Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Resizable virtual memory filesystem for Linux. | |
3 | * | |
4 | * Copyright (C) 2000 Linus Torvalds. | |
5 | * 2000 Transmeta Corp. | |
6 | * 2000-2001 Christoph Rohland | |
7 | * 2000-2001 SAP AG | |
8 | * 2002 Red Hat Inc. | |
6922c0c7 HD |
9 | * Copyright (C) 2002-2011 Hugh Dickins. |
10 | * Copyright (C) 2011 Google Inc. | |
0edd73b3 | 11 | * Copyright (C) 2002-2005 VERITAS Software Corporation. |
1da177e4 LT |
12 | * Copyright (C) 2004 Andi Kleen, SuSE Labs |
13 | * | |
14 | * Extended attribute support for tmpfs: | |
15 | * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net> | |
16 | * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com> | |
17 | * | |
853ac43a MM |
18 | * tiny-shmem: |
19 | * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com> | |
20 | * | |
1da177e4 LT |
21 | * This file is released under the GPL. |
22 | */ | |
23 | ||
853ac43a MM |
24 | #include <linux/fs.h> |
25 | #include <linux/init.h> | |
26 | #include <linux/vfs.h> | |
27 | #include <linux/mount.h> | |
250297ed | 28 | #include <linux/ramfs.h> |
caefba17 | 29 | #include <linux/pagemap.h> |
853ac43a MM |
30 | #include <linux/file.h> |
31 | #include <linux/mm.h> | |
b95f1b31 | 32 | #include <linux/export.h> |
853ac43a | 33 | #include <linux/swap.h> |
e2e40f2c | 34 | #include <linux/uio.h> |
853ac43a MM |
35 | |
36 | static struct vfsmount *shm_mnt; | |
37 | ||
38 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
39 | /* |
40 | * This virtual memory filesystem is heavily based on the ramfs. It | |
41 | * extends ramfs by the ability to use swap and honor resource limits | |
42 | * which makes it a completely usable filesystem. | |
43 | */ | |
44 | ||
39f0247d | 45 | #include <linux/xattr.h> |
a5694255 | 46 | #include <linux/exportfs.h> |
1c7c474c | 47 | #include <linux/posix_acl.h> |
feda821e | 48 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 49 | #include <linux/mman.h> |
1da177e4 LT |
50 | #include <linux/string.h> |
51 | #include <linux/slab.h> | |
52 | #include <linux/backing-dev.h> | |
53 | #include <linux/shmem_fs.h> | |
1da177e4 | 54 | #include <linux/writeback.h> |
1da177e4 | 55 | #include <linux/blkdev.h> |
bda97eab | 56 | #include <linux/pagevec.h> |
41ffe5d5 | 57 | #include <linux/percpu_counter.h> |
83e4fa9c | 58 | #include <linux/falloc.h> |
708e3508 | 59 | #include <linux/splice.h> |
1da177e4 LT |
60 | #include <linux/security.h> |
61 | #include <linux/swapops.h> | |
62 | #include <linux/mempolicy.h> | |
63 | #include <linux/namei.h> | |
b00dc3ad | 64 | #include <linux/ctype.h> |
304dbdb7 | 65 | #include <linux/migrate.h> |
c1f60a5a | 66 | #include <linux/highmem.h> |
680d794b | 67 | #include <linux/seq_file.h> |
92562927 | 68 | #include <linux/magic.h> |
9183df25 | 69 | #include <linux/syscalls.h> |
40e041a2 | 70 | #include <linux/fcntl.h> |
9183df25 | 71 | #include <uapi/linux/memfd.h> |
304dbdb7 | 72 | |
1da177e4 | 73 | #include <asm/uaccess.h> |
1da177e4 LT |
74 | #include <asm/pgtable.h> |
75 | ||
caefba17 | 76 | #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) |
1da177e4 LT |
77 | #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) |
78 | ||
1da177e4 LT |
79 | /* Pretend that each entry is of this size in directory's i_size */ |
80 | #define BOGO_DIRENT_SIZE 20 | |
81 | ||
69f07ec9 HD |
82 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
83 | #define SHORT_SYMLINK_LEN 128 | |
84 | ||
1aac1400 | 85 | /* |
f00cdc6d HD |
86 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
87 | * inode->i_private (with i_mutex making sure that it has only one user at | |
88 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
89 | */ |
90 | struct shmem_falloc { | |
8e205f77 | 91 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
92 | pgoff_t start; /* start of range currently being fallocated */ |
93 | pgoff_t next; /* the next page offset to be fallocated */ | |
94 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
95 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
96 | }; | |
97 | ||
285b2c4f | 98 | /* Flag allocation requirements to shmem_getpage */ |
1da177e4 | 99 | enum sgp_type { |
1da177e4 LT |
100 | SGP_READ, /* don't exceed i_size, don't allocate page */ |
101 | SGP_CACHE, /* don't exceed i_size, may allocate page */ | |
a0ee5ec5 | 102 | SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ |
1635f6a7 HD |
103 | SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */ |
104 | SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */ | |
1da177e4 LT |
105 | }; |
106 | ||
b76db735 | 107 | #ifdef CONFIG_TMPFS |
680d794b | 108 | static unsigned long shmem_default_max_blocks(void) |
109 | { | |
110 | return totalram_pages / 2; | |
111 | } | |
112 | ||
113 | static unsigned long shmem_default_max_inodes(void) | |
114 | { | |
115 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
116 | } | |
b76db735 | 117 | #endif |
680d794b | 118 | |
bde05d1c HD |
119 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
120 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
121 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 HD |
122 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
123 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type); | |
124 | ||
125 | static inline int shmem_getpage(struct inode *inode, pgoff_t index, | |
126 | struct page **pagep, enum sgp_type sgp, int *fault_type) | |
127 | { | |
128 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
129 | mapping_gfp_mask(inode->i_mapping), fault_type); | |
130 | } | |
1da177e4 | 131 | |
1da177e4 LT |
132 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
133 | { | |
134 | return sb->s_fs_info; | |
135 | } | |
136 | ||
137 | /* | |
138 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
139 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
140 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
141 | * consistent with the pre-accounting of private mappings ... | |
142 | */ | |
143 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
144 | { | |
0b0a0806 | 145 | return (flags & VM_NORESERVE) ? |
191c5424 | 146 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
147 | } |
148 | ||
149 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
150 | { | |
0b0a0806 | 151 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
152 | vm_unacct_memory(VM_ACCT(size)); |
153 | } | |
154 | ||
77142517 KK |
155 | static inline int shmem_reacct_size(unsigned long flags, |
156 | loff_t oldsize, loff_t newsize) | |
157 | { | |
158 | if (!(flags & VM_NORESERVE)) { | |
159 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
160 | return security_vm_enough_memory_mm(current->mm, | |
161 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
162 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
163 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
164 | } | |
165 | return 0; | |
166 | } | |
167 | ||
1da177e4 LT |
168 | /* |
169 | * ... whereas tmpfs objects are accounted incrementally as | |
170 | * pages are allocated, in order to allow huge sparse files. | |
171 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, | |
172 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
173 | */ | |
174 | static inline int shmem_acct_block(unsigned long flags) | |
175 | { | |
0b0a0806 | 176 | return (flags & VM_NORESERVE) ? |
191c5424 | 177 | security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0; |
1da177e4 LT |
178 | } |
179 | ||
180 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
181 | { | |
0b0a0806 | 182 | if (flags & VM_NORESERVE) |
1da177e4 LT |
183 | vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); |
184 | } | |
185 | ||
759b9775 | 186 | static const struct super_operations shmem_ops; |
f5e54d6e | 187 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 188 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
189 | static const struct inode_operations shmem_inode_operations; |
190 | static const struct inode_operations shmem_dir_inode_operations; | |
191 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 192 | static const struct vm_operations_struct shmem_vm_ops; |
1da177e4 | 193 | |
1da177e4 | 194 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 195 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 196 | |
5b04c689 PE |
197 | static int shmem_reserve_inode(struct super_block *sb) |
198 | { | |
199 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
200 | if (sbinfo->max_inodes) { | |
201 | spin_lock(&sbinfo->stat_lock); | |
202 | if (!sbinfo->free_inodes) { | |
203 | spin_unlock(&sbinfo->stat_lock); | |
204 | return -ENOSPC; | |
205 | } | |
206 | sbinfo->free_inodes--; | |
207 | spin_unlock(&sbinfo->stat_lock); | |
208 | } | |
209 | return 0; | |
210 | } | |
211 | ||
212 | static void shmem_free_inode(struct super_block *sb) | |
213 | { | |
214 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
215 | if (sbinfo->max_inodes) { | |
216 | spin_lock(&sbinfo->stat_lock); | |
217 | sbinfo->free_inodes++; | |
218 | spin_unlock(&sbinfo->stat_lock); | |
219 | } | |
220 | } | |
221 | ||
46711810 | 222 | /** |
41ffe5d5 | 223 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
224 | * @inode: inode to recalc |
225 | * | |
226 | * We have to calculate the free blocks since the mm can drop | |
227 | * undirtied hole pages behind our back. | |
228 | * | |
229 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
230 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
231 | * | |
232 | * It has to be called with the spinlock held. | |
233 | */ | |
234 | static void shmem_recalc_inode(struct inode *inode) | |
235 | { | |
236 | struct shmem_inode_info *info = SHMEM_I(inode); | |
237 | long freed; | |
238 | ||
239 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
240 | if (freed > 0) { | |
54af6042 HD |
241 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
242 | if (sbinfo->max_blocks) | |
243 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 244 | info->alloced -= freed; |
54af6042 | 245 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 246 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
247 | } |
248 | } | |
249 | ||
7a5d0fbb HD |
250 | /* |
251 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
252 | */ | |
253 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
254 | pgoff_t index, void *expected, void *replacement) | |
255 | { | |
256 | void **pslot; | |
6dbaf22c | 257 | void *item; |
7a5d0fbb HD |
258 | |
259 | VM_BUG_ON(!expected); | |
6dbaf22c | 260 | VM_BUG_ON(!replacement); |
7a5d0fbb | 261 | pslot = radix_tree_lookup_slot(&mapping->page_tree, index); |
6dbaf22c JW |
262 | if (!pslot) |
263 | return -ENOENT; | |
264 | item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock); | |
7a5d0fbb HD |
265 | if (item != expected) |
266 | return -ENOENT; | |
6dbaf22c | 267 | radix_tree_replace_slot(pslot, replacement); |
7a5d0fbb HD |
268 | return 0; |
269 | } | |
270 | ||
d1899228 HD |
271 | /* |
272 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
273 | * that an entry was not already brought back from swap by a racing thread. | |
274 | * | |
275 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
276 | * might be reused, and again be SwapCache, using the same swap as before. | |
277 | */ | |
278 | static bool shmem_confirm_swap(struct address_space *mapping, | |
279 | pgoff_t index, swp_entry_t swap) | |
280 | { | |
281 | void *item; | |
282 | ||
283 | rcu_read_lock(); | |
284 | item = radix_tree_lookup(&mapping->page_tree, index); | |
285 | rcu_read_unlock(); | |
286 | return item == swp_to_radix_entry(swap); | |
287 | } | |
288 | ||
46f65ec1 HD |
289 | /* |
290 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
291 | */ | |
292 | static int shmem_add_to_page_cache(struct page *page, | |
293 | struct address_space *mapping, | |
fed400a1 | 294 | pgoff_t index, void *expected) |
46f65ec1 | 295 | { |
b065b432 | 296 | int error; |
46f65ec1 | 297 | |
309381fe SL |
298 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
299 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
46f65ec1 | 300 | |
b065b432 HD |
301 | page_cache_get(page); |
302 | page->mapping = mapping; | |
303 | page->index = index; | |
304 | ||
305 | spin_lock_irq(&mapping->tree_lock); | |
46f65ec1 | 306 | if (!expected) |
b065b432 HD |
307 | error = radix_tree_insert(&mapping->page_tree, index, page); |
308 | else | |
309 | error = shmem_radix_tree_replace(mapping, index, expected, | |
310 | page); | |
46f65ec1 | 311 | if (!error) { |
b065b432 HD |
312 | mapping->nrpages++; |
313 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
314 | __inc_zone_page_state(page, NR_SHMEM); | |
315 | spin_unlock_irq(&mapping->tree_lock); | |
316 | } else { | |
317 | page->mapping = NULL; | |
318 | spin_unlock_irq(&mapping->tree_lock); | |
319 | page_cache_release(page); | |
46f65ec1 | 320 | } |
46f65ec1 HD |
321 | return error; |
322 | } | |
323 | ||
6922c0c7 HD |
324 | /* |
325 | * Like delete_from_page_cache, but substitutes swap for page. | |
326 | */ | |
327 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
328 | { | |
329 | struct address_space *mapping = page->mapping; | |
330 | int error; | |
331 | ||
332 | spin_lock_irq(&mapping->tree_lock); | |
333 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
334 | page->mapping = NULL; | |
335 | mapping->nrpages--; | |
336 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
337 | __dec_zone_page_state(page, NR_SHMEM); | |
338 | spin_unlock_irq(&mapping->tree_lock); | |
339 | page_cache_release(page); | |
340 | BUG_ON(error); | |
341 | } | |
342 | ||
7a5d0fbb HD |
343 | /* |
344 | * Remove swap entry from radix tree, free the swap and its page cache. | |
345 | */ | |
346 | static int shmem_free_swap(struct address_space *mapping, | |
347 | pgoff_t index, void *radswap) | |
348 | { | |
6dbaf22c | 349 | void *old; |
7a5d0fbb HD |
350 | |
351 | spin_lock_irq(&mapping->tree_lock); | |
6dbaf22c | 352 | old = radix_tree_delete_item(&mapping->page_tree, index, radswap); |
7a5d0fbb | 353 | spin_unlock_irq(&mapping->tree_lock); |
6dbaf22c JW |
354 | if (old != radswap) |
355 | return -ENOENT; | |
356 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
357 | return 0; | |
7a5d0fbb HD |
358 | } |
359 | ||
24513264 HD |
360 | /* |
361 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
362 | */ | |
363 | void shmem_unlock_mapping(struct address_space *mapping) | |
364 | { | |
365 | struct pagevec pvec; | |
366 | pgoff_t indices[PAGEVEC_SIZE]; | |
367 | pgoff_t index = 0; | |
368 | ||
369 | pagevec_init(&pvec, 0); | |
370 | /* | |
371 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
372 | */ | |
373 | while (!mapping_unevictable(mapping)) { | |
374 | /* | |
375 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
376 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
377 | */ | |
0cd6144a JW |
378 | pvec.nr = find_get_entries(mapping, index, |
379 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
380 | if (!pvec.nr) |
381 | break; | |
382 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 383 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
384 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
385 | pagevec_release(&pvec); | |
386 | cond_resched(); | |
387 | } | |
7a5d0fbb HD |
388 | } |
389 | ||
390 | /* | |
391 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 392 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 393 | */ |
1635f6a7 HD |
394 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
395 | bool unfalloc) | |
1da177e4 | 396 | { |
285b2c4f | 397 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 398 | struct shmem_inode_info *info = SHMEM_I(inode); |
285b2c4f | 399 | pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
83e4fa9c HD |
400 | pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT; |
401 | unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1); | |
402 | unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); | |
bda97eab | 403 | struct pagevec pvec; |
7a5d0fbb HD |
404 | pgoff_t indices[PAGEVEC_SIZE]; |
405 | long nr_swaps_freed = 0; | |
285b2c4f | 406 | pgoff_t index; |
bda97eab HD |
407 | int i; |
408 | ||
83e4fa9c HD |
409 | if (lend == -1) |
410 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
411 | |
412 | pagevec_init(&pvec, 0); | |
413 | index = start; | |
83e4fa9c | 414 | while (index < end) { |
0cd6144a JW |
415 | pvec.nr = find_get_entries(mapping, index, |
416 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
417 | pvec.pages, indices); | |
7a5d0fbb HD |
418 | if (!pvec.nr) |
419 | break; | |
bda97eab HD |
420 | for (i = 0; i < pagevec_count(&pvec); i++) { |
421 | struct page *page = pvec.pages[i]; | |
422 | ||
7a5d0fbb | 423 | index = indices[i]; |
83e4fa9c | 424 | if (index >= end) |
bda97eab HD |
425 | break; |
426 | ||
7a5d0fbb | 427 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
428 | if (unfalloc) |
429 | continue; | |
7a5d0fbb HD |
430 | nr_swaps_freed += !shmem_free_swap(mapping, |
431 | index, page); | |
bda97eab | 432 | continue; |
7a5d0fbb HD |
433 | } |
434 | ||
435 | if (!trylock_page(page)) | |
bda97eab | 436 | continue; |
1635f6a7 HD |
437 | if (!unfalloc || !PageUptodate(page)) { |
438 | if (page->mapping == mapping) { | |
309381fe | 439 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
440 | truncate_inode_page(mapping, page); |
441 | } | |
bda97eab | 442 | } |
bda97eab HD |
443 | unlock_page(page); |
444 | } | |
0cd6144a | 445 | pagevec_remove_exceptionals(&pvec); |
24513264 | 446 | pagevec_release(&pvec); |
bda97eab HD |
447 | cond_resched(); |
448 | index++; | |
449 | } | |
1da177e4 | 450 | |
83e4fa9c | 451 | if (partial_start) { |
bda97eab HD |
452 | struct page *page = NULL; |
453 | shmem_getpage(inode, start - 1, &page, SGP_READ, NULL); | |
454 | if (page) { | |
83e4fa9c HD |
455 | unsigned int top = PAGE_CACHE_SIZE; |
456 | if (start > end) { | |
457 | top = partial_end; | |
458 | partial_end = 0; | |
459 | } | |
460 | zero_user_segment(page, partial_start, top); | |
461 | set_page_dirty(page); | |
462 | unlock_page(page); | |
463 | page_cache_release(page); | |
464 | } | |
465 | } | |
466 | if (partial_end) { | |
467 | struct page *page = NULL; | |
468 | shmem_getpage(inode, end, &page, SGP_READ, NULL); | |
469 | if (page) { | |
470 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
471 | set_page_dirty(page); |
472 | unlock_page(page); | |
473 | page_cache_release(page); | |
474 | } | |
475 | } | |
83e4fa9c HD |
476 | if (start >= end) |
477 | return; | |
bda97eab HD |
478 | |
479 | index = start; | |
b1a36650 | 480 | while (index < end) { |
bda97eab | 481 | cond_resched(); |
0cd6144a JW |
482 | |
483 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 484 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 485 | pvec.pages, indices); |
7a5d0fbb | 486 | if (!pvec.nr) { |
b1a36650 HD |
487 | /* If all gone or hole-punch or unfalloc, we're done */ |
488 | if (index == start || end != -1) | |
bda97eab | 489 | break; |
b1a36650 | 490 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
491 | index = start; |
492 | continue; | |
493 | } | |
bda97eab HD |
494 | for (i = 0; i < pagevec_count(&pvec); i++) { |
495 | struct page *page = pvec.pages[i]; | |
496 | ||
7a5d0fbb | 497 | index = indices[i]; |
83e4fa9c | 498 | if (index >= end) |
bda97eab HD |
499 | break; |
500 | ||
7a5d0fbb | 501 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
502 | if (unfalloc) |
503 | continue; | |
b1a36650 HD |
504 | if (shmem_free_swap(mapping, index, page)) { |
505 | /* Swap was replaced by page: retry */ | |
506 | index--; | |
507 | break; | |
508 | } | |
509 | nr_swaps_freed++; | |
7a5d0fbb HD |
510 | continue; |
511 | } | |
512 | ||
bda97eab | 513 | lock_page(page); |
1635f6a7 HD |
514 | if (!unfalloc || !PageUptodate(page)) { |
515 | if (page->mapping == mapping) { | |
309381fe | 516 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 517 | truncate_inode_page(mapping, page); |
b1a36650 HD |
518 | } else { |
519 | /* Page was replaced by swap: retry */ | |
520 | unlock_page(page); | |
521 | index--; | |
522 | break; | |
1635f6a7 | 523 | } |
7a5d0fbb | 524 | } |
bda97eab HD |
525 | unlock_page(page); |
526 | } | |
0cd6144a | 527 | pagevec_remove_exceptionals(&pvec); |
24513264 | 528 | pagevec_release(&pvec); |
bda97eab HD |
529 | index++; |
530 | } | |
94c1e62d | 531 | |
1da177e4 | 532 | spin_lock(&info->lock); |
7a5d0fbb | 533 | info->swapped -= nr_swaps_freed; |
1da177e4 LT |
534 | shmem_recalc_inode(inode); |
535 | spin_unlock(&info->lock); | |
1635f6a7 | 536 | } |
1da177e4 | 537 | |
1635f6a7 HD |
538 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
539 | { | |
540 | shmem_undo_range(inode, lstart, lend, false); | |
285b2c4f | 541 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1da177e4 | 542 | } |
94c1e62d | 543 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 544 | |
44a30220 YZ |
545 | static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry, |
546 | struct kstat *stat) | |
547 | { | |
548 | struct inode *inode = dentry->d_inode; | |
549 | struct shmem_inode_info *info = SHMEM_I(inode); | |
550 | ||
d0424c42 HD |
551 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
552 | spin_lock(&info->lock); | |
553 | shmem_recalc_inode(inode); | |
554 | spin_unlock(&info->lock); | |
555 | } | |
44a30220 | 556 | generic_fillattr(inode, stat); |
44a30220 YZ |
557 | return 0; |
558 | } | |
559 | ||
94c1e62d | 560 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 561 | { |
75c3cfa8 | 562 | struct inode *inode = d_inode(dentry); |
40e041a2 | 563 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 LT |
564 | int error; |
565 | ||
db78b877 CH |
566 | error = inode_change_ok(inode, attr); |
567 | if (error) | |
568 | return error; | |
569 | ||
94c1e62d HD |
570 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
571 | loff_t oldsize = inode->i_size; | |
572 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 573 | |
40e041a2 DH |
574 | /* protected by i_mutex */ |
575 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
576 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
577 | return -EPERM; | |
578 | ||
94c1e62d | 579 | if (newsize != oldsize) { |
77142517 KK |
580 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
581 | oldsize, newsize); | |
582 | if (error) | |
583 | return error; | |
94c1e62d HD |
584 | i_size_write(inode, newsize); |
585 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
586 | } | |
afa2db2f | 587 | if (newsize <= oldsize) { |
94c1e62d | 588 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
589 | if (oldsize > holebegin) |
590 | unmap_mapping_range(inode->i_mapping, | |
591 | holebegin, 0, 1); | |
592 | if (info->alloced) | |
593 | shmem_truncate_range(inode, | |
594 | newsize, (loff_t)-1); | |
94c1e62d | 595 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
596 | if (oldsize > holebegin) |
597 | unmap_mapping_range(inode->i_mapping, | |
598 | holebegin, 0, 1); | |
94c1e62d | 599 | } |
1da177e4 LT |
600 | } |
601 | ||
db78b877 | 602 | setattr_copy(inode, attr); |
db78b877 | 603 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 604 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
605 | return error; |
606 | } | |
607 | ||
1f895f75 | 608 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 609 | { |
1da177e4 LT |
610 | struct shmem_inode_info *info = SHMEM_I(inode); |
611 | ||
3889e6e7 | 612 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
613 | shmem_unacct_size(info->flags, inode->i_size); |
614 | inode->i_size = 0; | |
3889e6e7 | 615 | shmem_truncate_range(inode, 0, (loff_t)-1); |
1da177e4 | 616 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 617 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 618 | list_del_init(&info->swaplist); |
cb5f7b9a | 619 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 620 | } |
69f07ec9 HD |
621 | } else |
622 | kfree(info->symlink); | |
b09e0fa4 | 623 | |
38f38657 | 624 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 625 | WARN_ON(inode->i_blocks); |
5b04c689 | 626 | shmem_free_inode(inode->i_sb); |
dbd5768f | 627 | clear_inode(inode); |
1da177e4 LT |
628 | } |
629 | ||
46f65ec1 HD |
630 | /* |
631 | * If swap found in inode, free it and move page from swapcache to filecache. | |
632 | */ | |
41ffe5d5 | 633 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 634 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 635 | { |
285b2c4f | 636 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 637 | void *radswap; |
41ffe5d5 | 638 | pgoff_t index; |
bde05d1c HD |
639 | gfp_t gfp; |
640 | int error = 0; | |
1da177e4 | 641 | |
46f65ec1 | 642 | radswap = swp_to_radix_entry(swap); |
e504f3fd | 643 | index = radix_tree_locate_item(&mapping->page_tree, radswap); |
46f65ec1 | 644 | if (index == -1) |
00501b53 | 645 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 646 | |
1b1b32f2 HD |
647 | /* |
648 | * Move _head_ to start search for next from here. | |
1f895f75 | 649 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 650 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 651 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
652 | */ |
653 | if (shmem_swaplist.next != &info->swaplist) | |
654 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 655 | |
bde05d1c HD |
656 | gfp = mapping_gfp_mask(mapping); |
657 | if (shmem_should_replace_page(*pagep, gfp)) { | |
658 | mutex_unlock(&shmem_swaplist_mutex); | |
659 | error = shmem_replace_page(pagep, gfp, info, index); | |
660 | mutex_lock(&shmem_swaplist_mutex); | |
661 | /* | |
662 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
663 | * allocation, but the inode might have been freed while we |
664 | * dropped it: although a racing shmem_evict_inode() cannot | |
665 | * complete without emptying the radix_tree, our page lock | |
666 | * on this swapcache page is not enough to prevent that - | |
667 | * free_swap_and_cache() of our swap entry will only | |
668 | * trylock_page(), removing swap from radix_tree whatever. | |
669 | * | |
670 | * We must not proceed to shmem_add_to_page_cache() if the | |
671 | * inode has been freed, but of course we cannot rely on | |
672 | * inode or mapping or info to check that. However, we can | |
673 | * safely check if our swap entry is still in use (and here | |
674 | * it can't have got reused for another page): if it's still | |
675 | * in use, then the inode cannot have been freed yet, and we | |
676 | * can safely proceed (if it's no longer in use, that tells | |
677 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
678 | */ |
679 | if (!page_swapcount(*pagep)) | |
680 | error = -ENOENT; | |
681 | } | |
682 | ||
d13d1443 | 683 | /* |
778dd893 HD |
684 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
685 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
686 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 687 | */ |
bde05d1c HD |
688 | if (!error) |
689 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 690 | radswap); |
48f170fb | 691 | if (error != -ENOMEM) { |
46f65ec1 HD |
692 | /* |
693 | * Truncation and eviction use free_swap_and_cache(), which | |
694 | * only does trylock page: if we raced, best clean up here. | |
695 | */ | |
bde05d1c HD |
696 | delete_from_swap_cache(*pagep); |
697 | set_page_dirty(*pagep); | |
46f65ec1 HD |
698 | if (!error) { |
699 | spin_lock(&info->lock); | |
700 | info->swapped--; | |
701 | spin_unlock(&info->lock); | |
702 | swap_free(swap); | |
703 | } | |
1da177e4 | 704 | } |
2e0e26c7 | 705 | return error; |
1da177e4 LT |
706 | } |
707 | ||
708 | /* | |
46f65ec1 | 709 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 710 | */ |
41ffe5d5 | 711 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 712 | { |
41ffe5d5 | 713 | struct list_head *this, *next; |
1da177e4 | 714 | struct shmem_inode_info *info; |
00501b53 | 715 | struct mem_cgroup *memcg; |
bde05d1c HD |
716 | int error = 0; |
717 | ||
718 | /* | |
719 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 720 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 721 | */ |
0142ef6c | 722 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 723 | goto out; |
778dd893 HD |
724 | |
725 | /* | |
726 | * Charge page using GFP_KERNEL while we can wait, before taking | |
727 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
728 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 729 | */ |
00501b53 | 730 | error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg); |
778dd893 HD |
731 | if (error) |
732 | goto out; | |
46f65ec1 | 733 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 734 | error = -EAGAIN; |
1da177e4 | 735 | |
cb5f7b9a | 736 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
737 | list_for_each_safe(this, next, &shmem_swaplist) { |
738 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 739 | if (info->swapped) |
00501b53 | 740 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
741 | else |
742 | list_del_init(&info->swaplist); | |
cb5f7b9a | 743 | cond_resched(); |
00501b53 | 744 | if (error != -EAGAIN) |
778dd893 | 745 | break; |
00501b53 | 746 | /* found nothing in this: move on to search the next */ |
1da177e4 | 747 | } |
cb5f7b9a | 748 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 749 | |
00501b53 JW |
750 | if (error) { |
751 | if (error != -ENOMEM) | |
752 | error = 0; | |
753 | mem_cgroup_cancel_charge(page, memcg); | |
754 | } else | |
755 | mem_cgroup_commit_charge(page, memcg, true); | |
778dd893 | 756 | out: |
aaa46865 HD |
757 | unlock_page(page); |
758 | page_cache_release(page); | |
778dd893 | 759 | return error; |
1da177e4 LT |
760 | } |
761 | ||
762 | /* | |
763 | * Move the page from the page cache to the swap cache. | |
764 | */ | |
765 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
766 | { | |
767 | struct shmem_inode_info *info; | |
1da177e4 | 768 | struct address_space *mapping; |
1da177e4 | 769 | struct inode *inode; |
6922c0c7 HD |
770 | swp_entry_t swap; |
771 | pgoff_t index; | |
1da177e4 LT |
772 | |
773 | BUG_ON(!PageLocked(page)); | |
1da177e4 LT |
774 | mapping = page->mapping; |
775 | index = page->index; | |
776 | inode = mapping->host; | |
777 | info = SHMEM_I(inode); | |
778 | if (info->flags & VM_LOCKED) | |
779 | goto redirty; | |
d9fe526a | 780 | if (!total_swap_pages) |
1da177e4 LT |
781 | goto redirty; |
782 | ||
d9fe526a | 783 | /* |
97b713ba CH |
784 | * Our capabilities prevent regular writeback or sync from ever calling |
785 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
786 | * its underlying filesystem, in which case tmpfs should write out to | |
787 | * swap only in response to memory pressure, and not for the writeback | |
788 | * threads or sync. | |
d9fe526a | 789 | */ |
48f170fb HD |
790 | if (!wbc->for_reclaim) { |
791 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
792 | goto redirty; | |
793 | } | |
1635f6a7 HD |
794 | |
795 | /* | |
796 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
797 | * value into swapfile.c, the only way we can correctly account for a | |
798 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
799 | * |
800 | * That's okay for a page already fallocated earlier, but if we have | |
801 | * not yet completed the fallocation, then (a) we want to keep track | |
802 | * of this page in case we have to undo it, and (b) it may not be a | |
803 | * good idea to continue anyway, once we're pushing into swap. So | |
804 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
805 | */ |
806 | if (!PageUptodate(page)) { | |
1aac1400 HD |
807 | if (inode->i_private) { |
808 | struct shmem_falloc *shmem_falloc; | |
809 | spin_lock(&inode->i_lock); | |
810 | shmem_falloc = inode->i_private; | |
811 | if (shmem_falloc && | |
8e205f77 | 812 | !shmem_falloc->waitq && |
1aac1400 HD |
813 | index >= shmem_falloc->start && |
814 | index < shmem_falloc->next) | |
815 | shmem_falloc->nr_unswapped++; | |
816 | else | |
817 | shmem_falloc = NULL; | |
818 | spin_unlock(&inode->i_lock); | |
819 | if (shmem_falloc) | |
820 | goto redirty; | |
821 | } | |
1635f6a7 HD |
822 | clear_highpage(page); |
823 | flush_dcache_page(page); | |
824 | SetPageUptodate(page); | |
825 | } | |
826 | ||
48f170fb HD |
827 | swap = get_swap_page(); |
828 | if (!swap.val) | |
829 | goto redirty; | |
d9fe526a | 830 | |
b1dea800 HD |
831 | /* |
832 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
833 | * if it's not already there. Do it now before the page is |
834 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 835 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
836 | * we've incremented swapped, because shmem_unuse_inode() will |
837 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 838 | */ |
48f170fb HD |
839 | mutex_lock(&shmem_swaplist_mutex); |
840 | if (list_empty(&info->swaplist)) | |
841 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 842 | |
48f170fb | 843 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
aaa46865 | 844 | swap_shmem_alloc(swap); |
6922c0c7 HD |
845 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); |
846 | ||
847 | spin_lock(&info->lock); | |
848 | info->swapped++; | |
849 | shmem_recalc_inode(inode); | |
826267cf | 850 | spin_unlock(&info->lock); |
6922c0c7 HD |
851 | |
852 | mutex_unlock(&shmem_swaplist_mutex); | |
d9fe526a | 853 | BUG_ON(page_mapped(page)); |
9fab5619 | 854 | swap_writepage(page, wbc); |
1da177e4 LT |
855 | return 0; |
856 | } | |
857 | ||
6922c0c7 | 858 | mutex_unlock(&shmem_swaplist_mutex); |
0a31bc97 | 859 | swapcache_free(swap); |
1da177e4 LT |
860 | redirty: |
861 | set_page_dirty(page); | |
d9fe526a HD |
862 | if (wbc->for_reclaim) |
863 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
864 | unlock_page(page); | |
865 | return 0; | |
1da177e4 LT |
866 | } |
867 | ||
868 | #ifdef CONFIG_NUMA | |
680d794b | 869 | #ifdef CONFIG_TMPFS |
71fe804b | 870 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 871 | { |
095f1fc4 | 872 | char buffer[64]; |
680d794b | 873 | |
71fe804b | 874 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 875 | return; /* show nothing */ |
680d794b | 876 | |
a7a88b23 | 877 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
878 | |
879 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 880 | } |
71fe804b LS |
881 | |
882 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
883 | { | |
884 | struct mempolicy *mpol = NULL; | |
885 | if (sbinfo->mpol) { | |
886 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
887 | mpol = sbinfo->mpol; | |
888 | mpol_get(mpol); | |
889 | spin_unlock(&sbinfo->stat_lock); | |
890 | } | |
891 | return mpol; | |
892 | } | |
680d794b | 893 | #endif /* CONFIG_TMPFS */ |
894 | ||
41ffe5d5 HD |
895 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
896 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 897 | { |
1da177e4 | 898 | struct vm_area_struct pvma; |
18a2f371 | 899 | struct page *page; |
52cd3b07 | 900 | |
1da177e4 | 901 | /* Create a pseudo vma that just contains the policy */ |
c4cc6d07 | 902 | pvma.vm_start = 0; |
09c231cb NZ |
903 | /* Bias interleave by inode number to distribute better across nodes */ |
904 | pvma.vm_pgoff = index + info->vfs_inode.i_ino; | |
c4cc6d07 | 905 | pvma.vm_ops = NULL; |
18a2f371 MG |
906 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
907 | ||
908 | page = swapin_readahead(swap, gfp, &pvma, 0); | |
909 | ||
910 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
911 | mpol_cond_put(pvma.vm_policy); | |
912 | ||
913 | return page; | |
1da177e4 LT |
914 | } |
915 | ||
02098fea | 916 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 917 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
918 | { |
919 | struct vm_area_struct pvma; | |
18a2f371 | 920 | struct page *page; |
1da177e4 | 921 | |
c4cc6d07 HD |
922 | /* Create a pseudo vma that just contains the policy */ |
923 | pvma.vm_start = 0; | |
09c231cb NZ |
924 | /* Bias interleave by inode number to distribute better across nodes */ |
925 | pvma.vm_pgoff = index + info->vfs_inode.i_ino; | |
c4cc6d07 | 926 | pvma.vm_ops = NULL; |
41ffe5d5 | 927 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
52cd3b07 | 928 | |
18a2f371 MG |
929 | page = alloc_page_vma(gfp, &pvma, 0); |
930 | ||
931 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
932 | mpol_cond_put(pvma.vm_policy); | |
933 | ||
934 | return page; | |
1da177e4 | 935 | } |
680d794b | 936 | #else /* !CONFIG_NUMA */ |
937 | #ifdef CONFIG_TMPFS | |
41ffe5d5 | 938 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 939 | { |
940 | } | |
941 | #endif /* CONFIG_TMPFS */ | |
942 | ||
41ffe5d5 HD |
943 | static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
944 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 945 | { |
41ffe5d5 | 946 | return swapin_readahead(swap, gfp, NULL, 0); |
1da177e4 LT |
947 | } |
948 | ||
02098fea | 949 | static inline struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 950 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 | 951 | { |
e84e2e13 | 952 | return alloc_page(gfp); |
1da177e4 | 953 | } |
680d794b | 954 | #endif /* CONFIG_NUMA */ |
1da177e4 | 955 | |
71fe804b LS |
956 | #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) |
957 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
958 | { | |
959 | return NULL; | |
960 | } | |
961 | #endif | |
962 | ||
bde05d1c HD |
963 | /* |
964 | * When a page is moved from swapcache to shmem filecache (either by the | |
965 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
966 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
967 | * ignorance of the mapping it belongs to. If that mapping has special | |
968 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
969 | * we may need to copy to a suitable page before moving to filecache. | |
970 | * | |
971 | * In a future release, this may well be extended to respect cpuset and | |
972 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
973 | * but for now it is a simple matter of zone. | |
974 | */ | |
975 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
976 | { | |
977 | return page_zonenum(page) > gfp_zone(gfp); | |
978 | } | |
979 | ||
980 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
981 | struct shmem_inode_info *info, pgoff_t index) | |
982 | { | |
983 | struct page *oldpage, *newpage; | |
984 | struct address_space *swap_mapping; | |
985 | pgoff_t swap_index; | |
986 | int error; | |
987 | ||
988 | oldpage = *pagep; | |
989 | swap_index = page_private(oldpage); | |
990 | swap_mapping = page_mapping(oldpage); | |
991 | ||
992 | /* | |
993 | * We have arrived here because our zones are constrained, so don't | |
994 | * limit chance of success by further cpuset and node constraints. | |
995 | */ | |
996 | gfp &= ~GFP_CONSTRAINT_MASK; | |
997 | newpage = shmem_alloc_page(gfp, info, index); | |
998 | if (!newpage) | |
999 | return -ENOMEM; | |
bde05d1c | 1000 | |
bde05d1c HD |
1001 | page_cache_get(newpage); |
1002 | copy_highpage(newpage, oldpage); | |
0142ef6c | 1003 | flush_dcache_page(newpage); |
bde05d1c | 1004 | |
bde05d1c | 1005 | __set_page_locked(newpage); |
bde05d1c | 1006 | SetPageUptodate(newpage); |
bde05d1c | 1007 | SetPageSwapBacked(newpage); |
bde05d1c | 1008 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1009 | SetPageSwapCache(newpage); |
1010 | ||
1011 | /* | |
1012 | * Our caller will very soon move newpage out of swapcache, but it's | |
1013 | * a nice clean interface for us to replace oldpage by newpage there. | |
1014 | */ | |
1015 | spin_lock_irq(&swap_mapping->tree_lock); | |
1016 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
1017 | newpage); | |
0142ef6c HD |
1018 | if (!error) { |
1019 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
1020 | __dec_zone_page_state(oldpage, NR_FILE_PAGES); | |
1021 | } | |
bde05d1c | 1022 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1023 | |
0142ef6c HD |
1024 | if (unlikely(error)) { |
1025 | /* | |
1026 | * Is this possible? I think not, now that our callers check | |
1027 | * both PageSwapCache and page_private after getting page lock; | |
1028 | * but be defensive. Reverse old to newpage for clear and free. | |
1029 | */ | |
1030 | oldpage = newpage; | |
1031 | } else { | |
45637bab | 1032 | mem_cgroup_replace_page(oldpage, newpage); |
0142ef6c HD |
1033 | lru_cache_add_anon(newpage); |
1034 | *pagep = newpage; | |
1035 | } | |
bde05d1c HD |
1036 | |
1037 | ClearPageSwapCache(oldpage); | |
1038 | set_page_private(oldpage, 0); | |
1039 | ||
1040 | unlock_page(oldpage); | |
1041 | page_cache_release(oldpage); | |
1042 | page_cache_release(oldpage); | |
0142ef6c | 1043 | return error; |
bde05d1c HD |
1044 | } |
1045 | ||
1da177e4 | 1046 | /* |
68da9f05 | 1047 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1048 | * |
1049 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1050 | * vm. If we swap it in we mark it dirty since we also free the swap | |
1051 | * entry since a page cannot live in both the swap and page cache | |
1052 | */ | |
41ffe5d5 | 1053 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
68da9f05 | 1054 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type) |
1da177e4 LT |
1055 | { |
1056 | struct address_space *mapping = inode->i_mapping; | |
54af6042 | 1057 | struct shmem_inode_info *info; |
1da177e4 | 1058 | struct shmem_sb_info *sbinfo; |
00501b53 | 1059 | struct mem_cgroup *memcg; |
27ab7006 | 1060 | struct page *page; |
1da177e4 LT |
1061 | swp_entry_t swap; |
1062 | int error; | |
54af6042 | 1063 | int once = 0; |
1635f6a7 | 1064 | int alloced = 0; |
1da177e4 | 1065 | |
41ffe5d5 | 1066 | if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT)) |
1da177e4 | 1067 | return -EFBIG; |
1da177e4 | 1068 | repeat: |
54af6042 | 1069 | swap.val = 0; |
0cd6144a | 1070 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1071 | if (radix_tree_exceptional_entry(page)) { |
1072 | swap = radix_to_swp_entry(page); | |
1073 | page = NULL; | |
1074 | } | |
1075 | ||
1635f6a7 | 1076 | if (sgp != SGP_WRITE && sgp != SGP_FALLOC && |
54af6042 HD |
1077 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { |
1078 | error = -EINVAL; | |
1079 | goto failed; | |
1080 | } | |
1081 | ||
66d2f4d2 HD |
1082 | if (page && sgp == SGP_WRITE) |
1083 | mark_page_accessed(page); | |
1084 | ||
1635f6a7 HD |
1085 | /* fallocated page? */ |
1086 | if (page && !PageUptodate(page)) { | |
1087 | if (sgp != SGP_READ) | |
1088 | goto clear; | |
1089 | unlock_page(page); | |
1090 | page_cache_release(page); | |
1091 | page = NULL; | |
1092 | } | |
54af6042 | 1093 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1094 | *pagep = page; |
1095 | return 0; | |
27ab7006 HD |
1096 | } |
1097 | ||
1098 | /* | |
54af6042 HD |
1099 | * Fast cache lookup did not find it: |
1100 | * bring it back from swap or allocate. | |
27ab7006 | 1101 | */ |
54af6042 HD |
1102 | info = SHMEM_I(inode); |
1103 | sbinfo = SHMEM_SB(inode->i_sb); | |
1da177e4 | 1104 | |
1da177e4 LT |
1105 | if (swap.val) { |
1106 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1107 | page = lookup_swap_cache(swap); |
1108 | if (!page) { | |
1da177e4 | 1109 | /* here we actually do the io */ |
68da9f05 HD |
1110 | if (fault_type) |
1111 | *fault_type |= VM_FAULT_MAJOR; | |
41ffe5d5 | 1112 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1113 | if (!page) { |
54af6042 HD |
1114 | error = -ENOMEM; |
1115 | goto failed; | |
1da177e4 | 1116 | } |
1da177e4 LT |
1117 | } |
1118 | ||
1119 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1120 | lock_page(page); |
0142ef6c | 1121 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1122 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1123 | error = -EEXIST; /* try again */ |
d1899228 | 1124 | goto unlock; |
bde05d1c | 1125 | } |
27ab7006 | 1126 | if (!PageUptodate(page)) { |
1da177e4 | 1127 | error = -EIO; |
54af6042 | 1128 | goto failed; |
1da177e4 | 1129 | } |
54af6042 HD |
1130 | wait_on_page_writeback(page); |
1131 | ||
bde05d1c HD |
1132 | if (shmem_should_replace_page(page, gfp)) { |
1133 | error = shmem_replace_page(&page, gfp, info, index); | |
1134 | if (error) | |
1135 | goto failed; | |
1da177e4 | 1136 | } |
27ab7006 | 1137 | |
00501b53 | 1138 | error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg); |
d1899228 | 1139 | if (!error) { |
aa3b1895 | 1140 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1141 | swp_to_radix_entry(swap)); |
215c02bc HD |
1142 | /* |
1143 | * We already confirmed swap under page lock, and make | |
1144 | * no memory allocation here, so usually no possibility | |
1145 | * of error; but free_swap_and_cache() only trylocks a | |
1146 | * page, so it is just possible that the entry has been | |
1147 | * truncated or holepunched since swap was confirmed. | |
1148 | * shmem_undo_range() will have done some of the | |
1149 | * unaccounting, now delete_from_swap_cache() will do | |
93aa7d95 | 1150 | * the rest. |
215c02bc HD |
1151 | * Reset swap.val? No, leave it so "failed" goes back to |
1152 | * "repeat": reading a hole and writing should succeed. | |
1153 | */ | |
00501b53 JW |
1154 | if (error) { |
1155 | mem_cgroup_cancel_charge(page, memcg); | |
215c02bc | 1156 | delete_from_swap_cache(page); |
00501b53 | 1157 | } |
d1899228 | 1158 | } |
54af6042 HD |
1159 | if (error) |
1160 | goto failed; | |
1161 | ||
00501b53 JW |
1162 | mem_cgroup_commit_charge(page, memcg, true); |
1163 | ||
54af6042 | 1164 | spin_lock(&info->lock); |
285b2c4f | 1165 | info->swapped--; |
54af6042 | 1166 | shmem_recalc_inode(inode); |
27ab7006 | 1167 | spin_unlock(&info->lock); |
54af6042 | 1168 | |
66d2f4d2 HD |
1169 | if (sgp == SGP_WRITE) |
1170 | mark_page_accessed(page); | |
1171 | ||
54af6042 | 1172 | delete_from_swap_cache(page); |
27ab7006 HD |
1173 | set_page_dirty(page); |
1174 | swap_free(swap); | |
1175 | ||
54af6042 HD |
1176 | } else { |
1177 | if (shmem_acct_block(info->flags)) { | |
1178 | error = -ENOSPC; | |
1179 | goto failed; | |
1da177e4 | 1180 | } |
0edd73b3 | 1181 | if (sbinfo->max_blocks) { |
fc5da22a | 1182 | if (percpu_counter_compare(&sbinfo->used_blocks, |
54af6042 HD |
1183 | sbinfo->max_blocks) >= 0) { |
1184 | error = -ENOSPC; | |
1185 | goto unacct; | |
1186 | } | |
7e496299 | 1187 | percpu_counter_inc(&sbinfo->used_blocks); |
54af6042 | 1188 | } |
1da177e4 | 1189 | |
54af6042 HD |
1190 | page = shmem_alloc_page(gfp, info, index); |
1191 | if (!page) { | |
1192 | error = -ENOMEM; | |
1193 | goto decused; | |
1da177e4 LT |
1194 | } |
1195 | ||
07a42788 | 1196 | __SetPageSwapBacked(page); |
54af6042 | 1197 | __set_page_locked(page); |
66d2f4d2 | 1198 | if (sgp == SGP_WRITE) |
eb39d618 | 1199 | __SetPageReferenced(page); |
66d2f4d2 | 1200 | |
00501b53 | 1201 | error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg); |
54af6042 HD |
1202 | if (error) |
1203 | goto decused; | |
5e4c0d97 | 1204 | error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK); |
b065b432 HD |
1205 | if (!error) { |
1206 | error = shmem_add_to_page_cache(page, mapping, index, | |
fed400a1 | 1207 | NULL); |
b065b432 HD |
1208 | radix_tree_preload_end(); |
1209 | } | |
1210 | if (error) { | |
00501b53 | 1211 | mem_cgroup_cancel_charge(page, memcg); |
b065b432 HD |
1212 | goto decused; |
1213 | } | |
00501b53 | 1214 | mem_cgroup_commit_charge(page, memcg, false); |
54af6042 HD |
1215 | lru_cache_add_anon(page); |
1216 | ||
1217 | spin_lock(&info->lock); | |
1da177e4 | 1218 | info->alloced++; |
54af6042 HD |
1219 | inode->i_blocks += BLOCKS_PER_PAGE; |
1220 | shmem_recalc_inode(inode); | |
1da177e4 | 1221 | spin_unlock(&info->lock); |
1635f6a7 | 1222 | alloced = true; |
54af6042 | 1223 | |
ec9516fb | 1224 | /* |
1635f6a7 HD |
1225 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1226 | */ | |
1227 | if (sgp == SGP_FALLOC) | |
1228 | sgp = SGP_WRITE; | |
1229 | clear: | |
1230 | /* | |
1231 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1232 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1233 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb HD |
1234 | */ |
1235 | if (sgp != SGP_WRITE) { | |
1236 | clear_highpage(page); | |
1237 | flush_dcache_page(page); | |
1238 | SetPageUptodate(page); | |
1239 | } | |
a0ee5ec5 | 1240 | if (sgp == SGP_DIRTY) |
27ab7006 | 1241 | set_page_dirty(page); |
1da177e4 | 1242 | } |
bde05d1c | 1243 | |
54af6042 | 1244 | /* Perhaps the file has been truncated since we checked */ |
1635f6a7 | 1245 | if (sgp != SGP_WRITE && sgp != SGP_FALLOC && |
54af6042 HD |
1246 | ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { |
1247 | error = -EINVAL; | |
1635f6a7 HD |
1248 | if (alloced) |
1249 | goto trunc; | |
1250 | else | |
1251 | goto failed; | |
e83c32e8 | 1252 | } |
54af6042 HD |
1253 | *pagep = page; |
1254 | return 0; | |
1da177e4 | 1255 | |
59a16ead | 1256 | /* |
54af6042 | 1257 | * Error recovery. |
59a16ead | 1258 | */ |
54af6042 | 1259 | trunc: |
1635f6a7 | 1260 | info = SHMEM_I(inode); |
54af6042 HD |
1261 | ClearPageDirty(page); |
1262 | delete_from_page_cache(page); | |
1263 | spin_lock(&info->lock); | |
1264 | info->alloced--; | |
1265 | inode->i_blocks -= BLOCKS_PER_PAGE; | |
59a16ead | 1266 | spin_unlock(&info->lock); |
54af6042 | 1267 | decused: |
1635f6a7 | 1268 | sbinfo = SHMEM_SB(inode->i_sb); |
54af6042 HD |
1269 | if (sbinfo->max_blocks) |
1270 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
1271 | unacct: | |
1272 | shmem_unacct_blocks(info->flags, 1); | |
1273 | failed: | |
d1899228 HD |
1274 | if (swap.val && error != -EINVAL && |
1275 | !shmem_confirm_swap(mapping, index, swap)) | |
1276 | error = -EEXIST; | |
1277 | unlock: | |
27ab7006 | 1278 | if (page) { |
54af6042 | 1279 | unlock_page(page); |
27ab7006 | 1280 | page_cache_release(page); |
54af6042 HD |
1281 | } |
1282 | if (error == -ENOSPC && !once++) { | |
1283 | info = SHMEM_I(inode); | |
1284 | spin_lock(&info->lock); | |
1285 | shmem_recalc_inode(inode); | |
1286 | spin_unlock(&info->lock); | |
27ab7006 | 1287 | goto repeat; |
ff36b801 | 1288 | } |
d1899228 | 1289 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1290 | goto repeat; |
1291 | return error; | |
1da177e4 LT |
1292 | } |
1293 | ||
d0217ac0 | 1294 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1295 | { |
496ad9aa | 1296 | struct inode *inode = file_inode(vma->vm_file); |
1da177e4 | 1297 | int error; |
68da9f05 | 1298 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1299 | |
f00cdc6d HD |
1300 | /* |
1301 | * Trinity finds that probing a hole which tmpfs is punching can | |
1302 | * prevent the hole-punch from ever completing: which in turn | |
1303 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1304 | * faulting pages into the hole while it's being punched. Although |
1305 | * shmem_undo_range() does remove the additions, it may be unable to | |
1306 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1307 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1308 | * | |
1309 | * It does not matter if we sometimes reach this check just before the | |
1310 | * hole-punch begins, so that one fault then races with the punch: | |
1311 | * we just need to make racing faults a rare case. | |
1312 | * | |
1313 | * The implementation below would be much simpler if we just used a | |
1314 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1315 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1316 | */ |
1317 | if (unlikely(inode->i_private)) { | |
1318 | struct shmem_falloc *shmem_falloc; | |
1319 | ||
1320 | spin_lock(&inode->i_lock); | |
1321 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1322 | if (shmem_falloc && |
1323 | shmem_falloc->waitq && | |
1324 | vmf->pgoff >= shmem_falloc->start && | |
1325 | vmf->pgoff < shmem_falloc->next) { | |
1326 | wait_queue_head_t *shmem_falloc_waitq; | |
1327 | DEFINE_WAIT(shmem_fault_wait); | |
1328 | ||
1329 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1330 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1331 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1332 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1333 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1334 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1335 | } |
8e205f77 HD |
1336 | |
1337 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1338 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1339 | TASK_UNINTERRUPTIBLE); | |
1340 | spin_unlock(&inode->i_lock); | |
1341 | schedule(); | |
1342 | ||
1343 | /* | |
1344 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1345 | * stack of the hole-punching task: shmem_falloc_waitq | |
1346 | * is usually invalid by the time we reach here, but | |
1347 | * finish_wait() does not dereference it in that case; | |
1348 | * though i_lock needed lest racing with wake_up_all(). | |
1349 | */ | |
1350 | spin_lock(&inode->i_lock); | |
1351 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1352 | spin_unlock(&inode->i_lock); | |
1353 | return ret; | |
f00cdc6d | 1354 | } |
8e205f77 | 1355 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
1356 | } |
1357 | ||
27d54b39 | 1358 | error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); |
d0217ac0 NP |
1359 | if (error) |
1360 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1361 | |
456f998e YH |
1362 | if (ret & VM_FAULT_MAJOR) { |
1363 | count_vm_event(PGMAJFAULT); | |
1364 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); | |
1365 | } | |
68da9f05 | 1366 | return ret; |
1da177e4 LT |
1367 | } |
1368 | ||
1da177e4 | 1369 | #ifdef CONFIG_NUMA |
41ffe5d5 | 1370 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 1371 | { |
496ad9aa | 1372 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1373 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
1374 | } |
1375 | ||
d8dc74f2 AB |
1376 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
1377 | unsigned long addr) | |
1da177e4 | 1378 | { |
496ad9aa | 1379 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1380 | pgoff_t index; |
1da177e4 | 1381 | |
41ffe5d5 HD |
1382 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
1383 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
1384 | } |
1385 | #endif | |
1386 | ||
1387 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
1388 | { | |
496ad9aa | 1389 | struct inode *inode = file_inode(file); |
1da177e4 LT |
1390 | struct shmem_inode_info *info = SHMEM_I(inode); |
1391 | int retval = -ENOMEM; | |
1392 | ||
1393 | spin_lock(&info->lock); | |
1394 | if (lock && !(info->flags & VM_LOCKED)) { | |
1395 | if (!user_shm_lock(inode->i_size, user)) | |
1396 | goto out_nomem; | |
1397 | info->flags |= VM_LOCKED; | |
89e004ea | 1398 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
1399 | } |
1400 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
1401 | user_shm_unlock(inode->i_size, user); | |
1402 | info->flags &= ~VM_LOCKED; | |
89e004ea | 1403 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
1404 | } |
1405 | retval = 0; | |
89e004ea | 1406 | |
1da177e4 LT |
1407 | out_nomem: |
1408 | spin_unlock(&info->lock); | |
1409 | return retval; | |
1410 | } | |
1411 | ||
9b83a6a8 | 1412 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
1413 | { |
1414 | file_accessed(file); | |
1415 | vma->vm_ops = &shmem_vm_ops; | |
1416 | return 0; | |
1417 | } | |
1418 | ||
454abafe | 1419 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 1420 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
1421 | { |
1422 | struct inode *inode; | |
1423 | struct shmem_inode_info *info; | |
1424 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
1425 | ||
5b04c689 PE |
1426 | if (shmem_reserve_inode(sb)) |
1427 | return NULL; | |
1da177e4 LT |
1428 | |
1429 | inode = new_inode(sb); | |
1430 | if (inode) { | |
85fe4025 | 1431 | inode->i_ino = get_next_ino(); |
454abafe | 1432 | inode_init_owner(inode, dir, mode); |
1da177e4 | 1433 | inode->i_blocks = 0; |
1da177e4 | 1434 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
91828a40 | 1435 | inode->i_generation = get_seconds(); |
1da177e4 LT |
1436 | info = SHMEM_I(inode); |
1437 | memset(info, 0, (char *)inode - (char *)info); | |
1438 | spin_lock_init(&info->lock); | |
40e041a2 | 1439 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 1440 | info->flags = flags & VM_NORESERVE; |
1da177e4 | 1441 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 1442 | simple_xattrs_init(&info->xattrs); |
72c04902 | 1443 | cache_no_acl(inode); |
1da177e4 LT |
1444 | |
1445 | switch (mode & S_IFMT) { | |
1446 | default: | |
39f0247d | 1447 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
1448 | init_special_inode(inode, mode, dev); |
1449 | break; | |
1450 | case S_IFREG: | |
14fcc23f | 1451 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
1452 | inode->i_op = &shmem_inode_operations; |
1453 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
1454 | mpol_shared_policy_init(&info->policy, |
1455 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
1456 | break; |
1457 | case S_IFDIR: | |
d8c76e6f | 1458 | inc_nlink(inode); |
1da177e4 LT |
1459 | /* Some things misbehave if size == 0 on a directory */ |
1460 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
1461 | inode->i_op = &shmem_dir_inode_operations; | |
1462 | inode->i_fop = &simple_dir_operations; | |
1463 | break; | |
1464 | case S_IFLNK: | |
1465 | /* | |
1466 | * Must not load anything in the rbtree, | |
1467 | * mpol_free_shared_policy will not be called. | |
1468 | */ | |
71fe804b | 1469 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
1470 | break; |
1471 | } | |
5b04c689 PE |
1472 | } else |
1473 | shmem_free_inode(sb); | |
1da177e4 LT |
1474 | return inode; |
1475 | } | |
1476 | ||
0cd6144a JW |
1477 | bool shmem_mapping(struct address_space *mapping) |
1478 | { | |
f0774d88 SL |
1479 | if (!mapping->host) |
1480 | return false; | |
1481 | ||
97b713ba | 1482 | return mapping->host->i_sb->s_op == &shmem_ops; |
0cd6144a JW |
1483 | } |
1484 | ||
1da177e4 | 1485 | #ifdef CONFIG_TMPFS |
92e1d5be | 1486 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 1487 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 1488 | |
6d9d88d0 JS |
1489 | #ifdef CONFIG_TMPFS_XATTR |
1490 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
1491 | #else | |
1492 | #define shmem_initxattrs NULL | |
1493 | #endif | |
1494 | ||
1da177e4 | 1495 | static int |
800d15a5 NP |
1496 | shmem_write_begin(struct file *file, struct address_space *mapping, |
1497 | loff_t pos, unsigned len, unsigned flags, | |
1498 | struct page **pagep, void **fsdata) | |
1da177e4 | 1499 | { |
800d15a5 | 1500 | struct inode *inode = mapping->host; |
40e041a2 | 1501 | struct shmem_inode_info *info = SHMEM_I(inode); |
800d15a5 | 1502 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
40e041a2 DH |
1503 | |
1504 | /* i_mutex is held by caller */ | |
1505 | if (unlikely(info->seals)) { | |
1506 | if (info->seals & F_SEAL_WRITE) | |
1507 | return -EPERM; | |
1508 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
1509 | return -EPERM; | |
1510 | } | |
1511 | ||
66d2f4d2 | 1512 | return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); |
800d15a5 NP |
1513 | } |
1514 | ||
1515 | static int | |
1516 | shmem_write_end(struct file *file, struct address_space *mapping, | |
1517 | loff_t pos, unsigned len, unsigned copied, | |
1518 | struct page *page, void *fsdata) | |
1519 | { | |
1520 | struct inode *inode = mapping->host; | |
1521 | ||
d3602444 HD |
1522 | if (pos + copied > inode->i_size) |
1523 | i_size_write(inode, pos + copied); | |
1524 | ||
ec9516fb HD |
1525 | if (!PageUptodate(page)) { |
1526 | if (copied < PAGE_CACHE_SIZE) { | |
1527 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
1528 | zero_user_segments(page, 0, from, | |
1529 | from + copied, PAGE_CACHE_SIZE); | |
1530 | } | |
1531 | SetPageUptodate(page); | |
1532 | } | |
800d15a5 | 1533 | set_page_dirty(page); |
6746aff7 | 1534 | unlock_page(page); |
800d15a5 NP |
1535 | page_cache_release(page); |
1536 | ||
800d15a5 | 1537 | return copied; |
1da177e4 LT |
1538 | } |
1539 | ||
2ba5bbed | 1540 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 1541 | { |
6e58e79d AV |
1542 | struct file *file = iocb->ki_filp; |
1543 | struct inode *inode = file_inode(file); | |
1da177e4 | 1544 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
1545 | pgoff_t index; |
1546 | unsigned long offset; | |
a0ee5ec5 | 1547 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 1548 | int error = 0; |
cb66a7a1 | 1549 | ssize_t retval = 0; |
6e58e79d | 1550 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
1551 | |
1552 | /* | |
1553 | * Might this read be for a stacking filesystem? Then when reading | |
1554 | * holes of a sparse file, we actually need to allocate those pages, | |
1555 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
1556 | */ | |
777eda2c | 1557 | if (!iter_is_iovec(to)) |
a0ee5ec5 | 1558 | sgp = SGP_DIRTY; |
1da177e4 LT |
1559 | |
1560 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1561 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1562 | ||
1563 | for (;;) { | |
1564 | struct page *page = NULL; | |
41ffe5d5 HD |
1565 | pgoff_t end_index; |
1566 | unsigned long nr, ret; | |
1da177e4 LT |
1567 | loff_t i_size = i_size_read(inode); |
1568 | ||
1569 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1570 | if (index > end_index) | |
1571 | break; | |
1572 | if (index == end_index) { | |
1573 | nr = i_size & ~PAGE_CACHE_MASK; | |
1574 | if (nr <= offset) | |
1575 | break; | |
1576 | } | |
1577 | ||
6e58e79d AV |
1578 | error = shmem_getpage(inode, index, &page, sgp, NULL); |
1579 | if (error) { | |
1580 | if (error == -EINVAL) | |
1581 | error = 0; | |
1da177e4 LT |
1582 | break; |
1583 | } | |
d3602444 HD |
1584 | if (page) |
1585 | unlock_page(page); | |
1da177e4 LT |
1586 | |
1587 | /* | |
1588 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 1589 | * are called without i_mutex protection against truncate |
1da177e4 LT |
1590 | */ |
1591 | nr = PAGE_CACHE_SIZE; | |
1592 | i_size = i_size_read(inode); | |
1593 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
1594 | if (index == end_index) { | |
1595 | nr = i_size & ~PAGE_CACHE_MASK; | |
1596 | if (nr <= offset) { | |
1597 | if (page) | |
1598 | page_cache_release(page); | |
1599 | break; | |
1600 | } | |
1601 | } | |
1602 | nr -= offset; | |
1603 | ||
1604 | if (page) { | |
1605 | /* | |
1606 | * If users can be writing to this page using arbitrary | |
1607 | * virtual addresses, take care about potential aliasing | |
1608 | * before reading the page on the kernel side. | |
1609 | */ | |
1610 | if (mapping_writably_mapped(mapping)) | |
1611 | flush_dcache_page(page); | |
1612 | /* | |
1613 | * Mark the page accessed if we read the beginning. | |
1614 | */ | |
1615 | if (!offset) | |
1616 | mark_page_accessed(page); | |
b5810039 | 1617 | } else { |
1da177e4 | 1618 | page = ZERO_PAGE(0); |
b5810039 NP |
1619 | page_cache_get(page); |
1620 | } | |
1da177e4 LT |
1621 | |
1622 | /* | |
1623 | * Ok, we have the page, and it's up-to-date, so | |
1624 | * now we can copy it to user space... | |
1da177e4 | 1625 | */ |
2ba5bbed | 1626 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 1627 | retval += ret; |
1da177e4 LT |
1628 | offset += ret; |
1629 | index += offset >> PAGE_CACHE_SHIFT; | |
1630 | offset &= ~PAGE_CACHE_MASK; | |
1631 | ||
1632 | page_cache_release(page); | |
2ba5bbed | 1633 | if (!iov_iter_count(to)) |
1da177e4 | 1634 | break; |
6e58e79d AV |
1635 | if (ret < nr) { |
1636 | error = -EFAULT; | |
1637 | break; | |
1638 | } | |
1da177e4 LT |
1639 | cond_resched(); |
1640 | } | |
1641 | ||
1642 | *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; | |
6e58e79d AV |
1643 | file_accessed(file); |
1644 | return retval ? retval : error; | |
1da177e4 LT |
1645 | } |
1646 | ||
708e3508 HD |
1647 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, |
1648 | struct pipe_inode_info *pipe, size_t len, | |
1649 | unsigned int flags) | |
1650 | { | |
1651 | struct address_space *mapping = in->f_mapping; | |
71f0e07a | 1652 | struct inode *inode = mapping->host; |
708e3508 HD |
1653 | unsigned int loff, nr_pages, req_pages; |
1654 | struct page *pages[PIPE_DEF_BUFFERS]; | |
1655 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
1656 | struct page *page; | |
1657 | pgoff_t index, end_index; | |
1658 | loff_t isize, left; | |
1659 | int error, page_nr; | |
1660 | struct splice_pipe_desc spd = { | |
1661 | .pages = pages, | |
1662 | .partial = partial, | |
047fe360 | 1663 | .nr_pages_max = PIPE_DEF_BUFFERS, |
708e3508 HD |
1664 | .flags = flags, |
1665 | .ops = &page_cache_pipe_buf_ops, | |
1666 | .spd_release = spd_release_page, | |
1667 | }; | |
1668 | ||
71f0e07a | 1669 | isize = i_size_read(inode); |
708e3508 HD |
1670 | if (unlikely(*ppos >= isize)) |
1671 | return 0; | |
1672 | ||
1673 | left = isize - *ppos; | |
1674 | if (unlikely(left < len)) | |
1675 | len = left; | |
1676 | ||
1677 | if (splice_grow_spd(pipe, &spd)) | |
1678 | return -ENOMEM; | |
1679 | ||
1680 | index = *ppos >> PAGE_CACHE_SHIFT; | |
1681 | loff = *ppos & ~PAGE_CACHE_MASK; | |
1682 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
a786c06d | 1683 | nr_pages = min(req_pages, spd.nr_pages_max); |
708e3508 | 1684 | |
708e3508 HD |
1685 | spd.nr_pages = find_get_pages_contig(mapping, index, |
1686 | nr_pages, spd.pages); | |
1687 | index += spd.nr_pages; | |
708e3508 | 1688 | error = 0; |
708e3508 | 1689 | |
71f0e07a | 1690 | while (spd.nr_pages < nr_pages) { |
71f0e07a HD |
1691 | error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL); |
1692 | if (error) | |
1693 | break; | |
1694 | unlock_page(page); | |
708e3508 HD |
1695 | spd.pages[spd.nr_pages++] = page; |
1696 | index++; | |
1697 | } | |
1698 | ||
708e3508 HD |
1699 | index = *ppos >> PAGE_CACHE_SHIFT; |
1700 | nr_pages = spd.nr_pages; | |
1701 | spd.nr_pages = 0; | |
71f0e07a | 1702 | |
708e3508 HD |
1703 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { |
1704 | unsigned int this_len; | |
1705 | ||
1706 | if (!len) | |
1707 | break; | |
1708 | ||
708e3508 HD |
1709 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); |
1710 | page = spd.pages[page_nr]; | |
1711 | ||
71f0e07a | 1712 | if (!PageUptodate(page) || page->mapping != mapping) { |
71f0e07a HD |
1713 | error = shmem_getpage(inode, index, &page, |
1714 | SGP_CACHE, NULL); | |
1715 | if (error) | |
708e3508 | 1716 | break; |
71f0e07a HD |
1717 | unlock_page(page); |
1718 | page_cache_release(spd.pages[page_nr]); | |
1719 | spd.pages[page_nr] = page; | |
708e3508 | 1720 | } |
71f0e07a HD |
1721 | |
1722 | isize = i_size_read(inode); | |
708e3508 HD |
1723 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; |
1724 | if (unlikely(!isize || index > end_index)) | |
1725 | break; | |
1726 | ||
708e3508 HD |
1727 | if (end_index == index) { |
1728 | unsigned int plen; | |
1729 | ||
708e3508 HD |
1730 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
1731 | if (plen <= loff) | |
1732 | break; | |
1733 | ||
708e3508 HD |
1734 | this_len = min(this_len, plen - loff); |
1735 | len = this_len; | |
1736 | } | |
1737 | ||
1738 | spd.partial[page_nr].offset = loff; | |
1739 | spd.partial[page_nr].len = this_len; | |
1740 | len -= this_len; | |
1741 | loff = 0; | |
1742 | spd.nr_pages++; | |
1743 | index++; | |
1744 | } | |
1745 | ||
708e3508 HD |
1746 | while (page_nr < nr_pages) |
1747 | page_cache_release(spd.pages[page_nr++]); | |
708e3508 HD |
1748 | |
1749 | if (spd.nr_pages) | |
1750 | error = splice_to_pipe(pipe, &spd); | |
1751 | ||
047fe360 | 1752 | splice_shrink_spd(&spd); |
708e3508 HD |
1753 | |
1754 | if (error > 0) { | |
1755 | *ppos += error; | |
1756 | file_accessed(in); | |
1757 | } | |
1758 | return error; | |
1759 | } | |
1760 | ||
220f2ac9 HD |
1761 | /* |
1762 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
1763 | */ | |
1764 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 1765 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
1766 | { |
1767 | struct page *page; | |
1768 | struct pagevec pvec; | |
1769 | pgoff_t indices[PAGEVEC_SIZE]; | |
1770 | bool done = false; | |
1771 | int i; | |
1772 | ||
1773 | pagevec_init(&pvec, 0); | |
1774 | pvec.nr = 1; /* start small: we may be there already */ | |
1775 | while (!done) { | |
0cd6144a | 1776 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
1777 | pvec.nr, pvec.pages, indices); |
1778 | if (!pvec.nr) { | |
965c8e59 | 1779 | if (whence == SEEK_DATA) |
220f2ac9 HD |
1780 | index = end; |
1781 | break; | |
1782 | } | |
1783 | for (i = 0; i < pvec.nr; i++, index++) { | |
1784 | if (index < indices[i]) { | |
965c8e59 | 1785 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
1786 | done = true; |
1787 | break; | |
1788 | } | |
1789 | index = indices[i]; | |
1790 | } | |
1791 | page = pvec.pages[i]; | |
1792 | if (page && !radix_tree_exceptional_entry(page)) { | |
1793 | if (!PageUptodate(page)) | |
1794 | page = NULL; | |
1795 | } | |
1796 | if (index >= end || | |
965c8e59 AM |
1797 | (page && whence == SEEK_DATA) || |
1798 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
1799 | done = true; |
1800 | break; | |
1801 | } | |
1802 | } | |
0cd6144a | 1803 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
1804 | pagevec_release(&pvec); |
1805 | pvec.nr = PAGEVEC_SIZE; | |
1806 | cond_resched(); | |
1807 | } | |
1808 | return index; | |
1809 | } | |
1810 | ||
965c8e59 | 1811 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
1812 | { |
1813 | struct address_space *mapping = file->f_mapping; | |
1814 | struct inode *inode = mapping->host; | |
1815 | pgoff_t start, end; | |
1816 | loff_t new_offset; | |
1817 | ||
965c8e59 AM |
1818 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
1819 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 HD |
1820 | MAX_LFS_FILESIZE, i_size_read(inode)); |
1821 | mutex_lock(&inode->i_mutex); | |
1822 | /* We're holding i_mutex so we can access i_size directly */ | |
1823 | ||
1824 | if (offset < 0) | |
1825 | offset = -EINVAL; | |
1826 | else if (offset >= inode->i_size) | |
1827 | offset = -ENXIO; | |
1828 | else { | |
1829 | start = offset >> PAGE_CACHE_SHIFT; | |
1830 | end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
965c8e59 | 1831 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
220f2ac9 HD |
1832 | new_offset <<= PAGE_CACHE_SHIFT; |
1833 | if (new_offset > offset) { | |
1834 | if (new_offset < inode->i_size) | |
1835 | offset = new_offset; | |
965c8e59 | 1836 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
1837 | offset = -ENXIO; |
1838 | else | |
1839 | offset = inode->i_size; | |
1840 | } | |
1841 | } | |
1842 | ||
387aae6f HD |
1843 | if (offset >= 0) |
1844 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
220f2ac9 HD |
1845 | mutex_unlock(&inode->i_mutex); |
1846 | return offset; | |
1847 | } | |
1848 | ||
05f65b5c DH |
1849 | /* |
1850 | * We need a tag: a new tag would expand every radix_tree_node by 8 bytes, | |
1851 | * so reuse a tag which we firmly believe is never set or cleared on shmem. | |
1852 | */ | |
1853 | #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE | |
1854 | #define LAST_SCAN 4 /* about 150ms max */ | |
1855 | ||
1856 | static void shmem_tag_pins(struct address_space *mapping) | |
1857 | { | |
1858 | struct radix_tree_iter iter; | |
1859 | void **slot; | |
1860 | pgoff_t start; | |
1861 | struct page *page; | |
1862 | ||
1863 | lru_add_drain(); | |
1864 | start = 0; | |
1865 | rcu_read_lock(); | |
1866 | ||
1867 | restart: | |
1868 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { | |
1869 | page = radix_tree_deref_slot(slot); | |
1870 | if (!page || radix_tree_exception(page)) { | |
1871 | if (radix_tree_deref_retry(page)) | |
1872 | goto restart; | |
1873 | } else if (page_count(page) - page_mapcount(page) > 1) { | |
1874 | spin_lock_irq(&mapping->tree_lock); | |
1875 | radix_tree_tag_set(&mapping->page_tree, iter.index, | |
1876 | SHMEM_TAG_PINNED); | |
1877 | spin_unlock_irq(&mapping->tree_lock); | |
1878 | } | |
1879 | ||
1880 | if (need_resched()) { | |
1881 | cond_resched_rcu(); | |
1882 | start = iter.index + 1; | |
1883 | goto restart; | |
1884 | } | |
1885 | } | |
1886 | rcu_read_unlock(); | |
1887 | } | |
1888 | ||
1889 | /* | |
1890 | * Setting SEAL_WRITE requires us to verify there's no pending writer. However, | |
1891 | * via get_user_pages(), drivers might have some pending I/O without any active | |
1892 | * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages | |
1893 | * and see whether it has an elevated ref-count. If so, we tag them and wait for | |
1894 | * them to be dropped. | |
1895 | * The caller must guarantee that no new user will acquire writable references | |
1896 | * to those pages to avoid races. | |
1897 | */ | |
40e041a2 DH |
1898 | static int shmem_wait_for_pins(struct address_space *mapping) |
1899 | { | |
05f65b5c DH |
1900 | struct radix_tree_iter iter; |
1901 | void **slot; | |
1902 | pgoff_t start; | |
1903 | struct page *page; | |
1904 | int error, scan; | |
1905 | ||
1906 | shmem_tag_pins(mapping); | |
1907 | ||
1908 | error = 0; | |
1909 | for (scan = 0; scan <= LAST_SCAN; scan++) { | |
1910 | if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED)) | |
1911 | break; | |
1912 | ||
1913 | if (!scan) | |
1914 | lru_add_drain_all(); | |
1915 | else if (schedule_timeout_killable((HZ << scan) / 200)) | |
1916 | scan = LAST_SCAN; | |
1917 | ||
1918 | start = 0; | |
1919 | rcu_read_lock(); | |
1920 | restart: | |
1921 | radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, | |
1922 | start, SHMEM_TAG_PINNED) { | |
1923 | ||
1924 | page = radix_tree_deref_slot(slot); | |
1925 | if (radix_tree_exception(page)) { | |
1926 | if (radix_tree_deref_retry(page)) | |
1927 | goto restart; | |
1928 | ||
1929 | page = NULL; | |
1930 | } | |
1931 | ||
1932 | if (page && | |
1933 | page_count(page) - page_mapcount(page) != 1) { | |
1934 | if (scan < LAST_SCAN) | |
1935 | goto continue_resched; | |
1936 | ||
1937 | /* | |
1938 | * On the last scan, we clean up all those tags | |
1939 | * we inserted; but make a note that we still | |
1940 | * found pages pinned. | |
1941 | */ | |
1942 | error = -EBUSY; | |
1943 | } | |
1944 | ||
1945 | spin_lock_irq(&mapping->tree_lock); | |
1946 | radix_tree_tag_clear(&mapping->page_tree, | |
1947 | iter.index, SHMEM_TAG_PINNED); | |
1948 | spin_unlock_irq(&mapping->tree_lock); | |
1949 | continue_resched: | |
1950 | if (need_resched()) { | |
1951 | cond_resched_rcu(); | |
1952 | start = iter.index + 1; | |
1953 | goto restart; | |
1954 | } | |
1955 | } | |
1956 | rcu_read_unlock(); | |
1957 | } | |
1958 | ||
1959 | return error; | |
40e041a2 DH |
1960 | } |
1961 | ||
1962 | #define F_ALL_SEALS (F_SEAL_SEAL | \ | |
1963 | F_SEAL_SHRINK | \ | |
1964 | F_SEAL_GROW | \ | |
1965 | F_SEAL_WRITE) | |
1966 | ||
1967 | int shmem_add_seals(struct file *file, unsigned int seals) | |
1968 | { | |
1969 | struct inode *inode = file_inode(file); | |
1970 | struct shmem_inode_info *info = SHMEM_I(inode); | |
1971 | int error; | |
1972 | ||
1973 | /* | |
1974 | * SEALING | |
1975 | * Sealing allows multiple parties to share a shmem-file but restrict | |
1976 | * access to a specific subset of file operations. Seals can only be | |
1977 | * added, but never removed. This way, mutually untrusted parties can | |
1978 | * share common memory regions with a well-defined policy. A malicious | |
1979 | * peer can thus never perform unwanted operations on a shared object. | |
1980 | * | |
1981 | * Seals are only supported on special shmem-files and always affect | |
1982 | * the whole underlying inode. Once a seal is set, it may prevent some | |
1983 | * kinds of access to the file. Currently, the following seals are | |
1984 | * defined: | |
1985 | * SEAL_SEAL: Prevent further seals from being set on this file | |
1986 | * SEAL_SHRINK: Prevent the file from shrinking | |
1987 | * SEAL_GROW: Prevent the file from growing | |
1988 | * SEAL_WRITE: Prevent write access to the file | |
1989 | * | |
1990 | * As we don't require any trust relationship between two parties, we | |
1991 | * must prevent seals from being removed. Therefore, sealing a file | |
1992 | * only adds a given set of seals to the file, it never touches | |
1993 | * existing seals. Furthermore, the "setting seals"-operation can be | |
1994 | * sealed itself, which basically prevents any further seal from being | |
1995 | * added. | |
1996 | * | |
1997 | * Semantics of sealing are only defined on volatile files. Only | |
1998 | * anonymous shmem files support sealing. More importantly, seals are | |
1999 | * never written to disk. Therefore, there's no plan to support it on | |
2000 | * other file types. | |
2001 | */ | |
2002 | ||
2003 | if (file->f_op != &shmem_file_operations) | |
2004 | return -EINVAL; | |
2005 | if (!(file->f_mode & FMODE_WRITE)) | |
2006 | return -EPERM; | |
2007 | if (seals & ~(unsigned int)F_ALL_SEALS) | |
2008 | return -EINVAL; | |
2009 | ||
2010 | mutex_lock(&inode->i_mutex); | |
2011 | ||
2012 | if (info->seals & F_SEAL_SEAL) { | |
2013 | error = -EPERM; | |
2014 | goto unlock; | |
2015 | } | |
2016 | ||
2017 | if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) { | |
2018 | error = mapping_deny_writable(file->f_mapping); | |
2019 | if (error) | |
2020 | goto unlock; | |
2021 | ||
2022 | error = shmem_wait_for_pins(file->f_mapping); | |
2023 | if (error) { | |
2024 | mapping_allow_writable(file->f_mapping); | |
2025 | goto unlock; | |
2026 | } | |
2027 | } | |
2028 | ||
2029 | info->seals |= seals; | |
2030 | error = 0; | |
2031 | ||
2032 | unlock: | |
2033 | mutex_unlock(&inode->i_mutex); | |
2034 | return error; | |
2035 | } | |
2036 | EXPORT_SYMBOL_GPL(shmem_add_seals); | |
2037 | ||
2038 | int shmem_get_seals(struct file *file) | |
2039 | { | |
2040 | if (file->f_op != &shmem_file_operations) | |
2041 | return -EINVAL; | |
2042 | ||
2043 | return SHMEM_I(file_inode(file))->seals; | |
2044 | } | |
2045 | EXPORT_SYMBOL_GPL(shmem_get_seals); | |
2046 | ||
2047 | long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg) | |
2048 | { | |
2049 | long error; | |
2050 | ||
2051 | switch (cmd) { | |
2052 | case F_ADD_SEALS: | |
2053 | /* disallow upper 32bit */ | |
2054 | if (arg > UINT_MAX) | |
2055 | return -EINVAL; | |
2056 | ||
2057 | error = shmem_add_seals(file, arg); | |
2058 | break; | |
2059 | case F_GET_SEALS: | |
2060 | error = shmem_get_seals(file); | |
2061 | break; | |
2062 | default: | |
2063 | error = -EINVAL; | |
2064 | break; | |
2065 | } | |
2066 | ||
2067 | return error; | |
2068 | } | |
2069 | ||
83e4fa9c HD |
2070 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2071 | loff_t len) | |
2072 | { | |
496ad9aa | 2073 | struct inode *inode = file_inode(file); |
e2d12e22 | 2074 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2075 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2076 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2077 | pgoff_t start, index, end; |
2078 | int error; | |
83e4fa9c | 2079 | |
13ace4d0 HD |
2080 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2081 | return -EOPNOTSUPP; | |
2082 | ||
83e4fa9c HD |
2083 | mutex_lock(&inode->i_mutex); |
2084 | ||
2085 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2086 | struct address_space *mapping = file->f_mapping; | |
2087 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2088 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2089 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2090 | |
40e041a2 DH |
2091 | /* protected by i_mutex */ |
2092 | if (info->seals & F_SEAL_WRITE) { | |
2093 | error = -EPERM; | |
2094 | goto out; | |
2095 | } | |
2096 | ||
8e205f77 | 2097 | shmem_falloc.waitq = &shmem_falloc_waitq; |
f00cdc6d HD |
2098 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
2099 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
2100 | spin_lock(&inode->i_lock); | |
2101 | inode->i_private = &shmem_falloc; | |
2102 | spin_unlock(&inode->i_lock); | |
2103 | ||
83e4fa9c HD |
2104 | if ((u64)unmap_end > (u64)unmap_start) |
2105 | unmap_mapping_range(mapping, unmap_start, | |
2106 | 1 + unmap_end - unmap_start, 0); | |
2107 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2108 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2109 | |
2110 | spin_lock(&inode->i_lock); | |
2111 | inode->i_private = NULL; | |
2112 | wake_up_all(&shmem_falloc_waitq); | |
2113 | spin_unlock(&inode->i_lock); | |
83e4fa9c | 2114 | error = 0; |
8e205f77 | 2115 | goto out; |
e2d12e22 HD |
2116 | } |
2117 | ||
2118 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2119 | error = inode_newsize_ok(inode, offset + len); | |
2120 | if (error) | |
2121 | goto out; | |
2122 | ||
40e041a2 DH |
2123 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2124 | error = -EPERM; | |
2125 | goto out; | |
2126 | } | |
2127 | ||
e2d12e22 HD |
2128 | start = offset >> PAGE_CACHE_SHIFT; |
2129 | end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
2130 | /* Try to avoid a swapstorm if len is impossible to satisfy */ | |
2131 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2132 | error = -ENOSPC; | |
2133 | goto out; | |
83e4fa9c HD |
2134 | } |
2135 | ||
8e205f77 | 2136 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2137 | shmem_falloc.start = start; |
2138 | shmem_falloc.next = start; | |
2139 | shmem_falloc.nr_falloced = 0; | |
2140 | shmem_falloc.nr_unswapped = 0; | |
2141 | spin_lock(&inode->i_lock); | |
2142 | inode->i_private = &shmem_falloc; | |
2143 | spin_unlock(&inode->i_lock); | |
2144 | ||
e2d12e22 HD |
2145 | for (index = start; index < end; index++) { |
2146 | struct page *page; | |
2147 | ||
2148 | /* | |
2149 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2150 | * been interrupted because we are using up too much memory. | |
2151 | */ | |
2152 | if (signal_pending(current)) | |
2153 | error = -EINTR; | |
1aac1400 HD |
2154 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2155 | error = -ENOMEM; | |
e2d12e22 | 2156 | else |
1635f6a7 | 2157 | error = shmem_getpage(inode, index, &page, SGP_FALLOC, |
e2d12e22 HD |
2158 | NULL); |
2159 | if (error) { | |
1635f6a7 HD |
2160 | /* Remove the !PageUptodate pages we added */ |
2161 | shmem_undo_range(inode, | |
2162 | (loff_t)start << PAGE_CACHE_SHIFT, | |
2163 | (loff_t)index << PAGE_CACHE_SHIFT, true); | |
1aac1400 | 2164 | goto undone; |
e2d12e22 HD |
2165 | } |
2166 | ||
1aac1400 HD |
2167 | /* |
2168 | * Inform shmem_writepage() how far we have reached. | |
2169 | * No need for lock or barrier: we have the page lock. | |
2170 | */ | |
2171 | shmem_falloc.next++; | |
2172 | if (!PageUptodate(page)) | |
2173 | shmem_falloc.nr_falloced++; | |
2174 | ||
e2d12e22 | 2175 | /* |
1635f6a7 HD |
2176 | * If !PageUptodate, leave it that way so that freeable pages |
2177 | * can be recognized if we need to rollback on error later. | |
2178 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2179 | * than free the pages we are allocating (and SGP_CACHE pages |
2180 | * might still be clean: we now need to mark those dirty too). | |
2181 | */ | |
2182 | set_page_dirty(page); | |
2183 | unlock_page(page); | |
2184 | page_cache_release(page); | |
2185 | cond_resched(); | |
2186 | } | |
2187 | ||
2188 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2189 | i_size_write(inode, offset + len); | |
e2d12e22 | 2190 | inode->i_ctime = CURRENT_TIME; |
1aac1400 HD |
2191 | undone: |
2192 | spin_lock(&inode->i_lock); | |
2193 | inode->i_private = NULL; | |
2194 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2195 | out: |
83e4fa9c HD |
2196 | mutex_unlock(&inode->i_mutex); |
2197 | return error; | |
2198 | } | |
2199 | ||
726c3342 | 2200 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2201 | { |
726c3342 | 2202 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2203 | |
2204 | buf->f_type = TMPFS_MAGIC; | |
2205 | buf->f_bsize = PAGE_CACHE_SIZE; | |
2206 | buf->f_namelen = NAME_MAX; | |
0edd73b3 | 2207 | if (sbinfo->max_blocks) { |
1da177e4 | 2208 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2209 | buf->f_bavail = |
2210 | buf->f_bfree = sbinfo->max_blocks - | |
2211 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2212 | } |
2213 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2214 | buf->f_files = sbinfo->max_inodes; |
2215 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2216 | } |
2217 | /* else leave those fields 0 like simple_statfs */ | |
2218 | return 0; | |
2219 | } | |
2220 | ||
2221 | /* | |
2222 | * File creation. Allocate an inode, and we're done.. | |
2223 | */ | |
2224 | static int | |
1a67aafb | 2225 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2226 | { |
0b0a0806 | 2227 | struct inode *inode; |
1da177e4 LT |
2228 | int error = -ENOSPC; |
2229 | ||
454abafe | 2230 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2231 | if (inode) { |
feda821e CH |
2232 | error = simple_acl_create(dir, inode); |
2233 | if (error) | |
2234 | goto out_iput; | |
2a7dba39 | 2235 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2236 | &dentry->d_name, |
6d9d88d0 | 2237 | shmem_initxattrs, NULL); |
feda821e CH |
2238 | if (error && error != -EOPNOTSUPP) |
2239 | goto out_iput; | |
37ec43cd | 2240 | |
718deb6b | 2241 | error = 0; |
1da177e4 LT |
2242 | dir->i_size += BOGO_DIRENT_SIZE; |
2243 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2244 | d_instantiate(dentry, inode); | |
2245 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2246 | } |
2247 | return error; | |
feda821e CH |
2248 | out_iput: |
2249 | iput(inode); | |
2250 | return error; | |
1da177e4 LT |
2251 | } |
2252 | ||
60545d0d AV |
2253 | static int |
2254 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2255 | { | |
2256 | struct inode *inode; | |
2257 | int error = -ENOSPC; | |
2258 | ||
2259 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
2260 | if (inode) { | |
2261 | error = security_inode_init_security(inode, dir, | |
2262 | NULL, | |
2263 | shmem_initxattrs, NULL); | |
feda821e CH |
2264 | if (error && error != -EOPNOTSUPP) |
2265 | goto out_iput; | |
2266 | error = simple_acl_create(dir, inode); | |
2267 | if (error) | |
2268 | goto out_iput; | |
60545d0d AV |
2269 | d_tmpfile(dentry, inode); |
2270 | } | |
2271 | return error; | |
feda821e CH |
2272 | out_iput: |
2273 | iput(inode); | |
2274 | return error; | |
60545d0d AV |
2275 | } |
2276 | ||
18bb1db3 | 2277 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
2278 | { |
2279 | int error; | |
2280 | ||
2281 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
2282 | return error; | |
d8c76e6f | 2283 | inc_nlink(dir); |
1da177e4 LT |
2284 | return 0; |
2285 | } | |
2286 | ||
4acdaf27 | 2287 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2288 | bool excl) |
1da177e4 LT |
2289 | { |
2290 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
2291 | } | |
2292 | ||
2293 | /* | |
2294 | * Link a file.. | |
2295 | */ | |
2296 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
2297 | { | |
75c3cfa8 | 2298 | struct inode *inode = d_inode(old_dentry); |
5b04c689 | 2299 | int ret; |
1da177e4 LT |
2300 | |
2301 | /* | |
2302 | * No ordinary (disk based) filesystem counts links as inodes; | |
2303 | * but each new link needs a new dentry, pinning lowmem, and | |
2304 | * tmpfs dentries cannot be pruned until they are unlinked. | |
2305 | */ | |
5b04c689 PE |
2306 | ret = shmem_reserve_inode(inode->i_sb); |
2307 | if (ret) | |
2308 | goto out; | |
1da177e4 LT |
2309 | |
2310 | dir->i_size += BOGO_DIRENT_SIZE; | |
2311 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 2312 | inc_nlink(inode); |
7de9c6ee | 2313 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
2314 | dget(dentry); /* Extra pinning count for the created dentry */ |
2315 | d_instantiate(dentry, inode); | |
5b04c689 PE |
2316 | out: |
2317 | return ret; | |
1da177e4 LT |
2318 | } |
2319 | ||
2320 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
2321 | { | |
75c3cfa8 | 2322 | struct inode *inode = d_inode(dentry); |
1da177e4 | 2323 | |
5b04c689 PE |
2324 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
2325 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
2326 | |
2327 | dir->i_size -= BOGO_DIRENT_SIZE; | |
2328 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 2329 | drop_nlink(inode); |
1da177e4 LT |
2330 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
2331 | return 0; | |
2332 | } | |
2333 | ||
2334 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
2335 | { | |
2336 | if (!simple_empty(dentry)) | |
2337 | return -ENOTEMPTY; | |
2338 | ||
75c3cfa8 | 2339 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 2340 | drop_nlink(dir); |
1da177e4 LT |
2341 | return shmem_unlink(dir, dentry); |
2342 | } | |
2343 | ||
37456771 MS |
2344 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
2345 | { | |
e36cb0b8 DH |
2346 | bool old_is_dir = d_is_dir(old_dentry); |
2347 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
2348 | |
2349 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
2350 | if (old_is_dir) { | |
2351 | drop_nlink(old_dir); | |
2352 | inc_nlink(new_dir); | |
2353 | } else { | |
2354 | drop_nlink(new_dir); | |
2355 | inc_nlink(old_dir); | |
2356 | } | |
2357 | } | |
2358 | old_dir->i_ctime = old_dir->i_mtime = | |
2359 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 DH |
2360 | d_inode(old_dentry)->i_ctime = |
2361 | d_inode(new_dentry)->i_ctime = CURRENT_TIME; | |
37456771 MS |
2362 | |
2363 | return 0; | |
2364 | } | |
2365 | ||
46fdb794 MS |
2366 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
2367 | { | |
2368 | struct dentry *whiteout; | |
2369 | int error; | |
2370 | ||
2371 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
2372 | if (!whiteout) | |
2373 | return -ENOMEM; | |
2374 | ||
2375 | error = shmem_mknod(old_dir, whiteout, | |
2376 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
2377 | dput(whiteout); | |
2378 | if (error) | |
2379 | return error; | |
2380 | ||
2381 | /* | |
2382 | * Cheat and hash the whiteout while the old dentry is still in | |
2383 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
2384 | * | |
2385 | * d_lookup() will consistently find one of them at this point, | |
2386 | * not sure which one, but that isn't even important. | |
2387 | */ | |
2388 | d_rehash(whiteout); | |
2389 | return 0; | |
2390 | } | |
2391 | ||
1da177e4 LT |
2392 | /* |
2393 | * The VFS layer already does all the dentry stuff for rename, | |
2394 | * we just have to decrement the usage count for the target if | |
2395 | * it exists so that the VFS layer correctly free's it when it | |
2396 | * gets overwritten. | |
2397 | */ | |
3b69ff51 | 2398 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 2399 | { |
75c3cfa8 | 2400 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
2401 | int they_are_dirs = S_ISDIR(inode->i_mode); |
2402 | ||
46fdb794 | 2403 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
2404 | return -EINVAL; |
2405 | ||
37456771 MS |
2406 | if (flags & RENAME_EXCHANGE) |
2407 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
2408 | ||
1da177e4 LT |
2409 | if (!simple_empty(new_dentry)) |
2410 | return -ENOTEMPTY; | |
2411 | ||
46fdb794 MS |
2412 | if (flags & RENAME_WHITEOUT) { |
2413 | int error; | |
2414 | ||
2415 | error = shmem_whiteout(old_dir, old_dentry); | |
2416 | if (error) | |
2417 | return error; | |
2418 | } | |
2419 | ||
75c3cfa8 | 2420 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 2421 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 2422 | if (they_are_dirs) { |
75c3cfa8 | 2423 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 2424 | drop_nlink(old_dir); |
b928095b | 2425 | } |
1da177e4 | 2426 | } else if (they_are_dirs) { |
9a53c3a7 | 2427 | drop_nlink(old_dir); |
d8c76e6f | 2428 | inc_nlink(new_dir); |
1da177e4 LT |
2429 | } |
2430 | ||
2431 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
2432 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
2433 | old_dir->i_ctime = old_dir->i_mtime = | |
2434 | new_dir->i_ctime = new_dir->i_mtime = | |
2435 | inode->i_ctime = CURRENT_TIME; | |
2436 | return 0; | |
2437 | } | |
2438 | ||
2439 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
2440 | { | |
2441 | int error; | |
2442 | int len; | |
2443 | struct inode *inode; | |
9276aad6 | 2444 | struct page *page; |
1da177e4 LT |
2445 | char *kaddr; |
2446 | struct shmem_inode_info *info; | |
2447 | ||
2448 | len = strlen(symname) + 1; | |
2449 | if (len > PAGE_CACHE_SIZE) | |
2450 | return -ENAMETOOLONG; | |
2451 | ||
454abafe | 2452 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
2453 | if (!inode) |
2454 | return -ENOSPC; | |
2455 | ||
9d8f13ba | 2456 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 2457 | shmem_initxattrs, NULL); |
570bc1c2 SS |
2458 | if (error) { |
2459 | if (error != -EOPNOTSUPP) { | |
2460 | iput(inode); | |
2461 | return error; | |
2462 | } | |
2463 | error = 0; | |
2464 | } | |
2465 | ||
1da177e4 LT |
2466 | info = SHMEM_I(inode); |
2467 | inode->i_size = len-1; | |
69f07ec9 HD |
2468 | if (len <= SHORT_SYMLINK_LEN) { |
2469 | info->symlink = kmemdup(symname, len, GFP_KERNEL); | |
2470 | if (!info->symlink) { | |
2471 | iput(inode); | |
2472 | return -ENOMEM; | |
2473 | } | |
2474 | inode->i_op = &shmem_short_symlink_operations; | |
60380f19 | 2475 | inode->i_link = info->symlink; |
1da177e4 LT |
2476 | } else { |
2477 | error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); | |
2478 | if (error) { | |
2479 | iput(inode); | |
2480 | return error; | |
2481 | } | |
14fcc23f | 2482 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 2483 | inode->i_op = &shmem_symlink_inode_operations; |
9b04c5fe | 2484 | kaddr = kmap_atomic(page); |
1da177e4 | 2485 | memcpy(kaddr, symname, len); |
9b04c5fe | 2486 | kunmap_atomic(kaddr); |
ec9516fb | 2487 | SetPageUptodate(page); |
1da177e4 | 2488 | set_page_dirty(page); |
6746aff7 | 2489 | unlock_page(page); |
1da177e4 LT |
2490 | page_cache_release(page); |
2491 | } | |
1da177e4 LT |
2492 | dir->i_size += BOGO_DIRENT_SIZE; |
2493 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2494 | d_instantiate(dentry, inode); | |
2495 | dget(dentry); | |
2496 | return 0; | |
2497 | } | |
2498 | ||
6e77137b | 2499 | static const char *shmem_follow_link(struct dentry *dentry, void **cookie) |
1da177e4 LT |
2500 | { |
2501 | struct page *page = NULL; | |
75c3cfa8 | 2502 | int error = shmem_getpage(d_inode(dentry), 0, &page, SGP_READ, NULL); |
680baacb AV |
2503 | if (error) |
2504 | return ERR_PTR(error); | |
2505 | unlock_page(page); | |
2506 | *cookie = page; | |
2507 | return kmap(page); | |
1da177e4 LT |
2508 | } |
2509 | ||
5f2c4179 | 2510 | static void shmem_put_link(struct inode *unused, void *cookie) |
1da177e4 | 2511 | { |
680baacb AV |
2512 | struct page *page = cookie; |
2513 | kunmap(page); | |
2514 | mark_page_accessed(page); | |
2515 | page_cache_release(page); | |
1da177e4 LT |
2516 | } |
2517 | ||
b09e0fa4 | 2518 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 2519 | /* |
b09e0fa4 EP |
2520 | * Superblocks without xattr inode operations may get some security.* xattr |
2521 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
2522 | * like ACLs, we also need to implement the security.* handlers at |
2523 | * filesystem level, though. | |
2524 | */ | |
2525 | ||
6d9d88d0 JS |
2526 | /* |
2527 | * Callback for security_inode_init_security() for acquiring xattrs. | |
2528 | */ | |
2529 | static int shmem_initxattrs(struct inode *inode, | |
2530 | const struct xattr *xattr_array, | |
2531 | void *fs_info) | |
2532 | { | |
2533 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2534 | const struct xattr *xattr; | |
38f38657 | 2535 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
2536 | size_t len; |
2537 | ||
2538 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 2539 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
2540 | if (!new_xattr) |
2541 | return -ENOMEM; | |
2542 | ||
2543 | len = strlen(xattr->name) + 1; | |
2544 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
2545 | GFP_KERNEL); | |
2546 | if (!new_xattr->name) { | |
2547 | kfree(new_xattr); | |
2548 | return -ENOMEM; | |
2549 | } | |
2550 | ||
2551 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
2552 | XATTR_SECURITY_PREFIX_LEN); | |
2553 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
2554 | xattr->name, len); | |
2555 | ||
38f38657 | 2556 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
2557 | } |
2558 | ||
2559 | return 0; | |
2560 | } | |
2561 | ||
bb435453 | 2562 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
b09e0fa4 | 2563 | #ifdef CONFIG_TMPFS_POSIX_ACL |
feda821e CH |
2564 | &posix_acl_access_xattr_handler, |
2565 | &posix_acl_default_xattr_handler, | |
b09e0fa4 | 2566 | #endif |
39f0247d AG |
2567 | NULL |
2568 | }; | |
b09e0fa4 EP |
2569 | |
2570 | static int shmem_xattr_validate(const char *name) | |
2571 | { | |
2572 | struct { const char *prefix; size_t len; } arr[] = { | |
2573 | { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN }, | |
2574 | { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN } | |
2575 | }; | |
2576 | int i; | |
2577 | ||
2578 | for (i = 0; i < ARRAY_SIZE(arr); i++) { | |
2579 | size_t preflen = arr[i].len; | |
2580 | if (strncmp(name, arr[i].prefix, preflen) == 0) { | |
2581 | if (!name[preflen]) | |
2582 | return -EINVAL; | |
2583 | return 0; | |
2584 | } | |
2585 | } | |
2586 | return -EOPNOTSUPP; | |
2587 | } | |
2588 | ||
2589 | static ssize_t shmem_getxattr(struct dentry *dentry, const char *name, | |
2590 | void *buffer, size_t size) | |
2591 | { | |
75c3cfa8 | 2592 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
b09e0fa4 EP |
2593 | int err; |
2594 | ||
2595 | /* | |
2596 | * If this is a request for a synthetic attribute in the system.* | |
2597 | * namespace use the generic infrastructure to resolve a handler | |
2598 | * for it via sb->s_xattr. | |
2599 | */ | |
2600 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2601 | return generic_getxattr(dentry, name, buffer, size); | |
2602 | ||
2603 | err = shmem_xattr_validate(name); | |
2604 | if (err) | |
2605 | return err; | |
2606 | ||
38f38657 | 2607 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
2608 | } |
2609 | ||
2610 | static int shmem_setxattr(struct dentry *dentry, const char *name, | |
2611 | const void *value, size_t size, int flags) | |
2612 | { | |
75c3cfa8 | 2613 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
b09e0fa4 EP |
2614 | int err; |
2615 | ||
2616 | /* | |
2617 | * If this is a request for a synthetic attribute in the system.* | |
2618 | * namespace use the generic infrastructure to resolve a handler | |
2619 | * for it via sb->s_xattr. | |
2620 | */ | |
2621 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2622 | return generic_setxattr(dentry, name, value, size, flags); | |
2623 | ||
2624 | err = shmem_xattr_validate(name); | |
2625 | if (err) | |
2626 | return err; | |
2627 | ||
38f38657 | 2628 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
2629 | } |
2630 | ||
2631 | static int shmem_removexattr(struct dentry *dentry, const char *name) | |
2632 | { | |
75c3cfa8 | 2633 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
b09e0fa4 EP |
2634 | int err; |
2635 | ||
2636 | /* | |
2637 | * If this is a request for a synthetic attribute in the system.* | |
2638 | * namespace use the generic infrastructure to resolve a handler | |
2639 | * for it via sb->s_xattr. | |
2640 | */ | |
2641 | if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) | |
2642 | return generic_removexattr(dentry, name); | |
2643 | ||
2644 | err = shmem_xattr_validate(name); | |
2645 | if (err) | |
2646 | return err; | |
2647 | ||
38f38657 | 2648 | return simple_xattr_remove(&info->xattrs, name); |
b09e0fa4 EP |
2649 | } |
2650 | ||
2651 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
2652 | { | |
75c3cfa8 | 2653 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
38f38657 | 2654 | return simple_xattr_list(&info->xattrs, buffer, size); |
b09e0fa4 EP |
2655 | } |
2656 | #endif /* CONFIG_TMPFS_XATTR */ | |
2657 | ||
69f07ec9 | 2658 | static const struct inode_operations shmem_short_symlink_operations = { |
b09e0fa4 | 2659 | .readlink = generic_readlink, |
60380f19 | 2660 | .follow_link = simple_follow_link, |
b09e0fa4 EP |
2661 | #ifdef CONFIG_TMPFS_XATTR |
2662 | .setxattr = shmem_setxattr, | |
2663 | .getxattr = shmem_getxattr, | |
2664 | .listxattr = shmem_listxattr, | |
2665 | .removexattr = shmem_removexattr, | |
2666 | #endif | |
2667 | }; | |
2668 | ||
2669 | static const struct inode_operations shmem_symlink_inode_operations = { | |
2670 | .readlink = generic_readlink, | |
2671 | .follow_link = shmem_follow_link, | |
2672 | .put_link = shmem_put_link, | |
2673 | #ifdef CONFIG_TMPFS_XATTR | |
2674 | .setxattr = shmem_setxattr, | |
2675 | .getxattr = shmem_getxattr, | |
2676 | .listxattr = shmem_listxattr, | |
2677 | .removexattr = shmem_removexattr, | |
39f0247d | 2678 | #endif |
b09e0fa4 | 2679 | }; |
39f0247d | 2680 | |
91828a40 DG |
2681 | static struct dentry *shmem_get_parent(struct dentry *child) |
2682 | { | |
2683 | return ERR_PTR(-ESTALE); | |
2684 | } | |
2685 | ||
2686 | static int shmem_match(struct inode *ino, void *vfh) | |
2687 | { | |
2688 | __u32 *fh = vfh; | |
2689 | __u64 inum = fh[2]; | |
2690 | inum = (inum << 32) | fh[1]; | |
2691 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
2692 | } | |
2693 | ||
480b116c CH |
2694 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
2695 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 2696 | { |
91828a40 | 2697 | struct inode *inode; |
480b116c | 2698 | struct dentry *dentry = NULL; |
35c2a7f4 | 2699 | u64 inum; |
480b116c CH |
2700 | |
2701 | if (fh_len < 3) | |
2702 | return NULL; | |
91828a40 | 2703 | |
35c2a7f4 HD |
2704 | inum = fid->raw[2]; |
2705 | inum = (inum << 32) | fid->raw[1]; | |
2706 | ||
480b116c CH |
2707 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
2708 | shmem_match, fid->raw); | |
91828a40 | 2709 | if (inode) { |
480b116c | 2710 | dentry = d_find_alias(inode); |
91828a40 DG |
2711 | iput(inode); |
2712 | } | |
2713 | ||
480b116c | 2714 | return dentry; |
91828a40 DG |
2715 | } |
2716 | ||
b0b0382b AV |
2717 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
2718 | struct inode *parent) | |
91828a40 | 2719 | { |
5fe0c237 AK |
2720 | if (*len < 3) { |
2721 | *len = 3; | |
94e07a75 | 2722 | return FILEID_INVALID; |
5fe0c237 | 2723 | } |
91828a40 | 2724 | |
1d3382cb | 2725 | if (inode_unhashed(inode)) { |
91828a40 DG |
2726 | /* Unfortunately insert_inode_hash is not idempotent, |
2727 | * so as we hash inodes here rather than at creation | |
2728 | * time, we need a lock to ensure we only try | |
2729 | * to do it once | |
2730 | */ | |
2731 | static DEFINE_SPINLOCK(lock); | |
2732 | spin_lock(&lock); | |
1d3382cb | 2733 | if (inode_unhashed(inode)) |
91828a40 DG |
2734 | __insert_inode_hash(inode, |
2735 | inode->i_ino + inode->i_generation); | |
2736 | spin_unlock(&lock); | |
2737 | } | |
2738 | ||
2739 | fh[0] = inode->i_generation; | |
2740 | fh[1] = inode->i_ino; | |
2741 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
2742 | ||
2743 | *len = 3; | |
2744 | return 1; | |
2745 | } | |
2746 | ||
39655164 | 2747 | static const struct export_operations shmem_export_ops = { |
91828a40 | 2748 | .get_parent = shmem_get_parent, |
91828a40 | 2749 | .encode_fh = shmem_encode_fh, |
480b116c | 2750 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
2751 | }; |
2752 | ||
680d794b | 2753 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
2754 | bool remount) | |
1da177e4 LT |
2755 | { |
2756 | char *this_char, *value, *rest; | |
49cd0a5c | 2757 | struct mempolicy *mpol = NULL; |
8751e039 EB |
2758 | uid_t uid; |
2759 | gid_t gid; | |
1da177e4 | 2760 | |
b00dc3ad HD |
2761 | while (options != NULL) { |
2762 | this_char = options; | |
2763 | for (;;) { | |
2764 | /* | |
2765 | * NUL-terminate this option: unfortunately, | |
2766 | * mount options form a comma-separated list, | |
2767 | * but mpol's nodelist may also contain commas. | |
2768 | */ | |
2769 | options = strchr(options, ','); | |
2770 | if (options == NULL) | |
2771 | break; | |
2772 | options++; | |
2773 | if (!isdigit(*options)) { | |
2774 | options[-1] = '\0'; | |
2775 | break; | |
2776 | } | |
2777 | } | |
1da177e4 LT |
2778 | if (!*this_char) |
2779 | continue; | |
2780 | if ((value = strchr(this_char,'=')) != NULL) { | |
2781 | *value++ = 0; | |
2782 | } else { | |
2783 | printk(KERN_ERR | |
2784 | "tmpfs: No value for mount option '%s'\n", | |
2785 | this_char); | |
49cd0a5c | 2786 | goto error; |
1da177e4 LT |
2787 | } |
2788 | ||
2789 | if (!strcmp(this_char,"size")) { | |
2790 | unsigned long long size; | |
2791 | size = memparse(value,&rest); | |
2792 | if (*rest == '%') { | |
2793 | size <<= PAGE_SHIFT; | |
2794 | size *= totalram_pages; | |
2795 | do_div(size, 100); | |
2796 | rest++; | |
2797 | } | |
2798 | if (*rest) | |
2799 | goto bad_val; | |
680d794b | 2800 | sbinfo->max_blocks = |
2801 | DIV_ROUND_UP(size, PAGE_CACHE_SIZE); | |
1da177e4 | 2802 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 2803 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
2804 | if (*rest) |
2805 | goto bad_val; | |
2806 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 2807 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
2808 | if (*rest) |
2809 | goto bad_val; | |
2810 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 2811 | if (remount) |
1da177e4 | 2812 | continue; |
680d794b | 2813 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
2814 | if (*rest) |
2815 | goto bad_val; | |
2816 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 2817 | if (remount) |
1da177e4 | 2818 | continue; |
8751e039 | 2819 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2820 | if (*rest) |
2821 | goto bad_val; | |
8751e039 EB |
2822 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
2823 | if (!uid_valid(sbinfo->uid)) | |
2824 | goto bad_val; | |
1da177e4 | 2825 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 2826 | if (remount) |
1da177e4 | 2827 | continue; |
8751e039 | 2828 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2829 | if (*rest) |
2830 | goto bad_val; | |
8751e039 EB |
2831 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
2832 | if (!gid_valid(sbinfo->gid)) | |
2833 | goto bad_val; | |
7339ff83 | 2834 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
2835 | mpol_put(mpol); |
2836 | mpol = NULL; | |
2837 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 2838 | goto bad_val; |
1da177e4 LT |
2839 | } else { |
2840 | printk(KERN_ERR "tmpfs: Bad mount option %s\n", | |
2841 | this_char); | |
49cd0a5c | 2842 | goto error; |
1da177e4 LT |
2843 | } |
2844 | } | |
49cd0a5c | 2845 | sbinfo->mpol = mpol; |
1da177e4 LT |
2846 | return 0; |
2847 | ||
2848 | bad_val: | |
2849 | printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", | |
2850 | value, this_char); | |
49cd0a5c GT |
2851 | error: |
2852 | mpol_put(mpol); | |
1da177e4 LT |
2853 | return 1; |
2854 | ||
2855 | } | |
2856 | ||
2857 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
2858 | { | |
2859 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 2860 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
2861 | unsigned long inodes; |
2862 | int error = -EINVAL; | |
2863 | ||
5f00110f | 2864 | config.mpol = NULL; |
680d794b | 2865 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 2866 | return error; |
1da177e4 | 2867 | |
0edd73b3 | 2868 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 2869 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 2870 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 2871 | goto out; |
680d794b | 2872 | if (config.max_inodes < inodes) |
0edd73b3 HD |
2873 | goto out; |
2874 | /* | |
54af6042 | 2875 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
2876 | * but we must separately disallow unlimited->limited, because |
2877 | * in that case we have no record of how much is already in use. | |
2878 | */ | |
680d794b | 2879 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 2880 | goto out; |
680d794b | 2881 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
2882 | goto out; |
2883 | ||
2884 | error = 0; | |
680d794b | 2885 | sbinfo->max_blocks = config.max_blocks; |
680d794b | 2886 | sbinfo->max_inodes = config.max_inodes; |
2887 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 2888 | |
5f00110f GT |
2889 | /* |
2890 | * Preserve previous mempolicy unless mpol remount option was specified. | |
2891 | */ | |
2892 | if (config.mpol) { | |
2893 | mpol_put(sbinfo->mpol); | |
2894 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
2895 | } | |
0edd73b3 HD |
2896 | out: |
2897 | spin_unlock(&sbinfo->stat_lock); | |
2898 | return error; | |
1da177e4 | 2899 | } |
680d794b | 2900 | |
34c80b1d | 2901 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 2902 | { |
34c80b1d | 2903 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b | 2904 | |
2905 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
2906 | seq_printf(seq, ",size=%luk", | |
2907 | sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); | |
2908 | if (sbinfo->max_inodes != shmem_default_max_inodes()) | |
2909 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
2910 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 2911 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
2912 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
2913 | seq_printf(seq, ",uid=%u", | |
2914 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
2915 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
2916 | seq_printf(seq, ",gid=%u", | |
2917 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
71fe804b | 2918 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b | 2919 | return 0; |
2920 | } | |
9183df25 DH |
2921 | |
2922 | #define MFD_NAME_PREFIX "memfd:" | |
2923 | #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1) | |
2924 | #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN) | |
2925 | ||
2926 | #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING) | |
2927 | ||
2928 | SYSCALL_DEFINE2(memfd_create, | |
2929 | const char __user *, uname, | |
2930 | unsigned int, flags) | |
2931 | { | |
2932 | struct shmem_inode_info *info; | |
2933 | struct file *file; | |
2934 | int fd, error; | |
2935 | char *name; | |
2936 | long len; | |
2937 | ||
2938 | if (flags & ~(unsigned int)MFD_ALL_FLAGS) | |
2939 | return -EINVAL; | |
2940 | ||
2941 | /* length includes terminating zero */ | |
2942 | len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1); | |
2943 | if (len <= 0) | |
2944 | return -EFAULT; | |
2945 | if (len > MFD_NAME_MAX_LEN + 1) | |
2946 | return -EINVAL; | |
2947 | ||
2948 | name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY); | |
2949 | if (!name) | |
2950 | return -ENOMEM; | |
2951 | ||
2952 | strcpy(name, MFD_NAME_PREFIX); | |
2953 | if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) { | |
2954 | error = -EFAULT; | |
2955 | goto err_name; | |
2956 | } | |
2957 | ||
2958 | /* terminating-zero may have changed after strnlen_user() returned */ | |
2959 | if (name[len + MFD_NAME_PREFIX_LEN - 1]) { | |
2960 | error = -EFAULT; | |
2961 | goto err_name; | |
2962 | } | |
2963 | ||
2964 | fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0); | |
2965 | if (fd < 0) { | |
2966 | error = fd; | |
2967 | goto err_name; | |
2968 | } | |
2969 | ||
2970 | file = shmem_file_setup(name, 0, VM_NORESERVE); | |
2971 | if (IS_ERR(file)) { | |
2972 | error = PTR_ERR(file); | |
2973 | goto err_fd; | |
2974 | } | |
2975 | info = SHMEM_I(file_inode(file)); | |
2976 | file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; | |
2977 | file->f_flags |= O_RDWR | O_LARGEFILE; | |
2978 | if (flags & MFD_ALLOW_SEALING) | |
2979 | info->seals &= ~F_SEAL_SEAL; | |
2980 | ||
2981 | fd_install(fd, file); | |
2982 | kfree(name); | |
2983 | return fd; | |
2984 | ||
2985 | err_fd: | |
2986 | put_unused_fd(fd); | |
2987 | err_name: | |
2988 | kfree(name); | |
2989 | return error; | |
2990 | } | |
2991 | ||
680d794b | 2992 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
2993 | |
2994 | static void shmem_put_super(struct super_block *sb) | |
2995 | { | |
602586a8 HD |
2996 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
2997 | ||
2998 | percpu_counter_destroy(&sbinfo->used_blocks); | |
49cd0a5c | 2999 | mpol_put(sbinfo->mpol); |
602586a8 | 3000 | kfree(sbinfo); |
1da177e4 LT |
3001 | sb->s_fs_info = NULL; |
3002 | } | |
3003 | ||
2b2af54a | 3004 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
3005 | { |
3006 | struct inode *inode; | |
0edd73b3 | 3007 | struct shmem_sb_info *sbinfo; |
680d794b | 3008 | int err = -ENOMEM; |
3009 | ||
3010 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3011 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b | 3012 | L1_CACHE_BYTES), GFP_KERNEL); |
3013 | if (!sbinfo) | |
3014 | return -ENOMEM; | |
3015 | ||
680d794b | 3016 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
3017 | sbinfo->uid = current_fsuid(); |
3018 | sbinfo->gid = current_fsgid(); | |
680d794b | 3019 | sb->s_fs_info = sbinfo; |
1da177e4 | 3020 | |
0edd73b3 | 3021 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3022 | /* |
3023 | * Per default we only allow half of the physical ram per | |
3024 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3025 | * but the internal instance is left unlimited. | |
3026 | */ | |
ca4e0519 | 3027 | if (!(sb->s_flags & MS_KERNMOUNT)) { |
680d794b | 3028 | sbinfo->max_blocks = shmem_default_max_blocks(); |
3029 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
3030 | if (shmem_parse_options(data, sbinfo, false)) { | |
3031 | err = -EINVAL; | |
3032 | goto failed; | |
3033 | } | |
ca4e0519 AV |
3034 | } else { |
3035 | sb->s_flags |= MS_NOUSER; | |
1da177e4 | 3036 | } |
91828a40 | 3037 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 3038 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
3039 | #else |
3040 | sb->s_flags |= MS_NOUSER; | |
3041 | #endif | |
3042 | ||
0edd73b3 | 3043 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3044 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3045 | goto failed; |
680d794b | 3046 | sbinfo->free_inodes = sbinfo->max_inodes; |
0edd73b3 | 3047 | |
285b2c4f | 3048 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
1da177e4 LT |
3049 | sb->s_blocksize = PAGE_CACHE_SIZE; |
3050 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
3051 | sb->s_magic = TMPFS_MAGIC; | |
3052 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3053 | sb->s_time_gran = 1; |
b09e0fa4 | 3054 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3055 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3056 | #endif |
3057 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
3058 | sb->s_flags |= MS_POSIXACL; |
3059 | #endif | |
0edd73b3 | 3060 | |
454abafe | 3061 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3062 | if (!inode) |
3063 | goto failed; | |
680d794b | 3064 | inode->i_uid = sbinfo->uid; |
3065 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3066 | sb->s_root = d_make_root(inode); |
3067 | if (!sb->s_root) | |
48fde701 | 3068 | goto failed; |
1da177e4 LT |
3069 | return 0; |
3070 | ||
1da177e4 LT |
3071 | failed: |
3072 | shmem_put_super(sb); | |
3073 | return err; | |
3074 | } | |
3075 | ||
fcc234f8 | 3076 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3077 | |
3078 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3079 | { | |
41ffe5d5 HD |
3080 | struct shmem_inode_info *info; |
3081 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3082 | if (!info) | |
1da177e4 | 3083 | return NULL; |
41ffe5d5 | 3084 | return &info->vfs_inode; |
1da177e4 LT |
3085 | } |
3086 | ||
41ffe5d5 | 3087 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
3088 | { |
3089 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
3090 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3091 | } | |
3092 | ||
1da177e4 LT |
3093 | static void shmem_destroy_inode(struct inode *inode) |
3094 | { | |
09208d15 | 3095 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3096 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 3097 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
3098 | } |
3099 | ||
41ffe5d5 | 3100 | static void shmem_init_inode(void *foo) |
1da177e4 | 3101 | { |
41ffe5d5 HD |
3102 | struct shmem_inode_info *info = foo; |
3103 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3104 | } |
3105 | ||
41ffe5d5 | 3106 | static int shmem_init_inodecache(void) |
1da177e4 LT |
3107 | { |
3108 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3109 | sizeof(struct shmem_inode_info), | |
41ffe5d5 | 3110 | 0, SLAB_PANIC, shmem_init_inode); |
1da177e4 LT |
3111 | return 0; |
3112 | } | |
3113 | ||
41ffe5d5 | 3114 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3115 | { |
1a1d92c1 | 3116 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3117 | } |
3118 | ||
f5e54d6e | 3119 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3120 | .writepage = shmem_writepage, |
76719325 | 3121 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3122 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3123 | .write_begin = shmem_write_begin, |
3124 | .write_end = shmem_write_end, | |
1da177e4 | 3125 | #endif |
1c93923c | 3126 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3127 | .migratepage = migrate_page, |
1c93923c | 3128 | #endif |
aa261f54 | 3129 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3130 | }; |
3131 | ||
15ad7cdc | 3132 | static const struct file_operations shmem_file_operations = { |
1da177e4 LT |
3133 | .mmap = shmem_mmap, |
3134 | #ifdef CONFIG_TMPFS | |
220f2ac9 | 3135 | .llseek = shmem_file_llseek, |
2ba5bbed | 3136 | .read_iter = shmem_file_read_iter, |
8174202b | 3137 | .write_iter = generic_file_write_iter, |
1b061d92 | 3138 | .fsync = noop_fsync, |
708e3508 | 3139 | .splice_read = shmem_file_splice_read, |
f6cb85d0 | 3140 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3141 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3142 | #endif |
3143 | }; | |
3144 | ||
92e1d5be | 3145 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3146 | .getattr = shmem_getattr, |
94c1e62d | 3147 | .setattr = shmem_setattr, |
b09e0fa4 EP |
3148 | #ifdef CONFIG_TMPFS_XATTR |
3149 | .setxattr = shmem_setxattr, | |
3150 | .getxattr = shmem_getxattr, | |
3151 | .listxattr = shmem_listxattr, | |
3152 | .removexattr = shmem_removexattr, | |
feda821e | 3153 | .set_acl = simple_set_acl, |
b09e0fa4 | 3154 | #endif |
1da177e4 LT |
3155 | }; |
3156 | ||
92e1d5be | 3157 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3158 | #ifdef CONFIG_TMPFS |
3159 | .create = shmem_create, | |
3160 | .lookup = simple_lookup, | |
3161 | .link = shmem_link, | |
3162 | .unlink = shmem_unlink, | |
3163 | .symlink = shmem_symlink, | |
3164 | .mkdir = shmem_mkdir, | |
3165 | .rmdir = shmem_rmdir, | |
3166 | .mknod = shmem_mknod, | |
3b69ff51 | 3167 | .rename2 = shmem_rename2, |
60545d0d | 3168 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3169 | #endif |
b09e0fa4 EP |
3170 | #ifdef CONFIG_TMPFS_XATTR |
3171 | .setxattr = shmem_setxattr, | |
3172 | .getxattr = shmem_getxattr, | |
3173 | .listxattr = shmem_listxattr, | |
3174 | .removexattr = shmem_removexattr, | |
3175 | #endif | |
39f0247d | 3176 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3177 | .setattr = shmem_setattr, |
feda821e | 3178 | .set_acl = simple_set_acl, |
39f0247d AG |
3179 | #endif |
3180 | }; | |
3181 | ||
92e1d5be | 3182 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 EP |
3183 | #ifdef CONFIG_TMPFS_XATTR |
3184 | .setxattr = shmem_setxattr, | |
3185 | .getxattr = shmem_getxattr, | |
3186 | .listxattr = shmem_listxattr, | |
3187 | .removexattr = shmem_removexattr, | |
3188 | #endif | |
39f0247d | 3189 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3190 | .setattr = shmem_setattr, |
feda821e | 3191 | .set_acl = simple_set_acl, |
39f0247d | 3192 | #endif |
1da177e4 LT |
3193 | }; |
3194 | ||
759b9775 | 3195 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
3196 | .alloc_inode = shmem_alloc_inode, |
3197 | .destroy_inode = shmem_destroy_inode, | |
3198 | #ifdef CONFIG_TMPFS | |
3199 | .statfs = shmem_statfs, | |
3200 | .remount_fs = shmem_remount_fs, | |
680d794b | 3201 | .show_options = shmem_show_options, |
1da177e4 | 3202 | #endif |
1f895f75 | 3203 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3204 | .drop_inode = generic_delete_inode, |
3205 | .put_super = shmem_put_super, | |
3206 | }; | |
3207 | ||
f0f37e2f | 3208 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3209 | .fault = shmem_fault, |
d7c17551 | 3210 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3211 | #ifdef CONFIG_NUMA |
3212 | .set_policy = shmem_set_policy, | |
3213 | .get_policy = shmem_get_policy, | |
3214 | #endif | |
3215 | }; | |
3216 | ||
3c26ff6e AV |
3217 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
3218 | int flags, const char *dev_name, void *data) | |
1da177e4 | 3219 | { |
3c26ff6e | 3220 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
3221 | } |
3222 | ||
41ffe5d5 | 3223 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3224 | .owner = THIS_MODULE, |
3225 | .name = "tmpfs", | |
3c26ff6e | 3226 | .mount = shmem_mount, |
1da177e4 | 3227 | .kill_sb = kill_litter_super, |
2b8576cb | 3228 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 3229 | }; |
1da177e4 | 3230 | |
41ffe5d5 | 3231 | int __init shmem_init(void) |
1da177e4 LT |
3232 | { |
3233 | int error; | |
3234 | ||
16203a7a RL |
3235 | /* If rootfs called this, don't re-init */ |
3236 | if (shmem_inode_cachep) | |
3237 | return 0; | |
3238 | ||
41ffe5d5 | 3239 | error = shmem_init_inodecache(); |
1da177e4 LT |
3240 | if (error) |
3241 | goto out3; | |
3242 | ||
41ffe5d5 | 3243 | error = register_filesystem(&shmem_fs_type); |
1da177e4 LT |
3244 | if (error) { |
3245 | printk(KERN_ERR "Could not register tmpfs\n"); | |
3246 | goto out2; | |
3247 | } | |
95dc112a | 3248 | |
ca4e0519 | 3249 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
3250 | if (IS_ERR(shm_mnt)) { |
3251 | error = PTR_ERR(shm_mnt); | |
3252 | printk(KERN_ERR "Could not kern_mount tmpfs\n"); | |
3253 | goto out1; | |
3254 | } | |
3255 | return 0; | |
3256 | ||
3257 | out1: | |
41ffe5d5 | 3258 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 3259 | out2: |
41ffe5d5 | 3260 | shmem_destroy_inodecache(); |
1da177e4 LT |
3261 | out3: |
3262 | shm_mnt = ERR_PTR(error); | |
3263 | return error; | |
3264 | } | |
853ac43a MM |
3265 | |
3266 | #else /* !CONFIG_SHMEM */ | |
3267 | ||
3268 | /* | |
3269 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
3270 | * | |
3271 | * This is intended for small system where the benefits of the full | |
3272 | * shmem code (swap-backed and resource-limited) are outweighed by | |
3273 | * their complexity. On systems without swap this code should be | |
3274 | * effectively equivalent, but much lighter weight. | |
3275 | */ | |
3276 | ||
41ffe5d5 | 3277 | static struct file_system_type shmem_fs_type = { |
853ac43a | 3278 | .name = "tmpfs", |
3c26ff6e | 3279 | .mount = ramfs_mount, |
853ac43a | 3280 | .kill_sb = kill_litter_super, |
2b8576cb | 3281 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
3282 | }; |
3283 | ||
41ffe5d5 | 3284 | int __init shmem_init(void) |
853ac43a | 3285 | { |
41ffe5d5 | 3286 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 3287 | |
41ffe5d5 | 3288 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
3289 | BUG_ON(IS_ERR(shm_mnt)); |
3290 | ||
3291 | return 0; | |
3292 | } | |
3293 | ||
41ffe5d5 | 3294 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
3295 | { |
3296 | return 0; | |
3297 | } | |
3298 | ||
3f96b79a HD |
3299 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
3300 | { | |
3301 | return 0; | |
3302 | } | |
3303 | ||
24513264 HD |
3304 | void shmem_unlock_mapping(struct address_space *mapping) |
3305 | { | |
3306 | } | |
3307 | ||
41ffe5d5 | 3308 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 3309 | { |
41ffe5d5 | 3310 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
3311 | } |
3312 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
3313 | ||
0b0a0806 HD |
3314 | #define shmem_vm_ops generic_file_vm_ops |
3315 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 3316 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
3317 | #define shmem_acct_size(flags, size) 0 |
3318 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
3319 | |
3320 | #endif /* CONFIG_SHMEM */ | |
3321 | ||
3322 | /* common code */ | |
1da177e4 | 3323 | |
3451538a | 3324 | static struct dentry_operations anon_ops = { |
118b2302 | 3325 | .d_dname = simple_dname |
3451538a AV |
3326 | }; |
3327 | ||
c7277090 EP |
3328 | static struct file *__shmem_file_setup(const char *name, loff_t size, |
3329 | unsigned long flags, unsigned int i_flags) | |
1da177e4 | 3330 | { |
6b4d0b27 | 3331 | struct file *res; |
1da177e4 | 3332 | struct inode *inode; |
2c48b9c4 | 3333 | struct path path; |
3451538a | 3334 | struct super_block *sb; |
1da177e4 LT |
3335 | struct qstr this; |
3336 | ||
3337 | if (IS_ERR(shm_mnt)) | |
6b4d0b27 | 3338 | return ERR_CAST(shm_mnt); |
1da177e4 | 3339 | |
285b2c4f | 3340 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
3341 | return ERR_PTR(-EINVAL); |
3342 | ||
3343 | if (shmem_acct_size(flags, size)) | |
3344 | return ERR_PTR(-ENOMEM); | |
3345 | ||
6b4d0b27 | 3346 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
3347 | this.name = name; |
3348 | this.len = strlen(name); | |
3349 | this.hash = 0; /* will go */ | |
3451538a | 3350 | sb = shm_mnt->mnt_sb; |
66ee4b88 | 3351 | path.mnt = mntget(shm_mnt); |
3451538a | 3352 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 3353 | if (!path.dentry) |
1da177e4 | 3354 | goto put_memory; |
3451538a | 3355 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 3356 | |
6b4d0b27 | 3357 | res = ERR_PTR(-ENOSPC); |
3451538a | 3358 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 3359 | if (!inode) |
66ee4b88 | 3360 | goto put_memory; |
1da177e4 | 3361 | |
c7277090 | 3362 | inode->i_flags |= i_flags; |
2c48b9c4 | 3363 | d_instantiate(path.dentry, inode); |
1da177e4 | 3364 | inode->i_size = size; |
6d6b77f1 | 3365 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
3366 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
3367 | if (IS_ERR(res)) | |
66ee4b88 | 3368 | goto put_path; |
4b42af81 | 3369 | |
6b4d0b27 | 3370 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 3371 | &shmem_file_operations); |
6b4d0b27 | 3372 | if (IS_ERR(res)) |
66ee4b88 | 3373 | goto put_path; |
4b42af81 | 3374 | |
6b4d0b27 | 3375 | return res; |
1da177e4 | 3376 | |
1da177e4 LT |
3377 | put_memory: |
3378 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
3379 | put_path: |
3380 | path_put(&path); | |
6b4d0b27 | 3381 | return res; |
1da177e4 | 3382 | } |
c7277090 EP |
3383 | |
3384 | /** | |
3385 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
3386 | * kernel internal. There will be NO LSM permission checks against the | |
3387 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
3388 | * higher layer. The users are the big_key and shm implementations. LSM |
3389 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
3390 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
3391 | * @size: size to be set for the file | |
3392 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3393 | */ | |
3394 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
3395 | { | |
3396 | return __shmem_file_setup(name, size, flags, S_PRIVATE); | |
3397 | } | |
3398 | ||
3399 | /** | |
3400 | * shmem_file_setup - get an unlinked file living in tmpfs | |
3401 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
3402 | * @size: size to be set for the file | |
3403 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3404 | */ | |
3405 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
3406 | { | |
3407 | return __shmem_file_setup(name, size, flags, 0); | |
3408 | } | |
395e0ddc | 3409 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 3410 | |
46711810 | 3411 | /** |
1da177e4 | 3412 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
3413 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
3414 | */ | |
3415 | int shmem_zero_setup(struct vm_area_struct *vma) | |
3416 | { | |
3417 | struct file *file; | |
3418 | loff_t size = vma->vm_end - vma->vm_start; | |
3419 | ||
66fc1303 HD |
3420 | /* |
3421 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
3422 | * between XFS directory reading and selinux: since this file is only | |
3423 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
3424 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
3425 | */ | |
3426 | file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE); | |
1da177e4 LT |
3427 | if (IS_ERR(file)) |
3428 | return PTR_ERR(file); | |
3429 | ||
3430 | if (vma->vm_file) | |
3431 | fput(vma->vm_file); | |
3432 | vma->vm_file = file; | |
3433 | vma->vm_ops = &shmem_vm_ops; | |
3434 | return 0; | |
3435 | } | |
d9d90e5e HD |
3436 | |
3437 | /** | |
3438 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
3439 | * @mapping: the page's address_space | |
3440 | * @index: the page index | |
3441 | * @gfp: the page allocator flags to use if allocating | |
3442 | * | |
3443 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
3444 | * with any new page allocations done using the specified allocation flags. | |
3445 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
3446 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
3447 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
3448 | * | |
68da9f05 HD |
3449 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
3450 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
3451 | */ |
3452 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
3453 | pgoff_t index, gfp_t gfp) | |
3454 | { | |
68da9f05 HD |
3455 | #ifdef CONFIG_SHMEM |
3456 | struct inode *inode = mapping->host; | |
9276aad6 | 3457 | struct page *page; |
68da9f05 HD |
3458 | int error; |
3459 | ||
3460 | BUG_ON(mapping->a_ops != &shmem_aops); | |
3461 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL); | |
3462 | if (error) | |
3463 | page = ERR_PTR(error); | |
3464 | else | |
3465 | unlock_page(page); | |
3466 | return page; | |
3467 | #else | |
3468 | /* | |
3469 | * The tiny !SHMEM case uses ramfs without swap | |
3470 | */ | |
d9d90e5e | 3471 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 3472 | #endif |
d9d90e5e HD |
3473 | } |
3474 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |