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. | |
0edd73b3 HD |
9 | * Copyright (C) 2002-2005 Hugh Dickins. |
10 | * Copyright (C) 2002-2005 VERITAS Software Corporation. | |
1da177e4 LT |
11 | * Copyright (C) 2004 Andi Kleen, SuSE Labs |
12 | * | |
13 | * Extended attribute support for tmpfs: | |
14 | * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net> | |
15 | * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com> | |
16 | * | |
17 | * This file is released under the GPL. | |
18 | */ | |
19 | ||
20 | /* | |
21 | * This virtual memory filesystem is heavily based on the ramfs. It | |
22 | * extends ramfs by the ability to use swap and honor resource limits | |
23 | * which makes it a completely usable filesystem. | |
24 | */ | |
25 | ||
1da177e4 LT |
26 | #include <linux/module.h> |
27 | #include <linux/init.h> | |
1da177e4 | 28 | #include <linux/fs.h> |
39f0247d | 29 | #include <linux/xattr.h> |
a5694255 | 30 | #include <linux/exportfs.h> |
39f0247d | 31 | #include <linux/generic_acl.h> |
1da177e4 LT |
32 | #include <linux/mm.h> |
33 | #include <linux/mman.h> | |
34 | #include <linux/file.h> | |
35 | #include <linux/swap.h> | |
36 | #include <linux/pagemap.h> | |
37 | #include <linux/string.h> | |
38 | #include <linux/slab.h> | |
39 | #include <linux/backing-dev.h> | |
40 | #include <linux/shmem_fs.h> | |
41 | #include <linux/mount.h> | |
42 | #include <linux/writeback.h> | |
43 | #include <linux/vfs.h> | |
44 | #include <linux/blkdev.h> | |
45 | #include <linux/security.h> | |
46 | #include <linux/swapops.h> | |
47 | #include <linux/mempolicy.h> | |
48 | #include <linux/namei.h> | |
b00dc3ad | 49 | #include <linux/ctype.h> |
304dbdb7 | 50 | #include <linux/migrate.h> |
c1f60a5a | 51 | #include <linux/highmem.h> |
680d794b | 52 | #include <linux/seq_file.h> |
92562927 | 53 | #include <linux/magic.h> |
304dbdb7 | 54 | |
1da177e4 LT |
55 | #include <asm/uaccess.h> |
56 | #include <asm/div64.h> | |
57 | #include <asm/pgtable.h> | |
58 | ||
1da177e4 LT |
59 | #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long)) |
60 | #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE) | |
61 | #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) | |
62 | ||
63 | #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1)) | |
64 | #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT) | |
65 | ||
66 | #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) | |
67 | ||
68 | /* info->flags needs VM_flags to handle pagein/truncate races efficiently */ | |
69 | #define SHMEM_PAGEIN VM_READ | |
70 | #define SHMEM_TRUNCATE VM_WRITE | |
71 | ||
72 | /* Definition to limit shmem_truncate's steps between cond_rescheds */ | |
73 | #define LATENCY_LIMIT 64 | |
74 | ||
75 | /* Pretend that each entry is of this size in directory's i_size */ | |
76 | #define BOGO_DIRENT_SIZE 20 | |
77 | ||
1da177e4 LT |
78 | /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */ |
79 | enum sgp_type { | |
1da177e4 LT |
80 | SGP_READ, /* don't exceed i_size, don't allocate page */ |
81 | SGP_CACHE, /* don't exceed i_size, may allocate page */ | |
a0ee5ec5 | 82 | SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ |
1da177e4 LT |
83 | SGP_WRITE, /* may exceed i_size, may allocate page */ |
84 | }; | |
85 | ||
b76db735 | 86 | #ifdef CONFIG_TMPFS |
680d794b | 87 | static unsigned long shmem_default_max_blocks(void) |
88 | { | |
89 | return totalram_pages / 2; | |
90 | } | |
91 | ||
92 | static unsigned long shmem_default_max_inodes(void) | |
93 | { | |
94 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
95 | } | |
b76db735 | 96 | #endif |
680d794b | 97 | |
1da177e4 LT |
98 | static int shmem_getpage(struct inode *inode, unsigned long idx, |
99 | struct page **pagep, enum sgp_type sgp, int *type); | |
100 | ||
6daa0e28 | 101 | static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) |
1da177e4 LT |
102 | { |
103 | /* | |
104 | * The above definition of ENTRIES_PER_PAGE, and the use of | |
105 | * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: | |
106 | * might be reconsidered if it ever diverges from PAGE_SIZE. | |
769848c0 | 107 | * |
e12ba74d | 108 | * Mobility flags are masked out as swap vectors cannot move |
1da177e4 | 109 | */ |
e12ba74d | 110 | return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO, |
769848c0 | 111 | PAGE_CACHE_SHIFT-PAGE_SHIFT); |
1da177e4 LT |
112 | } |
113 | ||
114 | static inline void shmem_dir_free(struct page *page) | |
115 | { | |
116 | __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); | |
117 | } | |
118 | ||
119 | static struct page **shmem_dir_map(struct page *page) | |
120 | { | |
121 | return (struct page **)kmap_atomic(page, KM_USER0); | |
122 | } | |
123 | ||
124 | static inline void shmem_dir_unmap(struct page **dir) | |
125 | { | |
126 | kunmap_atomic(dir, KM_USER0); | |
127 | } | |
128 | ||
129 | static swp_entry_t *shmem_swp_map(struct page *page) | |
130 | { | |
131 | return (swp_entry_t *)kmap_atomic(page, KM_USER1); | |
132 | } | |
133 | ||
134 | static inline void shmem_swp_balance_unmap(void) | |
135 | { | |
136 | /* | |
137 | * When passing a pointer to an i_direct entry, to code which | |
138 | * also handles indirect entries and so will shmem_swp_unmap, | |
139 | * we must arrange for the preempt count to remain in balance. | |
140 | * What kmap_atomic of a lowmem page does depends on config | |
141 | * and architecture, so pretend to kmap_atomic some lowmem page. | |
142 | */ | |
143 | (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); | |
144 | } | |
145 | ||
146 | static inline void shmem_swp_unmap(swp_entry_t *entry) | |
147 | { | |
148 | kunmap_atomic(entry, KM_USER1); | |
149 | } | |
150 | ||
151 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) | |
152 | { | |
153 | return sb->s_fs_info; | |
154 | } | |
155 | ||
156 | /* | |
157 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
158 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
159 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
160 | * consistent with the pre-accounting of private mappings ... | |
161 | */ | |
162 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
163 | { | |
731572d3 AC |
164 | return (flags & VM_ACCOUNT) ? |
165 | security_vm_enough_memory_kern(VM_ACCT(size)) : 0; | |
1da177e4 LT |
166 | } |
167 | ||
168 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
169 | { | |
170 | if (flags & VM_ACCOUNT) | |
171 | vm_unacct_memory(VM_ACCT(size)); | |
172 | } | |
173 | ||
174 | /* | |
175 | * ... whereas tmpfs objects are accounted incrementally as | |
176 | * pages are allocated, in order to allow huge sparse files. | |
177 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, | |
178 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
179 | */ | |
180 | static inline int shmem_acct_block(unsigned long flags) | |
181 | { | |
731572d3 AC |
182 | return (flags & VM_ACCOUNT) ? |
183 | 0 : security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)); | |
1da177e4 LT |
184 | } |
185 | ||
186 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
187 | { | |
188 | if (!(flags & VM_ACCOUNT)) | |
189 | vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); | |
190 | } | |
191 | ||
759b9775 | 192 | static const struct super_operations shmem_ops; |
f5e54d6e | 193 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 194 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
195 | static const struct inode_operations shmem_inode_operations; |
196 | static const struct inode_operations shmem_dir_inode_operations; | |
197 | static const struct inode_operations shmem_special_inode_operations; | |
1da177e4 LT |
198 | static struct vm_operations_struct shmem_vm_ops; |
199 | ||
6c231b7b | 200 | static struct backing_dev_info shmem_backing_dev_info __read_mostly = { |
1da177e4 | 201 | .ra_pages = 0, /* No readahead */ |
4f98a2fe | 202 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED, |
1da177e4 LT |
203 | .unplug_io_fn = default_unplug_io_fn, |
204 | }; | |
205 | ||
206 | static LIST_HEAD(shmem_swaplist); | |
cb5f7b9a | 207 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 LT |
208 | |
209 | static void shmem_free_blocks(struct inode *inode, long pages) | |
210 | { | |
211 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
0edd73b3 | 212 | if (sbinfo->max_blocks) { |
1da177e4 LT |
213 | spin_lock(&sbinfo->stat_lock); |
214 | sbinfo->free_blocks += pages; | |
215 | inode->i_blocks -= pages*BLOCKS_PER_PAGE; | |
216 | spin_unlock(&sbinfo->stat_lock); | |
217 | } | |
218 | } | |
219 | ||
5b04c689 PE |
220 | static int shmem_reserve_inode(struct super_block *sb) |
221 | { | |
222 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
223 | if (sbinfo->max_inodes) { | |
224 | spin_lock(&sbinfo->stat_lock); | |
225 | if (!sbinfo->free_inodes) { | |
226 | spin_unlock(&sbinfo->stat_lock); | |
227 | return -ENOSPC; | |
228 | } | |
229 | sbinfo->free_inodes--; | |
230 | spin_unlock(&sbinfo->stat_lock); | |
231 | } | |
232 | return 0; | |
233 | } | |
234 | ||
235 | static void shmem_free_inode(struct super_block *sb) | |
236 | { | |
237 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
238 | if (sbinfo->max_inodes) { | |
239 | spin_lock(&sbinfo->stat_lock); | |
240 | sbinfo->free_inodes++; | |
241 | spin_unlock(&sbinfo->stat_lock); | |
242 | } | |
243 | } | |
244 | ||
46711810 | 245 | /** |
1da177e4 | 246 | * shmem_recalc_inode - recalculate the size of an inode |
1da177e4 LT |
247 | * @inode: inode to recalc |
248 | * | |
249 | * We have to calculate the free blocks since the mm can drop | |
250 | * undirtied hole pages behind our back. | |
251 | * | |
252 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
253 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
254 | * | |
255 | * It has to be called with the spinlock held. | |
256 | */ | |
257 | static void shmem_recalc_inode(struct inode *inode) | |
258 | { | |
259 | struct shmem_inode_info *info = SHMEM_I(inode); | |
260 | long freed; | |
261 | ||
262 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
263 | if (freed > 0) { | |
264 | info->alloced -= freed; | |
265 | shmem_unacct_blocks(info->flags, freed); | |
266 | shmem_free_blocks(inode, freed); | |
267 | } | |
268 | } | |
269 | ||
46711810 | 270 | /** |
1da177e4 | 271 | * shmem_swp_entry - find the swap vector position in the info structure |
1da177e4 LT |
272 | * @info: info structure for the inode |
273 | * @index: index of the page to find | |
274 | * @page: optional page to add to the structure. Has to be preset to | |
275 | * all zeros | |
276 | * | |
277 | * If there is no space allocated yet it will return NULL when | |
278 | * page is NULL, else it will use the page for the needed block, | |
279 | * setting it to NULL on return to indicate that it has been used. | |
280 | * | |
281 | * The swap vector is organized the following way: | |
282 | * | |
283 | * There are SHMEM_NR_DIRECT entries directly stored in the | |
284 | * shmem_inode_info structure. So small files do not need an addional | |
285 | * allocation. | |
286 | * | |
287 | * For pages with index > SHMEM_NR_DIRECT there is the pointer | |
288 | * i_indirect which points to a page which holds in the first half | |
289 | * doubly indirect blocks, in the second half triple indirect blocks: | |
290 | * | |
291 | * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the | |
292 | * following layout (for SHMEM_NR_DIRECT == 16): | |
293 | * | |
294 | * i_indirect -> dir --> 16-19 | |
295 | * | +-> 20-23 | |
296 | * | | |
297 | * +-->dir2 --> 24-27 | |
298 | * | +-> 28-31 | |
299 | * | +-> 32-35 | |
300 | * | +-> 36-39 | |
301 | * | | |
302 | * +-->dir3 --> 40-43 | |
303 | * +-> 44-47 | |
304 | * +-> 48-51 | |
305 | * +-> 52-55 | |
306 | */ | |
307 | static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) | |
308 | { | |
309 | unsigned long offset; | |
310 | struct page **dir; | |
311 | struct page *subdir; | |
312 | ||
313 | if (index < SHMEM_NR_DIRECT) { | |
314 | shmem_swp_balance_unmap(); | |
315 | return info->i_direct+index; | |
316 | } | |
317 | if (!info->i_indirect) { | |
318 | if (page) { | |
319 | info->i_indirect = *page; | |
320 | *page = NULL; | |
321 | } | |
322 | return NULL; /* need another page */ | |
323 | } | |
324 | ||
325 | index -= SHMEM_NR_DIRECT; | |
326 | offset = index % ENTRIES_PER_PAGE; | |
327 | index /= ENTRIES_PER_PAGE; | |
328 | dir = shmem_dir_map(info->i_indirect); | |
329 | ||
330 | if (index >= ENTRIES_PER_PAGE/2) { | |
331 | index -= ENTRIES_PER_PAGE/2; | |
332 | dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; | |
333 | index %= ENTRIES_PER_PAGE; | |
334 | subdir = *dir; | |
335 | if (!subdir) { | |
336 | if (page) { | |
337 | *dir = *page; | |
338 | *page = NULL; | |
339 | } | |
340 | shmem_dir_unmap(dir); | |
341 | return NULL; /* need another page */ | |
342 | } | |
343 | shmem_dir_unmap(dir); | |
344 | dir = shmem_dir_map(subdir); | |
345 | } | |
346 | ||
347 | dir += index; | |
348 | subdir = *dir; | |
349 | if (!subdir) { | |
350 | if (!page || !(subdir = *page)) { | |
351 | shmem_dir_unmap(dir); | |
352 | return NULL; /* need a page */ | |
353 | } | |
354 | *dir = subdir; | |
355 | *page = NULL; | |
356 | } | |
357 | shmem_dir_unmap(dir); | |
358 | return shmem_swp_map(subdir) + offset; | |
359 | } | |
360 | ||
361 | static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) | |
362 | { | |
363 | long incdec = value? 1: -1; | |
364 | ||
365 | entry->val = value; | |
366 | info->swapped += incdec; | |
4c21e2f2 HD |
367 | if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { |
368 | struct page *page = kmap_atomic_to_page(entry); | |
369 | set_page_private(page, page_private(page) + incdec); | |
370 | } | |
1da177e4 LT |
371 | } |
372 | ||
46711810 | 373 | /** |
1da177e4 | 374 | * shmem_swp_alloc - get the position of the swap entry for the page. |
1da177e4 LT |
375 | * @info: info structure for the inode |
376 | * @index: index of the page to find | |
377 | * @sgp: check and recheck i_size? skip allocation? | |
46711810 RD |
378 | * |
379 | * If the entry does not exist, allocate it. | |
1da177e4 LT |
380 | */ |
381 | static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) | |
382 | { | |
383 | struct inode *inode = &info->vfs_inode; | |
384 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
385 | struct page *page = NULL; | |
386 | swp_entry_t *entry; | |
387 | ||
388 | if (sgp != SGP_WRITE && | |
389 | ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) | |
390 | return ERR_PTR(-EINVAL); | |
391 | ||
392 | while (!(entry = shmem_swp_entry(info, index, &page))) { | |
393 | if (sgp == SGP_READ) | |
394 | return shmem_swp_map(ZERO_PAGE(0)); | |
395 | /* | |
396 | * Test free_blocks against 1 not 0, since we have 1 data | |
397 | * page (and perhaps indirect index pages) yet to allocate: | |
398 | * a waste to allocate index if we cannot allocate data. | |
399 | */ | |
0edd73b3 | 400 | if (sbinfo->max_blocks) { |
1da177e4 LT |
401 | spin_lock(&sbinfo->stat_lock); |
402 | if (sbinfo->free_blocks <= 1) { | |
403 | spin_unlock(&sbinfo->stat_lock); | |
404 | return ERR_PTR(-ENOSPC); | |
405 | } | |
406 | sbinfo->free_blocks--; | |
407 | inode->i_blocks += BLOCKS_PER_PAGE; | |
408 | spin_unlock(&sbinfo->stat_lock); | |
409 | } | |
410 | ||
411 | spin_unlock(&info->lock); | |
769848c0 | 412 | page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); |
4c21e2f2 HD |
413 | if (page) |
414 | set_page_private(page, 0); | |
1da177e4 LT |
415 | spin_lock(&info->lock); |
416 | ||
417 | if (!page) { | |
418 | shmem_free_blocks(inode, 1); | |
419 | return ERR_PTR(-ENOMEM); | |
420 | } | |
421 | if (sgp != SGP_WRITE && | |
422 | ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { | |
423 | entry = ERR_PTR(-EINVAL); | |
424 | break; | |
425 | } | |
426 | if (info->next_index <= index) | |
427 | info->next_index = index + 1; | |
428 | } | |
429 | if (page) { | |
430 | /* another task gave its page, or truncated the file */ | |
431 | shmem_free_blocks(inode, 1); | |
432 | shmem_dir_free(page); | |
433 | } | |
434 | if (info->next_index <= index && !IS_ERR(entry)) | |
435 | info->next_index = index + 1; | |
436 | return entry; | |
437 | } | |
438 | ||
46711810 | 439 | /** |
1da177e4 | 440 | * shmem_free_swp - free some swap entries in a directory |
1ae70006 HD |
441 | * @dir: pointer to the directory |
442 | * @edir: pointer after last entry of the directory | |
443 | * @punch_lock: pointer to spinlock when needed for the holepunch case | |
1da177e4 | 444 | */ |
1ae70006 HD |
445 | static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, |
446 | spinlock_t *punch_lock) | |
1da177e4 | 447 | { |
1ae70006 | 448 | spinlock_t *punch_unlock = NULL; |
1da177e4 LT |
449 | swp_entry_t *ptr; |
450 | int freed = 0; | |
451 | ||
452 | for (ptr = dir; ptr < edir; ptr++) { | |
453 | if (ptr->val) { | |
1ae70006 HD |
454 | if (unlikely(punch_lock)) { |
455 | punch_unlock = punch_lock; | |
456 | punch_lock = NULL; | |
457 | spin_lock(punch_unlock); | |
458 | if (!ptr->val) | |
459 | continue; | |
460 | } | |
1da177e4 LT |
461 | free_swap_and_cache(*ptr); |
462 | *ptr = (swp_entry_t){0}; | |
463 | freed++; | |
464 | } | |
465 | } | |
1ae70006 HD |
466 | if (punch_unlock) |
467 | spin_unlock(punch_unlock); | |
1da177e4 LT |
468 | return freed; |
469 | } | |
470 | ||
1ae70006 HD |
471 | static int shmem_map_and_free_swp(struct page *subdir, int offset, |
472 | int limit, struct page ***dir, spinlock_t *punch_lock) | |
1da177e4 LT |
473 | { |
474 | swp_entry_t *ptr; | |
475 | int freed = 0; | |
476 | ||
477 | ptr = shmem_swp_map(subdir); | |
478 | for (; offset < limit; offset += LATENCY_LIMIT) { | |
479 | int size = limit - offset; | |
480 | if (size > LATENCY_LIMIT) | |
481 | size = LATENCY_LIMIT; | |
1ae70006 HD |
482 | freed += shmem_free_swp(ptr+offset, ptr+offset+size, |
483 | punch_lock); | |
1da177e4 LT |
484 | if (need_resched()) { |
485 | shmem_swp_unmap(ptr); | |
486 | if (*dir) { | |
487 | shmem_dir_unmap(*dir); | |
488 | *dir = NULL; | |
489 | } | |
490 | cond_resched(); | |
491 | ptr = shmem_swp_map(subdir); | |
492 | } | |
493 | } | |
494 | shmem_swp_unmap(ptr); | |
495 | return freed; | |
496 | } | |
497 | ||
498 | static void shmem_free_pages(struct list_head *next) | |
499 | { | |
500 | struct page *page; | |
501 | int freed = 0; | |
502 | ||
503 | do { | |
504 | page = container_of(next, struct page, lru); | |
505 | next = next->next; | |
506 | shmem_dir_free(page); | |
507 | freed++; | |
508 | if (freed >= LATENCY_LIMIT) { | |
509 | cond_resched(); | |
510 | freed = 0; | |
511 | } | |
512 | } while (next); | |
513 | } | |
514 | ||
f6b3ec23 | 515 | static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) |
1da177e4 LT |
516 | { |
517 | struct shmem_inode_info *info = SHMEM_I(inode); | |
518 | unsigned long idx; | |
519 | unsigned long size; | |
520 | unsigned long limit; | |
521 | unsigned long stage; | |
522 | unsigned long diroff; | |
523 | struct page **dir; | |
524 | struct page *topdir; | |
525 | struct page *middir; | |
526 | struct page *subdir; | |
527 | swp_entry_t *ptr; | |
528 | LIST_HEAD(pages_to_free); | |
529 | long nr_pages_to_free = 0; | |
530 | long nr_swaps_freed = 0; | |
531 | int offset; | |
532 | int freed; | |
a2646d1e | 533 | int punch_hole; |
1ae70006 HD |
534 | spinlock_t *needs_lock; |
535 | spinlock_t *punch_lock; | |
a2646d1e | 536 | unsigned long upper_limit; |
1da177e4 LT |
537 | |
538 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
f6b3ec23 | 539 | idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
1da177e4 LT |
540 | if (idx >= info->next_index) |
541 | return; | |
542 | ||
543 | spin_lock(&info->lock); | |
544 | info->flags |= SHMEM_TRUNCATE; | |
f6b3ec23 BP |
545 | if (likely(end == (loff_t) -1)) { |
546 | limit = info->next_index; | |
a2646d1e | 547 | upper_limit = SHMEM_MAX_INDEX; |
f6b3ec23 | 548 | info->next_index = idx; |
1ae70006 | 549 | needs_lock = NULL; |
a2646d1e | 550 | punch_hole = 0; |
f6b3ec23 | 551 | } else { |
a2646d1e HD |
552 | if (end + 1 >= inode->i_size) { /* we may free a little more */ |
553 | limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> | |
554 | PAGE_CACHE_SHIFT; | |
555 | upper_limit = SHMEM_MAX_INDEX; | |
556 | } else { | |
557 | limit = (end + 1) >> PAGE_CACHE_SHIFT; | |
558 | upper_limit = limit; | |
559 | } | |
1ae70006 | 560 | needs_lock = &info->lock; |
f6b3ec23 BP |
561 | punch_hole = 1; |
562 | } | |
563 | ||
1da177e4 | 564 | topdir = info->i_indirect; |
f6b3ec23 | 565 | if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { |
1da177e4 LT |
566 | info->i_indirect = NULL; |
567 | nr_pages_to_free++; | |
568 | list_add(&topdir->lru, &pages_to_free); | |
569 | } | |
570 | spin_unlock(&info->lock); | |
571 | ||
572 | if (info->swapped && idx < SHMEM_NR_DIRECT) { | |
573 | ptr = info->i_direct; | |
574 | size = limit; | |
575 | if (size > SHMEM_NR_DIRECT) | |
576 | size = SHMEM_NR_DIRECT; | |
1ae70006 | 577 | nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); |
1da177e4 | 578 | } |
92a3d03a BP |
579 | |
580 | /* | |
581 | * If there are no indirect blocks or we are punching a hole | |
582 | * below indirect blocks, nothing to be done. | |
583 | */ | |
a2646d1e | 584 | if (!topdir || limit <= SHMEM_NR_DIRECT) |
1da177e4 LT |
585 | goto done2; |
586 | ||
1ae70006 HD |
587 | /* |
588 | * The truncation case has already dropped info->lock, and we're safe | |
589 | * because i_size and next_index have already been lowered, preventing | |
590 | * access beyond. But in the punch_hole case, we still need to take | |
591 | * the lock when updating the swap directory, because there might be | |
592 | * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or | |
593 | * shmem_writepage. However, whenever we find we can remove a whole | |
594 | * directory page (not at the misaligned start or end of the range), | |
595 | * we first NULLify its pointer in the level above, and then have no | |
596 | * need to take the lock when updating its contents: needs_lock and | |
597 | * punch_lock (either pointing to info->lock or NULL) manage this. | |
598 | */ | |
599 | ||
a2646d1e | 600 | upper_limit -= SHMEM_NR_DIRECT; |
1da177e4 LT |
601 | limit -= SHMEM_NR_DIRECT; |
602 | idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; | |
603 | offset = idx % ENTRIES_PER_PAGE; | |
604 | idx -= offset; | |
605 | ||
606 | dir = shmem_dir_map(topdir); | |
607 | stage = ENTRIES_PER_PAGEPAGE/2; | |
608 | if (idx < ENTRIES_PER_PAGEPAGE/2) { | |
609 | middir = topdir; | |
610 | diroff = idx/ENTRIES_PER_PAGE; | |
611 | } else { | |
612 | dir += ENTRIES_PER_PAGE/2; | |
613 | dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; | |
614 | while (stage <= idx) | |
615 | stage += ENTRIES_PER_PAGEPAGE; | |
616 | middir = *dir; | |
617 | if (*dir) { | |
618 | diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % | |
619 | ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; | |
a2646d1e | 620 | if (!diroff && !offset && upper_limit >= stage) { |
1ae70006 HD |
621 | if (needs_lock) { |
622 | spin_lock(needs_lock); | |
623 | *dir = NULL; | |
624 | spin_unlock(needs_lock); | |
625 | needs_lock = NULL; | |
626 | } else | |
627 | *dir = NULL; | |
1da177e4 LT |
628 | nr_pages_to_free++; |
629 | list_add(&middir->lru, &pages_to_free); | |
630 | } | |
631 | shmem_dir_unmap(dir); | |
632 | dir = shmem_dir_map(middir); | |
633 | } else { | |
634 | diroff = 0; | |
635 | offset = 0; | |
636 | idx = stage; | |
637 | } | |
638 | } | |
639 | ||
640 | for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { | |
641 | if (unlikely(idx == stage)) { | |
642 | shmem_dir_unmap(dir); | |
643 | dir = shmem_dir_map(topdir) + | |
644 | ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; | |
645 | while (!*dir) { | |
646 | dir++; | |
647 | idx += ENTRIES_PER_PAGEPAGE; | |
648 | if (idx >= limit) | |
649 | goto done1; | |
650 | } | |
651 | stage = idx + ENTRIES_PER_PAGEPAGE; | |
652 | middir = *dir; | |
1ae70006 HD |
653 | if (punch_hole) |
654 | needs_lock = &info->lock; | |
a2646d1e | 655 | if (upper_limit >= stage) { |
1ae70006 HD |
656 | if (needs_lock) { |
657 | spin_lock(needs_lock); | |
658 | *dir = NULL; | |
659 | spin_unlock(needs_lock); | |
660 | needs_lock = NULL; | |
661 | } else | |
662 | *dir = NULL; | |
a2646d1e HD |
663 | nr_pages_to_free++; |
664 | list_add(&middir->lru, &pages_to_free); | |
665 | } | |
1da177e4 LT |
666 | shmem_dir_unmap(dir); |
667 | cond_resched(); | |
668 | dir = shmem_dir_map(middir); | |
669 | diroff = 0; | |
670 | } | |
1ae70006 | 671 | punch_lock = needs_lock; |
1da177e4 | 672 | subdir = dir[diroff]; |
1ae70006 HD |
673 | if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { |
674 | if (needs_lock) { | |
675 | spin_lock(needs_lock); | |
676 | dir[diroff] = NULL; | |
677 | spin_unlock(needs_lock); | |
678 | punch_lock = NULL; | |
679 | } else | |
680 | dir[diroff] = NULL; | |
681 | nr_pages_to_free++; | |
682 | list_add(&subdir->lru, &pages_to_free); | |
683 | } | |
684 | if (subdir && page_private(subdir) /* has swap entries */) { | |
1da177e4 LT |
685 | size = limit - idx; |
686 | if (size > ENTRIES_PER_PAGE) | |
687 | size = ENTRIES_PER_PAGE; | |
688 | freed = shmem_map_and_free_swp(subdir, | |
1ae70006 | 689 | offset, size, &dir, punch_lock); |
1da177e4 LT |
690 | if (!dir) |
691 | dir = shmem_dir_map(middir); | |
692 | nr_swaps_freed += freed; | |
1ae70006 | 693 | if (offset || punch_lock) { |
1da177e4 | 694 | spin_lock(&info->lock); |
1ae70006 HD |
695 | set_page_private(subdir, |
696 | page_private(subdir) - freed); | |
1da177e4 | 697 | spin_unlock(&info->lock); |
1ae70006 HD |
698 | } else |
699 | BUG_ON(page_private(subdir) != freed); | |
1da177e4 | 700 | } |
1ae70006 | 701 | offset = 0; |
1da177e4 LT |
702 | } |
703 | done1: | |
704 | shmem_dir_unmap(dir); | |
705 | done2: | |
706 | if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { | |
707 | /* | |
708 | * Call truncate_inode_pages again: racing shmem_unuse_inode | |
709 | * may have swizzled a page in from swap since vmtruncate or | |
710 | * generic_delete_inode did it, before we lowered next_index. | |
711 | * Also, though shmem_getpage checks i_size before adding to | |
712 | * cache, no recheck after: so fix the narrow window there too. | |
16a10019 HD |
713 | * |
714 | * Recalling truncate_inode_pages_range and unmap_mapping_range | |
715 | * every time for punch_hole (which never got a chance to clear | |
716 | * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, | |
717 | * yet hardly ever necessary: try to optimize them out later. | |
1da177e4 | 718 | */ |
f6b3ec23 | 719 | truncate_inode_pages_range(inode->i_mapping, start, end); |
16a10019 HD |
720 | if (punch_hole) |
721 | unmap_mapping_range(inode->i_mapping, start, | |
722 | end - start, 1); | |
1da177e4 LT |
723 | } |
724 | ||
725 | spin_lock(&info->lock); | |
726 | info->flags &= ~SHMEM_TRUNCATE; | |
727 | info->swapped -= nr_swaps_freed; | |
728 | if (nr_pages_to_free) | |
729 | shmem_free_blocks(inode, nr_pages_to_free); | |
730 | shmem_recalc_inode(inode); | |
731 | spin_unlock(&info->lock); | |
732 | ||
733 | /* | |
734 | * Empty swap vector directory pages to be freed? | |
735 | */ | |
736 | if (!list_empty(&pages_to_free)) { | |
737 | pages_to_free.prev->next = NULL; | |
738 | shmem_free_pages(pages_to_free.next); | |
739 | } | |
740 | } | |
741 | ||
f6b3ec23 BP |
742 | static void shmem_truncate(struct inode *inode) |
743 | { | |
744 | shmem_truncate_range(inode, inode->i_size, (loff_t)-1); | |
745 | } | |
746 | ||
1da177e4 LT |
747 | static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) |
748 | { | |
749 | struct inode *inode = dentry->d_inode; | |
750 | struct page *page = NULL; | |
751 | int error; | |
752 | ||
39f0247d | 753 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
1da177e4 LT |
754 | if (attr->ia_size < inode->i_size) { |
755 | /* | |
756 | * If truncating down to a partial page, then | |
757 | * if that page is already allocated, hold it | |
758 | * in memory until the truncation is over, so | |
759 | * truncate_partial_page cannnot miss it were | |
760 | * it assigned to swap. | |
761 | */ | |
762 | if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { | |
763 | (void) shmem_getpage(inode, | |
764 | attr->ia_size>>PAGE_CACHE_SHIFT, | |
765 | &page, SGP_READ, NULL); | |
d3602444 HD |
766 | if (page) |
767 | unlock_page(page); | |
1da177e4 LT |
768 | } |
769 | /* | |
770 | * Reset SHMEM_PAGEIN flag so that shmem_truncate can | |
771 | * detect if any pages might have been added to cache | |
772 | * after truncate_inode_pages. But we needn't bother | |
773 | * if it's being fully truncated to zero-length: the | |
774 | * nrpages check is efficient enough in that case. | |
775 | */ | |
776 | if (attr->ia_size) { | |
777 | struct shmem_inode_info *info = SHMEM_I(inode); | |
778 | spin_lock(&info->lock); | |
779 | info->flags &= ~SHMEM_PAGEIN; | |
780 | spin_unlock(&info->lock); | |
781 | } | |
782 | } | |
783 | } | |
784 | ||
785 | error = inode_change_ok(inode, attr); | |
786 | if (!error) | |
787 | error = inode_setattr(inode, attr); | |
39f0247d AG |
788 | #ifdef CONFIG_TMPFS_POSIX_ACL |
789 | if (!error && (attr->ia_valid & ATTR_MODE)) | |
790 | error = generic_acl_chmod(inode, &shmem_acl_ops); | |
791 | #endif | |
1da177e4 LT |
792 | if (page) |
793 | page_cache_release(page); | |
794 | return error; | |
795 | } | |
796 | ||
797 | static void shmem_delete_inode(struct inode *inode) | |
798 | { | |
1da177e4 LT |
799 | struct shmem_inode_info *info = SHMEM_I(inode); |
800 | ||
801 | if (inode->i_op->truncate == shmem_truncate) { | |
fef26658 | 802 | truncate_inode_pages(inode->i_mapping, 0); |
1da177e4 LT |
803 | shmem_unacct_size(info->flags, inode->i_size); |
804 | inode->i_size = 0; | |
805 | shmem_truncate(inode); | |
806 | if (!list_empty(&info->swaplist)) { | |
cb5f7b9a | 807 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 808 | list_del_init(&info->swaplist); |
cb5f7b9a | 809 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 LT |
810 | } |
811 | } | |
0edd73b3 | 812 | BUG_ON(inode->i_blocks); |
5b04c689 | 813 | shmem_free_inode(inode->i_sb); |
1da177e4 LT |
814 | clear_inode(inode); |
815 | } | |
816 | ||
817 | static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) | |
818 | { | |
819 | swp_entry_t *ptr; | |
820 | ||
821 | for (ptr = dir; ptr < edir; ptr++) { | |
822 | if (ptr->val == entry.val) | |
823 | return ptr - dir; | |
824 | } | |
825 | return -1; | |
826 | } | |
827 | ||
828 | static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) | |
829 | { | |
830 | struct inode *inode; | |
831 | unsigned long idx; | |
832 | unsigned long size; | |
833 | unsigned long limit; | |
834 | unsigned long stage; | |
835 | struct page **dir; | |
836 | struct page *subdir; | |
837 | swp_entry_t *ptr; | |
838 | int offset; | |
d9fe526a | 839 | int error; |
1da177e4 LT |
840 | |
841 | idx = 0; | |
842 | ptr = info->i_direct; | |
843 | spin_lock(&info->lock); | |
1b1b32f2 HD |
844 | if (!info->swapped) { |
845 | list_del_init(&info->swaplist); | |
846 | goto lost2; | |
847 | } | |
1da177e4 LT |
848 | limit = info->next_index; |
849 | size = limit; | |
850 | if (size > SHMEM_NR_DIRECT) | |
851 | size = SHMEM_NR_DIRECT; | |
852 | offset = shmem_find_swp(entry, ptr, ptr+size); | |
2e0e26c7 | 853 | if (offset >= 0) |
1da177e4 | 854 | goto found; |
1da177e4 LT |
855 | if (!info->i_indirect) |
856 | goto lost2; | |
857 | ||
858 | dir = shmem_dir_map(info->i_indirect); | |
859 | stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; | |
860 | ||
861 | for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { | |
862 | if (unlikely(idx == stage)) { | |
863 | shmem_dir_unmap(dir-1); | |
cb5f7b9a HD |
864 | if (cond_resched_lock(&info->lock)) { |
865 | /* check it has not been truncated */ | |
866 | if (limit > info->next_index) { | |
867 | limit = info->next_index; | |
868 | if (idx >= limit) | |
869 | goto lost2; | |
870 | } | |
871 | } | |
1da177e4 LT |
872 | dir = shmem_dir_map(info->i_indirect) + |
873 | ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; | |
874 | while (!*dir) { | |
875 | dir++; | |
876 | idx += ENTRIES_PER_PAGEPAGE; | |
877 | if (idx >= limit) | |
878 | goto lost1; | |
879 | } | |
880 | stage = idx + ENTRIES_PER_PAGEPAGE; | |
881 | subdir = *dir; | |
882 | shmem_dir_unmap(dir); | |
883 | dir = shmem_dir_map(subdir); | |
884 | } | |
885 | subdir = *dir; | |
4c21e2f2 | 886 | if (subdir && page_private(subdir)) { |
1da177e4 LT |
887 | ptr = shmem_swp_map(subdir); |
888 | size = limit - idx; | |
889 | if (size > ENTRIES_PER_PAGE) | |
890 | size = ENTRIES_PER_PAGE; | |
891 | offset = shmem_find_swp(entry, ptr, ptr+size); | |
2e0e26c7 | 892 | shmem_swp_unmap(ptr); |
1da177e4 LT |
893 | if (offset >= 0) { |
894 | shmem_dir_unmap(dir); | |
895 | goto found; | |
896 | } | |
1da177e4 LT |
897 | } |
898 | } | |
899 | lost1: | |
900 | shmem_dir_unmap(dir-1); | |
901 | lost2: | |
902 | spin_unlock(&info->lock); | |
903 | return 0; | |
904 | found: | |
905 | idx += offset; | |
2e0e26c7 HD |
906 | inode = igrab(&info->vfs_inode); |
907 | spin_unlock(&info->lock); | |
908 | ||
1b1b32f2 HD |
909 | /* |
910 | * Move _head_ to start search for next from here. | |
911 | * But be careful: shmem_delete_inode checks list_empty without taking | |
912 | * mutex, and there's an instant in list_move_tail when info->swaplist | |
913 | * would appear empty, if it were the only one on shmem_swaplist. We | |
914 | * could avoid doing it if inode NULL; or use this minor optimization. | |
915 | */ | |
916 | if (shmem_swaplist.next != &info->swaplist) | |
917 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 HD |
918 | mutex_unlock(&shmem_swaplist_mutex); |
919 | ||
920 | error = 1; | |
921 | if (!inode) | |
922 | goto out; | |
69029cd5 | 923 | /* Precharge page using GFP_KERNEL while we can wait */ |
82369553 | 924 | error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); |
b409f9fc HD |
925 | if (error) |
926 | goto out; | |
82369553 | 927 | error = radix_tree_preload(GFP_KERNEL); |
69029cd5 KH |
928 | if (error) { |
929 | mem_cgroup_uncharge_cache_page(page); | |
930 | goto out; | |
931 | } | |
b409f9fc | 932 | error = 1; |
2e0e26c7 HD |
933 | |
934 | spin_lock(&info->lock); | |
935 | ptr = shmem_swp_entry(info, idx, NULL); | |
69029cd5 | 936 | if (ptr && ptr->val == entry.val) { |
e286781d | 937 | error = add_to_page_cache_locked(page, inode->i_mapping, |
b409f9fc | 938 | idx, GFP_NOWAIT); |
69029cd5 KH |
939 | /* does mem_cgroup_uncharge_cache_page on error */ |
940 | } else /* we must compensate for our precharge above */ | |
941 | mem_cgroup_uncharge_cache_page(page); | |
942 | ||
d9fe526a HD |
943 | if (error == -EEXIST) { |
944 | struct page *filepage = find_get_page(inode->i_mapping, idx); | |
2e0e26c7 | 945 | error = 1; |
d9fe526a HD |
946 | if (filepage) { |
947 | /* | |
948 | * There might be a more uptodate page coming down | |
949 | * from a stacked writepage: forget our swappage if so. | |
950 | */ | |
951 | if (PageUptodate(filepage)) | |
952 | error = 0; | |
953 | page_cache_release(filepage); | |
954 | } | |
955 | } | |
956 | if (!error) { | |
73b1262f HD |
957 | delete_from_swap_cache(page); |
958 | set_page_dirty(page); | |
1da177e4 | 959 | info->flags |= SHMEM_PAGEIN; |
2e0e26c7 HD |
960 | shmem_swp_set(info, ptr, 0); |
961 | swap_free(entry); | |
962 | error = 1; /* not an error, but entry was found */ | |
1da177e4 | 963 | } |
2e0e26c7 HD |
964 | if (ptr) |
965 | shmem_swp_unmap(ptr); | |
1da177e4 | 966 | spin_unlock(&info->lock); |
b409f9fc | 967 | radix_tree_preload_end(); |
2e0e26c7 HD |
968 | out: |
969 | unlock_page(page); | |
970 | page_cache_release(page); | |
971 | iput(inode); /* allows for NULL */ | |
972 | return error; | |
1da177e4 LT |
973 | } |
974 | ||
975 | /* | |
976 | * shmem_unuse() search for an eventually swapped out shmem page. | |
977 | */ | |
978 | int shmem_unuse(swp_entry_t entry, struct page *page) | |
979 | { | |
980 | struct list_head *p, *next; | |
981 | struct shmem_inode_info *info; | |
982 | int found = 0; | |
983 | ||
cb5f7b9a | 984 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 LT |
985 | list_for_each_safe(p, next, &shmem_swaplist) { |
986 | info = list_entry(p, struct shmem_inode_info, swaplist); | |
1b1b32f2 | 987 | found = shmem_unuse_inode(info, entry, page); |
cb5f7b9a | 988 | cond_resched(); |
2e0e26c7 HD |
989 | if (found) |
990 | goto out; | |
1da177e4 | 991 | } |
cb5f7b9a | 992 | mutex_unlock(&shmem_swaplist_mutex); |
2e0e26c7 | 993 | out: return found; /* 0 or 1 or -ENOMEM */ |
1da177e4 LT |
994 | } |
995 | ||
996 | /* | |
997 | * Move the page from the page cache to the swap cache. | |
998 | */ | |
999 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1000 | { | |
1001 | struct shmem_inode_info *info; | |
1002 | swp_entry_t *entry, swap; | |
1003 | struct address_space *mapping; | |
1004 | unsigned long index; | |
1005 | struct inode *inode; | |
1006 | ||
1007 | BUG_ON(!PageLocked(page)); | |
1da177e4 LT |
1008 | mapping = page->mapping; |
1009 | index = page->index; | |
1010 | inode = mapping->host; | |
1011 | info = SHMEM_I(inode); | |
1012 | if (info->flags & VM_LOCKED) | |
1013 | goto redirty; | |
d9fe526a | 1014 | if (!total_swap_pages) |
1da177e4 LT |
1015 | goto redirty; |
1016 | ||
d9fe526a HD |
1017 | /* |
1018 | * shmem_backing_dev_info's capabilities prevent regular writeback or | |
1019 | * sync from ever calling shmem_writepage; but a stacking filesystem | |
1020 | * may use the ->writepage of its underlying filesystem, in which case | |
1021 | * tmpfs should write out to swap only in response to memory pressure, | |
1022 | * and not for pdflush or sync. However, in those cases, we do still | |
1023 | * want to check if there's a redundant swappage to be discarded. | |
1024 | */ | |
1025 | if (wbc->for_reclaim) | |
1026 | swap = get_swap_page(); | |
1027 | else | |
1028 | swap.val = 0; | |
1029 | ||
1da177e4 | 1030 | spin_lock(&info->lock); |
1da177e4 LT |
1031 | if (index >= info->next_index) { |
1032 | BUG_ON(!(info->flags & SHMEM_TRUNCATE)); | |
1033 | goto unlock; | |
1034 | } | |
1035 | entry = shmem_swp_entry(info, index, NULL); | |
d9fe526a HD |
1036 | if (entry->val) { |
1037 | /* | |
1038 | * The more uptodate page coming down from a stacked | |
1039 | * writepage should replace our old swappage. | |
1040 | */ | |
1041 | free_swap_and_cache(*entry); | |
1042 | shmem_swp_set(info, entry, 0); | |
1043 | } | |
1044 | shmem_recalc_inode(inode); | |
1da177e4 | 1045 | |
d9fe526a | 1046 | if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
73b1262f | 1047 | remove_from_page_cache(page); |
1da177e4 LT |
1048 | shmem_swp_set(info, entry, swap.val); |
1049 | shmem_swp_unmap(entry); | |
1b1b32f2 HD |
1050 | if (list_empty(&info->swaplist)) |
1051 | inode = igrab(inode); | |
1052 | else | |
1053 | inode = NULL; | |
1da177e4 | 1054 | spin_unlock(&info->lock); |
73b1262f | 1055 | swap_duplicate(swap); |
d9fe526a | 1056 | BUG_ON(page_mapped(page)); |
73b1262f HD |
1057 | page_cache_release(page); /* pagecache ref */ |
1058 | set_page_dirty(page); | |
1da177e4 | 1059 | unlock_page(page); |
1b1b32f2 HD |
1060 | if (inode) { |
1061 | mutex_lock(&shmem_swaplist_mutex); | |
1062 | /* move instead of add in case we're racing */ | |
1063 | list_move_tail(&info->swaplist, &shmem_swaplist); | |
1064 | mutex_unlock(&shmem_swaplist_mutex); | |
1065 | iput(inode); | |
1066 | } | |
1da177e4 LT |
1067 | return 0; |
1068 | } | |
1069 | ||
1070 | shmem_swp_unmap(entry); | |
1071 | unlock: | |
1072 | spin_unlock(&info->lock); | |
1073 | swap_free(swap); | |
1074 | redirty: | |
1075 | set_page_dirty(page); | |
d9fe526a HD |
1076 | if (wbc->for_reclaim) |
1077 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1078 | unlock_page(page); | |
1079 | return 0; | |
1da177e4 LT |
1080 | } |
1081 | ||
1082 | #ifdef CONFIG_NUMA | |
680d794b | 1083 | #ifdef CONFIG_TMPFS |
71fe804b | 1084 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1085 | { |
095f1fc4 | 1086 | char buffer[64]; |
680d794b | 1087 | |
71fe804b | 1088 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1089 | return; /* show nothing */ |
680d794b | 1090 | |
71fe804b | 1091 | mpol_to_str(buffer, sizeof(buffer), mpol, 1); |
095f1fc4 LS |
1092 | |
1093 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1094 | } |
71fe804b LS |
1095 | |
1096 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1097 | { | |
1098 | struct mempolicy *mpol = NULL; | |
1099 | if (sbinfo->mpol) { | |
1100 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1101 | mpol = sbinfo->mpol; | |
1102 | mpol_get(mpol); | |
1103 | spin_unlock(&sbinfo->stat_lock); | |
1104 | } | |
1105 | return mpol; | |
1106 | } | |
680d794b | 1107 | #endif /* CONFIG_TMPFS */ |
1108 | ||
02098fea HD |
1109 | static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, |
1110 | struct shmem_inode_info *info, unsigned long idx) | |
1da177e4 | 1111 | { |
52cd3b07 | 1112 | struct mempolicy mpol, *spol; |
1da177e4 | 1113 | struct vm_area_struct pvma; |
c4cc6d07 | 1114 | struct page *page; |
1da177e4 | 1115 | |
52cd3b07 LS |
1116 | spol = mpol_cond_copy(&mpol, |
1117 | mpol_shared_policy_lookup(&info->policy, idx)); | |
1118 | ||
1da177e4 | 1119 | /* Create a pseudo vma that just contains the policy */ |
c4cc6d07 | 1120 | pvma.vm_start = 0; |
1da177e4 | 1121 | pvma.vm_pgoff = idx; |
c4cc6d07 | 1122 | pvma.vm_ops = NULL; |
52cd3b07 | 1123 | pvma.vm_policy = spol; |
02098fea | 1124 | page = swapin_readahead(entry, gfp, &pvma, 0); |
1da177e4 LT |
1125 | return page; |
1126 | } | |
1127 | ||
02098fea HD |
1128 | static struct page *shmem_alloc_page(gfp_t gfp, |
1129 | struct shmem_inode_info *info, unsigned long idx) | |
1da177e4 LT |
1130 | { |
1131 | struct vm_area_struct pvma; | |
1da177e4 | 1132 | |
c4cc6d07 HD |
1133 | /* Create a pseudo vma that just contains the policy */ |
1134 | pvma.vm_start = 0; | |
1da177e4 | 1135 | pvma.vm_pgoff = idx; |
c4cc6d07 HD |
1136 | pvma.vm_ops = NULL; |
1137 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); | |
52cd3b07 LS |
1138 | |
1139 | /* | |
1140 | * alloc_page_vma() will drop the shared policy reference | |
1141 | */ | |
1142 | return alloc_page_vma(gfp, &pvma, 0); | |
1da177e4 | 1143 | } |
680d794b | 1144 | #else /* !CONFIG_NUMA */ |
1145 | #ifdef CONFIG_TMPFS | |
71fe804b | 1146 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p) |
680d794b | 1147 | { |
1148 | } | |
1149 | #endif /* CONFIG_TMPFS */ | |
1150 | ||
02098fea HD |
1151 | static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, |
1152 | struct shmem_inode_info *info, unsigned long idx) | |
1da177e4 | 1153 | { |
02098fea | 1154 | return swapin_readahead(entry, gfp, NULL, 0); |
1da177e4 LT |
1155 | } |
1156 | ||
02098fea HD |
1157 | static inline struct page *shmem_alloc_page(gfp_t gfp, |
1158 | struct shmem_inode_info *info, unsigned long idx) | |
1da177e4 | 1159 | { |
e84e2e13 | 1160 | return alloc_page(gfp); |
1da177e4 | 1161 | } |
680d794b | 1162 | #endif /* CONFIG_NUMA */ |
1da177e4 | 1163 | |
71fe804b LS |
1164 | #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) |
1165 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1166 | { | |
1167 | return NULL; | |
1168 | } | |
1169 | #endif | |
1170 | ||
1da177e4 LT |
1171 | /* |
1172 | * shmem_getpage - either get the page from swap or allocate a new one | |
1173 | * | |
1174 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1175 | * vm. If we swap it in we mark it dirty since we also free the swap | |
1176 | * entry since a page cannot live in both the swap and page cache | |
1177 | */ | |
1178 | static int shmem_getpage(struct inode *inode, unsigned long idx, | |
1179 | struct page **pagep, enum sgp_type sgp, int *type) | |
1180 | { | |
1181 | struct address_space *mapping = inode->i_mapping; | |
1182 | struct shmem_inode_info *info = SHMEM_I(inode); | |
1183 | struct shmem_sb_info *sbinfo; | |
1184 | struct page *filepage = *pagep; | |
1185 | struct page *swappage; | |
1186 | swp_entry_t *entry; | |
1187 | swp_entry_t swap; | |
02098fea | 1188 | gfp_t gfp; |
1da177e4 LT |
1189 | int error; |
1190 | ||
1191 | if (idx >= SHMEM_MAX_INDEX) | |
1192 | return -EFBIG; | |
54cb8821 NP |
1193 | |
1194 | if (type) | |
83c54070 | 1195 | *type = 0; |
54cb8821 | 1196 | |
1da177e4 LT |
1197 | /* |
1198 | * Normally, filepage is NULL on entry, and either found | |
1199 | * uptodate immediately, or allocated and zeroed, or read | |
1200 | * in under swappage, which is then assigned to filepage. | |
5402b976 | 1201 | * But shmem_readpage (required for splice) passes in a locked |
ae976416 HD |
1202 | * filepage, which may be found not uptodate by other callers |
1203 | * too, and may need to be copied from the swappage read in. | |
1da177e4 LT |
1204 | */ |
1205 | repeat: | |
1206 | if (!filepage) | |
1207 | filepage = find_lock_page(mapping, idx); | |
1208 | if (filepage && PageUptodate(filepage)) | |
1209 | goto done; | |
1210 | error = 0; | |
02098fea | 1211 | gfp = mapping_gfp_mask(mapping); |
b409f9fc HD |
1212 | if (!filepage) { |
1213 | /* | |
1214 | * Try to preload while we can wait, to not make a habit of | |
1215 | * draining atomic reserves; but don't latch on to this cpu. | |
1216 | */ | |
1217 | error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); | |
1218 | if (error) | |
1219 | goto failed; | |
1220 | radix_tree_preload_end(); | |
1221 | } | |
1da177e4 LT |
1222 | |
1223 | spin_lock(&info->lock); | |
1224 | shmem_recalc_inode(inode); | |
1225 | entry = shmem_swp_alloc(info, idx, sgp); | |
1226 | if (IS_ERR(entry)) { | |
1227 | spin_unlock(&info->lock); | |
1228 | error = PTR_ERR(entry); | |
1229 | goto failed; | |
1230 | } | |
1231 | swap = *entry; | |
1232 | ||
1233 | if (swap.val) { | |
1234 | /* Look it up and read it in.. */ | |
1235 | swappage = lookup_swap_cache(swap); | |
1236 | if (!swappage) { | |
1237 | shmem_swp_unmap(entry); | |
1da177e4 | 1238 | /* here we actually do the io */ |
83c54070 | 1239 | if (type && !(*type & VM_FAULT_MAJOR)) { |
f8891e5e | 1240 | __count_vm_event(PGMAJFAULT); |
83c54070 | 1241 | *type |= VM_FAULT_MAJOR; |
1da177e4 | 1242 | } |
f8891e5e | 1243 | spin_unlock(&info->lock); |
02098fea | 1244 | swappage = shmem_swapin(swap, gfp, info, idx); |
1da177e4 LT |
1245 | if (!swappage) { |
1246 | spin_lock(&info->lock); | |
1247 | entry = shmem_swp_alloc(info, idx, sgp); | |
1248 | if (IS_ERR(entry)) | |
1249 | error = PTR_ERR(entry); | |
1250 | else { | |
1251 | if (entry->val == swap.val) | |
1252 | error = -ENOMEM; | |
1253 | shmem_swp_unmap(entry); | |
1254 | } | |
1255 | spin_unlock(&info->lock); | |
1256 | if (error) | |
1257 | goto failed; | |
1258 | goto repeat; | |
1259 | } | |
1260 | wait_on_page_locked(swappage); | |
1261 | page_cache_release(swappage); | |
1262 | goto repeat; | |
1263 | } | |
1264 | ||
1265 | /* We have to do this with page locked to prevent races */ | |
529ae9aa | 1266 | if (!trylock_page(swappage)) { |
1da177e4 LT |
1267 | shmem_swp_unmap(entry); |
1268 | spin_unlock(&info->lock); | |
1269 | wait_on_page_locked(swappage); | |
1270 | page_cache_release(swappage); | |
1271 | goto repeat; | |
1272 | } | |
1273 | if (PageWriteback(swappage)) { | |
1274 | shmem_swp_unmap(entry); | |
1275 | spin_unlock(&info->lock); | |
1276 | wait_on_page_writeback(swappage); | |
1277 | unlock_page(swappage); | |
1278 | page_cache_release(swappage); | |
1279 | goto repeat; | |
1280 | } | |
1281 | if (!PageUptodate(swappage)) { | |
1282 | shmem_swp_unmap(entry); | |
1283 | spin_unlock(&info->lock); | |
1284 | unlock_page(swappage); | |
1285 | page_cache_release(swappage); | |
1286 | error = -EIO; | |
1287 | goto failed; | |
1288 | } | |
1289 | ||
1290 | if (filepage) { | |
1291 | shmem_swp_set(info, entry, 0); | |
1292 | shmem_swp_unmap(entry); | |
1293 | delete_from_swap_cache(swappage); | |
1294 | spin_unlock(&info->lock); | |
1295 | copy_highpage(filepage, swappage); | |
1296 | unlock_page(swappage); | |
1297 | page_cache_release(swappage); | |
1298 | flush_dcache_page(filepage); | |
1299 | SetPageUptodate(filepage); | |
1300 | set_page_dirty(filepage); | |
1301 | swap_free(swap); | |
e286781d NP |
1302 | } else if (!(error = add_to_page_cache_locked(swappage, mapping, |
1303 | idx, GFP_NOWAIT))) { | |
1da177e4 LT |
1304 | info->flags |= SHMEM_PAGEIN; |
1305 | shmem_swp_set(info, entry, 0); | |
1306 | shmem_swp_unmap(entry); | |
73b1262f | 1307 | delete_from_swap_cache(swappage); |
1da177e4 LT |
1308 | spin_unlock(&info->lock); |
1309 | filepage = swappage; | |
73b1262f | 1310 | set_page_dirty(filepage); |
1da177e4 LT |
1311 | swap_free(swap); |
1312 | } else { | |
1313 | shmem_swp_unmap(entry); | |
1314 | spin_unlock(&info->lock); | |
1315 | unlock_page(swappage); | |
c9b0ed51 | 1316 | page_cache_release(swappage); |
82369553 HD |
1317 | if (error == -ENOMEM) { |
1318 | /* allow reclaim from this memory cgroup */ | |
c9b0ed51 KH |
1319 | error = mem_cgroup_shrink_usage(current->mm, |
1320 | gfp); | |
1321 | if (error) | |
82369553 HD |
1322 | goto failed; |
1323 | } | |
1da177e4 LT |
1324 | goto repeat; |
1325 | } | |
1326 | } else if (sgp == SGP_READ && !filepage) { | |
1327 | shmem_swp_unmap(entry); | |
1328 | filepage = find_get_page(mapping, idx); | |
1329 | if (filepage && | |
529ae9aa | 1330 | (!PageUptodate(filepage) || !trylock_page(filepage))) { |
1da177e4 LT |
1331 | spin_unlock(&info->lock); |
1332 | wait_on_page_locked(filepage); | |
1333 | page_cache_release(filepage); | |
1334 | filepage = NULL; | |
1335 | goto repeat; | |
1336 | } | |
1337 | spin_unlock(&info->lock); | |
1338 | } else { | |
1339 | shmem_swp_unmap(entry); | |
1340 | sbinfo = SHMEM_SB(inode->i_sb); | |
0edd73b3 | 1341 | if (sbinfo->max_blocks) { |
1da177e4 LT |
1342 | spin_lock(&sbinfo->stat_lock); |
1343 | if (sbinfo->free_blocks == 0 || | |
1344 | shmem_acct_block(info->flags)) { | |
1345 | spin_unlock(&sbinfo->stat_lock); | |
1346 | spin_unlock(&info->lock); | |
1347 | error = -ENOSPC; | |
1348 | goto failed; | |
1349 | } | |
1350 | sbinfo->free_blocks--; | |
1351 | inode->i_blocks += BLOCKS_PER_PAGE; | |
1352 | spin_unlock(&sbinfo->stat_lock); | |
1353 | } else if (shmem_acct_block(info->flags)) { | |
1354 | spin_unlock(&info->lock); | |
1355 | error = -ENOSPC; | |
1356 | goto failed; | |
1357 | } | |
1358 | ||
1359 | if (!filepage) { | |
69029cd5 KH |
1360 | int ret; |
1361 | ||
1da177e4 | 1362 | spin_unlock(&info->lock); |
02098fea | 1363 | filepage = shmem_alloc_page(gfp, info, idx); |
1da177e4 LT |
1364 | if (!filepage) { |
1365 | shmem_unacct_blocks(info->flags, 1); | |
1366 | shmem_free_blocks(inode, 1); | |
1367 | error = -ENOMEM; | |
1368 | goto failed; | |
1369 | } | |
b2e18538 | 1370 | SetPageSwapBacked(filepage); |
1da177e4 | 1371 | |
82369553 HD |
1372 | /* Precharge page while we can wait, compensate after */ |
1373 | error = mem_cgroup_cache_charge(filepage, current->mm, | |
1374 | gfp & ~__GFP_HIGHMEM); | |
1375 | if (error) { | |
1376 | page_cache_release(filepage); | |
1377 | shmem_unacct_blocks(info->flags, 1); | |
1378 | shmem_free_blocks(inode, 1); | |
1379 | filepage = NULL; | |
1380 | goto failed; | |
1381 | } | |
1382 | ||
1da177e4 LT |
1383 | spin_lock(&info->lock); |
1384 | entry = shmem_swp_alloc(info, idx, sgp); | |
1385 | if (IS_ERR(entry)) | |
1386 | error = PTR_ERR(entry); | |
1387 | else { | |
1388 | swap = *entry; | |
1389 | shmem_swp_unmap(entry); | |
1390 | } | |
69029cd5 KH |
1391 | ret = error || swap.val; |
1392 | if (ret) | |
1393 | mem_cgroup_uncharge_cache_page(filepage); | |
1394 | else | |
1395 | ret = add_to_page_cache_lru(filepage, mapping, | |
1396 | idx, GFP_NOWAIT); | |
1397 | /* | |
1398 | * At add_to_page_cache_lru() failure, uncharge will | |
1399 | * be done automatically. | |
1400 | */ | |
1401 | if (ret) { | |
1da177e4 LT |
1402 | spin_unlock(&info->lock); |
1403 | page_cache_release(filepage); | |
1404 | shmem_unacct_blocks(info->flags, 1); | |
1405 | shmem_free_blocks(inode, 1); | |
1406 | filepage = NULL; | |
1407 | if (error) | |
1408 | goto failed; | |
1409 | goto repeat; | |
1410 | } | |
1411 | info->flags |= SHMEM_PAGEIN; | |
1412 | } | |
1413 | ||
1414 | info->alloced++; | |
1415 | spin_unlock(&info->lock); | |
e84e2e13 | 1416 | clear_highpage(filepage); |
1da177e4 LT |
1417 | flush_dcache_page(filepage); |
1418 | SetPageUptodate(filepage); | |
a0ee5ec5 HD |
1419 | if (sgp == SGP_DIRTY) |
1420 | set_page_dirty(filepage); | |
1da177e4 LT |
1421 | } |
1422 | done: | |
d3602444 | 1423 | *pagep = filepage; |
1da177e4 LT |
1424 | return 0; |
1425 | ||
1426 | failed: | |
1427 | if (*pagep != filepage) { | |
1428 | unlock_page(filepage); | |
1429 | page_cache_release(filepage); | |
1430 | } | |
1431 | return error; | |
1432 | } | |
1433 | ||
d0217ac0 | 1434 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1435 | { |
d3ac7f89 | 1436 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 | 1437 | int error; |
d0217ac0 | 1438 | int ret; |
1da177e4 | 1439 | |
d0217ac0 NP |
1440 | if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) |
1441 | return VM_FAULT_SIGBUS; | |
d00806b1 | 1442 | |
27d54b39 | 1443 | error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); |
d0217ac0 NP |
1444 | if (error) |
1445 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
1da177e4 | 1446 | |
83c54070 | 1447 | return ret | VM_FAULT_LOCKED; |
1da177e4 LT |
1448 | } |
1449 | ||
1da177e4 | 1450 | #ifdef CONFIG_NUMA |
d8dc74f2 | 1451 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) |
1da177e4 | 1452 | { |
d3ac7f89 | 1453 | struct inode *i = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 LT |
1454 | return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); |
1455 | } | |
1456 | ||
d8dc74f2 AB |
1457 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
1458 | unsigned long addr) | |
1da177e4 | 1459 | { |
d3ac7f89 | 1460 | struct inode *i = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 LT |
1461 | unsigned long idx; |
1462 | ||
1463 | idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
1464 | return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); | |
1465 | } | |
1466 | #endif | |
1467 | ||
1468 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
1469 | { | |
d3ac7f89 | 1470 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
1471 | struct shmem_inode_info *info = SHMEM_I(inode); |
1472 | int retval = -ENOMEM; | |
1473 | ||
1474 | spin_lock(&info->lock); | |
1475 | if (lock && !(info->flags & VM_LOCKED)) { | |
1476 | if (!user_shm_lock(inode->i_size, user)) | |
1477 | goto out_nomem; | |
1478 | info->flags |= VM_LOCKED; | |
89e004ea | 1479 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
1480 | } |
1481 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
1482 | user_shm_unlock(inode->i_size, user); | |
1483 | info->flags &= ~VM_LOCKED; | |
89e004ea LS |
1484 | mapping_clear_unevictable(file->f_mapping); |
1485 | scan_mapping_unevictable_pages(file->f_mapping); | |
1da177e4 LT |
1486 | } |
1487 | retval = 0; | |
89e004ea | 1488 | |
1da177e4 LT |
1489 | out_nomem: |
1490 | spin_unlock(&info->lock); | |
1491 | return retval; | |
1492 | } | |
1493 | ||
9b83a6a8 | 1494 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
1495 | { |
1496 | file_accessed(file); | |
1497 | vma->vm_ops = &shmem_vm_ops; | |
d0217ac0 | 1498 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
1499 | return 0; |
1500 | } | |
1501 | ||
1502 | static struct inode * | |
1503 | shmem_get_inode(struct super_block *sb, int mode, dev_t dev) | |
1504 | { | |
1505 | struct inode *inode; | |
1506 | struct shmem_inode_info *info; | |
1507 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
1508 | ||
5b04c689 PE |
1509 | if (shmem_reserve_inode(sb)) |
1510 | return NULL; | |
1da177e4 LT |
1511 | |
1512 | inode = new_inode(sb); | |
1513 | if (inode) { | |
1514 | inode->i_mode = mode; | |
76aac0e9 DH |
1515 | inode->i_uid = current_fsuid(); |
1516 | inode->i_gid = current_fsgid(); | |
1da177e4 | 1517 | inode->i_blocks = 0; |
1da177e4 LT |
1518 | inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; |
1519 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
91828a40 | 1520 | inode->i_generation = get_seconds(); |
1da177e4 LT |
1521 | info = SHMEM_I(inode); |
1522 | memset(info, 0, (char *)inode - (char *)info); | |
1523 | spin_lock_init(&info->lock); | |
1524 | INIT_LIST_HEAD(&info->swaplist); | |
1525 | ||
1526 | switch (mode & S_IFMT) { | |
1527 | default: | |
39f0247d | 1528 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
1529 | init_special_inode(inode, mode, dev); |
1530 | break; | |
1531 | case S_IFREG: | |
14fcc23f | 1532 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
1533 | inode->i_op = &shmem_inode_operations; |
1534 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
1535 | mpol_shared_policy_init(&info->policy, |
1536 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
1537 | break; |
1538 | case S_IFDIR: | |
d8c76e6f | 1539 | inc_nlink(inode); |
1da177e4 LT |
1540 | /* Some things misbehave if size == 0 on a directory */ |
1541 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
1542 | inode->i_op = &shmem_dir_inode_operations; | |
1543 | inode->i_fop = &simple_dir_operations; | |
1544 | break; | |
1545 | case S_IFLNK: | |
1546 | /* | |
1547 | * Must not load anything in the rbtree, | |
1548 | * mpol_free_shared_policy will not be called. | |
1549 | */ | |
71fe804b | 1550 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
1551 | break; |
1552 | } | |
5b04c689 PE |
1553 | } else |
1554 | shmem_free_inode(sb); | |
1da177e4 LT |
1555 | return inode; |
1556 | } | |
1557 | ||
1558 | #ifdef CONFIG_TMPFS | |
92e1d5be AV |
1559 | static const struct inode_operations shmem_symlink_inode_operations; |
1560 | static const struct inode_operations shmem_symlink_inline_operations; | |
1da177e4 LT |
1561 | |
1562 | /* | |
800d15a5 | 1563 | * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; |
ae976416 HD |
1564 | * but providing them allows a tmpfs file to be used for splice, sendfile, and |
1565 | * below the loop driver, in the generic fashion that many filesystems support. | |
1da177e4 | 1566 | */ |
ae976416 HD |
1567 | static int shmem_readpage(struct file *file, struct page *page) |
1568 | { | |
1569 | struct inode *inode = page->mapping->host; | |
1570 | int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); | |
1571 | unlock_page(page); | |
1572 | return error; | |
1573 | } | |
1574 | ||
1da177e4 | 1575 | static int |
800d15a5 NP |
1576 | shmem_write_begin(struct file *file, struct address_space *mapping, |
1577 | loff_t pos, unsigned len, unsigned flags, | |
1578 | struct page **pagep, void **fsdata) | |
1da177e4 | 1579 | { |
800d15a5 NP |
1580 | struct inode *inode = mapping->host; |
1581 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
1582 | *pagep = NULL; | |
1583 | return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); | |
1584 | } | |
1585 | ||
1586 | static int | |
1587 | shmem_write_end(struct file *file, struct address_space *mapping, | |
1588 | loff_t pos, unsigned len, unsigned copied, | |
1589 | struct page *page, void *fsdata) | |
1590 | { | |
1591 | struct inode *inode = mapping->host; | |
1592 | ||
d3602444 HD |
1593 | if (pos + copied > inode->i_size) |
1594 | i_size_write(inode, pos + copied); | |
1595 | ||
1596 | unlock_page(page); | |
800d15a5 NP |
1597 | set_page_dirty(page); |
1598 | page_cache_release(page); | |
1599 | ||
800d15a5 | 1600 | return copied; |
1da177e4 LT |
1601 | } |
1602 | ||
1da177e4 LT |
1603 | static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) |
1604 | { | |
d3ac7f89 | 1605 | struct inode *inode = filp->f_path.dentry->d_inode; |
1da177e4 LT |
1606 | struct address_space *mapping = inode->i_mapping; |
1607 | unsigned long index, offset; | |
a0ee5ec5 HD |
1608 | enum sgp_type sgp = SGP_READ; |
1609 | ||
1610 | /* | |
1611 | * Might this read be for a stacking filesystem? Then when reading | |
1612 | * holes of a sparse file, we actually need to allocate those pages, | |
1613 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
1614 | */ | |
1615 | if (segment_eq(get_fs(), KERNEL_DS)) | |
1616 | sgp = SGP_DIRTY; | |
1da177e4 LT |
1617 | |
1618 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1619 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1620 | ||
1621 | for (;;) { | |
1622 | struct page *page = NULL; | |
1623 | unsigned long end_index, nr, ret; | |
1624 | loff_t i_size = i_size_read(inode); | |
1625 | ||
1626 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1627 | if (index > end_index) | |
1628 | break; | |
1629 | if (index == end_index) { | |
1630 | nr = i_size & ~PAGE_CACHE_MASK; | |
1631 | if (nr <= offset) | |
1632 | break; | |
1633 | } | |
1634 | ||
a0ee5ec5 | 1635 | desc->error = shmem_getpage(inode, index, &page, sgp, NULL); |
1da177e4 LT |
1636 | if (desc->error) { |
1637 | if (desc->error == -EINVAL) | |
1638 | desc->error = 0; | |
1639 | break; | |
1640 | } | |
d3602444 HD |
1641 | if (page) |
1642 | unlock_page(page); | |
1da177e4 LT |
1643 | |
1644 | /* | |
1645 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 1646 | * are called without i_mutex protection against truncate |
1da177e4 LT |
1647 | */ |
1648 | nr = PAGE_CACHE_SIZE; | |
1649 | i_size = i_size_read(inode); | |
1650 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1651 | if (index == end_index) { | |
1652 | nr = i_size & ~PAGE_CACHE_MASK; | |
1653 | if (nr <= offset) { | |
1654 | if (page) | |
1655 | page_cache_release(page); | |
1656 | break; | |
1657 | } | |
1658 | } | |
1659 | nr -= offset; | |
1660 | ||
1661 | if (page) { | |
1662 | /* | |
1663 | * If users can be writing to this page using arbitrary | |
1664 | * virtual addresses, take care about potential aliasing | |
1665 | * before reading the page on the kernel side. | |
1666 | */ | |
1667 | if (mapping_writably_mapped(mapping)) | |
1668 | flush_dcache_page(page); | |
1669 | /* | |
1670 | * Mark the page accessed if we read the beginning. | |
1671 | */ | |
1672 | if (!offset) | |
1673 | mark_page_accessed(page); | |
b5810039 | 1674 | } else { |
1da177e4 | 1675 | page = ZERO_PAGE(0); |
b5810039 NP |
1676 | page_cache_get(page); |
1677 | } | |
1da177e4 LT |
1678 | |
1679 | /* | |
1680 | * Ok, we have the page, and it's up-to-date, so | |
1681 | * now we can copy it to user space... | |
1682 | * | |
1683 | * The actor routine returns how many bytes were actually used.. | |
1684 | * NOTE! This may not be the same as how much of a user buffer | |
1685 | * we filled up (we may be padding etc), so we can only update | |
1686 | * "pos" here (the actor routine has to update the user buffer | |
1687 | * pointers and the remaining count). | |
1688 | */ | |
1689 | ret = actor(desc, page, offset, nr); | |
1690 | offset += ret; | |
1691 | index += offset >> PAGE_CACHE_SHIFT; | |
1692 | offset &= ~PAGE_CACHE_MASK; | |
1693 | ||
1694 | page_cache_release(page); | |
1695 | if (ret != nr || !desc->count) | |
1696 | break; | |
1697 | ||
1698 | cond_resched(); | |
1699 | } | |
1700 | ||
1701 | *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; | |
1702 | file_accessed(filp); | |
1703 | } | |
1704 | ||
bcd78e49 HD |
1705 | static ssize_t shmem_file_aio_read(struct kiocb *iocb, |
1706 | const struct iovec *iov, unsigned long nr_segs, loff_t pos) | |
1707 | { | |
1708 | struct file *filp = iocb->ki_filp; | |
1709 | ssize_t retval; | |
1710 | unsigned long seg; | |
1711 | size_t count; | |
1712 | loff_t *ppos = &iocb->ki_pos; | |
1713 | ||
1714 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); | |
1715 | if (retval) | |
1716 | return retval; | |
1717 | ||
1718 | for (seg = 0; seg < nr_segs; seg++) { | |
1719 | read_descriptor_t desc; | |
1720 | ||
1721 | desc.written = 0; | |
1722 | desc.arg.buf = iov[seg].iov_base; | |
1723 | desc.count = iov[seg].iov_len; | |
1724 | if (desc.count == 0) | |
1725 | continue; | |
1726 | desc.error = 0; | |
1727 | do_shmem_file_read(filp, ppos, &desc, file_read_actor); | |
1728 | retval += desc.written; | |
1729 | if (desc.error) { | |
1730 | retval = retval ?: desc.error; | |
1731 | break; | |
1732 | } | |
1733 | if (desc.count > 0) | |
1734 | break; | |
1735 | } | |
1736 | return retval; | |
1da177e4 LT |
1737 | } |
1738 | ||
726c3342 | 1739 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 1740 | { |
726c3342 | 1741 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
1742 | |
1743 | buf->f_type = TMPFS_MAGIC; | |
1744 | buf->f_bsize = PAGE_CACHE_SIZE; | |
1745 | buf->f_namelen = NAME_MAX; | |
0edd73b3 HD |
1746 | spin_lock(&sbinfo->stat_lock); |
1747 | if (sbinfo->max_blocks) { | |
1da177e4 LT |
1748 | buf->f_blocks = sbinfo->max_blocks; |
1749 | buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; | |
0edd73b3 HD |
1750 | } |
1751 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
1752 | buf->f_files = sbinfo->max_inodes; |
1753 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
1754 | } |
1755 | /* else leave those fields 0 like simple_statfs */ | |
0edd73b3 | 1756 | spin_unlock(&sbinfo->stat_lock); |
1da177e4 LT |
1757 | return 0; |
1758 | } | |
1759 | ||
1760 | /* | |
1761 | * File creation. Allocate an inode, and we're done.. | |
1762 | */ | |
1763 | static int | |
1764 | shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) | |
1765 | { | |
1766 | struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); | |
1767 | int error = -ENOSPC; | |
1768 | ||
1769 | if (inode) { | |
570bc1c2 SS |
1770 | error = security_inode_init_security(inode, dir, NULL, NULL, |
1771 | NULL); | |
1772 | if (error) { | |
1773 | if (error != -EOPNOTSUPP) { | |
1774 | iput(inode); | |
1775 | return error; | |
1776 | } | |
39f0247d AG |
1777 | } |
1778 | error = shmem_acl_init(inode, dir); | |
1779 | if (error) { | |
1780 | iput(inode); | |
1781 | return error; | |
570bc1c2 | 1782 | } |
1da177e4 LT |
1783 | if (dir->i_mode & S_ISGID) { |
1784 | inode->i_gid = dir->i_gid; | |
1785 | if (S_ISDIR(mode)) | |
1786 | inode->i_mode |= S_ISGID; | |
1787 | } | |
1788 | dir->i_size += BOGO_DIRENT_SIZE; | |
1789 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
1790 | d_instantiate(dentry, inode); | |
1791 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
1792 | } |
1793 | return error; | |
1794 | } | |
1795 | ||
1796 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
1797 | { | |
1798 | int error; | |
1799 | ||
1800 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
1801 | return error; | |
d8c76e6f | 1802 | inc_nlink(dir); |
1da177e4 LT |
1803 | return 0; |
1804 | } | |
1805 | ||
1806 | static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, | |
1807 | struct nameidata *nd) | |
1808 | { | |
1809 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
1810 | } | |
1811 | ||
1812 | /* | |
1813 | * Link a file.. | |
1814 | */ | |
1815 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1816 | { | |
1817 | struct inode *inode = old_dentry->d_inode; | |
5b04c689 | 1818 | int ret; |
1da177e4 LT |
1819 | |
1820 | /* | |
1821 | * No ordinary (disk based) filesystem counts links as inodes; | |
1822 | * but each new link needs a new dentry, pinning lowmem, and | |
1823 | * tmpfs dentries cannot be pruned until they are unlinked. | |
1824 | */ | |
5b04c689 PE |
1825 | ret = shmem_reserve_inode(inode->i_sb); |
1826 | if (ret) | |
1827 | goto out; | |
1da177e4 LT |
1828 | |
1829 | dir->i_size += BOGO_DIRENT_SIZE; | |
1830 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 1831 | inc_nlink(inode); |
1da177e4 LT |
1832 | atomic_inc(&inode->i_count); /* New dentry reference */ |
1833 | dget(dentry); /* Extra pinning count for the created dentry */ | |
1834 | d_instantiate(dentry, inode); | |
5b04c689 PE |
1835 | out: |
1836 | return ret; | |
1da177e4 LT |
1837 | } |
1838 | ||
1839 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
1840 | { | |
1841 | struct inode *inode = dentry->d_inode; | |
1842 | ||
5b04c689 PE |
1843 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
1844 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
1845 | |
1846 | dir->i_size -= BOGO_DIRENT_SIZE; | |
1847 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 1848 | drop_nlink(inode); |
1da177e4 LT |
1849 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
1850 | return 0; | |
1851 | } | |
1852 | ||
1853 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
1854 | { | |
1855 | if (!simple_empty(dentry)) | |
1856 | return -ENOTEMPTY; | |
1857 | ||
9a53c3a7 DH |
1858 | drop_nlink(dentry->d_inode); |
1859 | drop_nlink(dir); | |
1da177e4 LT |
1860 | return shmem_unlink(dir, dentry); |
1861 | } | |
1862 | ||
1863 | /* | |
1864 | * The VFS layer already does all the dentry stuff for rename, | |
1865 | * we just have to decrement the usage count for the target if | |
1866 | * it exists so that the VFS layer correctly free's it when it | |
1867 | * gets overwritten. | |
1868 | */ | |
1869 | static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) | |
1870 | { | |
1871 | struct inode *inode = old_dentry->d_inode; | |
1872 | int they_are_dirs = S_ISDIR(inode->i_mode); | |
1873 | ||
1874 | if (!simple_empty(new_dentry)) | |
1875 | return -ENOTEMPTY; | |
1876 | ||
1877 | if (new_dentry->d_inode) { | |
1878 | (void) shmem_unlink(new_dir, new_dentry); | |
1879 | if (they_are_dirs) | |
9a53c3a7 | 1880 | drop_nlink(old_dir); |
1da177e4 | 1881 | } else if (they_are_dirs) { |
9a53c3a7 | 1882 | drop_nlink(old_dir); |
d8c76e6f | 1883 | inc_nlink(new_dir); |
1da177e4 LT |
1884 | } |
1885 | ||
1886 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
1887 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
1888 | old_dir->i_ctime = old_dir->i_mtime = | |
1889 | new_dir->i_ctime = new_dir->i_mtime = | |
1890 | inode->i_ctime = CURRENT_TIME; | |
1891 | return 0; | |
1892 | } | |
1893 | ||
1894 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1895 | { | |
1896 | int error; | |
1897 | int len; | |
1898 | struct inode *inode; | |
1899 | struct page *page = NULL; | |
1900 | char *kaddr; | |
1901 | struct shmem_inode_info *info; | |
1902 | ||
1903 | len = strlen(symname) + 1; | |
1904 | if (len > PAGE_CACHE_SIZE) | |
1905 | return -ENAMETOOLONG; | |
1906 | ||
1907 | inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); | |
1908 | if (!inode) | |
1909 | return -ENOSPC; | |
1910 | ||
570bc1c2 SS |
1911 | error = security_inode_init_security(inode, dir, NULL, NULL, |
1912 | NULL); | |
1913 | if (error) { | |
1914 | if (error != -EOPNOTSUPP) { | |
1915 | iput(inode); | |
1916 | return error; | |
1917 | } | |
1918 | error = 0; | |
1919 | } | |
1920 | ||
1da177e4 LT |
1921 | info = SHMEM_I(inode); |
1922 | inode->i_size = len-1; | |
1923 | if (len <= (char *)inode - (char *)info) { | |
1924 | /* do it inline */ | |
1925 | memcpy(info, symname, len); | |
1926 | inode->i_op = &shmem_symlink_inline_operations; | |
1927 | } else { | |
1928 | error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); | |
1929 | if (error) { | |
1930 | iput(inode); | |
1931 | return error; | |
1932 | } | |
d3602444 | 1933 | unlock_page(page); |
14fcc23f | 1934 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
1935 | inode->i_op = &shmem_symlink_inode_operations; |
1936 | kaddr = kmap_atomic(page, KM_USER0); | |
1937 | memcpy(kaddr, symname, len); | |
1938 | kunmap_atomic(kaddr, KM_USER0); | |
1939 | set_page_dirty(page); | |
1940 | page_cache_release(page); | |
1941 | } | |
1942 | if (dir->i_mode & S_ISGID) | |
1943 | inode->i_gid = dir->i_gid; | |
1944 | dir->i_size += BOGO_DIRENT_SIZE; | |
1945 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
1946 | d_instantiate(dentry, inode); | |
1947 | dget(dentry); | |
1948 | return 0; | |
1949 | } | |
1950 | ||
cc314eef | 1951 | static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) |
1da177e4 LT |
1952 | { |
1953 | nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); | |
cc314eef | 1954 | return NULL; |
1da177e4 LT |
1955 | } |
1956 | ||
cc314eef | 1957 | static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) |
1da177e4 LT |
1958 | { |
1959 | struct page *page = NULL; | |
1960 | int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); | |
1961 | nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); | |
d3602444 HD |
1962 | if (page) |
1963 | unlock_page(page); | |
cc314eef | 1964 | return page; |
1da177e4 LT |
1965 | } |
1966 | ||
cc314eef | 1967 | static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 LT |
1968 | { |
1969 | if (!IS_ERR(nd_get_link(nd))) { | |
cc314eef | 1970 | struct page *page = cookie; |
1da177e4 LT |
1971 | kunmap(page); |
1972 | mark_page_accessed(page); | |
1973 | page_cache_release(page); | |
1da177e4 LT |
1974 | } |
1975 | } | |
1976 | ||
92e1d5be | 1977 | static const struct inode_operations shmem_symlink_inline_operations = { |
1da177e4 LT |
1978 | .readlink = generic_readlink, |
1979 | .follow_link = shmem_follow_link_inline, | |
1da177e4 LT |
1980 | }; |
1981 | ||
92e1d5be | 1982 | static const struct inode_operations shmem_symlink_inode_operations = { |
1da177e4 LT |
1983 | .truncate = shmem_truncate, |
1984 | .readlink = generic_readlink, | |
1985 | .follow_link = shmem_follow_link, | |
1986 | .put_link = shmem_put_link, | |
1da177e4 LT |
1987 | }; |
1988 | ||
39f0247d | 1989 | #ifdef CONFIG_TMPFS_POSIX_ACL |
46711810 | 1990 | /* |
39f0247d AG |
1991 | * Superblocks without xattr inode operations will get security.* xattr |
1992 | * support from the VFS "for free". As soon as we have any other xattrs | |
1993 | * like ACLs, we also need to implement the security.* handlers at | |
1994 | * filesystem level, though. | |
1995 | */ | |
1996 | ||
1997 | static size_t shmem_xattr_security_list(struct inode *inode, char *list, | |
1998 | size_t list_len, const char *name, | |
1999 | size_t name_len) | |
2000 | { | |
2001 | return security_inode_listsecurity(inode, list, list_len); | |
2002 | } | |
2003 | ||
2004 | static int shmem_xattr_security_get(struct inode *inode, const char *name, | |
2005 | void *buffer, size_t size) | |
2006 | { | |
2007 | if (strcmp(name, "") == 0) | |
2008 | return -EINVAL; | |
42492594 | 2009 | return xattr_getsecurity(inode, name, buffer, size); |
39f0247d AG |
2010 | } |
2011 | ||
2012 | static int shmem_xattr_security_set(struct inode *inode, const char *name, | |
2013 | const void *value, size_t size, int flags) | |
2014 | { | |
2015 | if (strcmp(name, "") == 0) | |
2016 | return -EINVAL; | |
2017 | return security_inode_setsecurity(inode, name, value, size, flags); | |
2018 | } | |
2019 | ||
1f370a23 | 2020 | static struct xattr_handler shmem_xattr_security_handler = { |
39f0247d AG |
2021 | .prefix = XATTR_SECURITY_PREFIX, |
2022 | .list = shmem_xattr_security_list, | |
2023 | .get = shmem_xattr_security_get, | |
2024 | .set = shmem_xattr_security_set, | |
2025 | }; | |
2026 | ||
2027 | static struct xattr_handler *shmem_xattr_handlers[] = { | |
2028 | &shmem_xattr_acl_access_handler, | |
2029 | &shmem_xattr_acl_default_handler, | |
2030 | &shmem_xattr_security_handler, | |
2031 | NULL | |
2032 | }; | |
2033 | #endif | |
2034 | ||
91828a40 DG |
2035 | static struct dentry *shmem_get_parent(struct dentry *child) |
2036 | { | |
2037 | return ERR_PTR(-ESTALE); | |
2038 | } | |
2039 | ||
2040 | static int shmem_match(struct inode *ino, void *vfh) | |
2041 | { | |
2042 | __u32 *fh = vfh; | |
2043 | __u64 inum = fh[2]; | |
2044 | inum = (inum << 32) | fh[1]; | |
2045 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
2046 | } | |
2047 | ||
480b116c CH |
2048 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
2049 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 2050 | { |
91828a40 | 2051 | struct inode *inode; |
480b116c CH |
2052 | struct dentry *dentry = NULL; |
2053 | u64 inum = fid->raw[2]; | |
2054 | inum = (inum << 32) | fid->raw[1]; | |
2055 | ||
2056 | if (fh_len < 3) | |
2057 | return NULL; | |
91828a40 | 2058 | |
480b116c CH |
2059 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
2060 | shmem_match, fid->raw); | |
91828a40 | 2061 | if (inode) { |
480b116c | 2062 | dentry = d_find_alias(inode); |
91828a40 DG |
2063 | iput(inode); |
2064 | } | |
2065 | ||
480b116c | 2066 | return dentry; |
91828a40 DG |
2067 | } |
2068 | ||
2069 | static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, | |
2070 | int connectable) | |
2071 | { | |
2072 | struct inode *inode = dentry->d_inode; | |
2073 | ||
2074 | if (*len < 3) | |
2075 | return 255; | |
2076 | ||
2077 | if (hlist_unhashed(&inode->i_hash)) { | |
2078 | /* Unfortunately insert_inode_hash is not idempotent, | |
2079 | * so as we hash inodes here rather than at creation | |
2080 | * time, we need a lock to ensure we only try | |
2081 | * to do it once | |
2082 | */ | |
2083 | static DEFINE_SPINLOCK(lock); | |
2084 | spin_lock(&lock); | |
2085 | if (hlist_unhashed(&inode->i_hash)) | |
2086 | __insert_inode_hash(inode, | |
2087 | inode->i_ino + inode->i_generation); | |
2088 | spin_unlock(&lock); | |
2089 | } | |
2090 | ||
2091 | fh[0] = inode->i_generation; | |
2092 | fh[1] = inode->i_ino; | |
2093 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
2094 | ||
2095 | *len = 3; | |
2096 | return 1; | |
2097 | } | |
2098 | ||
39655164 | 2099 | static const struct export_operations shmem_export_ops = { |
91828a40 | 2100 | .get_parent = shmem_get_parent, |
91828a40 | 2101 | .encode_fh = shmem_encode_fh, |
480b116c | 2102 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
2103 | }; |
2104 | ||
680d794b | 2105 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
2106 | bool remount) | |
1da177e4 LT |
2107 | { |
2108 | char *this_char, *value, *rest; | |
2109 | ||
b00dc3ad HD |
2110 | while (options != NULL) { |
2111 | this_char = options; | |
2112 | for (;;) { | |
2113 | /* | |
2114 | * NUL-terminate this option: unfortunately, | |
2115 | * mount options form a comma-separated list, | |
2116 | * but mpol's nodelist may also contain commas. | |
2117 | */ | |
2118 | options = strchr(options, ','); | |
2119 | if (options == NULL) | |
2120 | break; | |
2121 | options++; | |
2122 | if (!isdigit(*options)) { | |
2123 | options[-1] = '\0'; | |
2124 | break; | |
2125 | } | |
2126 | } | |
1da177e4 LT |
2127 | if (!*this_char) |
2128 | continue; | |
2129 | if ((value = strchr(this_char,'=')) != NULL) { | |
2130 | *value++ = 0; | |
2131 | } else { | |
2132 | printk(KERN_ERR | |
2133 | "tmpfs: No value for mount option '%s'\n", | |
2134 | this_char); | |
2135 | return 1; | |
2136 | } | |
2137 | ||
2138 | if (!strcmp(this_char,"size")) { | |
2139 | unsigned long long size; | |
2140 | size = memparse(value,&rest); | |
2141 | if (*rest == '%') { | |
2142 | size <<= PAGE_SHIFT; | |
2143 | size *= totalram_pages; | |
2144 | do_div(size, 100); | |
2145 | rest++; | |
2146 | } | |
2147 | if (*rest) | |
2148 | goto bad_val; | |
680d794b | 2149 | sbinfo->max_blocks = |
2150 | DIV_ROUND_UP(size, PAGE_CACHE_SIZE); | |
1da177e4 | 2151 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 2152 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
2153 | if (*rest) |
2154 | goto bad_val; | |
2155 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 2156 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
2157 | if (*rest) |
2158 | goto bad_val; | |
2159 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 2160 | if (remount) |
1da177e4 | 2161 | continue; |
680d794b | 2162 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
2163 | if (*rest) |
2164 | goto bad_val; | |
2165 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 2166 | if (remount) |
1da177e4 | 2167 | continue; |
680d794b | 2168 | sbinfo->uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2169 | if (*rest) |
2170 | goto bad_val; | |
2171 | } else if (!strcmp(this_char,"gid")) { | |
680d794b | 2172 | if (remount) |
1da177e4 | 2173 | continue; |
680d794b | 2174 | sbinfo->gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2175 | if (*rest) |
2176 | goto bad_val; | |
7339ff83 | 2177 | } else if (!strcmp(this_char,"mpol")) { |
71fe804b | 2178 | if (mpol_parse_str(value, &sbinfo->mpol, 1)) |
7339ff83 | 2179 | goto bad_val; |
1da177e4 LT |
2180 | } else { |
2181 | printk(KERN_ERR "tmpfs: Bad mount option %s\n", | |
2182 | this_char); | |
2183 | return 1; | |
2184 | } | |
2185 | } | |
2186 | return 0; | |
2187 | ||
2188 | bad_val: | |
2189 | printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", | |
2190 | value, this_char); | |
2191 | return 1; | |
2192 | ||
2193 | } | |
2194 | ||
2195 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
2196 | { | |
2197 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 2198 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
2199 | unsigned long blocks; |
2200 | unsigned long inodes; | |
2201 | int error = -EINVAL; | |
2202 | ||
680d794b | 2203 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 2204 | return error; |
1da177e4 | 2205 | |
0edd73b3 HD |
2206 | spin_lock(&sbinfo->stat_lock); |
2207 | blocks = sbinfo->max_blocks - sbinfo->free_blocks; | |
2208 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; | |
680d794b | 2209 | if (config.max_blocks < blocks) |
0edd73b3 | 2210 | goto out; |
680d794b | 2211 | if (config.max_inodes < inodes) |
0edd73b3 HD |
2212 | goto out; |
2213 | /* | |
2214 | * Those tests also disallow limited->unlimited while any are in | |
2215 | * use, so i_blocks will always be zero when max_blocks is zero; | |
2216 | * but we must separately disallow unlimited->limited, because | |
2217 | * in that case we have no record of how much is already in use. | |
2218 | */ | |
680d794b | 2219 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 2220 | goto out; |
680d794b | 2221 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
2222 | goto out; |
2223 | ||
2224 | error = 0; | |
680d794b | 2225 | sbinfo->max_blocks = config.max_blocks; |
2226 | sbinfo->free_blocks = config.max_blocks - blocks; | |
2227 | sbinfo->max_inodes = config.max_inodes; | |
2228 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b LS |
2229 | |
2230 | mpol_put(sbinfo->mpol); | |
2231 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
0edd73b3 HD |
2232 | out: |
2233 | spin_unlock(&sbinfo->stat_lock); | |
2234 | return error; | |
1da177e4 | 2235 | } |
680d794b | 2236 | |
2237 | static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) | |
2238 | { | |
2239 | struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); | |
2240 | ||
2241 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
2242 | seq_printf(seq, ",size=%luk", | |
2243 | sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); | |
2244 | if (sbinfo->max_inodes != shmem_default_max_inodes()) | |
2245 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
2246 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
2247 | seq_printf(seq, ",mode=%03o", sbinfo->mode); | |
2248 | if (sbinfo->uid != 0) | |
2249 | seq_printf(seq, ",uid=%u", sbinfo->uid); | |
2250 | if (sbinfo->gid != 0) | |
2251 | seq_printf(seq, ",gid=%u", sbinfo->gid); | |
71fe804b | 2252 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b | 2253 | return 0; |
2254 | } | |
2255 | #endif /* CONFIG_TMPFS */ | |
1da177e4 LT |
2256 | |
2257 | static void shmem_put_super(struct super_block *sb) | |
2258 | { | |
2259 | kfree(sb->s_fs_info); | |
2260 | sb->s_fs_info = NULL; | |
2261 | } | |
2262 | ||
1da177e4 LT |
2263 | static int shmem_fill_super(struct super_block *sb, |
2264 | void *data, int silent) | |
2265 | { | |
2266 | struct inode *inode; | |
2267 | struct dentry *root; | |
0edd73b3 | 2268 | struct shmem_sb_info *sbinfo; |
680d794b | 2269 | int err = -ENOMEM; |
2270 | ||
2271 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
2272 | sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), | |
2273 | L1_CACHE_BYTES), GFP_KERNEL); | |
2274 | if (!sbinfo) | |
2275 | return -ENOMEM; | |
2276 | ||
2277 | sbinfo->max_blocks = 0; | |
2278 | sbinfo->max_inodes = 0; | |
2279 | sbinfo->mode = S_IRWXUGO | S_ISVTX; | |
76aac0e9 DH |
2280 | sbinfo->uid = current_fsuid(); |
2281 | sbinfo->gid = current_fsgid(); | |
71fe804b | 2282 | sbinfo->mpol = NULL; |
680d794b | 2283 | sb->s_fs_info = sbinfo; |
1da177e4 | 2284 | |
0edd73b3 | 2285 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
2286 | /* |
2287 | * Per default we only allow half of the physical ram per | |
2288 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
2289 | * but the internal instance is left unlimited. | |
2290 | */ | |
2291 | if (!(sb->s_flags & MS_NOUSER)) { | |
680d794b | 2292 | sbinfo->max_blocks = shmem_default_max_blocks(); |
2293 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
2294 | if (shmem_parse_options(data, sbinfo, false)) { | |
2295 | err = -EINVAL; | |
2296 | goto failed; | |
2297 | } | |
1da177e4 | 2298 | } |
91828a40 | 2299 | sb->s_export_op = &shmem_export_ops; |
1da177e4 LT |
2300 | #else |
2301 | sb->s_flags |= MS_NOUSER; | |
2302 | #endif | |
2303 | ||
0edd73b3 | 2304 | spin_lock_init(&sbinfo->stat_lock); |
680d794b | 2305 | sbinfo->free_blocks = sbinfo->max_blocks; |
2306 | sbinfo->free_inodes = sbinfo->max_inodes; | |
0edd73b3 | 2307 | |
1da177e4 LT |
2308 | sb->s_maxbytes = SHMEM_MAX_BYTES; |
2309 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
2310 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
2311 | sb->s_magic = TMPFS_MAGIC; | |
2312 | sb->s_op = &shmem_ops; | |
cfd95a9c | 2313 | sb->s_time_gran = 1; |
39f0247d AG |
2314 | #ifdef CONFIG_TMPFS_POSIX_ACL |
2315 | sb->s_xattr = shmem_xattr_handlers; | |
2316 | sb->s_flags |= MS_POSIXACL; | |
2317 | #endif | |
0edd73b3 | 2318 | |
680d794b | 2319 | inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0); |
1da177e4 LT |
2320 | if (!inode) |
2321 | goto failed; | |
680d794b | 2322 | inode->i_uid = sbinfo->uid; |
2323 | inode->i_gid = sbinfo->gid; | |
1da177e4 LT |
2324 | root = d_alloc_root(inode); |
2325 | if (!root) | |
2326 | goto failed_iput; | |
2327 | sb->s_root = root; | |
2328 | return 0; | |
2329 | ||
2330 | failed_iput: | |
2331 | iput(inode); | |
2332 | failed: | |
2333 | shmem_put_super(sb); | |
2334 | return err; | |
2335 | } | |
2336 | ||
fcc234f8 | 2337 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
2338 | |
2339 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
2340 | { | |
2341 | struct shmem_inode_info *p; | |
e94b1766 | 2342 | p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); |
1da177e4 LT |
2343 | if (!p) |
2344 | return NULL; | |
2345 | return &p->vfs_inode; | |
2346 | } | |
2347 | ||
2348 | static void shmem_destroy_inode(struct inode *inode) | |
2349 | { | |
2350 | if ((inode->i_mode & S_IFMT) == S_IFREG) { | |
2351 | /* only struct inode is valid if it's an inline symlink */ | |
2352 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); | |
2353 | } | |
39f0247d | 2354 | shmem_acl_destroy_inode(inode); |
1da177e4 LT |
2355 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
2356 | } | |
2357 | ||
51cc5068 | 2358 | static void init_once(void *foo) |
1da177e4 LT |
2359 | { |
2360 | struct shmem_inode_info *p = (struct shmem_inode_info *) foo; | |
2361 | ||
a35afb83 | 2362 | inode_init_once(&p->vfs_inode); |
39f0247d | 2363 | #ifdef CONFIG_TMPFS_POSIX_ACL |
a35afb83 CL |
2364 | p->i_acl = NULL; |
2365 | p->i_default_acl = NULL; | |
39f0247d | 2366 | #endif |
1da177e4 LT |
2367 | } |
2368 | ||
2369 | static int init_inodecache(void) | |
2370 | { | |
2371 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
2372 | sizeof(struct shmem_inode_info), | |
040b5c6f | 2373 | 0, SLAB_PANIC, init_once); |
1da177e4 LT |
2374 | return 0; |
2375 | } | |
2376 | ||
2377 | static void destroy_inodecache(void) | |
2378 | { | |
1a1d92c1 | 2379 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
2380 | } |
2381 | ||
f5e54d6e | 2382 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 2383 | .writepage = shmem_writepage, |
76719325 | 2384 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 2385 | #ifdef CONFIG_TMPFS |
ae976416 | 2386 | .readpage = shmem_readpage, |
800d15a5 NP |
2387 | .write_begin = shmem_write_begin, |
2388 | .write_end = shmem_write_end, | |
1da177e4 | 2389 | #endif |
304dbdb7 | 2390 | .migratepage = migrate_page, |
1da177e4 LT |
2391 | }; |
2392 | ||
15ad7cdc | 2393 | static const struct file_operations shmem_file_operations = { |
1da177e4 LT |
2394 | .mmap = shmem_mmap, |
2395 | #ifdef CONFIG_TMPFS | |
2396 | .llseek = generic_file_llseek, | |
bcd78e49 | 2397 | .read = do_sync_read, |
5402b976 | 2398 | .write = do_sync_write, |
bcd78e49 | 2399 | .aio_read = shmem_file_aio_read, |
5402b976 | 2400 | .aio_write = generic_file_aio_write, |
1da177e4 | 2401 | .fsync = simple_sync_file, |
ae976416 HD |
2402 | .splice_read = generic_file_splice_read, |
2403 | .splice_write = generic_file_splice_write, | |
1da177e4 LT |
2404 | #endif |
2405 | }; | |
2406 | ||
92e1d5be | 2407 | static const struct inode_operations shmem_inode_operations = { |
1da177e4 LT |
2408 | .truncate = shmem_truncate, |
2409 | .setattr = shmem_notify_change, | |
f6b3ec23 | 2410 | .truncate_range = shmem_truncate_range, |
39f0247d AG |
2411 | #ifdef CONFIG_TMPFS_POSIX_ACL |
2412 | .setxattr = generic_setxattr, | |
2413 | .getxattr = generic_getxattr, | |
2414 | .listxattr = generic_listxattr, | |
2415 | .removexattr = generic_removexattr, | |
2416 | .permission = shmem_permission, | |
2417 | #endif | |
2418 | ||
1da177e4 LT |
2419 | }; |
2420 | ||
92e1d5be | 2421 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
2422 | #ifdef CONFIG_TMPFS |
2423 | .create = shmem_create, | |
2424 | .lookup = simple_lookup, | |
2425 | .link = shmem_link, | |
2426 | .unlink = shmem_unlink, | |
2427 | .symlink = shmem_symlink, | |
2428 | .mkdir = shmem_mkdir, | |
2429 | .rmdir = shmem_rmdir, | |
2430 | .mknod = shmem_mknod, | |
2431 | .rename = shmem_rename, | |
1da177e4 | 2432 | #endif |
39f0247d AG |
2433 | #ifdef CONFIG_TMPFS_POSIX_ACL |
2434 | .setattr = shmem_notify_change, | |
2435 | .setxattr = generic_setxattr, | |
2436 | .getxattr = generic_getxattr, | |
2437 | .listxattr = generic_listxattr, | |
2438 | .removexattr = generic_removexattr, | |
2439 | .permission = shmem_permission, | |
2440 | #endif | |
2441 | }; | |
2442 | ||
92e1d5be | 2443 | static const struct inode_operations shmem_special_inode_operations = { |
39f0247d AG |
2444 | #ifdef CONFIG_TMPFS_POSIX_ACL |
2445 | .setattr = shmem_notify_change, | |
2446 | .setxattr = generic_setxattr, | |
2447 | .getxattr = generic_getxattr, | |
2448 | .listxattr = generic_listxattr, | |
2449 | .removexattr = generic_removexattr, | |
2450 | .permission = shmem_permission, | |
2451 | #endif | |
1da177e4 LT |
2452 | }; |
2453 | ||
759b9775 | 2454 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
2455 | .alloc_inode = shmem_alloc_inode, |
2456 | .destroy_inode = shmem_destroy_inode, | |
2457 | #ifdef CONFIG_TMPFS | |
2458 | .statfs = shmem_statfs, | |
2459 | .remount_fs = shmem_remount_fs, | |
680d794b | 2460 | .show_options = shmem_show_options, |
1da177e4 LT |
2461 | #endif |
2462 | .delete_inode = shmem_delete_inode, | |
2463 | .drop_inode = generic_delete_inode, | |
2464 | .put_super = shmem_put_super, | |
2465 | }; | |
2466 | ||
2467 | static struct vm_operations_struct shmem_vm_ops = { | |
54cb8821 | 2468 | .fault = shmem_fault, |
1da177e4 LT |
2469 | #ifdef CONFIG_NUMA |
2470 | .set_policy = shmem_set_policy, | |
2471 | .get_policy = shmem_get_policy, | |
2472 | #endif | |
2473 | }; | |
2474 | ||
2475 | ||
454e2398 DH |
2476 | static int shmem_get_sb(struct file_system_type *fs_type, |
2477 | int flags, const char *dev_name, void *data, struct vfsmount *mnt) | |
1da177e4 | 2478 | { |
454e2398 | 2479 | return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); |
1da177e4 LT |
2480 | } |
2481 | ||
2482 | static struct file_system_type tmpfs_fs_type = { | |
2483 | .owner = THIS_MODULE, | |
2484 | .name = "tmpfs", | |
2485 | .get_sb = shmem_get_sb, | |
2486 | .kill_sb = kill_litter_super, | |
2487 | }; | |
2488 | static struct vfsmount *shm_mnt; | |
2489 | ||
2490 | static int __init init_tmpfs(void) | |
2491 | { | |
2492 | int error; | |
2493 | ||
e0bf68dd PZ |
2494 | error = bdi_init(&shmem_backing_dev_info); |
2495 | if (error) | |
2496 | goto out4; | |
2497 | ||
1da177e4 LT |
2498 | error = init_inodecache(); |
2499 | if (error) | |
2500 | goto out3; | |
2501 | ||
2502 | error = register_filesystem(&tmpfs_fs_type); | |
2503 | if (error) { | |
2504 | printk(KERN_ERR "Could not register tmpfs\n"); | |
2505 | goto out2; | |
2506 | } | |
95dc112a | 2507 | |
1f5ce9e9 | 2508 | shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, |
1da177e4 LT |
2509 | tmpfs_fs_type.name, NULL); |
2510 | if (IS_ERR(shm_mnt)) { | |
2511 | error = PTR_ERR(shm_mnt); | |
2512 | printk(KERN_ERR "Could not kern_mount tmpfs\n"); | |
2513 | goto out1; | |
2514 | } | |
2515 | return 0; | |
2516 | ||
2517 | out1: | |
2518 | unregister_filesystem(&tmpfs_fs_type); | |
2519 | out2: | |
2520 | destroy_inodecache(); | |
2521 | out3: | |
e0bf68dd PZ |
2522 | bdi_destroy(&shmem_backing_dev_info); |
2523 | out4: | |
1da177e4 LT |
2524 | shm_mnt = ERR_PTR(error); |
2525 | return error; | |
2526 | } | |
2527 | module_init(init_tmpfs) | |
2528 | ||
46711810 | 2529 | /** |
1da177e4 | 2530 | * shmem_file_setup - get an unlinked file living in tmpfs |
1da177e4 LT |
2531 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
2532 | * @size: size to be set for the file | |
46711810 | 2533 | * @flags: vm_flags |
1da177e4 LT |
2534 | */ |
2535 | struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) | |
2536 | { | |
2537 | int error; | |
2538 | struct file *file; | |
2539 | struct inode *inode; | |
2540 | struct dentry *dentry, *root; | |
2541 | struct qstr this; | |
2542 | ||
2543 | if (IS_ERR(shm_mnt)) | |
2544 | return (void *)shm_mnt; | |
2545 | ||
2546 | if (size < 0 || size > SHMEM_MAX_BYTES) | |
2547 | return ERR_PTR(-EINVAL); | |
2548 | ||
2549 | if (shmem_acct_size(flags, size)) | |
2550 | return ERR_PTR(-ENOMEM); | |
2551 | ||
2552 | error = -ENOMEM; | |
2553 | this.name = name; | |
2554 | this.len = strlen(name); | |
2555 | this.hash = 0; /* will go */ | |
2556 | root = shm_mnt->mnt_root; | |
2557 | dentry = d_alloc(root, &this); | |
2558 | if (!dentry) | |
2559 | goto put_memory; | |
2560 | ||
2561 | error = -ENFILE; | |
2562 | file = get_empty_filp(); | |
2563 | if (!file) | |
2564 | goto put_dentry; | |
2565 | ||
2566 | error = -ENOSPC; | |
2567 | inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); | |
2568 | if (!inode) | |
2569 | goto close_file; | |
2570 | ||
2571 | SHMEM_I(inode)->flags = flags & VM_ACCOUNT; | |
2572 | d_instantiate(dentry, inode); | |
2573 | inode->i_size = size; | |
2574 | inode->i_nlink = 0; /* It is unlinked */ | |
ce8d2cdf DH |
2575 | init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ, |
2576 | &shmem_file_operations); | |
1da177e4 LT |
2577 | return file; |
2578 | ||
2579 | close_file: | |
2580 | put_filp(file); | |
2581 | put_dentry: | |
2582 | dput(dentry); | |
2583 | put_memory: | |
2584 | shmem_unacct_size(flags, size); | |
2585 | return ERR_PTR(error); | |
2586 | } | |
395e0ddc | 2587 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 2588 | |
46711810 | 2589 | /** |
1da177e4 | 2590 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
2591 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
2592 | */ | |
2593 | int shmem_zero_setup(struct vm_area_struct *vma) | |
2594 | { | |
2595 | struct file *file; | |
2596 | loff_t size = vma->vm_end - vma->vm_start; | |
2597 | ||
2598 | file = shmem_file_setup("dev/zero", size, vma->vm_flags); | |
2599 | if (IS_ERR(file)) | |
2600 | return PTR_ERR(file); | |
2601 | ||
2602 | if (vma->vm_file) | |
2603 | fput(vma->vm_file); | |
2604 | vma->vm_file = file; | |
2605 | vma->vm_ops = &shmem_vm_ops; | |
2606 | return 0; | |
2607 | } |