = shmem_default_max_blocks(); |
2624 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
2625 | if (shmem_parse_options(data, sbinfo, false)) { | |
2626 | err = -EINVAL; | |
2627 | goto failed; | |
2628 | } | |
ca4e0519 AV |
2629 | } else { |
2630 | sb->s_flags |= MS_NOUSER; | |
1da177e4 | 2631 | } |
91828a40 | 2632 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 2633 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
2634 | #else |
2635 | sb->s_flags |= MS_NOUSER; | |
2636 | #endif | |
2637 | ||
0edd73b3 | 2638 | spin_lock_init(&sbinfo->stat_lock); |
602586a8 HD |
2639 | if (percpu_counter_init(&sbinfo->used_blocks, 0)) |
2640 | goto failed; | |
680d794b | 2641 | sbinfo->free_inodes = sbinfo->max_inodes; |
0edd73b3 | 2642 | |
285b2c4f | 2643 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
1da177e4 LT |
2644 | sb->s_blocksize = PAGE_CACHE_SIZE; |
2645 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
2646 | sb->s_magic = TMPFS_MAGIC; | |
2647 | sb->s_op = &shmem_ops; | |
cfd95a9c | 2648 | sb->s_time_gran = 1; |
b09e0fa4 | 2649 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 2650 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
2651 | #endif |
2652 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
2653 | sb->s_flags |= MS_POSIXACL; |
2654 | #endif | |
0edd73b3 | 2655 | |
454abafe | 2656 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
2657 | if (!inode) |
2658 | goto failed; | |
680d794b AM |
2659 | inode->i_uid = sbinfo->uid; |
2660 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
2661 | sb->s_root = d_make_root(inode); |
2662 | if (!sb->s_root) | |
48fde701 | 2663 | goto failed; |
1da177e4 LT |
2664 | return 0; |
2665 | ||
1da177e4 LT |
2666 | failed: |
2667 | shmem_put_super(sb); | |
2668 | return err; | |
2669 | } | |
2670 | ||
fcc234f8 | 2671 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
2672 | |
2673 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
2674 | { | |
41ffe5d5 HD |
2675 | struct shmem_inode_info *info; |
2676 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
2677 | if (!info) | |
1da177e4 | 2678 | return NULL; |
41ffe5d5 | 2679 | return &info->vfs_inode; |
1da177e4 LT |
2680 | } |
2681 | ||
41ffe5d5 | 2682 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
2683 | { |
2684 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
2685 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
2686 | } | |
2687 | ||
1da177e4 LT |
2688 | static void shmem_destroy_inode(struct inode *inode) |
2689 | { | |
09208d15 | 2690 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 2691 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 2692 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
2693 | } |
2694 | ||
41ffe5d5 | 2695 | static void shmem_init_inode(void *foo) |
1da177e4 | 2696 | { |
41ffe5d5 HD |
2697 | struct shmem_inode_info *info = foo; |
2698 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
2699 | } |
2700 | ||
41ffe5d5 | 2701 | static int shmem_init_inodecache(void) |
1da177e4 LT |
2702 | { |
2703 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
2704 | sizeof(struct shmem_inode_info), | |
41ffe5d5 | 2705 | 0, SLAB_PANIC, shmem_init_inode); |
1da177e4 LT |
2706 | return 0; |
2707 | } | |
2708 | ||
41ffe5d5 | 2709 | static void shmem_destroy_inodecache(void) |
1da177e4 | 2710 | { |
1a1d92c1 | 2711 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
2712 | } |
2713 | ||
f5e54d6e | 2714 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 2715 | .writepage = shmem_writepage, |
76719325 | 2716 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 2717 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
2718 | .write_begin = shmem_write_begin, |
2719 | .write_end = shmem_write_end, | |
1da177e4 | 2720 | #endif |
304dbdb7 | 2721 | .migratepage = migrate_page, |
aa261f54 | 2722 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
2723 | }; |
2724 | ||
15ad7cdc | 2725 | static const struct file_operations shmem_file_operations = { |
1da177e4 LT |
2726 | .mmap = shmem_mmap, |
2727 | #ifdef CONFIG_TMPFS | |
220f2ac9 | 2728 | .llseek = shmem_file_llseek, |
2ba5bbed | 2729 | .read = new_sync_read, |
8174202b | 2730 | .write = new_sync_write, |
2ba5bbed | 2731 | .read_iter = shmem_file_read_iter, |
8174202b | 2732 | .write_iter = generic_file_write_iter, |
1b061d92 | 2733 | .fsync = noop_fsync, |
708e3508 | 2734 | .splice_read = shmem_file_splice_read, |
f6cb85d0 | 2735 | .splice_write = iter_file_splice_write, |
83e4fa9c | 2736 | .fallocate = shmem_fallocate, |
1da177e4 LT |
2737 | #endif |
2738 | }; | |
2739 | ||
92e1d5be | 2740 | static const struct inode_operations shmem_inode_operations = { |
94c1e62d | 2741 | .setattr = shmem_setattr, |
b09e0fa4 EP |
2742 | #ifdef CONFIG_TMPFS_XATTR |
2743 | .setxattr = shmem_setxattr, | |
2744 | .getxattr = shmem_getxattr, | |
2745 | .listxattr = shmem_listxattr, | |
2746 | .removexattr = shmem_removexattr, | |
feda821e | 2747 | .set_acl = simple_set_acl, |
b09e0fa4 | 2748 | #endif |
1da177e4 LT |
2749 | }; |
2750 | ||
92e1d5be | 2751 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
2752 | #ifdef CONFIG_TMPFS |
2753 | .create = shmem_create, | |
2754 | .lookup = simple_lookup, | |
2755 | .link = shmem_link, | |
2756 | .unlink = shmem_unlink, | |
2757 | .symlink = shmem_symlink, | |
2758 | .mkdir = shmem_mkdir, | |
2759 | .rmdir = shmem_rmdir, | |
2760 | .mknod = shmem_mknod, | |
2761 | .rename = shmem_rename, | |
60545d0d | 2762 | .tmpfile = shmem_tmpfile, |
1da177e4 | 2763 | #endif |
b09e0fa4 EP |
2764 | #ifdef CONFIG_TMPFS_XATTR |
2765 | .setxattr = shmem_setxattr, | |
2766 | .getxattr = shmem_getxattr, | |
2767 | .listxattr = shmem_listxattr, | |
2768 | .removexattr = shmem_removexattr, | |
2769 | #endif | |
39f0247d | 2770 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 2771 | .setattr = shmem_setattr, |
feda821e | 2772 | .set_acl = simple_set_acl, |
39f0247d AG |
2773 | #endif |
2774 | }; | |
2775 | ||
92e1d5be | 2776 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 EP |
2777 | #ifdef CONFIG_TMPFS_XATTR |
2778 | .setxattr = shmem_setxattr, | |
2779 | .getxattr = shmem_getxattr, | |
2780 | .listxattr = shmem_listxattr, | |
2781 | .removexattr = shmem_removexattr, | |
2782 | #endif | |
39f0247d | 2783 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 2784 | .setattr = shmem_setattr, |
feda821e | 2785 | .set_acl = simple_set_acl, |
39f0247d | 2786 | #endif |
1da177e4 LT |
2787 | }; |
2788 | ||
759b9775 | 2789 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
2790 | .alloc_inode = shmem_alloc_inode, |
2791 | .destroy_inode = shmem_destroy_inode, | |
2792 | #ifdef CONFIG_TMPFS | |
2793 | .statfs = shmem_statfs, | |
2794 | .remount_fs = shmem_remount_fs, | |
680d794b | 2795 | .show_options = shmem_show_options, |
1da177e4 | 2796 | #endif |
1f895f75 | 2797 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
2798 | .drop_inode = generic_delete_inode, |
2799 | .put_super = shmem_put_super, | |
2800 | }; | |
2801 | ||
f0f37e2f | 2802 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 2803 | .fault = shmem_fault, |
d7c17551 | 2804 | .map_pages = filemap_map_pages, |
1da177e4 LT |
2805 | #ifdef CONFIG_NUMA |
2806 | .set_policy = shmem_set_policy, | |
2807 | .get_policy = shmem_get_policy, | |
2808 | #endif | |
0b173bc4 | 2809 | .remap_pages = generic_file_remap_pages, |
1da177e4 LT |
2810 | }; |
2811 | ||
3c26ff6e AV |
2812 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
2813 | int flags, const char *dev_name, void *data) | |
1da177e4 | 2814 | { |
3c26ff6e | 2815 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
2816 | } |
2817 | ||
41ffe5d5 | 2818 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
2819 | .owner = THIS_MODULE, |
2820 | .name = "tmpfs", | |
3c26ff6e | 2821 | .mount = shmem_mount, |
1da177e4 | 2822 | .kill_sb = kill_litter_super, |
2b8576cb | 2823 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 2824 | }; |
1da177e4 | 2825 | |
41ffe5d5 | 2826 | int __init shmem_init(void) |
1da177e4 LT |
2827 | { |
2828 | int error; | |
2829 | ||
16203a7a RL |
2830 | /* If rootfs called this, don't re-init */ |
2831 | if (shmem_inode_cachep) | |
2832 | return 0; | |
2833 | ||
e0bf68dd PZ |
2834 | error = bdi_init(&shmem_backing_dev_info); |
2835 | if (error) | |
2836 | goto out4; | |
2837 | ||
41ffe5d5 | 2838 | error = shmem_init_inodecache(); |
1da177e4 LT |
2839 | if (error) |
2840 | goto out3; | |
2841 | ||
41ffe5d5 | 2842 | error = register_filesystem(&shmem_fs_type); |
1da177e4 LT |
2843 | if (error) { |
2844 | printk(KERN_ERR "Could not register tmpfs\n"); | |
2845 | goto out2; | |
2846 | } | |
95dc112a | 2847 | |
ca4e0519 | 2848 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
2849 | if (IS_ERR(shm_mnt)) { |
2850 | error = PTR_ERR(shm_mnt); | |
2851 | printk(KERN_ERR "Could not kern_mount tmpfs\n"); | |
2852 | goto out1; | |
2853 | } | |
2854 | return 0; | |
2855 | ||
2856 | out1: | |
41ffe5d5 | 2857 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 2858 | out2: |
41ffe5d5 | 2859 | shmem_destroy_inodecache(); |
1da177e4 | 2860 | out3: |
e0bf68dd PZ |
2861 | bdi_destroy(&shmem_backing_dev_info); |
2862 | out4: | |
1da177e4 LT |
2863 | shm_mnt = ERR_PTR(error); |
2864 | return error; | |
2865 | } | |
853ac43a MM |
2866 | |
2867 | #else /* !CONFIG_SHMEM */ | |
2868 | ||
2869 | /* | |
2870 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
2871 | * | |
2872 | * This is intended for small system where the benefits of the full | |
2873 | * shmem code (swap-backed and resource-limited) are outweighed by | |
2874 | * their complexity. On systems without swap this code should be | |
2875 | * effectively equivalent, but much lighter weight. | |
2876 | */ | |
2877 | ||
41ffe5d5 | 2878 | static struct file_system_type shmem_fs_type = { |
853ac43a | 2879 | .name = "tmpfs", |
3c26ff6e | 2880 | .mount = ramfs_mount, |
853ac43a | 2881 | .kill_sb = kill_litter_super, |
2b8576cb | 2882 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
2883 | }; |
2884 | ||
41ffe5d5 | 2885 | int __init shmem_init(void) |
853ac43a | 2886 | { |
41ffe5d5 | 2887 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 2888 | |
41ffe5d5 | 2889 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
2890 | BUG_ON(IS_ERR(shm_mnt)); |
2891 | ||
2892 | return 0; | |
2893 | } | |
2894 | ||
41ffe5d5 | 2895 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
2896 | { |
2897 | return 0; | |
2898 | } | |
2899 | ||
3f96b79a HD |
2900 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
2901 | { | |
2902 | return 0; | |
2903 | } | |
2904 | ||
24513264 HD |
2905 | void shmem_unlock_mapping(struct address_space *mapping) |
2906 | { | |
2907 | } | |
2908 | ||
41ffe5d5 | 2909 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 2910 | { |
41ffe5d5 | 2911 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
2912 | } |
2913 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
2914 | ||
0b0a0806 HD |
2915 | #define shmem_vm_ops generic_file_vm_ops |
2916 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 2917 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
2918 | #define shmem_acct_size(flags, size) 0 |
2919 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
2920 | |
2921 | #endif /* CONFIG_SHMEM */ | |
2922 | ||
2923 | /* common code */ | |
1da177e4 | 2924 | |
3451538a | 2925 | static struct dentry_operations anon_ops = { |
118b2302 | 2926 | .d_dname = simple_dname |
3451538a AV |
2927 | }; |
2928 | ||
c7277090 EP |
2929 | static struct file *__shmem_file_setup(const char *name, loff_t size, |
2930 | unsigned long flags, unsigned int i_flags) | |
1da177e4 | 2931 | { |
6b4d0b27 | 2932 | struct file *res; |
1da177e4 | 2933 | struct inode *inode; |
2c48b9c4 | 2934 | struct path path; |
3451538a | 2935 | struct super_block *sb; |
1da177e4 LT |
2936 | struct qstr this; |
2937 | ||
2938 | if (IS_ERR(shm_mnt)) | |
6b4d0b27 | 2939 | return ERR_CAST(shm_mnt); |
1da177e4 | 2940 | |
285b2c4f | 2941 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
2942 | return ERR_PTR(-EINVAL); |
2943 | ||
2944 | if (shmem_acct_size(flags, size)) | |
2945 | return ERR_PTR(-ENOMEM); | |
2946 | ||
6b4d0b27 | 2947 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
2948 | this.name = name; |
2949 | this.len = strlen(name); | |
2950 | this.hash = 0; /* will go */ | |
3451538a | 2951 | sb = shm_mnt->mnt_sb; |
66ee4b88 | 2952 | path.mnt = mntget(shm_mnt); |
3451538a | 2953 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 2954 | if (!path.dentry) |
1da177e4 | 2955 | goto put_memory; |
3451538a | 2956 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 2957 | |
6b4d0b27 | 2958 | res = ERR_PTR(-ENOSPC); |
3451538a | 2959 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 2960 | if (!inode) |
66ee4b88 | 2961 | goto put_memory; |
1da177e4 | 2962 | |
c7277090 | 2963 | inode->i_flags |= i_flags; |
2c48b9c4 | 2964 | d_instantiate(path.dentry, inode); |
1da177e4 | 2965 | inode->i_size = size; |
6d6b77f1 | 2966 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
2967 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
2968 | if (IS_ERR(res)) | |
66ee4b88 | 2969 | goto put_path; |
4b42af81 | 2970 | |
6b4d0b27 | 2971 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 2972 | &shmem_file_operations); |
6b4d0b27 | 2973 | if (IS_ERR(res)) |
66ee4b88 | 2974 | goto put_path; |
4b42af81 | 2975 | |
6b4d0b27 | 2976 | return res; |
1da177e4 | 2977 | |
1da177e4 LT |
2978 | put_memory: |
2979 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
2980 | put_path: |
2981 | path_put(&path); | |
6b4d0b27 | 2982 | return res; |
1da177e4 | 2983 | } |
c7277090 EP |
2984 | |
2985 | /** | |
2986 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
2987 | * kernel internal. There will be NO LSM permission checks against the | |
2988 | * underlying inode. So users of this interface must do LSM checks at a | |
2989 | * higher layer. The one user is the big_key implementation. LSM checks | |
2990 | * are provided at the key level rather than the inode level. | |
2991 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
2992 | * @size: size to be set for the file | |
2993 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
2994 | */ | |
2995 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
2996 | { | |
2997 | return __shmem_file_setup(name, size, flags, S_PRIVATE); | |
2998 | } | |
2999 | ||
3000 | /** | |
3001 | * shmem_file_setup - get an unlinked file living in tmpfs | |
3002 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
3003 | * @size: size to be set for the file | |
3004 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3005 | */ | |
3006 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
3007 | { | |
3008 | return __shmem_file_setup(name, size, flags, 0); | |
3009 | } | |
395e0ddc | 3010 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 3011 | |
46711810 | 3012 | /** |
1da177e4 | 3013 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
3014 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
3015 | */ | |
3016 | int shmem_zero_setup(struct vm_area_struct *vma) | |
3017 | { | |
3018 | struct file *file; | |
3019 | loff_t size = vma->vm_end - vma->vm_start; | |
3020 | ||
3021 | file = shmem_file_setup("dev/zero", size, vma->vm_flags); | |
3022 | if (IS_ERR(file)) | |
3023 | return PTR_ERR(file); | |
3024 | ||
3025 | if (vma->vm_file) | |
3026 | fput(vma->vm_file); | |
3027 | vma->vm_file = file; | |
3028 | vma->vm_ops = &shmem_vm_ops; | |
3029 | return 0; | |
3030 | } | |
d9d90e5e HD |
3031 | |
3032 | /** | |
3033 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
3034 | * @mapping: the page's address_space | |
3035 | * @index: the page index | |
3036 | * @gfp: the page allocator flags to use if allocating | |
3037 | * | |
3038 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
3039 | * with any new page allocations done using the specified allocation flags. | |
3040 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
3041 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
3042 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
3043 | * | |
68da9f05 HD |
3044 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
3045 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
3046 | */ |
3047 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
3048 | pgoff_t index, gfp_t gfp) | |
3049 | { | |
68da9f05 HD |
3050 | #ifdef CONFIG_SHMEM |
3051 | struct inode *inode = mapping->host; | |
9276aad6 | 3052 | struct page *page; |
68da9f05 HD |
3053 | int error; |
3054 | ||
3055 | BUG_ON(mapping->a_ops != &shmem_aops); | |
3056 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL); | |
3057 | if (error) | |
3058 | page = ERR_PTR(error); | |
3059 | else | |
3060 | unlock_page(page); | |
3061 | return page; | |
3062 | #else | |
3063 | /* | |
3064 | * The tiny !SHMEM case uses ramfs without swap | |
3065 | */ | |
d9d90e5e | 3066 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 3067 | #endif |
d9d90e5e HD |
3068 | } |
3069 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |